Divergent and convergent roles for kinases and phosphatases in neurofilament dynamics

Mammalian neurofilaments consist of three subunits, termed NF-H, NF-M and NF-L (corresponding to heavy, medium and light, according to their molecular mass) (Nixon and Shea, 1992; Pant and Veeranna, 1995), along with α-internexin and peripherin (Yuan et al., 2006; Yuan et al., 2012). NF-H and NF-M C-terminal phosphorylation fosters divalent-cation-dependent neurofilament–neurofilament interactions that mediate the formation of the stationary cytoskeleton, which provides support to the axon (Nixon, 1998; Yabe et al., 2001b; Shea and Lee, 2011; Shea and Lee, 2013). Extensively phosphorylated neurofilaments are normally segregated within axons through a complex interaction of kinases and phosphatases, disruption of which fosters aberrant accumulation of neurofilament spheroids within perikarya and proximal axons resembling those that occur during amyotrophic lateral sclerosis (ALS) (Pant and Veeranna, 1995; Julien and Mushynski, 1998; Sihag et al., 2007).

Kinases regulating neurofilament dynamics include cyclin-dependent protein kinase 5 (CDK5), mitogen-activated protein kinases (MAPKs), casein kinase 1 and 2 (CK1 and CK2), glycogen synthase kinase 3α and 3β (GSK3α and GSK3β), p38 MAPK and c-Jun terminal kinase (JNK) (Guan et al., 1991; Link et al., 1992; Guidato et al., 1996; Bajaj and Miller, 1997; Giasson and Mushynski, 1997; Veeranna et al., 1998; Bajaj et al., 1999; Li et al., 1999; Ackerley et al., 2000; Ackerley et al., 2003; Kesavapany et al., 2003; Waetzig and Herdegen, 2003; Chan et al., 2004; Barry et al., 2007; DeFuria and Shea, 2007; Perrot et al., 2008; Holmgren et al., 2012). Phosphatases that regulate neurofilament dynamics include protein phosphatase 1, 2A and 2B (PP1, PP2A and PP2B) (Shea et al., 1993; Saito et al., 1995; Strack et al., 1997; Jung and Shea, 1999; Gong et al., 2003; Veeranna et al., 2011).

Elucidation of the respective contribution of these kinases has been confounded because they mediate multiple integral roles in neuronal homeostasis (Maccioni et al., 2001; Galletti et al., 2009; Pucilowska et al., 2012; Shukla et al., 2012; Pan et al., 2013), are activated independently of neurofilament dynamics and display considerable crosstalk. For example, the MAPK pathway (i.e. MKK1 to ERK1/2) downregulates GSK3β (Ding et al., 2005). GSK3β activates MAPK pathway activity, whereas CDK5 inhibits the MAPK pathway (Noh et al., 2012) (Zheng et al., 2007). CK1-mediated phosphorylation primes several GSK3β substrates (Hagen et al., 2002; Harwood, 2002; Wang et al., 2002; Hergovich et al., 2006) and regulates CDK5 activity (Liu et al., 2001). Finally, the phosphatases that dephosphorylate neurofilaments can also positively and negatively regulate neurofilament kinases (Goldberg, 1999; Adams et al., 2005).

To surmount these difficulties, we utilized NB2a/d1 cells, which express and phosphorylate all neurofilament subunits and establish a stationary phase within axonal neurites. These cells are readily transfected, allowing selective manipulation of kinases and phosphatases, and can be grown in bulk to allow a combination of immunofluorescence analyses and immunological analyses of cellular fractions. This reductionist approach elucidated divergent and convergent roles for CK1, GSK3β, the MAPK pathway, CDK5 and PP1 in neurofilament dynamics.

Neurofilaments containing extensively phosphorylated NF-H are expressed by and incorporated into the cytoskeleton of differentiated NB2a/d1 cells (Fig. 1A,B). In efforts to identify which kinase(s) were predominantly responsible for generation of commonly studied neurofilament phospho-epitopes, we overexpressed the known neurofilament kinases CDK5, MAPK, GSK3β and CK1. Increased activity of CDK5 and the MAPK pathway have previously been demonstrated within these cells following transfection with p25 (a positive regulator of CDK5) and constitutively active MKK1 (Chan et al., 2004; Shea et al., 2004). We confirmed increased activity of GSK3β and CK1 following transfection, and their inhibition following Li⁺ and D4476 treatment (Fig. 1C,D). Homogenates of cells overexpressing these kinases were subjected to immunoblot analyses. Given that NB2a/d1 cells contain high levels of phospho- and nonphospho-neurofilaments (Fig. 1B), in order to facilitate detection of any changes in epitopes following selective kinase activation, total aliquots loaded onto gels were restricted, such that the levels in non-transfected cells were barely detectable (Fig. 1E). These analyses indicated that the SMI-31, SMI-34 and RT97 epitopes were predominantly generated by the MAPK pathway. By contrast, the RMO-24 epitope was predominantly generated by GSK3β (Fig. 1E).

To investigate further the potential roles of these kinases on neurofilament phosphorylation, we transfected cells with GFP-tagged neurofilament constructs, which facilitated the monitoring of the levels and distribution of newly expressed neurofilaments independently of differentially phosphorylated endogenous neurofilaments. In addition, expression of GFP-tagged constructs allowed us to more closely monitor the influence of kinase manipulation on neurofilament dynamics without the interference of endogenous neurofilaments.

Phosphorylation alters NF-H migration on SDS gels from 160 kDa to 200 kDa (Nixon and Shea, 1992). To determine the responsible kinase(s), we treated cells expressing GFP–NF-H with the above kinase inhibitors plus the casein kinase 2 (CK2) inhibitor tetrabromobenzotriazole (tBBT). The bulk of GFP–NF-H migrated at 195 kDa (Fig. 2), corresponding to hypophosphorylated (160 kDa) NF-H fused to 35-kDa GFP; GFP–NF-H was also detected at 260 kDa (extensively phosphorylated NF-H fused to GFP) (Lee et al., 2011). Phospho-NF-H immunoreactivity was concentrated at the 260-kDa isoform for untreated and Li⁺-, PD98059- and roscovitine-treated homogenates. D4476 increased the 195-kDa phospho-NF-H reactivity and reduced the 260-kDa phospho-NF-H reactivity (Fig. 2). D4476 did not alter GFP–NF-H distribution, indicating that this shift in phospho-reactivity was not due to altered migration of total NF-H.

This finding suggests that (1) phosphorylation events mediated by CK1 are crucial for inducing the migratory shift of NF-H to 200 kDa, (2) phosphorylation events mediated by the MAPK pathway, CDK5 and GSK3β are not sufficient to induce this migratory shift, and (3) the SMI-31, SMI-34 and RT97 phospho-epitopes can be generated in the absence of the migratory shift.

The broad band observed for extensively phosphorylated NF-H on SDS gels can be resolved into multiple phospho-dependent isoforms on gels with a low amount of acrylamide (Lewis and Nixon, 1988). GFP–NF-H, and its resultant phospho-isoforms, also appear as a relatively broad band; for example, the phospho-isoform migrating at 260 kDa spans ∼10 kDa (e.g. Fig. 3A). To examine further the impact of the above kinases on NF-H migration, we transfected cells with constructs expressing GFP-tagged C-terminal NF-H sidearms, including site-directed mutants in which the Ser residues of CDK5 consensus sequences had been replaced by Asp or Ala residues, termed GFP–NF-Hasp and GFP–NF-Hala sidearms, along with wild-type GFP–NF-H sidearms (i.e. the normal NF-H sidearm) to mimic permanently phosphorylated or nonphosphorylated states (Lee et al., 2011). These site-specific mutations allowed monitoring of potential impact of CDK5 on isoform migration. In addition, these ‘tail only’ constructs were more sensitive to subtle migratory shifts than were full-length constructs (e.g. Fig. 3B).

SMI-31 and RT97 in wild-type sidearms resolved on low-acrylamide SDS gels into two distinct isoforms migrating at ∼200 and 205 kDa. RMO-24 immunoreactivity was exclusively associated with the 205-kDa isoform and a 210-kDa isoform (Fig. 3C). The sidearm construct with Ala substitutions revealed a unique isoform migrating at ∼190 kDa, coupled with the loss of the SMI-31- and RT97-reactive 200-kDa isoform (Fig. 3D). RMO-24 immunoreactive isoforms were not affected. This suggests that the migratory shift of this isoform from 190 to 200 kDa is mediated by CDK5, given that the Ala substitution specifically blocked CDK5-mediated phosphorylation. Migration of the 205-kDa isoform was not affected, suggesting that its migration was not dependent upon CDK5 activity (Fig. 3D). No difference was detected between migration of GFP–NF-Hasp and wild-type GFP–NF-H sidearms, consistent with the presence of endogenous CDK5 activity in NB2a/d1 cells (Lee et al., 2011; Shea et al., 2004).

The majority of RMO-24 immunoreactivity was derived from GSK3β, because RMO-24 immunoreactivity was reduced by 61%±6 (mean±s.e.m.) following Li⁺ treatment; the 210-kDa RMO-24-reactive isoform was more severely depleted, and the 205-kDa isoform was relatively increased (Fig. 3E). This pattern of reduction in isoform reactivity was also observed following probing of additional replicas with a polyclonal antibody (R39) directed against all neurofilaments regardless of phosphorylation state. These findings indicate that GSK3β played a role in shifting the migration of the 205-kDa isoform to 210 kDa. By contrast, the 210-kDa isoform was not depleted in the Asp mutant sidearms (Fig. 3E). Given that Asp mutant sidearms were mutated exclusively at CDK5 consensus sites, this finding indicates that CDK5-mediated phosphorylation plays a role in generation of the 210-kDa isoform in addition to that of GSK3β. PD98059 treatment also depleted the RMO-24-reactive 210-kDa isoform, coupled with an increase in the 205-kDa isoform, suggesting that phosphorylation mediated by the MAPK pathway also contributes to generation of the 210-kDa form.

Although the major shift in migration of NF-H on SDS gels was apparently mediated by CK1 (Fig. 2), these findings demonstrate that generation of the full range of NF-H isoforms involves the MAPK pathway, CDK5 and GSK3β. In contrast to other phospho-dependent neurofilament antibodies, immunoreactivity of NF-H with RMO-24 in non-transfected cells is difficult to detect by immunoblot analyses. However, when the stacking gel was retained for transfer onto nitrocellulose, RMO-24 immunoreactivity was observed within the stack, suggesting that the majority of endogenous RMO-24-reactive NF-H isoforms are associated with SDS-resistant complexes (Fig. 3F). This possibility was confirmed by treatment of cytoskeletal preparations with 8 M urea to dissociate neurofilament complexes (Marston and Hartley, 1990), which eliminated RMO-24 immunoreactivity within the stack and increased immunoreactivity associated with the anticipated migratory position for extensively phosphorylated NF-H. Notably, some SMI-31 and SMI-34 reactivity was observed within this aggregate (Fig. 3G). This is also consistent with detection of the nearly all RMO-24 immunoreactivity within the axonal neurofilament bundle in immunofluorescence analyses (Fig. 1).

Axonal neurites contain a centrally situated bundle of closely opposed neurofilaments that undergo transport and turnover more slowly than the surrounding individual neurofilaments (Fig. 1A). Bundled neurofilaments can be separated from surrounding individual neurofilaments by sedimentation over sucrose, indicating that they are physically associated (Kushkuley et al., 2009) (Fig. 4A,B). Neurofilament–neurofilament associations leading to bundling are mediated by cation-dependent cross-bridging of phospho-neurofilaments (Kushkuley et al., 2009; Yabe et al., 2001a). To investigate which kinase(s) mediate the crucial events that promote neurofilament–neurofilament associations, we compared the impact of kinase overexpression or inhibition on intracellular GFP–NF-H distribution, and found that GSK3β activity and, to a lesser extent, CDK5 and MAPK pathway activity increased neurofilament–neurofilament associations, whereas, by contrast, CK1 activity decreased neurofilament–neurofilament associations (Fig. 4C,D).

We also examined the impact of kinase activity in cell-free analyses (Fig. 4E). Individual neurofilaments recovered from spinal cords were conjugated with Rhodamine and incubated with and without purified kinases, after which we quantified the percentage of neurofilaments that were associated with three or more other neurofilaments (Kushkuley et al., 2009). In these cell-free analyses, we were also able to treat isolated neurofilaments with multiple kinases; this was not practical to attempt within intact cells, because it would require transfection with multiple kinase constructs. In cell-free analyses, neither ERK1/2 nor GSK3β promoted neurofilament–neurofilament associations individually or in combination. Consistent with prior studies (Kushkuley et al., 2009), CDK5 induced an ∼40% increase in neurofilament–neurofilament associations. By contrast, CDK5 and GSK3β induced neurofilament–neurofilament associations for ∼90% of neurofilaments, indicating a synergistic impact of these kinases on neurofilament bundling. Notably, incubation with CDK5, GSK3β and ERK1/2 attenuated the extent of neurofilament–neurofilament associations observed following incubation with CDK5 and GSK3β to the level observed following incubation with CDK5 alone. This is consistent with the possibility that the MAPK pathway interfered specifically with the impact of GSK3β, and not that of CDK5, on neurofilament–neurofilament associations.

C-terminal phosphorylation inhibits neurofilament proteolysis (Grant et al., 2001; Pant and Veeranna, 1995). We therefore compared the relative contribution of the neurofilament kinases to protection against proteolysis. To accomplish this, we treated cells with the pharmacological inhibitors roscovitine, PD98059, Li⁺ and D4476, none of which altered GFP–NF-H levels in isolation (Fig. 5A). However, combined treatment with roscovitine and PD98059 significantly reduced total GFP–NF-H and RT97-reactive NF-H levels (Fig. 5A). By contrast, reduction was not observed following treatment with these compounds individually or in combination. Moreover, overexpression of GSK3β or CK1 could not prevent the depletion of GFP–NF-H following combined treatment with roscovitine and PD98059 (Fig. 5B). These findings indicate that phosphorylation events mediated by CDK5 and the MAPK pathway, but not by GSK3β or CK1, protect neurofilaments from proteolysis. Consistent with recent studies (Rao et al., 2012), phospho-neurofilaments within bundles were more resistant to proteolysis than were neurofilaments in homogenates (Fig. 5C). CDK5 and the MAPK pathway might therefore maintain a critical level of neurofilaments, which would indirectly promote neurofilament bundling.

MAPK pathway activity is required for transport of neurofilaments into and along axons (Chan et al., 2004) but the responsible mechanism remains unclear. Notably, the MAPK pathway inactivates GSK3β (Ding et al., 2005). Herein, we have demonstrated that GSK3β participates in neurofilament bundling, whereas MAPK pathway activity attenuated GSK3β-mediated bundling (Fig. 4). Given that neurofilaments that undergo bundling are withdrawn from the transporting pool (Shea and Lee, 2011), we hypothesized that the mechanism by which the MAPK pathway mediates neurofilament transport might be through inhibiting GSK3β-mediated neurofilament–neurofilament interactions.

If this were indeed the case, we hypothesized that inhibition of neurofilament transport by treatment with PD98059 coupled with overexpression of GSK3β might have an additive effect. To test this possibility, we first confirmed that the MAPK pathway phosphorylated GSK3β at Ser9 (Fig. 6A). Overexpression of GSK3β in which Ser9 was mutated to Ala (GSKala, which cannot be inactivated by phosphorylation and is therefore constitutively active) inhibited neurofilament transport into and along axonal neurites (Fig. 6B). Treatment with PD98059 (which inhibits MAPK pathway activity) inhibits neurofilament transport and fosters perikaryal neurofilament accumulation (Chan et al., 2004) (see also Fig. 6C for representative image). However, when cells overexpressing GSKala were treated with PD98059, we observed an increase in perikaryal neurofilament bundles (Fig. 6C). These findings indicate that the MAPK pathway promotes neurofilament axonal transport by inhibiting GSK3β activity, and in doing so, might preclude inappropriate perkaryal neurofilament bundling.

Overexpression of CDK5 induces the accumulation of phospho-neurofilament bundles within perikarya (Shea et al., 2004). Given the above demonstration of a crucial role for GSK3β in perikaryal neurofilament accumulation, we questioned whether or not GSK3β participated in CDK5-induced perikaryal neurofilament accumulation. To investigate this possibility, we overexpressed GFP–NF-H along with p25 or GSK3β, and treated cells with Li⁺ or roscovitine (inhibitors of GSK3β and CDK5, respectively) prior to harvest. Both p25 and GSK3β overexpression increased the percentage of cells with perikaryal neurofilament bundles. However, treatment with Li⁺ inhibited p25-induced perikaryal neurofilament accumulation, and treatment with roscovitine inhibited perikaryal neurofilament accumulation induced by GSK3β (Fig. 6D). These findings indicate that a combination of CDK5 and GSK3β activity is required to induce perikaryal neurofilament accumulations, and are consistent with our demonstration that GSK3β potentiated CDK5-induced neurofilament–neurofilament associations in cell-free analyses (Fig. 4).

The above results (Fig. 4) indicate that CK1 apparently inhibits neurofilament bundling. This finding seemed counterintuitive, because CK1 activity was responsible for the shift in NF-H migration on SDS gels that was attributed to extensive phosphorylation (Fig. 2), and bundles contain the bulk of extensively phosphorylated neurofilaments (Kushkuley et al., 2009; Yabe et al., 2001a). We therefore examined the levels and distribution of total (GFP–NF-H) and RT97-reactive NF-H within total homogenates and bundle fractions derived from cells following overexpression of CK1 and treatment with D4476 (a pharmacological inhibitor active against CK1) (Fig. 7A). Visual inspection, confirmed by densitometric analyses, revealed that overexpression of CK1 statistically reduced total RT97 immunoreactivity (P<0.01, ANOVA), whereas treatment with D4476 statistically increased total RT97 immunoreactivity (P<0.01, ANOVA). Overexpression of CK1 decreased the amount of total and RT97-reactive NF-H within bundles (trend towards significance, P<0.06, ANOVA), whereas treatment with D4476 increased the amount of total and RT97-reactive NF-H within bundles (trend towards significance, P<0.08, ANOVA). Immunofluorescence analyses demonstrated an increase in axonal and perikaryal RT97 reactivity following D4476 treatment (Fig. 7B).

Given that CK1 activates PP1 (Henry and Killilea, 1993), we probed whether or not CK1 regulated neurofilament dynamics through modulation of PP1 activity. D4476 treatment significantly (P<0.05) reduced PP1 but not PP2A activity in perikarya and axonal neurites (Fig. 7C). Calyculin, a pharmacological agent active against PP1, fostered the formation of SDS-resistant high-molecular-mass material that was reactive with RT97 but not the nonphospho-neurofilament antibody SMI-32 (Fig. 7D), and increased RT97 immunoreactivity within perikarya and axonal neurites (Fig. 7E). These findings indicate that PP1 activity regulates NF-H C-terminal phosphorylation, including restricting phospho-neurofilament accumulation within perikarya, and that CK1 might restrict neurofilament phosphorylation through activation of PP1.

The above findings highlight that the activity of the MAPK pathway, CDK5, CK1 and GSK3β exert divergent roles on neurofilament dynamics, including inhibition of proteolysis (MAPK and CDK5; Fig. 5), restriction of neurofilament phosphorylation within perikarya, promotion of neurofilament transport into axonal neurites (MAPK and CK1; Figs 6,7) and promotion of bundling (CDK5, GSK3β; Figs 6,7). Given that the collective impact of these diverse functions would foster an increase in phospho-neurofilaments within axonal neurites, we considered that increased activity of each of these kinases would contribute either indirectly or directly to the establishment and maintenance of axonal neurofilament bundles. In support of this notion, overexpression of each of these kinases both increased the percentage of neurites displaying neurofilament bundles and CDK5, CK1 and GSK3β each increased the association of GFP–NF-H with bundles (Fig. 8A,B). These findings confirm that each of these kinases contributes directly or indirectly to the incorporation of neurofilaments into axonal bundles.

To address further the role of GSK3β in bundling, we compared incorporation into the bundle-enriched fraction of full-length GFP–NF-H with incorporation of GFP–NF-H in which the terminal 187 amino acids [the region of the C-terminal sidearm reported to be essential for bundling (Chen et al., 2000)] had been deleted (NF-HΔ187). Cells expressing HΔ187 generated prominent GFP-reactive species migrating at ∼150 kDa on SDS gels, which corresponds to the anticipated migratory position of 115-kDa NF-H (i.e. lacking the terminal 187 amino acids) fused to GFP. Additional slower-migrating GFP-reactive species were observed between 155 and 175 kDa (Fig. 8C), and these displayed prominent immunoreactivity with antibodies directed against phospho-dependent neurofilament C-terminal epitopes (RT97, SMI34 and SMI31), confirming retention of these epitopes within the proximal portion of the sidearm. Retardation of migration of phospho-reactive HΔ187 isoforms demonstrates its ability to undergo phospho-mediated conformation alterations that foster retardation of full-length NF-H migration on SDS gels (Pant and Veeranna, 1995; Shea and Chan, 2008).

GFP–NF-HΔ187 co-assembled with the endogenous neurofilament network as shown by its distribution within the cytoskeleton and colocalization with filamentous profiles (Fig. 8D). Despite deletion of the portion of the NF-H sidearm purported to mediate neurofilament–neurofilament bundling, GFP–NF-HΔ187 was found within bundles both in cellular fractionation and immunofluorescence analyses (Fig. 5B,C), in a manner that was mediated by C-terminal crosslinking among endogenous (full-length) NF-H co-assembled into the same neurofilaments as shown previously (Kushkuley et al., 2009; Lee et al., 2011).

We next compared the influence of overexpression of GSK3β and CK1, because the majority of NF-H consensus sites for these kinases exist within the distal-most 187 amino acids (Chen et al., 2000; Hollander and Bennett, 1992; Hollander et al., 1996; Sasaki et al., 2002; Shaw et al., 1997), on bundling of GFP–NF-H and GFP–NF-HΔ187 (Fig. 8E). In the absence of kinase overexpression, both GFP–NF-H and GFP–NF-HΔ187 displayed an identical relative distribution within bundles versus the surrounding axoplasm. However, overexpression of GSK3β or CK1 each increased the relative amount of GFP–NF-H that was associated with axonal bundles, but did not alter the association of GFP–NF-HΔ187 within bundles. These findings suggest that GSK3β- and CK1-mediated phosphorylation of sites within the distal 187 amino acid residues of the NF-H C-terminal tail plays a crucial role in neurofilament bundling.

Key phosphorylation events foster the neurofilament–neurofilament associations that generate the stationary phase. Neurons are faced with the task of preventing or eliminating those events within perikarya, which would otherwise result in accumulation of perikaryal spheroids of phospho-neurofilaments, which are characteristic of conditions such as ALS, yet promoting these events within axons, without which the developing axon will not undergo stabilization (Julien and Mushynski, 1998; Pant and Veeranna, 1995; Shea and Lee, 2011; Shea and Lee, 2013). The findings presented herein elucidate divergent roles for neurofilament kinases and phosphatases that encompass axonal transport and the establishment and/or maintenance of the stationary phase. These functions were mediated in part by direct phosphorylation of neurofilaments, but also by interactions among kinases and phosphatases.

We demonstrated herein that GSK3β activity is essential for neurofilament–neurofilament interactions leading to incorporation of neurofilaments into the stationary phase, which is readily seen in NB2a/d1 cells and cultured neurons by the appearance of tightly associated ‘bundled’ neurofilaments (Kushkuley et al., 2009; Yabe et al., 2001a; Yuan et al., 2009). However, GSK3β-induced bundling was not restricted to axonal neurites but also occurred within perikarya. We demonstrated herein that the MAPK pathway, activity of which is essential for transport of neurofilaments into and along axonal neurites of NB2a/d1 cells (Chan et al., 2004), promotes neurofilament axonal transport by inhibiting GSK3β activity: increasing MAPK pathway activity increased GSK3β phosphorylation at Ser9 [which inactivates GSK3β (Sutherland et al., 1993)] and prevented GSK3β-induced neurofilament–neurofilament association, whereas inhibition of MAPK pathway activity potentiated GSK3β-induced inhibition of axonal transport and accumulation of neurofilament bundles within perikarya.

CK1 activated PP1, which was essential for translocation of neurofilaments out of perikarya and into axonal neurites; inhibition of CK1 activity or direct inhibition of PP1 fostered accumulation of phospho-neurofilament immunoreactivity and neurofilament bundles within perikarya. This finding suggests that the normal segregation of extensive phosphorylation and resultant neurofilament bundling to axons is achieved by maintaining a relatively higher ratio of phosphatase to kinase activity within perikarya than within axonal neurites. This possibility is consistent with the rapid de novo accumulation of phospho-neurofilament immunoreactivity within retinal ganglion cell perikarya and proximal axons in situ following inhibition of PP2A (Jung and Shea, 1999).

The above findings demonstrate essential roles for the MAPK pathway and CK1 in promotion of neurofilament transport out of perikarya. We did not determine whether or not the MAPK pathway and CK1 mediated continued transport of neurofilaments along axonal neurites by their respective inhibition of GSK3β and activation of PP1. However, transporting neurofilaments enter and leave the stationary cytoskeleton (Kushkuley et al., 2009; Lewis and Nixon, 1988; Nixon and Logvinenko, 1986; Trivedi et al., 2007; Yabe et al., 2001a; Yuan et al., 2009). This finding, coupled with the finding that there are increased phospho-neurofilaments along axons following inhibition of PP1 and PP2A (herein and Shea et al., 1993), suggests that there are ongoing cycles of neurofilament phosphorylation and dephosphorylation within axons.

In addition to regulation of PP1 activity, CK1 mediated an NF-H migratory shift on SDS gels from 160 kDa to 200 kDa. Herein, we observed prominent RT97 immunoreactivity with the entire range of NF-H isoforms, including the 160-kDa isoform, following CK1 inhibition. This was unexpected, because RT97 immunoreactivity, which is generated by the MAPK pathway (Veeranna et al., 2008), is normally associated with the most highly phosphorylated NF-H isoforms, migrating at ≥200 kDa (Jung and Shea, 1999; Veeranna et al., 2008; Yabe et al., 2000). However, prior studies have not simultaneously manipulated CK1 along with proline-directed CDK5 and MAPK activity. Given that CK1 is constitutively active, CK1-mediated phosphorylation events responsible for the shift of NF-H from 160 to 200 kDa would occur constitutively and independently of MAPK activity; whereas MAPK manipulation alters RT97 immunoreactivity, this immunoreactivity would routinely be detected on isoforms already phosphorylated by CK1. This line of reasoning is supported by the observation of 200-kDa NF-H within perikarya and axonal hillocks, whereas RT97 is normally restricted to the axonal shaft (Jung and Shea, 1999; Sánchez et al., 2000; Yabe et al., 2000).

The use of SDS gels containing a relatively low acrylamide content revealed an array of NF-H isoforms similar to those observed in optic pathway (Lewis and Nixon, 1988). Site-directed mutagenesis of CDK5-specific phosphorylation sites induced minor alterations in migration of NF-H migration on SDS gels. GSK3β activity induced RMO-24 reactivity and the appearance of the slowest-migrating neurofilament isoform, corresponding to the 210-kDa isoform in optic pathway (Lewis and Nixon, 1988). Inhibition of the MAPK pathway depleted this slowest-migrating isoform, indicating that a combination of MAPK and GSK3β activity generated this isoform.

Phosphorylation inhibits neurofilament proteolysis (Pant and Veeranna, 1995). Our findings have provided further insight into this regulation by demonstrating that phosphorylation by CDK5 and MAPK, and not by CK1 and/or GSK3β, mediated this protection. We observed increased NF-H proteolysis following simultaneous inhibition of CDK5 and the MAPK pathway but not upon inhibition of GSK3β or CK1, and overexpression of GSK3β and CK1 failed to prevent NF-H proteolysis following inhibition of CDK5 and the MAPK pathway. These findings indicate that phosphorylation of the KSP domains provides protection against proteolysis. Given that the terminal 187 amino acid residues, which contains the majority of the CK1 and GSK3β consensus sequences, is essential for bundling (Chen et al., 2000), one speculation arising from the differential impact of these kinases on proteolysis and bundling is that the C-terminal 187 amino acid residues, rather than the KSP-rich regions, might mediate the cation-dependent crosslinking that incorporates neurofilaments into bundles (Kushkuley et al., 2009), leaving the more proximal KSP-rich region, or a portion of it, exposed to potential protease activity. If this were indeed the case, dephosphorylation of the KSP-rich region could allow proteolysis of bundled neurofilaments, which might provide a means for localized remodeling of the cytoskeleton, including any required axonal branching (Xie et al., 2006).

Comparison of GFP–NF-H within axonal neurites and bundles indicated that CK1, GSK3β and CDK5 all increased bundling of neurofilaments, indicating that once neurofilaments were within axons, all of these kinases contributed to the establishment and maintenance of the stationary phase. However, our systematic overexpression and inhibition highlighted that these kinases regulated a hierarchical series of events encompassing proteases and phosphatases. GSK3β-mediated neurofilament bundling required prior phosphorylation of those neurofilaments by CDK5. MAPK-pathway-mediated downregulation of GSK3β activity within perikarya was essential to restrict segregation of bundling within axonal neurites. Like the MAPK pathway, CK1 also promoted neurofilament transport into axonal neurites by restricting the accumulation of extensive phosphorylation within perikarya, although CK1 mediated this by activation of PP1 rather than inhibition of GSK3β. In addition, inhibition of neurofilament proteolysis by CDK5 and MAPK also likely contributed to bundling by maintaining sufficient concentration of axonal neurofilaments. Promotion of neurofilament transport into axons by the MAPK pathway and CK1, inhibition of neurofilament proteolysis by the MAPK pathway and CDK5, and prior neurofilament phosphorylation at least by CDK5 represent mechanisms by which these kinases could contribute to GSK3β-induced neurofilament bundling within axons. These findings collectively indicate that the impact on neurofilament dynamics of the non-proline-directed kinases CK1 and GSK3β are functionally regulated by the proline-directed neurofilament kinases of the MAPK pathway and CDK5.

A limitation of these analyses is that we did not dephosphorylate neurofilaments prior to examination of the impact of kinase manipulation. Accordingly, we cannot completely exclude the possibility that prior phosphorylation events contributed to the changes observed following manipulation of individual kinases. Unfortunately, experimental neurofilament dephosphorylation results in rapid proteolysis within cells (Pant and Veeranna, 1995) and under cell-free conditions because of neurofilament-associated proteases (Kushkuley et al., 2009). A further limitation is that we were unable to examine the consequences of simultaneous overexpression of multiple kinases. We have routinely manipulated kinases in cells co-transfected with GFP–NF-H or GFP–NF-M by utilizing two times more of the respective kinase construct versus the GFP-tagged neurofilament construct. To examine the consequence of multiple kinases would require successful triple transfection (i.e. GFP–NF-H plus two or more kinase constructs), which is impractical.

The kinases studied herein have multiple substrates beyond neurofilaments. As such, we cannot exclude the possibility that at least some of the impact of kinase and/or phosphatase manipulation on neurofilaments, as seen hererin, was derived by indirect effects on other cellular pathways. For example, although neurofilament phosphorylation clearly provides resistance to proteolysis (Pant and Veeranna, 1995), we cannot exclude the possibility that manipulation of kinase and/or phosphatase activities also suppressed activity of proteolytic enzymes themselves, which would further contribute to increased neurofilament levels. Similarly, whereas phosphorylation promotes neurofilament bundling, and in doing so removes neurofilaments from the transporting pool (Shea and Lee, 2011; Kushkuley et al., 2009), the kinases examined herein also directly and indirectly have an impact on kinesin itself (Morfini et al., 2002). We therefore cannot completely exclude the possibility that at least some of the alterations in neurofilament distribution following kinase and/or phosphatase manipulations were derived at least in part from disruptions in overall axonal transport. This likelihood is reduced, however, by our prior demonstration that manipulation of the MAPK pathway and CDK5 by the same methodologies utilized herein did not alter axonal transport of tau (which is also dependent upon kinesin) in these cells (Moran et al., 2005; Dubey et al., 2008). Notably, even if the kinases and phosphatases studied herein mediate some of their impact on neurofilaments by modulation of proteolytic and transport systems, it does not diminish the overall conclusion that the collective impact of these divergent kinase activities increases the amount of phospho-neurofilaments within axonal neurites, which in turn contributes to establishment and maintenance of the stationary phase.

C-terminal neurofilament phosphorylation does not directly interfere with neurofilament axonal transport, but indirectly interferes with transport by allowing some neurofilaments to withdraw from the transporting pool to establish the stationary cytoskeleton, which is essential for axonal maturation. This avoids the need for any alteration in the transport system itself to foster axonal maturation. Notably, this allows continued activity of transport systems within axons. Using the same transport battery throughout neuronal differentiation and maturation simplifies these dynamics, allows stabilization of proximal axonal regions while distal regions are still elongating, and ultimately allows transition into a maintenance system that can repair and replace damaged or worn out neurofilaments at any locus along the axon.

The kinases studied herein share pivotal roles in neuronal homeostasis. Any requirement to modulate their activity to foster neurofilament transport and assembly of the stationary phase would impact on multiple essential neuronal pathways. However, given that extensively phosphorylated neurofilaments essentially undergo self-assembly to form the stationary phase within axons, there is no apparent requirement for a developing neuron to modulate the activity of these kinases, nor their corresponding phosphatases, for establishment or repair of the stationary phase. Intermediate filaments provide mechanical strength to cells and mediate the formation of tissues (<<citref rids="ref16">Fuchs, 1994</citref>). Perhaps nowhere can the need for long-term, stable structural support be more crucial than in axons, which, once synaptogenesis has occurred, remain in place for the lifetime of the individual (Shea and Lee, 2013). The interplay of kinase and phosphatase activities on neurofilament dynamics as demonstrated herein demonstrate how this class of intermediate filaments can establish a long-lasting and supportive macrostructure polarity during development without the need to alter activity of participating kinases or that of the overall transport system.

Mouse NB2a/d1 neuroblastoma cells were cultured in DMEM containing 10% fetal bovine serum and differentiated with 1 mM dibutyryl cyclic AMP (dbcAMP) (Yabe et al., 1999). For simplicity and clarity of writing, translocation of neurofilaments into and along axonal neurites of these cells is referred to as ‘axonal transport’.

Cells were transfected using Lipofectamine (Invitrogen, Carlsbad, CA) with constructs expressing GFP-tagged full-length NF-H (GFP–NF-H), the GFP-tagged isolated NF-H sidearm (i.e. lacking the rod domain, wild-type GFP–NF-H sidearms), GFP-tagged sidearms in which the serine residues in the C-terminal CDK5 consensus sites were mutated to aspartate residues (GFP–NF-Hasp) or alanine residues (GFP–NF-Hala) to mimic permanently phosphorylated or nonphosphorylated states (Ackerley et al., 2003; Lee et al., 2011), and GFP-tagged rat NF-H lacking the distal-most C-terminal 187 amino acids (GFP–NF-HΔ187). GFP–NF-HΔ187 was prepared from the above full-length GFP-tagged NF-H construct. Given that there is a unique AccI site in the C-terminal region of the rod domain of NF-H (Chen et al., 2000), the sequence between AccI and the last KSP motif was amplified by PCR using the following set of primers: 5′-AGAGTCGCCAAAGTGAACACGGATGCT-3′ and 5′-CTGAGGATCCTAAGGGGACTTCACTTC-3′. Amplified PCR fragments were digested with AccI and BamHI, for which consensus sites were included in the primers, then cloned into the prepared template plasmid digested with the same enzymes. A ≥70% transfection efficiency for GFP-tagged neurofilament constructs is attained under these conditions (Chan et al., 2004).

Cells were transfected with a plasmid expressing constitutively active mouse GSK3β (a generous gift from Chris Miller, Institute of Psychiatry, King's College, UK), mouse CK1-α (denoted CK1 herein, Origene, Rockville, MD), constitutively active MKK1 (an upstream activator of the MAPK pathway) (Li et al., 1999) and p25 [the truncated and constitutively active form of the CDK5 activator p35 (Patrick et al., 1999)]. Additional cultures were treated with the pharmacological inhibitors roscovitine (20 µM), PD98059 (10 µM), Li⁺ (10 mM), D4476 (100 µM) and tetrabromobenzotriazole (tBBT; 60 µM), which are active against CDK5, MKK1, GSK3β, CK1 and CK2 respectively (Cheng et al., 1983; Dudley et al., 1995; Meijer et al., 1997; Rena et al., 2004; Sarno et al., 2001). Given that the downstream impact of manipulation of MKK1 on neurofilaments is alteration of ERK1/2 kinases, we refer to MAPK pathway manipulation at points for simplicity of writing. For co-transfection with kinases and GFP–NF-H, more than two times of the kinase constructs versus GFP–NF-H were utilized (1 µg and 0.5 µg respectively); insuring that cells displaying GFP were co-transfected with the kinase construct (Chan et al., 2004; Shea et al., 2004).

Cells were homogenized in 50 mM Tris-HCl, pH 6.8, containing 1% Triton X-100, 5 mM EDTA, 1 mM PMSF and 50 µg/ml leupeptin and centrifuged (15,000 g; 15 min). The resulting pellet was defined as the Triton-insoluble cytoskeleton, and the resulting supernatant was defined as the Triton-soluble fraction. Spinal cord neurofilaments from adult C57BL6 mice of both genders (all animal procedures were carried out in accordance with the approval of our Institutional Animal Care and Use Committee) were resuspended in 0.1M MES, pH 6.8, containing 1 mM MgCl₂, 1 mM EGTA, 1 mM PMSF and 50 µg/ml leupeptin and labeled with Rhodamine as described previously (Kushkuley et al., 2009; Wagner et al., 2003).

To obtain fractions enriched in bundled neurofilaments, additional homogenates were layered over the same buffer containing 1 M sucrose, and centrifuged at 15,000 g for 15 min. Bundled neurofilaments sedimented through the sucrose cushion; ‘individual’ neurofilaments (not contained within bundles) were recovered at the sucrose interface (Kushkuley et al., 2009; Yabe et al., 2001a). Protein concentration was determined by a BCA assay (Thermo scientific).

Individual neurofilaments obtained as described above were incubated with or without purified kinases (Kushkuley et al., 2009) for 2 h at 37°C with 0.1 µg/µl CDK5 and its activator p35, ERK1/2 and/or GSK3β (Upstate Biochemicals, Lake Placid, NY). The percentage of neurofilaments aligned with three or more than other neurofilaments were quantified by using fluorescence microscopy (Kushkuley et al., 2009). Closely aligned neurofilaments typically splay apart at their ends, facilitating detection of multiple neurofilaments (Kushkuley et al., 2009).

Samples were normalized according to total protein, subjected to SDS gel electrophoresis and transferred onto nitrocellulose. The bundled fraction was solubilized with 8 M urea for electrophoresis. Membranes were blocked with 5% BSA and 5 mM sodium fluoride in Tris-buffered saline containing 0.1% Tween-20 for 1 h then incubated overnight at 4°C with antibodies directed against GFP (1∶1000, Invitrogen) and antibodies directed against neurofilament phospho-epitopes [RT97 (generous gift of Brian Anderton, Institute of Psychiatry, London, UK), SMI-34 and SMI-31 (Covance; Dedham, MA) and RMO-24 (Invitrogen)], nonphospho-epitopes (SMI-32) and an antibody directed against neurofilaments regardless of phosphorylation state (R39) (Jung and Shea, 1999). Membranes were washed with the same buffer then incubated with alkaline-phosphatase-conjugated secondary antibodies for 1 h at room temperature and developed using a NBT/BCIP substrate kit (Promega, Madison, WI). Immunoreactive species were quantified in digitized images of replicas using Image J; the background signal from an adjacent, identically sized, area in the identical lane was subtracted from each reactive species (Yabe et al., 2001a; Yabe et al., 1999). All samples to be compared were electrophoresed on the same gel, transferred onto nitrocellulose and visualized simultaneously.

Cells grown on poly-L-Lysine-treated coverslips were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS, pH 7.4) for 10 min at room temperature, rinsed two times in PBS (5 min/rinse), and blocked for 30 min in PBS containing 1% BSA and 2% normal goat serum. Cultures were then incubated overnight at 4°C in PBS containing 1% BSA and a 1∶500 dilution of anti-GFP antibody or against R39, or 1∶100 dilutions of antibodies against RT97, SMI34, SMI-31, SMI-32 and RMO-24. Cultures were rinsed three times with PBS, incubated for 30 min at 37°C in PBS containing 1% BSA and 1∶300 dilutions of appropriate secondary antibodies.

Neurofilament distribution was quantified using ImageJ. Translocation of neurofilament constructs into and along axonal neurites was quantified as an index of axonal transport. The shaft of axonal neurites was divided into ten equivalent segments, excluding the hillock and the growth cone, and the percentage of GFP within each segment was calculated relative to the total GFP within the axonal shaft (Yabe et al., 2001b). In additional studies, axonal neurites (>25 for each condition) were divided into three equivalent segments (defined as proximal, central and distal), excluding the hillock and growth cone. The translocation rate was determined by dividing the GFP intensity in the distal fragment by that of central fragment within each neurite (Chan et al., 2004). To monitor the extent of incorporation of GFP-tagged neurofilaments into axonal neurofilament bundles, the GFP intensity of the bundle and that of the area adjacent to the bundle were quantified within the central neurite segment for >20 cells for each condition (Chan et al., 2004; Chan et al., 2005).

GSK3β activity was assayed by monitoring phosphorylation of GSK3β at Ser9 (which inactivates GSK3β) and phosphorylation of the GSK3β substrate β-catenin at Ser33, Ser37 and Thr41. Homogenates of cells with or without GSK3β overexpression or 10 mM Li⁺ treatment were subjected to electrophoresis, transferred onto nitrocellulose and probed with antibodies directed against total GSK3β, GSK3β that had been phosphorylated at Ser9, total β-catenin, and β-catenin phosphorylated at Ser33, Ser37 and Thr41 (Cell Signaling), and, as a loading control, anti-tubulin antibody DM1A. Given that β-catenin is degraded once phosphorylated by GSK3β (Ding et al., 2005), a decrease in phospho-β-catenin (which can also be reflected in total β-catenin levels) provides an index of GSK3β activity.

CK1 activity was monitored by casein gel zymology (Cheng and Louis, 1999). Lysates from cells incubated for 4 h with or without 100 µM D4476 were separated on a 12% polyacrylamide gel polymerized in the presence of 1 mg/ml dephosphorylated casein. After electrophoresis, SDS was removed from the gel by rinsing twice for 30 min with 50 mM Tris-HCl, pH 8.0, containing 20% isopropanol. The gel was washed twice for 30 min in 50 mM Tris-HCl containing 5 mM 2-mercaptoethanol. Electrophoresed proteins were denatured by incubating the gel twice for 30 min in the above buffer containing 6 M guanidine-HCl, and were then renatured by 15 h of washing at 4°C in four changes of the above buffer containing 0.05% Tween 40. The gel was equilibrated in 10 mM Tris-HCl, pH 7.5, containing 10 mM MgCl₂ for 30 min at room temperature. Casein phosphorylation was carried out by incubation of the gel in 10 mM Tris-HCl, pH 7.5, containing 10 mM MgCl₂, 25 µM ATP and 2.5 µCi/ml, [γ-³²P]ATP for 2 h at room temperature followed by washing with 5% trichloroacetic acid and 1% sodium pyrophosphate until no radioactivity was detected in the wash. The gel was dried and exposed to X-ray film.