Tau missorting and spastin-induced microtubule disruption in neurodegeneration: Alzheimer Disease and Hereditary Spastic Paraplegia

To elucidate the pathological cascade leading to neurodegeneration in AD, we established a system to model the effects of Aβ on Tau. In mature rodent primary neurons, Tau is present in several isoforms, is sorted into the axon and excluded from the somatodendritic compartment, all of which is similar to the adult human brain. To trigger AD-like pathological changes of Tau, we exposed primary neurons to a pre-aggregation form of Aβ (oligomeric Aβ), which is thought to be the toxic trigger in AD [3, 24, 35]. We investigated changes of Tau and related cellular pathologies using biochemical methods, fluorescence microscopy including live cell imaging, and electron microscopy.

After exposure to oligomeric Aβ, Tau becomes mislocalized (missorted) into the somatodendritic compartment, a feature reminiscent of incipient AD [5]. Interestingly, the missorted Tau does not originate from the axon, but is newly synthesized, fails to become sorted into the axon, and instead spreads throughout the somatodendritic compartment. Missorting of Tau correlates with a loss of synapses, most pronounced in dendrites containing high amounts of Tau [81]. This provides a link between the mislocalization of Tau and the cognitive decline observed in mouse models of AD and in AD patients. One of the key features is a dramatic loss of microtubules in neurons with missorted Tau [80]. Microtubules are of great importance as tracks of the intracellular traffic system based on microtubule-dependent motor proteins. This is particularly so for neurons whose extended processes require highly efficient transport.

Treatment of neurons with Aβ oligomers leads to missorting of Tau into dendrites, with subsequent decay of microtubules and mature spines (Fig. 1a, b, c). Consequently, mitochondria, vesicles, and other cargo suffer from traffic jams in these dendrites. Neurofilaments that are usually transported along microtubules from the soma into the axon, become also mislocalized to the somatodendritic compartment, which indicates impaired transport.

Most of these pathological changes depend on the presence of Tau, as in neurons derived from Tau knockout mice (TauKO), loss of microtubules and synapses, aberrant mitochondria and neurofilament distribution do not occur [78].

In search for the causes of microtubule breakdown in dendrites, we found that the decay of microtubules is executed by spastin, a microtubule severing enzyme. Spastin in turn is recruited to microtubules by polyglutamylation of microtubules, conferred by the enzyme TTLL6 (Tubulin-Tyrosin-Ligase-Like-6) and possibly related isoforms. TTLL6 is a microtubule modifying ligase which catalyzes the addition of polyGlu residues to Glu residues near the C-terminal tail of α-tubulin. This highly acidic configuration leads to the recruitment of the microtubule severing protein spastin and subsequent destruction of microtubules. TTLL6 is mislocalized from the soma into dendrites by a Tau dependent mechanism, likely involving an interaction of TTLL6 with the N-terminal half of Tau [78] (Fig. 1d) [38].

Normal spastin levels seem to be essential for the microtubule breakdown induced by Tau missorting and Aβ-exposure. In control neurons, Aβ exposure leads to microtubule destruction, while knockdown of spastin via shRNA results in stable microtubules in spite of Aβ exposure (Fig. 2a). Of note, once the toxic stimulus of Aβ-oligomers disappears, microtubule density recovers, implying that without constant or repetitive exposure to toxic, oligomeric forms of Aβ, spastin over-activity returns to baseline levels without further intervention [78]. Thus, spastin is the effector of microtubule breakdown in this cell culture model of Alzheimer disease (Fig. 2b). The flip side of the spastin coin is the suggested mechanisms of HSP, where mutant or absent spastin activity impairs normal microtubule dynamics (see below), which results in bundled and overstable microtubules (Fig. 2c) [32].

Apart from the scenario described above, Tau has been implicated in several ways in the impairment of microtubules and synapses. Along with the related proteins MAP2 and MAP4, Tau belongs to the class of "structural MAPs" capable of binding to and stabilizing microtubules. Proper axonal sorting of Tau depends on intact Tau-microtubule interactions [34]. In the pathological condition of Tau missorting, however, Tau is altered by posttranslational modifications, notably phosphorylation at a number of sites [25, 48]. In contrast to low-level physiological phosphorylation, pathologically increased phosphorylation could lead to detachment of Tau from microtubules, with subsequent destabilization of microtubules (for discussion of physiological vs. pathological phosphorylation see [1, 77, 79]). An example is the phosphorylation of Tau at the KXGS-motifs in the repeat domain of Tau [71, 81]. These sites regulate the binding of Tau to microtubules and are targets of the Microtubule-Affinity-Regulating Kinase (MARK). In maturing neurons MARK is most active in growth cones, in the soma and the dendrites, but not in axons [72]. When phosphorylated at the KXGS-motifs, Tau cannot bind and stabilize microtubules, and can become missorted [43]. This in turn can lead to abberant interactions of missorted Tau with proteins in the somatodendritic compartment, e.g. the tyrosine kinase Fyn [30].

Tau has been shown to protect microtubules against attack by another microtubule severing protein, katanin [52]. Thus, detachment of Tau from microtubules as a result of Tau phosphorylation could further promote the destabilization of microtubules, either by lack of stabilization and/or by loss of protection against severing. Tau would then also be free to translocalize into spines and mediate the loss of synapses, as shown by experiments using pseudophosphorylated or mutated Tau [28, 78]. The precise mechanism of how phosphorylated Tau induces loss of mature spines and loss of f-actin in spines is unclear. Suggested mechanisms circle around a NMDA-receptor mediated calcium rise, as glutamate triggers similar pathological effects [81]. For example, NMDA-receptors could be released from the synapse to the dendritic shaft after f-actin disassembly mediated by phosphorylated Tau. This would cause long term depression and excitotoxicity, possibly by a calcineurin dependend mechanism [30, 57, 74, 79].

Mutations in the gene encoding spastin –SPAST– are the single most common cause of autosomal dominant hereditary spastic paraplegia (HSP) and cause about 40 % of familial HSP and about 20 % of sporadic HSP (for recent review on HSP see [21]). Spastin protein occurs in four isoforms (Fig. 3a), depending on alternative splicing of exon 4 and an alternative translation initiation codon, but the predominant isoforms are of 60 and 68 kDa size [11].

Spastin has diverse cellular functions. They include membrane shaping and membrane trafficking, but the best known function is microtubule severing. Spastin consists of two main structural domains (Fig. 3a): (i) a microtubule interacting and transport (MIT) domain at the N-terminus, and (ii) an AAA-ATPase domain at the C-terminus, which is responsible for microtubule severing (Fig. 3b). Attached to the AAA-ATPase domain is the microtubule binding domain (MTBD), thus spastin has two domains interacting with microtubules. Near the N-terminus there is also a transmembrane (TM) domain possibly involved in ER shaping and trafficking. Only the full length isoform (M1) with 616aa contains the TM domain, while the slightly shorter isoform M87 (M85 in rodents, hereafter referred to as M87), lacks the highly hydrophobic 86 N-terminal aa, which include the TM domain [11].

In HSP with SPAST mutations the MT-severing activity of spastin is decreased: Over 300 pathological mutations, including partial deletions, have been described, mostly in the AAA-ATPase domain [60]. In SPAST related HSP, haploinsufficiency of spastin results in impaired microtubule severing and as such represents a loss of function phenotype. As the MIT or the MTBD domain in HSP-patients is still intact, spastin mutants can still bind to and crosslink microtubules, which could result in bundles of stable microtubules [62]. This represents a gain of pathologic function (see below).

Thus, in a formal sense, the view that microtubule severing by spastin might be impaired in HSP (i.e. under-activity of spastin) is the opposite to our proposed pathological cascade of AD, which involves aberrant microtubule breakdown by spastin (i.e. over-activity of spastin; Fig. 2b, c, Fig. 3c). Under- and over-activity of spastin are the two sides of the same coin, both leading to neurodegeneration. In the following sections, we will explore common mechanisms and potential therapeutic avenues.

Spastin is recruited to microtubules by polyglutamylation, a specific type of posttranslational modification of microtubules [38]. Accordingly, we found that the polyglutamylation of microtubules precedes Aβ-induced microtubule loss. Tau appears to mediate the activation of spastin (=microtubule severing) via recruitment of TTLL6. Tau is necessary for an enrichement of TTLL6 into dendrites, where TTLL6 then catalyzes the ligation of polyGlu residues within the C-terminal tail of α-tubulin.

Polyglutamylation can also be induced by other members of the Tubulin-Tyrosin-Ligase-Like family, in particular TTLL1, 4, 5, 6, 7, 11, and 13 [31]]. Of these, mainly TTLL1, 4, 5 and 7 initiate polyglutamylation with different preferences for α-tubulin and β-tubulin. The main glutamylation sites are on Glu445 of α-tubulin (type A1A/B), and on Glu435 of β-tubulin (type B2A/B, Glu438 for B3-tubulin), but several other sites are possible due to the high glutamate content of the acidic C-terminal tail of tubulin [31]. TTLL4 or 7 catalyze the ligation of further Glu residues to microtubules, preferentially on β-tubulin, and TTLL5 on α-tubulin. This results in branching points and short polyGlu side chains that do not yet suffice for spastin mediated severing of microtubules. By contrast, only TTLL6 and 11 elongate polyGlu side chains on α-tubulin, in a manner which initiates spastin mediated severing [38].

It is currently not clear whether Tau-dependent mislocalization of TTLL6 in response to Aβ is the sole mediator of microtubule loss in dendrites, or whether this can be achieved by other TTLLs as well. Polyglutamylation of microtubules is reversible; the enzymes that are responsible for deglutamylation have recently been identified and belong to the class of cytosolic carboxypeptidases (CCP). Of these, CCP5 removes mainly branching points, while CCP1, 4, 6 reduce the glutamate chain length [55]. Thus, other polyglutamylases and deglutamylases could also be involved in Aβ and Tau mediated loss of microtubules.

The proposed mechanisms for the pathological changes in spastin activity in HSP and in our neuronal cell model of AD are flip sides of the same coin. In HSP and models thereof, spastin activity is decreased, whereas in the cell model of AD, spastin is aberrantly activated and induces breakdown of microtubules. Consistent with this, AD patients and several model systems of tauopathy show decreased microtubule density (see Table 1). Thus, inhibiting the microtubule severing activity of spastin could be beneficial in an AD setting. In HSP caused by SPAST gene mutations, spastin is unable to sever microtubules, but at least some forms of mutated spastin might still bind and cluster microtubules, resulting in overstable bundled microtubules (see Fig. 2c for schematics). In this context, it might be worthwhile mentioning that both spastinKO and TauKO mice display late and moderate motor deficits [13, 36, 41, 47, 70], possibly due to stalled microtubule-based transport. In neurons derived from spastinKO-mice or HSP-patients, axonal varicosities that contain organelles and microtubule fragments can be observed [19, 26, 70]. Similar varicosities were also observed in axons of TauKO mice when additional stressors (overexpression of mutAPP or traumatic brain injury) were present, but microtubule content was not investigated. Many different conditions may lead to late and moderate motor deficits, but this raises the question of possible shared mechanisms of spastin and Tau action on maintaining axonal health. In the following section we explore if and how inhibitors of spastin or upstream regulators of spastin could be beneficial for AD and SPAST-caused HSP.

The steps upstream of spastin recruitment in AD and HSP might be similar in spite of the different outcomes: In case of pathological severing as in the proposed model for AD, microtubules first need to be polyglutamylated (by TTLL6 and possibly also by TTLL11 and the priming ligases TTLL1/5) [38]. In case of aberrant microtubule bundling and impaired severing as proposed for HSP, polyglutamylation still might be necessary to guide the severing-deficient spastin to microtubules, but this question has not yet been addressed experimentally. Thus, targeting binding of spastin to microtubules, or microtubule polyglutamylation by TTLL1, 5, 6 or 11 could be a therapeutic option not just for AD and other tauopathies, but also for SPAST caused HSP.

Most mutations of spastin are present within the AAA-ATPase domain and lead to a loss of microtubule severing, therefore HSP caused by spastin mutation is considered a loss of function or haploinsufficiency phenotype [60]. But loss of function of the severing activity of spastin cannot explain important aspects of the disease and observations made in model systems of HSP. There is no correlation between spastin levels and the severity of neurodegeneration in disease and model systems [60, 76]. Knockout or truncation of spastin in mice only results in late and moderate motor deficits, which are a lot more subtle than those in human patients suffering from HSP [36, 70].

Also, HSP linked to spastin mutation affects mainly the corticospinal tracts without any involvement of the peripheral nervous system. In the central nervous system effects are most obvious in the spinal cord. Other parts of the central nervous system are not affected or only at later stages. There are, however, reports of dementia and cognitive decline associated with HSP, but these are very subtle, usually below clinical thresholds, and virtually absent when compared to intrafamilial controls [22, 68], but this is a matter of debate (see e.g. [46]). There is one report on a mutation of spastin (exon 10, R424G), in which authors also looked for other brain pathologies and identified Lewy bodies and tauopathy [73]. The tauopathy found corresponded to Braak stage 2, an AD neuropathology stage which is normally not associated with dementia, but this HSP patient did suffer from dementia [6, 73]. More human studies are needed to test the involvement of Tau in HSP.

Because a loss of function cannot well explain many of the phenotypes of HSP and model systems thereof, a pathological gain of function mechanism which also explains the spinal cord involvement has been proposed [62, 70]. The two major isoforms of spastin resulting from the two start codons show differential distribution and pathology involvement in model systems. M1 is only present in the adult spinal cord, while M87 is ubiquitous; M1 appears to be detrimental to axonal transport and MT dynamics, whereas M87 is not [64]. Also, in humans, significant levels of M1 were only detected in the adult spinal cord, not in the brain [63]. Only mutated M1, but not M87, decorates MTs and thereby could over-stabilize them, resulting in impaired (i.e. reduced) MT dynamics [62]. This mechanism has only been shown for some mutations, but as most of the mutations are present within the AAA-ATPase domain, spastin would retain its ability to bind to and bundle microtubules. Therefore, an inhibitor of M1 spastin activity or M1 binding to MTs could be of therapeutic value for at least some SPAST mutation based cases of HSP.

Spastin has diverse cellular function apart from MT-severing, many of which could result in a pathological gain of function if SPAST is mutated (for review see Baas and Solowska [61]). The M1 isoform specifically interacts with atlastin 1, a large integralmembrane GTPase, which mediates homotypic fusion of ER-tubules. Of note, mutations of atlastin 1 are the second most common cause of HSP [83], indicating that impaired ER-shaping along microtubules is of great importance in HSP. As most SPAST mutations do not affect the interaction motif of spastin for atlastin, impairment of ER-tubule shaping could also be a toxic gain of function mechanism in HSP independent of MT-severing. In a model system of mutated atlastin 1 (encoded by ATL1, also SPG3A), however, impairments could be prevented by the use of microtubule stability modulating drugs (vinblastine and taxol; [84]). This indicates that also in HSP cases that are not caused by mutation of spastin, modulation of microtubule stability could be of therapeutic value, but more studies are needed here to elucidate the molecular pathomechanisms.

Of note, increased spastin levels have been linked to increased malignancy and invasiveness of glioblastomas. Although the mechanisms are unclear, increased spastin expression resulted in the formation of protrusions and decreased cell proliferation of cancer cell lines. High-grade malignancy glioblastomas were linked to increased expression of spastin. Spastin could shift cancer populations from cancer cells showing high proliferation to cells that proliferate less, but show greater malignancy and invasiveness. This change in properties of cancer cells results in cancers more difficult to treat. Therefore spastin inhibitors might also be of therapeutic value for certain forms of brain cancer [16].

Physiological functions of spastin might be essential for cellular homeostasis, which would tamper hope of using spastin as a therapeutic target. Knocking down spastin during early stages of development results in impaired dendritic branching and outgrowth in cell culture [53]. However, there are no reports of developmental defects in human patients of HSP, indeed disease onset usually occurs after several asymptomatic decades of life. Loss of spastin function by knockout in mice only results in late and moderate motor deficits and no developmental deficits [36, 70].

In the context of axon regeneration, it has been reported that axotomized Drosophila sensory neurons are able to regenerate their axon from the axon stump or the dendrite only when spastin is present homozygously, but not when one allele is missing. Other microtubule severing enzymes (e.g. kat-60 (the Drosophila homologue of human katanin-p60), kat-60 L and fidgetin) did not have an effect on axon regeneration in case of heterozygous loss of function mutants. Other microtubule related functions, such as axonal outgrowth during development or microtubule polarity in axons and dendrites were unaltered in the same study [66]. Thus, at least in the context of axon regeneration, lowering spastin activity would not be beneficial.

Taken together, these notions indicate that inhibiting spastin might not harm overall neuronal function and thus could be beneficial for patients. If the gain of function mechanism proves valid for HSP, inhibiting specifically the microtubule severing function of spastin might not be of therapeutic value for HSP patients, as this would lead to even more MT bundling and overstabilization. For the treatment of HSP a microtubule binding inhibitor would be preferable, while for AD either an inhibitor of microtubule severing or microtubule binding inhibitor could be of therapeutic value. It remains unclear, however, whether microtubule bundling by mutated spastin is an in vitro artefact as this has not been demonstrated in vivo. Inhibiting mutant spastin could be beneficial over absence of spastin, but this has not been addressed experimentally.

Apart from polyglutamylation of microtubules by TTLL6 and TTLL11 (see above), little is known about the regulation of spastin activity in physiological and pathological settings. PAK3 is one of the six known p21-activated kinases (PAKs), which is part of the group I PAKs (PAK1,2,3). PAKs can be activated by multiple signaling molecules, particularly by Rho family small GTPases. In a neuronal context PAKs regulate neurite outgrowth, spine morphology and synaptic plasticity. Mutations in the PAK3 gene were linked to X-linked mental retardation (for review on physiology and pathology of PAKs see [45]). Interestingly, overexpression of PAK3 was found to enhance the Drosophila eye degeneration phenotype induced by overexpression of spastin. Conversely, loss of PAK3 function prevented spastin mediated defects in synaptic function. This indicates a tight genetic and functional interaction between spastin and PAK3 [50]. However, pharmacological manipulation of PAK3 with effects on spastin has not yet been demonstrated.

The Notch signaling agonist Jagged1 was found to induce increased microtubule stability, similar to low doses of the microtubule stabilizer Taxol. Jagged1 downregulates the expression of spastin, which results in higher stability of microtubules [20]. Conversely, exposure of primary neurons to Aβ induced the expression of the Notch antagonist Numb, which could be counteracted by the pro-survival Notch pathway induced by the endocannabinoids anandamide and 2-arachidonoglycerol. As a Notch signaling modulator, anandamide also shifts gamma-secretase processing from APP to Notch1, which would also be beneficial in an AD-setting ([69], see review for Notch signaling and its influence on microtubule stabilization and synaptic plasticity [4]).

Transcription of spastin was shown to be upregulated by the transcription factors NRF1 and SOX11, whereas miR-96 and miR-182 downregulate both mRNA stability and protein levels of spastin [27].

For AD, tauopathy and HSP models, both microtubule stabilizing and destabilizing drugs have been investigated and reported to show beneficial effects (see e.g. [14, 18, 19], and Table 1 for details and complete list). It is surprising that antagonizing drugs can lead to a similar end result, i.e. rescue from SPAST caused defects. The underlying mechanisms are unclear. Microtubule destabilizing agents used (nocodazole, noscapine, vinblastine) might compensate for decreased severing of mutant spastin by increasing the amount of dynamic labile microtubules, and by increasing microtubule mass by minus end detachment of microtubules from their organizing centers [75]. Microtubule stabilizing drugs (taxol, epothilone D) could rely on activation of other microtubule severing enzymes to compensate for decreased severing by mutant spastin. For instance, increased microtubule stabilization leads to increased acetylation of microtubules, which in turn leads to increased katanin based severing [67]. Microtubule-stabilizing drugs such as taxol and epothilone D are used in cancer therapy, as they overstabilize microtubules and thus induce mitotic arrest and apoptosis of dividing cells. At lower concentrations these drugs marginally stabilize microtubules just enough to counteract microtubule destabilization in models of AD, tauopathy and surprisingly also HSP (Table 1 and for review [8]). However, taxol is not blood–brain barrier penetrant and causes peripheral neuropathy, and epothilone D inhibits the P-glycoprotein (Pgp)-transporter (a widely distributed toxin/drug efflux ABC-transporter), both of which could be responsible for undesired neurotoxicity and/or drug-drug interaction. Other promising, non-naturally occurring microtubule-stabilizing drugs are being developed and tested [44].

Using microtubule stabilizing drugs for the treatment of AD and HSP has also some caveats: Common side effects of taxol (when used at elevated concentrations in cancer therapy) include peripheral neuropathy, likely due to impairment of microtubule based transport in the long peripheral neurons (for review see [58]). As dementia patients would have to be treated for decades, even the use of lower concentrations of microtubule stabilizing drugs might result in similar problems.

A clinical trial for epothilone D for the treatment of patients with mild AD was discontinued after a phase 1 trial in 2013. Also, one clinical trial aiming at microtubule stabilization with Davunetide (alias Neuroactive Peptide NAP, an eight amino acid neuropeptide) for the treatment of neurodegeneration (such as the tauopathy Progressive Supranuclear Palsy, PSP) was negative in all endpoints, which prompted the halt of other studies with the same compound for other applications (biomarkers) and dementia (Frontotemporal lobar degeneration (FTLD), Corticobasal Degeneration (CBD); for more information on ongoing and halted clinical trials on AD see www.​alzforum.​org/​clinical-trial-registries). Another microtubule stabilizing drug, TPI287, a synthetic taxane derivative that crosses the blood brain barrier, is used in a clinical phase I safety trial for several tauopathies (AD, CBD, PSP, and 4 Repeat Tauopathy (4RT)) as well as different brain cancer conditions (for more information on clinical trials of TPI287 see www.​clinicaltrials.​gov).

However, indirect microtubule-stabilization via inhibition of MT-severing by TTLLs and spastin (or other microtubule severing proteins) might be more feasible as these enzymes are present in far lower concentration than tubulin. Very recently, the microtubule severing enzyme fidgetin was shown to trim the growing ends of dynamic microtubules [42]. This implies that after spastin severing, microtubule fragments might be unable to grow due to fidgetin action (see Table 2 for complete list of proteins possibly involved in AD and HSP related microtubule disruption. Considering (i) the past failures of clinical trials on AD with microtubule stabilizing drugs, (ii) the possible gain of function mechanisms in HSP, and (iii) the possible detrimental action of spastin and fidgetin, it becomes clear that we need a better understanding of microtubule dynamics in neurodegeneration. A molecular toolbox of small molecule inhibitors of microtubule severing enzymes would serve the further investigation of pathomechanisms in AD, related tauopathies and HSP, and facilitate the development of new MT-based therapeutics.