Axodendritic sorting and pathological missorting of Tau are isoform-specific and determined by axon initial segment architecture

Subcellular mislocalization of the microtubule-associated protein Tau is a hallmark of Alzheimer disease (AD) and other tauopathies. Six Tau isoforms, differentiated by the presence or absence of a second repeat or of N-terminal inserts, exist in the human CNS, but their physiological and pathological differences have long remained elusive. Here, we investigated the properties and distributions of human and rodent Tau isoforms in primary forebrain rodent neurons. We found that the Tau diffusion barrier (TDB), located within the axon initial segment (AIS), controls retrograde (axon-to-soma) and anterograde (soma-to-axon) traffic of Tau. Tau isoforms without the N-terminal inserts were sorted efficiently into the axon. However, the longest isoform (2N4R-Tau) was partially retained in cell bodies and dendrites, where it accelerated spine and dendrite growth. The TDB (located within the AIS) was impaired when AIS components (ankyrin G, EB1) were knocked down or when glycogen synthase kinase-3β (GSK3β; an AD-associated kinase tethered to the AIS) was overexpressed. Using superresolution nanoscopy and live-cell imaging, we observed that microtubules within the AIS appeared highly dynamic, a feature essential for the TDB. Pathomechanistically, amyloid-β insult caused cofilin activation and F-actin remodeling and decreased microtubule dynamics in the AIS. Concomitantly with these amyloid-β–induced disruptions, the AIS/TDB sorting function failed, causing AD-like Tau missorting. In summary, we provide evidence that the human and rodent Tau isoforms differ in axodendritic sorting and amyloid-β–induced missorting and that the axodendritic distribution of Tau depends on AIS integrity.

The mechanisms proposed for the polarized distribution of Tau in adult neurons include (i) preferential binding to MTs in the axon, (ii) preferential degradation in the somatodendritic compartment (20), (iii) axonal transport of Tau mRNA (21), (iv) preferred protein translation in axons (22), and (v) preferred retention by an axonal barrier (23). Because most protein synthesis takes place in the cell body, the bulk of Tau must transit from the cell body to the axon through the axon initial segment (AIS), which roughly coincides with a retrograde Tau barrier. The AIS is enriched with cytoskeletal scaffolding proteins ankyrin G and ␤IV spectrin, which together cluster voltagegated sodium and potassium channels (Na v and K v , respectively) and link them to the actin cytoskeleton (24). Functionally, the AIS plays a role in action potential initiation and modulation as well as neuronal polarity (25). Several studies describe the barrier function of the AIS as being based on lipids, membranes, F-actin, or MTs, depending on the components studied (23, 26 -30). The internal structure of the AIS and its relationship to barrier functions has remained elusive since its discovery (31), but recent studies have revealed a dense, fibrillar/globular submembranous coat containing the classical AIS proteins (e.g. ankyrin G and ␤IV spectrin), which covers MT bundles (32).
The AIS and neuronal polarity are established early in development in vivo and can be mirrored in vitro. With the emergence of the axon, Tau becomes gradually enriched in the axon concurrent with AIS establishment (33,34). Similarly, Tau isoforms change from the fetal to the adult distribution, due to alternative splicing ( Table 1). The initially high Tau phosphorylation decreases with neuronal maturation, but in neurodegeneration Tau becomes "hyperphosphorylated," concomitant with missorting and aggregation. Hyperphosphorlyation of Tau in turn acts on the AIS localization within the axon, thereby decreasing neuronal excitability (35). A major kinase acting on Tau is GSK3␤, which phosphorylates many of the Ser-Pro or Thr-Pro motifs and is implicated in neurodegeneration (36). GSK3␤ is important for AIS function because it is an integral part of the AIS architecture, where it induces local tethering of ␤-catenin and ankyrin G (37).
We recently discovered a barrier that blocks the retrograde passage of Tau from the axon into the soma and dendrites (termed the Tau diffusion barrier (TDB), located at the AIS) (23). Here, we investigate the relationship between the TDB and the AIS and their effects on different isoforms of Tau. We found that the TDB has different retention efficiencies for different isoforms of Tau. The TDB functions not only in a retrograde fashion but also anterogradely; it partially retains the largest isoform of Tau in the dendrites, where it enhances dendritic growth and spine formation. Knockdown of structural components of the AIS decreases the Tau barrier efficiency. The Tau and AIS kinase GSK3␤ induces missorting of endogenous and exogenous Tau, but independently of Tau phosphorylation. STED nanoscopy and live-cell imaging show that MTs are major constituents of the TDB in the AIS, with unusual dynamics. Exposure of neurons to amyloid-␤ oligomers results in impairment of the F-actin-based filter in the AIS and decreased MT dynamics, resulting in failure of proper axodendritic Tau sorting.

The Tau diffusion barrier is isoform-selective
Previous studies have shown that there is a retrograde diffusion barrier for Tau at the AIS that retards the back-diffusion of Tau from the axon once it has entered it (23). These results were obtained using a Tau fusion protein with the photoconvertible GFP derivative Dendra2 (38), linked to the N terminus of fulllength Tau (Tau D2 ). This allowed us to observe the movement of axonal Tau after local UV irradiation. Phosphorylated forms of Tau D2 were able to penetrate the barrier to some extent. In the earlier study, we focused on the Tau barrier with regard to full-length Tau (2N4R) only and its modification by phosphorylation. However, Tau contains several domains that may also modulate its functions.
In the present study, we tested all six isoforms of Tau (see Table 1 for comparisons and nomenclature of the isoforms and Fig. 1A for a diagram). Transfected species of Tau D2 were distributed throughout the cell (Fig. 1B, 1 and 5, green emission of unconverted Tau D2 ). Then ϳ15 m of the proximal axon (ϳ50 m away from the cell body) were irradiated with UV light, causing the Dendra2 to convert from green to red fluorescence. This was then imaged for up to 1 h, and signals were measured in the axon or the cell body (indicated by red rectangles and red triangles, respectively) (Fig. 1B).
The reference values are represented by Dendra2 alone (barrier fully open, unrestricted diffusion from axon into cell body, Fig. 1C, fluorescence (F) ϳ100 in arbitrary units (a.u., see "Materials and methods")) and the 8-repeat construct of Tau (barrier fully closed, F ϳ2.5 (background), almost no detectable diffusion into cell body). For the 4R isoforms, the barrier is relatively tight (F ϳ5-10, teal bars), and for two of the 3R isoforms, there is intermediate leakiness (F ϳ25, blue bars). For 2N4R-Tau mutants, the leakiness is intermediate to high (F ϳ25-70, red bars). This is particularly apparent for the 4KXGE  (39). Tau carrying mutations derived from FTLD or PSP also shows decreased MT affinity (40). Because the TDB correlates with Tau-MT interaction, we tested the TDB for (i) 2N4R-Tau mutated at sites causing FTLD (pro-aggregant Tau with deletion of Lys-280), (ii) a control version of this mutation that cannot aggregate due to the insertion of two prolines (antiaggregant Tau ⌬K280PP), and (iii) the mutant A152T of 2N4R-Tau implicated in PSP (41). All of these mutations increased the leakiness of the TDB for retrograde diffusion back into the cell soma (Fig. 1C, red bars). The example of A152T-Tau is remark- Figure 1. The second repeat and inserts synergistically prevent Tau from retrograde passage through the TDB in the AIS, whereas mutations cause retrograde leakage into the soma. A, bar diagrams of the studied isoforms of Tau. Red, alternatively spliced N-terminal inserts (N1 and N2); green, repeat domain; yellow, repeat 2 (R2). Mutated 2N4R-Tau species are displayed in the bottom half, and mutations are represented by red vertical lines. B, photoconversion of primary rat cortical neurons transfected with control 2N4R-Tau D2 (top) or 2N4R-Tau D2 -KXGE mutant (bottom). 1 and 5, reference image of unconverted protein before photoconversion to identify soma and axon in the green channel. 2 and 6, before photoconversion, there is no fluorescence in the red channel. 3 and 4a, after photoconversion in the region of interest (dotted box, red), cells transfected with 2N4R-Tau D2 (top) show movement of Tau D2 beyond the region of interest but not into the soma (red triangle) and dendrites, indicating a retrograde barrier in the initial part of the axon. 7 and 8a, in contrast, cells transfected with 2N4R-Tau D2 -KXGE mutant show penetration of photoconverted protein into the somatodendritic compartment. Pictograms in 4b and 8b are a graphical representation of 2N4R-Tau D2 unable to cross the diffusion barrier within the AIS (4b, cell body remains dark) and 4KXGE-Tau D2 , which crosses the AIS and reaches the cell soma (8b, cell body lights up). C, level of Tau leaking through the barrier into the soma after photoconversion shown for several Tau isoforms (blue) and Tau mutations (red). Isoforms lacking both the second repeat (R2) and the second insert (N2) penetrate into the somatodendritic compartment. Pseudophosphorylated Tau (KXGE) or FTLD-Tau related mutations (A152T and ⌬K280) also cause incomplete axonal retention. Note that the tightness of the barrier does not depend on the propensity of Tau for ␤-structure that controls aggregation (compare ⌬K280 and ⌬K280PP). D, decay of a selection of photoconverted Dendra2 and Tau D2 species in the photoconverted area of the axon. Axonal diffusion rates correlate with the MT binding potential of the species involved in some cases (e.g. Dendra2 3 low and 8-repeat Tau  able, because a single point mutation outside the conventional MT-binding region (amino acids ϳ200 -370) can enhance the mobility of Tau in neurons. One exception to the above classification is 2N3R-Tau, where the barrier is almost tight (as in 2N4R), indicating that the second N-terminal insert (amino acids 74 -102, exon 3) contributes strongly to the functional retention by the Tau barrier, although it lies outside the MTbinding domain. Some of these variations correlate with the affinity of Tau for MTs, but not all, so that we conclude that MT affinity alone is not sufficient to explain Tau retention in the axon.
The diffusion rates of the Tau species can be assessed by the fluorescence increase in the cell body (as shown above), but also by the decay in the photoconverted area, which reflects the diffusion of Tau both toward the cell body (retrogradely) and away (anterogradely toward the axon terminal) (Fig. 1, D and E).
The reference values are set by Dendra2 alone (rapid decay because of fast diffusion, short half-life of ϳ2 min) and 8R-Tau (very slow diffusion because of tight MT binding, long half-life of ϳ46 min). Overall, tight MT binding correlates with slow diffusion within the axon (e.g. half-times Dendra2 Ͻ 4KXGE-Tau Ͻ 2N4R-Tau Ͻ 8-repeat Tau).

Exogenous Tau is efficiently sorted into axons of wild-type and TauKO neurons except the largest isoform, 2N4R
In the experiments described so far, we overexpressed the isoforms of Tau D2 at high levels (ϳ20-fold of endogenous Tau) to be able to investigate the barrier. In this setting, using primary neurons at an early stage of polarization (8 -9 days in vitro (DIV)), Tau is initially present in all compartments. Subsequently, we photoconverted a subset of molecules in the axon to observe their propagation. As an alternative approach, to confirm the functioning of the TDB at low Tau levels, we used a protocol that limits Tau expression to 1-3-fold of endogenous Tau (estimated at ϳ1 M (42)), which required antibody-mediated amplification of the transfected constructs. We also waited longer (4 -6 days) to allow the transfected versions of Tau to reach their destiny compartment. We also used more mature well-polarized neurons (13 DIV) with longer processes from TauKO and wild-type mice. We found that the shorter human 4R isoforms (0N4R and 1N4R) and the 3R isoforms were highly enriched in axons, whereas the longest isoform (2N4R) was in part excluded from axons ( Fig. 2 (A and B), supplemental Figs. S1 and S2 and supplemental Table S1).
To quantify effects, we used an enrichment factor (e.f.), which is calculated by the normalized ratio of axonal presence versus dendritic presence of transfected Tau (see "Materials and methods"). Quantitatively, axonal e.f. was ϳ75, 85, and 95 for 2N4R, 1N4R, and 0N4R, respectively. The 3R versions of Tau without or two inserts (i.e. 0N3R and 2N3R) showed axonal enrichment but to different extents (e.f. of ϳ90 and 95), whereas the 1N3R isoform was, with an e.f. of 70, less efficiently sorted (supplemental Table S1 and Fig. 2C). With the exception of 2N4R-Tau D2 , Tau isoforms and mutants that were able to partially penetrate the TDB retrogradely were also axonally enriched in neurons transfected with a specific isoform. For example, 4KXGE-2N4R-Tau D2 showed retrograde penetration of the TDB (Fig. 1, B and C) (23) and also axonal enrichment (i.e. anterograde penetration of the TDB), in contrast to other mutations (⌬K280, ⌬K280PP, A152T, and all 2N4R-Tau D2 ), which showed only slight retrograde penetration of the TDB and correspondingly only limited axonal sorting ( Fig. 2 (A-C), supplemental Figs. S1 and S2, and supplemental Table S1). Because 2N4R-Tau has a much higher affinity with MTs than the tested mutations, this confirms that strong MT binding can only partly explain axonal sorting of Tau. Surprisingly, we found no differences for TauKO (KO of all isoforms (43)) and wild-type cells (for full comparisons, see supplemental Figs. S1 and S2), indicating that even when Tau is absent during development and after establishment of polarity, Tau can be as efficiently sorted as in wild-type conditions.

Dendritic Tau accelerates spine formation and dendritic elongation
In rodents, 3R-Tau is mainly expressed during the first week of neuronal development and decreases during the second week, whereas 4R-Tau is mainly expressed after 2 weeks. Because mice overexpressing the 2N4R isoform of Tau show enhanced LTP early in life (44) and throughout life (45), we hypothesized that Tau has a dendritic function in synapse formation. Because dendritic synapses sit on spines, we chose to express the Tau isoforms in young neurons, before spines are formed (7 DIV). We found that some isoforms of Tau D2 induce accelerated spine formation, but with strikingly different phenotypes (Fig. 2D). For example, 0N3R induced elongated spine necks and mature spine heads, unlike filopodia (Fig. 2D, 2). 0N4R had almost no effect, whereas 2N4R-Tau induced mature spine heads ( Fig. 2D and supplemental Table S1). Expression of the other native isoforms of Tau showed intermediate effects on spine development, but all of them induced an increase in spine number and either an increase in spine maturity or spine length. In detail, 1N4R-Tau showed intermediate effect on spine maturation, number, and length, positioning it right between 0N4R and 2N4R-Tau (supplemental Table S1). 2N3R-Tau showed intermediate and 1N3R-Tau showed slight increase in all parameters investigated (spine number, length, and maturity), positioning them between 0N4R and 2N4R. The mutations tested (⌬K280, ⌬K280PP, A152T, KXGE, and KXGA, expressed on the basis of 2N4R-Tau D2 ) resulted in more spines of variable phenotype, indicating that these mutations might affect the spine maturation and also spine/filopodia dynamics ( Fig. 2D and supplemental Table S1). Because overexpression of Tau results in enhanced spine maturation, we tested whether there was an effect of acute knockdown of Tau via shRNA on spine maturation. When applied during the period of spine maturation (14 -21 DIV, which coincides with the transition from 3R-to 4R-Tau), knockdown of all Tau isoforms via shRNA results in impaired spine formation (consistent with Ref. 46) (Fig. 2D, 7 and 8). However, we found no differences between wild-type and TauKO neurons in terms of aberrant spine formation after transfection of individual Tau isoforms. This indicates that forced expression of different isoforms (thereby also neglecting the physiological mixture of isoforms) results in aberrant spine development independent of the presence or absence of endogenous Tau. Not surprisingly, expression of Tau induced elongation of dendrites, but to dif-ferent extents ( Fig. 2E; graphical summary of Fig. 2, D and E, in Fig. 2F). Elongation of dendrites required the presence of N-terminal inserts and the presence of intact KXGS motifs but were largely unaltered in the case of the tested mutations (⌬K280, ⌬K280PP, and A152T; all expressed on the basis of 2N4R-Tau D2 ). Thus, elongation of dendrites correlates roughly with increased dendritic presence for some of the native Tau isoforms (high dendritic presence for N-containing isoforms results in elongated dendrites, whereas 0N-Tau isoforms do not elongate dendrites), except for the Tau mutations tested and the 2N3R isoform (intermediate dendritic presence). This indicates that the presence of the two N-terminal inserts is sufficient to induce dendrite elongation.

Structural components of the axon initial segment are required for the tightness of the anterograde and retrograde Tau diffusion barrier
Because the position of the TDB coincides roughly with that of the AIS, we determined which of the major known components of the AIS are essential for TDB function. Interestingly, knockdown of any of the AIS components by shRNA, such as neuronal cell adhesion molecules (NrCAM), Na v , or scaffold proteins (␤IV spectrin or AnkG) (47), led to slight leakage of 2N4R-Tau D2 from the axon into the soma (4 -6-fold above background levels (supplemental Fig. S3, A and B)). This leakage is similar to the leakiness of one of the shorter isoforms of Tau (e.g. 0N3R; Fig. 1C). In addition, knockdown of the MT end-binding proteins EB1 and EB3, which are enriched at the AIS (48), led to similar enhanced retrograde leakage of Tau D2 (supplemental Fig. S3, A and B). Thus, each of these AIS components contribute to the maintenance of the retrograde TDB.
We next tested whether knockdown of the different structural components of the AIS would also lead to enhanced anterograde movement of Tau. We thus expressed the longest isoform of Tau (2N4R-Tau D2 ), which shows little anterograde axonal propagation compared with the other isoforms ( Fig. 2 (A and B) and supplemental Table S1), in combination with knockdown of the AIS components (i.e. NrCAM, NaV, ␤IV spectrin, AnkG, EB1, and EB3; supplemental  Fig. S3D). Interestingly, none of the knockdowns resulted in impaired axonal sorting of endogenous rat Tau (supplemental Fig. S4). Together, these results indicate that the AIS structural components are important for both the regulation of anterograde and retrograde passage of exogenous Tau but do not hinder anterograde passage of endogenous Tau.
The protein kinase GSK3␤ is a structural component tethered to the AIS, affects the activity of AIS-based sodium channels (37), and is crucial for cell polarity and pathological development of tauopathies, including Alzheimer's (49,50). We thus tested the effect of GSK3␤ on Tau distribution. Overexpression of GSK3␤ induces pronounced missorting even of endogenous Tau, in contrast to other structural components of the AIS mentioned above. Missorted Tau is phosphorylated at the KXGS motifs (supplemental Fig. S5A), despite the fact that GSK3␤ cannot phosphorylate Tau directly at these motifs. This is consistent with the hypothesis that Tau becomes phosphorylated by other kinases (e.g. MARK), after it has transited into the somatodendritic compartment (51). Missorting of Tau is recognized as a pathological sign and can be induced by exposure to A␤ (e.g. see supplemental Fig. S5A (1), arrowheads) and other forms of molecular stress (51). However, exposing GSK3␤-transfectedneuronstoA␤didnotfurtherincreasephosphorylation or missorting of Tau (supplemental Fig. S5A). This indicates that either A␤ exposure and GSK3␤ activity activate the same pathway, or GSK3␤ activity changes the cell in a different fashion so that A␤ cannot exert its toxicity anymore. We also tested the state of GSK3␤ activity after exposure of primary neurons to A␤, using phosphorylation at Ser-9 as a readout. Time slicing up to 24 h showed that there is a brief 30-min period of phosphorylation of GSK3␤ (inactivation), which then returns to baseline within 3 h (supplemental Fig. S5B). This indicates that there is no activation of GSK3␤ after exposure to A␤.
We previously found that A␤-induced Tau missorting requires loss of dendritic MTs (12). Here, we find that missorting of Tau is induced by overexpression of GSK3␤ (wild-type and constitutively active GSK3␤-S9A), but not by inactive GSK3␤-S9E, and thereby induced missorting of Tau does not lead to MT breakdown (supplemental Fig. S6A). Hence, MT loss cannot account for Tau missorting induced by GSK3␤. We thus tested whether missorting of Tau induced by GSK3␤ is due to a change of Tau or a change of the TDB. Cotransfection of 2N4R-Tau D2 and GSK3␤ results in enhanced retrograde propagation of axonal Tau D2 into the cell body, indicative of a dysfunctional barrier, whereas inactive GSK3␤ (S9E mutant) has no effect (Fig. 3, A and B). GSK3␤ can phosphorylate Tau at many sites, possibly changing the diffusion capability of Tau or its interaction with the TDB. To determine the effect of GSK3␤ on the TDB independently of Tau phosphorylation, we cotransfected GSK3␤ with non-phosphorylatable Tau D2 (17AP-Tau D2 ). B, wild-type neurons expressing endogenous Tau in addition to transfected exogenous Tau D2 (stained with CP27). Exogenous and endogenous Tau are stained with a pan-Tau antibody (K9JA), and dendrites are stained with an antibody against MAP2. Columns 1 and 2, 0N4R-Tau is enriched in the axon. As a result, axonal Tau concentration is higher than in neighboring untransfected axons where only endogenous Tau is present, but it can barely be detected in dendrites with a pan-Tau antibody. Column 2 (bottom) shows magnified images from the boxed areas, and axon is indicated by arrowheads. Columns 3 and 4, 2N4R-Tau is not enriched in axons. Column 5, 2N4R-Tau pseudophosphorylated at the KXGS motifs (KXGS motifs mutated to KXGE) shows strong axonal enrichment. C1, quantification of axonal enrichment of different Tau D2 isoforms and 2N4R-Tau D2 mutations reveals lower axonal presence of the longest isoform of Tau (2N4R) compared with the other isoforms. The exception is the human-specific isoform 1N3R, not present in rodents. The mutations 4KXGE, A152T, and ⌬K280PP of 2N4R-Tau D2 increase the axonal sorting, whereas the mutation ⌬K280 has no effect. C2, graphical sketch of the different distributions of the representative isoforms 0N4R (strong axonal enrichment) and 2N4R (weak axonal enrichment). Left panels, Tau distribution depicted in green. Right panels, Tau distribution in green presented relative to a volume marker with unbiased distribution in red, resulting in a ratiometric image. D and E, wild-type neurons with tdTomato as an unbiased volume/morphology marker. D, Tau D2 transfection induces spine formation and maturation in young (10 DIV; 1-6) neurons, and Tau knockdown abolishes spine maturation in old (21 DIV; 7 and 8) neurons. 1, neurons only transfected with a control vector (Dendra2) show no spine formation and very few filopodia, representative of neurons on the brink of (but before) spine development. 2, transfection of 0N3R-Tau results in elongated filopodia-like protrusions up to 10 m. 3, by contrast, 0N4R-Tau has no effect on dendritic spine/filopodia formation. 4, 2N4R-Tau-transfected cells show dendritic processes with mushroom heads (arrowheads), indicative of mature spines. 5 and 6, mutations of the KXGS motifs in the repeat domain to KXGE (5) or KXGA (6) of 2N4R-Tau prevent formation of protrusions and acceleration of spine maturation. For a comprehensive list of effects of all isoforms and several mutations, see supplemental Table S1. 7, aged control neurons display normal mature spines. 8, aged neurons transfected with shRNA against Tau display impaired spines. E1, Tau transfection induces dendritic outgrowth. Left, neuron transfected with a control vector (Dendra2). Right, neuron transfected with the 2N4R-Tau. The longest dendrites are indicated by red dashes. E2, quantification of the longest dendrite of cells transfected as indicated. F, graphical sketch of the dendritic effects of the representative Tau isoforms 0N3R (filopodia spines), 0N4R (no dendritic effects), and 2N4R (elongated dendrites and premature development of spines). *, p Ͻ 0.05. Error bars, S.E.
17AP-Tau D2 contains 17 proline-directed serines (putative GSK3␤ phosphorylation sites) mutated to alanine (17AP-Tau D2 ). 17AP-Tau D2 still results in enhanced retrograde propagation to the soma when cotransfected with GSK3␤, but not when expressed alone or co-expressed with a control (CFP; Fig.  3, A and B). Diffusion rates within axons are the same for 2N4R-Tau D2 irrespective of GSK3␤ activity (Fig. 3C). However, 17AP-Tau D2 shows reduced diffusion by 50% already after 10 min within the axon (Fig. 3C, purple). This points toward an under-estimation of the TDB dysfunction because there is less Tau available to cross the TDB. Taken together, this indicates that GSK3␤ influences Tau sorting via modifying either the TDB or some other factors maintaining the polarized distribution of Tau (e.g. inhibiting the Tau kinase MARK), but not by phosphorylating Tau.

AIS structure contributes to Tau diffusion barrier
Sequence motifs in the 5Ј-or 3Ј-UTRs of endogenous Tau mRNA have been implicated in the polarized trafficking and translation of Tau (21,22). Our exogenously expressed Tau lacks these motifs, yet the protein is also sorted in a polarized fashion. This suggests that the mRNA-based mechanisms cannot account alone for the sorting of Tau protein. Because known structural elements of the AIS are necessary to maintain the tightness of the Tau diffusion barrier (supplemental Fig.  S3), we hypothesized that there may be additional structural elements of the TDB not integrated into the AIS structure that determine the isoform-specific blockage of Tau. We previously showed that the efficiency of the TDB is decreased when MT dynamics are disturbed by the MT destabilizer nocodazole (i.e. loss of MTs made the barrier leaky) (23).
To examine MT-related properties, the beginning of the AIS was defined as the point where an AIS marker (NrCam) reaches 50% of its level within the AIS, typically 17 m away from the edge of the cell body. This coincides roughly with the point where MAP2 (a somatodendritic marker) decreases to ϳ50% of its value in the axon hillock (Fig. 4, A and B). Acetylated tubulin (a marker for stable MTs) starts low and increases slightly within the AIS, and polyglutamylated tubulin (stable MTs and a potential target for MT-severing enzymes) remains low throughout the AIS (Fig. 4 (A and C) and supplemental Fig. S7). Tyrosinated tubulin (a marker for dynamic MTs) continuously decreases from the beginning of the AIS onward to the axon, consistent with mainly dynamic MTs in the cell body and mainly stable MTs in the axon ( Fig. 4C; for volume-normalized calculations and pictograms, see supplemental Fig. S7). We next used fixation protocols that either facilitate the visualization of polymerized tubulin (MTs) or free tubulin. This revealed high levels of tubulin in the AIS compared with the axon or the cell soma (ϳ2.5-fold higher) but only a low fraction of stainable MTs (Fig. 4D). The ratio of free tubulin versus MTs is ϳ2:1 in the AIS and ϳ1.5:1 in the cell body, but only 1:1 in the axon (Fig.  4D). This indicates a predominance of unstable MTs or depolymerized tubulin in the AIS.
We next attempted to visualize MTs in the AIS with STED nanoscopy. Within the AIS, no MTs are observed when using monoclonal antibodies against unmodified ␣-tubulin (Fig. 4E) or acetylated tubulin, and very few MTs are visible with an antibody against tyrosinated tubulin (Fig. 4F) with normal fixation procedures (i.e. no application of MT-modifying substances or permeabilization prior to fixation). This indicates that MTs within the AIS are too dynamic to be visualized with standard staining techniques. Only coapplication of the molecular densifier PEG, which creates high osmotic pressure allowing the fixatives to fix MT before they fall apart, together with the MT stabilizer taxol prior to fixation allowed us to observe  (9 -12), mutating the GSK3␤ target sites of Tau to alanine (17 SP/TP motifs mutated to AP, AP17-Tau D2 ) results in a leaky barrier so that AP17-Tau D2 still shows retrograde propagation back into the cell body (quantified in B, purple bar). B, quantification of A as indicated. C, unable to be phosphorylated at its SP/TP motifs, AP17-Tau is revealed to have a very slow diffusion rate (purple trace). Penetration of the TDB can be seen, nonetheless, as was shown in panels 9 -12 of A and the purple bar in B. Error bars, S.E.
MTs within the AIS, emphasizing the transient nature of these MTs (Fig. 4G).
Next, we examined MT dynamics directly by imaging of transfected EB3-GFP within the AIS. Due to the high density of dynamic MTs in the AIS (compared with the low density of dynamic MTs in axons, which are easier to image), a specialized imaging procedure was required to visualize EB3 comets within the dendrites and the AIS, using high-frame rate imaging combined with noise reduction (see "Materials and methods"). This revealed that within the AIS, there is a high concen- tration of transfected EB3 and a high density of EB3 comets, compared with axons and dendrites (Fig. 5, A and B). This again illustrates that MTs are highly dynamic and short-lived within the AIS. All comets observed within the AIS moved anterogradely, indicating a homogeneous polarity of MTs with plus ends pointing into the axon (in contrast to dendrites that con-tain MTs with mixed orientations (52)). The high concentration of EB3 comets hints at a hub or organization center for MTs at the AIS, probably affecting MT-based transport. Costaining of EB3 in fixed cells with MAP2 and mitochondria illustrates that the EB3 hub lies within the AIS (Fig. 5, C and D).

Exposure of neurons to A␤ results in activation of cofilin, loss of F-actin, and decreased MT growth within the AIS
Missorting of Tau induced by A␤ oligomers is not caused by breakdown of the retrograde barrier, but by failure of newly synthesized Tau to sort into the axon (51). This could be due to either a tightening of the anterograde barrier or loss of sortingenabling structures. We thus tested whether A␤ exposure connects pathological missorting of Tau with defective elements of the AIS/TDB.
We first tested whether A␤ binds to the AIS. A␤ localizes to dendrites (primarily dendritic spines), but less to axons (90% less binding, judged by immunofluorescence). In contrast, A␤ localizes fairly well to the AIS (25% of dendritic localization; supplemental Fig. S8A). As a result of A␤ exposure, ankyrin G throughout the AIS and F-actin within the AIS decrease, indicative of a direct effect of A␤ on the AIS (supplemental Fig. S8B).
An F-actin meshwork represents a cytoplasmic barrier, which hinders non-axonal proteins and larger beads to enter the axon (27,28,53). We previously found that in dendritic spines, A␤ exposure leads to a rapid loss of F-actin (51). Testing F-actin content in the AIS both by live-cell imaging and phalloidin-based stainings reveals that F-actin is lost and remodeled rapidly after exposure to A␤ (i.e. within 15 min), but not, for example, in the dendritic shaft, where F-actin accumulates (Fig.  6, A and B). As cause for this actin remodeling, we found that cofilin,anF-actin-severingenzyme,becomesactivated(dephosphorylated) in the AIS with a time course comparable with that of decreasing F-actin (Fig. 6). Thus, activation of cofilin could result in loss or remodeling of F-actin. Live imaging with an F-actin-sensitive probe (f-tractin) confirmed that the area covered by F-actin within the AIS increased, but only after 1.5 h of treatment with A␤ (Fig. 7A), whereas in control conditions, there was no change. This timing is in agreement with the increase in missorting of Tau (51) and the decrease of cofilin activity after initial activation by A␤ (Fig. 6). We next tested whether loss of F-actin alone would lead to missorting of Tau, due to simple loss of its cytoplasmic filter function. Treatment of primary neurons with various concentrations (0.5-2.5 M) of the F-actin-destabilizing drug latrunculin A for various periods (5 min to 24 h) in no case lead to missorting of Tau, despite Ͼ90% reduction of F-actin staining upon treatment. Instead, co-treatment of primary neurons with latrunculin A and A␤ led to an attenuation (50%) of A␤-induced missorting. Conversely, treatment of cells with the F-actin overstabilizer jasplakinolide (1 M) was sufficient to induce missorting of Tau, even independently of A␤ (Fig. 7C).
Thus, only after cell stress ceases is cofilin activity attenuated, and actin in the AIS repolymerizes. Tau entry into the axon is then blocked by aberrant repolymerization of F-actin in the AIS. This finding would be in agreement with earlier data that indicate that A␤-oligomers lose their toxicity after a brief time (1-3 h) in cell culture and that the peak of missorting of Tau also appears only after the A␤-oligomers lose their toxicity (51,54). How can the aberrantly polymerized F-actin block Tau entry? EB3-live imaging of the AIS revealed that 1.5 h after A␤ insult, the EB3 comet speed was reduced by ϳ20% (0.1 m/s) (Fig. 7B). This indicates that repolymerizing actin might exert its effect on Tau sorting via strangulating MT polymerization within the AIS.

A␤ exposure disrupts Tau isoform-specific axodendritic sorting
Previously, we showed that A␤ causes impaired translocation of Tau into axons rather than retrograde leakage of Tau from the axon to the soma (51). The effect of A␤ on the TDB has not been determined so far. We transfected neurons with 2N4R-Tau D2 (strong dendritic presence) as well as 0N4R-Tau D2 (strong axonal presence) and measured axonal sorting in the presence and absence of 1 M A␤. Axonal translocation was determined by photoconversion of Tau D2 in the soma and imaging of axon. 2N4R-Tau D2 is poorly sorted into the axon, whereas 0N4R-Tau D2 is translocated to the axon and past the AIS (100 m from the axon hillock) within 10 min. A␤ treatment decreases the axonal sorting of 0N4R-Tau D2 (Fig. 7D). Conversely, 2N4R-Tau D2 and Dendra2 show increased axonal penetration after A␤ treatment (Fig. 7D, 2). This hints at an alteration of the AIS structure as a result of A␤, probably related to the restructuring of F-actin in the AIS.
To test whether the sorting of rodent (mouse) Tau (mTau) isoforms is different from that of the human isoforms, we cloned the four rodent Tau isoforms (0N3R-mTau, 0N4R-mTau, 1N4R-mTau, and 2N4R-mTau) tagged with citrine (mTau Cit ). We found that mTau isoforms behave slightly differently from human Tau isoforms; differences in axodendritic sorting among the isoforms are less pronounced, and only 0N4R-mTau Cit is more sorted into the axon (axonal e.f. of ϳ100 versus ϳ85 of the other isoforms (Fig. 7E)). This is in contrast to the human Tau isoforms, which show a wider variation between the isoforms (e.f. ranging from ϳ70 to ϳ95). We next tested the contribution of the isoforms to missorting by transfecting individual isoforms into neurons and then exposed cells to A␤. This revealed that the axonal sorting of 0N4R-mTau Cit is reduced (ϳ30%) after A␤ insult, whereas the other rodent isoforms (0N3R, 1N4R, and 2N4R) show no change (Fig. 7E). In summary, we argue that missorting of Tau is due to somatodendritic accumulation of the normally axonally targeted 0N4R-Tau. A␤-caused missorting can be induced at developmental stages where only 0NxR isoforms are present, but 0N3R-Tau distribution is unchanged after A␤ insult. Thus, missorted Tau must consist mainly of 0N4R-Tau.

The Tau diffusion barrier functions both anterogradely and retrogradely, and its tightness is isoform-specific
In the present study, we investigated the TDB and the function of the AIS with respect to physiological sorting and pathological missorting of Tau. We find that longer isoforms of Tau, 2N4R and 1N4R, show almost no retrograde passage from the axon into the soma, whereas all of the shorter splice variants of Tau are able to overcome the retrograde barrier (Fig. 1). The same short isoforms of Tau also show enhanced axonal enrichment (i.e. anterograde sorting) within 4 -6 days when expressed at nearly endogenous levels (Fig. 2). This finding was confirmed by photoconversion of Dendra2-tagged Tau in the soma, which did propagate more slowly into the axon in the case of the longest Tau isoform (2N4R) compared with a medium-sized, well-axonally sorted Tau (0N4R; Fig. 7D). This indicates that the TDB can function in both directions. Interestingly, isoform-specific sorting of Tau occurs both in TauKO and in wild-type neurons, even when exogenous Tau is transfected after the establishment of polarity. Thus, either the Tau sorting machinery is not Tau-specific and develops normally in the absence of Tau, or the cell can rapidly establish Tau sorting even after the process of neuronal polarization has been accomplished. The isoform-specific differences of anterograde sorting (i.e. axonal enrichment of somatically produced Tau) or retrograde passage from the axon to the soma are independent of MT affinity (55) or axonal diffusion rates (Figs. 1 and 2). In the case of 3R-Tau, the second insert appears to decrease the  Cells transfected with Dendra2 alone or 0N4R-Tau D2 show anterograde penetration of the TDB, whereas 2N4R-Tau D2 shows reduced anterograde movement into the axon in control conditions (black bars). Treatment with A␤ for 1 h reduces retention of Dendra2 and 2N4R-Tau D2 but shows a trend for enhanced retention of 0N4R-Tau D2 . E, axonal sorting of the 0N4R-Tau is disrupted by A␤ treatment. Primary cortical neurons 16 DIV were transfected with the different rodent isoforms of Tau tagged with citrine (mTau Cit ; green) and the volume marker tdTomato (red) for 5 days. E1, 2N4R-mTau Cit shows little enrichement in the axon (left panels), whereas the 0N4R-mTau Cit isoform is enriched in the axon (middle panels, arrow; green), but this enrichment is lost after A␤ treatment (right panels). E2, quantification of the axonal enrichment of all rodent Tau isoforms tagged with citrine (mTau Cit ) with and without treatment with 1 M A␤ for 3 h. The shorter 0N isoforms are enriched in the axon (arrow), but this enrichment is partially lost after A␤ treatment. *, p Ͻ 0.05; **, p Ͻ 0.01. Error bars, S.E. ability of Tau to pass the TDB. Considering that 2N3R-Tau is bigger than 0N4R-Tau, also increased size would correlate with inability to cross the TDB. This idea would be in agreement with a previously proposed nonspecific size-dependent filter (27,28). However, mutants of the longest isoform of Tau, derived from FTLD or PSP (i.e. ⌬K280 and A152T), and also a pseudophosphorylation mutant (4KXGE; all four KXGS motifs mutated to KXGE) show enhanced retrograde and anterograde passage of Tau, indicating that size alone is not the sole determinant.

Tau increases spine development and dendritic growth in an isoform-dependent manner
The longest isoform of Tau (2N4R-Tau) along with the other 4R isoforms is expressed only after the initial polarization has taken place (i.e. not during embryonal development), both in rodents and humans. When expressed in low endogenous-like levels, some Tau isoforms (e.g. 3R-Tau species, especially 0N3R) induce filopodia-like spine growth. Other isoforms (0N4R and 1N4R) had no or little effect on promoting spine development. 2N4R-Tau, however, induces striking maturation of spines and elongation of dendrites (Fig. 2, D-F). This is in line with the appearance of specific Tau isoforms during development; the short (3R) isoforms are expressed early in development, driving filopodia spine formation, whereas later in life, the longer isoforms (e.g. 2N4R) are expressed (6), which concurs with spine maturation. Accordingly, knockdown of Tau at the time of spine formation and maturation (14 DIV) results in loss of spine formation. This points toward a bona fide role for Tau in dendritic development. Future studies (e.g. by knocking down individual isoforms of Tau while maintaining the physiological ratio of the remaining isoforms) will have to elucidate which isoform of Tau during which state of development is responsible for or contributes to normal spine development.

Sorting of Tau can be achieved independently of regulatory translation but is modulated by phosphorylation of Tau and AIS structural integrity
We studied the mechanistic/structural manner in which the TDB is able to confine the Tau protein to the axonal compartment. Previously, Tau mRNA was shown to be translocated to the axon by a 3Ј-UTR axonal targeting signal, implying local translation of Tau (21). However, the exogenous Tau constructs that we used to investigate the TDB and the axonal enrichment do not carry the 3Ј-UTR. Thus, in our model, Tau mRNA does not contribute to the sorting of Tau.
Phosphorylation of Tau probably has a function in sorting of Tau; pseudophosphorylation at the KXGS motifs enhances anterograde and retrograde propagation of Tau in the axon (23,56). Activation of the kinase MARK (primarily responsible for phosphorylation of these epitopes) lags behind A␤-induced missorting of Tau, indicative of a role in resorting of Tau. Here we show that Tau pseudophosphorylated at the KXGS motifs shows enhanced axonal enrichment compared with wild-type Tau ( Fig. 2 and supplemental Figs. S1 and S2), indicating the importance of this phospho-site for sorting. Another prominent Tau kinase, GSK3␤, is a major player within the AIS, where it is anchored (37), but it does not phosphorylate the KXGS motifs of Tau. Testing the influence of GSK3␤ on the TDB revealed that GSK␤ induces accumulation of endogenous and exogenous Tau in the somatodendritic compartment ( Fig.  3 and supplemental Fig. S5). In the case of retrograde missorting of exogenous Tau, this effect is independent of direct phosphorylation of Tau by GSK3␤, because even non-phosphorylatable Tau shows the same retrograde propagation from the axon into the soma (Fig. 3). Thus, the interplay of GSK3␤ with components of the AIS/TDB is essential for sorting and missorting of Tau in our model. Knockdown of other structural components of the AIS results in weakening of the TDB. Tau transit through the TDB is facilitated when classical AIS components (e.g. AnkG and EB1) are knocked down via shRNA. All knockdowns also induce increased invasion of the dendritic marker MAP2 into the proximal part of the axon. This indicates that the classical AIS components contribute to the TDB, but also to the nonspecific filter function of the AIS.

Structures around the TDB: Highly dynamic MTs in the AIS
MT destabilization by nocodazol results in missorting and breakdown of the TDB, whereas MT stabilization by taxol prevents missorting after A␤-insult (23,51). We thus hypothesized that the structure of MTs within the AIS holds the key to the TDB. Surprisingly, MTs within the AIS are extremely dynamic. Markers of stable MTs (acetylation and polyglutamylation) are absent in the AIS. MTs could not be stained in the AIS without the use of the molecular densifier PEG and taxol, emphasizing the transient nature of these MTs (Fig. 4), similarly to previous studies (32). EB3-based live imaging of MT dynamics confirmed the dynamic nature of AIS MTs and revealed a hub of EB3 comet spawning at the central part of the AIS (Fig. 5), consistent with observations of enrichment of EB proteins in the AIS (29).
Of note, EB3 comet speed (and thus MT growth rates) is unchanged in the different parts of the AIS and also in the axon (constant at ϳ0.5 m/s for the proximal axon, the AIS itself, and the axon hillock); only after an AD-like insult with A␤ oligomers does the EB3 comet speed decrease by ϳ20% (Fig. 7B).

AD-like insult by A␤ causes axodendritic missorting of Tau due to activation of cofilin and loss of MT growth within the AIS
A␤ oligomers bind preferentially to dendritic spines. However, within axons, the AIS is a preferential target for A␤ (supplemental Fig. S8). Because the AIS is composed of a multitude of regulatory elements and receptors (57), A␤ binding may impair the function of the AIS. This is indeed the case for the TDB, which lies within the AIS. A␤ exposure causes activation of cofilin, resulting in increased severing, loss, and subsequent remodeling of F-actin. Whereas this leads to increased amounts of F-actin in the dendritic shaft probably due to redistribution from shriveling spines, the first effect in the AIS is a decrease of F-actin. Thus, the observed activation of cofilin and consequent loss of F-actin, particularly within the AIS, perturb the F-actin meshwork in the AIS. This meshwork acts as a filter to retain non-axonal proteins in the soma. The structural organization appears different, depending on the techniques employed, but probably consists of spectrin IV-bound regularly spaced F-actin rings; still, a large portion of the actin cytoskeleton in the AIS is Isoform-dependent (mis)sorting of Tau thought to be dynamic (58,59). In the case of the TDB, breakdown of F-actin results in anterograde penetration of the "dendritic" 2N4R-Tau, but also failure of proper sorting into the axon of the stronger axonally sorted 0N4R-Tau (Figs. 6 and 7). Thus, the failure of the TDB results in axodendritic missorting of Tau that is isoform-specific.
A␤-induced loss of F-actin within the AIS may also result in facilitated diffusion of other proteins; Dendra2, for example, shows ϳ2 times faster diffusion into the axon. A␤-induced disruption of F-actin within the AIS would explain the increased anterograde propagation of the longest isoform of Tau, but why is the normally strongly axonally targeted major isoform of Tau (0N4R) missorted after A␤ insult (Fig. 7)? First, because A␤ also disrupts the anterograde polymerization of MTs within the AIS, this would decrease the pumping mechanism that transports the Tau-tubulin complex into the axon (56). Second, because the peak of missorting of Tau (3-6 h (51)) lags substantially behind the loss of F-actin and activation of cofilin (0.1-1 h; Fig. 6), probably the subsequent (aberrant) remodeling of F-actin is responsible for missorting, rather than the loss of F-actin per se. Consistently, the F-actin stabilizer jasplikanolide induces missorting, but not the F-actin destabilizer latrunculin.
Remodeling of the F-actin cytoskeleton within the AIS would result also in the impairment of MT polymerization within the center of the AIS, although it is currently unclear whether there is a direct causal link and whether there is an effector protein.
Anyhow, the remodeling of F-actin within the AIS does not occur prior to 1 h after A␤ exposure. Thus, strangulation of MT growth by aberrant F-actin repolymerization within the AIS would start after A␤ has lost its toxicity and F-actin starts to repolymerize. This we indeed observed (Figs. 6 and 7A). Tau is then unable to traffic along MTs, decreasing its ability to pass the anterograde TDB, resulting in somatodendritic accumulation (classical missorting). The cause-effect relationship, however, is far from obvious, because it was shown recently that Tau stabilizes MT growth, judged by EB1 and EB3 comet number and comet path length (60). Thus, the reduced entry of Tau into the AIS (due to aberrant F-actin remodeling) may also contribute to reduced MT growth.

Conclusion
We show here that the barrier function of the TDB within the AIS can operate both in the anterograde and retrograde direction. For 90 -97% of Tau molecules, where the second insert is absent, Tau passes through the TDB anterogradely and is retained in the axon. In the case of the longest isoform, 2N4R (ϳ3-10% of Tau), Tau molecules are partially retained in the somatodendritic compartment, where they enhance dendritic outgrowth and spine formation. Passage of the TDB barrier in both directions is facilitated when Tau itself is mutated (i.e. in the case of pathological mutations, phosphorylation, knockdown of AIS components, or overexpression of GSK3␤ (an AIS and AD kinase)). Pathomechanistically, we show that AD-like A␤ insult leads to cofilin activation and remodeling of F-actin within the AIS, resulting in decreased MT mobility. This leads to somatodendritic accumulation of 0N4R-Tau, hence classical missorting, but also enhanced axonal penetration of 2N4R-Tau. In summary, we provide evidence for physiological and pathomechanistic differences of the different Tau isoforms and their role in sorting and missorting and demonstrate their functional importance for MT and actin dynamics within the AIS for the maintenance of the TDB.

Cell culture and transfection
Dissociated cortical neurons were prepared from embryonic day 18 Wistar rats or embryonic day 16 C57BL6JRj wild-type or TauKO mice according to Ref. 61 and plated on a poly-D-lysine (50 g/ml)-coated coverglass. After 4 DIV, 0.3 M cytosine arabinoside (Sigma, Munich, Germany) was added to the cultures to reduce glial growth. Cultures were maintained in NeuroBasal medium with B27 (Invitrogen) or NS21 (PanBiotech), L-glutamine (PAA), and penicillin/streptomycin (PAA/PanBiotech) at 37°C, 5% CO 2 . Lipofectamine-based transfection for experiments for retrograde Tau movement was conducted as described in Ref. 23 (high protein expression). Transfection for experiments investigating the Tau distribution after 4 -6 days of low expression was conducted as described in Ref. 62, making use of standard Lipofectamine 2000 (L2K; Invitrogen) reagent, but not OptiMEM. Briefly, for low protein expression, the DNA amount was limited to 1.2 g/construct/well of a 6-well multiwell plate or to 0.6 g/construct/well of 12-or 24-well plates. Instead of OptiMEM, pure unsupplemented Neurobasal medium (Gibco) was used. L2K/DNA mix was prepared as recommended (5 min ϩ 30 min of incubation). Next, cell culture medium already in the multiwell plates (conditioned medium) was carefully removed, leaving a small amount of conditioned medium in the well (1 ml for 6-well plates or 0.5 ml for 12-and 24-well plates). The removed conditioned medium was saved. Cells were then incubated for 6 h for cells older than 14 DIV or for 3 h for cells between 7 and 10 DIV) with the L2K transfection mix. Next, cells were washed twice in the multiwell plates with preheated and carboxylated unsupplemented neurobasal medium, taking great care that cells never fell dry (critically important for older cells). Next, the previously removed conditioned medium was used to allow expression of transfected constructs for 4 -6 days. Exposure of cells to reagents and media other than conditioned medium was kept to an absolute minimum to avoid cell degeneration (see also Ref. 62 for critical steps and modulation of expression).

Imaging, photoconversion, and analysis
A Fluoview1000 confocal microscope (Olympus, Hamburg, Germany) equipped with a SIM scanner (which allows for simultaneous stimulation and imaging) and a ϫ60 objective live-cell imaging chamber and ZDC system for z-drift compensation was used for photoconversion experiments and image acquisition. For time-lapse live-cell imaging, cells were plated on glass-bottom dishes and were kept in the imaging chamber (37°C, 65% humidity and supplied with 5% CO 2 ). Images were taken every 30 s for all applications. Before photoconversion, green fluorescence of Dendra2 was acquired by excitation at 488 nm and detection over the range of 500 -550 nm. Photoconversion was performed during the imaging interval by rapidly illuminating the selected area with a 405-nm laser scanner (1-2 exposures of 1 s each, 405-nm laser power 15%). Thereafter, red fluorescence was acquired by excitation at 561 nm and detection over the range 570 -670 nm. In the case of PS-CFP2 protein, cyan fluorescence was acquired by excitation at 405 nm with 0.2% laser power and detection over the range of 420 -510 nm. Photoconversion was performed similar to Dendra2, and green fluorescence of PS-CFP2 was acquired by excitation at 488 nm and detection over the range of 500 -550 nm. For comparison of Dendra2 and PSCFP, fluorescence intensities were normalized using Dendra2 or PS-CFP2 alone and hTau40 tagged with the different fluorescent proteins as controls; fluorescent photoswitchable protein (PSCFP2 and Dendra2) cell body intensities after axonal conversion were set to 100, and baseline levels were normalized by noise subtraction. For knockdown of the different AIS components, PSCFP2 was used because knockdown vectors contained mRFP, which had to be used to identify cells expressing the shRNA. Fluorescence intensity profiles and values according to the indicated time points and area were determined by the microscopy software (Olympus FV1000 Software). Microscopic settings as well as the selection of the regions of interest were kept similar to allow comparison between different experiments, as described previously (23). STED imaging was conducted as recommended by the manufacturer (Leica) with a gSTED system on TCS SP8 secondary antibodies conjugated with CF488 (VWR) or Alexa488 (Dianova) as chromophores and AntifadeGold mounting medium (Invitrogen). Images were analyzed using manufacturer software (Leica LAS AF Lite). Standard immunofluorescence was conducted essentially as described before (12,51), and see supplemental Table S2 for a list of antibodies used. Axons were identified by the absence of MAP2 and presence of Tau staining, branching pattern of ϳ90°, recurrence to cell body and crossing of processes of the same cell, length Ͼ 300 m, and constant diameter. Dendrites were identified by positive MAP2 stain, the presence of dendritic spines, branching pattern of ϳ60°, constantly decreasing diameter, and length Ͻ 300 m. Cell bodies, dendrites, and axons have very different diameters, which is impossible to judge by light microscopy. We therefore used a volume marker (tdTomato) and compared the intensities of transfected constructs (i.e. Tau D2 stained with human Tau-specific antibody CP27) with the intensities of the volume marker. The volume marker was assumed to be distributed throughout the cell in an unbiased manner, thus making comparisons of the different Tau constructs possible. Values given under "Results" and in the figures were calculated by dividing the normalized (i.e. CP27 fluorescence intensities (Tau) versus tdTomato (Tom) fluorescence intensities) axonal values (AV) versus dendritic values (DV). The formula, therefore, is (Tau AV /Tom AV )/(Tau DV /Tom DV ); for illustration purposes, values were multiplied by a factor of 50 (to attain a maximal value of 100).
For the evaluation of spine development after transfection of different Tau isoforms and mutations leading to Fig. 2D and supplemental Table S1, at least three cultures from each genotype of wild-type and TauKO hippocampal neurons and several cultures from rat cortical and hippocampal neurons were included.
MT stabilization during extraction was conducted using 3-min extraction at room temperature in MT-stabilizing buffer (PEM buffer: 2% polyethylene glycol (M r 10,000), 100 mM PIPES-KOH, pH 6.9, 1 mM MgCl 2 , 1 mM EGTA) containing 1% Triton X-100 and 2 M taxol (all Sigma or VWR) after a quick rinse in PBS (PAA/Sigma). Cells were then fixed with 3.7% formaldehyde and 0.5% glutaraldehyde in PEM buffer for 20 min. Mitochondria and EB3 live imaging was conducted at 2-s intervals with an MM-based Leica DM6000, processed, and analyzed with Metamorph software.

Other procedures
Western blotting, ATP and LDH measurements, and statistical analysis (*, p Ͻ 0.05; **, p Ͻ 0.01 versus control if not indicated otherwise) were done as described before (12,51). Dephosphorylation of endogenous rat Tau was done without phosphatase inhibitors but with protease inhibitors, making use of alkaline phosphate (FastAP, ThermoFisher Scientific) for 16 h. Oligomeric amyloid-␤ was prepared as described before (51). Briefly, A␤40 and A␤42 (both 21st peptide) were mixed in a 7:3 ratio and oligomerized via sequential dissolution in 40 mM NaOH to a 2 mM A␤ solution and consequently diluted in PBS to a 100 M solution. This solution was then incubated for 1 h at 37°C. For complete characterization and differences from unmixed preparations, see Ref. 51.