Missorting of Tau in Neurons Causes Degeneration of Synapses That Can Be Rescued by the Kinase MARK2/Par-1

Early hallmarks of Alzheimer's disease include the loss of synapses, which precedes the loss of neurons and the pathological phosphorylation and aggregation of tau protein. Mitochondrial dysfunction has been suggested as a reason, but evidence on the role of tau was lacking. Here, we show that transfection of tau in mature hippocampal neurons leads to an improper distribution of tau into the somatodendritic compartment with concomitant degeneration of synapses, as seen by the disappearance of spines and of presynaptic and postsynaptic markers. This is accompanied by transport inhibition of vesicles and organelles, concomitant with an increase and bundling of microtubules. Mitochondria degenerate, thus causing ATP levels to decrease. The tau-induced synaptic decay can be relieved by the activation of the kinase MARK2 (microtubule-associated protein/microtubule affinity regulating kinase 2)/Par-1 (protease-activated receptor 1), which can remove tau from the microtubule tracks and reverses the transport block. This leads to the rescue of dendritic spines, synapses, mitochondrial transport and ATP levels.


Introduction
Early changes in Alzheimer's disease (AD) brains include missorting of tau and loss of synapses. This precedes the anomalous aggregation of proteins (amyloid plaques, neurofibrillary tangles) and is correlated with cognitive dysfunction (Terry et al., 1991;Coleman and Yao, 2003). The reasons for synaptic loss are not well understood. The amyloid hypothesis holds that toxic aggregates of the A␤ peptide lead to the dysfunction of synapses (Hardy and Selkoe 2002;Spires and Hyman, 2004). The role of tau in this context has not been clear. Normal tau is a highly soluble microtubule (MT)-associated protein in axons, but in Alzheimer's disease and other tauopathies it accumulates in the somatodendritic compartment where it eventually aggregates (Delacourte and Buee, 2000). Furthermore, the distribution of neurofibrillary tau deposits correlates closely with the neuronal loss and the clinical progression of AD (Braak and Braak, 1991).
Dendritic spines, the postsynaptic parts of synapses, extend from the surfaces of the dendritic processes of neurons. Their morphology varies from small mushroom shaped extensions to "neckless" (stubby) spines and filopodia-like spines which occur mostly during spine morphogenesis. The function of spines in neurons is to propagate signal inputs (Yuste and Bonhoeffer, 2004). Spines are devoid of microtubules but rich in actin (Hering and Sheng, 2001;Matus, 2000). Nevertheless, microtubules also play a role in postsynaptic function, notably as tracks for the transport of synaptic components and organelles. Thus, activitydependent fusion of vesicles into the dendritic membrane is blocked by depolymerization of microtubules, and microtubule motors are responsible for dendritic transport of glutamate receptor subunits and other cargoes (Setou et al., 2000).
In transgenic mice, overexpression of tau leads to pathology, as seen by missorting of tau into dendrites, tau hyperphosphorylation, aggregation, and synaptic and neuronal loss (Duff and Sulemen, 2004;LaFerla and Oddo, 2005). Analogous findings have been reported for overexpression of tau in Drosophila (Wittmann et al., 2001;Chee et al., 2005). We showed previously that elevated tau leads to inhibition of anterograde traffic of vesicles and organelles in chick retinal ganglion cells. This is explained by an interference between tau and microtubule motors and can be relieved by kinases such as microtubule-associated protein (MAP)/microtubule affinity regulating kinase 2 (MARK2)/protease-activated receptor 1 (PAR-1), which phosphorylate tau in the microtubule-binding domain and cause its detachment from microtubules (Mandelkow et al., 2004). Because of the inhibitory effect of tau on traffic, we suspected that tau might also affect synapses. Here, we demonstrate a dramatic loss of synaptic markers and morphological changes caused by tau, illustrated by the disappearance of spines in mature hippocampal neurons. Postsynaptic markers such as the scaffolding protein PSD95, neuroligin, AMPA and NMDA receptor (NMDAR) subunits decrease in tau-transfected neurons, as well as presynaptic proteins such as bassoon and piccolo. This degeneration is accompanied by the dysfunction of mitochondria. The damage can be prevented by MARK2/PAR-1, which preserves dendritic spines and synapses.

Materials and Methods
Cloning. Adenoviruses were generated as described previously (Stamer et al., 2002). Briefly, the SalI-XbaI fragment containing the cDNAs [blue fluorescent protein (CFP)-htau40, CFP-ht40/KXGA (Ser to Ala mutations in all four KXGS motifs in the repeat domain) mutant, and yellow fluorescent protein (YFP)-MARK2 wild-type] were subcloned into the restriction sites of the pShuttle vector under control of the human cytomegalovirus promotor . Plasmids were linearized with PmeI and cotransfected with the pAdEasy 1 vector into Escherichia coli BJ 5183 for homologous recombination. Plasmid DNA was amplified in E. coli DH10B and digested with PacI to cut out the entire recombinant adenoviral DNA, which was transfected into human embryonic kidney 293 (HEK293) cells using Lipofectamine (Invitrogen, Karlsruhe, Germany). Generation of viruses in HEK293 cells was monitored by fluorescence microscopy. Cells were harvested, resuspended in PBS, and lysed by three freeze-thaw cycles. Cell debris and nuclei were removed by centrifugation, and the virus suspension purified by two CsCl gradient centrifugations. CsCl was removed by gel filtration and equilibrated in storage buffer containing (in mM) 10 Tris/Cl, 135 NaCl, 3 KCl, 1 MgCl 2 , 10% glycerol, pH 8.0.
Cell culture. Hippocampal neurons were isolated from embryonic day 18 (E18) Sprague Dawley rats, plated on poly-D-lysine (50 g/ml)-coated cover glass for immunofluorescence (IF) or on coated plastic for protein biochemistry at a density of 20,000 cells/cm 2 . Cultures were treated after 4 d with 5 g/ml cytosine arabinoside (Sigma, Munich, Germany) to reduce glial growth. Cells were maintained in NeuroBasal medium with B27 (Invitrogen) for 3-4 weeks. Cultures were then transfected with adenovirus and harvested or fixed up to 9 d after transfection. Cells were transfected with tau, MARK2/Par-1, or GFP (green fluorescent protein) using adenovirus. Ten microliters of virus suspension was added per 1.5 ϫ 10 5 cells and incubated for 1-10 d. Cells were fixed for immunofluorescence or harvested for SDS-PAGE. For adenoviral transfection of CFP-htau40 or YFP-MARK2, a 100-fold multiplicity of infection (3 ϫ 10 7 infectious particles) was applied to primary neurons. For double transfections, 3 ϫ 10 7 particles of each recombinant adenovirus were added. After 4 h incubation, the viral suspensions were removed.
Image analysis. Hippocampal neurons were observed with a 63ϫ objective on an LSM510 Meta confocal microscope (Zeiss, Jena, Germany) using lasers, beam splitters, and filters according to the fluorophores. For time-lapse imaging of spines and mitochondria, cells were plated on glass bottom dishes. Cells were kept at 37°C by air heating and supplied with 5% CO 2 . For the analysis of spine morphology and density, confocal stacks were taken and spines counted and measured.  (msp,19;spsp,52;tdn,4;ex,2;tdl,178.1 m),and KXGA plus MARK (msp,30;spsp,21;tdn,15;ex,24;tdl,752.8 m).
Spine densities for each dendrite were measured and the average was plotted in Figure 2 D. The error bars are SEs resulting from the spine densities per dendrite.
Western blot analysis. Neuronal cultures grown on plastic dishes were scraped into lysis buffer, concentrations were determined, and equal amounts subjected to SDS-PAGE. Quantification was done by densitometry using NIH Image, and normalized by the concentration of actin. Densities of untransfected cultures were set to 100%.
Cell treatment and ATP determination. ATP depletion of cell cultures required inhibition of the glycolytic and oxidative energy metabolism, accomplished by washing cultures with glucose-free PBS with 0.5 mM CaCl 2 and 1.0 mM MgCl 2 . Cultures were then kept in this buffer containing 1.0 M or 0.1 M antimycin (Sigma) for 2.5 h. Control cultures were kept in buffer without antimycin. For quantification of ATP, we used the sensitive ATP determination kit (Biaffin, Kassel, Germany) optimized for a total content of 0.1-100 pmol ATP (Sakakibara et al., 1999). Cells were plated at a density of 20,000 cells/cm 2 , cultured for 23-25 d, and transfected for 3 d with either tau alone or MARK2 plus tau. Cultures were kept in media until the ATP measurement or in glucose-free PBS supplemented with 0.5 mM CaCl 2 and 1.0 mM MgCl 2 for 2.5 h. Measurements were done in triplicate. Before harvesting, the cells were washed once with PBS. One-hundred microliters of lysis buffer was added, scraped into tubes, and immediately frozen in liquid nitrogen. They were then thawed at 37°C and spun down. The supernatant was transferred into a fresh tube. Ten microliters of lysate were diluted to 1000 l with water. Thirty microliters of the lysate were mixed with 30 l of reagent. After 10 s, the ATP content was determined using a luminometer (Eppendorf, Hamburg, Germany) at 562 nm and integrated over 20 s. All ATP measurements were normalized by the protein concentration. Values of untransfected cultures were set to 100%.
Electron microscopy. Neurons were grown on glass-bottom dishes for 3-4 weeks and infected. Cultures were fixed in 2.5% glutaraldehyde for 1 h, washed with 0.15 M cacodylate, and imaged. Transfected cells were marked and postfixed for 1 h with 1% osmium tetroxide and 1% uranyl acetate, dehydrated with increasing concentrations of ethanol, and embedded using Epon 812 (Fluka, Neu-Ulm, Germany). After curing the resin, cells of interest were located by phase contrast microscopy, circled with a needle, mounted and sectioned either parallel or perpendicular to the long axis of the processes at a thickness of 50 -80 nm, and viewed with a Phillips (Einhoven, The Netherlands) CM12 electron microscope.

Missorting of tau and effects on synapses
In a normal differentiated neuron, tau is confined to axons whereas MAP2 occurs in the somatodendritic compartment (Binder et al., 1985). To assess the effect of tau on synapses, we transfected hippocampal neurons containing mature synapses [25 d in vitro (div)] with adenovirus encoding CFP-tagged htau40, the longest isoform in the human CNS. After 24 d of culturing of cells, Tau is only present in the axons and MAP2 only in the dendrites of hippocampal neurons ( Fig. 1 A). Within 1 d after transfection the CFP-tagged tau appears in the entire neuron, axons as well as somatodendritic compartment (Fig. 1 B). This results in a dramatic change in the postsynaptic dendritic spines. Initially, the dendrites contain well defined mushroom spines filled with F-actin (as seen by phalloidin staining) (Fig.  1C). After 18 h of tau expression, tau is found along the whole dendrite including the mushroom spine heads (Fig. 1 D). The tau in spines is mostly detached from microtubules because microtubules are absent from spines. After 30 h, the spines change to a filopodia-like appearance and decrease in density. Finally, after 2-4 d of tau expression, the dendrites are nearly devoid of spines and have a smooth shaft, in strong contrast to untransfected cells ( Fig. 1 E, bottom). F-Actin (stained by phalloidin) retracts into the dendritic shaft and cell body, and in addition F-actin shows an irregular accumulation in the cell body ( Fig. 1 F). After 5-6 d, the cell processes disintegrate and cells degenerate and die.

Tau-induced degeneration of spines can be rescued by MARK2
Because tau can inhibit axonal transport by interfering with the attachment of microtubule motors, and because phosphorylation of tau in the repeat domain by the protein kinase MARK2/Par-1 can detach tau from microtubules (Stamer et al., 2002;Mandelkow et al., 2004), we suspected that the tau-induced synaptic decay was related to a transport defect in dendrites and axons that could be rescued by MARK2. To test this we transfected neurons with tau and MARK2 by adenovirus in different intervals. In contrast to transfection with tau alone, neurons transfected simultaneously with CFP-tau and YFP-MARK2 and observed after 3-4 d maintain numerous mushroom-shaped spines and many spiny spines. The mushroom-shaped spines contain actin, tau, and MARK2 ( Fig. 2 A, arrow). As a control, in neurons transfected with MARK2/PAR-1 alone, the spines appear filopodia-like, typical for spine morphogenesis. Actin-phalloidin label is seen in the shaft as well as in the elongated spines and filopodia-like structures (Fig. 2 B).
Spine density and length was assessed by counting mushroom-shaped spines and spiny spines per micrometer of dendrite ( Fig. 2 D, E). In control cells, essentially all observed spines are mushroom-shaped, whereas in tau-transfected neurons hardly any spines are seen with F-actin staining after 4 d of transfection, leaving mostly bare dendritic shafts. The density of spiny spines (lacking spine heads) increases dramatically for MARK2-transfected cells and for cells transfected with tau plus MARK2. Therefore, the total spine density (with or without spine heads) increases significantly for singly MARK2transfected and doubly tau plus MARK2-transfected cells (Fig.  2 D). When cells are incubated with CFP-tau adenovirus for 2 d before adding YFP-MARK2 adenovirus for an additional 2 d, MARK2 can surprisingly rescue spine morphology and density nearly to control levels. However, an expression of tau alone longer than 2 d cannot be rescued by MARK2. Typically, the protection of synapses and cell integrity by MARK2 against tauinduced damage is temporary and lasts for ϳ4 d. After that, cell processes and eventually cells disintegrate and die. By comparison, untransfected cells survive up to 6 weeks. As a control, transfection of tau-KXGA mutant (which lacks the KXGS sites phosphorylatable by MARK2) into neurons also leads to a loss of spines, similar to wild-type tau, but in this case the loss cannot be rescued by MARK2 because the KXGA mutant of tau cannot be phosphorylated by MARK2 (Fig. 2C-E).

Phosphorylation of tau by MARK2
Normal differentiated neurons show tau only in the axon, and this tau has no detectable phosphorylation in the repeat domain by immunofluorescence (KXGS motifs, visualized by 12E8 antibody) (Fig. 3A). In contrast, cells transfected with CFP-tau show elevated tau (missorted into the somatodendritic compartment), and this tau already contains substantial phosphorylation at KXGS motifs (Fig. 3B). When MARK2 is cotransfected together with tau ( Fig. 3C), phosphorylation of tau in the repeat domain increases strongly, concomitant with an increase in soluble cytoplasmic tau, which can diffuse into dendritic spines (Fig. 1C).
Quantification of untransfected cells shows the typical superposition of several endogenous rat tau isoforms (Fig. 3D, lane 1). Transfection with exogenous htau40 results in an additional band, distinguishable by its higher relative molecular weight because of the CFP label (lane 2). Its expression is approximately equal to endogenous tau (correcting for the fraction of trans-Figure 1. Tau distribution and changes in dendritic spine morphology after expression of tau. A, Untransfected hippocampal neurons. K9JA, an antibody for total Tau is localized only in axons, whereas MAP2 visualized by MAP2 clone HM2 antibody (Sigma) only localizes in dendrites. B, Expression of CFP-tau in hippocampal neurons leads to missorting into dendrites. CFP-tau appears in the soma and dendrites and colocalizes with MAP2 (HM2 antibody). C, Phalloidin fluorescence illustrates the concentration of F-actin in spine heads. D, After tau expression, CFP-tau appears along the dendrites and also in dendritic spines, followed by their disappearance (shown at 18, 30, and 48 h after transfection). E, F, Transfection by CFP-Tau for 3 d causes loss of spines and accumulation of F-actin in the dendritic shaft (E) and cell body (F ) (staining by CFP-tau, phalloidin, and merge). Scale bars, 10 m. fected cells, ϳ60%). Pronounced changes occur in the phosphorylation state (at the KXGS motifs) as judged by antibody 12E8 (Fig. 3E). Endogenous tau shows a low level of phosphorylation (lane 1), transfected human tau shows an enhanced level (lane 2), but with MARK2 (lane 3), the phosphorylation is dramatically enhanced for all tau species (sixfold, compared with expression of tau only). This illustrates that the rescue by MARK2 of the tauinduced synapse decay is strongly correlated with the higher level of tau phosphorylation in KXGS motifs and, thus, with tau's detachment from microtubules.

Loss of synaptic markers with tau and rescue with MARK2
Because the shape and density of synapses changes with tau expression, we investigated the influence on synaptic markers. As a control, Figure 4 A illustrates the juxtaposition of synaptophysin (a presynaptic marker) with AMPA receptors (AMPAR, postsyn- Image stacks were taken of each dendrite with the confocal microscope. The spine density changes significantly in tau-transfected and tau/KXGA transfected neurons. The number of mushroom-spines and spiny spines is significantly reduced. Neurons transfected with tau plus MARK2 for 4 d show an increase in density and length of spiny spines and therefore a 30% increase in total length, but only a minor increase in the length of mushroom spines and no significant difference in their density. Cells transfected for 2 d with tau and then for 2 more days with MARK2 also showed an increase in the number of spiny spines compared with control. Mushroom spine length and density was comparable with control. Spine density was strongly reduced in tau-transfected neurons, but higher in neurons transfected with tau plus MARK2 because of an increase in the number of spiny spines. In cells transfected first with tau and 2 d later with MARK2, the density of spines was similar to control, and the number of spiny spines was not elevated. Spiny spines increase with transfection of MARK2 alone and mushroom spines decrease. Scale bars: A-C, 10 m. Error bars indicate SE. aptic) and actin-phalloidin along dendrites. Likewise, there is a tight colocalization with postsynaptic markers PSD95 and AMPAR with F-actin. Actin and synaptophysin are prominent around the cell body (Fig. 4B). In contrast, when tau is expressed for 3 d, PSD95, actin-phalloidin, and AMPAR are strongly reduced (Fig. 4C). Synaptophysincontaining structures remain numerous along the dendrites. AMPAR and synaptophysin accumulations are seen in the cell body, indicating a transport defect in the cells (Fig. 4D). This confirms the general observation that most synaptic markers are strongly reduced after tau expression (Fig.  5A-D). The density of synapses along the dendrites (ϳ0.3-0.4 per micrometer in controls) (Fig. 4G) decreases sharply after tau expression; however, it stays at control values after coexpression of tau plus MARK2. Classification into shaft and spine synapses reveals that spine synapses disappear almost completely after tau expression, but are maintained in the presence of MARK2 (Fig. 4H). A remnant of a spine is seen in Figure 4I, but no PSD is detectable in this section or adjacent ones (open arrow), and the spine apparatus is dilated (filled arrow). Almost all observed synapses (98%) are shaft synapses which appear functional and have visible PSDs (Fig. 4J, open arrow). In doubly transfected cultures (tau plus MARK2) the relationship between shaft and spine synapses are comparable with control values. The tau-induced suppression of synaptic markers is in remarkable contrast to the rescue achieved by cotransfection with MARK2, illustrated here for the case of AM-PAR, PSD95 (postsynaptic), and synaptophysin (presynaptic) (Fig. 4E).
The picture emerging from the immunofluorescence data are confirmed by Western blot analysis of neuronal proteins ( Fig.  5A-D). Neuroligin and PSD95 (cell adhesion and scaffolding molecules in postsynaptic densities) decrease after tau expression, whereas in doubly transfected cultures (tau plus MARK2) the levels are preserved (Fig. 5A). Postsynaptic AMPA and NMDA receptors are reduced in tau-transfected neurons, whereas cotransfection with tau and MARK2 keeps the levels of the receptor subunits stable (Fig. 5B).
To understand the effect of tau on presynapses, we studied proteins of the presynaptic active zone such as bassoon and piccolo, two proteins of the cytoskeletal matrix colocalizing at active zones (Dresbach et al., 2001). A clear decrease is seen in tautransfected cultures after 2-4 d of transfection, but not in doubly transfected cultures (tau plus MARK2) (Fig. 5C,D). Thus, presynapses seem to be affected by tau and MARK2 directly at the active zone. In contrast, synaptophysin, a vesicle-bound presynaptic protein does not change with tau expression, but rather accumulates in the cell body.

Tau-induced synapse loss is not caused by apoptosis
There is a debate on how apoptosis or necrosis contribute to cell death in neurodegenerative diseases. Furthermore, neuronal cell processes can enter a preapoptotic state before their decay (Berliocchi et al., 2005;Mattson and Magnus, 2006). Because tau causes a loss of synapses, we hypothesized that this might be a prelude to apoptosis, and therefore searched for apoptotic markers (Fig. 5 E, F ). Tau-transfected hippocampal neurons show only a slight increase of DNA fragmentation [4,6-diamidino-2phenylindole (DAPI) staining] starting 5 d after transfection. Staining for active caspase-3 and the early apoptotic marker annexin does not increase in tau-transfected neurons. An increase of the number of caspase-3-and annexin-positive cells is observed in doubly transfected neurons (tau plus MARK2) 5 d after transfection (Fig. 5E). The cleavage of the nuclear protein PARP by caspases is a further marker of apoptosis. In tau-transfected cells a downregulation of PARP is observed, but no cleavage up to 6 d (Fig. 5F ). In tau plus MARK2-transfected cells a low level of PARP cleavage is seen starting at day 4. Thus, apoptotic effects do not explain the tau-induced decay of synapses shortly after transfection.

Movement of mitochondria is perturbed in tau-transfected neurons
Defects in mitochondrial metabolism have been proposed to underlie neurodegenerative diseases, and proper mitochondrial lo- calization is a prerequisite for neuronal function because of the high demand for energy and calcium regulation at synapses (Beal, 2005;Li et al., 2004). In previous studies, we showed that elevated tau is capable of inhibiting the transport of mitochondria (Stamer et al., 2002). Because the dynamics of spines depends critically on the assembly and disassembly of the actin cytoskeleton, and because actin polymerization is a major cause for ATP consumption in cells (Sarmiere and Bamburg, 2004), we hypothesized that the loss of synapses might be related to dysfunctional mitochondria. We stained neurons with marker dyes reporting on total mitochondria and on their state of activity. Control neurons display the normal uniform distribution of active mitochondria in cell bodies and dendrites (Fig. 6 A, B), but neurons transfected with tau are mostly devoid of mitochondria in the processes and cell bodies after 3 d (Fig. 6C,D). In contrast, tau plus MARK2 doubly transfected cells show again a normal distribution of active mitochondria (Fig. 6 E).
The change in the distribution of mitochondria in dendrites corresponds to a change in mobility. Untransfected cultures display saltatory movements of mitochondria in both directions (Fig. 6 F,  top), which are supported mostly by microtubule-dependent motor proteins (Hollenbeck and Saxton, 2005;Hirokawa, 1998). In contrast, in tau-transfected cells, the movement of mitochondria is strongly inhibited. At 2 d after transfection, the mitochondria become mostly stationary (Fig. 6 F, center) and subsequently disintegrate (Fig. 8K). However, in doubly transfected neurons (tau plus MARK2) the saltatory movement remains mostly intact (Fig. 6 F, bottom). Quantification shows that in control cells, 45% of the mitochondria are stationary (within an observation window of 3 h); this fraction rises to 85% with tau transfection but becomes normal again (ϳ60%) with tau plus MARK2 (Fig. 6G).

Tau causes the loss of ATP in neurons
To ensure the functioning of cell processes, mitochondria must be able to move to the sites of high ATP consumption (Hollenbeck and Saxton, 2005). Because mitochondria become immobile after tau transfection, we suspected a deficit in ATP supply and therefore checked ATP levels of cells. In the case of untransfected cell cultures, if the medium is replaced by Ringer solution for 2.5 h to shut down ATP production through glycolysis, the ATP concentration drops to 50%, indicating that mitochondrial and glycolytic ATP synthesis each contribute about equally to ATP production (a typical observation for neuronal cultures) (Fig. 7A). When the cell culture is transfected with tau for 3 d, the total ATP level drops to ϳ40%. The transfection rate is around 60%; therefore, the data can be explained by assuming that the ATP production of transfected cells drops to a low value, so that the observed ATP levels are mostly generated by the untransfected cell fraction. When glycolysis is blocked (in Ringer solution) the ATP level drops to ϳ20% in tau-transfected cultures, again compatible with the assumption that the 40% untransfected cells generate ATP at half speed (by mitochondria only). As control, if cultures are treated with antimycin (an inhibitor of mitochondrial respiration) plus Ringer solution, ATP levels become negligible within 2.5 h. In contrast, in tau-transfected culture the reduction of ATP levels (to ϳ40%) can be fully rescued by cotransfection with MARK2 (Fig. 7A). This argues that the traffic inhibition by tau has a direct effect on cellular ATP levels and can be prevented by keeping traffic paths open via phosphorylation of tau.
Next, we compared the effects of tau transfection and ATP reduction on synaptic proteins. Figure 7B shows the example of drebrin, an F-actin binding protein which plays a role in spine morphogenesis. After tau transfection for 2-3 d, drebrin disappears from the spines and a small fraction clusters together with F-actin in the dendritic shaft (Fig. 7D). The displacement of dre-brin and loss of spines can be prevented by cotransfection of tau plus MARK2 up to 3 d after expressing tau. Importantly, the changes in synaptic proteins are closely correlated with the level of ATP, illustrated in Figure 7B for the cases of drebrin, AMPAR, NMDAR, PSD95, and the presynaptic protein bassoon. Treatment of the cultures with Ringer solution for 2.5 h (to suppress glycolysis) has little effect on protein levels, but treatment with antimycin (to inhibit mitochondria) in addition to suppression of glycolysis leads to a marked reduction of drebrin and bassoon within 2.5 h (AMPAR, PSD95, and NMDAR have a slower response) (Fig. 7B, left). A comparable strong reduction of drebrin is achieved by expression of tau for 3 d, whereas expression of tau plus MARK2 leaves the protein near its control levels (Fig. 7B,  right).

Microtubule bundling, traffic inhibition, mitochondrial and synaptic decay
To elucidate the nature of the synaptic decay, we performed ultrathin sections of the transfected cells for electron microscopy Figure 5. Presynaptic and postsynaptic proteins are reduced by tau expression but preserved by transfection of tau plus MARK2. A, Western blots showing that postsynaptic proteins neuroligin and PSD95 are reduced in tau-expressing cultures and that loss is prevented by double transfection of tau plus MARK2. The PSD95 antibody (MA1-045) detects a ϳ95 kDa protein and a slightly larger species from rat brain extracts. Actin is shown as a control. B, C, Postsynaptic receptors NMDAR1, AMPAR, and the presynaptic proteins bassoon and piccolo are reduced in tau-expressing cultures during a time course of 2-3 d and rescued with double transfection of tau plus MARK2. Synaptophysin shows no overall change after tau transfection but accumulates in the cell body (Fig.  4 D). D, Densitometric analysis of the Western blots for synaptic proteins (neuroligin, PSD95, drebrin, AMPAR, NMDAR1, piccolo, bassoon), all normalized to ␤-actin. The first bracket shows the reduction of protein levels after tau expression, compared with untransfected controls (n ϭ number of culture preparations). The second set of parentheses shows the rescue in tau plus MARK2 transfected cultures: neuroligin (27 Ϯ 3%; n ϭ 10; *p Ͻ 0.001) (116 Ϯ 23%; n ϭ 7; *p Ͻ 0.05), PSD95 (17 Ϯ 7%; n ϭ 4; *p Ͻ 0.001) (88 Ϯ 4%; n ϭ 3; *p Ͻ 0.03), drebrin (30 Ϯ 9%; n ϭ 4; *p Ͻ 0.001) (115 Ϯ 7%; n ϭ 2; *p Ͻ 0.02), AMPAR (32 Ϯ 8%; n ϭ 5; *p Ͻ 0.001) (84 Ϯ 8%; n ϭ 5; *p Ͻ 0.005), NMDAR1 (26 Ϯ 9%; n ϭ 4; *p Ͻ 0.001) (77 Ϯ 9%; n ϭ 3; p Ͼ 0.05), piccolo (22 Ϯ 13%; n ϭ 2; *p Ͻ 0.001) (73 Ϯ 18%; n ϭ 2; p Ͼ 0.05), bassoon (26 Ϯ 10%; n ϭ 3; *p Ͻ 0.001) (77 Ϯ 16%; n ϭ 3; p Ͼ 0.05). E, F, Apoptotic markers are not significantly increased in tau-transfected neurons. E, Control cultures (25 div) stained for the markers annexin, active caspase, and DAPI. An increase of fragmented nuclei in tau-transfected neurons after 3 and 5 d of transfection is seen but no increase in annexin or active caspase. Doubly transfected neurons (tau plus MARK2) display no increase in these markers after 3 d, but a strong increase in staining after 5 d of transfection. F, PARP is downregulated in tau transfected neurons, but not degraded. In contrast, in doubly transfected neurons (tau plus MARK2), both the intact 116 kDa form and 85 kDa fragment of PARP are seen after 4 d of transfection. The control shows a cell extract of Jurkat cells induced to undergo Fas-mediated apoptosis by treatment with anti-human Fas mAb, clone DX2 (BD Biosciences, Franklin Lakes, NJ). Error bars indicate SE. (Fig. 8). The boxed areas (Fig. 8 E, J,O,P) show fluorescence images of the cells from which the sections were derived. Figure  8 A-D corresponds to untransfected cells, E-I to tau-transfection for 2 d, J-N to tau-transfection for 4 d, and O-T to transfection with tau plus MARK2 for 2 d. The control cells (left) show the typical appearance of microtubules running along the length of the dendrite, separated by a wide distance (typically ϳ100 -150 nm center-to-center, or ϳ75-125 nm wall-to-wall, sufficient for the passing of vesicles and organelles). Cross sections reveal an irregular lattice with a density of ϳ100 MT/m 2 , in good agreement with the literature (Bartlett and Banker, 1984;Baas et al., 1991;Chen, et al., 1992) (Fig. 8 D, bottom). The mitochondria have the typical healthy appearance with well developed cristae and are positioned throughout the volume of the dendrites (data not shown). This picture changes dramatically when tau is ex-pressed for ϳ2 d (Fig. 8 E-I ). Microtubules begin to form bundles in the interior of the dendrites, gradually become more numerous, and eventually fill a large fraction of the dendritic volume in a densely packed manner (fourfold compaction, ϳ400 MT/ m 2 , center-to-center spacing decreasing to ϳ50 -75 nm, leaving only a distance of ϳ25-50 nm between walls). Mitochondria still have a healthy appearance but are pushed outside the microtubule bundles and to the walls of the dendrites, as if the bundles were too dense for entry (Fig. 8F). Even before the point of bundle formation, the movements of mitochondria are already strongly reduced, as judged by comparing live cell movement assays with ultrathin sectioning of the same areas. At later time points (4 d, where cell processes begin to degenerate), the microtubules have mostly disintegrated (Fig. 8 J-N ) and mitochondria are swollen and have lost their internal structure, typical of their decay. By remarkable contrast, transfection of tau plus MARK2 prevents the appearance of microtubule bundles, the dislocation of mitochondria to the edges, and the subsequent disintegration of mitochondria and microtubules (Fig. 8O-T ).
How could tau trigger the dramatic fourfold increase in microtubules, which corresponds to a local tubulin concentration of Ͼ2 mM? To check this, we determined the tubulin concentration in the cell population. It increases approximately twofold after 2 d of tau transfection (data not shown). Because only ϳ50% of the cells are transfected, the increase per transfected cell is approximately fourfold, in good agreement with the observed increase in local density of microtubules. The additional generation of tubulin is consistent with the observation that tubulin synthesis is autoregulated via the stability of its mRNA (Cleveland et al., 1983). Thus, stabilization of microtubules (by tau in this case) lowers the pool of unpolymerized tubulin, and this in turn enhances the rate of tubulin production.

Discussion
Loss of synapses represents an early change in AD (Terry et al., 1991;Coleman and Yao, 2003). This occurs in brain regions (entorhinal region, hippocampus) characterized also by anomalous changes in tau protein, and the progression of tau pathology closely parallels the clinical stages of the disease [Braak stages (Braak and Braak, 1991)]. The abnormal aggregation of tau in dystrophic neurites or neurofibrillary tangles is the most conspicuous change; this is preceded by more subtle changes, such as missorting of tau into the somatodendritic compartment, enhanced phosphorylation, and pathological conformation (Delacourte and Buee, 2000). The toxic effect of tau on neurons is frequently ascribed to its aggregated state, however, it is unclear how the early synaptic decay could be caused by aggregation,  (prox, proximal; dist, distal) (horizontal arrow, 10 m; vertical arrow, 1 min). In neurons transfected with tau (middle), mitochondria show almost no movement (time point was chosen at 2 d after transfection to visualize mitochondria before they disappear). Doubly transfected neurons (bottom) display saltatory movements in both directions, comparable with control. G, Quantification of stationary mitochondria (immobile Ͼ10 min), and anterograde and retrograde movements of mitochondria in dendrites. Transfection of tau for 2 d increases the fraction of immobile mitochondria to ϳ100%, but movement is partially rescued in doubly transfected cells. Conversely, the mobile fraction decreases strongly after tau-transfection in both directions (middle and left), but is rescued by tau plus MARK2 transfection. (Number of mitochondria in three different neurons each: control, n static ϭ 18, n antero ϭ 12, n retro ϭ 9; Tau, n static ϭ 25, n antero ϭ 2, n retro ϭ 2; MARK2 plus tau, n static ϭ 22, n antero ϭ 7, n retro ϭ 9.) Scale bars: (in C, E) A-E, 10 m. Error bars indicate SE.
which occurs much later. One possibility is the exposure of neurons to toxic oligomeric forms of the A␤ peptide, another might be some preaggregated oligomeric form of tau that is not well defined at present. In contrast, nonaggregated tau has usually been considered beneficial because it stabilizes microtubules (Garcia and Cleveland, 2001). Our previous results have suggested, however, that seemingly normal tau can cause cellular damage because it can inhibit microtubule-based transport (Stamer et al., 2002). This would be particularly pertinent for the extended axons and dendrites of neurons, which depend on efficient transport (Stokin and Goldstein, 2006).
We therefore set out to test the link between tau and synaptic decay. As a model system, we chose primary hippocampal neurons cultured for 25 d when numerous synapses have developed. We report here that elevation of tau indeed leads to missorting into the somatodendritic compartment and decay of synapses, thus recapitulating two early events in AD brains. What are the molecular events linking tau expression and synaptic decay? To develop a working model we note the following features: (1) the tau-induced decay affects nearly all presynaptic and postsynaptic markers tested, including the F-actin cytoskeleton which retracts from the dendritic spines and accumulates in shafts and cell bodies. The effect is therefore systemic and not specific for some interaction between tau and a synaptic protein.
(2) Even at low levels of expression, tau penetrates into the somatodendritic compartment, showing that the normal sorting mechanism breaks down. (3) Transport processes are slowed down by tau. Because tau is also missorted, the inhibition of traffic affects all cell extensions, axons, and dendrites. (4) The phosphorylation of tau at the KXGS motifs of the repeat domain, which regulates its microtubule affinity, increases with tau expression (recapitulating similar events in AD). It indicates an altered balance of a signaling cascade, which involves MARK2 or a related kinase and lowers tau-microtubule interactions. However, this phosphorylation does not suffice to prevent the tau-induced traffic blockade. (5) Elevation of tau in dendrites and axons leads to an approximately fourfold increase of microtubules concomitant with MT bundling, swelling, and degeneration of mitochondria. (6) Tau causes a drastic decrease in ATP levels, which suffices to explain the decay of the highly ATP-dependent synapses. This effect can be mimicked by poisoning mitochondria. (7) Surprisingly, the tau-induced damage can be temporarily prevented or rescued by MARK2/Par-1. Because one of the main known functions of MARK2 is to phosphorylate the repeat domain of tau and detach it from microtubules, the protective effect of MARK2 in-dicates that tau-induced damage is closely related to taumicrotubule binding.
These observations can be explained by the combination of tau's ability to inhibit intracellular transport and its missorting at elevated concentration. The mechanisms for tau's sorting into axons in normal neurons and its missorting in pathological conditions are poorly understood. Tau moves by slow axonal transport (Mercken et al., 1995). Several hypotheses for sorting have been tested (axonal sorting of tau mRNA, preferential degradation in dendrites, selective stabilization or destabilization of tau by phosphorylation) (Hirokawa et al., 1996;Aronov et al., 2001;Kosik and Krichevsky, 2002); however, for our purposes it suffices to note that the capacity of the cell for axonal sorting is limited and therefore ceases to function when tau becomes elevated. As a result, tau, invades axons as well as dendrites and their Figure 7. ATP decrease caused by tau transfection leads to a decrease in synaptic proteins. A, ATP levels in tau transfected and tau plus MARK2-transfected cultures. After 3 d of tau transfection, ATP decreases by one-half (50 Ϯ 7%; n ϭ 5; *p Ͻ 0.001), but remains high in tau plus MARK2 doubly transfected cultures (114 Ϯ 7%; n ϭ 3; *p Ͻ 0.05). To assess the amount of ATP synthesized by glycolysis, cultures were treated for 2.5 h with Ringer solution before ATP measurements. ATP levels of untreated cultures were reduced by one-half (57 Ϯ 4%; n ϭ 8; *p Ͻ 0.001), indicating that glycolysis and mitochondrial synthesis contribute about equally to the ATP production. Tau-transfected cultures in Ringer show an additional decrease by one-half. B, Energy deprivation with 1 M antimycin (AM) and Ringer for 2.5 h cause a strong protein downregulation of drebrin, bassoon and a slight downregulation of NMDAR1, but not of PSD95 and AMPAR in Western blots. Tau expression also results in a reduction of drebrin after 3 d of transfection. Double transfection with tau and MARK2 prevents this loss. C, D, Immunostaining of drebrin in cultures transfected for3d(D) shows a strong reduction of drebrin staining compared with control (C). Drebrin staining disappears from the spines and often colocalizes with F-actin in the dendritic shaft (arrow). Scale bar, 10 m. Error bars indicate SE. level of MARK2, which phosphorylates tau and thus neutralizes its influence on microtubules.
In the context of our results, the previous study of Shahani et al. (2006) is of particular interest. These authors prepared hippocampal slices from mice and transfected them by Sindbis virus with tau or pseudophosphorylated tau (several Ser/Thr residues replaced by Glu). Although transfection with tau was well tolerated by the neurons, pseudophosphorylated tau variants caused apoptotic degeneration without major changes in spine density or morphology. Thus, their results differ in several aspects from ours. The most likely explanation is the difference in experimental system, for example, pure neuronal cultures without glia in our case versus a neuronal hippocampal tissue with a mixture of cell types maintaining their connectivities (Shahani et al., 2006), high tau expression in our case versus low tau expression, or use of different vectors. The comparison illustrates that the cellular environment exerts a major influence on the challenge of tau expression. In particular, the problem of an energy deficit, which causes the nonapoptic degeneration in our cell cultures, could be compensated in the multicellular system of a slice culture.
As concluding remarks we note that mitochondria, apart from generating chemical energy, contribute to calcium homeostasis, mediate apoptosis, and are sources of reactive oxygen species. These factors could contribute to synaptic decay, in addition to the rundown of ATP. It is even conceivable that tau binds directly to mitochondria and inhibits their functions, in analogy with the effects of SOD1 (superoxide dismutase 1) on mitochondria in amyotrophic lateral sclerosis (Liu et al., 2004). We also note that mutations in the tau gene cause dementias (e.g., FTDP-17 frontotemporal dementia with parkinsonism linked to chromosome 17). A puzzling feature is that the mutations can cause disease simply by shifting the tau isoform balance toward more fourrepeat (4R) isoforms, even without changing tau protein (Lee et al., 2001). This is difficult to explain by loss of microtubules, because 4R-tau stabilizes microtubules better than 3R-tau. However, the important point may be that 4R-tau inhibits transport more efficiently and could therefore contribute to premature synapse loss, long before other events such as hyperphosphorylation and aggregation become prominent.