Common antigenic motif recognized by human VH5-51/VL4-1 tau antibodies with distinct functionalities

Misfolding and aggregation of tau protein are closely associated with the onset and progression of Alzheimer’s Disease (AD). By interrogating IgG+ memory B cells from asymptomatic donors with tau peptides, we have identified two somatically mutated VH5-51/VL4-1 antibodies. One of these, CBTAU-27.1, binds to the aggregation motif in the R3 repeat domain and blocks the aggregation of tau into paired helical filaments (PHFs) by sequestering monomeric tau. The other, CBTAU-28.1, binds to the N-terminal insert region and inhibits the spreading of tau seeds and mediates the uptake of tau aggregates into microglia by binding PHFs. Crystal structures revealed that the combination of VH5-51 and VL4-1 recognizes a common Pro-Xn-Lys motif driven by germline-encoded hotspot interactions while the specificity and thereby functionality of the antibodies are defined by the CDR3 regions. Affinity improvement led to improvement in functionality, identifying their epitopes as new targets for therapy and prevention of AD.


INTRODUCTION
Intracellular neurofibrillary tangles (NFTs) consisting of aggregated tau protein are a hallmark of Alzheimer's disease (AD) and other neurogenerative disorders, collectively referred to as tauopathies (Lee et al, 2001). Tau is a microtubule-associated protein expressed predominantly in neuronal axons and promotes the assembly and stability of microtubules (Cleveland et al, 1977b;Weingarten et al, 1975). It is expressed in the adult human brain as six isoforms with zero, one or two N-terminal acidic inserts (0N, 1N, or 2N) and either three or four microtubule-binding repeats (3R or 4R) (Goedert et al, 1989). The tau protein contains many potential phosphorylation sites and the regulated phosphorylation and dephosphorylation of several of these has been shown to affect its interaction with tubulin and cytoskeleton function (Billingsley & Kincaid, 1997;Sanchez et al, 2000). Hyperphosphorylation of tau is thought to lead to microtubule dissociation and assembly of the normally disordered, highly soluble protein into β sheet-rich aggregated fibrils called paired helical filaments (PHFs) that make up NFTs (Cleveland et al, 1977a;Lee et al, 1991;Mandelkow et al, 2007). While the molecular mechanism of tau aggregation remains elusive, it is believed that its initial nucleation step is energetically unfavorable, whereas the subsequent fibril growth follows an energetically downhill landscape (Apetri et al, 2005;Crespo et al, 2012;Harper & Lansbury, 1997;Surewicz et al, 2006). Accumulating evidence indicates that these fibrils can transmit from cell to cell and spread tau pathology to distant brain regions by seeding the recruitment of soluble tau into de novo aggregates (Frost et al, 2009;Guo & Lee, 2011;Guo et al, 2016;Kfoury et al, 2012;Wu et al, 2013).
Several monoclonal antibodies that inhibit the spreading of tau fibrils have been described and are being developed for antibody-based therapies (Bright et al, 2015;Umeda et al, 2015;Yanamandra et al, 2013;Yanamandra et al, 2017). We have recently described the isolation of a panel of antibodies with such functional activity by interrogating the peripheral IgG + memory B cells of healthy human blood donors for reactivity to phosphorylated tau peptides (Pascual et al, 2017). To expand the arsenal of potential targets and include epitopes present in physiological tau, we used the BSelex technology in combination with a pool of unphosphorylated tau peptides as baits (Fig. 1A, and table S1 for peptide sequences)

Identification of naturally occurring anti-tau antibodies in healthy donors.
In total, 2.6 x 10 6 memory B cells from nine healthy blood donors aged 18-65 years were interrogated and 92 taureactive B cells were sorted. From 30 of these, antibody heavy and light variable chain sequences were recovered and full-length IgGs were cloned and expressed. Two unique tau binding antibodies, CBTAU-27.1 and CBTAU-28.1, which are both derived from the V H 5-51 and V L 4-1 germline families, were identified. Both antibodies carry high numbers of somatic mutations with 38 and 28 nucleotide substitutions for the heavy and light chains of and 19 and 16 for the heavy and light chains of CBTAU-28.1, respectively. Since memory B cell selections were performed using a peptide pool, an ELISA-based binding assay with the 10 individual tau peptides was performed and established that CBTAU-27.1 and CBTAU-28.1 bound to peptide A6897 (residues 299-369) and peptide A6940 (residues 42-103), respectively.
8 mutations improved the affinity by more than 50-fold to the low nanomolar range ( Fig. 2C and   fig. S2). For CBTAU-28.1, analysis of the structure did not reveal potential affinity-improving mutations and, therefore, a random mutagenesis strategy was employed (Fig. 2B). This approach led to the identification of Ser 32 Arg and Glu 35  Lys mutations in the light chain that combined led to an ~4-fold improvement in affinity compared to the parental antibody ( Fig. 2D and fig. S2). Co-crystal structures of the Fab fragments of the CBTAU-27.1 double mutant Ser 27D  Tyr / Thr 94  Ile (from here on referred to as dmCBTAU-27.1) and the CBTAU-28.1 double mutant Ser 32  Arg / Glu 35  Lys (from here on referred to as dmCBTAU-28.1) in complex with peptides A8119 and A7731, respectively, were determined at 3.0 and 2.85 Å resolution (table S3) The tau specificity of the antibodies was further assessed by immunohistochemical staining on post-mortem control and AD brain tissue (Fig. 3). CBTAU-27.1 did not show immunoreactivity in either control or AD cases, whereas dmCBTAU-27.1 showed immunoreactivity in the cytosol of neurons of the control cases and clear recognition of aggregated tau in neuropil threads and NFTs in AD brains, but only after heat pretreatment Heat pretreatment using sodium citrate buffer is a routine 'antigen retrieval' procedure to recover reactivity in formalin-fixed paraffin-embedded tissue sections. No immunoreactivity of CBTAU-28.1 was detected in the control cases, whereas dmCBTAU-28.1 showed diffuse immunoreactivity of neurons after heat pretreatment. In AD brains, both CBTAU-28.1 and dmCBTAU-28.1 recognized neuropil threads and NFTs regardless of the sample treatment. CBTAU-28.1 thus recognizes PHFs without heat pretreatment whereas CBTAU-27.1, even in its high affinity variant, requires heat pretreatment to recognize pathologic tau. This is in line with the epitope of CBTAU-27.1 being buried within the PHFs and becoming exposed upon heat pretreatment. The diffuse neuronal immunoreactivity of dmCBTAU-27.1 and dmCBTAU-28.1 observed in control brain tissue after heat pretreatment shows that these antibodies bind to physiological tau. The observed immunoreactivity under identical conditions shows a clear improvement in the detection of tau by affinity-improved antibodies relative to parental antibodies without affecting specificity (Fig. 3).
Binding domain-dependent functional activities. The observation that the epitope of CBTAU-27.1 forms part of the core of the PHFs led us to consider that CBTAU-27.1 might be able to prevent aggregation of tau by inhibiting the initial nucleation step. While the molecular mechanism of tau aggregation is not fully understood, the current paradigm is that it follows a nucleation-dependent polymerization (NDP) process (Apetri et al, 2005;Crespo et al, 2012;Harper & Lansbury, 1997;Surewicz et al, 2006). An NDP mechanism is characterized by an initial nucleation step (nuclei formation) followed by an exponential growth step (fibril elongation). Nucleation, the rate-limiting step of the aggregation process, is a stochastic phenomenon and refers to the formation of high energy nuclei. Once nuclei are formed, they rapidly recruit tau monomer (growth step), and convert into thermodynamically stable aggregates. These aggregates can undergo fragmentation generating more fibril ends that are capable of recruiting tau monomers and converting them into de novo fibrils. This process is in most general terms referred to as "seeding". To assess whether CBTAU-27.1 can interfere with the aggregation of tau, we have developed a robust and highly reproducible in vitro assay that monitors the heparin-induced aggregation of full-length recombinant Tau441 (rtau) by thioflavin T (ThT) fluorescence. The aggregation behavior of tau in our assay fulfills the expected features of an NDP: sigmoidal kinetic curves with a well-defined lag phase followed by exponential growth ending in a stationary phase ( fig. S4). The aggregation kinetics of tau were highly reproducible and displayed the expected concentration dependence ( fig. S4B). Furthermore, the obtained tau aggregates displayed PHF-like morphology as assessed by atomic force microscopy (AFM) (fig. S4C) and were extremely efficient in seeding de novo aggregation of tau ( fig. S4D and E).
In agreement with our hypothesis, CBTAU-27.1 inhibited tau aggregation, as reflected by longer lag phases and lower final ThT fluorescence signal in the kinetic curves. This inhibitory effect was strongly enhanced after affinity improvement (Fig. 4, A and B, and fig. S5). To shed further light on the mechanism, we assessed the ability of dmCBTAU-27.1 to alter the tau conversion when added at later times after initiating the aggregation. In all cases, our results show that stoichiometric amounts of dmCBTAU-27.1 can not only prevent tau aggregation, but also arrest it even in the exponential phase where significant amounts of seeds are already present, presumably by binding and sequestering monomeric tau ( fig. S9). The hypothesis that CBTAU-27.1 targets monomeric tau and does not interact with tau aggregates was further confirmed by sedimentation experiments followed by SEC-MALS size determination, which indicates that the antibody binds to two tau monomers using its two Fab arms while not being able to co-sediment with preformed tau aggregates (fig. S10). In sum, these results confirm the initial hypothesis that an antibody that targets the key PHF interface of the monomeric tau can block its misfolding and aggregation.
Both double mutant and parental CBTAU-28.1 showed comparable alterations in the aggregation kinetics (Fig. 4, E and F, and fig. S6). While somewhat longer lag phases and lower end-point fluorescent signals were observed in the presence of antibody, the effect did not seem to be dose dependent and the kinetics were strikingly irreproducible. Furthermore, visual inspection of reaction mixtures after 120 h incubation revealed that CBTAU-28. We next assessed the ability of the antibodies to bind tau aggregates and thus potentially block the propagation and spreading of tau pathology. We therefore incubated human AD brain homogenate containing PHFs with the antibodies and depleted the antibody-antigen complexes.
The residual seeding capacity was assessed using a cell-based biosensor assay (Holmes et al, 2014;Yanamandra et al, 2013). In line with its inability to bind PHFs, CBTAU-27.1 did not reduce the seeding activity of the AD brain homogenate, while dmCBTAU-27.1 showed only minor reduction and only at the highest concentration tested (Fig. 5A). In contrast, CBTAU-28.1, like mouse anti-PHF antibody AT8, depletes seeding activity from AD brain homogenate and this in vitro activity is enhanced for dmCBTAU-28.1 (Fig. 5B).
The observation that CBTAU-28.1 and dmCBTAU-28 can bind PHFs led us to explore the possibility that these antibodies may furthermore enhance the uptake of tau aggregates by microglia, the resident macrophage cells of central nervous system (Ginhoux et al, 2013).
Indeed, CBTAU-28.1 and dmCBTAU-28.1 promoted the uptake of aggregated rtau into mouse microglial BV2 cells and the affinity-improved antibody appeared to mediate tau uptake to a greater extent (Fig. 6A). The fact that Fabs of both parental and dmCBTAU-28.1 did not increase basal tau uptake indicates that the uptake is Fc mediated. As expected, CBTAU-27.1 and dmCBTAU-27.1 did not show activity in this assay (Fig. 6C). Antibody-mediated tau uptake and localization of rtau aggregates in the endolysosomal compartment by CBTAU-28.1 and dmCBTAU-28.1, but not CBTAU-27.1 and dmCBTAU-27.1, was confirmed by confocal microscopy (Fig.6, B and D, and fig. S12).

DISCUSSION
Implications for therapy and vaccines. By binding to the region that is critical for the aggregation of tau and which forms the core of PHFs, CBTAU-27.1 prevents aggregation of tau in vitro. This functional activity identifies its epitope as a potential target for immunotherapy and could perhaps allow earlier intervention than antibodies that inhibit the spreading of already formed tau seeds (Bright et al, 2015;Umeda et al, 2015;Yanamandra et al, 2013;Yanamandra et al, 2017). Evidence that interfering with tau aggregation through immunotherapy may be possible is provided by murine antibody DC8E8 which also targets monomeric tau, inhibits tau aggregation in vitro, and reduces tau pathology in a murine AD model (Kontsekova et al, 2014).
An alternative approach could be the development of a small molecule drug targeting monomeric tau that interacts with the CBTAU-27.1 epitope. This epitope does not overlap with the microtubule binding domains (Fig. 1F), suggesting that such a drug may not interfere with the normal function of physiological tau while preventing its aggregation. Furthermore, the epitope of CBTAU-27.1 is specific for tau, and is not present on MAP2, a microtubule stabilizing protein closely related to tau (Dehmelt & Halpain, 2005). Like previously described antibodies, CBTAU-28.1 may be able to inhibit the spread of tau pathology. In addition, it demonstrated an Fc-dependent enhancement of the uptake of tau aggregates by microglial BV2 cells. It is probably able to enhance the uptake of aggregates because its epitope is distant from the core of the PHFs and remains accessible. Both antibodies are derived from the V H 5-51 and V L 4-1 germline families and bind their respective epitopes through hotspot interactions that are remarkably alike, pointing towards a conserved structural motif in tau that could not have been predicted from sequence analysis alone. The motif containing proline and lysine separated by 4 to 7 amino acids, is commonly found in Tau, and appears nine times on 2N4R. The lack of somatic mutations around the hotspot proline and lysine suggests these two key tau residues could be responsible for the initial recognition of the V H 5-51 and V L 4-1 germline combination.
The same V H 5-51 or V L 4-1 hotspot interactions can be separately found in other crystallized antibody complexes (Chen, 2017;Gorny et al, 2011;Liao et al, 2013;Ofek et al, 2014), but the combination of the two germlines could be more prevalent against tau, as they seem to get triggered by the Pro -X n -Lys motif abundantly present on the protein. Identification and characterization of these antibodies may thus be exploited to develop antibody or drug regimens for distinct phases of progression of tau pathology and pave the way towards a peptide-based tau vaccine, by taking advantage of the apparent immunogenicity of the identified motif and presenting both hotspot residues in the right spacing and orientation.

Human PBMC isolation
Whole blood from healthy male and female donors was obtained from the San Diego Blood Bank (ages 18-65 years) after informed consent was obtained from the donors. PBMCs were isolated on Ficoll-Paque Plus (GE Healthcare) and cryopreserved at 50 million cells per ml in 90% FBS and 10% DMSO.

Single cell sorting and recovery of heavy and light variable light antibody genes from tauspecific memory B cells
Tau-peptide specific B cells were sorted and antibody chains were recovered as previously described (Pascual et al, 2017). Briefly, biotinylated peptides were mixed with streptavidin-APC or streptavidin-PE (Thermo Fisher) at a 35:1 molar ratio and free peptide removed using BioSpin 30 column (Biorad). PBMCs corresponding to 3 donors were thawed and B cells were enriched by positive selection with CD22 + magnetic beads (Miltenyi Biotec). B cells were subsequently labeled with extracellular markers IgG-FITC, CD19-PerCPCy5.5 and CD27-PECy7 (all from BD Biosciences) and incubated with labeled peptide baits. Doublets and dead cells were excluded and the CD19 + , IgG + , CD27hi, antigen double-positive cells were collected by single cell sorting into PCR plates containing RT-PCR reaction buffer and RNaseOUT (Thermo Fisher). Heavy and light chain (HC/LC) antibody variable regions were recovered using a twostep PCR approach from single sorted memory B cells using a pool of leader specific (Step I) and framework specific primers (Step II PCR). Heavy and light chain PCR fragments (380-400 kb) were linked via an overlap extension PCR and subsequently cloned into a dual-CMV-based human IgG1 mammalian expression vector (generated in house). Finally, all recovered antibodies were expressed and tested by ELISA against the tau peptides used in the sorting experiments to identify tau specific antibodies.

Peptide synthesis
All peptides were synthesized at Pepscan BV (Lelystad, The Netherlands) or New England Peptide, Inc. using routine Fmoc-based solid phase strategies on automated synthesizers. After acidic cleavage and deprotection, peptides were purified by reversed phase HPLC, lyophilized and stored as powders at -80 o C. Purity and identity were confirmed by LC-MS.

Recombinant IgG expression and purification
Human IgG1 antibodies were constructed by cloning the heavy (V H ) and light (V L ) chain variable regions into a single expression vector containing the wildtype IgG constant regions.
Plasmids encoding the sequences corresponding to human anti-tau mAbs were transiently transfected in human embryonic kidney 293-derived Expi293F TM cells (Thermo Fisher) and 7 days post transfection, the expressed antibodies were purified from the culture medium by MabSelect SuRe (GE Healthcare) Protein A affinity chromatography. IgGs were eluted from the column with 100 mM sodium citrate buffer, pH 3.5 which was immediately buffer exchanged into PBS, pH 7.4 using a self-packed Sephadex G-25 column (GE Healthcare). Mouse chimeric versions of CBTAU-27.1 and CBTAU-28.1 were generated cloning the variable light and heavy chain into an expression vector in which the human Fc was replaced with murine Fc. Fab fragments were obtained by papain digestion of antibodies (Thermo Fisher Scientific), followed by removal of the Fc using MabSelect SuRe resin (GE Healthcare). Each antibody was quality controlled by SDS-PAGE and size exclusion chromatography coupled with multi angle light scattering (SEC-MALS) and was further confirmed for reactivity to cognate tau peptide by Octet biolayer interferometry.

Recombinant tau (rtau) expression and purification
The gene encoding the human tau-441 isoform (2N4R) extended with an N-terminal His-tag and a C-terminal C-tag and containing the C291A and C322A mutations was cloned into a kanamycin resistant bacterial expression vector and transformed into BL21 (DE3) cells. A 3 ml 2-YT Broth (Invitrogen) preculture containing 25 mg/L kanamycin was inoculated from a single bacterial colony for 6 hours after which it was diluted to 300ml for overnight growth in a 1-liter shaker flask and subsequently diluted to 5 liters in a 10-liter wave bag. Protein expression was induced when the culture reached an OD 600nm 1.0 by addition of 2 mM IPTG. Three hours after induction, the bacterial pellets were harvested by centrifugation, and lysed with Bugbuster protein extraction reagent (Merck) supplemented with a protease inhibitor cocktail (cOmplete TM ULTRA tablets EDTA free, Roche). Purification was performed by affinity chromatography using self-packed Ni-Sepharose (GE Healthcare) and C-Tag resin (Thermo Fisher Scientific) columns.

Reactivity of anti-tau human mAbs to tau peptides by ELISA
Reactivity of recovered mAbs were tested against biotinylated tau peptides as previously described (Pascual et al, 2017). Briefly, tau peptides were captured on streptavidin-coated plates (Thermo Fisher Scientific) at 1 µg/ml in TBS and incubated for 2 h. Goat anti-human Fab (2 µg/ml, Jackson ImmunoResearch) to measure total IgG was used, and bovine actin (1 µg/ml TBS, Sigma) and irrelevant peptide were used to confirm specificity of the purified mAbs.
ELISA plates were blocked and purified IgGs were diluted to 5 µg/ml in TBS/0.25% BSA and titrated (5-fold serial dilutions) against peptides. Plates were subsequently washed and goat anti-human IgG F(ab')2 (Jackson ImmunoResearch) was used at 1:2000 dilution as secondary.

Following incubation, plates were washed four times in TBS-T and developed with Sure Blue
Reserve TMB Microwell Peroxidase Substrate (KPL) for approximately 2 min. The reaction was stopped by the addition of TMB Stop Solution and the absorbance at 450 nm was measured using an ELISA plate reader.

Qualitative association and dissociation measurements by Octet biolayer interferometry
The relative binding of the antibodies to tau peptides was assessed by biolayer interferometry antibody. To induce dissociation, the biosensor containing the antibody-tau peptide complex was immersed in assay buffer without antibody. The immobilization of peptides to sensors, the association and the dissociation steps, were followed in different ionic strength buffers containing 10 % FortéBio kinetics buffer as assay buffer. The relative association and dissociation kinetic curves were compared to qualitatively assess the efficiency of antibody binding to peptides encompassing different tau epitopes.

Affinity measurements by Isothermal Titration Calorimetry (ITC)
The affinities of antibodies for their corresponding tau peptides were determined in solution on a Before data collection, the crystals of CBTAU-27.1 Fab and CBTAU-28.1 Fab were soaked in the reservoir solution supplemented with 25% (v/v) and 15% (v/v) glycerol, respectively, for a few seconds and then flash-frozen in liquid nitrogen. X-ray diffraction data of CBTAU-27.1 Fab apo-form were collected to 1.9 Å resolution at beamline 23ID-D at the Advanced Photon Source (APS). X-ray diffraction data of CBTAU-27.1 Fab with 2833-1 and CBTAU-28.1 with W1805 were collected to 2.0 and 2.1 Å resolution, respectively, at beamline 12-2 at the Stanford Synchrotron Radiation Lightsource (SSRL). HKL2000 (HKL Research, Inc.) was used to integrate and scale the diffraction data (Table S2). Electron density for both peptides 2833-1 and W1805 were clear and the peptides were built in the later stages of the refinement. Final refinement statistics are summarized in Table S2. The dmCBTAU-27.1 diffraction data were indexed in space group C2 and the dmCBTAU-28.1 data in space group P2 1 2 1 2 1 . Data were processed using the programs XDS and XSCALE (see table S3). The phase information necessary to determine and analyze the structure was obtained by molecular replacement using previously solved structure of dmCBTAU-27.1 and dmCBTAU-28.1 Fabs were used search models. Subsequent model building and refinement was performed according to standard protocols with the software packages CCP4 and COOT. For calculation of R-free, 6.2 % of the measured reflections were excluded from refinement (see table S3). TLS refinement (using REFMAC5, CCP4) resulted in lower R-factors and higher quality of the electron density map for dmCBTAU-27.1-A8119 using five TLS groups. Waters were added using the "Find waters" algorithm of COOT into Fo-Fc maps contoured at 3.0 sigma followed by refinement with REFMAC5 and checking all waters with the validation tool of COOT. The occupancy of side chains, which were in negative peaks in the Fo-Fc map (contoured at -3.0 σ), were set to zero and subsequently to 0.5 if a positive peak occurred after the next refinement cycle. The Ramachandran plot of the final model shows 88.7% (dmCBTAU-27.1) and 88.1% (dmCBTAU-28.1) of all residues in the most favored region, 10.8% (dmCBTAU-27.1) and 11.6% (dmCBTAU-28.1) in the additionally allowed region, and 0.3% (dmCBTAU-27.1) and 0.0% (dmCBTAU-28.1) in the generously allowed region. Residue Ala51(L) is found in the disallowed region of the Ramachandran plot (table S3) for both Fabs, as is frequently found in other Fabs (Arevalo et al, 1993) . This was either confirmed from the electron density map or could not be modelled in another more favorable conformation. Statistics of the final structure and the refinement process are listed in table S3.

Affinity maturation by phage display
The coding sequence for scFv directed against CBTAU-28.1 epitope was cloned into an inducible prokaryotic expression vector containing the phage M13 pIII gene. Random mutations were introduced in the scFv by error prone PCR (Genemorph II EZClone Domain Mutagenesis kit) after which the DNA was transformed into TG1bacteria. The transformants were grown to mid-log phase and infected with CT helper phages that have a genome lacking the infectivity domains N1 and N2 of protein pIII, rendering phage particles which are only infective if they display the scFv linked to the full-length pIII (Kramer et al, 2003). Phage libraries were screened using magnetic beads coated with rtau in immunotubes. To deselect nonspecific binders, the tubes were coated with a tau peptide lacking the CBTAU-28.1 epitope. To ensure maturation against the correct epitope, selection was continued using beads coated with the cognate A6940 peptide. Eluted phages were used to infect XL1-blue F' E. coli XL1-blue F' which were cultured and infected with CT-helper (or VCSM13) phages to rescue phages used for subsequent selection rounds. After three rounds of panning, individual phage clones were isolated and screened in phage ELISA for binding to rtau and cognate CBTAU-28.1 peptide A6940.

Atomic Force Microscopy
For each sample, 20 μl of rtau solution was deposited onto freshly cleaved mica surface. After 3 minutes incubation, the surface was washed with double-distilled water and dried with air.
Samples were imaged using the Scanasyst-air protocol using a MultiMode 8-HR and Scanasystair silicon cantilevers (Bruker Corporation, Santa Barbara, USA). Height images of 1024x1024 pixels in size and surface areas of 10x10 µm were acquired under ambient environmental conditions with peak force frequency of 2 KHz.

Assessment of CBTAU-27.1 binding to rtau and PHFs by SEC-MALS
15 μM monomer or aggregated rtau were incubated with dmCBTAU-27.1 in a rtau:IgG1 = 1:0.6 ratio for 15 minutes and samples were subsequently centrifuged for 15 minutes at 20,000 g.

Immunohistochemistry
Brain samples were obtained from The Netherlands Brain Bank (NBB), Netherlands Institute for Neuroscience, Amsterdam. All donors had given written informed consent for brain autopsy and the use of material and clinical information for research purposes. Neuropathological diagnosis was assessed using histochemical stains (haematoxylin and eosin, Bodian and/or Gallyas silver stains (Uchihara, 2007) and immunohistochemistry for Amyloid beta, p-tau (AT8), α-synuclein, TDP-43 and P62, on formalin-fixed and paraffin-embedded tissue from different parts of the brain, including the frontal cortex (F2), temporal pole cortex, parietal cortex (superior and inferior lobule), occipital pole cortex, amygdala and the hippocampus, essentially CA1 and entorhinal area of the parahippocampal gyrus. Staging of pathology was assessed according to Braak and Braak for tau pathology and Thal for Amyloid beta pathology (Braak & Braak, 1991;Thal et al, 2002). For the Gallyas silver staining, 30 µm thick sections were rinsed in distilled water and incubated in 5% periodic acid for 30 minutes at RT, followed by an incubation in silver iodide solution (4% sodium hydroxide, 10% potassium iodide and 0.35% silver nitrate in distilled water) for 30 minutes at RT. Subsequently, sections were washed in 0.5% acetic acid and developed with developer working solution (10 volumes 5% sodium carbonate solution, 3 volumes solution 0.2% ammonium nitrate, 0.2% silver nitrate and 1% Tungstosilicic acid solution, and 7 volumes 0.2% ammonium nitrate, 0.2% silver nitrate, 1% Tungstosilicic acid and 0.3% formaldehyde solution. After color development, sections were rinsed in 0.5% acetic acid, after which sections were incubated in 5% sodium thiosulphate and rinsed in distilled water. Stained sections were mounted on coated glass slides (Menzel-Gläser) and dried for at least 2 hours at 37°C.
Subsequently sections were fixed in ethanol 70% for 10 minutes, counterstained with hematoxylin, dehydrated and mounted with Quick D mounting medium.

FRET based cellular immunodepletion assay
Cryopreserved brain tissue was acquired from the Newcastle Brain Tissue Resource biobank. Frozen brain tissue samples from 17 AD patients were homogenized in homogenization buffer (10 mM Tris (Gibco), 150 mM NaCl (Gibco) containing protease inhibitors (cOmplete TM ULTRA tablets EDTA free, Roche) to obtain a 10% w/v pooled brain homogenate. Individual antibody dilutions were prepared in PBS pH 7.4 (Sigma), mixed with brain extract in a 1:1 ratio in a 96 well PCR plate (Thermo Scientific), and incubated until the beads were washed. Protein-G DynaBeads (Life Technologies) were added in a 96-well PCR plate (Thermo Scientific) and washed twice with PBS, 0.01% Tween20 (Sigma) by pulling down the beads with a magnet (Life Technologies). Wash buffer was removed completely and 10 µl of PBS, 0.1% Tween20 were added to the beads together with 90 µl of the 1:1 antibody-brain extract mixture. Samples were incubated over night at 4°C, rotating at 5 rpm. The following day, the immunodepleted fractions were separated from the beads by pulling down the beads with the magnet, transferred to a new 96-well PCR plate and stored at -80°C until tested. Each condition was tested in duplicate. For the microscopy experiments, cells were seeded in 96-well µClear® plate (Greiner Bio-one).
After incubation with the immunocomplexes, nuclei were stained with Hoechst (Sigma) and the acidic cellular compartment with LysoTracker Red dye (Thermo Fisher). Live-cell imaging was performed using the Opera Phenix™ High Content Screening System (PerkinElmer) with temperature set to 37°C and in presence of 5% CO 2 . For high quality images, a 63x water immersion objective was used and 0.5 μm planes (20 per Z-stack) were acquired per imaged field.
Cleveland DW, Hwo SY, Kirschner MW (1977b) Purification of tau, a microtubule-associated protein that induces assembly of microtubules from purified tubulin.        is the scaled intensity of the i th measurement of reflection h, k, l, < I hkl > is the average intensity for that reflection, and N is the redundancy. c R pim = Σhkl (1/(n-1))1/2 Σi| Ihkl,i -<Ihkl> | / Σhkl Σi Ihkl,i, where n is the redundancy. d CC 1/2 = Pearson Correlation Coefficient between two random half datasets. e No. molecules for complexes refers to number of Fab molecules per asymmetric unit (ASU).
and F c are the observed and calculated structure factors, respectively. g R free was calculated as for R cryst , but with 5% of data excluded before refinement.  is the scaled intensity of the i th measurement of reflection h, k, l, < I hkl > is the average intensity for that reflection, and N is the redundancy. c R pim = Σhkl (1/(n-1))1/2 Σi| Ihkl,i -<Ihkl> | / Σhkl Σi Ihkl,i, where n is the redundancy. d CC 1/2 = Pearson Correlation Coefficient between two random half datasets. e No. molecules for complexes refers to number of Fab molecules per asymmetric unit (ASU).
and F c are the observed and calculated structure factors, respectively. g R free was calculated as for R cryst , but with 6% of data excluded before refinement.   with tau, binding of this antibody was significantly affected by ionic strength of the buffer and also observed for its affinity-improved mutant. For specificity, association and dissociation kinetics for the interactions of the antibodies with a panel of different tau peptides were determined. Lack of significant binding of the affinity-improved mutants to tau peptides to which the parental antibodies did not bind, and to additional peptides not containing the epitopes of the parental antibodies, indicate that the specificity was retained.