Tau Antibody Structure Reveals a Molecular Switch Defining a Pathological Conformation of the Tau Protein

Tau antibodies have shown therapeutic potential for Alzheimer’s disease and several are in clinical trials. As a microtubule-associated protein, tau relies on dynamic phosphorylation for its normal functions. In tauopathies, it becomes hyperphosphorylated and aggregates into toxic assemblies, which collectively lead to neurodegeneration. Of the phospho-epitopes, the region around Ser396 has received particular attention because of its prominence and stability in tauopathies. Here we report the first structure of a monoclonal tau antibody in complex with the pathologically important phospho-Ser396 residue. Its binding region reveals tau residues Tyr394 to phospho-Ser396 stabilized in a β-strand conformation that is coordinated by a phospho-specific antigen binding site. These details highlight a molecular switch that defines this prominent conformation of tau and ways to target it. Overall, the structure of the antibody-antigen complex clarifies why certain phosphorylation sites in tau are more closely linked to neurodegeneration than others.

Animals. rTg4510, tTA and non-Tg littermate F1 mice were bred at Taconic, Denmark. Mice expressing the tTA activator transgenes were maintained on 129S6 background strain (Taconic) and mutant tau responder mice were maintained in the FVB/NCrl background strain (Taconic). Mice were screened by PCR using the following primer pairs 5′-GATTAACAGCGCATTAGAGCTG-3′ and 5′-GCATATGATCAATTCAAGGCCGATAAG -3′ for the tTA activator transgene and 5′-TGAACCAGGATGGCTGAGCC-3′ and 5′-TTGTCATCGCTTCCA GTCCCCG-3′ for the mutant tau responder transgene. rTg4510, tTA and non-Tg littermates were group-housed (5 animals/cage) and received water and food ad libitum (Brogaarden, Denmark) as well as environmental enrichment. The light/dark cycle was 12 hours; room temperature was 21 ± 2 °C and a relative humidity of 55% ± 5%. All animal experiments were performed in accordance with the European Communities Council Directive #86/609, the directives of the Danish National Committee on Animal Research Ethics, and Danish legislation on experimental animals (license no. 2014-15-0201-00339).
Tissue collection. Mice were euthanized by cervical dislocation. Brains were quickly removed, weighed and divided sagittally down the midline to yield two hemispheres. One hemisphere was parted from the cerebellum and hindbrain, the forebrain containing cerebral cortex and hippocampus was snap-frozen on dry ice and stored at −80 °C until use for biochemical analysis. Frontal cortices from 4 AD donors (2 males and 2 females) and 4 non-demented healthy controls (HC, 2 males and 2 females) were obtained from Cambridge Bioscience (Cambridge, UK). The brain tissue samples were fresh-frozen and stored at −80 °C until used for biochemical analysis. All donors were of Caucasian race. Age of AD donors: 76-91 years. Age of the HC donors: 57-80 years. Post-mortem delays: 3-8 hours.
Tau Biochemistry. Brain tissue extraction was performed as described previously 32 . Tissues were homogenized in 10 volumes of Tris-buffered saline containing protease and phosphatase inhibitors as follows: 50 mM Tris/HCl (pH 7.4); 274 mM NaCl; 5 mM KCl; 1% protease inhibitor mixture (Roche); 1% phosphatase inhibitor cocktail I & II (Sigma); and 1 mM phenylmethylsulfonyl fluoride (PMSF; Sigma). The homogenates were centrifuged at 27,000 × g for 20 min at 4 °C to obtain supernatant (S1) and pellet fractions. Pellets were re-homogenized in 5 volumes of high salt/sucrose buffer (0.8 M NaCl, 10% sucrose, 10 mM Tris/HCl, [pH 7.4], 1 mM EGTA, 1 mM PMSF) and centrifuged as above. The supernatants were collected and incubated with sarkosyl (1% final concentration; Sigma) for one hour at 37 °C, followed by centrifugation at 150,000 × g for one hour at 4 °C to obtain sarkosyl-insoluble pellets, referred to as P3. The P3 pellet was re-suspended in TE buffer ( Western blot. Samples were boiled in 1 × LDS loading buffer and 100 mM DTT. A volume corresponding to 0.5 µl S1 and P3 isolated from 24 weeks old rTg4510 and non-transgenic littermate (non-tg) mice or 5 μl S1 and P3 isolated from four pooled Alzheimer's disease (AD) and healthy control (HC) brains was loaded on a 4-12% Bis-Tris NuPAGE Gel (LifeTech Novex). After electrophoresis, the proteins were blotted over to a Immobilon-FL PVDF membrane (0.45 μm, IPFL10100, Millipore). The membrane was blocked with SEA blocking buffer (Product #37527, Thermo Fisher). Tau and pTau levels were assessed in the samples using 1 μg/mL C5.2 overnight at 4 °C. A secondary fluorophore conjugated IgG antibody was used (IRDye 680 Goat anti-mouse, LICOR biosciences) and the signal was quantified using Odyssey CLx and Image studio software (LI-COR biosciences).
Dot blot. 1 μl sample of S1 and P3 isolated from 24 weeks old rTg4510 and non-transgenic littermate (non-tg) mice or from four pooled Alzheimer's disease (AD) and healthy control (HC) brains was spotted on nitrocellulose membranes. S1 and P3 isolated from mice was used at a 1:10 dilution. After blocking with a blocking solution containing 5% nonfat milk and 0.1% Triton-X100 in TBS, the membranes were incubated with 1 µg/mL C5.2 overnight at 4 °C. Membranes were washed, and incubated with a peroxidase-conjugated anti-mouse antibody (1:5000; Jackson ImmunoResearch, West Grove, PA). Bound antibodies were detected using an enhanced chemiluminescence system (ECL PLUS kit; Perkin Elmer).
Immunohistochemistry. Paraffin sections of 4 μm thickness were prepared from blocks of PFA-fixed prefrontal cortex from AD donors and non-demented controls obtained from Tissue Solutions (Glasgow, UK). The sections were deparaffinised, rehydrated and subjected to antigen retrieval by boiling in 10 mM citrate buffer pH 6 for 5 min. Endogenous peroxidase was quenched for 10 min with 1% H 2 O 2 in PBS. Sections were then incubated 20 min in blocking buffer followed by overnight incubation in a humidified chamber at 4 °C with primary antibody at 0.1 to 1 μg/mL in incubation buffer. Between each incubation step, the sections were washed 3 × 10 min in wash buffer. Subsequently, the sections were incubated at room temperature in the following solutions; 60 min with a biotinylated secondary Ab diluted 1:200 in wash buffer; 60 min in ABC; 20 min in DAB. Finally, the sections were counterstained with hematoxylin and coverslipped.
Enzyme-linked Immunosorbent Assay (ELISA). Immulon 4 HBX (extra high-binding) 96-well microtiter plates (Thermo Scientific) were coated with the same peptide used for crystallization diluted to 2 μg/mL in phosphate-buffered saline (PBS) and left overnight at 4 °C (Table 1). Plates were then washed three times in PBS-T (PBS containing 0.05% Tween-20) and blocked in 5% non-fat milk and 3% bovine serum albumin in PBS for 2 hours at room temperature. Plates were subsequently washed 3 times in PBS-T before C5.2 was serially diluted (8 μg/mL-0.05 ng/mL) in PBS and incubated for 1 hour at room temperature. Crystallization, data collection, and structure determination. Concentrated Fab C5.2 was mixed with the peptide pS396/pS404 at a 1:10 molar ratio. Crystallization conditions were screened and optimized using the vapor diffusion hanging drop method. Well-diffracted crystals were obtained in a solution containing 25.5% polyethylene glycol 4000, 0.17 M ammonium sulfate and 15% glycerol. X-ray diffraction data were collected at 100 K at a wavelength of 0.979 Å at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 14-1. The data set to 2.09 Å was processed using the XDS software package 33 and the structure determined by molecular replacement using an initial model with high sequence similarity ( Table 2). Multiple steps of refinement were carried out in COOT 34 and PHENIX 35 . The final structure analysis was carried out in ICM 36

C5.2 binding to pathological tau protein.
To complement and extend the ELISA/MSD screening for antibody binding to pathological tau fractions (see Methods), we characterized C5.2 binding by Western and dot blots in brain samples isolated from tau transgenic mice (rTg4510 mice overexpressing human tau with P301L mutation) and non-transgenic littermates, as well as from Alzheimer's disease patients and healthy control subjects. Further characterization was conducted by immunohistochemistry in brain sections from Alzheimer's disease patients and healthy control subjects. As shown on representative images, C5.2 is specific for pathological tau protein (Fig. 1a,b). It binds predominantly to an insoluble tau fraction (P3) but also has reactivity to a soluble tau fraction (S1) both from rTg4510 mouse and Alzheimer's disease brains. In the latter, C5.2 recognizes similar bands in both fractions (Fig. 1a). Normal tau in the soluble tau fraction (S1) from non-transgenic littermates and healthy control subjects is not detected with C5.2. On Alzheimer's disease brain sections, it binds strongly to tangles, pre-tangles and dystrophic neurites, while it does not bind to normal tau in brain sections from healthy control subjects (Fig. 1b).

C5.2 pSer396 phospho-specificity.
In order to evaluate the ability of C5.2 to exclusively target the phosphorylated epitope, we performed an enzyme-linked immunosorbent assay (ELISA). The peptides used in the assay are the same peptides used for crystallization and diluted to a concentration of 2 μg/mL (Table 1). Binding curves were generated from ELISA data by maintaining plate-bound peptide concentrations and serially diluting C5.2 IgG (Fig. 1c). They revealed C5.2's strong association with pS396/pS404 and pS396 peptides but lack of recognition of pS404 or the non-phosphorylated peptide, suggesting that C5.2 is a pS396 phosphorylation state-specific mAb.
Nanomolar K D values obtained using fluorescence polarization. To accurately measure the C5.2 binding affinity, we performed fluorescence polarization experiments to determine dissociation constants (K D ), using the Fab C5.2 and four differentially phosphorylated FITC-conjugated peptides ( Table 1). The peptide concentration was kept constant while that of Fab C5.2 was serially diluted from 675 nM to 0.08 nM (Fig. 1d)  Crystal structure of Fab C5.2 in complex with a phosphorylated Tau epitope. To define the C5.2 epitope, we determined the co-crystal structure of Fab C5.2 in complex with the pS396/pS404 di-phosphorylated tau peptide ( 386 TDHGAEIVYKpSPVVSGDTpSPRHL 408 ) and refined the structure to a resolution of 2.1 Å with an R work and R free of 21.8% and 27%, respectively ( Fig. 2 and Table 2). Although the peptide used in the crystallization is 23 residues in length, only 7 were visible in the electron density map ( P392 IVYKpSPV P398 ). In addition, an orphan phosphate is present near the C-terminal end of the peptide. We attempted to co-crystalize Fab C5.2 with all four peptides from Table 1 but given the low affinity for half of the peptides and the number of crystallization conditions screened, we were only able to resolve one co-crystal structure. The crystals grew in monoclinic space group C2 with three Fab:peptide complexes in the asymmetric unit. The three complexes are highly similar with an average root mean square deviation (RMSD) of 0.2 Å so we only describe one of the complexes hereafter. For clarity, the antibody residues in the text are numbered according to the Kabat convention 38 , with the light and heavy chains labeled with "L" and "H", but colored cyan and green in the figures, respectively.
Overall antibody-antigen interactions. C5.2 makes direct contacts with 5 tau residues, including Ile 392 , Val 393 , Tyr 394 , pSer 396 and Pro 397 . Although Lys 395 and Val 398 were also observed in the electron density, they do not form direct contacts with the antibody. The backbone of the epitope region of tau lays straight on top of a relatively shallow (~5 Å deep) antigen-binding groove formed by both heavy chain and light chain CDR regions, with the N-terminus of the epitope located at the heavy chain side and the C-terminus at the light chain side (Fig. 2). The contacts within the antibody-antigen (Ab-Ag) interface involve hydrogen bonds, salt bridges, π-stacking and hydrophobic forces. The negatively charged side chain of pSer396 is located in an electropositive, phosphate-binding pocket coordinated by C5.2 CDR-H residues Arg H94 , Arg H95 and Arg H32 (Fig. 2d). These features allow C5.2 to grip the peptide like a claw and facilitate the extended peptide conformation with CDR-H and CDR-L on either side. Although the peptide makes contact with both chains, a majority of them are made with the heavy chain (Fig. 3). We have also observed in the electron density an orphan phosphate close to the C-terminus of the epitope (Fig. 2). S1 rTg4510 P3 non-tg P3 rTg4510 S1 non-tg

CDR-L3 coordinates pTau main-chain interactions. CDR-L3 is the only light chain loop that makes
contact with the tau peptide. The backbone hydroxyl of Thr L92 forms a hydrogen bond with the backbone amine of Tyr 394 . The backbone amine of Leu L94 forms a hydrogen bond with the backbone hydroxyl of Ile 392 . In addition, Leu L94 and Tyr L93 contribute to a hydrophobic environment that facilitates the aliphatic epitope residues Val 393 and Ile 392 (Fig. 3a). Lastly, His L91 and Phe L96 , along with heavy chain residues Tyr H50 and His H35 , contribute to a network of aromatic residues in the paratope to provide a stacking interaction with the epitope. The interactions provided by CDR-L3 are main-chain to main-chain and water-mediated hydrogen bonds.
Tyrosine coordinates a majority of the side-chain interaction. C5.2 contains four tyrosine residues while the epitope contributes a fifth in the antibody-antigen interactions. Tyr H50 and Tyr 394 form a T-shaped stacking interaction that orients the side-chain hydroxyl of Tyr 394 towards the phosphate of pSer 396 (Fig. 3a,c). Tyr 394 of the epitope is then able to form a hydrogen bond with pSer 396 and Thr H33 , which further stabilizes the orientation of the epitope. Tyr H52 stacks with Pro 397 and contributes to overall aromatic stacking forces. Tyr L93 provides noncovalent contacts and facilitates the aliphatic N-terminal peptide residues. Overall, Tyr 394 , accounting for about 40% of the buried surface area of the epitope, is involved in the most interactions contacting four CDR loops and forming an intramolecular hydrogen bond with pSer 396 (Fig. 3a,c). Lastly, of interest, is a platform of the aromatic residues Tyr 394 , His L91 , Phe L96 , His H35 and Tyr H50 that stack to suitably mount the peptide for the C5.2 recognition. In addition, Tyr H27 is an essential component of the network fostering the orphan phosphate (Fig. 3b). The orphan phosphate ion is a buffer ion and does not suggest additional phospho-peptide interactions with pSer404.
Phosphorylation-specific antibody interactions. C5.2 is a PSSA that binds the hyperphosphorylated residue Ser 396 . The key feature of pSer 396 recognition is the turn of the CDR-H3 loop that contains a stretch of residues that coordinate a majority of the phospho-specificity. Arg H95 provides a charge-stabilizing salt bridge with a hydroxyl group on pSer 396 (Fig. 3a). The backbone amine of Thr H97 forms a hydrogen bond with another of the hydroxyl group. In between the two residues, Gly H96 is tactically located at CDR-H3 hairpin turn to coordinate this nest formation of contacts for pSer 396 recognition. Gly H96 also contributes a hydrogen bond to pSer 396 . While CDR-H3 provides pSer 396 phospho-specificity, CDR-H1 is responsible for recognition of the orphan phosphate (Fig. 3b). Polar residues Tyr H27 and Asp H31 provide noncovalent contacts while the backbone amine of Thr H28 forms a hydrogen bond with a phosphate hydroxyl. Finally, the side chain of Arg H32 forms a salt bridge to complete the nest of interactions orchestrating the orphan phospho-specificity. Local conformation of tau peptide. Pathological neurofibrillary tangles and tau proto-filaments are characterized by a high content of β-sheet in a cross-β/β-helix structure 39 . Under normal un-phosphorylated conditions, the tau 393-399 sequence will adopt an α-helical conformation due to the high contents of the β-branched amino-acid valine 40 . Phosphorylation of Ser 396 , like phosphorylation of Ser 199 with a similar YXpSP motif, can stabilize the secondary structure in a β-strand conformation, forming a seed structure, which may induce tangle filament formation (Supplementary Figs S1 and S2) 41 . In the antibody-tau peptide complex structure, the backbone conformation is locked through the Tyr 394 -pSer 396 sidechain interaction; the -OH group of Tyr 394 is only ~2.5 Å away from one of the phosphate oxygens. The dihedral backbone angles (φ, ψ) Tyr 394 (−78.7, 131.4), Lys 395 (−126.2, 130.1), and pSer 396 (−56.2, 131.0) are fixed in a β-strand conformation through the strong Tyr 394 -pSer 396 sidechain interaction. The structure of tau peptide ( 392 IVYK 395 ) in the C5.2 complex is found to closely resemble (Cα RMSD = 0.2 Å) the core of the tau proto-filament located in the microtubule binding region (MTB) R3 ( 306 VQIVYK 311 ) that has been recently described as having higher β-strand and amyloidogenic propensity ( Supplementary Fig. S2) [42][43][44] . However, the orientation of the side chain of Tyr 394 in the C5.2 complex is flipped from that in the β-strand so that Tyr 394 can interact with pSer 396 . We propose that the YXSP motif in tau constitutes a specific conformational switch which upon serine phosphorylation may change local protein backbone from α-helix to β-strand conformation. This structural phenomenon explains the specificity of this antibody for pathological tau protein.

Discussion
We have provided the first crystal structure of a mAb targeting the pSer 396 Tau epitope. The C5.2 Fab:peptide complex structure reveals the peptide in an extended β-strand conformation ranging from tau residues Ile392 to Val398 ( 392 IVYKpSPV 398 ) and pSer396 serving as a major specificity determinant (Fig. 2). Western blot, immunohistochemistry, ELISA and fluorescence polarization analyses confirm C5.2 to be: (1) specific for pathological tau protein, and; (2) phosphorylation state-specific with nanomolar affinity for the pS396/pS404 and pS396 peptides (K D : 15.4 nm and 14.9 nm, respectively) (Fig. 1). Previous studies have described similar pSer396 anti-pTau antibodies 45,46 . The first study characterized the mAbs PHF-1, PHF-13 and PHF-47, which bind di-phosphorylated peptides containing Ser396 phosphorylation. All three mAbs bound pSer396 but did not bind to the non-phosphorylated peptide or to pSer400, pThr403, or pSer404 phosphorylated peptides without the presence of pSer396 phosphorylation 46 . Another study described this pattern of recognition as primary phospho-residue recognition, in this case pSer396, followed by a weak, and possibly non-essential, recognition of a nearby phosphorylated residue 45 . The specificity of pSer396 antibodies for tau pathology was already described in these reports 47,48 , but the structural reason for this specificity has remained unexplained until now. There have been numerous reports of anti-pTau antibodies but only a few have been accompanied by structures at atomic resolution. The structure of the first anti-pTau mAb was published in 2012 and provided a preliminary structural analysis of an avian single chain Fv (scFv), generated using phage display methods 49 , bound to a pT231/pT235 Tau peptide 50 . It gave a comprehensive analysis of phospho-specific antibody recognition but avian antibodies are uncommon and possess a structurally unusual binding pattern. In 2016, the co-crystal structure was described of the AT8 mAb in a complex with a pS202/pT205/pS208 Tau peptide 51 . Although the AT8 epitope contains three phospho-residues, several key residues, including AT8 Gly H99 , Ser H100 , Tyr L39 and His L96 , provide the contacts for pThr205 recognition which are similar to the network of C5.2 contacts to pSer396 (Fig. 4a,b).
Our analysis of the antigen binding of C5.2 is an advancement in the field of PSSA that may lead to insights in phospho-serine recognition of PSSA in general. A recent report used phage display to develop PSSAs and designed scFv libraries starting with parent scaffolds of known physiological phospho-residue binding motifs 52 . Their analysis of contacts identified three major regions of recognition, the anchor residue (Ser H52 or Thr H52A ), the conformation stabilizing residue (Gly H54 ) and the more diverse residues conferring side-chain specificity (Arg H53 , Ser H55 and Ser/Thr H56 ). Combining this analysis of pSer specific mAb and that of our C5.2 complex structure, we begin to see a pattern in phospho-serine recognition. In the previous pSer specific scFv, Ser H52 , Arg H53 , Gly H54 and Ser H55 form hydrogen bonds to pSer while Thr H52A anchors the peptide (Fig. 4c). Instead of CDR-H2, C5.2 uses CDR-H3 for phospho-recognition but still generates a similar anion-binding nest for the phospho-serine (Fig. 3).
More interestingly, C5.2 binds the pTau epitope with a similar four-fingered grip as well-known physiological signaling proteins that function through recognition of post-translational phosphorylation. For example, 14-3-3 proteins are a family of dimeric α-helical proteins that recognize two different sequence motifs, R[S/Ar] XpSXP and RX[Ar/S]XpSXP (pS and Ar symbolizing phosphoserine and aromatic, respectively). The key feature of phospho-recognition is a positively charged pocket formed by the residues Lys 49 , Arg 56 , Arg 127 and Tyr 128 that provide the 'four-fingered grip' (Fig. 4d). The location of proline +2 residues to the C-terminal of pSer plays the important role of kinking the epitope in order to redirect the protein out of the binding interface and allow pSer to extend inward. The WW module is another example of phosphorylation-dependent signaling component that recognizes short proline-rich motifs of PPXY, PPLP and PPR. Intriguingly, the WW domain plays a significant role in tau biochemistry. Pin1, a regulator of cell-cycle progression, contains a WW domain that catalyzes a phosphorylation-dependent prolyl-peptide bond isomerization between pSer/pThr-Pro to render it available for dephosphorylation by the phosphatase PP2A 53,54 . A crystal structure of Pin1 in complex with a di-phosphorylated RNA Pol II CTD hexapeptide (YpSPTpSPS) identifies Ser 16 , Arg 17 and Tyr 23 as key contact residues 55 (Fig. 4e). The four-fingered grip is represented by C5.2 residues Arg H94 , Arg H95 , Gly H96 , and Thr H97 (Fig. 4a). Not only do these structures emulate the use of an electropositive binding pocket for phospho-recognition but they give insight into the importance of Pro 397 flanking pSer396 for epitope recognition and the high frequency of Arg, Tyr, Ser, Thr, His, Lys and Gly residues in the paratope.
The C5.2 structure suggests that the pSer396 phosphate is stabilized by the adjacent Tyr394. The consequence is that the YXpSP becomes a poorer substrate for PP2A. The result will be that phosphorylation at the pSer396 position will accumulate during cellular stress seen in neurodegenerative diseases such as Alzheimer's disease. Interestingly, Tau pSer199/pSer202 is another phosphorylation site ( 196 GYSSPGS 202 ) that is frequently associated with PHF pathology in Alzheimer's disease, which possesses the same sequence motif (YXpSP) as the pSer396 region. This suggests that pSer199 region can also be stabilized in a β-strand (tangle or proto-filament inducing conformation) backbone structure. Solution phase NMR 56 and solid-state NMR studies of full-length recombinant 2N4R tau 57 suggested that the preferred secondary structure of the repeat units in the microtubule binding domains (MTB) is β-strand. A recent cryo-EM structure of the MTB domain of tau 39 suggests that the pathological structure of tangle proto-filaments (tangle monomer) is a cross-β/β-helix unit. The pSer199 and pSer396 YXpSP motifs are flanking both sides of the MTB domains. We thus postulate that the YXpSP motif constitutes a key conformation, which is stabilized in a β-strand through phosphorylation. Specifically, the YXSP → YXpSP phosphorylation reaction amounts to an α-helix to β-strand switch which may change overall backbone conformation of the microtubule domains. Blocking this motif with tau antibodies may have therapeutic benefits as supported by prior work by us and others [15][16][17][18][19][20][21][22][23][24][25][26][27][28][29]58,59 .
PSSAs are extremely valuable tools for studying dynamic signaling events that require rapid assembly and disassembly of signaling complexes regulated by protein phosphorylation. This includes the microtubule stabilizing function of the tau protein that is disrupted in AD. Using PSSAs to target abnormal phosphorylation has the potential to serve as both a safe therapeutic and a much-needed diagnostic biomarker. These are particularly exciting times for pTau PSSA development with several such antibodies already in clinical trials and additional ones in late stages of preclinical assessment 13 . As more anti-pTau antibodies are developed and characterized, they will provide a framework for optimization of an immunotherapy, as well as a better understanding of AD treatment and prevention.