How short peptides can disassemble ultra-stable tau fibrils extracted from Alzheimer’s disease brain by a strain-relief mechanism

Reducing fibrous aggregates of protein tau is a possible strategy for halting progression of Alzheimer’s dis-ease (AD). Previously we found that in vitro the D-peptide D-TLKIVWC disassembles tau fibrils from AD brains (AD-tau) into benign segments with no energy source present beyond ambient thermal agitation. This disassembly by a short peptide was unexpected, given that AD-tau is sufficiently stable to withstand disas-sembly in boiling SDS detergent. To consider D peptide-mediated disassembly as a potential therapeutic for AD, it is essential to understand the mechanism and energy source of the disassembly action. We find as-sembly of D-peptides into amyloid-like fibrils is essential for tau fibril disassembly. Cryo-EM and atomic force microscopy reveal that these D-peptide fibrils have a right-handed twist and embrace tau fibrils which have a left-handed twist. In binding to the AD-tau fibril, the oppositely twisted D-peptide fibril produces a strain, which is relieved by the disassembly of both fibrils. This strain-relief mechanism appears to operate in other examples of amyloid fibril disassembly and provides a new direction for the development of first-in-class therapeutics for amyloid diseases.


Introduction
Tau pathology caused by the abnormal aggregation of tau is more strongly correlated with cognitive symptoms and severity in Alzheimer's disease (AD) than Aβ plaques [1][2][3] .Thus, disrupting tau aggregates emerges as a promising therapeutic strategy for AD 4,5 .To date, several types of disaggregators of tau brils have been reported and their disaggregation mechanisms are being sought.For example, the human Hsp70/DnaJ system can disaggregate tau brils in vitro in an ATP-dependent manner, but it generates small, seeding-competent species that accelerate the progression of disease 6 .Small molecules such as methylene blue 7 and EGCG, by binding to tau brils, apparently disrupt intermolecular interactions 8 .Our recent structural studies revealed EGCG stacks in the clefts formed at the junction of the two proto laments in AD tau brils 9 .However, small molecules including methylene blue/LMTX and EGCG have not been proven to be effective drugs, perhaps because of limited bioavailability, promiscuous protein binding and low blood-brain barrier permeability 10 .
A peptide-based disaggregator can offer advantages over small molecules including higher speci city and higher binding a nity 11 .Especially, compared to L-peptides, D-enantiomeric peptides are known to be less immunogenic, less protease-sensitive in vitro and more resistant to degradation in vivo 12 .A recent phase I clinical trial of a Dpeptide that disassembles Aβ oligomers has proved to be safe and well tolerated 13 .Previously, we reported that the D-peptide D-TLKIVWC can disassemble tau brils extracted from AD brains (AD-tau), neutralizing their seeding ability and rescuing behavioral de cits in a mouse model of Alzheimer's disease 14 .However, the underlying disassembly mechanism remains unknown, preventing further development of this type of drug candidate for AD.
Here, we designed a series of peptides of sequence D-TLKIVWX varying only at the seventh residue, X.These Dpeptides showed variable e cacy in disassembling AD-tau brils in vitro, with X = Ile as the best performer.From electron microscopy, we discovered that D-TLKIVWX peptides form amyloid-like brils themselves, and from atomic force microscopy we learned that these brils have a right-handed helical twist, in contrast to the left-handed helical twist of AD-tau.To learn the molecular interactions between the brils that participate in disassembly, we determined the cryo-EM structures of D-TLKIVWX proto laments bound to tau brils of opposing twist.Combining our structural data, we propose a strain-relief mechanism for AD-tau bril disassembly.D-TLKIVWX peptides disassemble AD-tau brils to non-seeding species In previous work, we found that D-TLKIVWC can disassemble tau brils, whereas its six-residue analog D-TLKIVW cannot 14 .Here we wondered whether the disassembling ability arises from the capacity of the cysteine to form a disul de bond with tau residue Cys-322 located in the structured core of tau brils.The formation of a disul de bond could alter the conformation of tau in brils, resulting in bril breakdown 15,16 .To evaluate this hypothesis, we measured disassembly activity as a function of disul de bond forming potential which we adjusted by including either glutathione (GSH) or glutathione disul de (GSSG).As shown in Extended Data Fig. 1a, D-TLKIVWC exhibited equivalent capacity to disassemble recombinant tau K18+ (residues Gln244-Glu380 of 4R tau) brils across all tested conditions, indicating the formation of a disul de bond is irrelevant to the disassembly of tau brils.
To determine if the cysteine residue of D-TLKIVWC is essential in disassembling tau brils, we substituted the Dcysteine with various other D-amino acid residues, including anionic residues (aspartate (D) and glutamate (E)), cationic residues (arginine (R) and lysine (K)), polar residues (serine (S) and threonine (T)), hydrophobic residues (alanine (A), isoleucine (I), and valine (V)), as well as a β-sheet interrupter (proline (P)) 17 .Their performance in disassembling AD-tau brils after 48 hours of incubation was evaluated by dot blot and electron microscopy (TEM).
The dot blot experiment was conducted with the conformational antibody GT38 18 , which speci cally recognizes ADtau brils and does not probe monomeric tau or D-peptide controls (Extended Data Fig. 1b).The D-peptide variants (D-TLKIVWX, X = A, S, D, I, V, R, K, E, T) showed varying e cacy in disassembling AD-tau brils depending on the type of residue in the seventh position (Fig. 1a-b).Speci cally, hydrophobic residues (X = I, V and A) most signi cantly reduced the level of AD-tau brils, with e cacy decreasing in the following order: hydrophobic > polar (X = S and T) > cationic (X = R and K) > anionic (X = D and E).Lastly, the β-sheet interrupter (X = P) and the deletion of the seventh residue both showed no reduction in the level of AD-tau brils.This trend was consistently supported by TEM characterization, where the number of AD-tau brils (labeled by red arrows in Fig. 1c and Extended Data Fig. 1c) was quanti ed after 48 hours of disassembly (Extended Data Fig. 1d).Notably, we observed new bril species (blue arrows in Fig. 1c and Extended Data Fig. 1c).Furthermore, we observed speci city of D-TLKIVWX in disassembling AD-tau brils as it cannot disrupt other amyloid brils such as α-syn brils or wild type hnRNPA2 brils (Extended Data Fig. 2).In summary, most tested residue types X in D-TLKIVWX demonstrate speci c disassembly action against AD-tau brils, with hydrophobic residues, especially Ile, being the best.
In addition to uncovering the mechanism of disassembly, it is important to ascertain whether disassembly produces pathologic products that seed the growth of additional brils 19 .Therefore, we systematically investigated the seeding capacity of the products of AD-tau brils after overnight incubation with variants of D-TLKIVWX in a HEK293T cell line stably expressing yellow uorescent protein (YFP)-fused tau 20 .As illustrated in Fig. 1d, AD-tau brils alone can seed the aggregation of endogenous uorescent tau, leading to the formation of bright puncta (indicated as white arrows) observable under uorescence microscopy.In contrast, the overnight disassembly products of AD-tau brils treated with D-TLKIVWX gradually lost their seeding ability in a dose-dependent manner (Fig. 1d and Extended Data Fig. 3a).Further, automated image analysis of visible puncta revealed a dependence of the seeding capacity on the seventh amino acid residue within D-TLKIVWX (Extended Data Fig. 3b-f).The observed dependence is consistent with the trends observed in our in vitro disassembly assays (Fig. 1b and Extended Data Fig. 1c,d).Additionally, because of disassembly activity, D-TLKIVWX (X = I and S) shows a dose-dependent effect in reducing AD-tau toxicity in mouse Neuro 2A (N2a) cells (Fig. 1e and Extended Data Fig. 3g).Our results demonstrate that the products of AD tau brils disassembled by D-TLKIVWX are not seeding-competent and are non-toxic.

Amyloid-like bril formation of D-TLKIVWX is essential for tau disassembly
To better understand the disassembly process of tau brils, we conducted a time-course dot blot and TEM experiment using D-TLKIVWI as a representative of the more e cient bril disassemblers.Dot blot data (Extended Data Fig. 4a) con rmed that D-TLKIVWI gradually reduced the level of AD-tau brils as a function of time.The timecourse TEM images in Fig. 2a and Extended Data Fig. 4b show AD-tau brils (red arrows) appearing to be covered by unknown species after one hour of incubation with D-TLKIVWI, and additional brillar structures (blue arrows) become increasingly evident at three-and six-hour timepoints.At 24 h, amorphous products (magenta arrows) appeared concomitant with the disappearance of AD-tau brils and reduction of the unidenti ed bril species.The amorphous products appeared more dispersed in micrographs acquired at 48 h, and a western blot showed that the disassembly products of AD-tau brils consisted primarily of insoluble species; denatured pelleted material migrated as dimers and other multimers, not as monomeric tau (Extended Data Fig. 4c).Notably, in the TEM images of D-TLKIVWX (X = I, S, R, D, E, K, T, C, A and V)-treated AD-tau samples (Fig. 1c and Extended Data Fig. 1c), we also observed the emergence of new brils (labeled by blue arrows) accompanying with the disappearance of AD-tau brils.We identi ed these new species as D-TLKIVWX brils since D-TLKIVWX (X = I, S, R, D, E, K, T, C, A and V) exhibited aggregation activity in the same buffer (Extended Data Fig. 5a).In contrast, neither D-TLKIVW nor D-TLKIVWP exhibited aggregation nor disassembled AD-tau brils.These observations suggest that the ability of D-TLKIVWX to brilize aids the disassembly of AD-tau brils.
To test the hypothesis that D-TLKIVWX form amyloid-like brils that disassemble AD-tau, we designed a negative control experiment by eliminating the ability of D-TLKIVWX to brilize.As shown in Fig. 2b and Extended Data Fig. 5b,c, 10 mM D-TLKIVWX (X = I, S and R) form well-de ned brils when they are left undisturbed for three days at room temperature.The corresponding X-ray diffraction analysis (Fig. 2c and insets in Extended Data Fig. 5b,c) con rm these D-TLKIVWX (X = I, S and R) brils exhibit the characteristic features of amyloid brils with a strong 4.7-4.8Å re ection corresponding to the distance between hydrogen-bonded β-strands, and a more diffuse 8-12 Å ring arising from the inter-sheet spacing 21 .As such, the aggregation activity of D-TLKIVWX might be prevented by eliminating hydrogen bonds between neighboring β-strands through N-methylation of peptide backbones 22 .Indeed, when we N-methylated the D-isoleucine of D-TLKIVWX (named as D-TLK(N-Me-I)VWX (X = I, S and R)), the peptides were unbale to form brils, as shown in Fig. 2d and Extended Data Fig. 5d,e.TEM and dot blot experiments further showed that these non-self-aggregating peptides D-TLK(N-Me-I)VWX (X = I, S and R) were unable to disassemble AD-tau brils (Fig. 2e-g and Extended Data Fig. 5f,g).Taken together, these experiments demonstrate the critical role of amyloid-like characteristics of D-TLKIVWX in disassembling AD-tau brils.

D-TLKIVWX peptides form right-handed brils with conserved motifs
We determined the structures of D-TLKIVWX amyloid-like brils, aiming to reveal features that facilitate disassembly of AD-tau brils.Generally, β-sheets formed by L-peptides adopt left-handed helical structures, while β-sheets formed by D-peptides adopt right-handed helical structures 23,24 .Here, atomic force microscopy (AFM) con rmed the twist is right-handed in all 18 polymorphs of D-TLKIVWI brils observed 25,26 , 27 (Fig. 3a and Extended Data Fig. 6a).Using cryo-EM, we were able to determine the structures of the predominant polymorphs of D-TLKIVWX (X = I, S and R) brils (indicated by red squares in Extended Data Fig. 6b-d) at 3.6 Å, 3.5 Å, 3.7 Å resolution, respectively (Fig. 3b-d).Data collection and re nement statistics are summarized in Supplementary Table 1.
The D-TLKIVWX (X = I, S and R) brils are each composed of different numbers of proto laments and these proto laments associate in different patterns, but all the proto laments share the same underlying structural motif known as a "steric zipper" 28 , a pair of β-sheets mated together by an interface of snugly tting sidechains (Fig. 3b-d and Extended Data Fig. 7a,b).In addition, the steric zippers formed by D-TLKIVWX (X = I, S and R) all share the same symmetry pattern in which antiparallel β-sheets mate together by interfacing sidechains of Leu2, Ile4, and Trp6 (an example of "class 5" symmetry 29 ).As a result, the helical rise of D-TLKIVWX (X = I, S and R) amyloid-like brils is 9.56 Å, instead of the 4.80 Å spacing that is common among pathogenic amyloid brils.Notably, the sidechain of the seventh residue faces outward from the steric zipper interface in all cases.The identity of the seventh sidechain appears to affect the geometry of association between proto laments but does not affect the symmetry of the proto lament itself.Thus, the ability to disassemble AD-tau brils seems linked more strongly to the steric zipper symmetry (which is conserved among all three functional D-peptides), rather than the pattern of association between zippers (which differs among the three D-peptides).
Importantly, the structures suggest why the absence of a seventh residue in D-TLKIVW prohibits its bril formation, and therefore presumably its inability to disassemble tau.Removal of the seventh residue would destabilize bril formation by reducing the number of backbone hydrogen bonds (10 vs. 14) and increasing the distance between the negative charge of the C-terminal carboxylate and the compensating positive charge at the N-terminal amine of the adjacent strand (5.0 Å vs. 2.8 Å) (Extended Data Fig. 7c).

D-TLKIVWX proto laments grow along left-handed AD-tau brils
To elucidate how D-TLKIVWX might interact with AD-tau brils to induce their disassembly, we cryogenically trapped complexes of D-TLKIVWX (X =I, S and R) with AD-tau brils at an intermediate 24-h time point with a lower ratio of D-TLKIVWX to tau (estimated 100:1) in comparison with the time-course experiment in Fig. 2a (500 :1).As a negative control, we also collected images of AD-tau brils in the absence of D-TLKIVWX.Helical reconstructions of the ADtau complexed with each of the three D-peptides revealed the paired helical lament (PHF) tau polymorph 30 (Extended Data Fig. 8), and the atoms modeled into the PHF density showed no signi cant structural deviations from the negative control (Supplementary Table 2).However, the cryo-EM map of PHF complexed with D-TLKIVWX (X =I, S and R) revealed residual density near Val313-Thr319 of PHF, which was absent from our control (Fig. 4a-c and Extended Data Fig. 9a,b).We attribute the residual density to D-TLKIVWX for two reasons: (1) the shape of the residual density resembles one steric zipper unit of D-TLKIVWX brils (Fig. 4a-c and Fig. 3b-d); (2) 1 H-15 N-HSQC NMR experiments indicated that D-TLKIVWX (X =I and S) interacted with 15 N, 13 C-labeled tau K18+ monomer 31 with chemical shift perturbation mapped to Val306-Lys311 11 and Val313-Thr319 (Fig. 4d and Extended Data Fig. 9c,d), consistent with the location of additional density next to tau PHF.Note that a rapid precipitation of D-TLKIVWX occurred upon mixing in our NMR experiments, corresponding to initial D-peptide bril formation, but chemical shift changes of monomeric tau were observed over time when soluble fraction of D-peptides increased.Re nement of the 3D reconstruction of D-TLKIVWX (X =I, S and R) complexed with Tau PHF achieved overall resolutions of 3.1 Å, 3.1 Å and 3.5 Å, respectively.We note that the cryo-EM density map of D-TLKIVWI is stronger than that of D-TLKIVWS/R and that D-TLKIVWR is slightly further from the core of tau PHF (Fig. 4a-c and Extended Data Fig. 9b).This difference in occupancy and positioning may explain their unequal e ciency in disassembling AD-tau brils (Fig. 1b).

Discussion
Strain-relief of D-TLKIVWX brils drives disassembly of AD-tau brils Our observations lead us to propose the strain-relief mechanism of amyloid bril disassembly illustrated in Fig. 5.As D-TLKIVWX binds to the Tau PHF, it stacks to form D-TLKIVWX amyloid-like proto laments and these proto laments are constrained by binding to adopt the left-handed helical twist of Tau PHF (Fig. 5a,b).Because D-TLKIVWX brils have an intrinsic right-handed twist (Fig. 3), the cryo-trapped structure of Figure 4a-c is metastable, and a strain develops.If not trapped by freezing, this metastable structure disassembles in hours.
We propose that the strain produced by the metastable left-hand twist of D-TLKIVWX is relieved by the proto lament reversing its twist from a left-to right-handed helix (Fig. 5c).Because the D-peptide brils is bound to Tau PHF bril, the reversal pulls against the Tau PHF structure.As a result, the tau residues that contact D-TLKIVWX are torn away from the native PHF contacts, thereby breaking backbone hydrogen bonds between tau molecules, and permitting solvent to invade the bril core and further dissociate tau PHF (Fig. 5d).The disassembly products are non-seeding species (Fig. 1); they are tau-D-TLKIVWX complexes, not tau monomers (Extended Data Fig. 4c).This dynamic process is visually depicted in the Supplementary Video 1.
Our proposed strain-relief mechanism is consistent with all our experimental ndings.First, D-TLKIVWX disassembles tau brils to which its precursor, D-TLKIVW, was designed to bind, but does not disassemble other left-handed amyloid brils, such as α-synuclein and wildtype hnRNPA2 LCD brils.Second, other analogs of D-TLKIVWX, including D-TLKIVW, D-TLKIVWP, and D-TLK(N-Me-I)VWX (X = I, S, R) fail to disassemble AD-tau brils, presumably because they lack the ability to form brils themselves.Third, the variation in e cacy of disassembly AD-tau brils among D-TLKIVWX can be in uenced by their binding strength with AD-tau brils.Fourth, the concurrent disappearance of AD-tau brils and the newly formed D-TLKIVWX brils, along with the emergence of amorphous products at later time points observed in the time-course EM images (Fig. 2a) strongly suggests that the AD-tau brils are disassembled.
An additional observation that supports this "strain-relief" mechanism of disassembly is that L-TLKIVWX (X = C, I, S, R) all display inferior e cacy in disassembling AD-tau brils compared to their enantiomers D-TLKIVWX (Extended Data Fig. 10a,b).Despite exhibiting the same bril-forming property as D-TLKIVWX 32 (Extended Data Fig. 10c) and nearly identical binding to tau monomers (Extended Data Fig. 10d,e), the L-TLKIVWI brils principally possess a lefthanded twist, same as tau PHF.Therefore, the L-TLKIVWX would have less structural torsion to release when they bind and assemble along the axis of tau PHF in comparison with D-TLKIVWX, resulting in their inferior e cacy in disassembling AD-tau brils.
The strain-relief mechanism may be a general theme of action of disruptors of amyloid brils.In previous work, we presented evidence that the polyphenolic compound EGCG disassembles AD-tau brils by stacking into a metastable amyloid-like bril on the surface of AD-tau brils 9 .A subsequent change in which aromatic rings of EGCG curve into a more stable conformation could provide the energy to disassemble stable AD-tau brils.Thus, both disassembling actions of the very different compounds EGCG and D-TLKIVWX on AD-tau brils may be considered examples of strain-relief mechanisms.
In summary, we nd the disassembly of AD-tau brils is not exclusive to D-TLKIVWC, because D-TLKIVWX (X = A, S, D, I, V, R, K, E, T) also displays this property, but with unequal e cacy.We nd that the amyloid-like, bril-forming property of D-TLKIVWX contributes to the disassembly of AD-tau brils.Based on cryo-EM, AFM, NMR, and other results reported here, we propose that the disassembly of AD-tau brils is driven by the release of torsion in D-TLKIVWX proto laments.It remains unknown whether D-TLKIVWX disassembles tau brils from other tauopathies 33 .The strain-relief mechanism of amyloid disassembly proposed here may explain how diverse small molecules can provide su cient energy to disrupt extremely stable pathogenic amyloid brils.This mechanism may be applied to the design of a new generation of disaggregators for tau and other pathological amyloids.

Chemicals and Materials
All the peptides were synthesized by GenScript and puri ed to ≥98%, as determined by mass spectrometry and HPLC (GenScript Corp, Piscataway, NJ).

Recombinant protein expression and puri cation
Unlabeled recombinant tau K18+ (residues Gln244-Glu380 of 4R tau) was expressed in a pNG2 vector in BL21-Gold E. coli cells grown in LB to an A 600 = 0.8.Cells were induced with 0.5 mM isopropyl 1-thio-β-D-galactopyranoside (IPTG) for 3 h at 37 °C and lysed by sonication in 20 mM MES buffer (pH 6.8) with 1 mM EDTA, 1 mM MgCl 2 , 1 mM dithiothreitol (DTT), and HALT protease inhibitor before the addition of NaCl (500 mM nal concentration).The lysate was boiled for 20 min and then clari ed by centrifugation at 15,000 rpm for 15 min and dialyzed to 20 mM MES buffer (pH 6.8) with 50 mM NaCl and 5 mM DTT. Dialyzed lysate was puri ed on a 5-mL HiTrap SP HP ion exchange column and eluted over a gradient of NaCl from 50 to 550 mM.Protein was further puri ed on a HiLoad 16/600 Superdex 75 pg column (GE Healthcare) in 10 mM Tris (pH 7.6) with 100 mM NaCl and 1 mM DTT and concentrated to 20-60 mg/mL by ultra ltration using a 3-kDa cutoff lter (Millipore-Sigma, Burlington, MA).
Isotopically labeled tau K18+ proteins for NMR were grown in M9 H 2 O media supplemented with 15 NH 4 Cl (and 13 Cglucose) as the sole nitrogen (and carbon) source.Protein expression was induced with 1 mM IPTG at 37 °C for 6 hours.The puri cation was same as for the unlabeled protein.
The construct for overexpression of mCherry-hnRNPA2-LCD fusion protein was provided by Dr. Masato Kato of University of Texas, Southwestern.Protein overexpression and puri cation procedures followed the same protocol reported previously 34 .

Thio avin T (ThT) assay
Kinetic uorescence data were collected in a microplate reader (FLUOstar Omega, BMG Labtech) at 37 ˚C with double orbital shaking at 700 rpm.Fluorescence measurements were recorded every 10 mins with excitation and emission wavelengths of 440 and 480 nm.All samples were added in triplicate and experiments were repeated at least twice.

Dot blot assay
Puri ed AD-tau brils from brain extract were incubated with 500 μM L/D-TLKIVW, D-TLKIVWX (X = C, A, S, D, I, V, R, K, E, P, T), D-TLK(N-Me-I)VWX (X =I, S and R), L-TLKIVWX (X = C, I, S and R) at 37 ℃ for 48 h, respectively.2.5 μL of samples were added on nitrocellulose membrane (0.2 µm, Bio-Rad).The membrane was blocked by 5% (w/v) nonfat dry milk in TBS-T (T = 0.1% (v/v) Tween-20) at room temperature for 1 hr.After blocking, the membrane was incubated with GT38 antibody obtained from Virginia Lee's lab (1:1000) in 5% (w/v) milk in TBS-T at 4 ℃ overnight.Then, the membrane was washed in TBS-T three times for 5 minutes each and incubated with goat anti-mouse IgG HRP (1:5000, cat# AB205719, Abcam) in TBS-T for 1 h at room temperature.The membrane was washed three more times, and the signal was developed with Pierce TM ECL western blotting substrate (170-5061, BioRad).
Negative stain transmission electron microscopy (TEM) 6 μL of sample was applied to a glow discharged carbon coated electron microscopy grid (CF150-Cu, Electron Microscopy Sciences) for 5 minutes.Then grids were stained with 2% uranyl acetate for 2 minutes.Samples were visualized using a FEI Tecnai T12 Quick room temperature transmission electron microscope equipped with a Gatan 2,048 x 2,048 CCD camera operated at an acceleration voltage of 120 kV.
AD brain tau bril seeding in tau biosensor cell line HEK293 cell lines stably expressing tau-K18-YFP were engineered by Marc Diamond's laboratory at the University of Texas Southwestern Medical Center and used without further characterization or authentication.Cells were maintained in Dulbecco's modi ed Eagle's medium (DMEM, Life Technologies, cat# 11965092) supplemented with 10% (v/v) fetal bovine serum (Life Technologies, cat# A3160401), 1% antibiotic-antimycotic (Life Technologies, cat# 15140122), and 1% Glutamax (Life Technologies, cat# 35050061) at 37 °C, 5% CO 2 in a humidi ed incubator.AD-tau brils were incubated with D-TLKIVWX (X = A, S, D, I, V, R, K, E, P, T) (2.5, 5, 10, 20, 50, 75, 100 μM) at 4 °C overnight and sonicated in a cup horn water bath for 3 min.Then these disassembly products of AD-tau were mixed with 1 volume of Lipofectamine 3000 prepared by diluting 1μL of Lipofectamine 3000 (Life Technologies, cat# 2729899) in 19 μL Opti-MEM (Life Technologies, cat# 31985070).After 20 min, 10 μL of brils were added to 90 μL tau biosensor cells.After 24 hours of incubation, the number of seeded aggregates was determined by imaging the entire well of a 96-well plate in triplicate using a Celigo image cytometer (Nexcelom) in the YFP channel.The data analysis was described before.For high-quality images, cells were photographed on a ZEISS Axio Observer D1 uorescence microscope using the EGFP uorescence channel.
Cell toxicity of AD-tau disassembly species AD-tau bril disassembly species were produced by incubating AD-tau brils (estimated 1 μM) with D-TLKIVWX (5, 10, 20, 50, or 100 μM) at 4 °C overnight.Neuro 2A (N2a) cells were cultured in DMEM supplemented with 10% (v/v) fetal bovine serum, 1% antibiotic-antimycotic, and 1% Glutamax in a 5% CO 2 humidi ed environment at 37 °C.Cells were plated at a density of roughly 6,000 cells/well on 96-well plates in 90 μL of fresh medium.After 24 h, 10 μL of the above AD-tau bril disassembly species were added and the cells were incubated for another 24 h at 37 °C.Cytotoxicity was measured utilizing an MTT assay.
X-ray diffraction (XRD) D-TLKIVWX (X = I, S, R) peptides were dissolved to 10 mM in deionized water and incubated at room temperature quiescently for three days.Peptide brils were aligned by pipetting the suspension in a 3 mm gap between two repolished glass rods and drying overnight.The aligned brils were cooled to 100 K. Diffraction data was collected on a FR-E+ rotating anode x-ray generator (Rigaku) equipped with a R-AXIS HTC imaging plate detector (Rigaku).Cu K-α x-ray beam with 1.5406 Å wavelength was used, and the detector was placed 78 mm from the sample.Diffraction images were visualized using ADXV (The Scripps Research Institute).
Atomic force microscopy (AFM) 4 mM D-TLKIVWI in deionized water was shaken at room temperature for three days and diluted into distilled water in a 1:10 ratio.Then, 5 μL of diluted sample was deposited onto freshly cleaved mica and incubated for 10 min.The sample was rinsed with Milli-Q water and dried under a stream of nitrogen gas.AFM images were collected using a Dimension Icon microscope (Bruker) in PeakForce Tapping mode using ScanAsyst-HR probes.Each collected image had a scan size of 3 x 3 μm and 2048 x 2048 pixels and was collected using a scan rate of 0.494 Hz.Nanoscope Analysis software (Version 2.0, Bruker) was used to process the image data by attening the height topology data to remove tilt and scanner bow.Fibrils were traced and computationally straightened from collected AFM images in Matlab using Trace_y 35 .
Cryo-EM samples D-TLKIVWI brils were optimized by shaking 4 mM D-TLKIVWI in deionized water at room temperature for three days.10 mM D-TLKIVWS/R in deionized water formed brils when left undisturbed for three days at room temperature.For D-TLKIVWR brils, the pH of the peptide solution was adjusted to 7.0.Prior to cryo-EM grid preparation, AD-tau brils in a buffer comprised of 20 mM Tris-HCl pH 7.4, 100 mM NaCl were pre-incubated at 37 °C with nal concentration of 100 µM D-TLKIVWX (X = I, S, R) from 10 mM stocking solution in water for 24 hours.Control tau brils from the same brain donor were treated identically except for the addition of D-TLKIVWX.

Cryo-EM data collection and processing
To prepare the cryo-EM grids, we applied 2.5 μl of sample solution onto Quantifoil 1.2/1.3200 mesh electron microscope grids glow-discharged for 2 minutes in a Pelco easiGlow unit before use.Grids were plunge-frozen into liquid nitrogen-cooled liquid ethane inside a Vitrobot Mark IV (FEI) vitri cation robot after blotting.Cryo-EM data of D-TLKIVWR and D-TLKIVWS brils were collected on a Titan Krios transmission electron microscope (Thermo Fisher Scienti c) located at the National Center for Cryo-EM Access and Training, which is equipped with a Bioquantum/K3 direct detection camera (Gatan), operated with 300 kV acceleration voltage and an energy slit width of 20 eV, automated with Leginon software package 36 .Super-resolution movies were collected with a calibrated pixel size of 1.067 Å/pixel (0.5335 Å/pixel in super-resolution movie frames) and a dose per frame of ~1.5 e-/Å 2 .A total of 40 frames with a frame rate of 12 Hz were taken for each movie, resulting in a nal dose of ~60 e-/Å 2 per image.D-TLKIVWI brils were collected on a Titan Krios located at the HHMI Janelia Research Campus, which is equipped with a cold-FEG source (CFEG), a Selectris X energy lter and a Falcon 4i direct detection camera (TFS), operated with 300 kV acceleration voltage and an energy slit width of 6 eV, and automated with the SerialEM software package 37 .Electron Event Representation (EER) les were collected with a calibrated pixel size of 0.94 Å/pixel and a dose per raw frame of 0.0244 e-/Å 2 , resulting in 55 e-/Å 2 per image.AD-tau/D-TLKIVWX (X = I, S and R) were collected similarly as D-TLKIVWI brils, although manually targeted in SerialEM package.The AD-tau control was collected on a Titan Krios/Bioquantum/K3 setup located at the Stanford-SLAC Cryo-EM Center, operated with 300 kV acceleration voltage and an energy slit width of 20 eV, automated with EPU (TFS).(See details in Supplementary Table 1,2).
Movies and EER les were motion-corrected in RELION 38 and binned to pixel sizes according to Supplementary Table 1,2.CTF estimation was performed using CTFFIND4 39 .AD-tau/D-TLKIVWX brils were manually picked using e2helixboxer.pyfrom EMAN2 40 .D-TLKIVWX brils and AD-tau control particle picking was initially done manually using e2helixboxer.pyfrom EMAN2 for about 100 images as a training set for crYOLO 41 .CrYOLO was then trained with default parameters and was used to pick the rest of the images.Particle extraction, two-dimensional classi cation, three-dimensional classi cation, and 3D re nement were performed in RELION 42 .Brie y, particles were initially extracted using a larger box size of 640 pixels with two-fold binning.2D classi cation was then performed with all particles to eliminate bad particles and group particles into polymorphs if necessary.Particles from each polymorph were selected, extracted with smaller box sizes at native pixel sizes of detectors (binning=1), further "puri ed" using 2D classi cation and subjected to 3D classi cation, which was done initially with one class and then with three classes, using a Gaussian cylinder as the initial model.The best 3D classes were selected, and corresponding particles were nally re ned with 3D auto-re ne for the reported maps.(See details in

Atomic model building
Our starting atomic model of D-TLKIVWI was an ideal β-strand.It was manually adjusted to t the electrostatic potential map using Coot 44 and automated re nement was performed using Phenix 45 .To facilitate good rotamer geometry, the initial building and re nement was performed using a map with handedness chosen so that the amino acid residues appeared to be levorotary rather than dextrorotary.In this way, we could take advantage of the rotamer library in Coot which exists for L-amino acids, but not for D-amino acids.In the nal step of re nement, the map and coordinates were inverted to the correct hand, consistent with D-amino acids.The starting models for D-TLKIVWS and D-TLKIVWR were adapted from the re ned D-TLKIVWI structure.All atomic models were re ned in successive rounds using Coot for manual building and Phenix for automated re nement.Model validation statistics of all three D-peptide structures are reported in Supplementary Table 1.
Our starting atomic model of the complex between AD-tau PHF and D-TLKIVWI was built by manually orienting coordinates of the tau PHF (PDB ID 7nrv) 46 to t the electrostatic potential map using Coot 44 and then re ned with Phenix.Coordinates of a pair of β-sheets were extracted from the D-TLKIVWI structure described above and manually docked on the surface of the PHF using guidance from the 3.1 Å cryoEM map, as well as the low-pass ltered map (7 Å).We noted a blob of residual density situated at the end of three lysine side chains: K317 and K321 of tau and K3 of the D-peptide.Whatever molecule produced this residual density does not depend on the presence of the D-peptide to bind to tau, since a similar blob was evident in our PHF negative control lacking D-peptide.
Indeed, the presence of this residual density was noted in the original structure report of AD-tau PHFs 30 , and even noted in maps from PHFs produced with recombinant tau (PDB ID 7ql4) 47 .The chemical environment of this blob suggests that the blob originates from an anion, but the density is not su ciently detailed to uniquely identify the chemical species.It is roughly the size of a pair of phosphate ions.We chose to model this residual density with ethylenediaminetetraacetate (EDTA) because it ts the density, caries the expected negative charges to complement the positive charge on K317 and K321, and we know that EDTA was included in the buffer used for PHF puri cation.The starting models for tau complexed with D-TLKIVWS and D-TLKIVWR were obtained using an analogous procedure.The nal re ned coordinates for these two complexes do not include the D-peptides because density for the peptides was visible only in the low-pass ltered maps, and not in the high-resolution map (Figure 4 and Extended data Figure 9a).For the chemical shift perturbation (CSP) analysis, the overall change in chemical shift Δ was calculated between the free and bound states of tau K18+ protein as 50 : where Δδ H and Δδ N are the differences between the 1 H N and 15 N chemical shifts of the two states being compared.

Statistical analysis
Graphs are expressed as means + standard deviation (SD) and data were analyzed using SPSS 25 statistical analysis software (SPSS, Chicago, IL, USA).The one-way analysis of variance (ANOVA) was used to analyze difference among multiple groups.Statistical differences for all tests were considered signi cant at the *p < 0.05, **p < 0.01, ***p < 0.001 levels, NS, not signi cant.
EMD-44186/9B4N (Tau-D-TLKIVWR).Cryo-EM map and atomic model of the paired helical lament from the same patient have been deposited under the accession codes EMD-44187/9B4O.All data presented in this article are available within the gures and Supplementary Information les.All other data are available from the corresponding authors upon reasonable request. Figures NMR spectroscopyNMR samples were ~0.5 mL of 0.1 mM 15 N,13 C-labeled tau K18+ protein in 100 mM KCl, 20 mM NaH 2 PO 4 , 1 mM TCEP, 5%/95% D 2 O/H 2 O, pH 7.0 without or with 1 mM D-TLKIVWI, D-TLKIVWS, L-TLKIVWI.All NMR spectra were acquired at 25 °C with Bruker Avance III HD 600 MHz spectrometer equipped with QCI HCNP cryoprobe or Avance Neo 800 MHz spectrometer equipped with TCI HCN cryoprobe.Backbone assignments for both free tau K18+ and D-TLKIVWI bound tau K18+ were carried out using HNCACB, CBCA(CO)NH and C(CO)NH NMR experiments.NMR spectra were acquired with Topspin (Bruker), processed with NMRPipe 48 , and analyzed with NMRFAM-Sparky 49 .

Figure 1 D
Figure 1

Figure 2 D
Figure 2

Figure 3 D
Figure 3

Table Supplementary Table 1 ,
2).Part of the Cryo-EM data processing used Expanse GPU at San Diego Supercomputer Center through allocation BIO230174 from the Advanced Cyberinfrastructure Coordination Ecosystem 43 .