Membrane-mediated fibrillation and toxicity of the tau hexapeptide PHF6

The aggregation of the tau protein into neurofibrillary tangles is believed to correlate with cognitive decline in several neurodegenerative disorders, including Alzheimer's disease. Recent studies suggest that tau's interactions with the cell membrane could serve as a toxicity pathway and also enhance fibrillation into paired helical filaments (PHFs). Conformational changes associated with tau-membrane interactions are poorly understood, and their characterization could improve our understanding of tau pathogenicity. In this study, we investigated the molecular level structural changes associated with the interaction of the tau hexapeptide PHF6 with model lipid membranes and characterized the effects of these interactions on membrane stability and peptide fibrillation. We used two PHF6 forms, the aggregation-prone PHF6 with N-terminal acetylation (Ac-PHF6) and the non-aggregation prone PHF6 with a standard N terminus (NH3+-PHF6). We found that both PHF6 peptides are neurotoxic and exhibit similar membrane-mediated changes, consisting of: 1) favorable interactions with anionic membranes, 2) membrane destabilization through lipid extraction, and 3) membrane-mediated fibrillation. The rate at which these changes occurred was the main difference between the two peptides. NH3+-PHF6 displayed slow membrane-mediated fibrillation after 6 days of incubation, whereas Ac-PHF6 adopted a β-sheet conformation at the surface of the membrane within hours. Ac-PHF6 interactions with the membrane were also accompanied by membrane invagination and rapid membrane destabilization. Overall, our results reveal that membrane interactions could play a critical role in tau toxicity and fibrillation, and highlight that unraveling these interactions is important for significantly advancing the development of therapeutic strategies to manage tau-associated neurodegenerative diseases.

The misfolding and aggregation of the microtubule-associated protein tau into neurofibrillary tangles (NFTs) 3 is implicated in the development of Alzheimer's disease (AD) and other tauopathies such as Down's syndrome and frontotemporal dementia (1). The tau protein is composed of a projection domain made of acidic and proline-rich regions and a microtubule-binding (MTB) domain containing up to four repeat units made of tubulin-binding motifs ( Fig. 1) (2, 3). The physiological function of tau is to stabilize microtubules (4,5) to ensure cytoskeletal organization and cellular trafficking (6). Tau's function is modulated by post-translational modifications such as phosphorylation, which controls its affinity to the microtubules (7,8). In AD, tau is hyperphosphorylated, which leads to the abnormal accumulation of unattached tau and promotes selfaggregation into insoluble paired helical filaments (PHFs) (6). The deposition of NFTs was found to correlate with cognitive decline and neurodegeneration (9), and is involved in the rapid transcellular propagation of the disease (10).
The MTB domain plays major roles in both tau's physiological function and aggregation. Indeed, the MTB domain forms the core of PHFs. In particular, two hexapeptide segments, 275 VQIINK 280 (PHF6*) and 306 VQIVYK 311 (PHF6) located in the second and third repeat units of the MTB domain, respectively, have been identified as aggregation promoting ␤-sheet forming motifs (11)(12)(13). Both hexapeptides form stable parallel ␤-sheet-twisted filaments similar to the ␤-sheet structures found in tau-oligomers and NFTs (14,15), which make them relevant models of the full-length tau to investigate tau aggregation pathways (16).
Tau's toxicity mechanism is not well-understood. However, recent studies indicate that tau binding to the lipid membrane disrupts cell membrane homeostasis and causes apoptosis (17,18). There are several hypothesized mechanisms by which amyloid proteins interact with and destabilize lipid membranes. Amyloid-␤ (A␤), ␣-synuclein and amylin proteins have been described to form "ion-channel-like structures" on the cell membrane (19), which permeates the membrane and allows calcium flux (20). A␤ and ␣-synuclein have also been shown to remodel lipid membranes through a "carpeting-effect mechanism" (21), wherein the amyloid proteins blanket the lipid membrane causing membrane thinning (22) or generating membrane curvature (23,24). Membrane disruption can also be caused by a "detergent-like mechanism" as observed with A␤ (21), where the peptide removes lipids from the membrane by forming micelle-like structures with the lipids. Tau protein was also shown to extract lipids through a similar mechanism leading to the formation of stable protein-lipid complexes (17). Molecular level details of lipid membrane destabilization caused by tau protein, however, remain poorly understood.
Tau interaction with the cell membrane is also suspected to nucleate fibrillation. Previous studies have shown that upon interaction with anionic lipid membrane, tau (25,26), A␤ (27), and ␣-synuclein (28,29) are more prone to fibril formation. Possible explanations for accelerated fibril nucleation include membrane-induced changes to protein structure, partial folding of the intrinsically disordered proteins via membrane association or penetration, and increased protein concentration at the membrane (30). In tau protein, three segments located in the MTB domain (253-261, 315-323, and 346 -355) have been found to bind to lipid membranes and adopt a helical structure that promotes protein aggregation by enhancing protein-protein interactions (18,31).
In this study, the hexapeptide PHF6 was used as a model tau protein because of its importance in promoting tau aggregation, and the peptide on its own has been shown to recapitulate important fibril forming and toxic properties of the full-length tau (14,15). Truncated tau constructs, for example, K18, K19, and K32, have been routinely used as model tau proteins to investigate tau aggregation and pathological properties. Thus, insights gained about the membrane interaction, fibrillation, and toxicity of the PHF6 peptide will enhance our understanding of tau's role in neurodegenerative diseases and could lead to significant advancement in the development of therapeutic strategies to manage tau-associated neurodegenerative diseases.
We investigated the molecular level structural changes associated with the insertion of two PHF6 isoforms into lipid membranes and explored the effects on membrane stability and peptide fibrillation. After characterizing fibrillation of both peptides, their neuronal cell toxicity was evaluated. Angstrom level structural changes associated with PHF6 insertion into a lipid monolayer membrane at the air/water interface were investigated by in situ synchrotron X-ray reflectivity (XR) and grazing incidence X-ray diffraction (GIXD). The membranemediated conformational changes and fibrillation of PHF6 were also characterized in the presence of large unilamellar vesicles (LUVs). We found that both peptides preferentially inserted into the negatively charged lipid membrane, which led to different conformational changes; NH 3 ϩ -PHF6 inserted into the lipid membrane, whereas Ac-PHF6 formed a layer at the surface. Despite different macroscale events, both PHF6s disrupted membrane stability and underwent membrane-mediated fibrillation.

Results
PHF6 is an amphiphilic hexapeptide (32) located in the third repeat unit of the tau MTB domain (Fig. 1A). This hexapeptide has been shown to drive tau fibrillation by forming the core of ␤-sheet-rich tau aggregates (11)(12)(13). Although not physiological, the advantages of using PHF6 as a model tau peptide is that we can study the specific amino acids that drive tau fibrillation and learn about their interactions with lipid membranes. Fulllength tau contains 351-441 amino acids, and it likely has additional membrane interactions. It would be difficult to isolate the impacts from the fibril-prone PHF6 sequence in the context of full-length tau. In this study, we characterized the structural changes associated with PHF6 upon interaction with lipid membranes and investigated its role in cell toxicity and peptide fibrillation. Two PHF6 isoforms were evaluated, nonacetylated PHF6 (NH 3 ϩ -PHF6) and N-terminal acetylated PHF6 (Ac-PHF6) (Fig. 1, B and C), to investigate the influence of peptide charge on fibrillation and membrane interaction.

N-acetylation of PHF6 promotes fibrillation
PHF6 has been previously described to form fibrillar aggregates rich in parallel ␤-sheets (13,14). In this study we evaluated A, hTau40, made of 441 amino acids, is the longest isoform found in the brain. It is composed of a projection domain (green) and microtubule-binding domain (blue). PHF6 is a hexapeptide found in the third repeat unit of the microtubule-binding domain (R3) and is amphiphilic. B and C, structures of NH 3 ϩ -PHF6 and Ac-PHF6 peptides at pH 6.0. The acetyl group, the positive and negative charges are indicated in black, blue, and red, respectively.

Membrane-mediated fibrillation and toxicity of PHF6
the capability of NH 3 ϩ -PHF6 (net charge at pH 6: ϩ1) and Ac-PHF6 (net charge at pH 6: 0) to form fibrils by incubating the peptides in MilliQ water at 3 mg/ml at 70°C for 3 days.
As shown by TEM images in Fig. 2, A and C, unincubated NH 3 ϩ -PHF6 and Ac-PHF6 showed very few features indicating that both peptides are soluble and likely monomeric. After 3 days of incubation at 70°C, only Ac-PHF6 formed characteristic amyloid fibrils (Fig. 2D), whereas NH 3 ϩ -PHF6 remained unaggregated (Fig. 2B). These results are consistent with previous findings (13) and indicate that Ac-PHF6 is more prone to fibrillation than NH 3 ϩ -PHF6, which could be due to electrostatic repulsions between positively charged NH 3 ϩ -PHF6 peptides. Shorter Ac-PHF6 fibrils were also produced from the long fibrils after 30 min of sonication (Fig. 2E). This short fibril solution was composed of smaller protofibrils with lengths in the 50 -250 nm range and small globular structures with diameters around 16 nm.

PHF6 peptides are cytotoxic
We next evaluated neuronal toxicity of unincubated NH 3 ϩ -PHF6, unincubated Ac-PHF6, and long and short Ac-PHF6 fibrils on cultured neuroblastoma SH-SY5Y cells by MTS assay. This assay is routinely used to evaluate cytotoxicity and has been previously used to evaluate toxicity of full-length tau (33). Normalized viability results are reported in Fig. 3. Compared with untreated cells (control), cells treated with unincubated and fibrillar PHF6 displayed lower levels of cell viability (p Յ 0.01), with final viabilities lower than the 70% threshold generally considered for cytotoxicity (34). No significant difference in cell viability was observed between cells treated with unincubated NH 3 ϩ -PHF6 and Ac-PHF6. Additionally, both short and long Ac-PHF6 fibrils exhibited high toxicity with the short fibrils being the most cytotoxic with less than 25% viability. The higher toxicity observed for the short fibrils is supported by previous studies showing that pre-fibrillar amyloid aggregates (oligomers and protofibrils) made of full-length tau, ␤-amyloid, or ␣-synuclein display higher toxicity compared with their fulllength fibril counterparts (35)(36)(37)(38).
To complement the MTS-based cell toxicity assay, a live/ dead stain assay was performed. Two stains were used, calcein-AM and ethidium homodimer-1 (EthD1). Live cells were stained green by the conversion of the cell-permeant calcein-AM to green fluorescent calcein dye (excitation at 494 nm, emission at 517 nm). Membrane-compromised cells were stained red from binding of the noncell-permeant EthD1 to DNA that induces fluorescence turn-on (excitation at 528 nm, emission at 617 nm). After incubating the neuroblastoma cells with 10 M unincubated NH 3 ϩ -PHF6, unincubated Ac-PHF6, or fibrillar Ac-PHF6 (long and short) for 24 h, fluorescence images were taken (Fig. 4A) and the number of dead cells was quantified (Fig. 4B). Compared with the untreated control, all PHF6 samples caused significant increases in dead counts (cells/mm 2 ) (p Յ 0.01), indicating that PHF6 caused membrane permeation that could contribute to the peptide's toxicity. Comparable results were obtained for both unincubated NH 3 ϩ -PHF6 and Ac-PHF6, indicating no significant difference in their membrane permeation activity. Additionally, the ␤-sheet-rich fibrils caused higher dead cell counts (Fig. 4B) than the unincubated peptides. Taken together, our results show that both NH 3 ϩ -PHF6 and Ac-PHF6 are cytotoxic and the fibrillar state is more toxic and membrane disruptive compared with the unaggregated state.

PHF6 inserts into and destabilizes lipid membranes
One hypothesized mechanism of amyloid protein toxicity is the disruption of cell membrane structure and function via direct protein-membrane interactions. Indeed, the live/dead staining assay results show a correlation between PHF6 peptide toxicity and membrane permeabilization. To characterize

Membrane-mediated fibrillation and toxicity of PHF6
PHF6 interactions with the cell membrane, we used lipid monolayers formed at the air/water interface in a Langmuir trough as a model membrane system. Two lipids were used, zwitterionic DMPC and anionic DMPG. The insertion of NH 3 ϩ -PHF6 or Ac-PHF6 into monolayers was measured, whereas the monolayer surface area was held constant (Fig. 5). As such, favorable interactions that cause tau to insert into the lipid monolayer led to increases in surface pressure.
Both PHF6 peptides readily inserted into anionic DMPG monolayers, causing initial fast increases in surface pressure after injecting the peptides into the subphase (Fig. 5). Consistent with previously published findings with hTau40 (25), the truncated Ac-PHF6 did not insert into zwitterionic DMPC monolayer (data not shown), indicating that the peptide preferentially interacts with anionic lipid membranes. The insertion of NH 3 ϩ -PHF6 resulted in higher surface pressure increases than Ac-PHF6 (Fig. 5A). As both peptides exhibited similar adsorption to a clean air/water interface (Fig. S1), the difference in membrane insertion was likely due to differences in their specific interactions with the lipid membrane rather than their intrinsic surface activities. After NH 3 ϩ -PHF6 rapidly inserted into the membrane, a slow but persistent decrease in surface pressure was observed over time, indicating that some material was being removed from the surface (lipids and/or peptides). Note that this pressure loss is not due to instability of the lipid film, as lipid monolayers held at 25 mN/m did not show appreciable pressure loss over time (data not shown). Interestingly, Ac-PHF6 displayed two unique phases of surface pressure loss after the initial rapid increase: a sharp decrease in surface pressure, followed by a slower loss of about 2 mN/m per h. This two-phase surface pressure loss might be caused by different mechanisms of membrane destabilization.
To monitor lipid monolayer morphology, fluorescence microscopy (FM) was used to image the monolayer before and after peptide insertion (Fig. 5, B and C); 0.5 mol % headgrouplabeled lipid dye Texas Red DHPE (TR-DHPE) was included in the monolayer for FM imaging. A DMPG monolayer compressed to 25 mN/m has undergone phase transition where ordered liquid-condensed (LC) domains formed in the disordered liquid-expanded (LE) phase. As the bulky fluorescent dye was excluded from the LC phase, it appeared as dark patches, whereas LE phase appeared bright.
The insertion of NH 3 ϩ -PHF6 caused negligible changes to the monolayer morphology as FM images appeared similar before and at different time points of peptide insertion (Fig. 5B). In contrast, insertion of Ac-PHF6 resulted in drastic changes to monolayer morphology, where at least three fluorescence intensity levels were observed in FM images (Fig. 5C). In addition to the bright LE and dark LC phases, a third even brighter phase was formed in the LE portion of the film at earlier time points (Fig. 5C, 15-45 min). These brighter features could be membrane invagination (39) caused by the insertion of Ac-PHF6. After 1 h of incubation, the third bright phase disappeared and the dark LC domains became

Membrane-mediated fibrillation and toxicity of PHF6
progressively reduced, which could be due to the disruption of these domains by the peptide or reduction of the monolayer surface pressure observed with time.
One cause for the higher level of insertion of NH 3 ϩ -PHF6 compared with Ac-PHF6 could be the different overall peptide charges due to acetylation. At pH 6.0 (approximate water pH), NH 3 ϩ -PHF6 had a net positive charge that promoted its insertion into the negatively charged lipid membrane. In contrast, Ac-PHF6 had a neutral charge and therefore had less attraction to the anionic DMPG lipid headgroups. Overall, both peptides favorably interacted with the anionic membrane and extracted material (lipids, peptides, or both) from the membrane. Additionally, our results show that although Ac-PHF6 inserted less into the membrane than NH 3 ϩ -PHF6, it caused faster and more disruptions to the membrane stability and morphology.

؉ -PHF6 inserts into the membrane while Ac-PHF6 accumulates at the membrane surface
It is clear that both PHF6 peptides destabilized the cell membrane and caused toxicity. The peptides favorably interact with anionic lipid monolayers and some differences are seen on their effects on membrane morphology. To gain further insights into the molecular level structural changes of the peptides and the lipid monolayer accompanying PHF6 insertion, we used in situ synchrotron X-ray scattering techniques to determine the angstrom-level structure and organization of lipids and PHF6 peptides at the air/water interface.
XR was employed to determine the electron density distribution of lipid-peptide films at the air/water interface in the direction perpendicular to the film surface (z-direction) with sub-Å resolution. Using this method one can distinguish between regions of materials with different electron densities such as, in our case, phospholipid tails, lipid headgroups, peptides, or mixtures of these components. This information can be used to reconstruct a model depicting the structure of the phospholipid/peptide film in the z-direction (averaged over the coherent volume of the X-ray beam) and determine the location of peptide in the membrane as previously described (27). A DMPG monolayer at 25 mN/m was modeled as two "slabs" of different electron densities (Fig. 6, A1 and A2, and supporting Table S3), a 15.9 Ϯ 0.2 Å thick slab of phospholipid tails at a normalized electron density (/ water ) value of 0.98 Ϯ 0.03 and a 9.1 Ϯ 0.3 Å thick slab of more electron dense phospholipid heads with a / water value of 1.58 Ϯ 0.02. The insertion of NH 3 ϩ -PHF6 into the lipid monolayer induced subtle structural changes ( Fig. 6). At 2.5 h after injection, / water of the DMPG lipid headgroups decreased from 1.58 Ϯ 0.02 (Fig. 6, A2) to 1.49 Ϯ 0.03 (DMPG ϩ NH 3 ϩ -PHF6) (Fig. 6, B2). This change could be caused by the insertion of NH 3 ϩ -PHF6 into the lipid headgroups, thus decreasing the / water value of this layer. The / water values of proteins, including A␤40 and hTau40, adsorbed to an air/water interface have been previously reported to be around 1.26 (25,27). Thus, the intermediate / water value of 1.49 Ϯ 0.03 (Fig. 6, B2) is likely a mix of peptides and lipid headgroups. Concurrently, / water of the lipid tails increased slightly from 0.98 Ϯ 0.03 to 1.02 Ϯ 0.004, suggesting some peptide insertion into the tails. A third slab was modeled beneath the lipid headgroups of electron density that was subtly distinguishable from the water subphase (/ water value of 1.041 Ϯ 0.005). This indicated a sparsely populated layer of NH 3 ϩ -PHF6 adsorbed to the lipid head groups. A similar observation was made at 4.5 and 6.5 h, a small change in electron density indicated NH 3 ϩ -PHF6 insertion into the membrane and a sparse layer of adsorbed peptide.
In contrast, insertion of Ac-PHF6 induced drastic changes to the lipid monolayer structure ( Fig. 7 and Table S4). After 2.5 h of Ac-PHF6 incubation with the monolayer, the membrane had clearly lost many features of its original structure. Three slabs were needed to fit the XR reflectivity data (Fig. 7, B2). The two slabs representing the lipid monolayer had decreased electron densities with increased roughness values indicating less distinct layers. Combined with the drastic decreases in surface pressure observed after the addition of Ac-PHF6 (Fig. 5), the decrease in electron density was likely due to loss of lipids at the surface. A third layer, extending 24.1 Ϯ 0.07 Å into the water subphase at a / water value of 1.21 Ϯ 0.0014 is likely a peptide

Membrane-mediated fibrillation and toxicity of PHF6
layer forming underneath the membrane (Fig. 7, B2). This highly perturbed monolayer structure could be the result of large-scale membrane deformations, such as invagination, which were also observed by FM imaging (Fig. 5C). At 4.5 h, the membrane-peptide film resembled a typical lipid monolayer with a less prominent peptide layer underneath the lipid headgroups (16.3 Ϯ 0.3 Å thick with a / water value of 1.06 Ϯ 0.0018). The electron density of the lipid membrane layers decreased by about 10% compared with that at 25 mN/m, reflecting the loss of material at the surface. Finally, at 6.5 h the lipid/peptide film structure was similar to that at 2.5 h, a highly perturbed lipid monolayer covered with a thick layer of bound peptides.

PHF6 forms ␤-sheet-rich structures and disrupts lipid packing
Although analysis of XR data provides electron density distributions averaged over both in-plane ordered and disordered parts of the monolayer, GIXD detects only ordered diffracting species present at the air/water interface. As the expected ordered structures included condensed phospholipid tails and repeating peptide structures such as ␤-sheet-rich fibrillar structures, GIXD was used to monitor assembly of PHF6 peptides and its impacts on lipid packing. Fitting of integrated dif-fraction peaks gave rise to the lattice spacing (d-spacing), coherence length (L c ), and relative amounts of diffracting material in the plane of the lipid/peptide film (27). L c is calculated from the full width at half-maximum peak intensity (FWHM) and L c value represents the average size of diffracting crystallites. GIXD diffraction images and a summary of the peak fitting parameters are reported under supporting "Experimental procedures", Fig. S2, and Tables S1 and S2.
DMPG monolayer at 25 mN/m showed a single diffraction peak indicating hexagonally packed lipid tails in the LC phase with a distance between acyl tails of 4.87 Ϯ 0.02 Å (Table 1) and a L c value of 513 Ϯ 15 Å (Table 1 and Fig. 8A). One hour after NH 3 ϩ -PHF6 injection, the ordered lipid tails adopted a distorted hexagonal packing as two diffraction peaks were observed. At 3.5 h, lipid tails again became hexagonally packed. However, the total amount of diffracting lipids decreased by ϳ60% based on peak intensity decrease. The in-plane lipid packing remained largely the same for the rest of the experiment and no other ordered structures were observed.
In contrast, the interaction of Ac-PHF6 with DMPG was more dynamic (Fig. 8B, Tables 1 and 2). At 1 h after Ac-PHF6 injection, the lipid diffraction peak shifted slightly to indicate larger d-spacing between lipid tails. The phospholipid diffrac-

Membrane-mediated fibrillation and toxicity of PHF6
tion peak also broadened significantly, corresponding to a more than 5-fold reduction of L c from about 500 to 73 Ϯ 11 Å. Concurrently, a second, nonlipid diffraction peak appeared at a q xy value of 1.33 Å Ϫ1 that corresponds to a d-spacing of 4.74 Ϯ 0.02 Å ( Table 2). As this d-spacing exactly matches the distance between ␤-strands of amyloid fibers previously measured by X-ray diffraction (40), the appearance of this peak indicates the formation of semi-crystalline ␤-strand lattice structures of Ac-PHF6 stabilized by hydrogen bonds. The L c calculated from the peptide diffraction peak was 125 Ϯ 19 Å, indicating that about 26 peptides are in positional registry in the crystallites. At 3.5 h, two lipid diffraction peaks were observed, indicating distorted hexagonal packing with tail-tail intermolecular distances of 4.841 Ϯ 0.016 and 4.966 Ϯ 0.018 Å with corresponding L c values of 340 Ϯ 50 and 95 Ϯ 14 Å, respectively. Interestingly, at 3.5 h, the peptide diffraction peak diminished significantly before returning to a high intensity peak at 5 h. At 5 h the lipid diffraction indicated a return to hexagonal packing with small ordered lipid domains (L c of 76 Ϯ 11 Å), and at 7 h both the lipid and peptide diffraction peaks were diminished significantly. We additionally extracted Bragg rod data by integrating X-ray diffraction images over q xy to determine the length of the coherently scattering molecular entity along the z-dimension (L m ). Analysis of the Ac-PHF6 diffraction peak at 1.33 Å Ϫ1 at 5 h after peptide injection yielded a L m value of 20.2 Ϯ 0.6 Å (Fig. S4). Thus, of the 24.12 Ϯ 0.07-Å thick peptide layer accumulated underneath the lipid headgroups as detected by XR, 20.2 Å of it was participating in diffraction.
Taken together, GIXD resolved three distinct stages of the Ac-PHF6 -membrane film after peptide addition. The first is low intensity diffraction from hexagonal lipid packing with strong peptide diffraction observed 1.0 h post-injection. At this stage, Ac-PHF6 is bound to the DMPG monolayer, and the binding disrupted lipid packing and induced the peptide to assemble in ␤-sheet-rich crystalline aggregates. This stage is followed by a second distinct stage of high intensity distorted hexagonal lipid packing with little peptide diffraction at 3.5 h. The interpretation of this drastic change in the peptide-lipid film is that as the ␤-sheet-rich Ac-PHF6 aggregates grew, they detached from the lipid monolayer and resulted in a diminished peptide diffraction peak at the air/water interface. This loss of peptides from the interface is corroborated by observed loss of surface pressure and changes in membrane morphology (Fig.  5). At this stage, another cycle of Ac-PHF6 insertion and assembly occurred (5.0 -6.5 h), leading to detachment and loss of material at the interface at 7.0 h post-injection where significant surface pressure loss and only a weak lipid diffraction peak were observed.
From the GIXD data, we additionally deduced tertiary structural information about the Ac-PHF6 ␤-sheet-rich aggregates (or protofibrils). In GIXD the X-rays travel in the plane of the lipid monolayer and they only diffract from ordered species with their unit cell perpendicular to that plane. Thus, the ␤-strand unit cell must be oriented perpendicular to the lipid monolayers, which suggests that the fibril growth axis was parallel to the lipid membrane (Fig. 9F). At 5.0 h, two additional diffraction peaks were observed at q xy values below 0.5 Å Ϫ1 (Fig. 8C), with d-spacings of 9.0 Ϯ 0.06 and 11.5 Ϯ 0.13 Å. Although the exact structures that gave rise to these peaks are unknown, we posit that at least one of the peaks arose from inter-␤-sheet packing of the protofibril as amyloid fibrils have been reported to have an approximate spacing of 10 Å between ␤-sheets (40). Thus, these diffraction peaks may have arisen from two types of inter-sheet packing arrangements. Alternatively, the diffraction could be from ordered phospholipid headgroups templated by electrostatic interactions with the protofibril. Although lipid headgroup diffraction is typically not observed by liquid surface X-ray scattering experiments, this study may be a unique case driven by the high degree of ordering in the bound protein protofibrils and their strong interaction with the negatively charged headgroups via positively charged lysine moieties. These observations, along with the XR data, corroborate dynamic Ac-PHF6 -membrane interactions and structural reorganization as described previously.
With the findings from GIXD and XR, we constructed a detailed structural model for PHF6 peptide interaction with the anionic lipid membrane (Fig. 9). Favorable electrostatic interactions cause NH 3 ϩ -PHF6 to fully insert into the phospholipid monolayer, as evidenced by modulated electron density of the headgroups and tails. A sparsely populated peptide layer was

Membrane-mediated fibrillation and toxicity of PHF6
observed outside of the DMPG monolayer. As a result of peptide interaction, packing of lipid tails became distorted (Fig. 9, A and C). Ac-PHF6 first interacted with DMPG to form a 16-Å thick layer of peptides underneath the lipid headgroups and induced distorted hexagonal packing of the lipid tails (Fig. 9D). This layer was possibly formed by burying the C-terminal cationic lysine in the anionic headgroups, leaving 16 Å of peptide to protrude below the lipid monolayer (Fig. 9, B and D). Ac-PHF6 quickly formed ␤-sheet-rich protofibrillar structures after inserting into the DMPG membrane as evidenced by a ␤strands d-spacing of 4.7 Å (Fig. 9F). These protofibrils are oriented with the growth axis parallel to the membrane, and likely had one terminal lysine inserted into the lipid headgroups. In addition to the ␤-strands d-spacing of 4.7 Å, two additional d-spacings of 9.0 and 11.5 Å were also observed between the ␤-sheets (Fig. 9F). The 24-Å thickness of the peptide layer under the monolayer was consistent with the PHF6 fibril structure (41) with one terminal lysine excluded from the measurement (Fig. 9, E and F, and Fig. S3). This model was further corroborated by Bragg rod analysis of the 5.0-h Ac-PHF6 1.328-Å Ϫ1 diffraction peak (Fig. S4). The result was that 20.2 Ϯ 0.6 Å of the peptide layer were contributing to the diffraction peak out of the 24-Å protein layer measured by XR studies.

PHF6 interaction with LUVs promotes fibrillation
NH 3 ϩ -PHF6 and Ac-PHF6 displayed selective favorable interactions with anionic DMPG membrane and Ac-PHF6 was also shown to assemble into ␤-sheet-rich protofibrils upon interaction. To investigate the effect of this interaction on peptide fibrillation, both hexapeptides were incubated in water with 100-nm LUVs composed of 70 mol % POPC and 30 mol% POPG. Peptide secondary structures, vesicle size distribution, and formation of fibrils were evaluated using circular dichroism (CD) spectroscopy, dynamic light scattering (DLS), and TEM imaging, respectively. We did not use the popular fluorescence probe thioflavin T to track fibril formation because of its poor binding to PHF6 fibrils.
Freshly solubilized NH 3 ϩ -PHF6 and Ac-PHF6 were rich in random coils as shown by the negative peak at around 195 nm in CD spectra (Fig. 10). After a 1-h incubation with vesicles (168 g/ml of peptide with 688 M lipids) at room temperature, drastic changes in secondary structure were observed for both peptides; positive and negative peaks at around 200 and 218 nm, respectively, appeared, indicating the formation of ␤-sheetrich structures. A stronger CD signal was observed in the case of Ac-PHF6 showing a higher ␤-sheet content compared with NH 3 ϩ -PHF6. The formation of ␤-sheets by Ac-PHF6 corroborates our finding from GIXD measurements (Fig. 8). However, GIXD did not detect ␤-sheet assembly of NH 3 ϩ -PHF6 at the pure DMPG membrane surface. CD is a bulk solution technique, thus, the measured ␤-sheet content in NH 3 ϩ -PHF6 incubated with LUVs could be from the peptide adopting a ␤-sheet conformation in solution, and not at the surface of the membrane. As NH 3 ϩ -PHF6 was shown to destabilize the lipid monolayer (Fig. 5), it is likely that the peptide could adopt a ␤-sheet conformation when complexed with dissolved lipids. Alterna-

Figure 9. Model for PHF6 interactions with DMPG monolayer based on XR and GIXD data. A, NH 3
ϩ -PHF6 structure extracted from a previously published structure of a paired helical filament (41). Positive charges on the peptide are circled in blue. B, Ac-PHF6 structure with the positive charge circled in blue. The length of the peptide (red) excluding the C-terminal lysine is estimated. C, model for nonfibrillar (unincubated) NH 3 ϩ -PHF6 interacting with a DMPG monolayer. Electrostatic interactions drive the insertion of the positively charged peptide into the anionic DMPG monolayer. Lipid tails are in a distorted hexagonal packing structure. D, model for unincubated Ac-PHF6 interactions with DMPG. Peptide insets with the cationic lysine (blue) into the anionic DMPG headgroup. A 16-Å layer of peptide is observed below the lipid monolayer, and lipids are in a distorted hexagonal structure. E, model of an Ac-PHF6 fibril derived from a previously published fibril structure (Fig. S3) (41). 2 peptides are shown here. Measured distance (red) excludes one terminal lysine, assuming that the lysine is embedded in the DMPG headgroups. Positive charges are circled in blue. F, model for inserted and assembled Ac-PHF6 protofibril interacting with DMPG monolayer. Lipid packing is disrupted due to peptide interactions, and one terminal lysine is assumed to insert into the anionic phospholipid headgroup. The fibril forms a 24-Å layer of peptide beneath the lipids. The fibril growth axis is parallel to the plane of the lipids, and there is a lattice spacing of 4.7 Å between ␤-strands, and 9.0 or 11.5 Å spacing between ␤-sheets.

Membrane-mediated fibrillation and toxicity of PHF6
tively, the membrane curvature found in LUVs may induce ␤-sheet structures more than what is observed in the flat lipid monolayer system (42,43).
LUV size distribution was monitored by DLS during incubation with both peptides (50 g/ml of PHF6 in the presence of 20 M lipids) (Fig. 11). LUVs were stable for over 7 days at room temperature where the hydrodynamic radius (R h ) remained unchanged at ϳ50 nm (Fig. S5). The addition of NH 3 ϩ -PHF6 to the LUVs did not induce any changes to the vesicle size distribution during the first 24 h, but the size distribution became broader and R h increased after 6 days (Fig. 11A). By day 8, the vesicles likely became unstable and agglomerated. In contrast, after mixing with Ac-PHF6, an immediate broadening of LUV size distribution and increase of R h were observed (Fig. 11B). By day 6, size distribution shifted toward larger sizes and became very broad. These changes are likely caused by immediate peptide binding and vesicle agglomeration, followed by peptide fibrillation at longer incubation times.
To further understand the increase of the LUV size induced by NH 3 ϩ -PHF6 and Ac-PHF6 and the changes in secondary structures, the samples were imaged by TEM (Fig. 12). The sample containing NH 3 ϩ -PHF6 incubated with LUVs for 1 h showed round, 100 -200 -nm sized features (Fig. 12A). These are likely peptide-coated LUVs as bare lipid vesicles are completely disrupted during the drying step in sample preparation for TEM (27). After 6 days of incubation, long fibrils were visible, which supports DLS measurements that showed the presence of larger structures in the sample. When incubated alone, NH 3 ϩ -PHF6 did not form fibrils in the same incubation period, indicating that the interaction with the lipid membrane promoted NH 3 ϩ -PHF6 fibrillation, possibly by reducing the repulsive electrostatic interactions between positively charge ammonium groups of NH 3 ϩ -PHF6. Also, by binding to the lipid membrane, the increased local concentration of NH 3 ϩ -PHF6 is also expected to promote peptide fibrillation.
In the presence of LUVs, Ac-PHF6 produced long fibrils after only a few hours of incubation (Fig. 12B), which is consistent with the high ␤-sheet content measured by CD spectroscopy (Fig. 10) and by GIXD (Fig. 8). Fibrils produced in the presence of LUVs formed large clusters connected to small dark agglomerates, which appear to be Ac-PHF6 -coated LUVs. In contrast, when the same peptide was incubated alone, significantly slower fibrillation was observed; short protofibrils were observed after 1 h and long fibrils were only observed after 6 days of incubation. Taken together, our results show that anionic vesicles accelerated fibrillation of both NH 3 ϩ -PHF6 and Ac-PHF6.

Discussion
The interaction of the tau protein with the lipid membrane is believed to play a key role in peptide toxicity and fibrillation. In this study, we investigated the lipid membrane interaction of PHF6, a hexapeptide located in the third repeat unit of the MTB domain forming the core of tau PHFs and revealed its effect on membrane structure and peptide fibrillation. PHF6 was chosen as a model tau peptide to isolate the effect of this aggregation promoting hexapeptide motif from the full-length tau protein.
We used two PHF6 peptides, one fibrillation prone (Ac-PHF6) and another not fibrillation prone (NH 3 ϩ -PHF6), to study the effect of these peptides on cell toxicity, membrane interaction, and membrane-mediated fibrillation.
We found that both unincubated PHF6 peptides are neurotoxic and pre-formed Ac-PHF6 fibrils exert higher toxicity with the short fibrils exhibiting highest cytotoxicity (Fig. 3). Membrane disruption via direct peptide interactions has been proposed as a mode of toxicity for amyloidogenic proteins, and is the central hypothesis tested in this work. Membrane destabilization caused by PHF6 is supported by Langmuir trough isotherm experiments where both Ac-PHF6 and NH 3 ϩ -PHF6 favorably interact with anionic, but not zwitterionic, lipid monolayers at the air/water interface (Fig. 5). This interaction leads to lipid extraction from the interface into the aqueous subphase as observed by a decrease in the membrane surface pressure after the insertion of PHF6. This model agrees with a previously published study showing that full-length tau inserts into negatively charged lipid membranes and extracts lipids (17). Furthermore, the interaction of PHF6 with lipid membranes also enhances fibrillation kinetics of both peptides. Strikingly, in the presence of anionic LUVs, Ac-PHF6 formed fibrils within 1 h of incubation, compared with 6 days required for complete fibrillation in the absence of vesicles. Likewise, the nonaggregating NH 3 ϩ -PHF6 was able to form fibrils after 6 days of incubation with LUVs. Based on these results, we propose a 3-step model for PHF6 membrane-mediated toxicity: 1) peptide insertion into the lipid membrane, 2) membrane destabilization through lipid extraction, and 3) membrane-mediated fibrillation.
Despite a similar membrane-mediated toxicity mechanism, NH 3 ϩ -PHF6 and Ac-PHF6 undergo different molecular level changes when interacting with the lipid membrane, which are summarized in Fig. 13. XR data (Fig. 6) indicate that the cationic NH 3 ϩ -PHF6 fully and rapidly inserts into the negatively charged DMPG monolayer and does not immediately form ␤-sheets at the membrane surface but forms fibrils within 6 days (Figs. 11 and 12). Binding of the posi-

Membrane-mediated fibrillation and toxicity of PHF6
tively charged peptides to the anionic membrane reduces the electrostatic repulsion between the peptides and increases its local concentration, both of which promote the peptide's assembly into fibrils. In contrast, Ac-PHF6 associated with the membrane, likely via lysine-mediated electrostatic interactions, but does not fully insert into the lipid headgroups (Fig. 9). The peptide quickly destabilizes the membrane (Fig.  5) and assembles into ␤-sheetenriched protofibrils on the surface of the lipid membrane. These aggregates promote extensive membrane destabilization through both membrane invagination and fast peptide fibrillation.
It is important to note the difference in kinetics observed between experiments performed with a lipid monolayer made of pure DMPG lipids and LUVs made of 7:3 POPC:POPG lipids. In the LUV system, Ac-PHF6 reached complete fibrillation and disruption of the vesicles within 1 h, but in the lipid monolayer system large changes were observed up to 7 h. The differences in lipid compositions, membrane fluidity, membrane curvature, and peptide concentration are likely to influence both peptide interaction with lipid membranes and membrane destabilization kinetics, as observed in the case of ␣-synuclein (42). However, both of our experimental systems show the same trends for membrane interaction and fibrillation, whereas acknowledging a difference in peptide interaction kinetics.
Overall, the aggregation prone Ac-PHF6 exhibits fast membrane-mediated fibrillation and rapid membrane disruption, whereas the nonaggregation prone NH 3 ϩ -PHF6 displays both slower membrane disruption and membrane-mediated fibrillation. Additionally, the pre-formed Ac-PHF6 fibrils displayed higher neurotoxicity than unincubated Ac-PHF6 and NH 3 ϩ -PHF6 (Fig. 3), indicating that the peptide's assembly state and secondary structures are involved in lipid membrane destabilization and toxicity. Also, it is important to note that the lipid models used in this study were mainly composed of negatively charged lipids that do not capture the full complexity of the cell membrane. This may account for some discrepancy between the in vitro experiments and the in vivo cell toxicity assays. The cell membrane is asymmetric with a zwitterionic extracellular leaflet and anionic cytosolic leaflet (44,45). The in vitro cell toxicity assay measures impacts from PHF6's interactions with the extracellular leaflet, however, they have previously been used to correlate the toxicity of full-length tau monomers, oligomers, and fibrils (33). Although the tau protein originates inside the cell, future investigations must also consider the extracellular lipid environment as it likely plays an important role in transmission of tau toxicity throughout the brain.
In this study, we have gained molecular level insights about PHF6-membrane interactions and effects of these interactions on peptide aggregation and membrane structure. With PHF6 as a model peptide for fibril-prone sequences in tau, this knowledge enhances our understanding of membrane-mediated peptide neurotoxicity and fibrillation. Our results, together with a previously published study (17), support a lipid extraction mechanism of tau-induced membrane destabilization, which results in the formation of stable peptide-lipid complexes. These complexes are also likely involved in the fast peptide fibrillation observed in the presence of vesicles. The production of such complexes could be linked to transcellular propagation as cell internalization of tau aggregates is believed to induce fibrillation of intracellular tau (10). Further study of tau-lipid complexes can shed light on the role of tau in neurodegenerative diseases and could lead to the development of novel therapeutic strategies for tau-associated neurodegenerative disorders. With this study we also introduced two PHF6 models, one fibrillation prone (Ac-PHF6) and one nonaggregation prone (NH 3 ϩ -PHF6). These models are useful tools to study the two

Membrane-mediated fibrillation and toxicity of PHF6
toxicity events associated with amyloid proteins: membrane disruption and protein fibrillation. These PHF6 models could be used to better understand the mechanism of action of therapeutic molecules and could help elucidate the toxicity mechanisms associated with amyloid aggregates.

Peptide synthesis, purification, and incubation
N-Acetyl-VQIVYK (Ac-PHF6) and nonacetylated VQIVYK (NH 3 ϩ -PHF6) were synthesized by a solid-phase method as previously described (13). Peptides were purified by reverse phase-HPLC (RP-HPLC) using an acetonitrile-water gradient containing 0.1% TFA. Purified PHF6 were lyophilized and stored at Ϫ70°C until use. Peptide molecular weights were confirmed by electrospray ionization MS and peptide purity was determined by RP-HPLC. Stock solutions were prepared by solubilizing the peptides in MilliQ water at 4 mg/ml and filtering with a 0.02-m filter. PHF6 concentration was determined by absorbance measurement using a Nanodrop (Thermo Fisher Scientific) (⑀280 nm ϭ 1195.9 cm Ϫ1 M Ϫ1 (46), MW Ac-PHF6 ϭ 790 g/mol, MW NH3ϩ-PHF6 ϭ 748 g/mol). PHF6 peptides were incubated at 70°C for 3 days at 3 mg/ml or at room temperature for up to 6 days at 100 g/ml to produce fibrils. Ac-PHF6 fibrils produced after 3 days of incubation at 70°C were sonicated for 30 min with a 550T Ultrasonic Cleaner (VWR International, Radnor, PA) to form shorter fibrils.

MTS-based cell viability assay
SH-SY5Y neuroblastoma cells at passage 20 were cultivated at 37°C and 5% CO 2 in DMEM F-12 media containing 10% FBS and 1% PS. 96-Well-plates were seeded with 20,000 cells/well. Cells were incubated for 16 h to allow cell attachment prior to serum deprivation over 24 h ensuring cell synchronization to a quiescent state. Cells were then treated with 20 M PHF6 for 48 h. After treatment, cell viability was determined by the MTS reduction assay. The assay was performed by adding 20 l of CellTiter 96 AQueous One Solution Cell Proliferation Assay to each well. The conversion of MTS into formazan by metabolically active cells was monitored through absorbance at 490 nm after 3 h of incubation at 37°C and 5% CO 2 . The optical density (OD) of formazan was measured using a Spectra Max M2 plate reader (Molecular Devices, Sunnyvale, CA). The % cell viability was calculated using Equation 1.
% Viability ϭ OD sample Ϫ OD background OD control Ϫ OD background ϫ 100 (Eq. 1) The control and background samples consisted of untreated cells and unseeded wells, respectively. Each condition was tested in quintuplet and data were statistically analyzed using JMP Pro 13 software (SAS Institute, Inc., Cary, NJ). The effect of the PHF6 on cell viability was analyzed using t test with a p value Յ 0.01 and 70% cell viability was used as a threshold to define cytotoxicity.

Live/dead staining assay
SH-SY5Y cells were grown in a 6-well-plate at 150,000 cells/well. After 24 h incubation at 37°C and 5% CO 2 , cells were treated with 10 M PHF6 peptide diluted in DMEM F-12 media containing 10% FBS and 1% PS. After another 24 h, cells were incubated with 1 M EthD-1 and 0.2 M calcein-AM in DMEM F-12 media over 30 min at 37°C and 5% CO 2 . Cells were rinsed with media and imaged using a Nikon Eclipse TS100 fluorescence microscope (Nikon Instruments, Inc., Melville, NY). Two filters blocks were used. A TRITC (G2E/C) filter cube was used to detect ϩ -PHF6 quickly inserts into lipid membrane and slightly destabilizes the membrane while promoting a slow fibrillation. In contrast, the aggregation prone Ac-PHF6 rapidly inserts into lipid membrane and quickly assembles into ␤-sheet-rich protofibrils. Membrane destabilization and invagination occur concurrently.

Membrane-mediated fibrillation and toxicity of PHF6
EthD-1 (red fluorescence) and a FITC (B2E/C) filter cube was used to detect calcein (green fluorescence). Images were then merged using ImageJ (National Institutes of Health, Bethesda, MD). Finally, dead cell density (number of dead cells/mm 2 ) was determined and the results were statistically analyzed using JMP Pro 13 software (SAS Institute, Inc., Cary, NJ) based on the average of four images taken for each condition with both filters. The effect of PHF6 on the number of dead cells/mm 2 was analyzed using t test with a p value Յ 0.01.

Lipid monolayer insertion assays
A Langmuir trough (KSV Instruments Ltd., Finland) Micromini model with Delrin barriers was used to measure insertion of NH 3 ϩ -PHF6 and Ac-PHF6 into a lipid monolayer composed of 100% DMPG. Water (50 ml) was used as the subphase, and experiments were performed at 30 Ϯ 0.5°C (temperature was controlled via a circulating water bath). DMPG was initially dissolved at 3.0 mg/ml in 7:3 (v/v) chloroform to methanol solution and sonicated. TR-DHPE was dissolved in 9:1 chloroform to methanol at 0.5 mg/ml. The final spreading solution for the trough was composed of 0.2 mg/ml of DMPG with 0.5 mol % TR-DHPE in 9:1 (v/v) chloroform to methanol. After spreading the lipids on the air/water interface, solvent was allowed to evaporate for 10 min before compressing the lipids with symmetric barriers at 3.0 mm/s to a surface pressure of 25 mN/m. For constant area experiments, the barrier position was fixed so that peptide insertion resulted in pressure increase. NH 3 ϩ -PHF6 and Ac-PHF6 were dissolved in water at 800 M, vortexed, and the supernatant was collected after 10 min of centrifugation at 14,000 rpm to remove insoluble peptides. Peptides were injected into the subphase to reach a final concentration of 5 M in the trough subphase. Fluorescence microscopy images were taken with an IX51 model inverted fluorescent microscope (Olympus) and with an Exi Aqua Bio-Imaging Camera (QImaging).

X-ray scattering experiments
X-ray scattering experiments were performed at the Advanced Photon Source at Argonne National Labs, Sector 15 NSF's Chem-MatCARS. Experiments were carried out in a 6.5 ϫ 6.5 cm 2 (20-ml subphase volume) Langmuir trough equipped with a Wilhelmy plate balance. Water subphase was degassed and maintained at room temperature (ϳ23°C) and DMPG was deposited at the air/ water interface to form a monolayer at 25 mN/m. An aliquot of NH 3 ϩ -PHF6 or Ac-PHF6 was then injected into the subphase to a final peptide concentration of 5 M in the trough. To prevent background scattering and oxidative beam damage to the monolayer, the trough was in a sealed canister and purged with helium gas to reduce the gaseous oxygen content to Ͻ2%. The wavelength of the X-ray beam was 1.24 Å, and X-rays were detected with a Dectris PILATUS 100K detector. Data were integrated using Python software built by beamline support scientists (https://github.com/ weibu/Liquid_Surface_ChemMatCARS). 4 A typical set of XR and GIXD scans required ϳ45-60 min to obtain. As an additional precaution against beam damage, the trough was systematically translated during data collection. The dimensions of the incoming X-ray beam footprint on the liquid surface were ϳ1 mm by 3-10 mm for XR, and ϳ1 mm by 29 mm for GIXD. XR and GIXD measurements were collected and analyzed as described in the supporting "Experimental procedures".

Preparation of LUVs
LUVs composed of 70 mol % POPC and 30 mol % POPG were prepared as previously described (27). A 4 mM lipid solution was prepared in chloroform and dried overnight under vacuum. Dried lipid film was rehydrated with MilliQ water to 2 mM concentration. Lipid solution was vortexed and incubated for 30 min at room temperature before being exposed to five freeze/thaw cycles. Finally, rehydrated lipids were extruded 19 times using an extruder with 100 nm pore size polycarbonate membranes (Avanti Polar Lipids, Alabaster, AL).

CD spectroscopy
PHF6 secondary structures were analyzed immediately after solubilization in water at 168 g/ml and after 1 h incubation in the presence of 688 M LUVs by CD spectroscopy using an AVIV 410 CD spectrometer (Aviv Biomedical Inc., Lakewood, NJ). A 0.1-cm path length quartz cell (Starna Cells, Atascaders, CA) was used and CD spectra were obtained in triplicates between 180 and 260 nm with an averaging time of 15 s. The average CD signal in millidegree was then converted to molar ellipticity () (47).

DLS
LUVs were analyzed by DLS to determine their size distribution when incubated alone at 20 M or in the presence of 50 g/ml of NH 3 ϩ -PHF6 or Ac-PHF6. Samples were analyzed using a DAWN HELEOS II light scattering instrument (Wyatt Technology Corporation, Santa Barbara, CA) in a quartz microcuvette. Scattering intensity at 90°was recorded over 10 min and analyzed using the regularization tool on ASTRA 5.3.4.20 software.

TEM imaging
PHF6 peptides were imaged by TEM to visualize fibrillation. Peptides were diluted to 50 g/ml with water and 4 l of the diluted sample was loaded onto a glow discharge (30 s) copper grid. Excess sample was wicked away and the grid was stained 4 times with 2% uranyl acetate, 4 l of stain for 3 min followed by 3 more stains of 1 min each. Between each staining step, excess of stain was wicked away. After staining, the grid was air dried and imaged using a HITACHI HT7700 transmission electron microscope (Hitachi High Technologies Corp., Tokyo, Japan).