Abstract
Tauopathies are defined by progressive accumulation of tau amyloids. These assemble around a protein seed, whose structure is unknown, but might explain the initiation of pathology. We have purified and characterized distinct forms of tau monomer—either seed-competent or inert. Recombinant tau that was seed-competent triggered intracellular tau aggregation, induced full length tau fibrillization in vitro, and exhibited intrinsic properties of self-assembly. Tau monomer from AD brain, but not from controls, similarly seeded aggregation, and self-assembled in vitro to form higher order, seed-competent structures. We used crosslinking with mass spectrometry to identify distinct conformers of both recombinant tau and human brain-derived protein. Theoretical models informed by this data suggest that VQIINK and VQIVYK sequences, which support amyloid formation, are uniquely exposed in all seed-competent structures. Our data imply that initiation of pathological aggregation begins with conversion of tau monomer from an inert to a seed-competent form.
Introduction
Amyloids are ordered protein assemblies, typically rich in beta sheet, that underlie multiple disorders including Alzheimer’s disease (AD). Amyloid-forming proteins include tau, synuclein, and expanded polyglutamine proteins such as huntingtin, among many others. It is unknown how or why intracellular amyloid proteins such as tau transition from a relatively inert form to one that efficiently self-assembles into ordered structures. This process begins with the formation of a pathogenic “seed,” a structure that serves as a template for homotypic fibril growth. This structural transition could be the critical event in the pathogenesis of neurodegeneration. Under defined conditions and relatively high concentrations (typically micromolar), recombinant tau monomer will form amyloid fibrils in vitro. However free monomer in cells is likely at a much lower concentration, and a complex intracellular protein milieu, with competing heterotypic interactions, theoretically should inhibit spontaneous self-assembly. The conversion of a protein from a monomer to a large, ordered multimer may occur by multiple mechanisms, but a proximal step involves the formation of a seed. This event, and indeed the actual form of the protein that constitutes the “minimal” seed, has remained obscure. This has led to the idea that a seed is potentially transitory, arising from equilibrium between two states: one relatively aggregation-resistant, and another that is short-lived. A seed could be a single molecule, or an assembly of molecules. Based on extrapolation from kinetic aggregation studies, it has been suggested that a critical seed for tau and polyglutamine peptide amyloid formation is a single molecule(1,2), while another study has proposed a tau multimer(3). Isolation of the seed-competent form of tau thus could be critical to understanding the initiation of disease.
Tau protein forms amyloids that underlie neurodegeneration in a variety of neuropathological syndromes, collectively termed tauopathies(4). These include AD and frontotemporal dementias, among many others. Multiple groups, including ours, have now observed that tau will propagate an aggregated state from the outside to the inside of a cell, between cells, across synapses, and within brain networks(5). In prior work we used size exclusion chromatography (SEC) to define tau trimers as the minimal unit of spontaneous cellular uptake and intracellular amyloid formation, and proposed this as the smallest particle capable of propagating aggregates between cells(6). This work involved application of “naked” protein assemblies derived from recombinant protein or human brain onto cultured “biosensor” HEK293 cells or primary neurons that express a tau aggregation reporter(7,8). These biosensor cells take up tau aggregates via macropinocytosis(9). The aggregates subsequently serve as highly specific templates to trigger intracellular amyloid formation(8,10). We have also determined that preincubation of cationic lipids such as Lipofectamine with tau seeds will directly transduce them into the cell, bypassing the physiologic uptake mechanism(8,11). Lipofectamine-mediated delivery into biosensor cells allows direct quantitation of seed titer for both tau and α-synuclein(9).
Tau is intrinsically disordered upon isolation from bacteria or mammalian cells, and is relatively inert in terms of spontaneous self-assembly. However under various conditions, including exposure to polyanions such as heparin, tau will form aggregates via nucleated self-assembly (12,13). It is unknown how these experimental conditions relate to the initiation of aggregation in human brain. We have now purified various stable forms of full-length tau monomer from recombinant protein and human brain. One type is relatively inert, and is stable for long periods. Another type is “seed-competent,” triggers amyloid formation in cells and in vitro, and exhibits intrinsic properties of self-assembly. We have used crosslinking with mass spectrometry (XL-MS) to probe the structures of these molecules and build theoretical structural models. These models imply that differential exposure of hexapeptide motifs previously known to be important for amyloid formation distinguish the two forms of tau. Identification of distinct and stable forms of tau monomer, including structural isoforms that are uniquely seed-competent, bears directly on how we understand the initiation of protein aggregation in the tauopathies.
MATERIALS AND METHODS
Tau expression, purification, fibrillization, and labeling
We utilized several forms of recombinant tau. Full-length (FL), wild-type (WT) tau contains two cysteines that create disulfide bridges, and complicate isolation of monomer. Thus in addition to preparing FL WT tau (2N4R), we created FL tau (2N4R) that contains two cysteine/alanine substitutions (C291A, C322A), termed tau (2A), and prepared recombinant protein as previously described(14). Additionally, for fluorescence correlation spectroscopy (FCS), we engineered a single cysteine at the amino terminus of this construct (Cys-Tau (2A)) for labeling via sulfhydryl chemistry. These modified proteins have fibrillization and seeding properties similar to FL WT tau. To initiate fibrillization, we incubated tau in 10 mM HEPES, 100 mM NaCl, and 8 µM heparin (1:1 ratio of FL tau to heparin) at 37°C for 72 h without agitation. For cysteine labeling, we incubated 200 µL of 8 µM fibrils (monomer equivalent) and monomer with 0.025 mg of Alexa Fluor-488 (AF488) C5-maleimide (Invitrogen) and 80µM Tetramethylrhodamine-5-maleimide (Sigma-Aldrich) overnight at 4°C with gentle rotation. We quenched excess dye with 10mM DTT for 1h at room temperature. We employ the following terminology to refer to four types of recombinant tau monomer, each comprised of identical primary amino acid (aa) sequences:
Mr: recombinant tau prepared from E. coli, without any further modification or size fractionation.
Mh: recombinant tau treated with heparin for 1min.
Mi: recombinant tau, subsequently isolated by SEC.
Ms: recombinant tau, treated with heparin, allowed to form fibrils, then sonicated, and isolated by SEC.
Sonication and size exclusion chromatography (SEC)
We sonicated labeled and non-labeled fibrils using a Q700 Sonicator (QSonica) at a power of 100-110 watt (Amplitude 50) at 4°C for 3h. Samples were then centrifuged at 10,000 x g for 10 min and 1 mL of supernatant was loaded into a Superdex 200 Increase 10/300 GL column (GE Healthcare) and eluted in PBS buffer at 4°C. After measuring the protein content of each fraction with a Micro BCA assay (Thermo Scientific) and/or fluorescence using a plate reader (Tecan M1000), we aliquoted and stored samples at -80°C until further use. Each aliquot was thawed immediately before use. The molecular weight of proteins in each fraction was estimated by running gel filtration standards (Bio-Rad): Thyroglobulin (bovine) 670 kDa; γ-globulin (bovine) 158 kDa; Ovalbumin (chicken) 44 kDa; myoglobin (horse) 17 kDa; and vitamin B12 1.35 kDa.
CD spectroscopy
Circular dichroism (CD) measurements were performed at 25°C on a Jasco J-815 spectropolarimeter using a 0.1 cm optical path length. 200µL of 2 µM Ms or Mi monomer was dialyzed onto 10 mM NaP and the spectra were measured at 0.10 nm intervals, with a band width of 1.0 nm, and scan speed of 10 nm/min. The spectrum represents the average of 4 scans in the range of 195 to 250 nm.
Enzyme linked immunosorbent assay
A total tau “sandwich” ELISA was performed similarly to that described previously(15). Antibodies were kindly provided by Dr. Peter Davies (Albert Einstein College of Medicine). 96-well round-bottom plates (Corning) were coated for 48 hours at 4°C with DA-31 (aa 150-190) diluted in sodium bicarbonate buffer (6 µg/mL). Plates were rinsed with PBS 3 times, blocked for 2 hours at room temperature with Starting Block (Pierce), and rinsed with PBS 5 additional times. SEC fractions were diluted in SuperBlock solution (Pierce; 20% SuperBlock, diluted in TBS), and 50 µL sample was added per well. DA-9 (aa 102-150) was conjugated to HRP using the Lighting-Link HRP Conjugation Kit (Innova Biosciences), diluted 1:50 in SuperBlock solution, and 50µL was added per well (15µg/mL). Sample + detection antibody complexes were incubated overnight at 4°C. Plates were washed with PBS 9 times with a 15 sec incubation between each wash, and 75 µL 1-Step Ultra TMB Substrate Solution (Pierce) was added. Plates were developed for 30min, and the reaction quenched with 2M sulfuric acid. Absorbance was measured at 450nm using an Epoch plate reader (BioTek). Each plate contained a standard curve, and all samples were run in triplicate.
Fluorescence correlation spectroscopy
FCS measurements were conducted on a Confocal/Multiphoton Zeiss LSM780 Inverted microscope (Carl Zeiss-Evotec, Jena, Germany), using a 40X water immersion objective as previously described (16). Fluorescently labeled tau from SEC fractions (in PBS) was excited at 488nm and 561nm for 30sec, recording 10 times(17). The data analysis was performed with Origin 7.0 (OriginLab, Northampton, MA).
Liposome-mediated transduction of tau seeds
Stable cell lines were plated at a density of 35,000 cells per well in a 96-well plate. After 18h, at 60% confluency, cells were transduced with protein seeds. Transduction complexes were made by combining [8.75 µL Opti-MEM (Gibco) +1.25 µL Lipofectamine 2000 (Invitrogen)] with [Opti-MEM + proteopathic seeds] for a total volume of 20µL per well. Liposome preparations were incubated at room temperature for 20min before adding to cells. Cells were incubated with transduction complexes for 24h.
FRET flow cytometry
Cells were harvested with 0.05% trypsin and fixed in 2% paraformaldehyde (Electron Microscopy Services) for 10min, then resuspended in flow cytometry buffer. The MACSQuant VYB (Miltenyi) was used to perform FRET flow cytometry. To measure CFP and FRET, cells were excited with a 405nm laser, and fluorescence was captured with 405/50nm and 525/50nm filters, respectively. To measure YFP, cells were excited with a 488nm laser and fluorescence was captured with a 525/50nm filter. To quantify FRET, we used a gating strategy similar to that previously described(8). The integrated FRET density (IFD), defined as the percentage of FRET-positive cells multiplied by the median fluorescence intensity of FRET-positive cells, was used for all analyses. For each experiment, ∼20,000 cells were analyzed in triplicate. Analysis was performed using FlowJo v10 software (Treestar).
Tau seeding in vitro
Recombinant full length (0N4R) tau monomer was purified as previously described(18) at 1mg/mL in BRB80 buffer (80mM PIPES, 1mM MgCl2, 1mM EGTA, pH 6.8 with 0.3M NaCl) and boiled at 100°C for 5min with 25mM β-mercaptoethanol. The tau protein solution was then rapidly diluted 1:5 and cooled to 20°C in PBS, pH 7.4, to a final concentration of 0.2mg/mL of tau and 5mM β-mercaptoethanol. This solution was supplemented with Thioflavin T (ThT) to a final concentration of 20µM and filtered through a sterile 0.2µm filter. Reaction sizes of 195µL were aliquoted from the prepared protein stock and thoroughly mixed with 5µL of each sample at 100nM monomer equivalent, or 5µL of buffer control. For each sample, three different technical replicates were prepared. An opaque 96-well plate was prepared with a 3mm glass bead added to each well to increase agitation. The recombinant tau solution was added to the plate in 200µl reaction volumes. The plate was sealed with sealing tape to prevent evaporation and incubated in the plate reader (SpectraMax M2) at 37°C. ThT fluorescence was monitored over time with excitation and emission filters set to 444nm and 485nm, respectively. Fluorescence readings were taken every 5min, with agitation for 5sec before each reading.
Tau extraction from brain and characterization by SEC
0.5g frontal lobe sections from AD patients and age-matched controls lacking evident tau pathology were gently homogenized at 4°C in 5mL of TBS buffer containing protease inhibitor cocktails (Roche) using a dounce homogenizer. Samples were centrifuged at 21,000 x g for 15 min at 4°C to remove cellular debris. Supernatant was partitioned into aliquots, snap frozen and stored at -80°C. Immunopurification was performed with HJ8.5 at a ratio of 1:50 (1µg mAb per 50µg of total protein), incubating overnight at 4°C while rotating. To each 1mL of mAb/brain homogenate we added 200µL of a 50% slurry protein G-agarose beads (Santa-Cruz). We washed the bead with TBS buffer before overnight incubation at 4°C. We then centrifuged the complexes at 1000 x g for 3min and discarded the supernatant. Beads were washed with Ag/Ab Binding Buffer, pH 8.0 (Thermo Scientific) three times. Tau bound to the beads was eluted in 100 µL low pH elution buffer (Thermo Scientific), incubated at room temperature for 7min, followed by neutralization with 10µL Tris-base pH 8.5. This elution step was repeated once more with 50 µL elution buffer and 5µL Tris-base pH 8.5 for a total of 165µL. Samples were then centrifuged at 10,000 x g for 10min, and the supernatant loaded onto a Superdex 200 Increase 10/300 GL column (GE Healthcare). SEC fractions were frozen at -80°C after evaluation of protein content by Micro BCA assay (Thermo Scientific).
To compare different extraction methods, fresh frozen frontal lobe section from an AD patient was suspended in TBS buffer containing protease inhibitor cocktails (Roche) at 10% w/vol in 4 portions. Samples were homogenized using 3 different devices: a dounce homogenizer, probe sonicator (Omni International), and tissue homogenizer (Power Gen 125, Fischer Scientific). We also included one more condition of homogenizing with tissue homogenizer followed by probe sonication for 10min. Samples were centrifuged at 21,000 x g for 15min at 4°C to remove cellular debris. Supernatant was partitioned into aliquots followed by immunopurification.
Analysis of heat denaturation data
We analyzed the IFD from measurements of temperature dependent seeding using global fits to a proposed unimolecular heat denaturation reaction. This analysis rests on the Arrhenius equation(19): where kU is the unfolding rate constant, E is the activation energy, R is the gas constant, T is the temperature, and A is the pre-exponential factor. For the unimodal model, the data were fit globally to:
Here, t is the heat denaturation time and τ = 1/kU is the unfolding time. A second, multimodal model was deployed to account for discrepancies in the early time points which appeared to suggest the presence of a lag phase in denaturation. In this model, the data were fit globally to where lt is the lag time given by and B is a pre-exponential factor. We used the Akaike information criterion (AIC) to evaluate the best model as it quantifies the trade-off between goodness of fit and the complexity of the model (20). For least squares model fitting, AIC can be reduced to: where p is the number of parameters in the model, n is the number of observations, and RSS is the residual sum of squares. The preferred model is the one with the minimum AIC. Here, we find AIC = 123 for the unimodal model and AIC = 105 for the multimodal model, which suggests the multimodal model is a better description of the denaturation data.
Crosslinking, sample processing and LC-MS/MS analysis
Mi and Ms tau samples were prepared as described above. In all cases, tau preparations were crosslinked at a total protein concentration of ∼0.1mg/mL using 10 – 20µg starting material. The crosslinking buffer was 50 mM HEPES-KOH (pH 7.4) containing 300mM NaCl and 1mM DTT. The crosslinking reaction was initiated by adding disuccinimidyl suberate (DSS) stock solution (25 mM DSS-d0 and –d12, Creative Molecules) in DMF to a final concentration of 1 mM. Samples were incubated at 37°C for 1min. For the Mh sample, heparin sulfate (Sigma) was added to a final concentration of 5µM, followed by 1mM DSS and the samples were inclubated for 1min at 37°C. Excess reagent was quenched by addition of ammonium hydrogen carbonate to 50mM and incubation at 37°C for 30min, and then flash frozen at -80°C. After the quenching step, samples were evaporated to dryness in a vacuum centrifuge and resuspended in 8M urea. Proteins were reduced with 2.5mM TCEP (37°C, 30 min) and alkylated with 5mM iodoacetamide (30min, room temperature, protected from light). The sample solutions were diluted to 1M urea with 50mM ammonium hydrogen carbonate and trypsin (Promega) was added at an enzyme-to-substrate ratio of 1:50.Proteolysis was carried out at 37°C overnight followed by acidification with formic acid to 2%(v/v). Samples were then purified by solid-phase extraction using Sep-Pak tC18 cartridges (Waters) according to standard protocols. Samples were fractionated by size exclusion chromatography (SEC) on a Superdex Peptide column as described elsewhere (21). Two fractions collected from SEC were evaporated to dryness and reconstituted in water/acetonitrile/formic acid (95:5:0.1, v/v/v) to a final concentration of approximately 0.5 µg/µl. 2µL each were injected for duplicate LC-MS/MS analyses on an Eksigent 1D-NanoLC-Ultra HPLC system coupled to a Thermo Orbitrap Fusion Tribrid system. Peptides were separated on self-packed New Objective PicoFrit columns (11cm x 0.075mm I.D.) containing Magic C18 material (Michrom, 3µm particle size, 200Å pore size) at a flow rate of 300nL/min using the following gradient. 0-5min = 5 %B, 5-95min = 5-35 %B, 95-97min = 35-95 %B and 97-107min = 95 %B, where A = (water/acetonitrile/formic acid, 97:3:0.1) and B = (acetonitrile/water/formic acid, 97:3:0.1). The mass spectrometer was operated in data-dependent mode by selecting the five most abundant precursor ions (m/z 350-1600, charge state 3+ and above) from a preview scan and subjecting them to collision-induced dissociation (normalized collision energy = 35%, 30ms activation). Fragment ions were detected at low resolution in the linear ion trap. Dynamic exclusion was enabled (repeat count 1, exclusion duration 30sec).
Analysis of mass spectrometry data
Thermo .raw files were converted into the open .mzXML format using msconvert (proteowizard.sourceforge.net) and analyzed using an in-house version of xQuest(22). Spectral pairs with a precursor mass difference of 12.075321 Da were extracted and searched against the respective FASTA databases containing Tau (TAU_HUMAN P10636-8). xQuest settings were as follows: Maximum number of missed cleavages (excluding the crosslinking site) = 2, peptide length = 5-50 aa, fixed modifications = carbamidomethyl-Cys (mass shift = 57.021460 Da), mass shift of the light crosslinker = 138.068080 Da, mass shift of mono-links = 156.078644 and 155.096428 Da, MS1 tolerance = 10 ppm, MS2 tolerance =0.2 Da for common ions and 0.3 Da for crosslink ions, search in ion-tag mode. For brain derived samples we also included variable modifications including: Methionine oxidation = 15.99491, Ser/Thr/Tyr Phosphorylation = 79.96633 and Lysine Ubiquitylation = 114.043 with nvariable_mod = 1. Post-search manual validation and filtering was performed using the following criteria: xQuest score > 7, mass error between -4 and +7ppm, %TIC > 10, and a minimum peptide length of six aa. In addition, at least four assigned fragment ions (or at least three contiguous fragments) were required on each of the two peptides in a crosslink. False discovery rates for the identified crosslinks were estimated using xprophet(22). Crosslink data was visualized using Xvis(23).
Generation of structural models using XL-MS-derived constraints
High confidence crosslink pairs identified above were used to generate an ensemble of possible structures using a Rosetta protocol employing the crosslink pairs as structural restraints. The integration of XL-MS derived restraints have been previously used to refine structural models of large complexes(21) and simpler heterodimeric complexes(24). Based on distance distributions of crosslink pairs mapped onto crystallographic structures we set a lower bound of 15Å and an upper bound of 25Å for lysine Cα pairs in our simulations.Importantly, in our simulations we weighted the constraint pairs as to allow some distances above the upper bound limit. The fragment library was supplanted by using chemical shifts derived from tau (bmrb entry 17920) using csrosetta(25). We generated 1000 models for each of the four XL-MS datasets on a high performance cluster (biohpc.swmed.edu). Because crosslinks were absent in the N- and C-termini we selected models that excluded additional contacts from residues 1-150 and 400-441. Low energy structures with radius of gyrationFL/Frag ratios >2 and contact order FL/Frag ratios between 0.9 and 1.1 were further analyzed. Relative contact order (i.e. intramolecular crosslink distance) was computed using a perl script and is defined as the average sequence distance between all pairs of contacting residues normalized by the total sequence length. Contact maps were computed using a python script and all figures were generated using Pymol. All plots were generated using gnuplot.
Commandline used for ab initio protocol calculations with XL-MS restraints
AbinitioRelax.default.linuxgccrelease -in:file:fasta tau.fasta -file:frag3 tau.frags3.dat -file:frag9 tau.frags9.dat -nstruct 1000 -abinitio::increase_cycles 0.5 -abinitio::relax -score::weights score13_env_hb -abinitio::rg_reweight 0.5 -abinitio::rsd_wt_helix 0.5 -abinitio::rsd_wt_loop -disable_co_filter true -out:file:silent csrosetta.out -constraints:cst_fa_file tau.cst - constraints:cst_file tau.cst -constraints:cst_weight 0.1 -constraints:cst_fa_weight 0.1 - loopfcst::coord_cst_weight 10.0
Statistical analysis
Group mean values were analyzed by one-way ANOVA with Bonferroni post hoc significant differences test using GraphPad prism 5 software. Data in text and figures are represented as mean ± SEM.
RESULTS
Isolation of monomer from unfibrillized or fibrillized sources
We initially sought to define the tau seeding unit that would trigger intracellular aggregation upon direct delivery to the cell interior. We had previously observed that a tau trimer is the minimal assembly size that triggers endocytosis and intracellular seeding(6). These experiments depended on spontaneous cell uptake, since no Lipofectamine was added to the reactions. A prior study had also indicated the role of disulfide linkages in promoting tau aggregation, potentially by dimer formation(3). Thus, for our initial studies we engineered and purified full-length (FL) tau monomer that lacks any internal cysteines due to alanine substitution (C299A and C322A), termed tau (2A). FL tau (2A) cannot self-associate based on disulfide linkages, which helped prevent the formation of cryptic dimers that could confound our studies. These substitutions do not affect tau purification, heparin-induced fibrillization, and sonication protocols, which we performed as described previously(6). We covalently labeled the fibril preps prior to sonication and isolation of recombinant FL tau (2A) assemblies of various sizes by size exclusion chromatography (SEC)(6). In parallel, we also studied FL wild type (WT) tau. We used methods identical to those described previously, which included controls of fluorescence correlation spectroscopy (FCS), and crosslinking to confirm purification of bona fide monomer(6). We purified unfibrillized recombinant FL tau (2A) monomer by SEC (Fig. 1A), and isolated SEC fractions of sonicated fibrils that contained putative monomer, dimer, trimer and ∼10-mer (Fig. 1B).
Fibril-derived monomer exhibits seeding activity in cells and in vitro
To test the seeding activity of the tau preparations, we used a previously described “biosensor” cell reporter line(8).These cells stably express 4R tau repeat domain (RD) containing the disease-associated P301S mutation fused to cyan and yellow fluorescent proteins (tau-CFP/YFP). Exogenously applied seeds induce intracellular aggregation with resultant fluorescence resonance energy transfer (FRET) measured via flow cytometry(8,11). The degree of aggregation is scored using “integrated FRET density” (IFD), which is the product of the percent positive cells and the mean fluorescence intensity of FRET-positive cells, and from this we determine a titer of tau seeding activity(8). Lipofectamine directly transduces tau assemblies across the plasma membrane and increases the assay’s sensitivity by approximately 100-fold. Upon incubation with Lipofectamine, we readily observed seeding by larger assemblies, whether FL WT or 2A. Surprisingly, we also observed seeding activity in the monomer fractions (Fig. 1C). There was no difference between FL WT tau, and FL tau (2A) (Fig. 1C). We termed the inert monomer “Mi,” and the seed-competent monomer “Ms.” We confirmed our observations using epifluorescence microscopy, which revealed induction of intracellular inclusions after exposure of cells to Ms, but not Mi (Fig. 1D). These surprising results inspired us to check whether heparin itself could lead to the formation of a seed-competent monomer. We exposed FL WT tau to heparin for varying amounts of time, before purifying different assembly sizes by SEC and testing for seeding activity. After 15min of heparin exposure, we detected low but significant amounts of seed-competent monomer, while larger assemblies were more rare (Fig. 1E). Recombinant monomer not treated with heparin had no seeding activity at any time point (Fig. 1E). At later time points (1h, 4h) monomer fractions as well as larger assemblies all had strong seeding activity (Fig. 1E). To rule out an artifact of the Lipofectamine transduction, we tested FL (2A) tau preparations in an in vitro seeding assay that induces fibril formation by full-length tau (0N4R) through iterative polymerization and agitation steps(18). Mi had no intrinsic seeding activity. However Ms induced amyloid formation, albeit more slowly than trimer or unfractionated fibrils (Fig. 1F). We concluded that the Ms fraction contains seeding activity that enables intracellular aggregation of tau RD-CFP/YFP in cells, or full-length tau in vitro, and that heparin-exposed monomer also developed seeding activity prior to its formation of larger assemblies.
Comparison of Mi and Ms by CD and FCS
We were intrigued to observe seeding activity in a monomer fraction, and thus attempted to detect multimers. To begin, we tested for obvious structural differences between Mi and Ms using CD spectroscopy, which revealed none (Fig. 2A). We confirmed the sizes of species isolated after sonication using FCS, which measures particle diffusion through a fixed volume (Fig. 2B). As we previously observed(6), we accurately estimated the sizes of small assemblies (=10-mer), but not larger assemblies (>10-mer) (Fig. 2B). Next, we used double-label FCS in an attempt to identify multimers within the Ms fraction. We engineered a cysteine onto the amino terminus of FL tau (2A) to enable covalent modification (Cys-Tau (2A)). We then prepared Cys-tau (2A) fibrils, or monomer, and labeled them simultaneously with Alexa488 (green) and tetramethylrhodamine (TMR, red) via maleimide chemistry. We carried out sonication and purification by SEC as before, isolating assemblies of various sizes. We evaluated each for cross-correlation between red and green signal, which indicates the presence of at least two tau molecules in a particle. We analyzed >300 events for each assembly. When we evaluated Mi and Ms, 100% of events in each case showed a diffusion time consistent with a molecule of ∼50kD, which corresponds to the tau monomer (Fig.2C,D). Furthermore we observed no cross-correlation between red and green signal, indicating that neither preparation had detectable multimeric assemblies (Fig. 2C,D,H). By contrast, when we evaluated larger species such as dimer, trimer, or ∼10-mer, we observed longer diffusion times consistent with the predicted assembly sizes, and significant cross-correlation values (Fig. 2E-H), consistent with the presence of multimers. In summary, we found no evidence of multimers in the Ms fraction using FCS.
SEC preparation efficiently purifies Ms monomer
To test for contamination of the Ms preparation with larger seed-competent assemblies during SEC, we took three approaches. First, we determined what degree of contamination would be required to give a substantial signal. We titrated Ms to determine an EC50 of ∼10nM (Fig.3A), and also titrated dimer and trimer into a solution of 100nM Mi (assuming that those species would most likely account for contamination, if there were any)(Fig. 3B). We observed similar seeding efficiencies from dimer and trimer, with an EC50 of ∼10nM monomer equivalent (Fig. 3B). In our estimation, these data indicated that to account for signal observed in the seeding assay, contamination of an otherwise inert monomer with larger seed-competent assemblies would have to be substantial.
We next tested the ability of SEC to exclude larger seeds from the monomer fraction. We first isolated Ms and larger assemblies from a sonicated fibril preparation. Removing the fraction that contained Ms (B5), we then pooled the remaining fractions, and spiked them with Mi. We re-fractionated the material on SEC to isolate again the monomer in fraction B5 (Fig. 3C). As previously observed, Ms and other fibril-derived assemblies had seeding activity (Fig. 3D). However, in the second case, while we observed seeding activity in larger assemblies, the monomer (which we take to be Mi) re-isolated from a pool of larger fibril-derived assemblies had no seeding activity (Fig. 3E). This confirmed that larger, seed-competent assemblies do not appreciably contaminate the monomer fraction during SEC.
Finally, we used heat-mediated dissociation of oligomeric assemblies to test for the possibility that Ms in fact represented a uniquely compact multimer that somehow purifies as a monomer. We collected Ms by SEC, and heated the sample to 95°C for 3h. We then re-isolated the sample via SEC. We carried out the same procedure with trimer and ∼20-mer. In each case, we tested the resultant fractions for seeding activity. In the first instance, after heating we re-isolated Ms purely as monomer that retained its seeding activity (Fig. 3F). The trimer assembly (fraction B8) broke down to smaller assemblies, predominantly monomer, each of which retained seeding activity (Fig. 3G). The ∼20-mer (fraction A5) was largely stable following heat treatment, and retained its seeding activity (Fig. 3H). Based on these experiments, we concluded that our purification protocols leave virtually no contamination of Ms by larger assemblies, and that all detectable seeding activity of Ms in fact derives from monomer.
Differential heat lability of tau assemblies
In the preceding experiment Ms retained seeding activity even after 3h at 95°C, a condition sufficient to dissociate trimers. These experiments implied that Ms consists of a surprisingly stable seed-competent structure, largely resistant to heat denaturation. Consequently, we used more nuanced heat denaturation of seeding activity to probe the relative stabilities of Ms, dimer, trimer, and larger assemblies of FL WT tau. We first isolated tau monomer, dimer, trimer, ∼10-mer, and ∼20-mer on SEC. We then incubated the various assemblies at a range of temperatures (65, 75, 85, 95°C) and times (0, 3, 12, 18, 24, 48, 72h) before measuring seeding activity. Lower temperatures only slightly reduced seeding activity, whereas exposure of Ms, dimer, and trimer to temperatures ≥85°C for 18-24h eliminated most of it (Fig. 4A-D). By contrast, the seeding activities of ∼10-mer and ∼20-mer were relatively heat-resistant (Fig. 4A-D). This is consistent with our prior observations that tau seeds derived from cultured cells are resistant to boiling (10). Ms, dimer, and trimer lost seeding activity at roughly the same rate, while larger assemblies remained intact. To determine a putative energy barrier between Ms and an inert form, we evaluated the denaturation data for Ms, comparing two models for the transition of Ms to an inert form (which we assumed to be an unfolding reaction): a unimodal unfolding model vs. a multimodal model that assumes intermediate seed-competent states. The unimodal model did not account for the data at early time points, which indicated a lag phase in denaturation, whereas the multimodel model performed better (Fig. 4E). The lag phase in denaturation implies an ensemble of seed-competent states that define Ms, each separated by smaller barriers. Using this multimodal model, we calculated the barrier to conversion of Ms to an inert form to be ∼18 kcal / mol.
Ms has unique properties of self-assembly
Aggregation of Mi in vitro is relatively slow, requires high protein concentration (micromolar), and polyanions such as heparin(12,13). Based on the seeding activity of Ms we predicted that it might more readily self-associate. We incubated FL WT tau Mi and Ms alone, or dimer or trimer at equimolar ratios, keeping total particle concentration constant at 500nM. We then monitored change in assembly size over 24h. Mi, dimer, and trimer showed no evidence of self-association in this timeframe (Fig. 5A,C,D). By contrast, when incubated alone, Ms readily formed larger assemblies (Fig. 5B). When we incubated Mi with dimer or trimer, we saw no change in the assembly population over 24h (Fig. 5E,F). By contrast, when we mixed Ms with dimer or trimer we observed a growth of larger assemblies with a concomitant reduction in dimer and trimer peaks (Fig. 5G,H). We conclude that Mi, dimer, and trimer do not form larger assemblies at an appreciable rate, while Ms self-assembles or adds on to larger assemblies.
AD brain contains seed-competent monomer
Given our experiments with Mi and Ms, we wished to test whether similar structures exist in vivo, and thus isolated tau monomer from AD and control patient brains. We extracted brain samples using a dounce homogenizer gently to try to avoid liberating significant monomer from fibrils. We immunoprecipitated tau using an antibody that targets the amino-terminus, and resolved the eluates by SEC, followed by ELISA to determine tau levels (Fig. 6A,B). Whereas tau from control brain extract eluted in the monomer fraction (Fig. 6A), tau from AD brain distributed across multiple fractions, corresponding to monomer and larger assemblies (Fig. 6B). When we tested each fraction for seeding activity, we observed none in any control brain fraction, including monomer (Fig. 6C). However all AD fractions contained seeding activity, including monomer (Fig. 6C). A seed-competent tau monomer is thus present in AD brain, but not in control brain. To test for its self-association in vitro, we incubated control and AD monomer for up to 24h. We then resolved the assemblies via SEC and tested each fraction for seeding activity. At 0h, AD monomer exhibited seeding activity (Fig. 6D). Over time we observed a transition to larger seed-competent assemblies (Fig. 6E, F). Tau monomer derived from AD brain thus has an intrinsic capacity for self-association into seed-competent assemblies.
XL-MS reveals distinct, stable conformational states in recombinant tau monomer
To probe the structures of various forms of recombinant tau monomer, we employed crosslinking with mass spectrometry (XL-MS). This method creates restraints for structural models of single proteins or protein complexes(21,26,27). We studied FL WT tau monomer from a variety of conditions: recombinant protein purified from E. coli (Mr); the same protein exposed to heparin for 1min and purified by SEC (Mh); recombinant monomer subsequently purified by SEC (Mi); and seed-competent monomer purified after sonication of fibrils (Ms). In each case, we performed reactions at low protein concentrations and short crosslink times, avoiding inter-protein crosslinks. We reacted samples with DSS, digested them with trypsin, enriched crosslinked peptides with by SEC, and analyzed them by capillary liquid chromatography tandem mass spectrometry. We assigned the complex fragment ion spectra to the corresponding peptide sequences using xQuest (22). Denaturation of recombinant tau with 8M urea prior to crosslinking produced no intramolecular cross-links (data not shown), indicating that crosslinks observed under native conditions represent local structure.
Given our observation that four forms of WT FL tau (Mr, Mh, Mi, Ms) had two distinct activities (inert vs. seed-competent), we compared their structures using XL-MS. Mr and Mh, despite different seeding activities, exhibited relatively similar intramolecular crosslink patterns. These were dominated by short-range crosslinks (Fig. 7A,B). In Mr, we identified 21 unique crosslinks evenly distributed within the repeat domain (RD: aa243-367, composed of four repeated ∼30aa sequences, RD1-4), and the projection domain (PD: aa1-242), with an average sequence separation of 16.5 aa between modified lysines (Fig. 7A; S1A; Table S1). In Mh, 57 unique crosslinks had a similar average sequence separation of 16 aa between modified lysines (Fig. 7B; S1B; Table S1). We observed an increased abundance of short range crosslinks clustered into two main regions: PD (aa130-175), RD2 (aa270-290), and RD4 (aa350-400) (Fig. 7B, S7B and Table S1). Mi exhibited 7 crosslinks, and Ms exhibited 8 (Fig. 7C,D; S1C,D; Table S1). Mi and Ms shared only two common crosslinks, K224-K347 and K225-K347, and crosslinks differed within the RD and in regions just upstream and downstream (Fig. 7C,D). The average sequence separation for Mi was 113 aa, and for Ms it was 136 aa. We note differences between Mi (purified by SEC), and Mr (which was not). This may be due to residual cell components (e.g. RNA) bound to Mr. In summary, we observed mostly short-range crosslink patterns for Mr and Mh with different distributions in the RD. Importantly, the crosslink pattern of Mh was very different from that of Ms, indicating that heparin exposure alone does not account for the structure of Ms.
To test patient-derived tau monomer, we immunopurified tau from brains of 3 AD brains and 3 age-matched controls, using the established SEC protocol to purify seed-competent monomer followed by our XL-MS protocol. In control tau we observed 5-7 intramolecular crosslinks (Fig. 7E; Table S1) and in AD-derived tau we observed 10-13 intramolecular crosslinks (Fig. 7F; Table S1). Similar to recombinant Ms, in all AD tau monomer samples we observed a discrete set of crosslinks between aa150 and aa259-290 (Fig. 7E, red lines). To evaluate the variability in the preparation of the brain derived samples more precisely, we homogenized tissue from a single AD brain using different approaches ranging in severity: dounce homogenization, pulse sonication, mechanical homogenization, and mechanical homogenization with pulse sonication (Fig. S2A-D, Table S1). While there was variation in the total number of crosslinks (3, 11, 57, and 8, respectively) the core set of crosslinks between aa150 and 259-290 were present independent of homogenization method (Fig.S2A-D, red lines). The consistency of the crosslinks between aa150 and 259-290 focused our attention on the domains they implicated, especially the amyloid-forming sequences within the RD.
Models of seed-competent monomers suggest exposure of VQIINK and VQIVYK
To understand how core elements of tau might play a role in its aggregation, we employed ROSETTA to create models of tau structure for Mr, Mh, Mi and Ms, using restraints from the crosslink patterns, and length of the DSS crosslinker (Fig. 8A-D, Table S1, Supp. Movies: Mr,Mh, Mi, Ms). We prioritized low energy structures that excluded additional contacts with the N- and C-termini, as suggested by the crosslink patterns (see methods for details). We next calculated protein contact maps for representative Mr and Mi (inert), and Mh, and Ms (seed competent) structures (Fig. S3A-D), which highlight how the XL-MS derived restraints drive the topology of each tau structure (Fig. S3E-H, green dots). The Mr structural model was based on short range contacts clustered throughout the entire RD (Fig. 8A), and this predicted masking of VQIINK and VQIVYK sequences, which have previously been proposed as critical for tau amyloid formation(28,29). The model of Mi structure was based on a dramatically different crosslink pattern, with the RD1/2 (Fig. 8C, red-green interface) and RD2/3 (Fig. 8C, green-blue interface) regions in discretely formed hairpins similar to the conformation of tau bound to microtubules indicated by NMR (30). In this model, the core elements of VQIINK (green spheres) at the beginning of the RD2, and VQIVYK (blue spheres) at the beginning of the RD3(29) are buried in hairpins (Fig. 8C, S3C). Remarkably, despite different crosslink patterns, our models suggested that both Mr and Mi feature VQIINK/VQIVYK sequences relatively obscured from solvent accessibility. Crosslink studies of tau monomer from control patients also suggested a structural model in which these core residues are less accessible (Fig. 8E, Table S1; Supp. Movie: Control 1), although we place important caveats on these interpretations. Our ability to resolve high confidence structural models using XL-MS in patient-derived samples is more limited due to fewer high confidence crosslinks, possibly from sample heterogeneity (multiple isoforms, and multiple tau structures).
For seed-competent forms of tau, and despite distinct crosslink patterns, we also observed common predicted effects on VQIINK/VQIVYK exposure. The Mh structural model was predominantly influenced by short-range crosslinks, consistent with local contacts within the PD and the RD2 that result in burial of the VQIINK sequence but exposure of the VQIVYK sequence (Fig. 8B; Supp. Movie: Mh). This is consistent with recent observations that heparin binds directly to residues 270-290 which include VQIINK (33). By contrast the model of Ms was influenced by more long-range crosslinks, which nonetheless predicted an exposed conformation of VQIINK/VQIVYK (Fig. 8D; Supp. Movie: Ms). Lastly, in AD brain-derived tau we also observed crosslinks between aa ∼150 to aa259-290, which we think help define an exposed configuration of VQIINK/VQIVYK in a seed-competent monomer (Fig. 8F, Table S1; Supp. Movie: AD1). In summary, despite profound differences in overall crosslink patterns, the models suggested by XL-MS highlight relative exposure of VQIINK/VQIVYK in forms of tau with intrinsic seeding activity (Mh, Ms), while forms of tau that are inert (Mr, Mi) feature VQIINK/VQIVYK buried in hairpins. These models will obviously require additional tests using advanced biophysical methods, but provide a plausible explanation for the distinct activities of tau monomer that we have observed.
Discussion
The simplest interpretation of our data is that consistent with it being an intrinsically disordered protein, tau monomer can occupy several unique and stable conformational states. One set of structures is relatively inert, while another has intrinsic ability to self-assemble, and acts as a template, or seed, for fibril growth in vitro and in cells. We confirmed with multiple controls that Ms, derived from sonicated fibrils, was in fact a monomer and not a larger assembly, and that there was no detectable cross-contamination of larger species into the putative monomer fraction upon SEC. Heat denaturation of seeding activity was consistent with a complex folding state for Ms, comprised of several seed-competent intermediates. Tau monomer purified from AD brain also had intrinsic seeding activity, and self-associated to produce seed-competent assemblies. Finally, we used XL-MS to compare conformations of different tau monomers. A model restrained by our XL-MS data suggested that VQIVYK and VQIINK sequences might assume an open configuration in Ms, Mh, and AD-derived monomer, all of which have intrinsic seeding activity. By contrast, the model suggested that Mi and Mr, and control brain monomer, all of which are inert, lack VQIINK/VQIVYK exposure. Taken together, these data establish a new concept for tau: this intrinsically disordered protein has multiple, stable monomeric states, functionally distinguished by the presence or absence of seeding activity.
Amyloid proteins form progressively larger assemblies over time, and it has been difficult to define the composition of the minimal seed. Mandelkow and colleagues studied tau aggregation in vitro and concluded that a seed of 8-12 molecules existed in their experimental system(3). By contrast, Kuret and colleagues posited an “intermediate” of tau that could subsequently initiate self-assembly, and their data, based on extrapolation of tau concentrations needed to enable development of thioflavin fluorescence in vitro, were consistent with a monomeric seed(1). Wetzel and colleagues also proposed that a monomer is the basis of a “thermodynamic nucleus” that templates the aggregation of synthetic polyglutamine peptides(31). However, no prior study has previously identified stable forms of tau that seed amyloid formation. This provides an opportunity to study the earliest events in aggregate formation.
The actual cause of tau aggregation in tauopathies is unknown. It has been proposed that dissociation of tau monomer from microtubules, possibly due to phosphorylation, allows self-association to form pathogenic assemblies(32). In this study, using a single source of recombinant protein, we define distinctly structured seed-competent and inert forms of tau. We have similarly identified seed-competent species in patient-derived preparations. We fully recognize that in reality “seed-competent” and “inert” forms of tau represent multiple structural ensembles separated by defined energy or kinetic barriers. The barrier blocking conversion of an inert to a seed-competent form of tau can apparently be overcome by incubation with heparin and/or incorporation into a fibril. In neurons, other factors such as post-translational modifications and heterologous binding events likely play a role. Identification of the factors that trigger conversion from inert to seed-competent forms will thus have obvious implications for understanding disease mechanisms.
Isolation of seed-competent monomer from AD brain, with a very mild purification that explicitly excludes sonication or vigorous tissue homogenization, strongly suggests that this form of monomer freely exists in vivo. Furthermore, we observed that both recombinant Ms and AD -derived monomer build multimeric assemblies in vitro far more efficiently than Mi or control-derived monomer. Thus we hypothesize that a uniquely structured form of tau may be required for efficient assembly growth in cells. This contrasts with the idea that multimeric assemblies uniquely stabilize the conformation of otherwise unstructured proteins as they incorporate into the growing fibril. Taken together, we imagine that the initiation of aggregation in human brain might begin with conversion of tau monomer from an inert to a seed-competent form. To fully study this process will require more extensive biochemical purification of tau monomer from the earliest stages of disease.
Ms has a remarkably stable structure, as it resists heat denaturation at 95°C for up to 3h. This suggests a heretofore unrecognized conformation of tau that, to account for its slow heat denaturation, likely involves multiple intra-molecular interactions involving short and long range amino acid contacts. XL-MS provides some indication of what these might be, and indicates a predominance of more long-range intramolecular interactions in Ms vs. Mi. In agreement with the XL-MS results, we observed that heat inactivation of Ms seeding activity occurs with a lag phase, rather than first order time-dependent decay. This is implies a complex tertiary structure in which Ms has multiple seed-competent intermediates. Future XL-MS studies performed at different temperatures could reveal these structures. With more advanced methods to interrogate the structure of monomeric tau in patient material, we imagine that “seed-competent monomer” will in fact represent myriad structures, depending on the underlying disease. This could provide an explanation for how a single protein might self-assemble into diverse amyloid strains.
Without further studies to identify structures of tau at higher resolution, we cannot know for certain why one form acts as a seed and the other does not. However, we gained important insights when we modeled tau structures using ROSETTA, using XL-MS results as restraints. Despite obvious caveats, our models predict that the local environment surrounding two hexapeptide motifs, VQIINK and VQIVYK, which are required for tau to form amyloid structures, may explain the differences between seed-competent and inert forms. In the models of Mi and Mr, and control brain-derived tau, these motifs lie buried in hairpin structures, whereas in Ms and AD-derived tau, both are exposed, and in Mh VQIVYK is exposed. VQIINK/VQIVYK thus might serve as substrates for intermolecular interaction in a growing assembly. Interestingly, Ms has a very different pattern of crosslinks compared to the Mh, yet both seed aggregation. Indeed, it has been recently observed that heparin binding involves residues spanning 270-290, and promotes expansion of the remainder of the molecule (33), which is consistent with our predictions of relative exposure of VQIINK/VQIVYK. We hypothesize that conformers of tau monomer differentially present the core VQIINK/VQIVYK sequences. The diversity of exposed core elements (potentially beyond VQIINK/VQIVYK) could specify the formation of assemblies that give rise to distinct strains. Consistent with this idea, a recent structural study indicates that in AD-derived tau fibrils the VQIVYK sequence plays a key role in the core amyloid structure along with adjacent amino acids, but the VQIINK sequence does not (34). We also note that multiple disease-associated mutations in tau affect residues in close proximity to VQIINK/VQIVYK. For example, our models predict that serine or leucine substitutions at P301 (which cause dominantly inherited tauopathy) would destabilize the local structure, and promote exposure of the VQIINK/VQIVYK sequences. Future experiments will test these ideas more definitively.
Conclusion
Our findings indicate that tau monomer has at least two dominant structural configurations, one of which has intrinsic activity as a seed and a self-assembly factor. We cannot exclude other models of seed formation based on these studies, and they may also play a key role in amyloid formation. Thus even temporary, critical local elevations in the concentration of an aggregation-prone protein may enable clustering, with an as yet unspecified conformational transition to form a seed. This would be consistent with a “molten globule” as the origin of a seed, perhaps reflecting the multimeric tau assembly described originally by Mandelkow and colleagues (3). However, our observations in this case suggest an alternative possibility, whereby events triggered by aging, genetics, beta amyloid toxicity, brain trauma, or other insults, enable tau monomer to transition to a stable conformation that mediates self-assembly and initiates tauopathy.
Competing Interests
A patent disclosure has been filed by H.M., L.A.J. and M.I.D. related to the use of unique crosslinks to create biomarkers for neurodegenerative diseases.
Acknowledgements
We thank Peter Davies for generously providing antibody reagents and ELISA protocol guidance. This work was supported by grants from the Tau Consortium and NIH grants awarded to 1R01NS071835 (M.I.D.), R01NS089932 (R.V.P. and M.I.D.), and the Effie Marie Cain Endowed Scholarship (L.A.J.). We appreciate the help of the Live Cell Imaging Core Facility administered by Katherine Luby-Phelps, Ph.D., and the Proteomics Core Facility at the University of Texas Southwestern Medical Center.