Tau molecular diversity contributes to clinical heterogeneity in Alzheimer’s disease

Abstract

Alzheimer’s disease (AD) causes unrelenting, progressive cognitive impairments, but its course is heterogeneous, with a broad range of rates of cognitive decline¹. The spread of tau aggregates (neurofibrillary tangles) across the cerebral cortex parallels symptom severity^2,3. We hypothesized that the kinetics of tau spread may vary if the properties of the propagating tau proteins vary across individuals. We carried out biochemical, biophysical, MS and both cell- and animal-based-bioactivity assays to characterize tau in 32 patients with AD. We found striking patient-to-patient heterogeneity in the hyperphosphorylated species of soluble, oligomeric, seed-competent tau. Tau seeding activity correlates with the aggressiveness of the clinical disease, and some post-translational modification (PTM) sites appear to be associated with both enhanced seeding activity and worse clinical outcomes, whereas others are not. These data suggest that different individuals with ‘typical’ AD may have distinct biochemical features of tau. These data are consistent with the possibility that individuals with AD, much like people with cancer, may have multiple molecular drivers of an otherwise common phenotype, and emphasize the potential for personalized therapeutic approaches for slowing clinical progression of AD.

Tauopathies are a group of brain diseases, such as progressive supranuclear palsy, chronic traumatic encephalopathy, Pick’s disease and AD, defined by the accumulation of disease-specific pathological conformations of tau proteins that have distinct clinical presentations, neuropathological features and patterns of neurodegeneration^4–8. These differences in tauopathies are reflected in unique patterns of the seeding of tau into abnormally folded seeds^9–15.

We tested the hypothesis that the principle of varied tau conformations leading to different clinical phenotypes could be extended even within a single disease syndrome, such as AD. Despite a relatively uniform anatomical pattern of tau progression as described by the Braak staging scheme¹⁶, AD is remarkably heterogeneous clinically, exemplified by a broad range of rate of cognitive decline¹. We hypothesized that this heterogeneity in AD might be partially explained if individuals have different tau species that show variation in properties related to spreading and propagation across the cerebral cortex.

To investigate the biochemical diversity of tau in the human AD brain, we selected 32 study participants on the basis of both clinical and pathological characteristics. These individuals were all diagnosed antemortem and postmortem with typical AD (see Supplementary Table 1 for inclusion and exclusion criteria). We analyzed age of onset (mean (M) = 69 years old; minimum–maximum, 45–81 years old), the duration of disease from symptom onset to death (M = 12 years; minimum–maximum, 5–19 years) and the trajectory and linear estimate of the rate of clinical progression extrapolated longitudinally from the Clinical Dementia Rating Scale Sum of Boxes (CDR-SOB) scores over a minimum of 3 annual research visits at the Massachusetts Alzheimer’s Disease Research Center (Fig. 1). Some individuals reached the maximum CDR-SOB score (18) in as few as 6 years, or declined for as many as 17 years from date of diagnosis; however, others never reached a CDR-SOB score of 18 before death, indicative of the clinical heterogeneity of AD¹. Although none of the patients with AD reported a strong family history of the disease, 1 participant (whose age of onset was 45) was found to harbor a potentially pathogenic c.811C>T mutation in presenilin 2 gene (PSEN2), encoding p.L271F. All study participants had typical neuropathological patterns of advanced AD (Braak V/VI).

Fig. 1 |. Heterogeneity of clinical progression and age of onset in Alzheimer’s disease.

a, Dementia progression in this study’s participants. Each line corresponds to an individual participant, and each point to an individual visit at the memory clinic. Participants were evaluated for their CDR-SOB scores at each visit. CDR-SOB = 0 is a cognitively normal individual; CDR-SOB = 18 is the maximum of this scale and corresponds to an individual with very advanced dementia. b, Linear estimate of the rate of clinical progression, extrapolated longitudinally from the data in a, from onset to CDR-SOB = 18 or from onset to last visit if CDR-SOB = 18 was not reached. c, Distribution of the ages of disease onset across the 32 participants in this study.

To test the hypothesis that there is tau heterogeneity in these individuals, all of whom were selected to represent ‘pure AD’ from a clinical and neuropathological perspective, we prepared a soluble extract from each participant’s frontal cortices and measured total tau amounts, tau seeding bioactivity (using a widely used Forster resonance energy transfer (FRET)-based biosensor¹⁰) and biophysical and biochemical parameters of tau. We found substantial person-to-person heterogeneity in tau seeding activity, with differences of up to tenfold among individuals, despite the addition of the same amount of tau to each assay (Fig. 2a,b).

Fig. 2 |. Heterogeneity of tau seeding in the human brain.

a, Schematic of the FRET-biosensor seeding assay. HEK293 cells were stably expressing the tau repeat domain fused with a cyan fluorescent protein or a yellow fluorescent protein. When seeds from human brain extract are added, these constructs aggregate and allow for the generation of a FRET signal that is detectable and quantifiable by flow cytometry and imaging. b, Seeding in 34 human participants (32 participants from Fig. 1, as well as a positive and a negative control) was quantified using flow cytometry in the seeding assay described in a. Samples were normalized to total tau levels before being added to the seeding assay, and the integrated FRET densities obtained were normalized to both positive and negative controls. The positive control is a human AD Braak VI brain that has been characterized in previous studies (Takeda et al.²⁴). The negative control is a human control brain without tau pathology. Error bars represent s.d. of the mean; n = 4 independent experiments. c, Images from the FRET-biosensor assay for tau seeding over time. White arrowheads show the aggregates forming inside cells. This experiment was repeated two times for each participant, with similar results. Scale bar, 100 μm. d, Tau seeding was quantified by live imaging over 72 h in a subset of 9 human study participants. Data for all 32 participants are available in Extended Data Fig. 1. Samples were normalized to total tau levels before being added to the seeding assay, and the number of seeds obtained was normalized to both positive and negative controls. A sigmoidal, four-parameter logistic (4PL) nonlinear regression in which x is log (concentration) was applied before data were plotted. e, Schematic of the human Tau (hTau) seeding assay in mice. hTau-expressing mice primary neurons were plated and incubated with human brain extract for 3 h, and then were further incubated for 11 d before a fixing and permeabilization step in which soluble tau species were washed and hyperphosphorylated tau was stained for, evidencing intracellular insoluble aggregates. f, Representative images of the hTau seeding assay after neurons were stained with NeuN (blue), MAP2 (red) and AT8 antibody against hyperphosphorylated tau (green), showing insoluble aggregates of tau (white arrowheads) when AD brain extract is incubated with neurons. This experiment was repeated six times with similar results. g, Quantification of hyperphosphorylated tau staining with the AT8 antibody in nine human participants in this study. These individuals were separated into groups of high (red, n = 4), moderate (green, n = 2) and low (blue, n = 3) seeders on the basis of the value obtained in Fig. 2b. A representation of data by participant, as well as a time course of aggregate appearance and NeuN and MAP2 burdens, is available in Extended Data Fig. 2. Error bars represent s.d. of the mean. h, Representative images of the presence of AT8-positive immunostaining 2 months after injection of brain extract from 3 patients—a low seeder (upper panels), a moderate seeder (middle panels) and a high seeder (lower panels)—in 4-month-old P301S mice; stars represent the injection sites. Middle and right panels show two different magnifications. Five mice were injected with each human AD extract. i, Two-month-old P301S transgenic mice were stereotactically injected with human AD brain PBS extracts from three high seeders (red), three moderate seeders (green) and three low seeders (blue), determined on the basis of the tau seeding activities obtained in Fig. 2b. Five mice were injected with each human AD extract, for a total of 15 mice in each group. Five mice were injected with human control brain extracts as negative control. Mice were euthanized 2 months later, and their brains were processed for AT8 immunohistochemistry to assess the in vivo seeding potential of human AD brain extracts. The bar graph depicts the stereological quantification of the number of AT8⁺ neurons in the cortex and hippocampus. A representation by patient is available in Extended Data Fig. 2f. One-way analysis of variance (ANOVA) (P < 0.0001) with a Tukey’s multiple-comparison post hoc test revealed a significantly higher number of NFTs after injection with brain extracts from high seeders compared with those from moderate (P < 0.0001) and low (P < 0.0001) seeders. Error bars represent s.d. of the mean.

To examine the possibility that the heterogeneity of seeding resulted from an unexpected characteristic of this cell-reporter assay, which measures seed-induced aggregation of tau repeat domain at a single time point (~24 h), we developed two alternative in vitro assays to measure tau seeding bioactivity. The first uses live-cell imaging and image-analysis algorithms to monitor and quantitate the formation of aggregates in real time over 72 h (Fig. 2c and Supplementary Video 1). Tau seeding in this system appears to follow the seeding-nucleation polymerization model as classically described for prions or amyloids¹⁷ (Fig. 2d, Extended Data Fig. 1a and Supplementary Table 2). The plateau phase reflects the amount of seeds added initially to the system, and correlates with the extent of seeding measured by the standard flow-cytometry protocol (Extended Data Fig. 1).

As an alternative approach to the assay with the FRET-based biosensor, we used primary neurons obtained from transgenic mice that express several wild-type human tau isoforms, including 3R and 4R, in a mouse tau-knockout background¹⁸. We incubated these neurons with brain extract from 9 of the patients (3 high, 3 moderate and 3 low ‘seeders’ as defined in the FRET-based-biosensor assay). Eleven days after extract addition, neurons were fixed and specifically permeabilized to remove soluble tau, and then were stained with an antibody that recognized a hyperphosphorylated pathological form of tau (AT8) to reveal tau aggregates (Fig. 2e). A parallel kinetic analysis showed that these aggregates appeared between the second and seventh day following incubation with brain extracts (Extended Data Fig. 2a). Neuronal viability was assessed by quantifying neuronal nuclei (NeuN) and microtubule-associated protein 2 (MAP2) signal (Extended Data Fig. 2b,c). The quantification of AT8-positive, Triton-X-100-insoluble aggregates in cells (Fig. 2f,g and Extended Data Fig. 2d) shows that the numbers of aggregates that were present in neurons follow the same relative trend of the intensities observed in the FRET-biosensor assay, as high seeders appeared to induce a substantially higher intensity and area of AT8-positive human tau aggregates than did low seeders within a given period of time (Extended Data Fig. 2e).

To examine whether the wide range of tau seeding bioactivity across patients with AD measured with these in vitro assays correlates with a differential tau seeding potential in vivo, we injected human AD brain extracts, obtained from high, medium and low seeders and normalized to the total tau amount, into the cortex and hippocampus of young P301S transgenic mice prior to pathological development in this mouse model of tauopathy, as previously described^19–21. Two months following injection, immunostaining for tau pathology (AT8) (Fig. 2h) and stereological quantification demonstrated that numbers of AT8-positive neurons in animals injected with brain extract from high seeders were higher than those in the mice injected with control extract or that from low/moderate seeders (Fig. 2i and Extended Data Fig. 2f) (P < 0.0001). Thus, this experiment added to our in vitro observations by confirming that variability in tau seeding activity of human AD brain extracts also occurs in vivo.

It is not known what causes the tau from a particular patient with AD to drive higher versus lower seeding. To address this question, we characterized tau derived from the various individuals in our study using biochemical assays. First, we quantified the amount of total tau, oligomeric tau (HT7/HT7 assay recognizing multimeric tau) and phosphorylated tau (at epitopes Thr181 or Ser202/Thr205) by AlphaLISA^22,23 and correlated these measures with their tau seeding potential measured by FRET-based-biosensor assay. We found that, although tau seeding was only weakly negatively correlated with the total amount of tau (Fig. 3a, upper panel; r = −0.41, P = 0.019), it was strongly correlated with the amount of oligomeric tau (Fig. 3a, lower panel; r = 0.86, P < 0.0001) and phosphorylated tau (Extended Data Fig. 3a,b; Thr181: r = 0.80, P < 0.0001; Ser202/Thr205: r = 0.85, P < 0.0001). Although correlation does not mean causation, these findings are consistent with the idea that tau seeds are made of oligomeric hyperphosphorylated species of tau.

Fig. 3 |. Tau seeding relies on an oligomeric form of tau.

a, In brain extracts from study participants, total tau (BT2-HT7 antibodies) and oligomeric tau (HT7-HT7 antibodies) were quantified using AlphaLISA. Prior to this assay, samples were normalized for total protein amount as determined using a bicinchoninic acid (BCA) assay. Absolute amounts of total tau are plotted against relative light units for oligomeric tau. The color-gradient scale bar indicates the seeding quantities determined in Fig. 2b. Both total tau and oligomeric tau show, respectively, a negative and positive statistically significant association with tau seeding, determined using a two-tailed Spearman’s rank non-parametric correlation. Bars represent the average of two independent experimental replicates. AlphaLISA data on tau hyperphosphorylation are available in Extended Data Fig. 3a,b. b, Fourteen participants in our study were selected from across the seeding spectrum, and brain extracts were run on a semi-denaturing detergent agarose gel electrophoresis to separate the HMW tau species (that is, tau oligomers and aggregates) and the LMW tau species (that is, tau monomers and proteolysis products) and revealed by a total tau antibody (Dako). These study participants were separated into groups of high (red, n = 5), moderate (green, n = 4) and low (blue, n = 5) seeders on the basis of the value obtained in Fig. 2b. Densitometry data for tau signals were quantified and plotted in the right panel, which shows a difference in the quantity of HMW tau species in the high seeders versus in the moderate and low seeders. Error bars represent s.d. of the mean. See Extended Data Fig. 3e for the association between the quantities of HMW tau from SDD-AGE and of oligomeric tau from AlphaLISA c, Schematic of the SEC technique. Human brain extracts were applied to a gel-filtration column, and fractions were retrieved with separation of HMW and LMW tau species. d, Upper panel, Tau seeding was quantified in SEC fractions of nine study participants using the assay described in Fig. 1a. Participants were grouped into high (red, n = 3), moderate (green, n = 3) and low (blue, n = 3) seeders on the basis of seeding quantities obtained in Fig. 2b. Tau seeding seems to be present mostly in HMW fractions, and not in LMW fractions. Error bars represent s.d. of the mean. Lower panel, Distribution of tau seeding across fractions for the three groups, indicating that there is widespread tau seeding in high seeders compared with that in moderate and low seeders. Boxes represent the 25th and 75th percentile, and the median is indicated by the midline; whiskers represent the 5th and 95th percentiles. e, Total and oligomeric tau levels were quantified in SEC fractions from nine study participants by AlphaLISA, showing the abundance of LMW tau species and the rarity of HMW tau. However, high seeders have more HMW oligomeric tau than do moderate and low seeders. Error bars represent s.d. of the mean. AlphaLISA data on tau hyperphosphorylation are available in Extended Data Fig. 3c,d. RLU, relative light unit. f–h, Twelve participants’ brain extracts were incubated with increasing doses of proteinase K (PK) and were run on a western blot to find evidence for the differential stabilities of tau species. An antibody recognizing total tau protein was used as the detection antibody. Full blots are available in Extended Data Fig. 4. i, Degradation products from the condition with 2.5 μg ml⁻¹ PK were quantified and compared using a one-way ANOVA with a Tukey’s multiple-comparison post hoc test. There was a significant association between high seeders (red, n = 4) and a higher amount of degradation products compared with moderate (green, n = 4) and in low (blue, n = 4) seeders. Error bars represent s.d. of the mean.

We and others have previously reported that tau seeding from the human AD brain is primarily due to a soluble high-molecular-weight (HMW) form of tau^23–25. Consistent with these findings, semi-denaturing detergent agarose gel electrophoresis (SDD-AGE)^9,26 (Fig. 3b) and oligomeric tau AlphaLISA data (Extended Data Fig. 3e) show variability among study participants, with high seeders apparently enriched in oligomeric species. Size-exclusion chromatography (SEC)²⁴ (Fig. 3c) followed by the FRET-based-biosensor seeding assay confirmed that tau seeding activity is highly enriched in the HMW fractions (fractions 2–5) (Fig. 3d), with little seeding activity observed in the low-molecular-weight (LMW) fractions (monomeric tau eluted across fractions 12–15), and that seed-competent species of tau are enriched in high seeders compared with moderate seeders and low seeders (Fig. 3d). The distribution pattern of seeding activity across fractions differs among high, moderate and low seeders (Fig. 3d, lower panel) such that high seeders have a wider distribution of seeding-active fractions, and moderate and low seeders’ activity is more narrowly localized to the HMW fractions. We quantified the amount of total tau, oligomeric tau and phosphorylated tau in the SEC fractions by AlphaLISA. As expected, the majority of total tau in the brain extracts tracked with monomeric tau (Fig. 3e, upper panel). Low levels of total tau were detected in HMW fractions 2–5. We interpret the absolute quantitation in these fractions with caution, as it is possible that there is a decreased sensitivity owing to steric hindrance of epitopes. Nonetheless, high seeders displayed substantially greater total tau signal in HMW fractions, reflecting much higher levels of tau in these fractions. This is corroborated by the presence of a larger oligomeric tau signal in high seeders than in moderate and low seeders (Fig. 3e, lower panel). Analysis of the same fractions for various phosphorylated tau epitopes indicated that the HMW tau fractions contained highly phosphorylated tau (Extended Data Fig. 3c,d).

To further characterize the physical properties of tau in these diverse brain extracts, we used a protease-sensitivity digestion assay that differentiates protein conformations^14,27. Human brain extracts were incubated with varying concentrations of proteinase K, and the resulting banding patterns of tau were examined by SDS–polyacrylamide gel electrophoresis (PAGE) and western blot. Although the protein extracts had similar banding patterns in the absence of proteinase K (Fig. 3f), banding patterns diverged among samples after treatment with proteinase K (Fig. 3g,h and Extended Data Fig. 4), indicating individual-level differences in protease sensitivity (Fig. 3i).

To examine the potential biochemical basis for these differences in molecular properties in tau across AD brains, we utilized label-free mass spectrometry (MS) to definitively identify sites of tau PTMs. We detected peptides that included phosphorylation sites Thr181, Ser198/Ser199/Ser202, Thr217, Thr231, Ser262 and Ser400/Thr403/Ser404, allowing identification of singly phosphorylated peptides and the doubly phosphorylated Thr231 and Ser235. Tau phosphorylation of the soluble, HMW species examined here appears to be heterogenous, with notable person-to-person differences (Fig. 4a). We examined whether this heterogeneity was associated with the postmortem interval or other technical variables, but found no or only very weak associations in this dataset (see Extended Data Fig. 5a,b for a more complete analyses).

Fig. 4 |. Tau seeding is associated with rate of disease progression and intensity of tau phosphorylation.

a, Brain extracts from the 32 study participants (rows) were sedimented by ultracentrifugation to precipitate seeding-active tau species. Pellets were digested in trypsin and analyzed on a Q Exactive Mass Spectrometer. Peptides modified by phosphorylation were identified and quantified, and data were normalized to the summed intensity of three unmodified tau peptides (columns). Color scales from the lowest value (light green) to the highest value (dark green) for each peptide. One sample (participant no. 9) was excluded owing to poor data quality. b, Intensities of phosphorylation for different phosphorylation sites were correlated to tau seeding. n = 31 individual human participants. Correlations were performed using a two-tailed Spearman’s rank non-parametric test, and P and r values are indicated on each plot. c, Tau seeding, as quantified in Fig. 2b, positively correlates with rate of disease progression, as quantified in Fig. 1b. n = 32 individual human participants. IFD, integrated FRET density. The correlation was performed using a two-tailed Spearman’s rank non-parametric test, and P and r values are indicated on the plot. d, Intensity of phosphorylation at different phosphoepitopes were also correlated with rate of disease progression, showing significant associations. n = 31 individual human participants. A two-tailed Spearman’s rank non-parametric correlation test was used, and r and P values are indicated on the plots. See Extended Data Figs. 5, 7, 8 and 9 for additional analysis regarding intensities of phosphorylation, rates of progressions and ages of onset.

Notably, the amount of phosphorylation on doubly phosphorylated Thr231 and Ser235, and on singly phosphorylated Ser262, positively correlated with the seeding capacity of the tau species (Fig. 4b and Extended Data Fig. 5c) (respectively, r = 0.74, P < 0.0001; r = 0.65, P < 0.0001), validating that phosphorylation at specific sites likely involves conformational changes impacting the seeding activity of tau²⁸. Of note, some phosphorylation sites appear to be negatively correlated with tau seeding—for example, a single phosphorylation within the cluster of sites represented on peptides Ser198/Ser199/Ser202 (r = −0.72, P < 0.0001) or Ser400/Thr403/Ser404 (r = −0.79, P < 0.0001). By comparison, antibodies that recognize multiple phosphorylated epitopes within these sites are strongly positively correlated (Extended Data Fig. 3a,b), suggesting that the exact phosphoepitope landscape can dramatically impact seeding potential.

We also evaluated whether the heterogeneity of tau seeding is reflected in histological findings in the brains of the participants (Extended Data Fig. 6). We found that tau seeding was, as expected, significantly positively correlated with tau burden (r = 0.66, P < 0.0001) and GFAP burden (r = 0.40, P = 0.02), but not with β-amyloid burden, CD68-positive microglia burden or cortical thickness (r = 0.29, P = 0.1; r = 0.19, P = 0.29; r = −0.10, P = 0.57).

Taken together, these results indicate that tau seeding is tightly linked with biochemical markers of HMW soluble pathological tau in the human AD brain. If tau seeding is a key pathway by which tau pathology and toxicity spread through the central nervous system to drive AD symptoms, it may correlate with clinical progression of AD. We tested this hypothesis by correlating tau seeding with the rate of clinical AD progression, determined in each individual by the longitudinal linear approximation of their CDR-SOB scores (Fig. 1b). Interestingly, the higher the tau seeding activity, the steeper the rate of decline (Fig. 4c). We also found some indication that other markers of disease aggressiveness correlate with seeding activity, including younger age of onset (Extended Data Fig. 7). The amount of phosphorylation on specific peptides correlated not only with seeding activity but also with rate of disease progression (Fig. 4d and Extended Data Fig. 8; see Extended Data Fig. 9 for additional analyses). Moreover, these robust markers of clinical disease activity also relate to the amount of oligomeric and phosphorylated tau in the brains of these patients (Extended Data Fig. 9c–h); in a secondary analysis, we found that high seeding seemed to be particularly present in carriers of apolipoprotein E (APOE)-ε4/ε4 (Extended Data Fig. 9j), emphasizing the likely importance of genetic risk factors in the mechanisms of tau aggregation and pathology propagation^29,30.

Taken together, these data further support the idea that tau seeding in soluble brain fractions can account for a moderate (~25%) amount of clinical heterogeneity, and thus support the idea that individual tau seeding characteristics are clinically relevant, and that tau seeding is deeply implicated in the mechanisms of disease progression. Consequently, suppression of tau seeding may be a valuable therapeutic strategy. We explored this by using immunodepletion with antibodies recognizing diverse tau epitopes (Extended Data Fig. 10a,b) combined with the tau FRET-based-biosensor assay. Seven commercially available antibodies with various epitopes along the protein tau were utilized (Extended Data Fig. 10b), which were in the amino terminus (Tau-12), middle domain (HT7) and carboxy terminus (Tau-46), or with distinct phosphoepitopes of tau present on hyperphosphorylated tau species (AT270, AT8, pS262 and PHF1). Non-immune immunoglobulin G (IgG) was a negative control. The affinity- and kinetic-binding properties of these antibodies were determined using a Biacore surface plasmon resonance (SPR) strategy (Supplementary Table 3). We used these antibodies to deplete seeding activity from the brain extracts of 15 of 32 participants for whom material was available. First, antibodies coupled to magnetic beads were incubated with brain extracts normalized to the amount of total soluble tau. After capture by magnetic beads, the supernatant was tested in the FRET-based-biosensor assay (Extended Data Fig. 10a). As shown in the heatmap of tau seeding reduction (Extended Data Fig. 10c), some antibodies reduce tau seeding more effectively than do others overall, and there is substantial variability from patient to patient, consistent with previous reports^31,32. Indeed, some individuals’ seeding activity was readily suppressed, whereas others were much less sensitive to antibody-mediated depletion. Overall, no single antibody was capable of fully depleting tau seeding activity in all study participants; however, antibodies recognizing pathological forms of tau (for example, AT8 and PHF1) appeared to be the most consistently effective. An alternative assay configuration broadly confirmed these results (Supplementary Fig. 1). Notably, we did not observe tight correlations for individual phosphoepitopes between blocking of seeding and mass-spectroscopy results, possibly because any individual phosphosite does not necessarily reflect other (linked) phosphorylation sites within each molecule, or the overall burden of PTM on a molecule (Supplementary Fig. 2).

Previous studies in different tauopathies report differing aggregation phenotypes consistent with pathological ‘strains’ of tau^9,15, which correlate with differing clinical presentations. Here, we provide multiple orthogonal lines of evidence supporting the idea that individuals with AD may have differing characteristics of soluble tau. Two separate in vitro bioactivity assays, confirmation in an in vivo seeding assay in a tauopathy mouse model and biochemical characterization, including SEC, protease sensitivity, alphaLISA and mass spectrometry, demonstrate surprisingly substantial patient-to-patient heterogeneity in the tau species present in the seed-competent soluble HMW brain fraction. The clinical relevance of this observation is suggested by two independent analyses. First, the measured tau seeding bioactivity (amount of seeding per unit of tau) correlates with a clinical measure of aggressiveness of disease: rate of cognitive decline. Second, tau immunodepletion experiments show that patients with AD respond differently to a panel of diverse anti-tau antibodies, with potential implications for anti-tau immunotherapy experiments³³. Thus, there appears to be heterogeneity in tau seeding properties, tau conformation and tau PTMs among individuals with otherwise apparently equivalent advanced AD.

It is interesting to note that cryogenic electron microscopy studies show differences in fibrils in AD, Pick’s disease, corticobasal degeneration and chronic traumatic encephalopathy, but have yielded a single set of two conformations of tau fibrils in AD brains^8,34, in contrast to our present findings. However, it should be noted that the data presented in the current study pertain to soluble HMW tau oligomers, rather than to the insoluble tau fibrils studied with cryo-EM. Thus, we envision that soluble tau can adopt multiple conformations that, when ultimately deposited in a long-lived fibril, remodel to a stable paired helical or straight filament conformation. Another possibility is that, despite a single common ultrastructure, PTM patterns still cause biological differences in seeding.

If different biologically active conformers of tau exist across AD brains, it raises the question of whether different underlying pathophysiological mechanisms occur in subsets of patients. Although the biological mechanisms underlying such heterogeneity are unknown, we postulate that differences in kinase activity (leading to different combinatorial phosphorylation patterns), or other post-translational modifications such as ubiquitinylation, might affect conformation^4,35. Particularly worth noting is that the pS262 epitope, one of the rare phosphorylation sites inside the microtubule-binding region, is known to promote oligomerization of tau and to alter stability of tau binding to microtubules^28,36–39. Regardless of the valence or influence on seeding and other relevant tau bioactivities associated with individual (or combinations of individual) PTMs, these data argue that the existence of specific patterns of PTMs can impact the biophysical properties of the protein (oligomerization, seeding potential and/or aggregation propensity). A caveat is that the detection of these ‘increased risk’ or even ‘protective’ sites may simply be a proxy for other post-translational tau modifications (multiple clustered phosphorylation, acetylation, ubiquitinylation, methylation, glycation or truncation, and so on), a marker for a combinatorial complex set of PTMs or a surrogate for other unmeasured variables.

In sum, we find unexpected biochemical and biophysical differences in seed-competent tau across individuals with otherwise typical AD, and show that the different tau forms correlate with differences in aggressiveness of clinical course. By analogy to cancer, in which different molecular drivers instruct individualized therapeutic approaches in patients with an otherwise common phenotype, it may be that these observations will inform subclassification of AD patients⁴⁰, and prove useful in understanding clinical heterogeneity and in designing the next generation of clinical trials for AD.

Methods

Human tissue and data collection.

Thirty-two human participants with AD were selected from the Massachusetts Alzheimer’s Disease Research Center Longitudinal Cohort study on the basis of the following criteria: (1) clinically diagnosed with dementia due to probable AD; (2) cognitive status assessed at least three times by a neurologist at the MGH Memory Clinic Unit and scored using CDR-SOB; (3) diagnosis of AD confirmed postmortem by an MGH neuropathologist; (4) Braak Neurofibrillary tangles (NFTs) stage V or VI as determined by the location of NFTs with a total tau immunostain and Bielchowsky’s silver stain¹⁶; and (5) the least possible concurrent pathologies. Age of onset, age at death, postmortem interval, sex and APOE genotype were also collected and are listed in Supplementary Table 1⁴¹. Additional information on the sample size and data exclusion can be found in the Reporting Summary. We approximated the rates of progression by applying a linear-regression model to the longitudinal CDR-SOB scores (from onset to CDR-SOB = 18, or from onset to the last available score when CDR-SOB = 18 was not reached; Fig. 1b). Autopsy tissue from human brains were collected at Massachusetts General hospital, with informed consent of patients or their relatives and approval of local institutional review boards.

Human brains were processed as previously described⁴². Briefly, all brains were separated into 2 hemispheres, one of which was postfixed in 10% formalin for 3 weeks, and regions of interest were embedded in paraffin following standard protocols^43,44. Seven-micrometer-thick paraffin-embedded frontal association cortex tissue sections (Brodmann area 8/9) were cut and subsequently placed on slides (Fisherbrand Superfrost Plus slides; Thermo Fisher Scientific) for histological analysis. The contralateral hemisphere was initially sliced coronally at the time of autopsy and 1-cm-thick slabs were immediately flash frozen and stored at −80 °C. Approximately 1 g of frontal cortex was dissected out of the frozen brain section corresponding to Brodmann Area 8/9 and kept at −80 °C until homogenization.

Tissue homogenization.

Frozen human tissue (300 mg) was thawed on wet ice and then immediately homogenized in 1,500 μl of PBS + protease inhibitor (Roche) in a 2-ml glass Dounce homogenizer. Tissue was Dounce homogenized with 30 up and down strokes on ice by hand. The homogenate was transferred to a 1.5-ml Eppendorf tube and centrifuged at 10,000g for 10 min at 4 °C. The supernatant was collected and aliquoted to avoid excessive freeze–thaw cycles. A BCA assay (Thermo Scientific Pierce) was performed to determine total protein concentration, and an ELISA (Meso Scale Diagnostics) was performed to determine the concentration of total tau in the samples, following the manufacturer’s protocol.

For in vivo experiments, PBS-soluble brain extracts containing 800 ng of total tau were pelleted at 150,000g for 30 min at 4 °C in order to sediment seeding species, as previously described²⁴. Pellets were subsequently resuspended in 50 μl of 1× PBS and stored at −80 °C until further use.

Seeding assays.

In vitro tau seeding FRET-biosensor assay.

The in vitro seeding assay has been previously described and widely characterized^10,42. Briefly, the Tau RD P301S FRET Biosensor (ATCC CRL-3275) cells stably expressing the repeat domain of Tau with the p.P301S mutation conjugated to either cyan fluorescent protein (CFP) or yellow fluorescent protein (YFP) (TauRD-P301S-CFP/YFP) were cultured at 37 °C, 5% CO₂ in DMEM, 10% vol/vol FBS, 0.5% vol/vol penicillin–streptomycin. Additional information on the cell line can be found in the Reporting Summary. Cells were plated on Costar Black, clear-bottom 96-well plates (previously coated with 1:20 poly-d-lysine) at a density of 40,000 cells per well. Brain extracts (8 ng of total tau quantified by total tau ELISA (MSD) per well) were then incubated with Lipofectamine 2000 (Invitrogen, final concentration 1% vol/vol) in opti-MEM (final volume of 50 μl per well) for 10 min at room temperature before being added to the cells. Each condition was applied in triplicate or quadruplicate. For seeding experiments using anti-tau antibodies, 1 μg of each anti-tau antibody was incubated with the brain extract for 20 min at 37 °C before incubation with lipofectamine. The effect of anti-tau antibodies on tau seeding was then analyzed by normalizing to carry-over samples and input samples.

Tau seeding was subsequently analyzed using flow cytometry or live imaging.

Flow-cytometry seeding analysis.

After 24 h, medium was removed, and 50 μl trypsin 1× was added for 7 min at 37 °C. Chilled DMEM + 10% FBS (150 μl) was added to the trypsin, and cells were transferred to 96-well U-bottom plates (Corning). Cells were pelleted at 500g, resuspended in freshly-made 2% vol/vol paraformaldehyde in PBS (Electron Microscopy Services) for 10 min at room temperature in the dark, and pelleted at 500g. Cells were resuspended in chilled PBS and run on the MACSQuant VYB (Miltenyi) flow cytometer. CFP and FRET were both measured by exciting the cells using the 405-nm laser and reading fluorescence emission at the 405/50-nm and 525/50-nm filters, respectively. To quantify the FRET signal, a bivariate plot of FRET versus the CFP donor was generated, and cells that received control brain extract alone were used to identify the FRET-negative population. Using this gate, the tau seeding value for each well was calculated by multiplying the percentage of FRET-positive cells by the median fluorescence intensity of that FRET-positive population. We analyzed 40,000 events per well. Data were analyzed using the MACSQuantify software (Miltenyi).

Four carry-over samples were used for each plate to normalize across multiple plates run at the same time. Additional information on the gating strategy can be found in the Reporting Summary.

Live imaging seeding analysis.

Immediately following addition of brain extract to the cells, the plate was transferred to a BioTek Cytation5 Imaging Multi-Mode Reader, using the Gen5 Image Prime 3.03 software. Cells were kept at 37 °C and were supplied with CO₂ at a concentration of 5%. Green fluorescent protein images of the cells were captured every 30 min for 72 h using the ×20 objective. Four images were taken per well. Images were exported from the Cytation5 Imaging machine for further analysis using Fiji/ImageJ. Images were first converted from individual files into time-series stacks of 145 images, each representing a single field of a single well over the 72-h period. After conversion, stacks were classified on the basis of three attributes: aggregates, cells or background, representing the three main features of each image. Classification was done using the Trainable Weka Segmentation⁴⁵. A classifier model was generated prior to experimentation using a representative set of seven images, each of which were manually classified into aggregates, cells or background. After manually classifying these images, this data were input into a machine-learning-based image-analysis algorithm, consisting of several smoothing and edge-sharpening algorithms. The specific algorithms used for this model classification were Gaussian blur, Hessian, mean, maximum, variance, minimum, median, Laplacian and entropy. The specific type of classifier used was a FastRandomForest (a multi-threaded version of the Random Forest classifier). After the manual implementation the classifier model itself, the classification job and the images were exported to a high-performance cluster developed by the Harvard Medical School Research Computing Group allowing for large dataset analysis. After image classification was completed on the cluster servers, we exported back the images to generate quantitative results. Classification returns an 8-bit grayscale image, with different grayscale values for the pixels classified as cells, aggregates or background (Supplementary Video 1). Using the ‘Analyze Particles’ macro built into Fiji/ImageJ, aggregate and cell qualities were quantitatively measured, including number, size, bounding perimeter, kurtosis and fraction of area occupied. After the quantitative information was determined by Fiji/ImageJ, it was exported and analyzed using Microsoft Excel and GraphPad Prism version 7 (GraphPad). Both the number of seeds per field and the number of seeds normalized to the area of cells were analyzed, showing similar results. For representation in Fig. 2 and Extended Data Fig. 1, a sigmoidal, 4PL non-linear regression in which x is log (concentration) was applied and data were plotted. Plateau values were obtained by averaging the raw data for the last 24 h, and slope values reflected the slope of the linear regression of the lines between 0.5 h and 30 h.

In vitro tau seeding assay in hTau mouse primary mouse neurons.

Mixed neuronal cultures were prepared from embryonic day 18 mouse brains from transgenic mice (B6.Cg-Mapt^{tm1.1(MAPT)Itl} Tg(MAPT)8cPdav), which are the offspring of a cross between hTau mice¹⁸ and an hTau tau knock-in model resulting in a model overexpressing several human tau isoforms (around 6-fold overexpression of human tau), knocked out for the mouse Mapt gene and on a pure C57BL/6 background. In contrast with the hTau mice, these mice don’t carry any copy of the enhanced green fluorescent protein gene. All animal care, housing, and experiments were performed in compliance with guidelines established by the Merck institutional animal care and use committee and in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Cortical tissues were dissected and treated with papain-based dissociation mixture in HBSS media (Hyclone) for 30 min at 37 °C. After rinsing, cells were mechanically dissociated in DMEM supplemented with 10% FBS and 1% penicillin–streptomycin. The cell suspension was filtered with 70-μm and 40-μm filters (Corning) and centrifuged at 250g for 5 min, and the pellet was resuspended in DMEM. The cells were plated in 96-well poly-d-lysine-coated plates and incubated in a 37 °C humidified incubator with 5% CO₂. After 3 h, the medium was replaced with Neurobasal medium (Neurobasal, 2% B27, 0.25 mM GlutaMAX, 1% penicillin–streptomycin, 10 mM sodium pyruvate) and the cells were further incubated at 37 °C. After 10 d in vitro (DIV10), cells were treated with PBS-extracted human brain extracts (containing 100 ng ml⁻¹ Tau monomer equivalent concentration based on BT2/HT7 AlphaLISA; see below for detailed procedure) diluted in Neurobasal medium. After 3 h, the medium was removed and replaced with fresh Neurobasal medium. The cells were incubated at 37 °C for a further 11 d.

At DIV21, the cells were fixed with Prefer (glyoxal-based fixing solution; Anatech) supplemented with 1% vol/vol Triton X-100 to wash out soluble proteins (15 min at room temperature (RT)). Primary antibodies against AT8 (monoclonal, 1/1,000, Thermo Fisher Scientific), MAP2 (polyclonal, Biolegend, 1/10,000) and NeuN (polyclonal, Millipore, 1/1,000), diluted with blocking buffer, were added for 1 h at RT or overnight at 4 °C. After washing with PBS, secondary antibodies against Alexa Fluor (anti-mouse Alexa Fluor 488 from Thermo Fisher Scientific, anti-chicken Alexa Fluor 647 from Thermo Fisher Scientific, anti-guinea-pig Alexa Fluor 405 from Abcam) diluted in blocking buffer were added for 1 h at RT.

For the time-course experiment (Extended Data Fig. 2a,b), human brain PBS extracts were incubated with 1% sarkosyl and 10% sucrose, and then were pelleted at 100,000g. The pellets were resuspended in PBS and sonicated before being added to the primary neurons at DIV8 at a total protein concentration of 75 μg ml⁻¹. Cells were then fixed and stained as indicated above after 1 h, 1 d, 2 d and 7 d following addition of brain extracts.

Images were acquired using the Phenix Opera imaging system (PerkinElmer) and analyzed using the Columbus 2.5.2 HCS image-analysis system (PerkinElmer). NeuN-positive nuclei were detected using the ‘find nuclei’ analysis module. The total intensities of AT8-positive tau and total AT8 spots were identified and calculated using ‘find region’ and ‘find spots’ modules, respectively. Total MAP2 intensity was calculated using the ‘find region’ module.

In vivo seeding assay in tau transgenic mice.

Two-month-old PS19 mice (B6;C3-Tg(Prnp-MAPT*P301S)PS19Vle/J) overexpressing the human 1N4R isoform of tau with a p.P301S mutation were obtained from The Jackson Laboratory. After deeply anesthetizing the mice of either sex with 1.5% (vol/vol) isoflurane, they were immobilized in a stereotaxic frame and brain extracts were injected at the following coordinates from the bregma suture (Anteroposterior, −2.5 mm; mediolateral, ±2 mm; dorsoventral, −2 and −0.8 mm) with a Hamilton syringe under aseptic conditions. Injections consisted of 1.5 μl brain extract from 10 human participants (n = 5 animals per participant), which were injected bilaterally at a flow rate of 0.1 μl min⁻¹. All injected animals were monitored throughout the surgery and postsurgery, until full recovery. After surgery, the skin was sutured and an analgesic was administered.

Two months postinjection, mice were euthanized with CO₂ and transcardially perfused first with cold 0.9% PBS followed by cold 4% PFA for 10 min. The brains were immediately removed, fixed for 48 h in 4% PFA and placed in 30% sucrose for 24 h. Coronal sections (40 μm thick) were sliced using a freezing microtome and were collected at 400 μm intervals for immunohistochemical analysis. All animal care, housing and experiments were performed in compliance with guidelines established by the Massachusetts General Hospital institutional animal care and use committee and in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Immunohistochemistry.

Seven-micrometer-thick formalin-fixed, paraffin-embedded human brain tissue sections from the frontal association cortex (Brodmann area 8/9) were allowed to air dry for at least 24 h after being cut, and were immunostained for β-amyloid, total tau, CD68 (activated phagocytic microglia) and GFAP (reactive astrocytes). β-amyloid slides were pretreated for 20 minutes with an EDTA, pH 9.0 as epitope retrieval step, whereas CD68 slides were pre-treated for 30 minutes with a citrate, pH 6.0 epitope retrieval step. All slides were then stained in a Leica Bond RX Autostainer (Leica) according to the manufacturer’s instructions. Briefly, slides were blocked using the peroxide block solution (Leica) for 20 min at RT, followed by an incubation in primary antibodies for 45 min (β-amyloid and CD68) or 35 min (tau and GFAP) at the following antibody concentrations: total tau (Dako, polyclonal), 1:6,000; β-amyloid (Dako, monoclonal), 1:600 μl; CD68 (Dako, monoclonal), 1:500; and GFAP (Sigma-Aldrich, monoclonal), 1:20,000. Slides were then incubated for 10 min with the Post Primary Solution (Leica) containing secondary antibodies (rabbit anti-mouse IgG) before being washed 3 times for 2 min at RT with the BOND Wash Solution (Leica) and incubation for 8 min at RT with Polymer Solution (Leica) containing anti-rabbit polymeric-horseradish-peroxidase-conjugated IgG. Slides were washed 2 times for 2 min with the Bond Wash solution and then incubated in 2 successive 30-s baths of DAB part 1 (Leica) and DAB part 2 (Leica) containing 3,3′-diaminobenzidine tetrahydrochloride hydrate. Slides were finally incubated with a hematoxylin solution (Leica) for 10 min (β-amyloid and CD68) or 15 min (tau and GFAP) at RT. Slides were then transferred into water and dehydrated by 10 quick incubations (around 1-s long) into 2 baths of 70% ethanol, 2 baths of 95% ethanol, 2 baths of 100% ethanol and 2 baths of 100% xylene. Slides were then cover slipped with Permount Mounting Medium (Fisher Scientific) and scanned using an Olympus VS120-S6-W virtual slide microscope, at a magnification of ×40. Scanned slide images were then analyzed in Olympus’ cellSens analysis software. Regions of interest were drawn to capture as much cortical ribbon as was available. Once regions of interest were drawn, HSV-thresholding was performed manually to obtain the burden of targeted pathologies. Slides for all individuals within one antibody condition were stained at the same time and microscopy exposure times as well as analysis thresholds were kept stable for all individuals within one antibody condition. Data were then transferred to GraphPad Prism version 7 for further statistical analysis.

For mouse studies, 40-μm-thick coronal free-floating sections were washed in TBS and endogenous peroxidases were quenched with 0.3% H₂O₂. Sections were then blocked with 3% normal goat serum (Vector Laboratories) in 0.25% Triton X-100 TBS and then incubated with mouse monoclonal AT8-biotin (Thermo Fisher Scientific) in blocking solution at 4 °C overnight. On day 2, sections were washed in TBS and incubated in streptavidin–horseradish-peroxidase-containing ABC reagent (Vector Laboratories). Fresh diaminobenzidine was prepared from a tablet (Sigma-Aldrich) and incubated with tissue for 45 s to precipitate. Sections were mounted onto slides, dehydrated in ascending ethanol/xylene and then coverslipped with Cytoseal XYL (Thermo Fisher Scientific).

Stereology.

After immunohistochemistry, stereological quantifications were performed using the Computer Assisted Stereological Toolbox (CAST, version 2.3.1.5, Olympus America). AT8-positive neurons were counted around the injection sites in all sections from bregma −1 to −2.7 mm in the hippocampi and in the cortex overlaying the hippocampi (mainly visual, auditory, retrosplenial, posterior parietal association and primary somatosensory cortex). From each section, 10% of the total area of the cortex was counted using a randomized algorithm to yield count per mm². All counting was conducted while researchers were blinded to experimental conditions.

Cortical thickness.

Cortical thickness was measured in immunostained sections as previously described⁴⁶. Briefly, the image-analysis software CAST, mounted on an upright BX51 Olympus microscope (Olympus) and coupled with a motorized stage and a charge-coupled device camera, was used to randomly sample the cortex of the entire section and measure the full thickness of the cortex. The measurements of full cortical thickness from 20 random sites were averaged.

AlphaLISA.

We performed four versions of the AlphaLISA sandwich immunoassays with different acceptor bead antibodies/biotinylated detection antibody combinations:

• Total tau: acceptor antibody BT2 (monoclonal), biotinylated antibody HT7 (bHT7; monoclonal);

• AT8-phosphorylated-tau: acceptor antibody HT7, biotinylated antibody AT8 (monoclonal);

• HF6-phosphorylated tau: acceptor antibody PHF6 (monoclonal), biotinylated antibody HT7;

• Oligomeric tau: acceptor antibody HT7, biotinylated antibody HT7.

Beads (1:200 dilution) and biotin-detector antibodies (1:266 dilution) were diluted into 1× AlphaLISA assay buffer (PerkinElmer) and placed into a 384-well Opti-plate (PerkinElmer) at 20 μl per well. Five microliters of each sample were then added. The HT7–bAT8, HT7–bHT7 and PHF6–bHT7 assays were all run with undiluted fraction assay input material. The BT2–bHT7 t-tau assay was run with 1:100 diluted SEC fraction assay input material in 1× PBS 0.2% Tween 20. The plate was then covered with aluminum sealing tape, centrifuged at 500g for 1 min and incubated with shaking overnight at 550 r.p.m., 4 °C. After incubation, the plate was centrifuged at 500g for 1 min. Two micrograms of streptavidin-donor beads (PerkinElmer) diluted in 1× AlphaLISA assay buffer (final volume, 25 μl per well) were added to each well. The plate was then covered with aluminum sealing tape, centrifuged at 500g for 1 min and incubated with shaking for 30 min at 550 r.p.m., RT. After incubation, the plate was centrifuged at 500g for 1 min and immediately read on a PHERAstar (BMG Labtech) 1.11 plate reader S/N 471–0028, software version 3.00 R2, at 680-nm and 615-nm wavelengths for excitation and detection, respectively. The resulting raw data in the BT2 total-tau assay were transformed to ng ml⁻¹ values based on the standard curve using the GraphPad Prism version 7.02 and 4PL format curve interpolation.

SDD-AGE.

Semi-denaturing detergent agarose gel electrophoresis (SDD-AGE) was performed as previously described^9,26. Briefly, brain extract normalized to contain 340 ng total tau was loaded onto a 1.5% agarose gel with Laemmli buffer (20 mmol l⁻¹ Tris-base, 200 mmol l⁻¹ glycine and 0.02% sodium dodecyl sulfate). The gel was run on ice at 35 V for 16 h and then transferred onto a polyvinylidene difluoride (PVDF) membrane overnight in TBS using filter paper and capillary action. The membrane was blocked in TBS containing 0.25% Tween 20 and 5% non-fat dry milk and incubated overnight with rabbit polyclonal anti-tau antibody (1:1,000; Dako). Horseradish-peroxidase-conjugated goat anti-rabbit IgG secondary antibody was applied in blocking solution on day 2 and was detected using chemiluminescent horseradish-peroxidase substrate (Thermo Fisher Scientific) and film (GE Healthcare Life Sciences). For quantification, grayscale images were scanned and imported into Fiji/ImageJ. Using the gel quantification densitometry tool, the gel was separated into 14 bands (bins) of equal size that were all quantified and plotted.

LC–MS/MS measurements.

For liquid chromatography–tandem mass spectrometry (LC–MS/MS), for each human participant, 50 μl of TBS soluble brain extract containing 75 μg of total protein was pelleted at 150,000g for 30 min at 4 °C in order to sediment seeding species as previously described²⁴. Pellets were subsequently resuspended in 1× PBS. Resuspended pellets were diluted in 8 M urea, 50 mM ammonium bicarbonate, and were processed separately using filter-aided sample preparation (FASP)⁴⁷ on filters with a 10-kDa molecular-weight cutoff (Millipore). Samples were reduced with 20 mM dithiothreitol and alkylated using 1% acrylamide⁴⁸. Protein mixtures were digested with 100 μl of 10 μg ml⁻¹ trypsin (sequencing-grade-modified trypsin) overnight at 37 °C. Peptide solutions were then acidified with formic acid (2% vol/vol), desalted using C18 Microspin columns (Nest Group) and vacuum-dried overnight. Peptides were reconstituted in sample buffer (5% formic acid, 5% acetonitrile (ACN)) and analyzed in random order on a Q Exactive mass spectrometer (Thermo Fisher Scientific). Peptides were separated on a LC system consisting of a micro-autosampler AS2 and a nanoflow high-performance LC (HPLC) pump (Eksigent) using a capflow PicoChip column (150 μm × 10 cm Acquity BEH C18 column, 1.7-μm particle size, 130-Å pore size; New Objective) during a 120-min gradient from 98% buffer A (0.1% (vol/vol) formic acid in HPLC-grade H₂O), 2% buffer B (0.2% (vol/vol) formic acid in ACN) to 70% buffer A, 30% buffer B at a flowrate of 1 μl min⁻¹. The PicoChip was heated to 50 °C and mounted directly at the inlet to the mass spectrometer. The Q Exactive was operated in positive-ion mode, with a data-dependent acquisition top 12 method. The first spectrometer (MS1) scan settings were as follows: mass-to-charge (m/z) range, 375–1,400; resolution 70,000 at m/z 400; AGC target, 3 × 10⁶; max IT, 60 ms. The second spectrometer (MS2) scan settings were: resolution 17,500 at m/z 400; AGC target, 5 × 10⁴; max IT, 100 ms; isolation window, m/z 1.6; NCE, 27; underfill ration, 2% (intensity threshold 1 × 10⁴); charge state exclusion unassigned, 1, >8; peptide match preferred; exclude isotopes on; dynamic exclusion, 25 s.

LC–MS/MS data analysis.

To identify phosphorylation sites on tau, raw files generated by Xcalibur were analyzed using MaxQuant⁴⁹ software version 1.6.5.0 searched against the UniProtKB/Swiss-Prot protein sequence human proteome database (October 2017) including isoform entries. The following settings were applied: digestion- trypsin with up to 2 missed cleavages; mass tolerances, 20 parts per million (ppm) for the first peptide search, 4.5 ppm for the second search; and fragment ion mass tolerance, 20 ppm. Propionamide on cysteine was set as fixed modification, and oxidation of methionine (M), N-terminal acetylation and phosphorylation of serine, threonine and tyrosine were set as variable modifications. The false-discovery rate was set to 1% on peptide-spectrum match, protein and site decoy fraction levels, and was determined by searching a reverse database. For all other search parameters, the default settings were used. In order to quantify modified peptides in each sample even in the absence of fragmentation of the precursor, an MS1 filtering approach using Skyline was used. A Skyline document creating entries for modified peptides representing phosphorylation sites identified on tau using MaxQuant was created. In cases where multiple coeluting phosphoisomers of the same peptide were identified by MaxQuant (Ser198/Ser199/Ser202, Ser400/Thr403/Ser403), they could not be distinguished on the MS1 level and therefore were treated as one entity. To generate a spectral library to facilitate peak identification, Mascot 2.6 was used to search ten individual data files with the following settings: fixed propionamide on cysteine, phosphorylation on serine, threonine, and tyrosine, N-terminal acetylation and methionine oxidation as variable modifications. Peptide mass tolerance 20 ppm, and fragment mass tolerance was 20 millimass units. The Human UniProtKB/Swiss-Prot database (October 2017) including isoforms was used with specific trypsin digestion and a maximum two missed cleavages. The results files were downloaded from Mascot and imported as a peptide search in Skyline to generate a spectral library using the DDA with MS1 filtering workflow⁵⁰. The following peptides were selected in the Skyline document: pThr181-TPPAPKT[+80 Da] PPSSGEPPK, pSer198/pSer199/pSer202-SGYSS[+80 Da]PGSPGTPGSR, pThr217-TPSLPT[+80 Da]PPTR, pThr231-VAVVRT[+80 Da] PPKSPSSAK, pThr231 and pThr235-VAVVRT[+80]PPKS[+80 Da]PSSAK, pSer262-IGS[+80 Da]TENLK pSer400/pThr403/perS404-SPVVSGDTS[+80 Da] PR, unmodified-LQTAPVPMPDLK, unmodified-C[+71 Da] GSLGNIHHKPGGGQVEVK, unmodified-IGSLDNITHVPGGGNK. All raw files were imported in Skyline extracting chromatograms within 5 min of Mascot IDs. Extracted ion chromatograms were examined and peak boundaries were manually adjusted for each sample and peptide using retention time, mass error (<10 ppm), isotope dot product (>0.8), fragment ions and local peak landscape to guide peak picking. The sum of the peak area of the three unmodified peptides was used to normalize the modified peptide peak area to total tau. Owing to a quality-control issue on the mass spectrometry data, participant no. 9 was excluded from all analyses. Analyses of the liquid chromatography–mass spectrometry data were performed blinded.

SEC.

Brain soluble extracts were separated by SEC as previously described²⁴ on single Superdex200 10/300GL columns (no. 17-5175-01, GE Healthcare) in PBS (no. P3813, Sigma-Aldrich, filtered through a 0.2-μm membrane filter), at a flow rate of 0.5 ml min⁻¹, with an AKTA purifier 10 (GE Healthcare). Each brain extract was diluted with PBS to contain 6,000 ng of human tau to a final volume of 500 μl, which was filtered through a 0.2-μm membrane filter and then loaded onto an SEC column. Fractions of 500 μl were retrieved, analyzed by AlphaLISA and subjected to the in vitro seeding assay.

Proteinase K resistance assay.

Proteinase K digestion was carried out as previously described¹⁴. Briefly, brain extracts were diluted in proteinase K buffer (50 mM Tris-HCL pH 8.0, 1 mM CaCl2, 3 mM DTT and 2 M urea) and proteinase K (Thermo Fisher Scientific) was added to the appropriate concentration (0–2.5 μg ml⁻¹). Samples were incubated in proteinase K buffer for 30 min at room temperature. The digestion was halted with 5 mM phenylmethylsulfonyl fluoride (Sigma-Aldrich). NuPAGE LDS Sample Buffer (4× concentration) (Invitrogen) and 10× NuPAGE Sample Reducing Agent (Invitrogen) were added to 5 μg of digested extracts, boiled for 10 min at 95 °C and run on 4–12% Bis-Tris SDS–PAGE (Invitrogen) in MES buffer (Invitrogen). The gels were transferred to a PVDF membrane. Blots were incubated with anti-total tau polyclonal (anti-C-terminus tau polyclonal, 1:2,000; Dako) and anti-phosphorylated tau (p-tau^{Ser396/Ser404} PHF1 polyclonal, 1:2,000; kindly donated by P. Davies) antibodies overnight at 4 °C. Blots were washed three times for 10 min with TBS-T and incubated with secondary anti-rabbit IRDye680 (1:1,000) for 1.5 h at room temperature and imaged on the Odyssey infrared imaging system (Li-Cor). For quantification, images were converted to grayscale in FIJI/ImageJ. Using the gel quantification densitometry tool, the full lane of tau was quantified, as well as tau fragments under 50 kDa and the monomeric tau (band from 50 kDa to 70 kDa). Quantifications were normalized to the full-length band of tau at 0 μg ml⁻¹ of proteinase K of each sample to take into account the amount of total tau in the sample.

Tau immunodepletion.

Tau immunodepletions were performed using the following antibodies: Tau-12 (monoclonal, Biolegend), HT7 (monoclonal, Thermo Fisher Scientific), Tau-46 (monoclonal, Cell signaling), AT270 (monoclonal, Thermo Fisher Scientific), AT8 (monoclonal, Thermo Fisher Scientific), pS262 (monoclonal, Thermo Fisher Scientific) and PHF1 (monoclonal, kind gift from P. Davies). The negative control for tau binding used was a random human immunoglobulin G (IgG). Additional information on antibodies can be found in the Reporting Summary. Tau immunodepletions from brain extracts were carried out using the Dynabeads Protein G Immunoprecipitation Kit according to manufacturer’s instructions with slight modifications (Invitrogen). Dynabeads (750 μg) were resuspended by rotating for 5 min. The supernatant was removed while the beads were immobilized by a magnetic field. Beads were resuspended in 100 μl PBS with 5 μg antibody and incubated for 10 min at RT with rotation. To remove the supernatant, we pipetted it out while the bead–antibody complexes were immobilized by a magnetic field. Bead–antibody complexes were resuspended in 100 μl PBS. The supernatant was removed while the bead–antibody complexes were immobilized by a magnetic field. Human brain extracts (normalized to a final total amount of 800 ng of tau protein) were diluted in PBS to a final volume of 100 μl, and were then added to the beads and gently homogenized by pipetting and were incubated for 10 min at RT with rotation. Supernatant was retrieved while the bead–antibody–antigen complexes were immobilized by a magnetic field. This supernatant was used subsequently for the in vitro FRET seeding assay to determine the amount of seeding reduction achieved by tau immunodepletion (1 μl of supernatant per well in the FRET-based-biosensor assay). Bead–antibody–antigen complexes were washed 3 times using 100 μl PBS for each wash and resuspended in 100 μL 1× LDS sample buffer/1× reducing agent/distilled water. After a 5-min incubation at 95 °C, beads were removed using a magnetic field and supernatant was collected and stored for further analyses.

Biacore.

Kinetic affinities of anti-tau antibodies were generated by Biacore T200 (GE Healthcare), a SPR-based technology, as previously described⁵¹. Briefly, a C1 sensor chip (GE Healthcare, BR-1005–35) was directly immobilized via amine coupling kit (GE Healthcare, BR-1000–50) with low density of human tau-441 protein (Sigma-Aldrich T0576) with 10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.05% P20 pH7.4 (1× HBS-EP+) running buffer. Human tau-441 protein was diluted to 0.5 μg ml⁻¹ in 10 mM sodium acetate pH 4.5 and injected at 10 μl min⁻¹ twice for 15 s each over Flow Cell 2 that had been activated for 7 min with EDC–NHS per the amine coupling kit instructions. The immobilized human tau-441 surface was neutralized with ethanolamine–HCl for 7 min. The reference surface, Flow Cell 1, was similarly activated with EDC–NHS and neutralized with ethanolamine–HCl but did not receive the human tau-441 protein. Serially diluted anti-tau antibodies were injected at 30 μl min⁻¹ over the human tau-441 surface for 4 min, followed by dissociation in 1× HBS-EP+ for 7 min. The human tau-441 sensor surface was regenerated with two 5-s pulses of 10 mM glycine pH 1.5 for 5 s at 100 μl min⁻¹. Binding kinetics were analyzed with Biacore T100 Evaluation Software using the 1:1 binding model.

Whole-exome sequencing.

DNA was extracted from cerebellum samples from all 32 study participants plus one control individual using the Qiagen DNeasy Blood & Tissue Kit All DNA samples were sent to Novagene, and the 17 that successfully passed quality control (from participant nos. 2, 3, 5, 6, 7, 9, 13, 15, 16, 18, 21, 22, 23, 26, 29 and 30, and the control). A total amount of 1.0 μg genomic DNA per sample was used as input material for the DNA sample preparation. Sequencing libraries were generated using Agilent SureSelect Human All ExonV6 kit (Agilent Technologies) following the manufacturer’s recommendations. Briefly, fragmentation was carried out by hydrodynamic shearing system (Covaris) to generate 180- to 280-bp fragments. Remaining overhangs were converted into blunt ends via exonuclease/polymerase activities, and enzymes were removed. After adenylation of the 3′ ends of DNA fragments, adapter oligonucleotides were ligated. DNA fragments with ligated adapter molecules on both ends were selectively enriched in a PCR reaction. Captured libraries were enriched in a PCR reaction to add index tags to prepare for hybridization. Products were purified using AMPure XP system (Beckman Coulter) and were quantified using the Agilent High Sensitivity DNA Assay on the Agilent Bioanalyzer 2100 system. The qualified libraries were fed into Illumina sequencers after pooling, according to effective concentration and expected data volume. The original data obtained from the high-throughput sequencing platforms were transformed to sequenced reads by base calling. Raw data were recorded in a FASTQ file which contains sequenced reads and corresponding sequencing quality information.

Statistics.

Basic statistics and plots have been generated using GraphPad Prism 7 (GraphPad). Unless otherwise stated, two-tailed Spearman r non-parametric correlations were used to correlate different variables obtained from single individuals. The Spearman correlation coefficient r and the P value are indicated in the figures. For the proteinase K resistance assay and the in vivo seeding assay, the quantifications were analyzed using a one-way ANOVA test with a Tukey’s multiple-comparisons post hoc test.

Reporting Summary.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1 |. Kinetics of tau seeding in FRET biosensor assay relates to Fig. 2d.

a, Tau seeding was quantified by live imaging over 72 h of FRET biosensor cells exposed to PBS extracts of the 32 human AD brains. This process is divided into three phases: a nucleation/lag phase followed by the exponential polymerization/elongation phase and ending by a plateau phase. Samples were normalized to total tau levels before being added to the seeding assay and the number of aggregates obtained was normalized to both positive and negative controls. A sigmoidal, 4PL, X is log(concentration) nonlinear regression was applied before plotting the data b, Tau seeding dose response was investigated in two of these AD brain extracts. Tau seeding was quantified with 8 ng of total tau in the sample (dose used for Fig. 2d and in panel A) but also with 0.08 ng, 0.8 ng, 4 ng, 16 ng and 40 ng to demonstrate the dose dependence of the plateau phase and the ceiling effect of the assay. c, Statistically significant two-tailed Spearman’s rank nonparametric correlation between the plateau value for each of the 32 AD brain extracts measured in this manner and the seeding value obtained on the FRET biosensor assay by flow cytometry (Fig. 2b). d, Statistically significant two-tailed Spearman’s rank nonparametric correlation between the values of the plateau and of the slope for the 32 AD brains.

Extended Data Fig. 2 |. hTau seeding assay cell viability and association with FRET biosensor assay.

a, Timecourse of insoluble AT8 aggregates appearance in the hTau primary neuron seeding assay 1 hour, 1 day, 2 days and 7 days post incubation with the AD brain extracts. The right panel only represents the time points from 1 h to 2 days post incubation with the brain extracts. Error bars represents Standard deviations to the mean. n = 10 human subjects. Quantification of immunolabelling with NeuN (b) and MAP2 (c) in hTau mouse primary neurons incubated with 9 different AD brain extracts from our study’s cohort show a general viability of neuronal cells after brain extract incubation beside a possible toxicity with subject #32 brain extract. n = 4 independent experiments, error bars represent standard deviations to the mean d, Quantification of AT8 hyperphosphorylated tau staining on 9 human subjects from Fig. 2g. n = 4 independent experiments. Error bars represents Standard deviations to the mean. The color gradient scale bar relates to seeding quantities obtained in Fig. 2b. e, Statistically significant two-tailed Spearman’s rank nonparametric correlation between the value of seeding activity obtained in Fig. 2b and the AT8 signal intensity obtained with the hTau primary neuron seeding assay for 9 subjects AD brain extracts. f, Tau seeding in a mouse model of tauopathy- Relates to Fig. 2h,i. Two-month-old P301S transgenic mice were stereotactically injected with human AD brain PBS extracts from 10 human AD subject. Mice were euthanized 2 months later, and their brains processed for AT8 immunohistochemistry to assess the in vivo seeding potential of human AD brain extracts. n = 5 animals per human subject. Bar graph depicts the stereological quantification of the number of AT8-positive neurons in the cortex and hippocampus. The color gradient scale bar relates to seeding quantities obtained in Fig. 2b. Error bars represents Standard deviations to the mean.

Extended Data Fig. 3 |. Hyperphosphorylation is closely associated with tau seeding.

Brain extracts from the 32 AD subject were quantified for two epitopes of tau using alphaLISA (AT8 (a) and PHF6 (b)). Prior to this assay, samples were normalized for total protein amount as obtained using a BCA assay. Background Corrected Relative Light Unit (RLU) are plotted here. The color gradient scale bar relates to seeding quantities obtained in Fig. 2b. Both phospho-epitopes show a positive statistically significant association with tau seeding using a two-tailed Spearman’s rank non parametric correlation. SEC fractions from 9 AD brain extracts grouped into high seeders (red, n = 3), moderate seeders (green, n = 3) and low seeders (blue, n = 3) were tested for the presence of epitopes of tau hyperphosphorylation by alphaLISA (AT8 (d) and PHF6 (c)). Error bars represent the standard deviation to the mean. Showing the enriched presence of these epitopes in HMW fractions, especially in high and moderate seeders. e, HMW tau species quantified from the SDD-AGE (bin1–6, see Fig. 3b) were correlated with oligomeric tau levels from the alphaLISA showing a significant two-tailed Spearman’s rank nonparametric correlation. The r coefficient and p value are indicated on the plot. n = 14 individual subjects.

Extended Data Fig. 4 |. Proteinase K digestion Western blots.

Relates to Fig. 3f–i. 12 AD brain extracts from our study’s cohort were incubated with increasing doses of proteinase K and run on a Western blot in order to investigate differential stabilities of tau species. Antibodies recognizing total tau proteins as well as hyperphosphorylated tau proteins were used as detection antibodies. This experiment was repeated two times with similar results.

Extended Data Fig. 5 |. Correlation of postmortem interval and longevity versus intensities of phosphorylation.

Intensity of phosphorylation of phospho-sites T181, S198/S199/S202, T217, T231, T231&S235, S262, S400/T403/S404 (ordinate) were correlated with postmortem interval⁵³ (a) or age at death⁵⁴ (b) (abscissa). n = 31 individual subjects. Two-tailed Spearman’s rank nonparametric correlation tests were used, and r coefficient and p values are indicated on the tables. c, Some phospho-epitopes do not correlate with seeding- Intensity of phosphorylation of phospho-sites T181, T217 and T231 (ordinate) were correlated with tau seeding activity (abscissa). n = 31 individual subjects. Two-tailed Spearman’s rank nonparametric correlation test was used, and r coefficient and p value are indicated on the plots.

Extended Data Fig. 6 |. Tau seeding activity correlates with tau and GFAP-immunoreactive burdens.

Immunohistochemical staining of tau (indicative of NFTs, neuropil threads and plaque-associated neuritic dystrophies) a, Amyloid-β (indicative of Aβ plaques) c, GFAP + reactive astrocytes e, and CD68 + phagocytic microglia g, formalin-fixed paraffin-embedded sections from the frontal association cortex (BA8/9) of the 32 AD subjects and their respective burden quantifications (b,d,f,h). The color gradient scale bar relates to seeding activities obtained in Fig. 2b. Both the tau burden and the GFAP burden show a statistically significant positive association with tau seeding using a two-tailed Spearman’s rank nonparametric correlation. i, Cortical thickness measured as a proxy for neurodenegeration⁴⁵ in the same sections did not significantly correlated with tau seeding activity using a two-tailed Spearman’s rank nonparametric correlation. n = 32 individual human subjects.

Extended Data Fig. 7 |. Age of disease onset correlates with tau seeding activity but not with intensity of tau phosphorylation.

a, Tau seeding (on the abscissa) as quantified in Fig. 2b negatively correlates with age of onset. n = 32 individual subjects. b, Intensities of phosphorylation of different phospho-sites (ordinate) were positively or negatively correlated with age of onset (abscissa). n = 31 individual subjects. Two-sided Spearman’s rank nonparametric correlation test was used and r coefficient and p value are indicated on the plots.

Extended Data Fig. 8 |. Some phospho-epitopes do not correlate with rate of disease clinical progression.

Intensities of phosphorylation of phospho-sites T181, T217 and T231 (ordinate) were correlated with rate of clinical disease progression as indicated by the slope of the linearized CDR-SOB score trajectories (abscissa). Two-sided Spearman’s rank nonparametric correlation test was used and r coefficient and p value are indicated on the plots. n = 31 individual human subjects.

Extended Data Fig. 9 |. Rate of clinical disease progression and age of symptom onset correlate with tau burden, oligomeric tau and phosphorylated tau levels.

The relationship between age of onset and rate of progression undoubtedly has many contributors, hence it is not surprising that some relationships are not evident statistically in a relatively small sample that was not selected to examine this question. As expected and previously established^2,3,55,56, Rate of disease clinical progression as indicated by the slope of the linearized CDR-SOB score trajectories and age of symptom onset as quantified in Fig. 1b, c significantly correlates with tau burden from Supplementary Fig. 8a,b (respectively a, and b) but also oligomeric tau levels from Fig. 3a (respectively c, and d, p = 0.057) and 2 epitopes of tau hyperphosphorylation: PHF6, from Supplementary Fig. 4b (respectively e, and f, p = 0.051) and AT8 from Supplementary Fig. 4a (respectively g, and h) in the 32 subjects of this study’s cohort⁵⁴. i, As recently described and probably not independent of the age of onset, tau seeding correlates with age at death⁵⁴. Correlations were carried out using a two-sided Spearman’s rank nonparametric correlation, r coefficient and p values are indicated on the plots. n = 32 individual human subjects. j, Analysis of tau seeding activity by APOE genotype showed a statistically significant difference with higher seeding activity in APOEε4/ε4 subjects (n = 5) compared to APOEε3/ε4 (n = 20, p = 0.0013) and APOEε4 non-carriers (n = 8, p = 0.0072). Groups were compared using a one-way ANOVA with a Tukey’s multiple comparison post-test. Error bars represent standard deviation to the mean.

Extended Data Fig. 10 |. Reduction of tau seeding by antibodies is epitope and subject-to-subject dependent.

a, Schematic representation of the paradigm of tau seeding reduction via immunodepletion. Antibodies were coupled with magnetic beads and incubated with AD brain extract. Beads/antibodies/antigens complexes were depleted and the supernatant placed on the FRET biosensor seeding assay. FRET was quantified by flow cytometry. b, Schematic representation of tau protein with alternative exons 2 (yellow), 3 (green) and 10 (red) as well as the repeated regions of the microtubule binding domains (black). Antibodies used in this study are indicated below. Green antibodies (Tau12, HT7 and Tau46) target the total protein when red antibodies (AT270, AT8, pS262 and PHF1) target phospho-epitopes known to be associated with AD tau pathology. c, Antibody-mediated reduction of tau seeding across 15 AD subjects from our study’s cohort (left column). Antibodies are organized in columns. IgG serve as negative control for seeding reduction. Percentage of seeding reduction and standard deviation are indicated for each individual/antibody association. The color code of seeding reduction is indicated in the lower panel.