Twelve phosphomimetic mutations induce the assembly of recombinant full-length human tau into paired helical filaments

Reviewer #1 (Public review):

Summary and Strengths:

The very well-written manuscript by Lövestam et al. from the Scheres/Goedert groups entitled "Twelve phosphomimetic mutations induce the assembly of recombinant full-length human tau into paired helical filaments" demonstrates the in vitro production of the so-called paired helical filament Alzheimer's disease (AD) polymorph fold of tau amyloids through the introduction of 12 point mutations that attempt to mimic the disease-associated hyper-phosphorylation of tau. The presented work is very important because it enables disease-related scientific work, including seeded amyloid replication in cells, to be performed in vitro using recombinant-expressed tau protein.

Comments on revised version:

The manuscript is significantly improved, as also indicated by Reviewer 2, with the 100% formation of the PHF and the additional experiments to elucidate on the potential mechanism by the PTMs. This is a great work.

Reviewer #2 (Public review):

Summary:

This manuscript addresses an important impediment in the field of Alzheimer's disease (AD) and tauapathy research by showing that 12 specific phosphomimetic mutations in full-length tau allow the protein to aggregate into fibrils with the AD fold and the fold of chronic traumatic encephalopathy fibrils in vitro. The paper presents comprehensive structural and cell based seeding data indicating the improvement of their approach over previous in vitro attempts on non-full-length tau constructs. The main weaknesses of this work results from the fact that only up to 70% of the tau fibrils form the desired fibril polymorphs. In addition, some of the figures are of low quality and confusing.

Strengths:

This study provides significant progress towards a very important and timely topic in the amyloid community, namely the in vitro production of tau fibrils found in patients.

The 12 specific phosphomimetic mutations presented in this work will have an immediate impact in the field since they can be easily reproduced.

Multiple high-resolution structures support the success of the phosphomimetic mutation approach.

Additional data show the seeding efficiency of the resulting fibrils, their reduced tendency to bundle, and their ability to be labeled without affecting core structure or seeding capability.

Comments on revised version:

Generally, I am satisfied with the revisions. Specifically, the new results showing 100% formation of PHF is a significant improvement.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public review):

Summary and Strengths:

The very well-written manuscript by Lövestam et al. from the Scheres/Goedert groups entitled "Twelve phosphomimetic mutations induce the assembly of recombinant fulllength human tau into paired helical filaments" demonstrates the in vitro production of the so-called paired helical filament Alzheimer's disease (AD) polymorph fold of tau amyloids through the introduction of 12 point mutations that attempt to mimic the disease-associated hyper-phosphorylation of tau. The presented work is very important because it enables disease-related scientific work, including seeded amyloid replication in cells, to be performed in vitro using recombinant-expressed tau protein.

Weaknesses:

The following points are asked to be addressed by the authors:

(i) In the discussion it would be helpful to note the findings that in AD the chemical structure tau (including phosphorylation) is what defines the polymorph fold and not the buffer/cellular environment. It would be further interesting to discuss these findings in respect to the relationship between disease and structure. The presented findings suggest that due to a cellular/organismal alteration, such as aging or Abeta aggregation, tau is specifically hyper-phosphorylated which then leads to its aggregation into the paired helical filaments that are associated with AD.

We have added an extra sentence to the Introduction to emphasise this possibility: “Besides the cellular environment in which they assemble, different tau folds may also be determined by chemical modifications of tau itself.”

In addition, the last paragraph of the Discussion now reads: “It could be that, besides different cellular environments in which the filaments assemble, different posttranslational modification patterns are also important for the assembly of tau into protofilament folds that are specific for the other tauopathies.”

(ii) The conditions used for each assembly reaction are a bit hard to keep track of and somewhat ambiguous. In order to help the reader, I would suggest making a table to show conditions used for each type of assembly (including the diameter / throw of the orbital shaker) and the results (structural/biological) of those conditions. For example, presumably the authors did not have ThT in the samples used for cryo-EM but the methods section does not specify this. Also, the presence of trace NaCl is proposed as a possible cause for the CTE fold to appear in the 0N4R sample (page 4) but no explanation of why this particular sample would have more NaCl than the others. Furthermore, it appears that NaCl was actually used in the seeded assembly reactions that produced the PHF and not the CTE fold. This would seem to indicate the CTE structure of 0N4RPAD12 is not actually induced by NaCl (like it was for tau297-391). In order for the reader to better understand the reproducibility of the polymorphs, it would be helpful to indicate in how many different conditions and how many replicates with new protein preparations each polymorph was observed (could be included in the same table)

We have added a new table (Table 1) with the buffer conditions, protein concentration and shaking speed and time, for all structures described in this paper. We never added ThT to assembly reactions that were used for cryo-EM.

We did not use NaCl in the seeded assembly reactions (we used sodium citrate). We don’t really know why 0N4R PAD12 tau more readily forms the CTE fold. The observation that it does so prompted us to use 0N3R for all ensuing experiments.

(iii) It is not clear how the authors calculate the percentage of each filament type. In Figure 1 it is stated "discarded solved particles (coloured) and discarded filaments in grey" which leaves the reviewer wondering what a "discarded solved particle" is and which filaments were discarded. From the main text one guesses that the latter is probably false positives from automated picking but if so, these should not be referred to as filaments. Also, are the percentages calculated for filaments or segments? In any case, it would be more helpful in such are report to know the best estimate of the ratio of identified filament types without confusing the reader with a measure of the quality of the picking algorithm. Please clarify. Also, a clarification is asked for the significance of the varying degrees of PHF and AD monomer filaments in the various assembly conditions. It could be expected that there is significant variability from sample to sample but it would be interesting to know if there has been any attempt to reproduce the samples to measure this variability. If not, it might be worth mentioning so that the % values are taking with the appropriate sized grain of salt. Finally, the representation of the data in Figure 1 would seem to imply that the 0N3R forms less or no monofilament AD fold because no cross-section is shown for this structure, however it is very similar to (or statistically the same as) the 1:1 mix of 0N3R:0N4R.

In the revised manuscript, we have used bi-hierchical clustering of filaments, where each segment (or particle) is classified based on both 2D class assignment and to which filament it belongs (this method is based on [Porthula et al (2019), Ultramicroscopy 203, 132-138] and was further developed in [Lövestam et al (2024) Nature 7993, 119-125]. Based on the assumption that filament type does not change within a single filament type, we have observed that this gives excellent classification results, and that this approach allows classification of many, even small minority, filament types. Using this approach, we now quantify the different filament types on the number of segments extracted from filaments classified in this way.

Moreover, we have also addressed the problem of having singlets among the PHF preparation: it turns out that waiting longer, just by transferring samples out of the shaker after one week and incubating it quiescently at 37 ºC for two more weeks, the singlets disappear and only PHFs remain. Filaments made for the fluorophore labelling in the revised Figure 3 were also done using the new protocol. In total, we have N=7 replicates with a mean of 95.3% PHFs and a standard deviation of 9.4%. The revised text in the Results section reads:

“To further increase the proportions of PHFs-to-singlet ratio, we removed the plate from the shaker after one week and incubated it quiescently at 37 ºC for two more weeks. This resulted in 100% PHFs formed (Figure 1 – figure supplement 4). When repeated seven times, on average 95.3% PHFs formed, with 25% of singlets formed in a single outlier (Figure 1 – figure supplement 5)”

(iv) The interpretation of the NMR data on soluble tau that the mutations on the second site are suppressing in part long range dynamic interaction around the aggregationinitiation site (FIA) is sound. It is in particular interesting to find that the mutations have a similar effect as the truncation at residue 391. An additional experiment using solvent PREs to elaborate on the solvent exposed sequence-resolved electrostatic potential and the intra-molecular long range interactions would likely strengthen the interpretation significantly (Iwahara, for example, Yu et al, in JACS 2024). Figure 6D Figure supplement shows the NMR cross peak intensities between tau 151-391 and PAD12tau151-391. Overall the intensities of the PAD12 tau construct are more intense which could be interpreted with less conformational exchange between long range dynamic interactions. There are however several regions which do not show any intensity anymore when compared with the corresponding wildtype construct such as 259-262, 292-294 which should be discussed/explained.

While long-range intramolecular interactions of tau have previously been reported through the use of spin labels (Mukrasch et al 2009 PLoS Biol 7(2): e1000034), we have been hesitant to introduce paramagnetic agents into our samples for two reasons. First, the bulky size of the spin label may affect filament formation or influence the dynamic properties of the protein. Second, covalent addition of the spin label requires mutation of the primary sequence to both remove native cysteine residues and add cysteines at the desired label location. We have previously shown that mutation of cysteine 322 to alanine leads to the formation of tau filaments with a structure that is different from the PHF (Santambrogio et al (2025) bioRxiv 2025.03.29.646137).

Instead, we have included in the revised manuscript new NMR and cryo-EM data that provide further support for the model that a FIA-like interaction between residues ₃₉₂IVYK₃₉₅ and residues ₃₀₆VQIVYK₃₁₁ has an inhibiting effect on filament nucleation in unmodified full-length tau. A mutant of tau297-441 where residues ₃₉₂IVYK₃₉₅ have been deleted and that does not contain the four PAD12 mutations in the carboxy-terminal domain behaves similarly in the NMR experiment as the tau297-441 construct with those four PAD12 mutations. Moreover, full-length 0N3R tau with the eight PAD12 mutations in the amino-terminal fuzzy coat and with the deletion of₃₉₂IVYK₃₉₅, but without the four PAD12 mutations in the carboxy-terminal domain, assembles readily into amyloid filaments (of which we also solved a cryo-EM structure, see the revised Figure 6B). These observations provide mechanistic insights into the previously proposed paper-clip model [Jeganathan (2008), J Biol Chem 283, 32066-32076], where interactions between the fuzzy coat inhibit filament formation of unmodified full-length tau, and phosphorylation in the fuzzy coat interferes with these interactions, thus leading to filament nucleation. Of course, the identification of residues ₃₉₂IVYK₃₉₅ for this interaction also explain why truncation of tau at residue 391 leads to spontaneous assembly. We have introduced a new Figure 7 to the revised manuscript to explain this model in more detail. The corresponding new section in the Results reads:

“To investigate this further, we also tested a tau construct comprising residues tau297-441 without the phosphomimetic mutations, but with a deletion of residues (Δ392-395). Filaments formed rapidly and the cryo-EM structure showed that the ordered core consisted of the amino-terminal part of the construct spanning residues 297-318 (Figure 6B). NMR analysis (Figure 6 – figure supplement 5B) showed that the tau297441 Δ392-395 construct exhibited similar backbone rigidity properties to the tau297-441 PAD12 construct, despite peak locations and local secondary structural propensities being more similar to the wildtype tau297-441 (Figure 6 – figure supplement 5A; Figure 6 – figure supplement 6). HSQC peak intensities in the 297-319 and 392-404 regions of tau297-441 Δ392-395 (Figure 6A, expanded from Figure 6 - figure supplement 5C) were like those in the tau297-441 PAD12. These data suggest that the IVYK deletion has a similar effect as the phosphomimetics on residues 396, 400, 403 and 404 on disrupting an intra-molecular interaction between the FIA core region and the carboxy-terminal domain, which may therefore be mediated by interactions between the two IVYK motifs that are similar to those observed in the FIA (Lövestam et al, 2024).”

A new section in the Discussion now reads:

“Our NMR data provide insights into the mechanism by which phosphorylation in the fuzzy coat of tau, or truncations of tau, lead to the formation of filaments with ordered cores of residues that are themselves not phosphorylated. HSQC peak intensity differences between unmodified tau 297-441, PAD12 tau 297-441 and tau297-391 suggest that phosphorylation of the fuzzy coat, particularly near the ₃₉₂IVYK₃₉₅ motif in the carboxy-terminal domain, a7ects the conformation of the residues of tau that become ordered in the FIA (Lövestam et al., 2024). Removal of residues ₃₉₂IVYK₃₉₅ in the carboxyterminal domain of tau 297-441 led to rapid filament formation in the absence of phosphomimetics, while HSQC peak intensity di7erences for this construct indicate similar backbone rigidity compared to tau 297-441 without the deletion, but with the four PAD12 mutations in the carboxy-terminal domain. Combined, these observations support a model where the ₃₉₂IVYK₃₉₅ motif in unmodified full-length tau monomers interacts with the ₃₀₈IVYK₃₁₁ motif, thus inhibiting filament formation by preventing the formation of the nucleating species, the FIA. Phosphorylation of nearby residues 396, 400, 403 and 404, or truncation at residue 391, disrupt this interaction and lead to filament formation. This model agrees with the previously proposed hairpin-like model of tau (Jeganathan et al., 2008), although the corresponding interaction between the aminoterminal domain of tau and the core-forming region remains unknown (Figure 7).”

Due to the challenging nature of the assignment, it was not possible to assign all residues in the HSQC of the tau151-391 and the PAD12 tau151-391 samples, including residues 259-262 and 292-294 for PAD12 tau151-391. To make this clearer, we have marked residues that are not assigned with an asterisk in the revised version of Figure 6 – figure supplement 1.

(v) Concerning the Cryo-EM data from the different hyper-phosphorylation mimics, it would seem that the authors could at least comment on the proportion of monofilament and paired-filaments even if they could not solve the structures. Nonetheless, based on their previous publications, one would also expect that they could show whether the nontwisted filaments are likely to have the same structure (by comparing the 2D classes to projections of non-twisted models). Also, it is very interesting to note that the twist could be so strongly controlled by the charge distribution on the non-structured regions (and may be also related to the work by Mezzenga on twist rate and buffer conditions). Is the result reported in Figure 2 a one-oT case or was it also reproducible?

As also indicated in the main text, the assembly conditions for the PAD12+4, PAD12-4 and PAD12+/-4 constructs were kept the same as those for the PAD12 construct. It is possible that further optimisation of the conditions could again lead to twisting filaments, but we chose not to pursue this route. With unlimited resources and time, one could assess in detail which of the PAD12 mutations are required and which ones could be omitted to form PHFs. However, this would require a lot of work and cryo-EM time. For now, we chose to prioritise reporting conditions that do work to reproducibly make PHFs in the laboratory (using the PAD12 construct) and leave the more detailed analysis of other constructs for future studies.

Reviewer #2 (Public review):

Summary:

This manuscript addresses an important impediment in the field of Alzheimer's disease (AD) and tauapathy research by showing that 12 specific phosphomimetic mutations in full-length tau allow the protein to aggregate into fibrils with the AD fold and the fold of chronic traumatic encephalopathy fibrils in vitro. The paper presents comprehensive structural and cell based seeding data indicating the improvement of their approach over previous in vitro attempts on non-full-length tau constructs. The main weaknesses of this work results from the fact that only up to 70% of the tau fibrils form the desired fibril polymorphs. In addition, some of the figures are of low quality and confusing.

As also explained in our response to reviewer #1, we have performed better quantification of filament types in the revised manuscript, and we have investigated how to get rid of the singlets. In the revised manuscript, we report that singlets disappear as time passes and that one can obtain 100% pure PHFs by quiescently incubating samples for another two weeks, after shaking for a week.

Strengths:

This study provides significant progress towards a very important and timely topic in the amyloid community, namely the in vitro production of tau fibrils found in patients.

The 12 specific phosphomimetic mutations presented in this work will have an immediate impact in the field since they can be easily reproduced.

Multiple high-resolution structures support the success of the phosphomimetic mutation approach. Additional data show the seeding efficiency of the resulting fibrils, their reduced tendency to bundle, and their ability to be labeled without affecting core structure or seeding capability.

Weaknesses:

Despite the success of making full-length AD tau fibrils, still ~30% of the fibrils are either not PHF, or not accounted for. A small fraction of the fibrils are single filaments and another ~20% are not accounted for. The authors mention that ~20% of these fibrils were not picked by the automated algorithm. However, it would be important to get additional clarity about these fibrils. Therefore, it would improve the impact of the paper if the authors could manually analyze passed-over particles to see if they are compatible with PHF or fall into a different class of fibrils. In addition, it would be helpful if the authors could comment on what can be done/tried to get the PHF yield closer to 90-100%

As mentioned above, in the revised manuscript we show that the singlets disappear over time and we now include a description of a method that leads to 100% PHF formation.

Reviewer #1 (Recommendations for the authors):

Minor points:

(a) In Figure 6 the dashed purple vertical lines overlap with the black bars, rendering a grey color which is confusing because the grey bars used for the shorter construct. It is suggested to improve the colors (remove transparency on the purple?)

We thank the reviewers for their suggestions for improving the visualisation of our data. We have recoloured the tau297-391 data from grey to gold and moved the dashed lines to the back of image to remove the apparent colour changes.

(b) Is there any support for the suggestion that "part of the second microtubule-binding repeat is ordered" being "related to this construct forming filaments with only a single protofilament"? It seemed to have come out of nowhere.

There is no further support for this statement, but we thought it would be worth hypothesizing about this observation.

(c) Figures 1 and 4 E is better described as a "main chain trace" or "backbone trace" although the latter usually refers to only CA positions. Ribbon usually refers to something else in representations of protein structures.

This has been changed into “main chain trace” in Figures 1 and 4.

(d) Figure 1 Supplement 3: Panel letters in the legend do not match.

This has been fixed.

Reviewer #2 (Recommendations for the authors):

The introduction is a bit lengthy (e.g. 3rd paragraph of introduction) and could benefit by focusing specific question the manuscript addresses.

We have shortened the Introduction. It now contains ~1150 words, which we hope provides a better compromise between length and sufficient background information.

Figure captions are generally not helpful in conveying a message to the reader.

Figure 1 - figure supplement 3 is quite confusing. The 4 structures in A) do not correspond to the grids in B-E. What is this figure supposed to show?

This confusion was probably the result of incorrect labelling of panels in the legend, which was also pointed out by reviewer #1. This has been fixed in the revised manuscript.

Page 11: Although I know what you mean, 'linear increase of ThT fluorescence' is not the correct term.

We have replaced “linear” with “rapid”.

Page 15: Although line shape and peak intensity can be related you are not reporting on line shape or width but simply on peak intensity. Therefore, I wouldn't talk about the result of a 'line shape analysis'.

We have changed the wording accordingly.

Figure 6 (and supplement 1) are confusing and too small to be readable in print. It might be sufficient to show the CSP and upload the remaining data to the BMRB.

We have made a clearer version of the main NMR Figure 6 in the revised manuscript showing the most pertinent NMR data and have moved the previous version into the figure supplements. We designed these figures to be viewed as full page A4 panels, ideally seen in one image as they show multiple comparisons of different experiments and constructs.

As such we feel these will be best viewed on screen as part of the eLife web document. We have uploaded HSQC spectra and assignments to the BMRB (see below).

Figure 6 supplement 3 might benefit from pointing out key residues in the overlay.

We have added the labels (this is now Figure 6 supplement 4).

Data availability: Please upload the assignments to the BMRB together with key spectra (e.g. HSQCs).

We have uploaded HSQC data along with our assignments to the BMRB, the accession codes are 52694 – tau297-441 wt; 52695 – tau297-441 PAD-12; 52696 – tau151-391 wt; 52697 – tau151-391 PAD-12; and 53230 – tau297-441 delta392-395. These accession codes have been added to the manuscript.

The quality of some of the figures (specifically Figure 1 - supplement 3 and Figure 6) is not suitable for publication.

For the original submission to bioRxiv, we produced a single PDF with a manageable file size. We will liaise with the eLife staff to ensure the images used in the version of record will be suitable for publication.