Proteostasis as a fundamental principle of Tau immunotherapy

Abstract

The microtubule-associated protein Tau is a driver of neuronal dysfunction in Alzheimer’s disease and other tauopathies. In this process, Tau initially undergoes subtle changes to its abundance, subcellular localization and a vast array of post-translational modifications including phosphorylation that progressively result in the protein’s somatodendritic accumulation and dysregulation of multiple Tau-dependent cellular processes.

Given the various loss- and gain-of-functions of Tau in disease and the brain-wide changes in the proteome that characterize tauopathies, we asked whether targeting Tau would restore the alterations in proteostasis observed in disease.

Therefore, by phage display, we generated a novel pan-Tau antibody, RNJ1, that preferentially binds human Tau and neutralizes proteopathic seeding activity in multiple cell lines and benchmarked it against a clinically tested pan-Tau antibody, HJ8.5 (murine version of tilavonemab). We then evaluated both antibodies, alone and in combination, in the K3 tauopathy mouse model, showing reduced Tau pathology and improvements in neuronal function following 14 weekly treatments, without obtaining synergy for the combination. These effects were more pronounced in female mice.

To investigate the molecular mechanisms contributing to improvements in neuronal function, we employed quantitative proteomics, phosphoproteomics and kinase prediction analysis to first establish alterations in K3 mice relative to wild-type controls at the proteome level. In female K3 mice, we found 342 differentially abundant proteins, which are predominantly involved in metabolic and microtubule-associated processes, strengthening previously reported findings of defects in several functional domains in multiple tauopathy models. We next asked whether antibody-mediated Tau target engagement indirectly affects levels of deregulated proteins in the K3 model. Importantly, both immunotherapies, in particular RNJ1, induced abundance shifts towards a restoration to wild-type levels (proteostasis). A total of 257 of 342 (∼75%) proteins altered in K3 were closer in abundance to wild-type levels after RNJ1 treatment, and 73% after HJ8.5 treatment. However, the magnitude of these changes was less pronounced than that observed with RNJ1. Furthermore, analysis of the phosphoproteome showed an even stronger restoration effect with RNJ1, with ∼82% of altered phosphopeptides in K3 showing a shift to wild-type levels, and 75% with HJ8.5. Gene set over-representation analysis further confirmed that proteins undergoing restoration are involved in biological pathways affected in K3 mice. Together, our study suggests that a Tau immunotherapy-induced restoration of proteostasis links target engagement and treatment efficacy.

Alzheimer’s disease, tau antibodies, neurodegenerative diseases, brain proteomics, brain phosphoproteomics, quantitative mass spectrometry

See Canet and Planel (https://doi.org/10.1093/brain/awae382) for a scientific commentary on this article.

Introduction

A histopathological hallmark not only of Alzheimer’s disease but also of more than two dozen neurodegenerative diseases, collectively termed tauopathies, is the intracellular aggregation of Tau, a neuronally enriched protein. Alzheimer's disease is further characterized by a second hallmark lesion in the form of amyloid-β (Aβ)-containing plaques in the extracellular space,¹ with both pathologies impairing multiple cellular functions.^2-6 In the pursuit of clearing these aggregates, substantial efforts have been invested into developing active and passive immunotherapies targeting the two molecules.^7,8 This has resulted in the recent approval of two anti-Aβ antibodies, aducanumab and lecanemab, by the US Food and Drug Administration (FDA), with additional antibodies following suit.^9-11 In contrast, Tau antibodies have, to date, failed to demonstrate clinical efficacy, necessitating the development of more potent Tau-targeting antibodies.

Developing an effective Tau immunotherapy, however, poses significant challenges. Tau is highly heterogeneous as in the human brain the protein exists as six isoforms that undergo a myriad of post-translational modifications, including hyperphosphorylation, and how the different forms of Tau each contribute to pathogenicity across tauopathies is not fully understood.¹ Furthermore, Tau is differentially distributed across distinct neuronal sub-compartments and is also found in the extracellular milieu and in glial cells. Adding to this complexity, Tau pathology varies significantly between the transgenic models that are used as validation tools for therapeutic interventions.¹² This makes it challenging if one explores an anti-Tau antibody in a transgenic mouse model and focuses the validation largely or solely on changes to abundance and specific phosphorylation of Tau, rather than on the impairments that are elicited at multiple and often subtle levels in different functional domains.

In this study, we generated a novel Tau antibody, RNJ1, and compared it in vitro and in vivo with the clinically tested Tau antibody HJ8.5 (tilavonemab).¹³ We first validated the antibodies’ capacity to bind monomeric and aggregated forms of human Tau and to neutralize aggregation induced by Tau seeds from tauopathy mice and human Alzheimer’s disease tissue in two independent Tau biosensor cell lines.¹⁴ We next found that RNJ1, but not HJ8.5, reduced total and phospho-Tau levels and improved behavioural deficits in the K369I mutant human Tau transgenic model K3. This strain displays Tau pathology and pronounced motor impairment with an early age-of-onset at 4 weeks, that manifests as a lack of locomotor ability and coordination and can be readily assessed by the Rotarod test. Given that protein changes in Alzheimer's disease and animal models are proxies for neuronal dysfunction,^6,15,16 in addition to focusing solely on changes to tau pathology, we conducted proteomic and phosphoproteomic analyses to investigate how the Tau antibodies induce multiple subtle changes to Tau and its interactions and thereby collectively contribute to improvements in different functional domains. Importantly, we found that both antibodies shifted the balance in multiple domains towards reinstalling a wild-type (WT) state of homeostasis in K3 mice. The shift elicited by antibody-treatment was induced at the level of both the proteome and the phosphoproteome. Notably, RNJ1 elicited more pronounced shifts in both domains, as evidenced by a significantly higher number of differentially abundant proteins and phosphopeptides. Our study highlights the potential for using proteomic changes as a correlate of Tau immunotherapeutic efficacy and advocates RNJ1 for clinical testing.

Materials and methods

The following materials and methods are described in the Supplementary material: antibodies; preparation of recombinant human Tau and single chain variable fragment (scFv) library panning; preparation of monoclonal scFvs for ELISA; purification of scFvs and western blot analysis; generation of mouse IgG RNJ1; epitope mapping of RNJ1 and confirmation of HJ8.5 epitope through indirect ELISA; determination of antibody EC₅₀ against monomeric hTau-441; determination of antibody binding curves against aggregated sarkosyl-insoluble Tau; surface plasmon resonance; generation of Tau RD P301S SH-SY5Y FRET biosensor cells; cell culture and Tau seed neutralization assays; large scale production of RNJ1 and HJ8.5 for treatment study; animals, immunization and behavioural tests; tissue processing for immunofluorescence; immunofluorescence imaging and image analysis; protein extraction from whole forebrain lysates and western blot analysis; sample preparation for mass spectrometry (MS); liquid chromatography-tandem MS analysis; MS data processing; MaxQuant data processing of the proteome and phosphoproteome of the antibody treatment study; protein kinase activity prediction; and other bioinformatic and statistical analyses.

Results

Isolation of the Tau antibody RNJ1 from a human phage display library and epitope mapping

To isolate a Tau-specific antibody, the Tomlinson I+J human synthetic single-fold single-chain variable fragment (scFv) library was panned in multiple rounds against full-length human recombinant Tau (hTau-441), yielding an enrichment of positive scFvs including RNJ1 (Supplementary Fig. 1A–D). This scFv bound to recombinant hTau-411 (the longest human brain Tau isoform), shown by ELISA and western blot analysis (Supplementary Fig. 1E and F). To characterize RNJ1 and determine its preclinical efficacy, its variable heavy and variable light chains were reformatted into a mouse IgG1 backbone containing a kappa light chain.

Using sequentially truncated forms of recombinant hTau-441 (Fig. 1A), we mapped the RNJ1 epitope by indirect ELISA and also confirmed the epitope of the clinically tested Tau antibody HJ8.5,¹³ which we had cloned and purified for benchmarking purposes. Both RNJ1 and HJ8.5 bound exclusively to forms of Tau containing the first 44 amino acids (aa); however, whereas RNJ1 had a binding site within aa 1–22, HJ8.5 recognized the variant containing aa 23–44, consistent with its reported epitope (aa 25–30) (Fig. 1B and C).¹³ Next, we employed an overlapping synthetic peptide library spanning aa 1–22 with peptide lengths of 10 aa and an offset of two aa to map the RNJ1 epitope at higher resolution (Fig. 1D). The peptides were conjugated to bovine serum albumin (BSA) and used as antigens in an indirect ELISA. Maximal binding was detected for peptides spanning aa 9–18 and 11–20, with residual binding observed to aa 13–22. Similarly, surface plasmon resonance (SPR) using the BSA-conjugated peptides as ligands and RNJ1 as analyte showed maximal binding for aa 9–18 and 11–20, with some binding to aa 13–22, and nearly no detectable signal for the remaining peptides (Supplementary Fig. 2). Together, these results show that RNJ1 binds to a segment of Tau spanning aa 9–22.

Figure 1

Antibody RNJ1 binds to an amino-terminal epitope of Tau and has stronger reactivity to aggregated Tau than the amino-terminal antibodies HJ8.5 and RNF5. (A) Schematic depicting Tau fragments expressed as fusion proteins with maltose-binding protein for epitope mapping of novel antibody RNJ1 and benchmarking antibody HJ8.5. (B) Epitope mapping of HJ8.5 by indirect ELISA using the Tau fusion proteins as antigens. (C) Epitope mapping of RNJ1 by indirect ELISA with the Tau fusion proteins. (D) Higher resolution mapping using 10 amino acid (aa) long overlapping peptides of Tau’s amino-terminal 22 aa conjugated to bovine serum albumin (BSA) (middle). (E) Indirect ELISA epitope mapping using BSA-conjugated non-phosphorylated and phosphorylated peptides (right). (F) Determination of RNJ1 and HJ8.5 concentration yielding half-maximal binding, EC₅₀, to monomeric recombinant human 2N4R Tau (hTau-441) by indirect ELISA. (G) Surface plasmon resonance (SPR) sensorgrams of RNJ1 and HJ8.5 (analytes) binding to monomeric recombinant hTau-441. Sensorgrams shown for antibody injections at 1.0, 3.0, 9.1, 27.3 and 82.0 nM. Both sensorgrams were fitted to a bivalent binding model. Experimental curves for RNJ1 are shown in red and for HJ8.5 in blue, and model fits in grey. (H) Binding curves of RNJ1, HJ8.5 and RNF5 from indirect ELISAs using sarkosyl-insoluble rTg4510 transgenic and human Alzheimer’s disease Tau seeds as antigens. (I) SPR sensorgrams of RNJ1, HJ8.5 and RNF5 (analytes) binding to sarkosyl-insoluble Tau seeds derived from rTg4510 mice. In B, C and E, n = 4 technical replicates. In D, F and H, n = 3 technical replicates. Data shown as mean ± standard error of the mean.

In addition, given that tyrosine 18 (Y18) is a substrate of Fyn and phosphorylation of this epitope is found in tauopathy and in relevant animal models,^17,18 we tested RNJ1 reactivity against the synthetic peptides aa 9–18, 11–20 and 13–22 phosphorylated at Y18. However, we did not detect binding, suggesting that RNJ1 binds to a segment spanning aa 9–22 that is not phosphorylated at Y18 (Fig. 1E).

RNJ1 displays stronger reactivity to aggregated Tau than HJ8.5 and RNF5

Given that our epitope mapping revealed adjacent epitopes for RNJ1 (aa 9–22) and HJ8.5 (aa 25–30) within Tau’s amino-terminal domain, we next interrogated whether this difference affects the binding profile to either monomeric or aggregated Tau. We first estimated the half-maximal binding concentration, EC₅₀, through indirect ELISAs using recombinant hTau-441 in a monomeric form as antigen, and found that the EC₅₀ of HJ8.5 (0.09 nM, 95% CI: 0.077–0.093 nM) was approximately five times lower than that of RNJ1 (0.51 nM, 95% CI: 0.46–0.57 nM) (Fig. 1F). We then measured the binding kinetic constants, using monomeric hTau-441 as ligand and the two antibodies at various concentrations as analytes in multi-cycle kinetic SPR analyses. Based on SPR sensorgrams, the antibodies displayed a good fit with a bivalent analyte model, with HJ8.5 showing higher affinity (equilibrium dissociation constant, K_D = 1.8 nM) to monomeric Tau than RNJ1 (K_D = 19.15 nM) (Fig. 1G).

To assess binding to aggregated Tau, we coated ELISA plates with sarkosyl-insoluble Tau obtained from P301L mutant Tau transgenic rTg4510 mice¹⁹ and human Alzheimer's disease brain lysates, and tested, in addition to RNJ1 and HJ8.5, a third amino-terminal antibody we have described previously, RNF5, which exhibits specificity for aa 35–44.²⁰ HJ8.5 and RNF5 both reached binding saturation at low antibody concentrations, whereas RNJ1 did not reach saturation within the concentration range tested, resulting in a significantly higher maximal response against sarkosyl-insoluble Tau from both mouse and human brain (Fig. 1H). To gain further insight into the binding properties of the antibodies to aggregated Tau, we performed SPR using sarkosyl-insoluble rTg4510 Tau as ligand and the antibodies as analytes. While the structural complexity of aggregated Tau precludes deriving binding kinetic constants with high confidence in SPR, it was evident from the sensorgrams that RNJ1 produced a much higher binding signal at equivalent antibody concentrations than HJ8.5 and RNF5 (Fig. 1I), reflecting the higher reactivity of RNJ1 against aggregated Tau determined by indirect ELISA (Fig. 1H). Together, the differential binding profile of RNJ1 to aggregated Tau determined by ELISA and SPR suggested that RNJ1 could have a higher capacity to neutralize templated aggregation induced by proteopathic Tau seeds.

RNJ1 shows remarkable efficiency at neutralizing proteopathic Tau seeds in vitro

Given that in Alzheimer's disease, pathological Tau has seeding capacity,²¹ and that RNJ1 displayed increased reactivity against aggregated Tau compared with HJ8.5 and RNF5, we sought to determine whether RNJ1 would have a higher capacity to neutralize Tau seeds in a biosensor system. For this, we employed the widely used Tau RD P301S FRET HEK293 (human embryonic kidney) cells which co-express the microtubule-binding repeat domain (RD) of P301S mutant Tau, fused to cyan fluorescent protein (CFP) or yellow fluorescent protein (YFP). In this system, internalization of exogenous Tau seeds induces aggregation of the recombinant RD Tau expressed by the cells, producing a Förster resonance energy transfer (FRET) signal that can be measured by microscopy or flow cytometry.²²

We treated the HEK293 biosensor cells with sarkosyl-insoluble rTg4510 Tau seeds in the presence or absence of the antibodies to assess the effect of antibody treatment on the formation of Tau inclusions, and measured the FRET signal by flow cytometry (Fig. 2A and B). Using Tau seeds at 200 ng/ml, we tested the antibodies at concentrations from 10 to 300 nM. When comparing the effects of the antibodies on Tau seeded aggregation, RNJ1 treatment was most effective at decreasing the mean integrated fluorescence intensities (IFI) for the FRET signal across all concentrations tested, reducing the IFI to ∼30% of the ‘no antibody’ control at the highest antibody concentration of 300 nM, compared with ∼42% for HJ8.5 and ∼70% for RNF5 (Fig. 2A–C).

Figure 2

RNJ1 inhibits seeding activity of sarkosyl-insoluble Tau in two distinct Tau biosensor cells. (A) Representative epifluorescence microscopy images of Tau repeat domain (RD) P301S FRET HEK293 biosensor cells 48 h post-exposure to rTg4510 transgenic mouse-derived Tau seeds in the presence or absence of 300 nM RNJ1, HJ8.5 or RNF5. All samples were treated with the same amount of Tau seeds. Scale bar = 100 µm. (B) Representative Förster resonance energy transfer (FRET)-based flow cytometry scatter plots of the human embryonic kidney (HEK)293 biosensor cells treated with rTg4510 Tau seeds with and without antibody. The percentage of FRET-positive cells is shown as mean ± standard error of the mean. No FRET signal was detected in biosensor cells without Tau seed exposure. (C) HEK293 biosensor cells treated with rTg4510 seeds and Tau antibodies. The percentage reduction in integrated FRET intensity values from flow cytometry analysis relative to the no antibody control values is shown. (D) Complementary Tau seed neutralization assay performed in a novel Tau RD P301S SH-SY5Y FRET biosensor system. (E) Seeding neutralization assay in HEK293 cells performed as in C but with Alzheimer’s disease (AD) Tau seeds. In B and C, n = 4 technical replicates for each concentration, 50 000 cells analysed per replicate. In D and E, n = 3 technical replicates for each concentration, 50 000 cells analysed per replicate. Statistical analysis was performed using two-way ANOVA with Holm–Sidak’s multiple comparison correction. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

We next used a complementary FRET-biosensor system we had established in human SH-SY5Y neuroblastoma cells to assess neutralization of Tau seeding in a more neuron-like model. Again, RNJ1 displayed superior neutralization capacity, decreasing the IFI of the FRET signal to ∼29% of the ‘no antibody’ control at the highest antibody concentration tested (300 nM) compared with ∼45% and ∼55% for HJ8.5 and RNF5, respectively (Fig. 2D). The mean IFI was consistently lower for RNJ1 than for HJ8.5 across all concentrations tested, although the difference achieved statistical significance only at 100 nM. Of note, RNJ1 was significantly superior to RNF5 across the four tested concentrations.

We then asked whether the antibodies were also capable of neutralizing Alzheimer’s disease brain-derived Tau seeds. We therefore treated the HEK293 biosensors with sarkosyl-insoluble Alzheimer’s disease Tau (Fig. 2E). Treatment with 100 nM RNJ1 reduced seeding intensity to nearly 60% of the seeding detected with the ‘no antibody’ control, whereas inhibition with equivalent doses of HJ8.5 only achieved a reduction to 88%. Treatment with RNJ1 achieved significantly higher seeding inhibition relative to HJ8.5 at both 20 and 100 nM concentrations. Together, these results show that RNJ1 exhibits a superior capacity to neutralize Tau seeds obtained from both a tauopathy mouse model and human Alzheimer’s disease brain tissue. Importantly, because the three tested antibodies bind to an amino-terminal epitope which is lacking in the Tau RD constructs expressed in the biosensor cells, the observed seeding inhibition is likely to occur by binding to exogenous Tau seeds rather than to RD Tau expressed by the biosensor cells.

RNJ1 treatment improves motor functions in the K3 mouse model of tauopathy

Having established that HJ8.5 and RNJ1 differ in their binding profiles to and neutralization capacity of seed-competent Tau, we next compared their efficacy, alone and together, in a longitudinal study using the K369I frontotemporal dementia-mutant Tau transgenic K3 mouse model of tauopathy.²³ K3 mice display Tau pathology and pronounced motor impairment with an early age-of-onset at 4 weeks, that manifests as a lack of locomotor ability and coordination as assessed, for example, in the Rotarod test.²⁴

For treatment group allocation, 48 K3 mice and 12 age-matched WT littermates (even male-to-female ratio) were subjected to a Rotarod baseline test at 4 weeks of age, and then allocated based on an equivalent average performance to five treatment groups (Fig. 3A): K3 mice treated with RNJ1 (K3^RNJ1), HJ8.5 (K3^HJ8.5), an antibody combination (K3^R+H) and vehicle (K3^Veh), plus vehicle-treated WT mice (WT^Veh). The mice received 14 weekly intraperitoneal doses of the antibody at 50 mg/kg for the single treatments and 25 mg/kg each for the combination, starting at 5 weeks of age. Rotarod performance was tracked longitudinally every 4 weeks, with a final test at the end of treatment. The K3 mice in all groups showed a significantly impaired performance in the Rotarod test compared to WT^Veh, with WT^Veh mice reaching a maximum latency-to-fall (180 s) in nearly every test session (Fig. 3B, top). K3^RNJ1 mice displayed a higher mean latency-to-fall than all other K3 groups at every test session after treatment start, but these differences did not reach statistical significance (Fig. 3B, top). To assess the overall performance of the groups over the course of treatment, we also calculated the area under the curve (AUC) of the latency-to-fall. Notably, K3^RNJ1 mice showed a 27% increase in the mean AUC compared to K3^Veh, without reaching statistical significance (P = 0.0911) (Fig. 3B, bottom). K3^HJ8.5 and K3^R+H mice showed no improvements throughout treatment or at time of completion compared to K3^Veh (Fig. 3B). Interestingly, we found that K3^Veh females performed better than K3^Veh males, with the second and third Rotarod tests (8 and 12 weeks of age) displaying significant differences (Fig. 3C, top), and the AUC being significantly higher for K3^Veh females (Fig. 3C, bottom). This prompted us to stratify the treatment groups based on sex to increase the sensitivity of the behavioural read-out (Fig. 3D).

Figure 3

RNJ1 treatment improves motor function in female K3 mice. (A) Experimental design of longitudinal treatment study: Baseline behaviour and locomotor function were determined in 4 week-old K369I Tau mutant K3 mice with a Tau pathology and wild-type (WT) littermate controls to assign mice into experimental groups with equivalent average Rotarod performance. The experimental groups were vehicle-treated WT mice (WT^Veh), and K3 mice treated with vehicle (K3^Veh), RNJ1 (K3^RNJ1), HJ8.5 (K3^HJ8.5) or an antibody combination (K3^R+H). Treatments were done weekly, and Rotarod performance was assessed every 4 weeks for 14 weeks. (B) Rotarod mean latency-to-fall (top) for all study arms across the treatment period (n = 12 mice per group) were compared by calculating the corresponding area under the curve (AUC) (bottom). (C) Mean latency-to-fall off the Rotarod displayed separately for male and female mice in the K3^Veh group (top) and corresponding area under the curve (AUC) (bottom). Female K3 mice showed improved locomotor function and balance compared to K3 males (n = 6 mice per group). (D) Group-based sex-stratified mean latency-to-fall across the treatment period (n = 6 mice per group) (top) and corresponding AUC (bottom). Female K3^RNJ1 mice showed improved locomotor performance compared with female K3^Veh mice. (E) Mean weight per group across the study. (F) Mean forelimb grip strength per group at baseline and at the end of treatment. In B–F, data are represented as mean ± standard error of the mean. Statistical comparisons for Rotarod AUC and grip strength in the treatment groups were performed using one-way ANOVA with Holm–Sidak’s multiple comparison correction or an unpaired t-test for comparison of Rotarod AUC based on sex in C. Comparison of the Rotarod latency-to-fall across time points was performed with a two-way ANOVA with Holm–Sidak’s multiple comparison correction. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Sex-stratified analysis of the treatment groups revealed that K3^RNJ1 females significantly improved in their Rotarod performance, with the second and third tests (8 and 12 weeks of age) showing a trend towards improvement, that reached statistical significance by the fourth test (16 weeks of age) (Fig. 3D, top right). This improvement was also reflected by a statistically significant increase in the AUC of latency-to-fall (Fig. 3D, bottom left). In contrast, the treated males showed no improvements (Fig. 3D, top left and bottom left). Weight (Fig. 3E) and grip strength averages (Fig. 3F) were significantly different between WT and all K3 groups, with no treatment effect. Together, our findings demonstrate that RNJ1 effectively improved motor function in female K3 mice.

RNJ1 treatment causes subtle reductions in total and phospho-Tau levels in brain homogenates of female K3 mice

Given that the RNJ1 treatment improved Rotarod performance in female K3 mice, we evaluated the biochemical and histopathological changes induced by the antibody treatment in the female cohort. Tau accumulates and undergoes hyperphosphorylation in disease and in K3 mice.²⁵ We therefore assessed immunoreactivity for total (mouse and human) Tau, for Tau phosphorylated at the pathological AT8 (pS202/pT205) and AT180 (pT231) epitopes by western blotting using whole forebrain lysates (Fig. 4A). This revealed a subtle, yet statistically significant mean reduction of ∼12% in total Tau (P = 0.0102) and ∼29% in AT180 (P = 0.0350) reactivity in the K3^RNJ1 group compared with K3^Veh (Fig. 4A and B), with the K3^R+H group also showing a statistically significant reduction of 8.5% in total Tau (P = 0.0490). Levels of AT8 displayed greater variability, with K3^RNJ1 mice showing a ∼27% reduction; however, this reduction did not reach statistical significance (P = 0.1069) (Fig. 4A and B).

Figure 4

Total and phospho-Tau levels are reduced in RNJ1-treated female K3 mice. (A) Immunoblots of whole forebrain lysates from female K3 mice probed for total Tau (DakoTau), S202/T205-phosphorylated Tau (AT8) and T231-phosphorylated Tau (AT180). (B) Quantification of A using total protein stain (Revert^TM 700) for normalization, represented as fold-change relative to the K3^Veh control group (n = 6 per group). (C) Representative (×20) images of AT8 immunoreactivity with DAPI counterstaining for nuclei in the hippocampus and motor cortex of female mice. Scale bars = 400 µm (top) and 100 µm (bottom). (D) Quantification of integrated fluorescence intensity in hippocampus and cortex in C. K3^Veh, n = 6; K3^RNJ1, n = 6; K3^HJ8.5, n = 6; K3^R+H, n = 5 (one mouse excluded due to tissue damage). (E) Rotarod performance across the treatment study [Rotarod area under the curve (AUC)] reveals a moderate inverse correlation with AT8 immunoreactivity for the hippocampus (left) (Pearson’s correlation r = −0.62, P = 0.0018) and cortex (right) (Pearson’s correlation r = −0.60, P = 0.0025). (F) Mass spectrometry quantification of protein abundance for the K369I human recombinant Tau (hTau) transgene and endogenous mouse Tau (mTau, n = 4 female mice per group). (G) Levels of hTau K369I show a strong inverse correlation with performance in Rotarod test (Pearson’s correlation r = −0.81, P = 0.0013), whereas endogenous mouse Tau levels only show a weak inverse correlation (Pearson’s correlation r = −0.32, P = 0.32). (H) Heat map of Tau phosphopeptides detected through quantitative mass spectrometry. Phosphopeptides with fold-changes that achieved P < 0.05 are highlighted with an asterisk. Data in the heat map represented as mean log₂ (fold-change) relative to K3^Veh. All other data represented as mean ± standard error of the mean. Statistical comparisons were performed with one-way ANOVA with Holm–Sidak’s post hoc multiple comparison’s test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. K3^Veh, K3^RNJ1, K3^HJ8.5, K3^R+H = K3 mice treated with vehicle, RNJ1, HJ8.5 or an antibody combination (R+H); WT^Veh = vehicle-treated wild-type mice.

We complemented the biochemical analysis of the forebrains of antibody-treated K3 females by assessing the levels of total and phospho-Tau in the sarkosyl-soluble and -insoluble fraction (Supplementary Figs 3A and B, 4A and B). In the sarkosyl-soluble fraction, the AT180 and pS262/pS356 (12E8) phospho-Tau epitopes were reduced in all treatment groups. AT8 immunoreactivity showed no differences across groups, whereas total Tau was reduced by ∼17% in the K3^RNJ1 group (P = 0.0103) (Supplementary Fig. 3A and B). In the sarkosyl-insoluble fraction, all treatment groups showed decreased immunoreactivity for 12E8, with RNJ1 showing the highest reduction at ∼26% of K3^Veh control levels (P = 0.0002). K3^RNJ1 mice also showed reductions in mean immunoreactivity for AT8 (∼43%; P = 0.4626) and total Tau (∼13%; P = 0.2703), but these changes were not statistically significant (Supplementary Fig. 4A and B). Moreover, we found that AT8 (Pearson’s correlation r = −0.55, P = 0.0058) and DakoTau (Pearson’s correlation r = −0.29, P = 0.018) immunoreactivity in the total lysate presented a significant inverse correlation with Rotarod performance of K3 female mice (Supplementary Fig. 5).

We also assessed AT8 reactivity by immunofluorescence in motor cortex and hippocampus (Fig. 4C), brain areas characterized by Tau deposition in K3 mice.^20,24 The mossy fibres of dentate granule cells, which innervate hilar cells and CA3 pyramidal cells, exhibit particularly pronounced reactivity (Fig. 4C). The integrated density of AT8 immunofluorescence in the hippocampus was decreased in some K3^RNJ1 mice, but this reduction was not statistically significant (Fig. 4C and D). In the motor cortex, AT8 immunoreactivity showed predominantly diffuse neuropil staining, with additional intense reactivity in the soma of a subset of neurons (Fig. 4C). The integrated density of AT8 immunofluorescence in the motor cortex showed a high degree of variability within treatment arms, and group differences were not statistically significant (Fig. 4C and D). However, when inspecting the integrated AT8 fluorescence intensity against the AUC of the Rotarod latency-to-fall for individual K3 mice in all treatment groups, we found a moderate correlation for both the hippocampus (Pearson’s correlation r = −0.62, P = 0.0018) and cortex (Pearson’s correlation r = −0.60, P = 0.0025) with Rotarod performance (AUC) (Fig. 4E). Taken together, these findings demonstrate subtle reductions in Tau pathology in RNJ1-treated K3 mice, which were less pronounced in the other treatment groups, correlating with motor impairment.

RNJ1 treatment reduces Tau levels and phosphorylation as shown by proteomics/phospho-proteomics

We next performed a quantitative proteomic and phosphoproteomic MS analysis from whole forebrain lysates of treated female mice as an unbiased approach to gain insight into differences between K3 and WT mice, as well as the effect of antibody treatments. To this end, we employed a TMTpro 18-plex labelling approach for quantitative analysis by tandem mass spectrometry of both tryptic peptides and phosphopeptides enriched by the ‘TiSH’ method.²⁶ Given the sample-size restrictions of the labelling method, we focused the MS analysis on the WT^Veh, K3^Veh, K3^RNJ1 and K3^HJ8.5 groups, and excluded the combination group (K3^R+H) as it had not shown synergistic effects in behavioural improvements or in Tau pathology determined by western blotting.

We found that human Tau showed a reduction of 17.7% in mean abundance in K3^RNJ1 compared with K3^Veh (P = 0.0559), whereas K3^HJ8.5 displayed a 9.1% reduction (P = 0.2124). Changes in levels of endogenous mouse Tau were more subtle, with the K3^RNJ1 group showing a 6.7% reduction in mean abundance (P = 0.1960) (Fig. 4F). Interestingly, the levels of human Tau displayed a strong inverse correlation with the Rotarod performance (Pearson’s correlation r = −0.81, P = 0.0013), which was particularly pronounced for the K3^RNJ1 group (Fig. 4G). Mouse Tau levels, on the other hand, showed only a weak inverse correlation which was not statistically significant (Pearson’s correlation r = −0.32, P = 0.32). This, together with the western blot analysis, suggests that a reduction in human Tau may contribute to therapeutic efficacy in K3 mice.

We next analysed the quantitative phosphopeptide MS data for changes in the phosphorylation of individual Tau phosphosites. We detected a total of 41 Tau phosphopeptides, comprising 27 unique phosphosites in Tau (including the well-known disease-relevant epitopes AT270, AT8, AT100, AT180, 12E8, PHF1 and PG5) (Fig. 4H). The antibody treatments affected phosphorylation at these sites to various degrees. Four phosphopeptides, pS202 being part of the AT8 epitope, pS404 and pS400/pS404 being part of the PHF1 epitope, and pS409 being part of the PG5 epitope, achieved statistically significant reductions in abundance in K3^RNJ1 compared with K3^Veh, whereas in K3^HJ8.5 mice, pS262 and pS356 (together forming the 12E8 epitope) were decreased. Together, our immunoblotting, histopathological and quantitative mass spectrometry analyses revealed that RNJ1-treated female K3 mice, different from HJ8.5-treated mice, had decreased levels of total and hyperphosphorylated Tau, and that these reductions correlated with Rotarod performance.

Quantitative proteomics reveals deregulated metabolic and microtubule-related proteins in K3 mice

To elucidate how RNJ1 treatment affects proteins with which Tau directly or indirectly interacts, we first aimed to delineate how the brain-wide proteome of K3 mice differs from that of wild-type mice. We found that 342 proteins out of 6622 proteins detected in our analysis exhibited differential abundance [DA, defined as fold-change in protein levels that achieve P < 0.05 after false discovery rate (FDR) correction] in K3^Veh relative to WT^Veh mice, of which 165 were decreased and 177 increased in abundance (Fig. 5A). Intriguingly, of these DA proteins, only ∼10% have been reported to interact with Tau via proximity-labelling MS,⁶ encoding predominantly cytoskeletal proteins. This distribution implies that the majority of the 342 DA proteins may be affected indirectly (Fig. 5B and C).

Figure 5

Quantitative mass spectrometry analysis shows deregulation of proteins involved in metabolic pathways and microtubule-based processes in female K3 mice. (A) Volcano plot of proteins with differential abundance (DA) in K3 mice treated with vehicle (K3^Veh) relative to vehicle-treated wild-type mice (WT^Veh) brains showing 342 DA proteins (165 proteins with lower and 177 with higher abundance), of 6622 total proteins detected in this study (6281 with no significant changes). (B) Venn diagram depicting that only a small subset of the 342 proteins have recently been reported as Tau interactors via proximity-labelling mass spectrometry.⁶ (C) Volcano plot displaying the subset of proteins in A that are reported Tau interactors. (D) Over-representation analysis (ORA) for proteins decreased in K3^Veh mice revealed Gene Ontology (GO) terms and Reactome pathways related to metabolic processes as being highly enriched, in particular proteins annotated to the terms ‘Mitochondrial matrix’ (GO Cellular Component) and the ‘Pyruvate metabolism and citric acid (TCA) cycle’ (Reactome pathway). GO ORA for proteins with higher abundance in K3^Veh mice shows a strong over-representation of microtubule-based processes, with axonal transport being highly enriched. (E) Gene-concept networks showing DA proteins in the comparison between K3^Veh and WT^Veh associated with the most highly enriched GO terms, the latter being represented as grey nodes with connecting lines to differentially abundant proteins in K3^Veh mice. BP = Biological Process; CC = Cellular Component; ER = endoplasmic reticulum; MF = Molecular Function.

We next subjected these DA proteins to a Gene Ontology (GO) and Reactome pathway over-representation analysis (ORA) to identify functional domains that are potentially affected by these DA proteins in K3 mice. By applying ORA to the protein set with lower abundance in K3 mice, several metabolic processes were found to be strongly over-represented (Fig. 5D, left), with the most significantly enriched term in Reactome pathways being ‘Pyruvate metabolism and citric acid (TCA) cycle’. Conversely, in the protein set with higher abundance, over-represented terms were predominantly associated with microtubule-related processes, particularly comprising proteins with roles in axonal transport (Fig. 5D, right). Several kinesins, tubulin subtypes and dynactin subunits were upregulated, as was stathmin 1, a key regulator of microtubule dynamics, as shown by gene-concept networks of proteins annotated to enriched terms in this comparative analysis (Fig. 5E).^27,28 Moreover, in addition to identifying small clusters of DA proteins involved in neurite outgrowth and clathrin-dependent endocytosis, STRING analysis revealed protein-protein interaction (PPI) networks of DA proteins involved in metabolic and microtubule-related processes (Supplementary Fig. 6). The pronounced decrease in proteins involved in metabolic processes, together with the increase in microtubule-related proteins, mirrors phenotypic impairments in mitochondrial function and axonal transport previously reported by us and other groups in this and other tauopathy models.^29-31 These analyses underscore the notion that proteomic alterations in K3 mice accurately model impairments across diverse functional domains.

Antibody treatment restores deregulated protein levels in K3 mice

We then asked whether the antibody treatment would affect the proteome (and thereby functionality) in K3 mice to restore the deregulated protein levels and associated pathways in this model. Firstly, principal component analysis (PCA) revealed that the K3^RNJ1 mice clustered separately from the other K3 groups (K3^Veh and K3^HJ8.5) (Fig. 6A). Additionally, whereas the levels of only five proteins (three up and two down) were DA in the K3^HJ8.5 group compared to K3^Veh, RNJ1 treatment induced DA changes to 323 proteins (186 up and 137 down) relative to the K3^Veh group (Fig. 6B). Together, these two analyses indicate substantial proteome-level differences in K3^RNJ1 relative to K3^Veh mice.

Figure 6

Brain-wide proteomic changes in response to antibody treatment in female K3 mice. (A) Principal component analysis (PCA) reveals three clusters for PC2 and PC3, with the K3 mice treated with RNJ1 (K3^RNJ1) cluster shifting towards vehicle-treated wild-type mice (WT^Veh). (B) Volcano plot for K3^RNJ1 mice showing differentially abundant (DA) proteins (P < 0.05 after false discovery rate correction), with 186 decreased and 137 increased proteins relative to K3^Veh (left). The K3 mice treated with HJ8.5 (K3^HJ8.5) group shows higher similarity to vehicle-treated K3 mice (K3^Veh), with only five proteins being DA (right). (C) Analysis of therapeutic impact by focusing on the set of 342 DA proteins in K3^Veh compared to WT^Veh (pie chart and Fig. 5A). (D) The heat maps represent the fold-change of proteins in the antibody-treated mice compared to K3^Veh, with the top heat map showing decreased and the bottom heat map increased DA proteins in K3^Veh compared with WT^Veh, showing a reversal by the antibodies towards WT^Veh. (E) Scatter plot (top: RNJ1; bottom: HJ8.5) displaying the direction of change in abundance of the 342 DA proteins in WT^Veh and antibody-treated mice relative to K3^Veh. The majority of the 342 proteins (∼75.1% for RNJ1; ∼72.5% for HJ8.5) fall into quadrants I and III, with quadrant I displaying proteins that are higher in WT^Veh and antibody-treated mice relative to K3^Veh, and quadrant III displaying proteins that are lower in WT^Veh and antibody-treated mice relative to K3^Veh, reiterating the reversal of the treatment groups towards WT. Proteins with significant fold-changes (DA) in both comparisons are represented as filled, and non-significant (NS) as open circles. The number of proteins falling into each quadrant is shown. (F) The Venn diagram represents the number of intersecting and non-intersecting proteins changed significantly (DA) in the K3^Veh versus WT^Veh and K3^RNJ1 versus K3^Veh comparisons. The heat map displays the row z-scores of log₂ transformed protein abundance in the overlap, showing a striking similarity in this subset for groups WT^Veh and K3^RNJ1, revealing proteins significantly affected in the K3 model, and by treatment, being restored to WT levels. Gene symbols in bold and underlined represent the corresponding DA proteins in K3^HJ8.5 compared with K3^Veh. Columns (individual mice) are clustered by correlation (top). Statistical comparisons of protein abundance were performed using a generalized linear model with Bayes shrinkage for group comparisons. P-values were derived from a moderated t-test and corrected for multi-hypotheses using the Benjamini and Hochberg (false discovery rate) method.³²

We then focused on the effects the treatment specifically had in relation to the 342 DA proteins in K3^Veh compared with WT^Veh given that this subset of proteins characterizes the K3 phenotype at the proteome level (Fig. 6C). In the first instance, we analysed changes with treatment in this subset of 342 proteins without considering statistical significance, as we posited that these changes could be subtle at the individual level, but could nonetheless result in a coordinated response. Interestingly, we found that the antibody treatments increased the abundance of the majority of the proteins that were downregulated (Fig. 6D, top, and Supplementary Fig. 7A) and decreased the abundance of the majority of the proteins that were upregulated in K3^Veh relative to WT^Veh (Fig. 6D, bottom, and Supplementary Fig. 7B), suggesting that antibody treatment shifts the levels of deregulated proteins in K3 towards those of WT. To further understand this coordinated shift, we examined the relationship between the levels of these 342 proteins in K3^RNJ1 and WT^Veh mice relative to K3^Veh mice (Fig. 6E, top). We first considered that antibody treatment increased or decreased protein levels with respect to K3^Veh mice when the fold-change values were larger or smaller than 1, respectively. We found that of the 165 proteins with decreased abundance in K3^Veh mice relative to WT^Veh, 131 (∼79.4%) were shifted to higher values in K3^RNJ1 mice (Fig. 6E, top, quadrant I). Similarly, 126 (∼71.2%) of the 177 proteins with higher levels in K3^Veh (quadrant III) were shifted to lower abundance in K3^RNJ1. Together, 257 of 342 (∼75.1%) DA proteins (K3^Veh versus WT^Veh) were closer in abundance to WT levels after RNJ1 treatment. The same analysis indicated a similar response in K3^HJ8.5 mice, with approximately 72.5% of these DA proteins (K3^Veh versus WT^Veh) also showing changes in the same direction as WT^Veh (see quadrants I and III) (Fig. 6E, bottom). Moreover, changes in protein levels in antibody-treated and WT^Veh mice relative to K3^Veh mice were positively correlated (Pearson’s correlation r = 0.42, P < 0.0001 for RNJ1, Fig. 6E, top; Pearson’s correlation r = 0.45, P < 0.0001, for HJ8.5, Fig. 6E, bottom).

We subsequently narrowed down the analysis to a subset of these 342 proteins that also met the criterion of being DA with antibody treatment compared with K3^Veh. We found 48 proteins that met this criterion in the K3^RNJ1 group (filled circles in Fig. 6E, top, and F), and only three in the K3^HJ8.5 group (filled circles in Fig. 6E, bottom, and F), with the latter also featuring in the 48 proteins identified in the K3^RNJ1 group. Remarkably, 45 of these 48 proteins in the RNJ1 group and all three in the HJ8.5 group reversed to WT levels with treatment. We again found a positive correlation between changes in protein levels in K3^RNJ1 and WT^Veh mice relative to K3^Veh mice (Pearson’s correlation r = 0.89, P < 0.0001; filled circles in Fig. 6E, top). Together, these analyses reinforce the notion that there is a restoration of protein levels in K3 towards WT following Tau antibody treatments.

In the subset of 45 overlapping DA proteins, we noticed clathrin light chain A (Clta) and clathrin coat assembly protein AP180 (Snap91), proteins involved in synaptic vesicle endocytosis, and both were decreased in abundance with RNJ1 treatment and in WT^Veh mice relative to K3^Veh. To further infer the functional domains that were altered in K3 mice and potentially restored by RNJ1 treatment, we performed ORA on the overlapping DA proteins and interestingly found the highest enrichment for the GO terms ‘Presynapse’ and ‘Endocytosis’, for proteins with lower abundance in WT^Veh and K3^RNJ1 relative to K3^Veh (Supplementary Fig. 8A and B). Furthermore, a functional PPI analysis revealed a network of 34 proteins within this subset, with six of those proteins annotated to the GO term ‘Presynapse’ and three annotated to the GO term ‘Synaptic vesicle’ (Supplementary Fig. 8C). In summary, our focused proteomic analysis indicates that both antibody treatments initiated a response to restore protein levels within the subset of proteins that characterize the K3 phenotype.

Differential analysis of the phosphoproteome also reveals a reversion of RNJ1-treated K3 mice to wild-type levels

To understand the additional effects of antibody treatment, we performed a differential quantitative analysis of the phosphoproteome. Consistent with the proteomic analysis (Fig. 6A), a PCA of the phosphoproteome revealed three clusters: the first formed by K3^RNJ1 mice, the second by WT^Veh and the third by both K3^HJ8.5 and K3^Veh, indicating that K3 mice have an altered phosphoproteome, which, in turn, is affected by RNJ1 treatment (Fig. 7A). We first identified 541 DA phosphopeptides in K3^Veh mice compared with WT^Veh, with 283 phosphopeptides showing lower and 258 higher abundance in K3^Veh (Fig. 7B, left). Interestingly, 653 phosphopeptides were DA in K3^RNJ1 compared with K3^Veh, of which 333 displayed lower and 320 higher abundance, reflecting a shift in the phosphoproteome with RNJ1 treatment in line with the PCA analysis (Fig. 7B, middle). In contrast, no DA phosphopeptide was detected in K3^HJ8.5 compared with K3^Veh (Fig. 7B, right).

Figure 7

Brain-wide phosphoproteomic changes in response to antibody treatment in female K3 mice. (A) Principal component analysis (PCA) of the phosphoproteome identifies three clusters, vehicle-treated wild-type mice (WT^Veh), K3 mice treated with RNJ1 (K3^RNJ1), and one formed by vehicle-treated K3 mice (K3^Veh) and K3 mice treated with HJ8.5 (K3^HJ8.5). (B) Volcano plots show differentially abundant (DA) phosphopeptides (P < 0.05 after false discovery rate correction) for the K3^Veh versus WT^Veh, K3^RNJ1 versus K3^Veh and K3^HJ8.5 versus K3^Veh comparisons. In total, 541 phosphopeptides were DA between K3^Veh and WT^Veh mice, and 653 between K3^RNJ1 and K3^Veh mice. The analysis of treatment effects on the phosphoproteome was focused on the 541 DA phosphopeptides in K3^Veh relative to WT^Veh. (C) Scatter plots (left: RNJ1, right: HJ8.5) display the direction of change in abundance of the 541 DA phosphopeptides in WT^Veh and antibody-treated mice relative to K3^Veh. The majority of the 541 DA phosphopeptides (∼82.1% for RNJ1; ∼75.4% for HJ8.5) fall into quadrants I and III, with quadrant I displaying phosphopeptides that are higher in abundance in WT^Veh and antibody-treated mice relative to K3^Veh, and quadrant III those that are lower in WT^Veh and antibody-treated mice relative to K3^Veh, indicating a reversal of the treatment groups towards WT. Phosphopeptides with significant fold-changes in both comparisons are represented as filled circles and non-significant as open circles. The number of phosphopeptides falling into each quadrant and the corresponding percentage of the total of DA phosphopeptides is shown. (D) Venn diagram representing the number of intersecting and non-intersecting DA phosphopeptides in both K3^Veh versus WT^Veh and K3^RNJ1 versus K3^Veh comparisons. The heat map displays the row z-scores of log₂ transformed phosphopeptide abundance in the overlap, showing striking similarity in this subset for WT^Veh and K3^RNJ1, thereby revealing that phosphopeptides significantly affected in the K3 model and by treatment were restored to WT levels. (E) Protein–protein interaction (PPI) network of the proteins containing DA phosphopeptides shown in C. Node size and colour relate linearly to the degree of connectivity of the node in the network. (F) SynGO analysis of proteins with DA phosphopeptides shown in C, showing enrichment of presynaptic and postsynaptic proteins. The number of proteins within the analysed set annotated to each category is shown. (G) S2448-phosphorylated mTOR and T188-phosphorylated MAPK1 display higher abundance in K3^RNJ1 and WT^Veh mice relative to K3^Veh. Data in G are shown as mean ± standard error of the mean. (H) Network-based kinase activity prediction analysis. Scatter plots (left: RNJ1, right: HJ8.5) display the overlap of kinases with differential activity (P < 0.2 for swing_k scores) in WT^Veh and antibody-treated mice relative to K3^Veh. Statistical comparisons of phosphosite abundance were performed using a generalized linear model with Bayes shrinkage for group comparisons. P-values were derived from a moderated t-test and corrected for multi-hypotheses using the Benjamini and Hochberg (false discovery rate) method.³² *P < 0.05, **P < 0.01, ***P < 0.001. cytoskel = cytoskeleton; interm = intermediate; postsyn = postsynaptic; presyn = presynaptic.

Akin to the proteomic analysis (Fig. 6C), we used the 541 phosphopeptides that are DA between K3^Veh and WT^Veh to determine how antibody treatments affect this subset. We first considered that antibody treatment increased or decreased phosphopeptide levels with respect to K3^Veh mice when the fold-change values were larger or smaller than 1, respectively (open and filled circles in Fig. 7C). We found that 205 out of 258 (∼79.4%) phosphopeptides with higher abundance in K3^Veh relative to WT^Veh were reverting towards lower abundance in the K3^RNJ1 group (Fig. 7C, left, quadrant III). Similarly, 239 of 283 phosphopeptides (∼84.5%) with lower abundance in K3^Veh relative to WT^Veh (quadrant I) had higher abundance in K3^RNJ1. Notably, with a total of 444 out of 541 (∼82.1%) phosphopeptides being shifted in abundance towards WT^Veh, this degree of reversal in the phosphoproteome was even more pronounced than in the above proteomic analysis (Fig. 6E). A similar albeit less prominent reversal was observed in K3^HJ8.5 mice, with 183 phosphopeptides being decreased and 217 increased both in K3^HJ8.5 and WT^Veh relative to the K3^Veh group (quadrants I and III), together amounting to 408 out of 541 (∼75.4%) phosphopeptides being reverted towards WT (Fig. 7C, right).

We then examined the overlap of phosphopeptides that were statistically different in both WT^Veh and K3^RNJ1 relative to K3^Veh and found that it comprised 101 phosphopeptides, originating from 90 proteins. Remarkably, 99 of these 101 phosphopeptides had restored levels in K3^RNJ1 mice (Fig. 7D), mirroring the above proteomic findings. This set of 99 phosphopeptides originated from a total of 89 proteins, which we subjected to a PPI analysis which revealed a network formed by 41 proteins, with multiple functional associations between these proteins that are affected in the K3 model and are restored to WT levels upon RNJ1 treatment (Fig. 7E). Moreover, an analysis of the 89 proteins with phosphopeptides restored to WT^Veh levels in K3^RNJ1 mice using the SynGO database³³ revealed that 28 of these are proteins found in synapses, with the highest enrichment and protein count (17) found for postsynaptic proteins (Fig. 7F).

Upon examination of key proteins within the PPI network, the increase in phosphorylation of MAPK1 at threonine 188 and of mTOR at serine 2448 in K3^RNJ1 mice is noteworthy, with both phosphopeptides also being elevated in WT^Veh compared with K3^Veh (Fig. 7G). To explore these findings further, we performed a phosphosite network-based analysis to infer the degree of activity of individual kinases, which again revealed a strong correlation between K3^RNJ1 and WT^Veh mice relative to K3^Veh when assessing the overlap of kinases with differential activity (Pearson’s correlation r = 0.88, P < 0.0001; Fig. 7H, left), whereas K3^HJ8.5 mice showed a weaker correlation (Pearson’s correlation r = 0.49, P < 0.0001; Fig. 7H, right). Furthermore, both MAPK1 and mTOR revealed increased predicted activity in K3^RNJ1 and WT^Veh mice, in line with the increased phosphorylation of MAPK1 pT188 and mTOR pS2448 sites. These phosphorylation events may point towards an upregulation of the mTOR and MAPK/ERK signalling pathways, both of which are involved in a multitude of neuronal functions including synaptic plasticity.^34,35

Differential dysproteostasis may be linked to different treatment outcomes in male and female K3 mice

To understand why treatment outcomes differed between males and females in relation to Rotarod performance and Tau pathology, we carried out a separate quantitative proteomic and phosphoproteomic analysis of the forebrains of male WT^Veh, K3^Veh, K3^RNJ1 and K3^HJ8.5 mice (n = 4 mice per group), obtaining a slightly higher coverage of the proteome than in females (7453 compared with 6622 proteins) (Supplementary Fig. 9A). Notably, we found a significantly higher number of DA proteins in the K3^Veh versus WT^Veh comparison for the males (1556 decreased and 604 increased in K3^Veh) than for the females (165 decreased and 177 increased in K3^Veh), consistent with the observation that males showed a worse performance in the Rotarod test (Fig. 3C and Supplementary Fig. 9B and C). ORA analysis using GO and Reactome terms for DA proteins in the K3^Veh versus WT^Veh male comparison showed enrichment for terms shared with the K3^Veh versus WT^Veh comparison for the females, such as ‘mitochondrial membrane’ and ‘the citric acid cycle and respiratory transport chain’ for proteins decreased in K3, and ‘COPI-independent Golgi-to-ER retrograde traffic’ and ‘dynactin complex’ for proteins increased in K3 (Fig. 5 and Supplementary Fig. 9D). Interestingly, enrichment of synaptic terms such as ‘postsynaptic membrane’ and ‘presynaptic membrane’ was found only in the male subset, suggesting a higher impact on synaptic function in males. Moreover, ORA of DA proteins with higher abundance in K3 unique to the male comparison revealed enrichment of GO terms not detected in females, such as ‘cytoplasmic translation’, ‘carbohydrate catabolic process’ and ‘glycolytic process’ (Supplementary Fig. 9E). MCODE analysis of the above subset also identified a PPI network involved in ‘SNARE complex’ related processes, further suggesting more pronounced synaptic dysfunction in K3 males (Supplementary Fig. 9F).

Next, we analysed the effect of antibody treatment on the abundance of DA proteins in K3^Veh males relative to WT^Veh males. Akin to our female analysis, we observed a strong reversal towards WT levels in K3^RNJ1 male mice, that was less pronounced in K3^HJ8.5 (Supplementary Fig. 10A–C). The intersection of DA proteins in both K3^Veh versus WT ^Veh and K3^RNJ1 versus K3^Veh comparisons comprised 208 proteins, of which 201 (∼96.6%) were changed to WT levels in K3^RNJ1, showing a strong positive correlation (Pearson’s correlation r = 0.94, P < 0.0001; Supplementary Fig. 10B and C). The fraction of reversed proteins in K3^RNJ1 males (201 of 2160 DA proteins; 9.3%), however, was lower than for K3^RNJ1 females (45 of 342 DA proteins; 13.2%) (Fig. 6F and Supplementary Fig. 10C). Importantly, the reversal in the phosphoproteome was far less pronounced than for the females, as reflected by a small set of DA phosphopeptides in the male K3^RNJ1 versus K3^Veh comparison (12), and an intersection with the WT^Veh versus K3^Veh comparison of only three DA phosphopeptides (Supplementary Fig. 10D–F). Despite this, abundance changes in the proteome and phosphoproteome of both antibody treatments showed significant positive correlations with WT^Veh abundance. Notably, the initiation of a reversal of dysproteostasis in males occurred without significant changes in AT180 or AT8 Tau phosphorylation in whole forebrain lysates or the sarkosyl-soluble fraction as assessed by immunoblotting (Supplementary Fig. 11A–D).

Our findings reveal that Tau transgenesis induces large-scale proteomic and phosphoproteomic changes, possibly due to the role of Tau as a major neuronal scaffolding protein. Importantly, Tau-antibody treatment, to an extent, reverts these large-scale changes linked to alterations in molecular and cellular functions, e.g. mitochondrial matrix and axonal-transport, and this reversion contributes to improvements in functional outcomes (Fig. 8A and B). As discussed above, the apparent lack of treatment effects in male K3 mice might be linked to the varied restoration in the proteome and phosphoproteome compared to the females (Fig. 8C). Taken together, our study indicates that restoration of proteostasis is a fundamental principle of Tau immunotherapies.

Figure 8

Effects of Tau transgenesis and Tau immunotherapy on proteomic and phosphoproteomic profiles in mouse models. (A) Tau transgenesis induces large-scale proteomic and phosphoproteomic changes that are partially reverted by Tau-antibody treatment. (B) In K3 mice, these changes include large-scale alterations in molecular and cellular functions, e.g. mitochondrial matrix and axonal transport. RNJ1-immunotherapy partially reverts these changes, contributing to improvements in functional outcomes. (C) The apparent lack of treatment effects in male K3 mice might be linked to the varied restoration in the proteome and phosphoproteome compared to the females. CRMP = collapsin response mediator protein; ER = endoplasmic reticulum; TCA = citric acid cycle.

Discussion

Our comparative in vitro and in vivo analysis of two Tau antibodies, one generated by us (RNJ1) and the other by Dr Holtzman and colleagues (HJ8.5, explored clinically as tilavonemab and included here for benchmarking purposes),^36-38 revealed that these antibodies (in particular RNJ1) work towards restoring proteomic homeostasis, which we posit may serve as an additional, highly informative read-out for assessing therapeutic outcomes, rather than assessing solely Tau abundance and phosphorylation. Here, we first validated RNJ1 in the established HEK293 and a novel SH-SY5Y Tau biosensor cell system, revealing the antibody’s superior capacity in neutralizing proteopathic Tau seeds derived from various tissue sources. We therefore trialled RNJ1 in the K3 mouse model of tauopathy, including three immunization arms, RNJ1 (K3^RNJ1), HJ8.5 (K3^HJ8.5), the two antibodies combined (K3^R+H), and two control arms, K3 vehicle control (K3^Veh) plus wild-type vehicle control (WT^Veh). This demonstrated reductions in Tau pathology and improved motor functions for the RNJ1 treatment specifically in the female mice, different from the HJ8.5 and combination groups, with no synergistic effects observed for the combination treatment.

To gain a better insight into the antibodies’ impact at the brain-wide level, we conducted a comprehensive proteomic and phospho-proteomic analysis of K3 compared to WT brains, as well as the treatment arms. In the female K3 mice, this revealed 342 DA proteins compared to WT mice, many of which encoded metabolic and microtubule-associated proteins, strengthening functional validations conducted in K3 mice previously and thereby underscoring the value of leveraging proteomic analyses to evaluate therapeutic effects.^23,39 Importantly, the antibody treatments restored the levels of most proteins that were altered in K3 mice closer to those in WT mice, i.e. they induced changes towards restoring homeostasis, with RNJ1 being more effective than HJ8.5 in reverting the K3 proteomic signature towards controls.

With regard to achieving therapeutic outcomes, recent antibody developments have almost collectively shifted from targeting predominantly the amino-terminus to the mid-region or phosphorylated epitopes characteristic of toxic species. Although several recent studies have demonstrated superiority of mid-region antibodies in blocking the proteopathic spread in vitro, this has yet to be validated in a clinical setting.^40,41 Antibodies targeting the amino-terminus of Tau have so far suffered setbacks in clinical development, with semorinemab, gosuranemab and tilavonemab being discontinued due to their inability to achieve their primary efficacy endpoints in phase II trials.^42-44 Nonetheless, for lack of better antibodies, what argues for not abandoning the amino-terminus as a target is this domain’s implication in pathological misfolding, thereby contributing to neurotoxicity and synaptic dysfunction.^45,46

Here, we determined that the epitope of RNJ1 is located in the amino-terminal region and is proximal to the epitopes of HJ8.5 and RNF5, the latter being another pan-Tau antibody generated previously in our laboratory,²⁰ with the three epitopes being non-overlapping. When comparing the three antibodies’ Tau seed neutralization capacity in vitro, RNJ1 was significantly more efficient at neutralizing seeds from both the rTg4510 mouse model of tauopathy and from human Alzheimer’s disease brain tissue. While we did not explore the molecular mechanisms accounting for these differential effects, we posit that they may be related to differences in how the antibodies bind to aggregated Tau. This is in agreement with the observation that antibodies targeting amino-terminal epitopes of Tau can display differential profiles in the detection of pathological conformations of Tau.⁴⁶ The structural basis for the differential binding of amino-terminal antibodies to aggregated pathological Tau remains unresolved; however, our study suggests that it may have important implications for treatment efficacy and in the selection of therapeutic candidates.

Tau exists as multiple isoforms and is heavily post-translationally modified which in part affects its subcellular localization and aggregation propensity. Given that transgenic models only partially recapitulate the human pathology, reductions of distinct forms of Tau (as revealed e.g. with specific anti-phospho-Tau antibodies) may not be fully informative or predictive of functional improvements. We therefore resorted to an unbiased and more comprehensive assessment of treatment effects through quantitative proteomic and phosphoproteomic analyses as well as kinase activity prediction assays. Our initial approach was to establish the set of DA proteins in K3^Veh mice relative to WT^Veh mice, to then evaluate whether a response to the antibody treatments affects the levels of these proteins. Interestingly, we identified DA proteins, that are in agreement with impairments that we (and others) had identified previously in this and other tauopathy models, including impaired anterograde axonal transport, which affects kinesin-dependent vesicles and mitochondria.^23,31 It is noteworthy to find that a large protein subset involved in axonal transport is strongly upregulated, presumably in response to altered microtubule dynamics. Moreover, the pronounced downregulation in proteins involved in cellular respiration and metabolism mirrors reported functional mitochondrial impairments in K3 mice. We have shown previously that anterograde transport of mitochondria in K3 neurons is impaired, and we and others have shown neuronal mitochondrial dysfunction in this and other tauopathy mouse models.^29-31 Together, this underscores the validity of employing proteomic analyses to identify alterations in functional domains and assess the impact of therapeutic interventions.

The proteomic profiling of female RNJ1-treated mice revealed notable differences to those of K3^Veh control and HJ8.5-treated mice. Importantly, by focusing on the set of DA proteins in the K3^Veh versus WT^Veh comparison, a key observation was that most proteins had a change in abundance in the direction of WT levels for both antibody treatment arms, although many of these were subtle, especially for the K3^HJ8.5 treatment arm. Only a fraction of these were found to be statistically significant (DA) in the K3^RNJ1 versus K3^Veh comparison; however, protein levels within this fraction showed a very high correlation between K3^RNJ1 and WT^Veh mice, suggesting that treatment efficacy is reflected by a reversion to homeostatic protein abundance levels. It was noticeable to find an even stronger correlation in the phospho-proteomic dataset when focusing on the subset of DA phosphopeptides in K3^Veh relative to WT^Veh, again for both antibody arms but with a more pronounced effect for RNJ1. The overall effects of RNJ1 on the phosphoproteome level mirrored what we had found for the proteome, i.e. that treatment elicited a change in the same direction as WT, of which a subset was altered at a statistically significant level. Remarkably, this subset was almost completely restored to WT levels, again consistent with the changes we observed at the proteome level.

The proteomic data in the male cohort also reveal an initiation towards restoration of proteostasis for both antibody treatments, with RNJ1 again showing a stronger effect. However, in K3^Veh males, there was a notable increase in deregulated proteins compared with K3^Veh females, indicating a more significant impairment across various functional domains, which likely underpins the more pronounced decline in locomotor ability in males as evidenced by their performance on the Rotarod test. Notably, despite a partial reversal towards WT levels in K3^RNJ1 males, this improvement did not translate into enhanced Rotarod performance. This suggests that the greater impairment observed in K3 males may require a more substantial reversal to manifest as improvements in behavioural outcomes, which may be achieved by increasing the antibody dose or enhancing blood–brain barrier penetration.

Whereas the exact mechanisms that underlie the diverse proteomic and phosphoproteomic effects were not explored in this study, our findings indicate that, by targeting Tau’s amino-terminus, both antibody treatments improved the K3 proteomic and phospho-proteomic signature. Notably, RNJ1 demonstrated a more pronounced restorative effect than HJ8.5, which likely reflects improved neuronal function and could account for the observed enhancements in motor function in our study.

More generally, and possibly not restricted to amino-terminal antibodies, our data suggest that anti-Tau immunization works towards restoring homeostasis in a tauopathy model, presenting quantitative proteomics (rather than or in addition to pathological Tau assessments) as a viable strategy to validate therapeutic interventions in preclinical models. We foresee a more important role for omics technologies⁴⁷ in preclinical studies and possibly, also in a clinical setting. In the future, this will be facilitated by an increased exploitation of fluid markers including plasma and cerebrospinal fluid for both diagnosis and therapeutic validation.^48-51

Data availability

The mass spectrometry proteomics data of female and male mice have been deposited to the ProteomeXchange Consortium via the PRIDE⁵² partner repository with the dataset identifiers PXD045799 and 10.6019/PXD045799 (females) and PXD051830 and 10.6019/PXD051830 (males).

Acknowledgements

We thank Rowan Tweedale, Dr Adam Walker and Dr Victor Anggono for critical reading of our manuscript. We thank Keisha Roffey for help with mouse immunizations and Tishila Palliyaguru for assistance with tissue processing for histological analysis. We thank Dr Juan Carlos Polanco for assistance with Tau seeding assays. The authors would also like to thank Dr Martina Jones for technical assistance in performing SPR experiments and data interpretation. Image acquisition and analysis was performed at the Queensland Brain Institute’s Advanced Microscopy Facility, using an automated slide scanner (Axio Imager Z2, Zeiss) with a Metafer Vslide Scanner program (Metasystems) supported by The University of Queensland through the Strategic Funding grant DVCR22052A.

Funding

We acknowledge support from the Estate of Dr Clem Jones, the State Government of Queensland (DSITI, Department of Science, Information Technology and Innovation), the National Health and Medical Research Council of Australia (GNT1176326 and GA39196), and the Terry and Maureen Hopkins Foundation to J.G. R.M.N. was a recipient of the Yulgilbar Alzheimer's Research Program Fellowship.

Competing interests

A.J.vW. is founder of i-Synapse. The authors declare no other competing interests.

Supplementary material

Supplementary material is available at Brain online.