High-throughput discovery of fluoroprobes that recognize amyloid fibril polymorphs

Abstract

Aggregation of microtubule-associated protein tau into conformationally distinct fibrils underpins neurodegenerative tauopathies. Fluorescent probes (fluoroprobes) such as thioflavin T have been essential tools for studying tau aggregation; however, most of them do not discriminate between amyloid fibril conformations (polymorphs). This gap is due, in part, to a lack of high-throughput methods for screening large, diverse chemical collections. Here we leverage advances in protein-adaptive differential scanning fluorimetry to screen the Aurora collection of 300+ fluoroprobes against multiple synthetic fibril polymorphs, including those formed from tau, α-synuclein and islet amyloid polypeptide. This screen-coupled with excitation-multiplexed bright-emission recording (EMBER) imaging and orthogonal secondary assays-revealed pan-fibril-binding chemotypes, as well as fluoroprobes selective for fibril subsets. One fluoroprobe recognized tau pathology in ex vivo brain slices from Alzheimer's disease and rodent models. We propose that these scaffolds represent entry points for developing fibril-selective ligands.

Fig. 1. A high-throughput screening platform reveals fluoroprobes that recognize tau fibril polymorphs.

a, Schematic of the primary screening workflow and summary of the results. Synthetic tau fibrils with diverse conformations were generated using either recombinant WT or P301S tau (0N4R splice isoform) mixed with 13 different inducers (see (b) and Supplementary Table 1). These 26 fibril samples were purified and incubated with the Aurora dye library in 384-well plates and then heated to generate temperature versus fluorescence plots. The resulting data were scored using a Python-based function (see Methods), with the top hits (score = 10) being dyes with high initial fluorescence, low background in the control (polyanion inducer; no tau) and a temperature-dependent decay. The highest-scoring hits across two biological replicates (cumulative score = 20) were then compared with the second control (monomeric tau alone; salmon-coloured arrows), yielding five hits that reacted with at least one of the WT or P301S fibrils (hashed green). This list was supplemented by manual curation of other top performing dyes (blue). b, List of the polyanions used to further diversify the fibril conformations and the concentration at which they were used.

Fig. 2. High-throughput screen results uncovers fluoroprobes that interact with either WT or P301S tau fibril polymorphs.

a, Hierarchically clustered heat map of the additive scores for screens performed using WT (top) or P301S (bottom) tau fibrils. Only the highest scoring fluoroprobes (score = 20) were taken forward for validation. The presentation of hierarchical clustering was performed using seaborn. b, Representative plots of relative fluorescence units (RFU, channel denoted per plot) versus temperature, highlighting the appearance of potentially pan-fibril dyes (L031) and potentially P301S-selective dyes (L017 and L033). FAM, fluorescein channel. Graphs are representative of the two biological replicates. Refer to ref. for the full dataset and Fig. 1b and Supplementary Table 1 for a list of inducers.

Fig. 3. EMBER analysis suggests that fluoroprobes bind in distinct chemical environments between WT and P301S tau fibrils.

a, Overview of the EMBER workflow. In the initial screen, the 27 hit dyes from the primary screen (see Fig. 2) were screened against twelve synthetic tau fibril samples in 384-well plates. In EMBER, fluorescence data are collected at a range of excitation and emission wavelengths to explore shifts in either the wavelength or intensity after dye binding to fibrils. b, Representative EMBER results, showing individual tau fibril particles composed of either WT tau (red) or P301S tau (green). Insets show the same particles at higher resolution. This example shows WT and P301S tau fibrils with inducer 1 and dye L031. c, For each particle, Bradley–Roth segmentation is performed across the full wavelength range to provide the EMBER plot. The particles are the same as in b. d, Individual EMBER plots are then concatenated for PCA, followed by quadratic discrimination to quantify polymorph selectivity. In this case, dye L031 was able to discriminate between WT and P301S tau fibrils (in the presence of inducer 1) with 88% accuracy. Boundaries pertaining to the fit discriminants are represented by the black line. PC1, principle component 1; PC2, principle component 2. e, Representative data showing that dye L031 is also able to discriminate between WT and P301S tau fibril samples created using other inducers (Ind). f, Discrimination heatmap for the full dataset, showing that a subset of the dyes can discriminate between tau fibril samples.

Fig. 4. paDSF fluoroprobes display selective recognition for diverse amyloid-forming proteins.

a, Heat map showing pDSF screening results for WT and S20G IAPP fibrils and WT α-synuclein fibrils. IAPP paDSF screens were performed and results were scored as described for tau fibrils, whereas α-synuclein fibrils were screened with an abbreviated procedure using manual scoring. b, Selected confocal micrographs (left) and EMBER profiles (right) validating fluoroprobe binding to WT and S20G IAPP fibrils and α-synuclein. c, Venn diagram summary of paDSF screening data for tau (all EMBER-validated inducer fibril types combined), IAPP (WT and S20G combined), and α-synuclein fibrils. Each amyloid-forming protein exhibits both unique and shared hits, with only one fluoroprobe hit shared between all three proteins.

Fig. 5. Validated fluoroprobe hits are chemically diverse and can detect tau fibril formation in real-time.

a, Chemical structures of the validated fluoroprobe hits, showing the two clusters (coumarins and polymethines). b, Histogram of pairwise Tanimoto similarities for all hits compared with one another (left), and with ThT (right). Tanimoto coefficients were calculated using a script created with the RDKit Python package (see ref. ). c–h, Kinetic aggregation assays. Either ThT (10 µM) or the hit dyes (50 µM, except for L031, which was 0.5 µM) were mixed with WT or P301S tau (10 µM), and aggregation was initiated with a polyanion inducer. Raw signal was normalized as a fraction of total signal to fall between 0 and 1 to facilitate comparisons. Data points are from single, representative experimental curves. c, Confirmation that P301S aggregates faster than WT, as shown using ThT and inducer 7. d, Example of a dye, L031, that has a similar profile to ThT. P301S tau with inducer 7 is shown. e, Example of a dye, MWC034, that only recognizes P301S tau, and not WT. Reactions contained inducer 7. f, Example of a dye, L016, that recognizes structures early in the process than ThT. Results from P301S and inducer 7. g, L033 recognizes relatively early structures, using WT tau and inducer 8. h, Dye MWC034 recognizes relatively late structures, using P301S tau and inducer 4.

Fig. 6. Fluoroprobe L095 recognizes tau pathology in brain tissue from mouse models and human patients with Alzheimer’s disease.

a, Representative micrographs, at two magnifications, from hippocampal sections of Tg4510 mice expressing human MAPT (tau) containing the P301L mutation under the control of the forebrain-specific Ca²⁺/calmodulin kinase II promoter. Samples are stained with an AT8 antibody for pathological tau (blue) and L095 (red), and are also shown merged. b, Two samples from a patient with Alzheimer’s disease, labelled with the AT8 antibody and L095. Note that L095 recognizes tangles (white arrows), but also labels a tau pathology that is consistent with neuropil threads (yellow arrow). c, Samples from two patients with Alzheimer’s disease, stained with antibodies for either αβ (4G8 antibody) or tau pathology (pS396 antibody). Note that L095 co-localizes with G48 at 610 nm, but with AT8 at 660 nm, allowing spectral discrimination between the two pathologies.

Extended Data Fig. 1. Analogues of the coumarin scaffold include both pan-fibril binding and potentially selective fluoroprobes.

(a) Overview of the limited medicinal chemistry campaign. Twenty four coumarins from the Max A. Weaver Collection (MWC) were screened against 26 fibril samples by paDSF using a pipeline that parallels Fig. 1, except that three biological replicates were used. After triage and removal of dyes that bound tau monomer, only 4 coumarins were identified that bind both WT and P301S fibrils. (b) Heat maps of the screening results, showing that most of the analogues failed to recognize either the WT or P301S tau fibril samples (white); whereas a subset produced reproducible signal (green). Here, the top Python score was 30 (10 for each replicate). (c) The chemical structures of the coumarins that bind to both WT and P301S fibrils, suggesting that they are pan-binders. (d) Chemical structures of coumarins with activity against at least one WT or P301S tau fibril conformer. The coumarin scaffold is highlighted in purple.