Patients fulfilling clinical and neuropathological criteria for Alzheimer’s disease (n = 11 for array tomography, 6 for ELISA) [45], or cognitively healthy control cases (n = 9 for array tomography, 6 for ELISA) were included in this study. Sample sizes were based on power calculations using effect size of 0.79 from our previous human array tomography studies looking at colocalization of clusterin and Aβ in synapses in Alzheimer’s [29] indicating that n = 6 per group is sufficient at power = 0.8 to detect a difference between colocalization of proteins at synapses between AD and controls (calculated using the WebPower package in R 4.1.2). A post-hoc power calculation using the results from the primary question in this study—whether Aβ and TMEM97 generate a FRET signal in human synapses—indicates that with our n and effect size, we have 100% power to detect a positive FRET signal (effect size 3.8 based on the % Aβ-TMEM97 FRET positive pixels within PSDs 38.4% ± 9.66 and the biological negative control—the % PSD-synaptophysin FRET positive pixels 1.75% ± 0.203). Clinical and neuropathological data were retrospectively obtained from the clinical charts available at the Edinburgh Brain Bank. Neuropathological stages were applied according to international recommendations [7, 45, 68]. Details of the human cases included are found in Table 1. Use of human tissue for post-mortem studies has been reviewed and approved by the Edinburgh Brain Bank ethics committee and the ACCORD medical research ethics committee, AMREC (ACCORD is the Academic and Clinical Central Office for Research and Development, a joint office of the University of Edinburgh and NHS Lothian, approval number 15-HV-016). The Edinburgh Brain Bank is a Medical Research Council funded facility with research ethics committee (REC) approval (16/ES/0084).

Mice expressing both human tau and the APP/PS1 transgene (APP/PS1 + Tau) were generated as previously described [51]. Briefly, two feeder lines were bred to produce experimental genotypes. The feeder lines were line 1: mice heterozygous for an APP/PS1 transgene and a CK-tTA driver transgene and homozygous for knockout of endogenous mouse tau; line 2: heterozygous for the Tg21221 human wild type tau transgene driven by CK-tTA and homozygous for knockout of endogenous mouse tau mouse tau [51]. APP/PS1 + Tau mice (n = 20) and littermate control mice not expressing APP/PS1 nor tau (n = 20) were aged to 9 months old before starting CT1812 treatment. Mice of both sexes were randomised into vehicle or control groups. Animal experiments were conducted in compliance with national and institutional guidelines including the Animals [Scientific Procedures Act] 1986 (UK), and the Council Directive 2010/63EU of the European Parliament and the Council of 22 September 2010 on the protection of animals used for scientific purposes, and had full Home Office ethical approval.

Mice were singly housed in a 12 h dark/light cycle with food and water ad libitum. Before dosing started, mice were habituated with double concentration Hartley’s strawberry jelly 4  days during which time all mice learned to eat the entire serving of jelly within 5 min. CT1812 fumarate was dissolved in dimethyl sulfoxide (DMSO) and added to cold jelly solution to make up the final volume of 0.6 mg/ml concentration before being allowed to set. Each week, a batch of Hartley’s strawberry jelly containing CT1812 or vehicle (plain triple strength jelly) was made. Mice were weighed at the beginning of each week to determine the weight of jelly to be given for that week, and were dosed daily for one month with jelly containing vehicle or CT1812 10 mg/kg/day (experimenters administering jelly were blind to condition). The jelly was delivered in a small petri dish on the floor of the home cage and mice were observed until all jelly was eaten to ensure the full dose was received following which the empty dish was removed.

After 28 days of treatment, mice were sacrificed by terminal anaesthesia. Blood was collected for drug levels then mice were perfused with phosphate-buffered saline (0.1 M) (PBS). Brains were removed and the cerebellum snap frozen for testing drug levels. One cerebral hemisphere (selected randomly) was drop fixed in 4% paraformaldehyde. The other hemisphere was dissected and entorhinal cortex processed for array tomography. The rest of the hemisphere was snap frozen for biochemical studies. Estimated percent receptor occupancy was calculated according to the formula (concentration/Ki)/[(concentration/Ki) + 1)], where Ki is determined by radioligand competition binding [27].

The main study combining array tomography and FRET experiments were performed on APP/PS1 + tau mice (n = 10) and control littermates (n = 8). Details are found in Table S1. Standard array tomography imaging (without FRET) was performed on APP/PS1 + tau mice (n = 9) and control littermates (n = 13) to test whether there were any drug effects on synapse density.

Fresh brain tissue samples from human and mouse cases were collected and processed as previously described [31]. Briefly, small pieces of brain tissue comprising all cortical layers were fixed in 4% paraformaldehyde and 2.5% sucrose in 20mM PBS pH 7.4 for up to 3  h. Samples were then dehydrated through ascending concentrations of cold ethanol until embedding into LR White resin (Electron Microscopy Sciences, EMS), which was allowed to polymerize overnight at > 50 °C. Tissue blocks were then stored at room temperature until used. For each case, two blocks corresponding to BA20/21 for human cases, or one from entorhinal cortex for mouse samples, were cut into 70 nm thick sections using an ultramicrotome (Leica) equipped with a Jumbo Histo Diamond Knife (Diatome, Hatfield, PA). Ribbons of at least 20 consecutive sections were collected in gelatine subbed coverslips.

Seventy  nm thick ribbons were immuno-labelled as described previously [31]. Briefly, coverslips were first incubated with Tris–glycine solution 5  min at room temperature followed by blocking of non-specific antigens with a cold-water fish blocking buffer (Sigma-Aldrich) for 30 min. Samples were then incubated for 2 h with primary antibodies, washed with Tris-buffered saline (TBS) solution and secondary antibodies applied for 30 min. After another TBS washing cycle, coverslips were mounted on microscope slides with Immu-Mount (Fisher Scientific) mounting media. For the detailed information of the primary and secondary antibodies used, please see Table S2.

Images of the same field of view of the consecutive sections were acquired using a Leica TCS8 confocal with 63 × 1.4 NA oil objective. In AD cases, images were acquired with a plaque in the field since our previous work demonstrated that synaptic Aβ accumulation is most prominent around plaques [29, 34]. Alexa fluor 488, Cy3 or Cy5 were sequentially excited with the 488, 552 or 638 laser lines and imaged in 500–550 nm, 570–634 nm or 649–710 nm spectral windows, respectively. For FRET analysis, the spectral window of the Cy5 (the acceptor, 649–710 nm) was also imaged under the excitation of Cy3 (the donor, 552 nm). This setting allowed us to record the transfer of energy from donor molecules to acceptors based on intensity (sensitized emission FRET, [23, 80], Fig. S1). Laser and detector settings were maintained through the whole study avoiding major saturation, which is only applied in figures for image visualization purposes.

Standard array tomography imaging (without FRET) was performed on APP/PS1 + tau mice (n = 9) and control littermates of mice (n = 13) to test whether there were any drug effects on synapse density. These images were acquired on a Ziess Axio Imager Z2 epifluorescence microscope with a 63 × 1.4 NA oil immersion objective and a CoolSNAP digital camera.

Images from consecutive sections were transformed into stacks using ImageJ [57, 59]. The following steps were performed using an in-house algorithm developed for array tomography image processing and analysis freely available (based on [13], available at https://github.com/Spires-Jones-Lab, Fig. S1). The consecutive images were first aligned using a rigid and affine registration. For the study of the immunoreactivity patterns, semi-automatic local threshold based on mean values was applied specifically for each channel yet common for all the included images. Importantly, only those objects detected in more than one consecutive Sect. (3D objects) were quantified, allowing us to reduce non-specific signals. The number of objects from each channel were quantified and neuropil concentration in mm3 of tissue established after removing confounding structures (i.e. blood vessels or cell bodies). to investigate the relationship between channels, colocalization was based on a minimum overlap of 10% of the area of the synaptic terminals. Finally, in Alzheimer’s cases, the effect of plaque proximity on concentration of objects in each channel or the colocalizing objects were also determined by calculating the Euclidean distance between the centroid of each object and the closest point to the plaque edge. The plaque edge was determined using a restrictive segmentation of the 6E10 channel to include only areas of contiguous staining (not including small Aβ positive puncta in the halo which we previously described [34]. Objects were then binned in 10 μm distances from the plaque edge.

For FRET analysis, donor-only (Cy3) and acceptor-only (Cy5) samples were imaged in each imaging session to calculate the donor-emission crosstalk with the acceptor emission (beta parameter) and the direct excitation of the acceptor by the donor excitation laser line (gamma parameter) [75, 80]. Aligned stacks of images corresponding to the acceptor emission under donor excitation line (FRET image) were first corrected for the above-mentioned parameters. Each pixel of the FRET image was corrected according to the pixel intensity of either donor-excited donor-emission images or acceptor-excited acceptor-emission images Fig. S1). Using the binary masks created before corresponding to post-synaptic terminals, donor and acceptor images, the pixels where the three objects were found overlapping were studied. The percent of pixels where any FRET signal was observed were quantified, allowing us to have a qualitative measure of the occurrence of the FRET effect.

iPSC lines derived from peripheral blood mononuclear cells from participants in the Lothian Birth Cohort 1936 (LBC1936) were used for this study as previously described [33, 69, 76]. In this study we used lines EDi030, EDi034, and EDi036. Neuronal differentiation was induced with dual SMAD inhibition (10mM SB431542, [Tocris, 1614] and 1mM dorsomorphin [R&D, 3093/10]) as published previously [61]. After 12 days induction, neuroepithelial cells were passaged mechanically onto 1:100 Matrigel (Corning, 354,230) and maintained in N2B27 media (1:1 of DMEM F12 Glutamax [Thermo Fisher Scientific, 10565018] and Neurobasal media [Thermo Fisher Scientific, 12348017], 1X N-2 (Thermo Fisher Scientific, 17,502–048], 1X B-27 [Thermo Fisher Scientific, 17,504–044], 1mM L-Glutamine [Thermo Fisher Scientific, 25,030–024], 5mg/mL insulin [Merck, I9278-5ML], 100mM 2-mercaptoethanol [Thermo Fisher Scientific, 31350010] 100mM non-essential amino acids [Thermo Fisher Scientific, 11,140–050]), and 1X anti-biotic/anti-mycotic [Thermo Fisher Scientific, 15240062]). Neural precursor cells were passaged with accutase (Thermo Fisher Scientific, A11105-01) at day 20, and day 25. A final passage was performed at day 30, with cells plated onto poly-L-ornithine (Merck, P4957) treated glass cover slips coated with 1:100 Matrigel, 10mg/mL laminin (Merck, L2020-1MG), and 10mg/mL fibronectin (Merck, F2006). Between days 35–49 maturing neurons N2B27 was supplemented with 10mM forskolin (Tocris, 1099). From day 50 on N2B27 was supplemented with 5ng/mL BDNF (R&D Systems, 248-BD) and 5ng/mL GDNF (R&D Systems, 212-GD).

Generation of brain homogenate from Alzheimer’s patients to challenge iPSC neurons was conducted according to a published protocol [26] with modifications. Human brain tissue was homogenised with a Dounce homogeniser and placed in a low protein binding 15mL tube (Thermo Fisher Scientific, 30,122,216) containing 10 mL 1X artificial CSF (pH 7.4) supplemented with 1 × cOmplete mini EDTA-free protease inhibitor cocktail tablet (Roche, 11,836,170,001) per 10 mL, per 2 g of tissue. The solution was placed on a roller for 30 min to extract soluble proteins, then centrifuged at 2000 RCF for 10 min to remove large, insoluble debris. The supernatant was transferred to ultracentrifuge tubes (Beckman, 355,647) and then centrifuged at 200,000 RCF for 110 min. The resulting supernatant, a homogenate fraction containing soluble Aβ forms, was then transferred to a Slide-A-Lyser G2, 2K MWCO 15mL dialysis cassette (Thermo Fisher Scientific, 87,719) and dialysis was conducted in 1X aCSF with magnetic stirring for three days at 4 °C to remove salts from the homogenate. During this time, the 1X aCSF was exchanged every 24 h. Dialysed brain homogenate was divided into two equal portions in low protein binding 15 mL tubes. Protein A Agarose (PrA) beads (Thermo 20,334) were washed three times in 1X aCSF. 30uL of Washed beads were then added per 1mL of homogenate. To create Aβ− treatment samples, Aβ was immunodepleted by adding 20 µL 4G8 antibody (Biolegend, 800,711) per 1 mL of homogenate and 20 µL 6E10 antibody (Biolegend, 803,001) per 1 mL of homogenate. To create Aβ + treatment samples, homogenate was ‘mock-immunodepleted’ with isotype control antibodies to non-human antigens by adding 20 µL of GFP (DSHB, DSHB-GFP-12A6) and GFP (DSHB, N86/38) antibody per 1ml of homogenate. Concentration of Aβ42was determined by ELISA (WAKO 4987481457102). Homogenate was then incubated for 24 h on a rocker, during which time the Aβ antibody complexes bind to the PrA beads in the immunodepleted portion. After 24 h incubation, homogenate was centrifuged at 2500 RCF for 5 min to remove the beads, and the supernatant was collected. The process of adding beads and antibody/serum to homogenate was repeated twice more. After the third centrifugation step, PrA beads alone were added to both Aβ + and Aβ− homogenate, incubated for 2 hours on a rocker, and then centrifuged at 2500 RCF for 5 min to clear any remaining antibody. Finally, homogenate from each portion was aliquoted at 0.5 mL into 1.5 mL low protein binding Eppendorf tubes (Thermo Fisher Scientific, 0030108116) and stored at – 80 °C. Concentration of Ab1-42 in Aβ + and Aβ− homogenate was quantified by sandwich ELISA (WAKO), according to manufacturer instructions using a ClarioSTAR spectrophotometer (BMG Labtech).

To determine whether Aβ treatments induce cell death, Click-iT™ Plus TUNEL Assay Kits for In Situ Apoptosis Detection (Thermo Fisher Scientific, C10617) was used to detect apoptotic cells according to the manufacturer’s protocols. Samples were fixed with 4% formalin (Polysciences, 04018–1), permeabilized with 0.3% Triton-X-100 in 1X PBS for 20 min at room temperature, incubated with TUNEL reaction buffer for 10 min at 37 °C, incubated with TUNEL reaction mixture for 1 h at 37 °C, blocked with 3% bovine serum albumin, and incubated with TUNEL reaction cocktail for 30 min at 37 °C. Immunocytochemisty was then performed for co-staining. All incubations were conducted in the dark.

Neurons from three iPSC donors were grown to approximately day 200 post-induction in 24 well plates. Cells were treated with media, Aβ + homogenate, or Aβ− homogenate diluted 1:4 in media for 24 h followed by addition of CT1812 (10 mM) or DMSO (Merck, D2438-50ML) vehicle treatment for a further 24 h. The final concentration of Aβ in the Aβ + treatment condition was 90 pM and in the Aβ− condition it was 8 pM. Cells were pre-incubated with Aβ + homogenate for 24 h before CT1812 treatment to model the human treatment condition in which people have Aβ accumulation before treatment with CT1812 begins. RNA was harvested from four pooled wells per treatment condition using trizol-chloroform extraction. Remaining coverslips were fixed for immunocytochemistry (ICC) as below. Each experiment was repeated with three different differentiations of each of the three lines.

Cells for ICC were fixed with 4% formalin (Polysciences, cat.04018–1) for 15 min, Washed thrice in 1X phosphate buffered saline (PBS). Fixed cells were permeabilized and blocked with in 1X PBS with 0.3% Triton-X and 3% bovine serum albumin (permeabilising block solution) for 30 min. Coverslips were incubated overnight at 4 °C with primary antibodies TMEM97, homer1, MAP2, GFAP, and Tuj1. Cells were washed with 1X PBS, and incubated in secondary antibodies diluted 1:500 in permeabilising block solution for 1 h in the dark. For the detailed information of the primary and secondary antibodies used, please see Table S2. Cells were incubated with 1:10,000 DAPI in 1X PBS with 0.3% Triton-X for 10 min, and washed 2 × in 1X PBS. Coverslips were mounted on slides (VWR, 631–0847) with mounting media (Merck, cat.345789-20ML) and imaged on a Leica TCS confocal microscope with an oil immersion 63 × objective.

For calcium imaging, cells were incubated for 7 days from approximately day 190 post-differentiation in GCamP6s AAV (pAAV.Syn.GCaMP6s.WPRE.SV40, Addgene 100843-AAV1). Images were acquired on an Leica DMI6000B inverted fluorescence microscope (20X objective, 2 frames per second). For acute treatments, cells were imaged at baseline followed by 2.5 min of treatment with Aβ + homogenate, Aβ− homogenate, 10 μM CT1812 or DMSO vehicle followed by repeat imaging of the same sites. For 24 h treatments, cells were incubated in Aβ + homogenate, Aβ− homogenate, or media for 24 h followed by baseline imaging, 2.5 min treatment with 10 μM CT1812 or DMSO vehicle, and re-imaging of the same sites.

RNA sequencing was performed on total RNA samples using TruSeq stranded mRNA-seq library preparation along with next-generation sequencing on NovaSeq6000 platform; sequencing was carried out by Edinburgh Genomics (Edinburgh, UK). Samples were sequenced to a depth of approximately 100 million 50-base pair, paired-end reads. The reads were mapped to the primary assembly of the human (hg38) reference genome contained in Ensembl release 106, using the STAR RNA-seq aligner, version 2.7.9a [18]. Tables of per-gene read counts were generated from the mapped reads with featureCounts, version 2.0.2 [39]. Differential gene expression was performed in R using DESeq2, version 1.30.1 [40]. Gene ontology analyses were run on the Gene Ontology online resource using their Panther online search tool for Biological Processes (http://geneontology.org/). MetaCore + MetaDrug version 22.3 build 71,000 was used to perform pathway analysis on Abeta vs. Vehicle, and Abeta + Drug vs. Abeta + Vehicle conditions (unadjusted p value < 0.05). STRING (Version 11.5) pathway analysis of Abeta + Drug vs. Abeta + Vehicle conditions (unadjusted p value < 0.05) [81].