The hiPSC fAD cell lines were derived from a patient carrying the fAD-linked A79V PSEN1 mutation (Li et al., 2016) and its gene corrected isogenic control A79V GC (Pires et al., 2016), a patient carrying the fAD-linked L150P PSEN1 mutation (Tubsuwan et al., 2016) and its gene corrected isogenic control L150P GC (Poon et al., 2016), a healthy control [K3P53, (Rasmussen et al., 2014)] and a CRISPR/Cas9 gene edited knock-in APP Swedish fAD line [BioSweden, (Frederiksen et al., 2019)]. All fAD lines have previously been published (Supplementary Table S1A). The sAD cell lines have not been previously published but have been characterized and assessed for pluripotency markers and genomic integrity (Supplementary Figure 2). The hiPSCs were cultured in Essential 8 (E8) media (A1517001, Thermo Fisher Scientific, United States) on Matrigel-coated plates (TH Gayer, 7643022), and media was replaced every day. The cells were passaged every third day for approximately two weeks, before neural induction was initiated.

Once hiPSC reached approximately 90% confluence, neural differentiation was performed according to a modified dual SMAD protocol (Shi et al., 2012). Neural induction was initiated by changing the media into neural induction media containing 50% DMEM/F12 (Thermo Fisher Scientific, 11330057), 50% advanced neurobasal medium (Thermo Fisher Scientific, 21103049), 1% N2 (Thermo Fisher Scientific, 17502048), 1% B27 without retinoic acid (Thermo Fisher Scientific, 1258010), 1% Glutamax™ (Thermo Fisher Scientific, 35050061), 1% non-essential amino acid (NEAA, Thermo Fisher Scientific, 11140-050), 0.1% Pen/Strep (Sigma-Aldrich, United States, P0781-100ML), supplemented with the inhibitors 10 μM SB431542 (SMAD inhibitor, SMS-gruppen, S1067) and 0.1 μM LDN193189 (Noggin analog, Sigma-Aldrich, SML0559). The cells were maintained in induction media for 12 days with daily media change. On day 12, a uniform neuroepithelial sheet appeared, and the neural progenitor cells (NPCs) were passaged with Accutase (Thermo Fisher Scientific, A1110501) into neural expansion media containing growth factors 10 ng/ml FGF2 (ProSpec, CYT-557) and 10 ng/ml EGF (ProSpec, CYT-217) instead of the inhibitors. NPCs were expanded and banked. Following expansion, NPCs were plated onto Poly-L-Ornithine (PLO, Sigma-Aldrich, P4957)/laminin (Sigma-Aldrich, L2020-1 mg) coated dishes with a seeding density of 50,000 cells/cm², and terminal neural differentiation was performed in neural maturation media, supplemented with 50 μM db-cAMP (Sigma Aldrich, D0627-100 mg), 200 μM Ascorbic acid (Sigma Aldrich, A4403-100MG), 20 ng/ml BDNF (ProSpec, CYT-207) and 10 ng/ml GDNF (ProSpec, CYT-305). The maturation process was carried out for 5 weeks for MitoTracker™ and Golgi ICC analysis and 7 weeks for assessment of Aβ secretion and Tau phosphorylation as well as MitoTracker™, Golgi and synaptic evaluation, with partial media change every third day, before the neurons were fixed or harvested for further analyses.

Secretion of Aβ peptides; Aβ38, Aβ40, and Aβ42, was assessed using the V-PLEX Plus Aβ Peptide Panel 1 Kit (MSD, 6E10), and the MESO QUICKPLEX SQ 120 Imager (MSD), connected to the DISCOVERY WORKBENCH 4.0 software, according to the manufacture’s protocol. The data was normalized to the RNA concentration of the individual samples, and data was obtained from three independent experiments.

For immunocytochemistry (ICC), neurons were plated and cultured on PLO/laminin-coated, double acid-treated coverslips for 5 weeks for MitoTracker™ and Golgi assessment and 7 weeks for neural verification, MitoTracker™, Golgi, and synaptic evaluation, fixed in 4% Paraformaldehyde (PFA) for 20 min at room temperature. Following, neurons were washed 3 × 5 min in PBS (Sigma, D8537). Neurons were then permeabilized with 0.2% Triton-X-100 solution for 20 min at room temperature, then blocked with 3% Bovine Serum Albumin (BSA, Sigma) for 30 min at room temperature. Neurons were incubated with primary antibodies (Supplementary Table S1B), diluted in 3% BSA, overnight at 4°C, washed with PBS 3 × 5 min at room temperature, and incubated with secondary antibodies (Supplementary Table S1B) for 1 h in the dark at room temperature. Another 3 × 5 min wash with PBS was performed, followed by DNA labeling with DAPI, diluted in PBS, for 7 min dark at room temperature. The neurons were then washed 3 × 5 min in PBS, before coverslips were mounted onto slides in DAKO fluorescence mounting solution and analyzed with the Leica confocal TCS SPE Microsystem implementing LAS × software. Obtained images were then processed in Fiji ImageJ 2.0.0-rc-65/1.51s. Data was obtained from three independent experiments.

After 5 days of maturation, neurons were fixed in 4% PFA and ICC was performed with Beta-III-Tubulin (Tuj1, Millipore, MAB1637). Neurite length was measured and analyzed using Neurite Tracer in Fiji ImageJ (Pool et al., 2008), from three independent experiments.

Neurons were plated on coverslips, cultured for seven weeks, then incubated with 50 nm MitoTracker™ Red CMXRos (Invitrogen/Molecular Probes, M7512) in DMEM/F12 for 15 min at 37°C. Following, the neurons were fixed in 4% PFA, permeabilized in 0.2% Triton X-100 in PBS, incubated with DAPI for DNA labeling, washed and then mounted onto slides in mounting media. Neurons from three independent experiments were imaged by confocal microscopy and analyzed in Fiji ImageJ. For detailed description see section: Immunocytochemistry and Confocal Microscopy.

Neurons were cultured on coverslips, and after 7 weeks of maturation fixed with 3% Glutaraldehyde (Merck, 1042390250) in 0.1 M Na-phosphate buffer with pH 7.2 at 4°C for 1 h. The neurons were then embedded in 4% agar and cut into 1–2 mm³ blocks under a stereomicroscope, then washed with 0.1 Na-phosphate buffer, followed by post-fixation in 1% osmium tetroxide (EMS) in 0.1% Na-phosphate buffer for 1 h at room temperature. Washing with MilliQ water was performed, followed by a stepwise dehydration in ethanol with increasing concentration. Propylene oxide (Merck) was used as an intermediate allowing for infiltration with Epon (TAAB, T031). The following day, neurons were embedded in Epon, and cured at 60°C for 48 h. Semi thin sections (2 μm) were cut on an ultramicrotome with glass knives (Leica Ultracut, Leica Microsystems, Wetzlar, Germany), then contrasted with 1% Toluidine blue (Millipore, 1159300025) in 1% Borax (LabChem, LC117101). Ultra-thin sections (50–70 nm) were cut on the microtome with a diamond knife (Jumdi, 2 mm), and the sections were collected onto grids, and then stained with 2% uranyl acetate (PolyScience) and lead citrate (Reynolds, 1963). Imaging and analysis were performed “blinded” using a Philips CM100 transmission electron microscope with a Morada digital camera equipment and iTEM software system (Olympus).

Morphometry was used for quantitative evaluation of mitochondria, and five grids from each cell line were used for the analysis. A total of 15 spots were randomly chosen at low magnification and positions were stored according to their X and Y coordinates using FEI/Philips CompuStage. Images of each spot were captured using high magnification (19,000X), resulting in 75 images per cell line, and a total of 300 images. Three mitochondria categories were established based on the qualitative evaluation: Normal mitochondria with well-defined cristae throughout the organelle, cristaless mitochondria where these structures were lacking, and an additional category with mitochondria not fitting either of the two former categories (undefined). Mitochondria morphology and size were evaluated, and Fiji ImageJ was used for analysis by application of a grid to each image. Intersections were counted for each mitochondria category, as well as the cytoplasm and the number of mitochondria to calculate the relative mitochondria/cytoplasm ratio and the relative individual mitochondria area. Data was obtained from three independent experiments.

Neurons were cultured for 7 weeks, washed with PBS, then lysed in M-PER mammalian protein extraction reagent (Thermo Fisher Scientific, 78501) with cOmplete™ Protease Inhibitor (Roche, 11873580001) and PhosSTOP™ phosphatase inhibitor (Roche, 04906845001). Protein content was determined using the Bradford Assay (Sigma, B6916). Separation of 10 μg protein was performed by NuPAGE™ Novex™ 4–12% Bis-Tris mini gel (Thermo Fisher Scientific, NP0335BOX), and immunoblotted with primary antibodies (Supplementary Table S1C) overnight at 4°C, followed by secondary antibody incubation (Supplementary Table S1C). The immunoblots were developed using Odyssey^R Fc Imaging System (LI-COR) and analyzed using Image Studio version 5.2.5 and Excel. Expression levels of target proteins were normalized against housekeeping proteins (GAPDH or β-actin), and data was obtained from three independent experiments.

To evaluate the implication of Aβ on the neurons, both the isogenic controls (A79V GC and L150P GC) and healthy control (K3P53) were treated with 2.5 μM Aβ peptide 1–42 (Sigma, PP69-0.25MG), incubated at 37°C for 24 h, then fixed and harvested for ICC and transmission electron microscopy (TEM). The analysis was performed in three individual experiments.

RNA was extracted from L150P and L150P GC as well as A79V and A79V GC neurons after 7 weeks of neural maturation, using the RNeasy^® Plus Mini Kit (Qiagen, The Netherlands, 74134), according to the manufacturer’s protocol, and the RNA quality was assessed using an Agilent 2100 Bioanalyzer system with RNA 6000 nano chip and reagents. Library preparation and sequencing (DNBseq, 2 × 100 nt, stranded paired-end) was done by an external provider for five independent replicates of fAD and isogenic control neurons (Beijing Genomics Institute, BGI, China).

The computational analysis of the RNA sequencing data is described in detail in a separate bioinformatics publication (Corsi et al., 2022b). Briefly, reads were pre-processed with Cutadapt v1.18 (Martin, 2011) to trim low quality 3’ ends (min. Phred = 30), remove residual adapters, and filter out resulting reads shorter than 20 nt (pair_filter = “any”). Sequences matching to rRNAs in SILVA v119.1 (Quast et al., 2013) were removed with BBDuk v38.22.¹ Genes were quantified from the reads pairs aligned with STAR v.2.6.1d (Dobin et al., 2013) on the human genome (hg38) using featureCounts (subread v.1.6.3, min. overlap 90 nt, min. fraction overlapping nt 90%, both aligned reads pairs non-chimeric) (Liao et al., 2014) provided with an extended set of annotations obtained by merging the GENCODE (Frankish et al., 2019) and the FANTOM-CAT gene models (Hon et al., 2017). Differential expression analysis was carried out using DESeq2 v1.22.2 (Love et al., 2014) and genes with Benjamini and Hochberg (1995) adjusted Wald test p-value < 0.05, absolute log2 fold change >1, and mean of normalized counts >10 were considered significant. The workflow includes RseQC v.4.0.0 (Wang et al., 2012) for the confirmation of the library strandness and the verification of the editing status at on-target and potential off-target sites with CRISPRroots v.1.2 (Corsi et al., 2022a). Potential off-targets identified by CRISPRroots as high-risk were validate by sequencing, and no off-target edit was found (Corsi et al., 2022b).

RNA from L150P and L150P GC, A79V and A79V GC as well as BioSweden and K3P53 hiPSC derived neurons after seven weeks of neural maturation was extracted using the RNeasy^® Plus Mini Kit (Qiagen, 74134). cDNA was synthesized from 1 μg of total RNA, according to Promega ImProm-II™ Reverse Transcription System (Promega, United States, A3800). RNA was mixed with 1:3 oligoT: random primer (0.5 μg/μl), heated at 70°C for 5 min, then immediately put on ice for 5 min. Reverse transcription mix containing ImProm II buffer, dNTP mix (10 mm), Nuclease-free water, RNasin Ribonuclease inhibitor (40 u/μl), MgCl₂ (2.5 mm) and ImProm II was added, following 5 min incubation at room temperature, then 1 h at 42°C. The enzyme was inactivated at 70°C for 15 min, and appropriate dilutions (1:5) of cDNA were prepared for further qPCR.

For qPCR analysis, nuclease-free water, QuantiFast SYBRGreen 2x (Qiagen), forward- and reverse primers (10 μM, Supplementary Table S1E) were mixed and added to each well of a qPCR plate, with a total of 8 μL per reaction and 2 μL of diluted cDNA was added. The qPCR was performed in triplicates, and a non-template control without cDNA was included for each target gene. The analysis was run on a LightCycler^® 480 real-time PCR system (Roche, Switzerland) with a total of 40 cycles, and the data was collected and processed using the Design and Analysis software 2.6.0. The data were normalized to GAPDH gene, and the analysis was performed in triplicates.

For proteomic assessment, cell pellets from L150P and L150P GC as well as A79V and A79V GC neurons were collected using ice-cold PBS and a cell scraper. Following, proteomic assessment was performed by mass spectrometry, according to a previously published protocol (Bogetofte et al., 2019). Data was obtained from three independent replicates.

From L150P and L150P GC, A79V and A79V GC as well as BioSweden, K3P53 and Aβ-treated K3P53 hiPSC derived neurons, total cell lysate protein N- and O-glycans were released and analyzed by C18 nanoflow liquid chromatography (LC) coupled to mass spectrometry (MS) as described previously, with minor adaptations (de Haan et al., 2022). Cell pellets were resuspended at ∼2 × 10⁴ cells/μL in lysis buffer [50 mm Tris HCl, 100 mm NaCl and 1× cOmplete™ protease inhibitor (EDTA-free)] and loaded on a preconditioned PVDF membrane. N-glycans were released using 2 U PNGase F in 30 μL water, eluted and dried. Sialic acids were derivatized by ethyl esterification (α2,6-linked sialic acids) and subsequent ammonia amidation (α2,3-linked sialic acids) (Lageveen-Kammeijer et al., 2019) and purified by hydrophilic interaction liquid chromatography (HILIC) solid phase extraction (SPE) (Selman et al., 2011). Next, 50 μL 2-aminobenzamide (2-AB) reagent (500 mm 2-AB, 116 mm 2-methylpyridine borane complex (PB) in 45:45:10 metanol:water:acetic acid) was added and the samples were incubated 2.5 h at 50°C. The glycans were purified by HILIC SPE and eluted in 50 μL water. Ten microliters of the eluates were diluted in 10 μL water for MS analysis. O-glycans were released from the same samples on the PVDF membrane using 20% hydroxylamine and 20% 1,8-diazabicyclo(5.4.0)undec-7-ene (DBU) for 1 h at 37°C, and enriched by hydrazide beads, 2-AB labeled as described above, and purified by HILIC and porous graphitized carbon (PGC) SPE (de Haan et al., 2022). Samples were resolved in 20 μL water for MS analysis. For both the N- and O-glycan preparations of each sample, 2 μL was injected for nanoLC-MS/MS analysis, using a single analytical column setup. The analytical column was prepared using a PicoFrit Emitter (New Objectives, 75 μM inner diameter), packed with Reprosil-Pure-AQ C18 phase (Dr. Maisch, 1.9 μM particle size, 22–25 cm column length). The emitter was interfaced to an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) via a nanoSpray Flex ion source. Samples were eluted in an 1 h method with a gradient from 3% to 32% of solvent B in 35 min, from 32% to 100% B in the next 10 min and 100% B for the last 15 min at 200 nl/min (solvent A: 0.1% formic acid in water; solvent B: 0.1% formic acid in 80% ACN). A precursor MS scan (m/z 200–1700, positive polarity) was acquired in the Orbitrap at a nominal resolution of 120,000, followed by Orbitrap higher-energy C-trap dissociation (HCD)-MS/MS at a nominal resolution of 50,000 of the 10 most abundant precursors in the MS spectrum (charge states 1 to 4). A minimum MS signal threshold of 30,000 was used to trigger data-dependent fragmentation events. HCD was performed with an energy of 27% ± 5%, applying a 20 s dynamic exclusion window. Data analysis and structural annotation were performed as described before (de Haan et al., 2022) using the Minora Feature Detector node in Thermo Proteome Discoverer 2.2.0.388 (Thermo Fisher Scientific Inc.), GlycoWorkbench 2.1 (build 146), (Damerell et al., 2012) Skyline 21.1.0.146 (ProteoWizard) (Adams et al., 2020) and the Thermo Xcalibur Qual browser 3.0.63. MS/MS spectra were manually assigned for each MS1 feature in at least one sample. N- and O- glycans were relatively quantified separately, by total area normalization. Derived traits were calculated based on specific glycosylation features, including for N-glycans the glycan type (paucimannose, oligomannose, complex or hybrid) and complex-type fucosylation (no or core and/or antenna), sialylation (no or α2,3- and/or α2,6-linked), bisection, LacdiNAc, galactosylation, and antenarity. For the O-glycans, the relative abundance of the O-glycan types was determined (O-GlacNAc, O-GalNAc, O-fucose, O-glycose or O-mannose), and specifically for the O-GalNAc glycans the relative abundance of the different cores (Tn, 1, 2, or 3) as well as the level of sialylation per core type. All values per cell type were represented as averages and standard deviations over three technical replicates.

Neuro 2A (N2A) cells were transiently transfected with either APP Swedish- or wild type human APP (hAPP), using Lipofectamine 2000 (Thermo Fisher Scientific, 11668030) following a similar protocol as described previously (Kedia et al., 2020). Briefly, after incubation, diluted DNA and Lipofectamine were combined and incubated for 20 min at room temperature to establish DNA-Lipofectamine complexes. The complex (500 μL) was then added to the N2A cells, which were further incubated for 24-72 h at 37°C. Media was replaced after 6 h. Following transfection, the N2A cells were fixed and immunocytochemical labeling for GM130 and γ-adaptin (Supplementary Table S1B) was performed. The N2A cells were mounted with Prolong (Molecular Probes, cat. no. MAN0010261) for Airyscan super-resolution microscopy. Airyscan super-resolution microscopy was performed as described previously (Kedia et al., 2021). Airyscan images were obtained on Zeiss LSM 880 equipped with 32 array detectors for acquisition of super-resolution images. For image acquisition, 405, 488, and, 633 nm lasers were used. The illumination parameters like intensities, digital and analogue gain of the detectors, sampling of the images, emission window for each fluorescent channel and their corresponding pinhole sizes were maintained constant across acquisition paradigm. The raw images acquired using Airyscan mode were processed using Zeiss ZEN 3.4 (Blue) software to generate final super-resolution images. The reconstruction parameters were also kept constant throughout the samples. The morphological and biophysical traits of cis-/trans-Golgi were quantified on super-resolution images through MetaMorph software (Molecular Devices).

To further validate our findings, male 5xFAD transgenic mice (N = 5) and male wild-type mice (N = 4) were sacrificed, and brains were dissected. The cortex and hippocampus were fixed in 3% Glutaraldehyde and embedded in Epon for TEM evaluation. For detailed description of TEM preparation and imaging, see the section TEM.

Statistical analyses were performed using GraphPad Prism (Version 9.2.0) with default options, and statistical significance was determined using a Student’s t test, multiple t test or two-way ANOVA with correction for multiple comparison. Data is presented as mean ± standard error of the mean (SEM) for all experiments with statistical significance *p < 0.05, ^**p < 0.01, ^***p < 0.001, and ^****p < 0.0001. Gene differential expression was performed using DESeq2 (Version 1.22.2) (Love et al., 2014) using the Benjamini-Hochberg method to adjust Wald test p-values in multiple testing.

Neurons were differentiated from hiPSCs derived from two patients carrying fAD-linked PSEN1 mutations and their respective isogenic controls corrected via CRISPR/Cas9 gene editing: L150P and L150P GC (gene corrected) (Poon et al., 2016; Tubsuwan et al., 2016), A79V and A79V GC (Li et al., 2016; Pires et al., 2016), as well as a healthy control- (K3P53) (Rasmussen et al., 2014) and a CRISPR/Cas9 knock-in hiPSC line carrying the Swedish APP mutation (BioSweden) (Frederiksen et al., 2019). fAD- and control-hiPSCs were successfully differentiated into cortical, glutamatergic forebrain neurons (Shi et al., 2012) (Figure 1A). Neurite outgrowth analysis revealed a non-significant tendency toward reduced neurite length in the AD lines compared to their respective isogenic controls after 1 week of differentiation (Figures 1B, C). Neural outgrowth capacity was initially lower for A79V and A79V GC, potentially indicating innate differences between the two hiPSC lines during the early stages of differentiation. However, all hiPSC derived neurons displayed comparable differentiation potential and maturity levels at week seven. This was assessed via expression of microtubule associated protein 2 (MAP2) and Tau (Figure 1D). Astrocyte content in our cultures ranged from 5% to 10%, detected by Glial Fibrillary Acidic Protein (GFAP) expression (Figure 1E). Most of the cultured neurons were glutamatergic neurons (VGLUT) with few GABAergic neurons (VGAT) (Figure 1E and Supplementary Figure 1).

Western blot (WB) analyses revealed significantly increased levels of phosphorylated Tau (p-Tau) isoforms pThr181 and pSer422 in the L150P neurons. A similar trend for pSer199/S202 was observed, but no difference was detected for pSer396. A79V demonstrated similar tendencies but showed no difference in pSer422 levels (Figure 1F). Mesoscale V-PLEX assessment of Aβ peptides 1–40 and 1–42 revealed a significant increase in Aβ42/40 ratio in L150P neurons, likely due to a substantial decrease in Aβ40. This is in line with other in vitro studies showing that fAD-linked mutations in PSEN1 can lead to reduced levels of individual Aβ peptides, with particularly low or even undetectable levels of Aβ40 (Sun et al., 2017). The increase in Aβ42/40 ratio was not significant in the A79V neurons; however, up-regulated secretion of both Aβ40 and Aβ42 was detected (Figure 1G), which aligns with other studies in different model systems of the A79V mutation (Kumar-Singh et al., 2006; Kauwe et al., 2007; Sun et al., 2017). Recapitulation of characteristic Aβ and Tau pathologies in our hiPSC derived cell models validates their relevance as in vitro AD models. Moreover, increased Aβ42/40 ratio was detected in BioSweden neurons, indicating that other fAD-linked mutations result in similar neuronal phenotypes (Figure 1H).

Next, we assessed the mitochondria ultrastructure in fAD and control neurons using TEM. Here, we demonstrated abnormal cristae-lacking organelle morphology (Figure 2A) and a significant increase of relative cristae-less mitochondria to cytoplasm ratio in both L150P and A79V neurons (Figure 2B). The relative individual mitochondria area was unaltered, indicating increased number of cristae-less organelles, and not mitochondria size (Figure 2C). TEM analysis revealed a perinuclear mitochondria accumulation in fAD neurons. MitoTracker™ confirmed a disrupted mitochondria distribution with a significant reduction in mean area of distribution in both L150P and A79V (Figures 2D, E). Rescue of mitochondria phenotypes in our isogenic controls shows that our findings can be attributed to the PSEN1 mutations. The same distribution pattern was demonstrated in BioSweden neurons, indicating that mitochondria defects are equally relevant for both PSEN1 and APP fAD-linked mutations (Figure 2F). Aberrant mitochondria morphology suggests impaired mitochondria function since the ATP production via electron transport chain takes place within the inner membrane, which depends on an intact membrane and correct cristae folding (Afzal et al., 2021). Abnormalities in energy production and metabolism were verified by computational analysis of RNA sequencing, which revealed differentially expressed genes related to mitochondria and oxidative stress, further validated by qPCR (Figures 2G, H), and supported by proteomics (Supplementary Table S2C). CKMT1, central to metabolism and energy transduction, was significantly increased in both L150P and A79V neurons (Figure 2G and Supplementary Table S2C). Such upregulation is commonly seen in conditions with compromised energy state (Schlattner et al., 2006), and might reflect a compensatory mechanism in response to impaired mitochondria function. Moreover, downregulation of SCARA3, normally protecting against reactive oxygen species (ROS), was detected in both L150P and A79V neurons (Figure 2G and Supplementary Table 2SC), indicating an impaired ability to handle oxidative stress and causing excessive neurotoxic effects (Peng et al., 2021). Combined, these findings validate mitochondria dysfunction early-on in AD pathogenesis in our seven-week fAD neurons, which are supported by previously described mitochondria phenotypes in the 5 × FAD mouse model (Andersen et al., 2021).

Transmission electron microscopy (TEM) assessment further revealed a reduced number of synapses, which were identified by electron dense regions at both the pre- and post-synaptic terminal, as well as synaptic vesicles in L150P and A79V neurons (Figure 2I). Together with reduced synaptophysin expression identified via ICC analysis (Figures 2J, M), these findings indicate reduced synaptic density. Interestingly, transcriptome analysis revealed mainly upregulation of synapse-related genes (Figure 2K), validated by qPCR (Figure 2L) and proteomics analysis (Supplementary Table S2C), suggesting altered synaptic function. Such upregulation has been demonstrated in mild cognitive impairment (MCI) and early AD, potentially acting as a compensatory mechanism for rebalancing synaptic transmission in early-stage AD (Berchtold et al., 2014). The fact that our in vitro cell models display both mitochondria and synaptic deficits, in addition to Aβ and Tau pathology, further stresses their potential in AD modeling, to better understand early cell-type specific disease perturbations.

Since Golgi fragmentation has been identified in neurodegenerative disorders (Martínez-Menárguez et al., 2019), we next investigated Golgi organization and function in L150P and A79V neurons. TEM analysis revealed an abnormal Golgi morphology with dilated cisternae and shortened Golgi stacks (Figure 3A). ICC analysis confirmed a disorganized Golgi pattern in L150P and A79V, compared to a centered Golgi at one pole in the perinuclear region in the isogenic controls, both for the cis- (GM130) (Figure 3B), and trans-Golgi (TGN46) (Figure 3C), indicating a total fragmentation, with an increase in cis- and trans-Golgi surface area (Figure 3D). This was particularly prominent for the trans-Golgi, with significant changes in both mutations at seven weeks. Our findings ware mirrored in BioSweden neurons (Figure 3E), demonstrating Golgi fragmentation as a characteristic phenotype common for both APP and PSEN1 fAD mutations. Moreover, Golgi abnormalities were consistent and present in other model systems. Multicolor Airyscan super-resolution imaging of N2A cells, transfected with hAPP Swedish, confirmed cis- (GM130) and trans-Golgi (γ-adaptin) fragmentation, with significant increase in average intensity and organelle volume (Figures 3F–H). Moreover, Golgi fragmentation was present in two independent male and female sAD hiPSC derived neuronal models (Figure 3I). Likewise, Golgi fragmentation was demonstrated in the 5 × FAD transgenic mouse model (Figure 3J). We thus validated our findings in multiple fAD and sAD in vitro, as well as fAD in vivo models, indicating Golgi fragmentation as a universal early event in AD pathology.

Differential expression of Golgi-related genes in L150P and A79V neurons (Figure 3K), validated by qPCR (Figure 3L), further supported Golgi abnormalities. This included GM130 and GORASP, involved in organization, assembly, and vesicle fusion, and FYCO1, containing a Golgi dynamics domain facilitating interactions and sorting. Additional differentially expressed genes related to glycosylation, which could indicate impaired Golgi function, include GALNTs, MAN1C1, and MGATs (Supplementary Table S2C).

To investigate the initial triggers and timeline of AD pathogenesis we assessed our hiPSC derived neurons at both 5 and 7 weeks during the terminal differentiation phase. At week five, the mitochondria distribution remained intact (Figure 4A), whilst Golgi fragmentation was clearly present and particularly prominent in A79V (Figure 4B). Interestingly, the cis-Golgi surface area was significantly increased in five-week A79V neurons, opposed to the seven-week neurons, suggesting a compensatory mechanism that could contribute to maintain Golgi structure and function. Importantly, our results suggest that Golgi fragmentation is one of the earliest neuronal AD events, preceding mitochondria deficits.

Aβ-mediated phosphorylation of the Golgi-matrix protein GRASP65 has been indicated as a trigger of Golgi fragmentation (Joshi et al., 2014). Additionally, Golgi organization is dependent on an intact cytoskeleton, and Tau hyperphosphorylation might exacerbate the disorganization. Fragmentation can lead to a vicious cycle enhancing Aβ and Tau pathologies, ultimately promoting neurodegeneration (Joshi et al., 2014, 2015). To confirm if increased Aβ is the mechanism underlying the observed Golgi fragmentation, we treated seven-week isogenic control neurons with Aβ42 peptide for 24 h. Strikingly, this treatment induced fragmentation of both cis- and trans-Golgi, mirroring the seven-week fAD neurons (Figure 4C), clearly indicating that Aβ42 is a trigger of fragmentation. The potential contribution of Tau phosphorylation is highlighted by the fact that L150P neurons demonstrated significant Tau hyperphosphorylation and consequently displayed a more profound fragmentation of the cis-Golgi compartment.

The observed Golgi fragmentation together with changes in genetic expression suggests compromised Golgi function. Therefore, we investigated glycosylation via total glycan profiling of our hiPSC derived neurons. Overall Golgi fragmentation appeared to have little impact, indicating a robust glycosylation machinery, and no consistent changes were observed in N-glycan profile for L150P and A79V. However, a few alterations could be observed within the L150P neurons, including a tendency of reduced high-mannose levels, low antenna fucosylation, bisecting GlcNAc, and LacdiNAc levels, increased antenna galactosylation (Figure 5A), as well as a sialic acid (SA) linkage shift from 2,3 SA to 2,6 SA (Figure 5B). The N-glycan profile of the A79V, as well as the BioSweden (Supplementary Figure 3), remained unchanged. A similar trend was observed in the O-glycan profile. L150P deviated slightly, showing an altered fucosylation pattern, displayed as absent H2N2F1 (Figure 5C), similar to the changes in antenna fucose N-glycans. These are likely regulated by the same enzymes, related to terminal glycosylation. Interestingly, L150P and A79V showed a comparable tendency toward increased core 2 O-GalNAc glycans (Figure 5C), which could imply that some stages of complex glycosylation could be affected by Golgi fragmentation. Taken together, the glycan alterations observed in the L150P neurons are likely to be disease-associated, but not exclusively linked to PSEN1 mutations or the Golgi fragmentation.

The finding of only minor glycan changes in A79V compared to L150P neurons suggests that other underlying mechanisms could maintain the function of the Golgi, despite fragmentation. Sortilin-related receptor 1 (SORL1) encodes a protein directly involved in APP processing, and mediates retrograde transport of APP between the trans-Golgi network (TGN) and early endosomes (Yin et al., 2015). Both genetic and functional alterations of SORL1 has been linked to AD (Rogaeva et al., 2007; Rovelet-Lecrux et al., 2021), and single nucleotide polymorphisms (SNPs) in SORL1 can confer increased risk of developing AD. Wild-type SORL1 is suggested to have a protective effect (Rogaeva et al., 2007; Rovelet-Lecrux et al., 2021). Therefore, we assessed the genetic variants of SORL1 in L150P and A79V hiPSC, via Sanger sequencing (Figure 5D) of the genomic locations within the SORL1 protein-coding region, harboring four SNP sites correlated to AD risk: rs2282649, rs2070045, rs12285364, and rs2298813. Strikingly, we identified the presence of two of these SNP variants in L150P (rs2282649 heterozygote and rs2070045 homozygote), which might contribute to the functional Golgi alteration, causing altered protein trafficking within the trans-Golgi. No SORL1-linked SNPs were identified in the A79V (Figure 5E), wild-type K3P53 or knock-in BioSweden hiPSCs. Previous studies have reported decreased SORL1 expression in AD (Dodson et al., 2008). Instead, A79V and BioSweden neurons showed upregulated SORL1 expression (Figure 5F). This suggests a beneficial compensatory mechanism that retain Golgi function, which could be characteristic for early stages of AD.