Modeling Late-Onset Alzheimer’s Disease Neuropathology via Direct Neuronal Reprogramming

Abstract

Late-onset Alzheimer’s disease (LOAD) is the most common form of AD. However, modeling sporadic LOAD, that endogenously captures hallmark neuronal pathologies, such as amyloid-β (Aβ) deposition, tau tangles, and neuronal loss, remains an unmet need. Here, we demonstrate that neurons generated by microRNA-based direct reprogramming of fibroblasts from individuals affected by autosomal dominant AD (ADAD) and LOAD in a three-dimensional (3D) environment, effectively recapitulate key neuropathological features of AD. Reprogrammed LOAD neurons exhibit Aβ-dependent neurodegeneration, and treatment with β- or γ-secretase inhibitors before (but not subsequent to) Aβ deposit formation mitigated neuronal death. Moreover, inhibiting age-associated retrotransposable elements (RTEs) in LOAD neurons reduced both Aβ deposition and neurodegeneration. Our study underscores the efficacy of modeling late-onset neuropathology of LOAD through high-efficiency microRNA-based neuronal reprogramming.

Main text

AD is the leading cause of dementia characterized by key neuronal hallmarks: extracellular deposition of Aβ plaques, formation of intracellular neurofibrillary tangles, and neuronal loss (1, 2). Modeling these pathological hallmarks has largely relied on transgenic cellular and animal models harboring genetic mutations linked to ADAD (3–5). In contrast, sporadic LOAD modeling remains underexplored although LOAD represents over 95% of AD cases, and a LOAD patient-specific model system that allows the investigation of adult-onset neuropathological features remains to be established.

Recent advances in stem cell and reprogramming technologies have enabled the generation of various human cell types, including neurons generated from induced pluripotent stem cells (iPSCs) (6). However, induction of pluripotency in iPSCs resets the age signatures of the donor cells to a fetal stage (7–9). This imposes challenges for modeling LOAD accounting for human age as a key contributing risk factor (10, 11). For instance, iPSC-derived neurons from individuals with AD exhibit increased production of Aβ42, active GSK-3β, phosphorylated tau, endosomal abnormalities, and oxidative stress (12–17), but late-stage neuropathological features, such as formation of insoluble tau and neurodegeneration have not been reported. In addition, ultra-structural analyses of tau filaments in brains from individuals with AD have shown that tau tangles contain both 3-repeat (3R)-tau and 4-repeat (4R)-tau isoforms (18), both of which are expressed in adult human brains (18, 19). However, fetal neurons and iPSCs-derived neurons primarily express 3R-tau (20, 21), imposing challenges to mimic tau isoform without introducing mutations within MAPT as it occurs in the brain of individuals with LOAD.

Direct cellular reprogramming retains age-related signatures stored in pre-reprogrammed non-neural somatic cells such as fibroblasts, including the epigenetic age, telomere length, and gene expression changes (22, 23). Recent studies have directly induced neurons from individuals with AD through NGN2/ASCL1 overexpression. These neurons exhibited several age-associated dysregulations such as instability of mature neuronal fate, increased cellular senescence, and metabolic shifts (11, 24, 25), but no AD hallmark neuropathology was shown. Here, we deploy microRNA (miRNA)-based direct reprogramming to convert patient fibroblasts into neurons (26, 27) in three-dimensional (3D) environment. MiR-9/9* and miR-124 (miR-9/9*−124), when expressed in human adult fibroblasts, induce chromatin reconfiguration leading to sequential steps of fibroblast identity erasure and neuronal program activation with high efficiency reaching over 80% of conversion rate (28, 29). This miRNA-induced neuronal state can synergize with additional transcription factors (TFs) to generate disease-relevant neuronal subtypes that recapitulate age-dependent neuropathology (26, 28, 30, 31), including primary tauopathy and adult-onset neurodegeneration of Huntington’s disease (20, 32, 33). We first evaluated the feasibility of modeling AD neuropathology by neuronal reprogramming of fibroblasts from individuals with ADAD, and then tested whether reprogrammed neurons from individuals clinically affected with LOAD would manifest neuropathological features of AD.

3D-direct neuronal reprogramming

Fibroblast samples from 10 individuals with AD (4 ADAD and 6 LOAD) and 17 age- and sex-matched healthy control (HC) donors (Table S1) were used for reprogramming using miR-9/9*−124 and TFs NEUROD2 and MYT1L to guide the miRNA-mediated conversion towards the cortical lineage as previously described (20, 26). This miRNA+TFs approach generated a highly enriched neuronal population as previously characterized by single cell and bulk RNA-seq (20, 28, 29) and electrically active cells in 2D culture, showing inward and outward currents and multiple action potentials measured by whole-cell recording (Fig. S1A). To help retain the secreted Aβ, we designed two 3D-culture methods: i) thin gel culture, which allows for morphological analyses of individual reprogrammed neurons, and ii) high cell-density assembly as spherical structures (termed as spheroids for simplicity hereto). For 3D thin-gel culture with reprogrammed cortical neurons (3D-CNs), cells were dissociated and resuspended in 15% Matrigel at post-induction day (PID) 7, followed by a 2–3 week reprogramming period in 96 well plate (~ 0.1×10⁶ cells per well) (Fig. 1A). 3D-CNs acquired neuronal morphologies marked by neuronal markers NCAM1, MAP2, and SLC17A7 (also known as VGLUT1) (Fig. 1B and S1B–C). RNA-seq of reprogramming cells at different post-induction days (PID) showed a continuous increase in the expression of long genes, a transcriptomic feature unique to neurons (29, 34) (Fig. 1C), decreased expression of fibroblast genes, and activation of pan-neuronal markers, synaptic markers, and cortical markers starting at around PID20 (Fig. 1D). In contrast, we did not detect the expression of non-neuronal cell markers (35) in 3D-CNs, such as astrocyte, microglia, oligodendrocyte, Schwann cells, oligodendrocyte precursor cells (OPC), radial glia, stem cells, and myocytes (Fig. S2A).

Fig. 1. Direct reprogramming of fibroblasts from individuals with AD into 3D-CNs and neuronal spheroids and ADAD 3D-CNs and spheroids showed higher amount of Aβ deposits.

(A) A schematic diagram of 3D-direct reprogramming of patient fibroblasts to cortical neurons in thin gel culture. (B) Representative images of a thin gel culture (PID28) in one whole well of a 96-well plate immunostained with TUBB3. (C) A LONGO plot depicting increased long gene expression (LGE) during neuronal reprogramming in thin gel. (D) Heatmap shows downregulation of fibroblast markers and upregulation of neuronal markers during neuronal reprogramming in thin gel. Pan-neu: Pan-neuronal markers. Syn: Synaptic markers. Two replicates from HC83 line (healthy individual) were used for analysis in C and D. (E) A schematic diagram of direct reprogramming of human fibroblasts to neuronal spheroids. (F) A representative image of a neuronal spheroid at PID28 stained with TUBB3 depicting neurite extensions radiating out from the core of the spheroids. (G) Expression of long genes in PID28 HC spheroids, and their starting fibroblasts. (H) A heatmap shows gene expression of fibroblast markers and neuronal markers in PID28 neuronal spheroids and their starting fibroblasts. Two replicates from HC83 line were used in G & H. (I) Extracellular Aβ deposits detected by 82E1 Aβ antibody in thin gel at PID30. Aβ deposits with size > 1000 μm³ were used for quantification. (J) Immunofluorescence images (left) and enlarged 3D reconstruction images (right) showing extracellular Aβ deposition (6E10) in thin gel at PID30. Aβ deposits with size > 1000 μm³ were used for quantification. (K) Aβ42 amounts were measured by electrochemiluminescence assay in ADAD and HC spheroids at PID28. (L) Whole mount immunostaining of Aβ in ADAD spheroids and HC spheroids at PID28. Aβ deposits with size > 1000 μm³ were used for quantification. (M) Representative images of spheroid sections (PID28) and sections of 5XFAD mouse brain cortex region (3-month-old) showing the extracellular Aβ deposition. For quantification in I, J, K, and L, n = 4 ADAD and 4 HC individuals, and 2–3 thin gels or spheroids per individual line were used. * p < 0.05 and ** p < 0.01 were calculated by unpaired t-test. Scale bar: 250 μm in B and F, 50 μm in I, J, and M, and 500 μm in L.

To produce neuronal spheroids, cells were dissociated and condensed to a pellet at PID7, and placed into a transwell insert to self-organize and form a high-cell density (~ 0.5×10⁶ cells) sphere, followed by 2–3 weeks of reprogramming (Fig. 1E and Fig. S1D). Spheroids at PID28 showed neurite projections radiating out from the core (Fig. 1F) and gene expression patterns similar to 3D-CNs at PID30 including downregulation of fibroblast marker genes and increased expression of long genes and neuronal markers as assessed by RNA-seq (Fig. 1G and 1H). The expression for glial or other non-neuronal markers was not detectable in spheroids derived from neuronal reprogramming (Fig. S2B). The expression of neuronal genes such as MAP2, MAPT (also known as TAU), SNAP25, ACTL6B (also known as BAF53B, a marker of mature neurons (27)), and cortical markers SLC17A7, TBR1 and POU3F2 (also known as BRN2, a cortical marker), was also confirmed by qPCR (Fig. S1E). Sections of PID28 spheroids showed over 90% of cells expressed neuronal markers NCAM1, MAP2, TAU and TUBBIII and more than 80% of cells expressed NEUN, whereas less than 1% of cells expressing fibroblast marker FSP1 (Fig. S1F). In summary, our results validate the neuronal identity of cells reprogrammed under 3D conditions.

Aβ deposition in ADAD 3D-CNs and spheroids

In ADAD, genetic mutations drive AD pathogenesis (36). Therefore, we first examined neurons reprogrammed from fibroblast samples of individuals with ADAD carrying mutations in PSEN1 or APP (Table S1). To assess Aβ deposition, we performed immunostaining analyses on 3D-CNs derived from individuals with ADAD and age-/sex-matched HC individuals using the 82E1 antibody specific for the cleaved N-terminus of APP. Intriguingly, ADAD 3D-CNs from four independent individuals consistently showed increased extracellular Aβ deposits (10–50 μm in diameter) compared to HC 3D-CNs (Fig. 1I and Fig. S3A). Similar results were observed using a different Aβ antibody, 6E10: ADAD 3D-CNs showed a threefold increase in total volume of Aβ deposits and higher numbers of Aβ deposits with size of 10,000–100,000 μm³ compared to HC 3D-CNs (Fig. 1J and Fig. S3B).

We examined whether the Aβ phenotype would be further amplified in a high-cell density by forming a 3D spheroid with five times more reprogrammed neurons than the thin gel condition. The Aβ42 amounts were measured by an electrochemiluminescence assay in spheroids after 4 weeks of reprogramming. HC spheroids displayed low Aβ42 expression, whereas ADAD spheroids contained substantial expression of Aβ42 (Fig. 1K), consistent with effects of ADAD mutations on APP processing. Whole mount immunostaining the 6E10 antibody on PID22 and PID28 spheroids, followed by spheroid clearing, revealed that ADAD spheroids had an increase in total volume of Aβ deposition compared to HC spheroids at both time points (Fig. 1L and Fig. S3D). Furthermore, ADAD spheroids showed a progressive increase in Aβ deposition from PID22 to PID 28 (Fig. S3D). Similar to the sizes of Aβ deposits detected in ADAD 3D-CNs (Fig. S3B), ADAD spheroids at PID28 also displayed higher numbers of Aβ deposits with volume sizes between 10,000–100,000 μm³ (Fig. S3C). Cryosections of PID28 ADAD spheroids immunostained with 6E10 or 82E1 Aβ antibody showed that Aβ deposits were formed in the extracellular space (Fig. 1M and Fig. S3E), in higher amounts compared to HC spheroids (Fig. S3E). We also validated the 6E10 Aβ antibody in detecting the extracellular Aβ signal by examining 5XFAD mouse cortical sections, which were positive for extracellular Ab signals in the cortex (Fig. 1M, right). Additionally, transmission electron microscopy (TEM) with immunogold labeling against Aβ showed gold signals in the extracellular domain in ADAD spheroids (Fig. S3F). Lastly, to confirm that Aβ deposits in ADAD spheroids resulted from APP cleavage, we treated the spheroids with β-secretase inhibitor IV or γ-secretase inhibitor DAPT from PID16 to 28, which reduced Aβ deposition (Fig. S3G). Altogether, both ADAD 3D-CNs and spheroids contain elevated extracellular Aβ deposition.

Tau dysregulation in ADAD 3D-CNs and spheroids

In AD, tau protein becomes hyperphosphorylated, resulting in tau dissociating from microtubules and forming insoluble tau aggregates in the neurites and cell bodies (3, 37). To investigate the amount of phosphorylated tau (p-tau) in ADAD 3D-CNs relative to HC 3D-CNs, we carried out immunostaining analyses using AT8 (phosphor-Ser202/Thr205) and PHF1 (phosphor-Ser396/Ser404) p-tau antibodies. Both ADAD and HC neurites displayed p-tau signals, mirroring the tau island patterns observed in primary neurons (20, 38, 39) (Fig. 2A and Fig. S4A). However, ADAD 3D-CNs showed significantly (p < 0.05) higher signals of p-tau compared to HC 3D-CNs (Fig. 2A). ADAD 3D-CNs also had increased signals for MC1 antibody, which detects conformationally misfolded tau (40), compared to HC 3D-CNs (Fig. S4B).

Fig. 2. Tauopathy, neurodegeneration, and transcriptomic features in ADAD 3D-CNs and spheroids.

(A) Immunostaining of p-tau (AT8 antibody) in ADAD and HC 3D-CNs (PID30). (B) Immunofluorescence images of pathogenic tau by co-staining of K63-linked ubiquitin and p-tau (PHF1) in ADAD and HC 3D-CNs. Arrows point to the enlarged beaded neurite bulges. (C) Transmission EM images of healthy neurites (top) and dystrophic “beaded” neurites (bottom) from HC and ADAD 3D-CNs, respectively. (D) Live-cell FRET images for detecting seed-competent tau from HC and ADAD 3D-CNs at PID26. Mean ± SEM. (E) Two spheroids were labeled by RFP or GFP and co-cultured starting at PID7. At PID17 and PID33, multiple live images were taken and compiled by Photoshop to display the whole spheroids. Left: Two spheroids derived from two different HC individuals. Right: ADAD spheroids were co-cultured with HC spheroids. HC G: HC spheroids with GFP; HC R: HC spheroids with RFP. Mean ± SEM. (F) TUNEL staining in ADAD or HC spheroid sections. 4 confocal images were taken from different areas on each section for 3 sections covering different planes of the spheroid. (G) TUBB3 immunostaining images show neurite outgrowth from the core of HC and ADAD spheroids at PID22 and PID28. Yellow dashed lines indicate the border between the core and the neurites. (H). A volcano plot displaying differentially expressed genes (DEGs) between ADAD and HC spheroids at PID25. 1411 DEGs were identified (p < 0.05, |log2fold change| > 0.58, base mean > 1). 24 samples from 4 ADAD and 4 HC individuals were analyzed. (I) Top gene ontology (GO) terms associated with upregulated and downregulated DEGs analyzed by DAVID. For quantification in A, B, D, F, and G, n = 3 or 4 ADAD and 3 or 4 HC individuals, and 2–3 thin gels or spheroids per individual line were used. For E: n = 3 pairs and 2–3 co-cultures per pair were used. Unpaired t-test for A, B, D, and F, adjusted p-values by two-way ANOVA with Šídák’s multiple comparisons test for E, and multiple paired t-tests for G. For all data: ns = p > 0.05, * p < 0.05, ** p < 0.01, and *** p < 0.001. Scale bars: 50 μm in A and F, 25 μm in B and D, 800 nm in C, 1 mm in E, and 250 μm in G.

In human AD brains, K63-linked accumulation of ubiquitinated tau is associated with tau seeding (41). By co-immunostaining using K63-linked ubiquitin and p-tau (PHF1) antibodies, we observed that ADAD 3D-CNs contained higher amount of K63-ubiquitin in PHF1-positive neurites than HC 3D-CNs (Fig. 2B). Beading or blebbing in neurites is a marker of dystrophic neurites in the AD brain (3, 42, 43). We detected spherical, beaded neurites in ADAD 3D-CNs and spheroids that are positive for p-tau/K63-ubiquitin staining (Fig. 2B and S4C, see arrows). The presence of beaded dystrophic neurites in ADAD neurons was also confirmed by high-resolution TEM (Fig. 2C).

To examine whether directly reprogrammed ADAD neurons contain seed-competent tau, we performed a Fluorescence Resonance Energy Transfer (FRET) assay by transducing neurons with tau FRET sensors composed of tau P301S-Ruby2 and -Clover reporters into 3D-CNs. These reporter proteins aggregate and transmit FRET signals in the presence of seed-competent tau species (44, 45).

As shown in Fig. 2D, ADAD 3D-CNs exhibited significantly (p < 0.05) increased FRET signals compared to HCs, suggesting that ADAD 3D-CNs contain more seed-competent tau. The FRET signal in ADAD 3D-CNs was specifically localized to beaded dystrophic neurite regions (Fig. 2D). Methanol fixation has previously been used to remove soluble proteins, including tau (20, 46). Immunostaining using PHF1 p-tau antibody following methanol fixation indicated that ADAD CNs exhibited elevated amounts of insoluble p-tau (Fig. S4D). Moreover, using GT-38 tau antibody, which selectively labels the AD-specific tau strain (47), we found that ADAD 3D-CNs showed GT-38-labeled tau albeit in low occurrences (Fig. S4E).

In AD, tau tangles contain a mixture of 3R- and 4R-tau isoforms to adopt an AD-characteristic topology (18, 19). Thus, the endogenous expression of both 3R and 4R tau isoforms is critical when using reprogrammed human neurons for modeling tau pathology in AD. To determine if neurons reprogrammed from individuals with ADAD express both 3R and 4R tau isoforms, we performed semi-quantitative PCR as previously described (20). 3R- and 4R-tau isoforms were detected with an approximately equal ratio in both ADAD and age-matched HC 3D-CNs (Fig. S4F). Lastly, given both Ab and tau phenotypes are present in ADAD 3D-CNs, we examined whether Aβ accumulation would contribute to tau dysregulation (3, 48) in patient-derived neurons. Treating ADAD 3D-CNs with β-secretase inhibitor IV or γ-secretase inhibitor reduced p-tau (AT8) and K63-ubiquitin⁺ p-tau (PHF1) signals in ADAD 3D-CNs (Fig. S4G). In summary, our findings indicate that ADAD neurons exhibit Aβ-dependent tau pathology.

Spontaneous neurodegeneration in ADAD 3D-CNs and spheroids

AD is a neurodegenerative disorder marked by neuronal loss, especially in the hippocampus and cerebral cortex (49, 50). To assess spontaneous neurodegeneration, we used Sytox-Green assay, a general cell death indicator, in ADAD 3D-CNs. Whereas ADAD and HC cells showed minimal cell death until PID20, substantial cell death was detected in ADAD 3D-CNs at PID30 and PID35, compared to HCs (Fig. S5A). These time points correspond to when neuronal identity is established during the conversion process (Fig. 1C, 1D and Fig. S1B). We also assessed neuronal death in ADAD spheroids. First, ADAD spheroids (labeled by turboRFP) cultured adjacent to HC spheroids (labeled by eGFP) shrank by 50% in size from PID17 to PID33, whereas HC spheroids only decreased by 10% in the same environment (Fig. 2E, right). However, adjacent HC spheroids derived from independent controls remained largely unchanged (Fig. 2E, left), highlighting the degeneration was specifically manifested in ADAD spheroids. To further validate that neurodegeneration becomes selectively manifested in ADAD spheroids, HC and ADAD spheroids were labeled with eGFP or turboRFP to mark live cells and cultured separately. Similar to co-culturing, ADAD-RFP spheroids exhibited a 50% reduction in RFP⁺ area at PID33, higher than the 10–20% reduction observed in HC spheroids (Fig. S5B). Labeling spheroids with a non-fluorescent reporter (lacZ) also resulted in a reduction in the size of ADAD spheroids (Fig. S5B). TUNEL staining of multiple sections (top, middle, bottom) of PID28 spheroids confirmed higher cell death in ADAD compared to HC spheroids (Fig. 2F). Since AD is also associated with progressive neurite degeneration (51), we examined neurite outgrowth in spheroids. Whereas HC spheroids maintained neurite length from PID22 to PID28, ADAD spheroids showed significant (p < 0.05 in distal and middle regions) retraction of neurite outgrowth at PID28 (Fig. 2G). Lastly, treating ADAD spheroids with β- or γ-secretase inhibitors reduced neuronal death (Fig. S5C), demonstrating that ADAD 3D-CNs and spheroids undergo Aβ-dependent neurodegeneration.

ADAD spheroids display distinct transcriptomic features

To explore the transcriptomic features in ADAD neurons, we carried out RNA-seq on PID25 spheroids derived from a different group of four individuals with ADAD and four age-/sex- matched HC. Comparing protein-coding gene expression between ADAD and HC spheroids, we detected 1411 differentially expressed genes (DEGs) (base mean >1, p < 0.05, |linear fold change| > 1.5) (Fig. 2H, Fig. S5F, and Table S4). 651 upregulated DEGs were identified in ADAD spheroids associated with gene ontology (GO) terms including cell-cell signaling, cell adhesion, immune response, synapse organization, and extrinsic apoptotic pathway (Fig. 2I). Among the most upregulated DEGs, higher amounts of MMP3 (52), IBSP (53), or TAGLN (54) in the brain have been linked to increased risk of AD. CCL20 (55), CSF3 (56) and IL10 (57) are associated with the inflammatory response in AD. A recent proteomic study identified HGF protein as one of the elevated proteins in the ADAD brain compared with cognitively normal individuals (Fig. S5F) (58). 760 downregulated DEGs were associated with cell adhesion, electron transport chain, NADP metabolism, and ubiquitin-dependent protein degradation. Among the top downregulated genes, common variants in ADH1B (59), and MS4A6E (60) have been associated with an increasing risk of AD. The expression of SST (61) and KCNIP1 (62) is reduced in the brains of individuals with AD, and downregulation of these two genes has been implicated in mediating the formation of Aβ oligomers. Downregulation of KCNIP3 in the brains of individuals with AD is associated with memory and learning (63).

LOAD neurons display increased Aβ deposition

Building on the success of replicating neuropathological phenotypes in ADAD neurons, we next asked whether neurons reprogrammed from fibroblasts of individuals with clinically affected LOAD could similarly reflect AD hallmarks. Currently, LOAD models that simultaneously capture Aβ, tau, and neurodegeneration characteristics are lacking. We directly reprogrammed fibroblasts from multiple independent individuals with LOAD composed of different APOE genotypes (APOE3/3, 4/3 and 4/4) and age-/sex-matched healthy individuals with APOE2/3, 3/3 or 3/4 (Table S1). It is important to note that Aβ deposition is not exclusive to individuals with AD as it can be found also in nondemented elderly individuals (64). Initially, we evaluated Aβ deposition in 3D-CNs from healthy individuals across different age groups. Aβ deposits were minimal in 3D-CNs derived from a young adult (22 years of age), but became more apparent with older age (Fig S6A). Thus, when comparing Aβ amounts between LOAD and HC neurons, it is critical to match the donor ages. Both LOAD and HC 3D-CNs contained extracellular Aβ deposits, as detected by 6E10 and 82E1 antibodies (Fig. 3A and Fig. S6B). Yet, LOAD 3D-CNs (derived from six independent individuals with LOAD, aged 60–89) showed higher amounts of Aβ deposition compared to HC 3D-CNs from six age-matched control individuals (Fig. 3A). Similarly, LOAD spheroids from six independent individuals exhibited significantly (p < 0.05) higher amounts of extracellular Aβ deposits at PID28 compared to age-matched HC spheroids (Fig. 3B, 3C and Fig. S7). These results underscore the potential of 3D neuronal reprogramming to model age-dependent Aβ deposition in LOAD neurons.

Fig. 3. Elevated Aβ deposition and pathogenic tau in LOAD 3D-CNs and spheroids.

(A) Aβ deposits in thin gel culture of PID30 LOAD and HC neurons. Boxed regions were highlighted on the right by 3D reconstruction showing extracellular Aβ deposition. Aβ deposits with a size > 1000 μm³ were used for quantification. (B) Whole mount immunostaining with Aβ antibody (6E10) in PID28 LOAD and HC spheroids. Aβ deposits with a size > 1000 μm³ were used for quantification. Boxed regions were magnified at the right panels to show Aβ deposition. (C) Immunostaining images of Aβ deposits on the sections of PID28 LOAD and HC spheroids using 6E10 Aβ antibody. (D) PCR analysis of 3R and 4R tau isoforms in LOAD and HC 3D-CNs at PID25. Numbers above the gel images are sample IDs. Data was shown as Mean ± SEM. (E) Immunostaining of p-tau (AT8 antibody) in LOAD and HC 3D-CNs at PID30. (F) Co-staining of K63-specific ubiquitin and p-tau (PHF1) in the neurites (PID30 3D-CNs). Arrows highlight the swelled dystrophic neurite bulges. (G) Live-cell imaging of FRET signal from HC and LOAD 3D-CNs containing Ruby2 and Clover reporters at PID28. For quantifications: n = 5– 6 LOAD and 5–6 HC individuals and 2–3 thin gels or spheroids per individual line were used in A, B, D, E, F and G. Unpaired t-tests were used for calculating p-values (* p < 0.05, ** p < 0.01 and ns p > 0.05). Scale bars: 50 μm in A and E, 500 μm in B, and 25 μm in C, F, and G.

LOAD neurons display tau pathology

AD is the most prevalent of tauopathies and is considered a secondary tauopathy because tau aggregation is regarded as a response to other pathological proteins or events (Aβ), but it is not driven by tau mutations (19, 65). We first examined if neurons reprogrammed from individuals with LOAD express both 3R and 4R tau isoforms by performing semi-quantitative PCR (20). Both 3R- and 4R-tau isoforms were detected with no noticeable difference in the 3R to 4R ratio between LOAD and HC 3D-CNs, consistent with the concept that AD is a mixed 3R/4R tauopathy (65, 66) (Fig. 3D). Several studies suggest that increased tau phosphorylation generally reflects brain aging (67, 68). We observed that tau phosphorylation increased with the age of control donors (Fig. S8A). Phosphorylated tau (AT8) was detected in both LOAD 3D-CNs and age-matched HC 3D-CNs when reprogrammed cells acquired the neuronal identity at PID30 and PID35, but not at PID10 and 20 (Fig. S8B). LOAD 3D-CNs displayed elevated p-tau amounts compared to HC 3D-CNs, as assessed by immunostaining with AT8 and PHF1 antibodies (Fig. 3E and Fig. S8C). LOAD 3D-CNs also showed higher signals for the MC1 tau antibody, which recognizes conformationally abnormal tau (40) (Fig. S8D). Additionally, LOAD 3D-CNs showed elevation of K63-ubiquitin⁺ pathogenic tau signals, primarily localized in beaded dystrophic neurites, compared to HC 3D-CNs (Fig 3F). Similar to ADAD 3D-CNs (Fig. 2D), LOAD 3D-CNs showed seed-competent tau in the beaded neurite regions, as detected by FRET signals (Fig. 3G). Altogether, these results suggest that neurons derived from individuals with LOAD can effectively capture tau dysregulation.

LOAD 3D-CNs and spheroids undergo spontaneous neurodegeneration.

Brain atrophy resulting from neuronal loss is a pathologic feature in LOAD (36). Sytox-Green assays revealed that the LOAD reprogrammed cells, but not age-matched HC cells, had an increase in cell death from PID20 to PID35. Specifically, LOAD and HC reprogramming cells did not display apparent cell death during the early phase of reprogramming (PID20), whereas PID30 and PID35 time points exhibited increased Sytox-Green signal in LOAD 3D-CNs compared to HC 3D-CNs (Fig. 4A), consistent with live-cell tracking over time showing neuronal loss of cell bodies and neurite length in LOAD 3D-CNs compared to HC 3D-CNs (Fig. S9A, movie S1). We cultured RFP-labeled LOAD spheroids next to GFP-labeled HC spheroids or separately. In both cases, we observed a ~ 50% reduction in the size of LOAD spheroids from PID17 to PID33, in contrast to ~10–20% reduction in size observed in the HC counterparts (Fig. 4B and Fig. S9B). TUNEL staining, assessed in various spheroid sections (top, middle, bottom), showed an increase in cell death in LOAD spheroids compared to control spheroids (Fig. 4C). Lastly, LOAD spheroids exhibited disintegration of neurites over time in comparison to HC spheroids (Fig. 4D), a phenomenon similarly observed in LOAD 3D-CNs (Fig. S9A). Taken together, our findings demonstrate the capability of modeling the neurodegeneration phenotype in neurons derived from individuals with LOAD.

Fig. 4. Spontaneous neurodegeneration in LOAD 3D-CNs and spheroids.

(A) Sytox-Green staining labeling dead cells in LOAD and HC 3D-CNs at different PIDs. Solid lines show the mean of each group whereas dotted lines represent each individual. Dead cells in LOAD 3D-CNs at PID20, PID30 and PID35 were compared to the HC 3D-CNs at the same PIDs. Mean ± SEM. p-values: two-way ANOVA with Šídák’s multiple comparisons. (B) LOAD spheroids and HC spheroids were co-cultured at PID7 and fluorescence images were taken at PID17 and PID33. Spheroid size (area, μm²) at PID33 was normalized to its size at PID17. Multiple live images were taken and stitched to display the whole spheroid. Mean ± SEM. Two-way ANOVA with Šídák’s multiple comparisons test. (C) TUNEL staining on sections of LOAD and HC spheroids (PID28). 4 confocal images were taken from different area on each section for 3 sections covering different planes of the spheroid. p-values: unpaired t-test. (D) The neurite outgrowth at proximal, middle, and distal regions in LOAD and HC spheroids was examined by immunofluorescence with TUBB3 antibody. TUBB3 signals in each region of the neurites were compared to the same region in HC89 spheroid at PID22 (set as 100%). Mean ± SEM. Multiple paired t-tests were calculated. For all quantifications: n = 5–6 LOAD and 5–6 HC individuals and 2–3 thin gels or spheroids per individual line were used. ns = p > 0.05, * p < 0.05, ** p < 0.01, *** p < 0.001, and **** p < 0.0001. Scale bars: 50 μm in A and C, 1 mm in B, and 250 μm in D.

Impairment of synapse formation in LOAD 3D-CNs

Pathogenic tau has been linked to synaptic dysfunction and loss (69). We detected decreased density of presynaptic marker SYN1 (Synapsin1), SYT1(Synaptotagmin1), or SYT9 (Synaptotagmin9)-positive puncta in LOAD 3D-CNs compared to HC 3D-CNs (Fig. S10A–C). We also compared synapse formation between LOAD and HC 3D-CNs. Cells were cultured in neuronal media supplement with human astrocyte-conditioned neuronal media starting at PID18 to promote synapse formation. LOAD and HC 3D-CNs were co-immunostained with postsynaptic marker PSD95 and presynaptic marker SYN1 at PID20 or PID30, and the number of SYN1/PSD95-double positive pairs was counted as the number of synapses (70). Whereas SYN1⁺ puncta were detected in PID20 and PID30 neurites, PSD95⁺ puncta were only apparent at PID30 (Fig. S10D), and largely absent at PID20. Thus, we counted the number of SYN1/PSD95-double positive signals at PID30 along the neurites as well as the percentage of PSD95⁺ puncta co-labeled with SYN1 in the neurites (Fig. S10D).

Quantification showed fewer numbers of SYN1⁺PSD95⁺ synapses and a reduction in PSD95⁺ signal in SYN1⁺ puncta in LOAD neurites compared to HC. SV2, a marker for mature synaptic vesicles and regulates presynaptic release (70, 71), was detected in PID20 HC and LOAD neurons with no overt difference in its expression. However, SV2 amount was reduced in LOAD 3D-CNs compared with HC 3D-CNs at PID30 (Fig. S10E). The amount of glutamate transporters labeled by VGLUT1 was not different between HC and LOAD 3D-CNs at PID20 or PID30 (Fig. S10F). Because PID30 aligns with the time point when LOAD 3D-CNs are undergoing degeneration (Fig. 4A and Fig. S9A), the impairment of synaptic formation likely reflects the degenerative state of LOAD neurons.

Effect of inhibiting APP processing in LOAD neurons on AD neuropathology.

Directly reprogrammed LOAD neurons provide an unprecedented patient neuron-based system that inherently captures AD-associated neuropathology without introducing genetic mutations or other perturbation of the system. We explored the interplay between Aβ deposition, tauopathy and neurodegeneration by examining the effects of reducing Aβ deposition on tau and neurodegeneration.

First, we treated reprogramming cells with β- or γ- secretase inhibitors at PID16, a time point before the observed onset of Aβ deposition (Fig. S6C). This early PID intervention substantially decreased Aβ deposits in the LOAD spheroids by 80–90% (Fig. 5A) and also reduced cleaved Aβ species in the media (Fig. S6D). In addition, we also treated reprogramming cells with β- or γ- secretase inhibitors at PID22, when some Aβ deposits had already formed (Fig. S6C). The reduction in Aβ deposition in LOAD spheroid treated at PID22 was either absent (with β-secretase inhibitor IV) or only modest (around 30% with DAPT) compared to DMSO-treated spheroids. Furthermore, treating b- or gsecretase inhibitors started at PID16 lowered p-tau and K63-ubiquitin/p-tau colocalization, whereas no effects were seen when treatment started at PID22 (Fig. 5B). Lastly, β- or γ- secretase inhibitors treated at PID16 effectively lowered neuronal death in LOAD spheroids by approximately 50%, whereas no effects were observed when treatment started at PID22 (Fig. 5C).

Fig. 5. Effects of inhibiting APP processing in LOAD 3D-CNs and spheroids on tauopathy and neurodegeneration.

(A) LOAD spheroids were treated with β-secretase inhibitor IV or DAPT starting at PID16 (before the observed onset of Aβ deposition) or PID22 (after the onset of Aβ deposition) and Aβ deposition was examined at PID28 by whole mount immunostaining using Aβ antibody (6E10). (B) Tau phosphorylation (AT8, top panel) and K63-ubiquitin/p-tau (PHF1) colocalization (bottom panel) were examined in PID28 LOAD 3D-CNs treated with β-secretase inhibitor IV or DAPT starting at PID16 or PID22. Arrows highlight swelled tau blebs in the bottom panel. (C) TUNEL staining of the sections of PID28 LOAD spheroids that were treated with β-secretase inhibitor IV or DAPT starting at PID16 or PID22. For quantifications in A, B, and C, n = 5 LOAD individuals and 2–3 thin gels or spheroids per line were used. Adjusted p-values were calculated by two-way ANOVA with Šídák’s multiple comparisons test. * p < 0.05; ** p < 0.01; **** p < 0.0001. Scale bars: 500 μm in A, 25 μm in B. and 50 μm in C.

We also tested whether treating ADAD neurons with β- or γ- secretase inhibitors starting either at PID16 or PID22 would lead to similar effects in reducing AD neuropathology as observed in LOAD. Overall, treatment started at PID16 was more effective in reducing neuronal death (Fig. S5D and E), Aβ deposition (Fig. S3H and I), and tauopathy (Fig. S4H and I) compared to starting the treatment at PID22. These results suggest that AD neurons undergo Aβ-dependent tauopathy and neurodegeneration, yet mitigation of neurodegeneration depends on the presence or absence of preformed Aβ deposits.

Activation of inflammatory pathways in LOAD spheroids

To infer transcriptional changes in LOAD neurons, we conducted RNA-seq on PID25 spheroids reprogrammed from five independent LOAD fibroblast lines and five sex-/age-matched HC lines. Principle component analysis revealed a separation of samples based on disease status (HC vs LOAD) (Fig. S11A). Comparing protein-coding gene expression between LOAD and HC spheroids, we identified 832 differentially expressed genes (DEGs) (p < 0.05, |linear fold change| > 1.5, base mean > 1) (Fig. 6A, Fig. S11B and Table S5). Functional gene annotation analysis revealed inflammation, Alzheimer’s disease, and cognition disorders are the top disease terms associated with all DEGs (Fig. 6B). Specifically, upregulated DEGs in LOAD spheroids were enriched for pathways such as inflammatory response and immune response (Fig. 6C). Among the most upregulated genes were matrix metallopeptidases (MMP) genes, including MMP1, MMP3 and MMP8 (Fig. 6A and Fig. S11B). MMPs are known to be linked to extracellular protein degradation including Aβ (72), whereas MMPs upregulated in AD may also confer neuroinflammation, synaptic dysfunction, and neuronal death (73, 74). Other upregulated DEGs included LEP, which encodes for Leptin, that has been linked to the sporadic form of AD (75). By gene ontology (GO), downregulated DEGs were enriched with terms linked to nucleosome assembly, telomere organization, and memory (Fig. 6C). Some of the downregulated genes were NPAS4, a TF that protects against age-dependent accumulation of DNA damages and somatic mutations (76), and PIWIL2, one of the four known human Piwi proteins (Fig. 6A and Fig. S11B). Depletion of piwi proteins was shown to increase transposable element (TE) transcription in a Drosophila model of tauopathy (77). Other DEGs also include DOC2A and FOXF1 (Table S5) that have been recently identified by a recent GWAS study as genes whose variants were linked to AD (78).

Fig. 6. LOAD spheroids display differentially expressed genes (DEGs) and differentially expressed TEs (DETEs).

(A) Volcano plot depicts differentially expressed genes between LOAD and HC spheroids. A total of 28 spheroid samples derived from 5 LOAD and 5 HC individuals were used for analysis. Each sample contains RNAs pooled from 2–3 spheroids. Down-regulated genes (p < 0.05, log2 (fold change) <−0.58), base mean > 1; Up-regulated genes (p < 0.05, log2 (fold change) > 0.58, base mean > 1). (B) Gene annotation of disease terms linked with all the significant DEGs analyzed by DisGeNET (C) Top GO terms associated with upregulated and downregulated DEGs analyzed by DAVID. (D) A heatmap of significant (p < 0.05, |log₂fold change| > 0.58) differentially expressed transposable elements (DETEs) in PID25 LOAD and HC spheroids (divided by two age groups) analyzed by RNA-seq. 270 DETEs were identified.

To further assess how the gene signatures of neurodegeneration in LOAD spheroids are reminiscent of its counterpart in the patient brain, we compared these DEGs with previously published gene expression profiling datasets of postmortem brains with AD as well as GO terms related to neurodegeneration (Table S7). First, we compared 832 DEGs identified in our LOAD spheroid RNA-seq dataset with the recently published gene list from single-nucleus RNA-seq (snRNA-seq) of postmortem human prefrontal cortex from individuals with AD (79). The signature protein-coding genes of neurodegeneration, upregulated in the excitatory neurons in the snRNA-seq, exhibited an overlap of 233 genes (Table S7, page 1). Among them, we found 170 commonly upregulated genes associated with aging, cell response to hypoxia, inflammatory response, and apoptosis (Fig. S11C). We further compared 832 DEGs with bulk RNA-seq datasets from various brain regions in individuals with AD (80–84), gene members in GO terms related to apoptosis/neurodegeneration, and AD risk genes from Alzgene database, revealing distinct sets of DEGs with overlaps ranging from 8 to 135 genes (Table S7, page 2, 3 and 4).

To explore common or unique gene expression changes between LOAD and ADAD, we compared the DEGs associated with LOAD versus old HC spheroids, and ADAD versus young HC spheroids. 103 commonly dysregulated genes were identified between LOAD and ADAD, which are associated with the regulation of synapse, autophagy, cytokine production, chromosome segregation, and voltage-gated potassium channels (Fig. S12A and Table S8). ADAD spheroid-specific DEGs were enriched with GO terms such as regulation of ERK1/2 cascade, calcium-dependent cell-cell adhesion, and response to calcium ion, whereas LOAD spheroids were uniquely associated with cellular response to cAMP, interferon-gamma, and lipopolysaccharide (Fig. S12A and Table S8). Furthermore, to investigate how aging is manifested in LOAD spheroids, we compared the DEGs between young and old HC spheroids (5365 genes, Table S9) to DEGs between young HC and LOAD spheroids (5219 genes, Table S9). 3983 common DEGs were identified between these two datasets, which were associated with translation, cell adhesion, response to hypoxia, axon guidance, ion transmembrane transport, and inflammatory response (Fig. S12B and Table S9). 1382 genes uniquely associated with old HC spheroids were linked to telomere organization, DNA methylation, anterior/posterior axis specification, extracellular matrix disassembly, and cell motility (Fig. S12B and Table S9). 1236 genes uniquely associated with LOAD spheroids were linked to extracellular matrix organization, cell-substrate adhesion, synapse assembly, cellular response to calcium ion, nervous system development, and learning (Fig. S12B and Table S9). Taken together, although common DEGs exist, LOAD spheroid transcriptome is characterized by the presence of age-associated pathways as well as unique DEGs that differ from ADAD spheroids.

Inhibition of retrotransposable element (RTE) suppresses neuropathologies in LOAD neurons

MiR-9/9*−124-induced neurons are primarily composed of neurons as previously characterized by single-cell RNA-seq (20, 29). We thus wondered why genes in inflammatory pathways, thought to be triggered by inflammatory signals from other cell types, were upregulated in LOAD neurons, and what neuron-intrinsic changes may trigger inflammation. As MMP1, MMP3 and MMP8 are among the top upregulated genes in LOAD spheroids (Fig. 6A and Fig. S11B) and upregulation of MMP genes has been linked neuroinflammation in AD (73, 85), we tested if repression of these MMP genes would ameliorate the neuropathology in LOAD neurons. Knocking down MMP1, MMP3 and MMP8 individually or together in LOAD neurons did not affect neuronal death, Aβ deposition and p-tau amount (Fig. S13), indicating targeting few select DEGs was not sufficient to reverse AD phenotypes.

It was previously shown that dysregulation of retrotransposon elements (RTEs) in aging may trigger inflammatory responses (86–88) and suppression of RTE was shown to reduce tau activation and tau-induced neurotoxicity in tau transgenic Drosophila (77). We thereby examined whether reprogrammed spheroids exhibited age-dependent changes in TE expression. TE analysis from RNA-seq datasets of HC spheroids divided by age (36–61 years vs. 66–90 years) and LOAD spheroids (66–90 years of age) revealed differentially expressed TEs (DETEs) primarily between age groups (Fig. 6D and Table S5), whereas subtle differences could be still detected between LOAD and age-matched HC (Fig. S11D and Table S5).

We then tested whether counteracting the global age-associated changes in RTE expression could influence AD phenotypes by treating LOAD neurons with the reverse transcriptase inhibitor, lamivudine (3TC (89)). LOAD neurons were treated with 3TC from PID16 to PID28, and the presence of single-stranded DNA (ssDNA), an intermediate product of retrotransposition and its amount has been served as an indicator for RTE synthesis (86), were reduced by ~50% at PID28 (Fig. 7A). The treatment with 3TC also reduced neuronal death in LOAD spheroids over the control H₂O treatment (Fig. 7B). 3TC also decreased Aβ formation and tau pathologies, as evidenced by the reduction of Aβ deposition, p-tau, and K63-linked ubiquitin-positive tau (Fig. 7C and 7D). Since ectopic retrotransposon activation has been associated with DNA damage and genomic instability (87, 90), we also examined DNA damage in 3TC-treated LOAD spheroids by immunostaining for 53BP1, a marker of double-stranded breaks (33). Reminiscent of persistent defective DNA repair in the brains of individuals with LOAD (79), LOAD spheroids exhibited an increase in DNA damage compared to age-matched HC spheroids. However, this increase in DNA damage was reduced by 3TC treatment (Fig 7E). We further confirmed these protective effects of 3TC were not due to the loss of neuronal identity, as 3TC-treated cells retained the expression of neuronal genes (Fig. S11E). To dissect the transcriptomic changes in spheroids treated with 3TC, we carried out RNA-Seq on LOAD spheroids derived from 4 individuals with LOAD and treated with 3TC or water (control). 534 DEGs (p < 0.05, |linear folder change| > 1.5, base mean > 1, protein-coding) were identified (248 upregulated and 286 downregulated DEGs) (Fig. 7F and Table S6). Both PI16 and ADAMDEC1, whose expression is strongly suppressed by inflammatory response (91, 92), were increased in 3TC-treated spheroids (Fig. 7F). GO analysis revealed that 3TC-induced DEGs were associated with G-protein coupled receptor signaling pathway, immune response, and inflammatory response (Fig. 7G), suggesting 3TC suppresses neurodegeneration in LOAD neurons through regulating RTE-associated neuroinflammation.

Fig. 7. 3TC (lamivudine), a reverse transcriptase inhibitor, ameliorates LOAD neuropathologies.

(A) Top: Schematic of 3TC treatment for inhibiting RTE formation in LOAD spheroids. Bottom: Immunofluorescent images of ssDNA detected in young HC, old HC, LOAD, and 3TC-treated LOAD spheroids. (B) TUNEL staining for cell deaths in 3TC- or H₂O- (control) treated spheroids (PID28). (C) Whole mount immunostaining images of Aβ deposition in 3TC and H₂O-treated spheroids at PID28. (D) p-Tau labeled by AT8 antibody (top panel) and co-localization of p-tau (PHF1) and K63-ubiquitin (bottom panel) in 3TC- or H₂O-treated LOAD 3D-CNs at PID28. Arrows highlight the dystrophic, bulged tau blebs. (E) Double-stranded DNA breaks labeled by 53BP1 antibody in age-matched HC, LOAD, and 3TC-treated LOAD spheroids. (F) Volcano plot of DEGs between 3TC and H₂O-treated spheroids at PID28. 3 biological replicates per cell line, 4 independent LOAD cell lines were used per treatment (3TC or H₂O), and a total 24 spheroid samples were used for the analysis. (G) Top GO terms for all DEGs (p < 0.05, |log2 (fold change)| > 0.58, base mean > 1) analyzed by DAVID. For quantifications: n = 4 HC or/and LOAD lines and 2–3 thin gels or spheroids per line were used in A, B, C, D, and E. Adjusted p-values in A and E were calculated by one-way ANOVA with Tukey multiple comparisons test. p-values in B, C, and D were calculated by paired t-test. * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001. Scale bar: 10 μm in A and E, 50 μm in B, 500 μm in C and 25 μm in D.

In addition, we treated old HC 3D-CNs and spheroids with 3TC and observed a mild yet significant (p < 0.05) reduction of neuronal death and Aβ deposition (Fig. S14A–D). However, 3TC treatment in ADAD neurons and spheroids did not affect AD neuropathology, likely reflecting the small number of 13 DETEs detected between ADAD spheroids and age-matched young HC spheroids (Fig. S14E–H), suggesting that the protective effect of RTE inhibition is more pronounced in the aged LOAD context. In summary, our findings highlight the potential of using patient-derived specific LOAD neurons to identify chemical or genetic interventions that may provide therapeutic benefits by increasing neuronal resilience against late-onset neurodegeneration in LOAD.

Conclusion

Findings in this study demonstrate the effectiveness of miRNA-based neuronal reprogramming in recapitulating Aβ deposition, tauopathy, and neurodegeneration phenotypes in both ADAD and LOAD patient-derived neurons in a 3D environment. Directly reprogrammed LOAD neurons, even without ADAD-associated mutations, can capture age-associated neuropathological features. This suggests neuron-intrinsic mechanisms influenced by the donor’s genetic background and epigenetic signatures that render susceptibility to neurons with aging. Whereas the inhibition of APP processing mitigated degeneration of LOAD neurons, the effect was clear only when cells were treated before the formation of Ab deposits, not after the onset. Identifying other factors contributing to degeneration of LOAD patient-specific neurons would provide invaluable insights in devising strategies to mitigate neurodegeneration. Future research should be directed to whole genome sequencing and chromatin profiling to facilitate the identification of genetic and epigenetic changes underlying degeneration of LOAD neurons. Moreover, genes involved in APP metabolism were identified as AD risk genes (78). MiRNA-induced AD neurons then can serve as a platform to test risk genes in neurons.

It is noteworthy that Aβ deposits and p-tau signals correlate with age even in neurons derived from healthy individuals (Fig. S6A and Fig. S8A), suggesting the age-associated alterations underlying the changes in APP metabolism and tau phosphorylation. This highlights the important role of age as the major risk factor for LOAD compared to ADAD, in which mutations associated APP processing drive the disease onset. Although mouse models overexpressing gene mutations associated with ADAD offer valuable insights into Aβ plaque formation (93), they show no obvious tau pathology. It typically requires additional expression of human tau harboring mutations associated with primary tauopathy (94–96) or seeding mutant human APP knock-in mouse brain with human AD brain-derived tau (97) to develop the tau pathology. As for human neurons, previous studies focused on overexpressing mutant APP and Presenilin in REN stem cells grown in a 3D culture (3), which differs from patient-specific LOAD model that relies on the endogenous recapitulation of AD-associated phenotypes as provided in this study.

The mechanism underlying 3TC leading to the reduction of Aβ formation remains unclear. As RTE difference was detected as a general age-associated process, the protective effect of 3TC likely represents an effect of perturbing an age-related mechanism that contributes to AD. Indeed, counteracting aging pathways has been shown to lower age- and disease stage-dependent neurodegeneration in a patient-derived Huntington disease model (32). Moreover, 3TC exerts a more robust effect in LOAD neurons compared with early-onset ADAD neurons, likely reflecting the general age-associated RTE dysregulation as a risk factor that is captured in LOAD spheroids. It remains unclear why LOAD neurons are selectively vulnerable to RTE dysregulation, whereas old HC neurons can tolerate it. Future research directions should focus on delineating this differential susceptibility. In addition, a recent study demonstrated that disruption of genome stability and 3D genome organization are pathological steps in progression of AD (79). This study revealed that 3TC treatment also reduces DNA damage in LOAD neurons. Thus, future studies should be directed to investigating the relationship between 3TC and changes in genome instability around the regions containing RTEs to gain further insights into the neuroprotection. Overall, results in this study demonstrate the effectiveness and sufficiency of 3D directly reprogrammed patient neurons to capture adult-onset neuropathology of LOAD. This groundwork now allows future avenues for studying how these neuropathological phenotypes may be modified when LOAD neurons interact with other cell types in the brain. In this regard, as aging may impact the function of astrocytes (98, 99) and microglia (100), robust direct reprogramming methods for generating aged astrocytes and microglia would enable the study of neuron-glia interactions in age-associated disease progression.

Materials and methods summary

Detailed materials and methods can be found in the supplemental materials. A brief of the key method is provided below.

3D direct neuronal reprogramming:

Human fibroblasts were spin-infected with a lentiviral cocktail containing rtTA, pTight-9–124-BclxL, MYT1L and NEUROD2. The fibroblast medium was changed every other day with 10% FBS, 1 μg/ml doxycycline and antibiotics. On day 7, cells were trypsinized, resuspended in transiting medium and counted for replating. For thin gel culture, 0.1×10⁶ cells resuspended in 100 μL transiting medium containing 15% Matrigel were immediately transferred to each well of an optically clear 96-well plate to form a thin gel. For spheroid culture, 0.5×10⁶ cells were centrifuged at 400g for 2.5 min at room temperature to form a cell pellet. The cell pellet was transferred to the center of a transwell insert in a 24 well plate by using a wide-bore tip to form a spheroid. After replating, half media changes were performed every other day for both thin gel culture and spheroid culture.

Data Availability:

Raw RNA-seq data presented in this study has been deposited and will be available through Gene Expression Omnibus (GEO) at NCBI (GSE267613, GSE252932, and GSE253174). Information for AD and healthy control individuals, tables for DEG and DETE from RNA-seq analysis are included as in Supplementary tables. Source data that supports all the statistics in this study are available in Supplemental Table S10.