Modelling APOE ɛ3/4 allele-associated sporadic Alzheimer’s disease in an induced neuron

Abstract

The recent generation of induced neurons by direct lineage conversion holds promise for in vitro modelling of sporadic Alzheimer’s disease. Here, we report the generation of induced neuron-based model of sporadic Alzheimer’s disease in mice and humans, and used this system to explore the pathogenic mechanisms resulting from the sporadic Alzheimer’s disease risk factor apolipoprotein E (APOE) ɛ3/4 allele. We show that mouse and human induced neurons overexpressing mutant amyloid precursor protein in the background of APOE ɛ3/4 allele exhibit altered amyloid precursor protein (APP) processing, abnormally increased production of amyloid-β₄₂ and hyperphosphorylation of tau. Importantly, we demonstrate that APOE ɛ3/4 patient induced neuron culture models can faithfully recapitulate molecular signatures seen in APOE ɛ3/4-associated sporadic Alzheimer’s disease patients. Moreover, analysis of the gene network derived from APOE ɛ3/4 patient induced neurons reveals a strong interaction between APOE ɛ3/4 and another Alzheimer’s disease risk factor, desmoglein 2 (DSG2). Knockdown of DSG2 in APOE ɛ3/4 induced neurons effectively rescued defective APP processing, demonstrating the functional importance of this interaction. These data provide a direct connection between APOE ɛ3/4 and another Alzheimer’s disease susceptibility gene and demonstrate in proof of principle the utility of induced neuron-based modelling of Alzheimer’s disease for therapeutic discovery.

Alzheimer’s disease, amyloid-β, neuroprotection, APP, APOE

Introduction

Alzheimer’s disease is the most common neurodegenerative disorder, characterized by cognitive decline with the loss of memory (Selkoe, 2002; Vetrivel et al., 2008). The major pathological features of Alzheimer’s disease include the accumulation of amyloid aggregates and hyperphosphorylation of tau protein (Selkoe, 2005). Known mutations causing Alzheimer’s disease act either by increasing the steady-state level of amyloid-β, altering the amyloid-β_42/40 ratio or altering the amyloidogenic potential of amyloid-β, which eventually leads to disease pathogenesis (Citron et al., 1998). Particularly, mutations in the amyloid precursor protein (APP), presenilin 1 (PSEN1) and presenilin 2 (PSEN2) genes cause an amino acid substitution in the sites at which the APP is cleaved, leading to accumulation of mainly amyloid-β₄₂ (Wolfe et al., 1999; Singleton et al., 2000; Tanzi et al., 2004; Shen and Kelleher, 2007). Additionally, the apolipoprotein E (APOE) ɛ3/4 allele is a major risk factor for sporadic Alzheimer’s disease and is closely associated with late onset of disease and increased amyloid plaques in Alzheimer’s disease (Reiman et al., 1996; Mayeux et al., 1998; Teasdale et al., 2005; Koffie et al., 2012). The aggregates of amyloid-β₄₂ are a major component of senile plaques, affecting the formation of neurofibrillary tangles and consequently neuronal loss (Glenner et al., 1984). Despite this fundamental understanding of Alzheimer’s disease pathology, a lack of access to appropriate human neurons for modelling Alzheimer’s disease remains a major challenge for moving therapeutic development forward.

Recent advances in direct reprogramming have garnered considerable interest for human disease modelling and cell replacement strategies (Vierbuchen and Wernig, 2011). Several studies have demonstrated the feasibility of generating human induced pluripotent stem cells from patients with rare, early-onset familial Alzheimer’s disease mutations and sporadic Alzheimer’s disease (Israel et al., 2012; Kondo et al., 2013; Young et al., 2015; Raja et al., 2016). However, the inefficient processes of induced pluripotent stem cell generation and directed differentiation into specific target cell populations limit the application of induced pluripotent stem cell technology in neurodegenerative disease modelling (HD iPSC Consortium, 2012; Arber et al., 2015). More recently, several studies have reported the direct conversion of one specific somatic cell type to another by forced expression of master transcriptional regulators (Ieda et al., 2010; Efe et al., 2011; Marro et al., 2011; Yang et al., 2013). Pioneering work on neuronal reprogramming demonstrated the feasibility of direct conversion of mouse and human fibroblasts into functional neurons (Vierbuchen et al., 2010), which can ultimately be applied to model neurological diseases and conduct drug screening against disease phenotypes more efficiently. Particularly, induced neuron-based modelling of late onset sporadic Alzheimer’s disease would provide a useful approach to understanding sporadic Alzheimer’s disease pathogenesis and facilitate therapeutic discovery.

Previously, we and others reported that both mouse and human fibroblasts can be directly converted into subtypes of functional neurons (Ambasudhan et al., 2011; Kim et al., 2011; Pfisterer et al., 2011). Several lines of evidence suggested that the induced neurons display morphological and functional properties of neurons indicating that induced neurons can be used for neurological disease modelling (Cremades et al., 2012; Meyer et al., 2014). However, direct conversion into human induced neurons occurs with very low efficiency (Ambasudhan et al., 2011; Pfisterer et al., 2011). To overcome these challenges, we applied nanopattern topography to cultures of human induced neurons as an efficient stimulant for direct lineage reprogramming (Yoo et al., 2015). We use this optimized lineage reprogramming system to demonstrate that induced neurons generating from sporadic patients with Alzheimer’s disease harbouring APOE ɛ3/4 allele exhibit increased production of amyloid-β peptides. Moreover, these induced neurons recapitulate the pathological features and molecular signatures found in the primary neuronal tissue of APOE ɛ3/4 patients with Alzheimer’s disease. We subsequently examined the transcriptional regulatory network in Alzheimer’s disease induced neurons and identify the desmoglein 2 (DSG2) gene as a candidate modifier of the Alzheimer’s disease phenotype. DSG2 was previously found to be a risk factor for late onset Alzheimer’s disease through genome-wide association studies (Lambert et al., 2013; Karch and Goate, 2015). We thus tested the functional significance of DSG2 dysregulation and observed that DSG2 inhibition can suppress APOE ɛ3/4-induced neurotoxicity in both human and mouse induced neuron models. Taken together, our results demonstrate that induced neurons can be used to effectively model Alzheimer’s disease and can provide a platform for the discovery and functional evaluation of candidate therapeutic modalities.

Materials and methods

Culture of mouse and human fibroblasts

Mouse embryonic fibroblasts were isolated from Tau-eGFP knock-in or wild-type embryonic Day 13 mouse embryos. After removing the head containing the spinal cord, dorsal root ganglia, and all internal organs, the remaining tissue was dissociated and incubated in trypsin (0.25%, Invitrogen) for 10–15 min. Single cells from each embryo were plated onto a 15 cm cell culture dish with mouse embryonic fibroblast medium [Dulbecco’s modified Eagle medium (DMEM) containing 10% foetal bovine serum and 1% penicillin/streptomycin]. Human control (GM23967, male, condition: healthy control, APOE ɛ3/3 genotype) and Alzheimer’s disease fibroblasts (AG11414, male, condition: Alzheimer’s disease early APOE ɛ3/4 genotype), (AG05810, female, condition: Alzheimer’s disease APOE ɛ3/4 genotype), (AG05770, male, condition: unknown, APOE ɛ3/3 genotype), (AG06840, male, condition: Alzheimer’s disease PSEN1 mutation, APOE ɛ3/3 genotype), (AG09908, female, condition: Alzheimer’s disease PSEN2 mutation, APOE ɛ3/3 genotype) were purchased from the Coriell Cell Repository. Human fibroblasts were maintained in human fibroblast medium [DMEM containing 10% foetal bovine serum, 1% non-essential amino acid (Gibco), 0.1% β-mercaptoethanol (Gibco) and 1% penicillin/streptomycin (Gibco)].

Lentivirus generation and viral infection

To overexpress human APOE ɛ4 in mouse induced neurons, APOE ɛ4 cDNA from APOE ɛ3/4 patient fibroblasts was cloned into lentiplasmid and the lentivirus was infected with reprogramming factors. Lentivirus was produced in HEK293T cells, which were grown in DMEM containing 10% foetal bovine serum and 1% penicillin/streptomycin. Subsequently, the cells were transfected with the lentivirus construct, Ascl1, Brn2, Myt1l, NeuroD1, mutant APP, M2rtTA and psPAX2, pMD2.G vectors using calcium phosphate co-precipitation. Following transfection for ∼24 h, the medium was changed and the viruses were harvested 48–72 h later. Mouse embryonic fibroblasts were infected with the lentivirus at passage 3, twice in 2 days. The infected mouse embryonic fibroblasts were cultured in N3 medium containing DMEM/F12, insulin (25 µg/ml), progesterone (20 nM), transferrin (50 µg/ml), putrescine (100 µM), laminin (1 µg/ml), basic fibroblast growth factor (25 µg/ml) and 1% penicillin/streptomycin.

Fabrication of nano-grooved substrates

Polyurethane acrylate nanoscale grooved patterns (300 nm ridge, groove width of 400 nm) were fabricated on glass substrates using UV-assisted capillary force lithography as previously described (Lee et al., 2010; Yoo et al., 2015). In brief, a UV curable polyurethane acrylate precursor solution (Minuta Tech. Inc.) was dispensed onto glass coverslips (diameter: 25 mm, thickness: 100 mm, Marienfeld) that had been coated with an adhesion promoter (phosphoric acrylate: isopropyl alcohol; 1:10 v/v). A flexible mould with nanoscale patterns was placed directly onto the surface, and the mould cavity was spontaneously filled with the polyurethane acrylate precursor via capillary action. The polyurethane acrylate nanoscale solution was cured by exposure to UV light (l ¼ 250 e400 nm) for ∼30 s (100 mJ/cm²) and the mould was peeled off using sharp tweezers. The as-synthesized nano-grooved substrates were cleaned using 70% ethanol, rinsed with distilled water and then coated with gelatin by immersion in a 0.1% (w/v) gelatin solution.

Direct conversion of human fibroblast into induced neurons

To generate human induced neurons, human fibroblasts were plated on nano-patterned substrates. Polybrene (Sigma) was used to improve the efficiency of the infection process. Plated human fibroblasts were infected with lentivirus FUW-Ascl1, Brn2, Myt1l, NeuroD1, Teto-mut APP, M2rtta three times in 2 days and were maintained in N3 medium containing DMEM/F12, insulin (25 µg/ml), progesterone (20 nM), transferrin (50 µg/ml), putrescine (100 µM), laminin (1 µg/ml), fibroblast growth factor basic (25 µg/ml), brain-derived neurotrophic factor (10 g/ml) and 1% penicillin/streptomycin. To overexpress mutant APP in induced neurons, doxycycline was treated at 7 days after initial factor infection.

Flow cytometry

Cells were dissociated using 0.125% trypsin-EDTA for 4 min and the single cells were then pelleted, resuspended in 4% paraformaldehyde, and incubated for 10 min at 4°C. Following this, the cells were washed with 1% bovine serum albumin twice, and resuspended in fluorescence-activated cell sorting buffer for analysis. Flow cytometry was performed using an Accuri instrument (Becton-Dickinson). All data were analysed with FlowJo vX software (TreeStar).

Immunofluorescent staining analysis

The cells were washed with phosphate-buffered saline (PBS) before being fixed in 4% paraformaldehyde in PBS. Cells were immunostained according to standard protocols using the following primary antibodies: Tuj1 (Sigma), MAP2 (Cell Signaling), VGluT1 (Abcam), NeuN (Millipore), Cleaved-caspase3 (Cell Signaling), Phosphorylated tau (AT8, Pierce), APP (Millipore), Nestin (Millipore), EEA1 (Millipore), Amyloid-β₄₂ (Millipore) and appropriate fluorescent secondary antibodies (Invitrogen). Representative images were captured using a Nikon Eclipse Ti microscope. The directions, lengths and 3D features of the induced neurons were analysed using Adobe Photoshop^® software.

Western blot analysis

Induced neurons were extracted using Tris-buffered saline (TBS) extraction buffer containing 1 × TBS buffer, 1 mM NaF, 1 mM NaVO₃, and protease inhibitor mixture (Roche) before being incubated on ice for 20 min. To obtain TBS-soluble fractions, the samples were centrifuged for 1 h at 20 000g at 4°C. The remaining pellets were then resuspended in 2% SDS extraction buffer containing 1 × TBS buffer, 1% Triton^™ X-100, 2% SDS, 1 mM NaF, 1 mM NaVO₃, and protease inhibitor mixture (GenDepot) before being incubated on ice for 30 min. To obtain TBS-insoluble/SDS-soluble fractions, the samples were centrifuged for 1 h at 20 000g at 4°C. The supernatants were extracted and analysed to detect the protein levels. The TBS-insoluble/SDS-soluble fractions were electrophoresed on 12% Bis-Tris gels before being transferred to nitrocellulose membranes (GE Healthcare Biosciences). Representative images were captured using Chemidoc TRS+ with image lab software (Bio-Rad). The primary monoclonal antibodies were used as follows: anti-Beta-amyloid 6E10 (1:400, BioLegend), beta-actin (1:1000, Abfrontier).

Quantitative real-time polymerase chain reaction analysis

Quantitative real-time polymerase chain reaction (qRT-PCR) was performed using Platinum^® SYBR^® green qPCR SuperMix (Invitrogen). qRT-PCR analysis was conducted by the reverse transcription reaction in a Rotor-gene Q (QIAGEN). Total RNA was purified from the samples and isolated using an RNeasy® Kit (QIAGEN) according to the manufacturer’s protocols. Gene expression of fibroblast and neuronal markers was normalized against GAPDH levels in each sample.

Amyloid-β ELISA

ELISA analysis was performed using a human amyloid-β assay ELISA kits (IBL). Neuronal culture media were collected from cultured induced neurons after 72–96 h before being centrifuged at 20 000g for min. To harvest intracellular amyloid-β₄₀ and amyloid-β₄₂, cells were lysed by ELISA sampling buffer (1 × TBS buffer, 1% Triton^™ X-100, 0.1% SDS). The collected samples were subjected to ELISA kits including amyloid-β (1–42) and amyloid-β (1–40), according to the manufacturer’s instructions.

Electrophysiology

For electrophysiological analysis, cells were cultured on nanopatterned glass coverslips before being transferred to recording medium containing the extracellular solution [in mM: 130 NaCl, 4 KCl, 2 CaCl₂, 1 MgCl₂, 10 HEPES, 10 glucose (pH 7.35)]. Patch pipettes were filled with the intracellular solution (in mM: 110 potassium gluconate, 20 KCl, 2 Mg-ATP, 10 sodium phosphocreatine, 1.0 EGTA, 0.3 GTP-Tris, and 20 HEPES) (pH 7.25). Electrodes were fabricated from borosilicate capillary glass tubing (Warner Instruments) with a capillary glass puller (Sutter Instruments). The patch electrodes were fire-polished using a microforge (Narishige) and had resistances of 2–4 MΩ. After establishing whole-cell mode, the cell membrane capacitance, and series resistance were compensated to 75% electronically using a patch clamp amplifier (Axopatch 200B; Molecular Devices). Data generation and acquisition were performed using the pClamp10 software on an IBM computer equipped with an analogue-to-digital converter (Digidata 1440 A; Axon Instruments). Once current-clamp mode was obtained, the cell was maintained at a potential of approximately −65 to −70 mV. Current injection protocol steps were applied ranging from −100 pA to +120 pA. Inward sodium current and outward potassium current were measured in voltage-clamp mode with voltage step protocols ranging from −55 to +55 mV.

Gene expression profiling, gene set enrichment analysis, and network analysis

Affymetrix GeneChip Human Gene 2.0 ST Array was performed according to the manufacturer’s protocol. Robust Multiarray Averaging method using affy R package was used for normalization and summarization. When multiple probes were available, averaged values per gene were applied. Gene set enrichment analysis (Subramanian et al., 2005) was performed using gene set enrichment analysis pre-ranked mode to determine whether gene sets were statistically enriched in both APOE ɛ3/4 sporadic Alzheimer’s disease patient induced neurons versus APOE ɛ3/3 control induced neurons and APOE ɛ3/4 sporadic Alzheimer’s disease patient versus APOE ɛ3/3 controls in human brain tissue (GSE48350). Curated gene sets (1691 genes of BLALOCK_ALZHEIMERS_DISEASE and 390 genes of BLALOCK _ALZHEIMERS_DISEASE_INCIPIENT) in Molecular Signatures Database v5.1 and differentially expressed genes (fold change ≥ 2) in APOE ɛ3/4 patient induced neurons versus APOE ɛ3/3 control induced neurons were used. The results of gene set enrichment analysis are considered significant when the false discovery rate and nominal P-value are < 0.05. The interactome of proteins in Homo sapiens was obtained from NCBI Gene (ftp://ftp.ncbi.nlm.nih.gov/gene/GeneRIF/). To identify the gene network found in Alzheimer's disease patients with APOE ɛ3/4 genotype-derived induced neurons, the protein-protein interactions among differentially expressed genes were imported into Cytoscape (http://www.Cytoscape.org, v3.3.0) (Shannon et al., 2003) and analysed using Molecular Complex Detection, v1.4.1 (Bader and Hogue, 2003) to find the protein network modules. The data supporting the findings of this study are available in the Supplementary material or from the corresponding author on request.

Statistical analysis

All data are presented as the mean ± standard deviation (SD) of three independent experiments. n-Values indicate the number of independent experiments performed or the number of individual experiments or mouse. For each independent in vitro experiment, at least three technical replicates were used and a minimum number of three independent experiments were performed to ensure adequate statistical power. In all of the analyses, group differences were considered statistically significant at P < 0.05. (*P < 0.05, **P < 0.01). ANOVA test was used for multi-component comparisons and Student’s t-test for two-component comparisons after the normal distribution was confirmed.

Results

Alzheimer’s disease modelling in mouse induced neurons

In proof-of-principle studies, we initially established induced neurons from mouse fibroblasts to model Alzheimer’s disease pathology. To develop mouse induced neurons that produce a high level of toxic amyloid-β, we overexpressed a human APP containing 670/M671 Swedish mutations. First, mouse fibroblasts were transduced with lentivirus constitutively expressing Ascl1, Brn2, and Myt1l along with a doxycycline-inducible mutant APP lentivirus (Fig. 1A). Seven days of Ascl1, Brn2, Myt1l expression was previously shown to be sufficient for converting mouse fibroblasts into functional induced neurons (Vierbuchen et al., 2010). We also confirmed that at Day 7, prior to doxycycline-induction, there was no distinction between the control FUW-Ascl1, Brn2, and Myt1l infected cultures and those also containing the doxycycline-inducible APP (Fig. 1B and Supplementary Fig. 1A–D). To avoid the effects of mutant APP on the formation induced neurons, doxycycline was introduced into the culture 7 days after Ascl1, Brn2, Myt1l expression (Fig. 1A). Fifteen days after Ascl1, Brn2, Myt1l expression, control induced neurons exhibited expression of neuronal markers such as Tuj1, VGluT1 and NeuN (Fig. 1C and D), which were not observed in fibroblasts transduced with control lentivirus (Fig. 1C and D, left panel). Consistent with previous reports, we observed that most Tuj1-positive mouse induced neurons were also positive for the glutamatergic neuron marker, VGluT1 and the mature post-mitotic neuron marker, NeuN (Fig. 1C and D, and Supplementary Fig. 1E). However, we were not able to observe expression of markers of neural progenitor/stem cells such as Nestin (Supplementary Fig. 1F), suggesting these induced neurons more closely resemble mature glutamatergic forebrain neurons.

Figure 1

Modelling Alzheimer’s disease phenotypes in mouse induced neurons. (A) Schematic of modelling Alzheimer’s disease (AD) phenotypes in mouse induced neurons (iNs). Mouse embryonic fibroblasts (MEFs) were transduced with lentivirus constitutively expressing Ascl1, Brn2, and Myt1l along with a doxycycline-inducible human mutant APP lentivirus and FUW-APOE ɛ4 lentivirus. To induce Alzheimer’s disease phenotypes in the reprogrammed induced neurons, doxycycline was introduced into the culture 7 days after initial Ascl1, Brn2, and Myt1l induction. (B) The number of Tuj1+ cells in fibroblasts, induced neurons and APP-expressing induced neurons at different time points. To induced mutant APP expression, doxycycline was treated on Day 7 after initial factor infection. Data represent mean ± SEM. Student’s t-test, *P < 0.05, **P < 0.01; n = 4 per each sample. (C) Immunofluorescence of Tuj1+ induced neurons in the presence and absence of mutant APP induction for 8 days. Scale bar = 100 µm. (D) Co-staining of induced neuron cultures with the Tuj1, VGluT1 and NeuN at 8 days after doxycycline induction. Scale bar = 100 µm. (E) Measurement of the neurite length of control and APP expressing induced neurons. Data represent mean ± SEM. Student’s t-test, **P < 0.01; n = 5 per each sample. (F) ELISA analysis of amyloid-β₄₂ (top) and amyloid-β₄₀ (bottom) secreted from controls and APP expressing induced neurons treated with β-secretase inhibitor and γ-secretase inhibitor. Data represent mean ± SEM. ANOVA test, *P < 0.05, **P < 0.01; n = 6 per each sample. (G) Immunofluorescence staining showing the production of amyloid-β₄₂ on 16 days after doxycycline induction. Scale bar = 100 µm. (H) The number of amyloid-β₄₂ positive cells per Tuj1+ at different time point after doxycycline induction. Data represent mean ± SEM. ANOVA test, *P < 0.05, **P < 0.01; n = 6 per each sample. (I) The ratio of amyloid-β₄₂/amyloid-β₄₀ in control, APP expressing induced neurons and APP expressing APOE ɛ4/4 induced neurons at 16 days after doxycycline induction. Data represent mean ± SEM. Student’s t-test, **P < 0.01; n = 6 per each sample. (J) The number of phosphorylated tau/Map2-positive cells after doxycycline induction. Data represent mean ± SEM. n = 3 per each sample. (K) The ratio of p-tau/total tau in control, APP expressing induced neurons and APP and human APOE ɛ4 expressing induced neurons. Data represent mean ± SEM. ANOVA test, *P < 0.05, **P < 0.01; n = 6 per each sample. Dox = doxycycline.

Remarkably, we observed that the induced neuron cultures overexpressing human mutant APP exhibited decreased neurite length and numbers of neuronal specific Tuj1-, VGluT1- and NeuN-positive cells (Figure 1B, D, E and Supplementary Fig. 1E). We also examined the number of Tuj1+ induced neurons in the cultures across different time points, and found that 8 days of APP induction resulted in a significant decrease in the number of Tuj1+ cells (Fig. 1B and Supplementary Fig. 1B), suggesting that mutant APP expression may lead to degeneration of the induced neurons. Consist with these results, expression of neuronal transcripts Mapt, Map2, Gad1 (GAD67), Tbr1, Syn1, and Bcl11b (Ctip2), were all markedly decreased in APP expressing induced neuron cultures relative to controls (Supplementary Fig. 1G). Additionally, induced neurons derived from Tau-eGFP fibroblasts were ∼7% GFP+, and the frequency of Tau-eGFP+ induced neurons was significantly decreased in APP-expressing cultures (Supplementary Fig. 1H).

As a proof of the utility of mouse induced neurons in Alzheimer’s disease modelling, we first examined Alzheimer’s disease-associated phenotypes in APP expressing induced neuron cultures. Immunocytochemical analysis conducted 8 days after APP induction revealed that most APP-expressing induced neurons express amyloid-β₄₂ (Fig. 1G). We monitored the number of amyloid-β₄₂+/Tuj1+ cells over a period of 3 weeks from the onset of doxycycline induction and observed a rapid, linear increase in amyloid-β₄₂+ and Tuj1+ double-positive neurons across the time course, but APP expressing induced neurons exhibited the most significant increase in amyloid-β₄₂+ neurons (Fig. 1H). Amyloid-β₄₂ is a major component of amyloid plaques found in the brains of patients with Alzheimer’s disease, and an increased amyloid-β₄₂:amyloid-β₄₀ ratio is a key feature of Alzheimer’s disease pathogenesis (Serneels et al., 2009). Thus, we examined extracellular amyloid-β secretion in the mutant APP-expressing mouse induced neurons. Importantly, we observed increased amyloid-β₄₂ secretion into the medium of APP-expressing induced neurons and a significantly elevated amyloid-β₄₂:amyloid-β₄₀ ratio compared with control and APP only expressing induced neurons (Fig. 1I). Hyperphosphorylation and abnormal accumulation of the microtubule-associated protein tau (MAPT) are key pathological features of Alzheimer’s disease (Gotz et al., 1995; Shi et al., 2012). Detection of tau phosphorylation, Ser202 and Thr205 with the AT8 antibody revealed increased p-tau accumulation in APP expressing induced neurons (Fig. 1J). Moreover, we observed co-staining of p-tau with MAP2, and the ratio of p-tau/tau was significantly increased in APP expressing induced neurons, suggesting that phosphorylated tau was aberrantly localized in these cells (Fig. 1J and K).

APP is largely processed in the vesicular endosomal compartments, and increased APP expression leads to alterations of intracellular trafficking, which has consequently been implicated in Alzheimer’s disease pathogenesis (Thinakaran and Koo, 2008). Thus, we examined whether APP-expressing induced neurons exhibit abnormal APP localization in vesicular endosomal compartments. We observed that localization of APP puncta positive for an early endosomal marker, EEA1, in induced neurons, and the number of APP/EEA1 double positive compartments was significantly increased in mutant APP-expressing induced neurons (Supplementary Fig. 2A and B), whereas no differences in the expression of the APP processing enzyme β-secretase-1 (BACE1) were observed (Supplementary Fig. 2C).

Next, we examine the susceptibility of APP-expressing induced neurons to hydrogen peroxide-induced oxidative damage. Eight days after doxycycline induction, APP-expressing induced neurons were treated with the various concentrations of hydrogen peroxide for 24 h and cells were fixed and stained for Tuj1 and cleaved caspase-3. We observed a decrease in the frequency of Tuj1+ cells and a concomitant increase in cleaved caspase-3+ apoptotic induced neurons upon APP overexpression at all concentrations of hydrogen peroxide (Supplementary Fig. 2D–F). Taken together, these data suggest that mouse induced neurons can faithfully recapitulate Alzheimer’s disease phenotypes at least in part as evidenced by the accumulation of amyloid-β₄₂, tau phosphorylation, abnormal endocytic function and susceptibility to oxidative stress-induced cell death. To examine whether APP-expressing induced neurons represent a tractable model for evaluating therapeutic candidates, we further assessed their response to pharmacological inhibition of the γ- and β-secretase complexes, which sequentially cleave APP into amyloid-β species. Indeed, inhibition of γ- and β-secretase complexes in APP-expressing induced neuron cultures significantly attenuated secretion of both amyloid-β₄₂ and amyloid-β₄₀, as well as the extracellular accumulation of amyloid-β₄₂ (Fig. 1F and Supplementary Fig. 2G and H). Furthermore, the expression of neuronal marker genes, Map2, Neurod1, Syn1, and Gad1 (GAD67) was partially restored in APP-expressing induced neurons treated with β-secretase inhibitors (Supplementary Fig. 2I). In addition, we observed that APP expression led to thioflavin T-positive deposits, and these deposits were significantly decreased by treatment with β-secretase inhibitors (Supplementary Fig. 2J–L), demonstrating that this model system can be used for testing therapeutic strategies against Alzheimer’s disease, as previously suggested (Frackowiak et al., 2003; Choi et al., 2014; Muratore et al., 2014).

Additionally, to determine whether the expression of the late onset Alzheimer’s disease risk factor APOE ɛ4 allele can directly affect APP processing and amyloid-β₄₂ generation in induced neurons, we next assessed the number of amyloid-β₄₂+ cells and the ratio of amyloid-β_42/40 in human APOE ɛ4-expressing mouse induced neurons (Fig. 1G–I). Importantly, we observed that APP/APOE ɛ4-expressing cultures had significantly increased numbers of amyloid-β₄₂ positive induced neurons and extracellular accumulation of amyloid-β₄₂ (Fig. 1G and H). Taken together, these results indicate that mouse induced neurons expressing mutant APP and/or APOE ɛ4 can represent a valid model for studying APOE ɛ4-associated sporadic Alzheimer’s disease pathogenesis and screening for potential therapeutics.

Modelling APOE ɛ3/4-associated sporadic Alzheimer’s disease in human induced neurons

Human induced neurons can be derived from fibroblasts using a specific combination of transcription factors (Vierbuchen et al., 2010; Ambasudhan et al., 2011; Pfisterer et al., 2011). We initially attempted to reprogram human fibroblasts into induced neurons using the previously reported factor combination ASCL1, BRN2, MYT1L and NEUROD1. However, in our laboratory, this combination of reprogramming factors leads to the inefficient generation of TUJ1+ human induced neurons, precluding the type of analysis performed on murine induced neurons presented above (Fig. 2A and B). Thus, we reasoned that additional variables need to be modified along with reprogramming factor expression in order to enhance the efficiency of generating human induced neurons. Recent evidence indicates that cell fate determination can be influenced by modulating biophysical cues such as surface nano-topography (Ravichandran et al., 2009; Pan et al., 2013; Solanki et al., 2013), and we previously found that nanoscale biophysical stimulation by nanopatterned substrates promotes highly efficient direct lineage reprogramming of mouse fibroblasts into induced functional neurons (Yoo et al., 2015). Thus, we examined whether human fibroblasts also can be more efficiently reprogrammed into the neuronal state using nano-topography, and subsequently whether resulting cells could serve as a model for human Alzheimer’s disease. To determine whether nano-topographical cues, when combined with ectopic factor expression, can improve reprogramming efficiency, we plated ASCL1, BRN2, MYT1L and NEUROD1-infected human fibroblasts on nano-grooved substrates, which were fabricated with a 400 nm groove width separated by 300 nm ridges (Supplementary Fig. 3A and B). Consistent with previous reports, reprogramming human fibroblasts on nanopatterned substrates resulted in apparent changes in cell alignment and elongation along the patterned surface (Supplementary Fig. 3C and F). In this context, direct lineage reprogramming on the nanopatterned substrates resulted in a greater number of cells taking on neuronal morphology and TUJ1 expression relative to non-patterned surfaces (Fig. 2A).

Figure 2

Efficient direct lineage conversion of human fibroblasts into human induced neurons on nanopatterned substrates. (A) Representative immunostaining showing the expression of TUJ1 (left) and compass plots (right) of ASCL1, BRN2, MYT1L and NEUROD1 derived human induced neurons (hiNs) on a non-patterned and a nanopatterned substrates in the absence and presence of APP expression. Scale bar = 100 µm. (B) The number of TUJ1+ human induced neurons on control and a nanopatterned substrates in the absence and presence of APP overexpression. Data represent mean ± SEM. ANOVA test, *P < 0.05; n = 6 per each sample. (C) The images show the MAP2 and VGLUT1 positive human induced neurons and APP expressing human induced neurons on the control and nanopatterned substrates at 8 days after doxycycline induction. Scale bar = 100 µm. (D) The percentage of VGLUT1/MAP2-positive cells on 8 days after doxycycline induction. Data represent mean ± SEM. (E) qRT-PCR analysis of neuronal markers (MAP2, CHAT, NEFL, and MAPT) at 8 days after doxycycline induction. Data represent mean ± SEM. ANOVA test, *P < 0.05, **P < 0.01; n = 3 per each sample. (F) Action potentials and sodium and potassium currents of human induced neurons and APP expressing human induced neurons on the nanopatterned substrates. n = 14 each group (G) Electrophysiological analysis of human induced neurons and APP expressing human induced neurons on the nanopatterned substrates. Data represent mean ± SEM, n = 12 each group. hFb = human fibroblasts.

Seven days after factor infection, doxycycline induction of the mutant APP lentivirus was used as in the murine system described above. Human induced neurons on the nanopatterned surface expressing mutant APP exhibited decreased TUJ1+ cell numbers relative to control human induced neurons (Fig. 2A and B). Importantly, in the absence of nanopatterning, these differences could not be observed due to the baseline inefficiency of the human induced neuron conversion, highlighting the importance of efficiently reprogrammed cultures for Alzheimer’s disease modelling (Fig. 2A and B). Consistent with the increased number of human induced neurons generated, immunocytochemistry conducted 15 days after reprogramming revealed that the nanopatterned substrates enable the expression of mature neuronal markers including MAP2 and VGLUT1 (Fig. 2C and D). Culture on nanopatterned substrates significantly increased the expression of pan-neuronal marker genes including MAP2, CHAT, NEFL, MAPT and the endogenous NEUROD1, and this expression was attenuated by induction of mutant APP (Fig. 2E). Additionally, to examine the reprogramming efficiency of human fibroblasts, we counted red fluorescent protein positive cells derived from cultures harbouring a synapsin-red fluorescent protein reporter by flow cytometry. Reprogramming for 15 days on nanopatterned substrates resulted in a 3-fold increase in the number of synapsin-red fluorescent protein positive human induced neurons, relative to control (Supplementary Fig. 3D). Next, we evaluated the morphological characteristics of control and APP-expressing induced neurons on nanopatterned substrates. Eight days after doxycycline induction, human fibroblasts developed into induced neurons with typical neuronal morphology, including dramatically decreased soma size and increased neurite length (Supplementary Fig. 3E). Importantly, we did not observe decreased soma size in APP-expressing induced neurons, and observed a decrease in neurite length on the nanopatterned substrates (Supplementary Fig. 3E). Electrophysiological recordings showed that both the control induced neurons and APP-expressing induced neurons (14 cells each group) cultured on nanopatterned substrates elicit action potentials as well as sodium and potassium currents (Fig. 2F). Electrophysiological characteristics of human induced neurons on nanopatterned substrates including resting membrane potential, input resistance, and action potential amplitude indicates that these cells are highly functionally similar to bona fide primary neurons (Fig. 2G). Taken together, these data indicate that human induced neurons on nanopatterned substrates exhibit well developed and functional neuronal characteristics, and APP-expressing human induced neurons can be used for Alzheimer’s disease modelling.

Next, we applied our human induced neuron conversion protocol to fibroblasts derived patients with sporadic forms of Alzheimer’s disease and examined sporadic Alzheimer’s disease-associated pathology in these induced neurons. We prepared human fibroblasts from both healthy control in APOE ɛ3/3 isoform genotype and sporadic Alzheimer’s disease patient harbouring an Arg112 and Arg158 in APOE ɛ3/4 isoform genotype (Patient AG11414). This APOE ɛ3/4 allele is a major risk factor for sporadic Alzheimer’s disease and is associated with earlier onset of Alzheimer’s disease and increased amyloid plaques (Kim et al., 2009; Duan et al., 2014).

As in previous experiments, we induced reprogramming of sporadic Alzheimer’s disease patient fibroblasts on the nanopatterned substrates after infection. The human induced neuron cultures on nanopatterned substrates aligned with groove direction and exhibited an increase in neuronal morphology and reprogramming efficiency assessed by the per cent of TUJ1+ and MAP2+ cells (Supplementary Fig. 3F and G). We did not observe significant differences in the number of TUJ1+/MAP2+ induced neurons between control (APOE ɛ3/3 allele) and patient (APOE ɛ3/4 allele)-derived cultures (Supplementary Fig. 3F and G). In addition, we did not observe significant changes in the number of TUJ1+ cells in long-term cultures (40 days, data not shown), suggesting that APOE ɛ3/4 patient induced neurons do not exhibit the Alzheimer’s disease-associated degeneration in culture in the absence of other perturbation. However, we did observe a modest decrease in the number of TUJ1+/MAP2+ induced neurons in APP-expressing APOE ɛ3/4 patient induced neurons (Supplementary Fig. 3F and G). Remarkably, sporadic Alzheimer’s disease patient induced neurons and APP-expressing induced neurons cultured on nanopatterned substrates exhibited accumulation of amyloid-β polymers (Fig. 3A), suggesting that efficient reprogramming of APOE ɛ3/4 Alzheimer’s disease patient induced neurons on the nanopatterned substrates is critical for the generation of Alzheimer’s disease phenotypes in the induced neurons. Moreover, APP-expressing APOE ɛ3/4 patient induced neurons show a greater accumulation of amyloid-β polymers (Fig. 3A), demonstrating this system can be used as a model for Alzheimer’s disease. Additionally, we observed significant increases in amyloid-β₄₂-positive cells and tau phosphorylation in both APP-expressing APOE ɛ3/3 human induced neurons and APOE ɛ3/4 patient induced neurons 10 days after doxycycline induction (Fig. 3B, D and E). Furthermore, the expression of neuronal genes was significantly decreased in mutant APP-expressing APOE ɛ3/4 patient induced neurons on nanopatterned substrates (Fig. 3C).

Figure 3

Pathological analysis of APOE ɛ3/4 patient induced neurons on a nanopatterned substrate. (A) Western blot analysis shows increase of amyloid-β oligomers in normal human fibroblasts and APOE ɛ3/4 patient, and APP expressing human induced neurons on nanopatterned substrates at 10 days after doxycycline induction. (B) Immunostaining of TUJ1 and amyloid-β₄₂ in the no intracellular amyloid-β₄₂ accumulation in human induced neurons on non-pattern, normal human fibroblasts (APOE ɛ3/3 genotype) and sporadic Alzheimer’s disease patient (APOE ɛ3/4 genotype) derived human induced neurons on nanopatterned substrates. Scale bar = 100 µm. (C) Expression profiling using qRT-PCR analysis of patient fibroblasts, APP expressing fibroblasts, APOE ɛ3/4 patient induced neurons, APP expressing APOE ɛ3/4 patient induced neurons on the control (yellow) and nanopatterned (purple) substrates. Ratios of expression differences are shown in different colours. Red and blue represent higher and lower gene expression levels; n = 6 per each sample. (D) The percentage of amyloid-β₄₂/TUJ1 positive APOE ɛ3/4 patient human induced neurons 10 days after doxycycline induction. Data represent mean ± SEM. (E) The percentage of p-tau/MAP2-positive APOE ɛ3/4 patient human induced neurons on 10 days after doxycycline induction. Data represent mean ± SEM. (F) Decrease in amyloid-β₄₂+/TUJ1 positive APOE ɛ3/4 patient induced neurons and APP expressing APOE ɛ3/4 patient induced neurons on a nano-grooved pattern by β-secretase inhibitor and γ-secretase inhibitor treatment. Data represent mean ± SEM. ANOVA test, *P < 0.05, **P < 0.01; n = 6 per each sample. (G) Survival rate of hydrogen peroxide treated APOE ɛ3/4 patient fibroblasts, APOE ɛ3/4 patient induced neurons, APP expressing APOE ɛ3/4 patient induced neurons on nanopatterned substrates. Data represent mean ± SEM. Student’s t-test, *P < 0.05, **P < 0.01; n = 6 per each sample. SAD = sporadic Alzheimer’s disease.

Consistent with these results, localization of APP-positive puncta in early endosomes was significantly increased in APP-expressing patient induced neurons as confirmed by containing with early endosome associated antigen-1, EEA1 (Supplementary Fig. 4A and B). As expected, BACE1 transcript levels were upregulated in nanopattern-derived patient induced neurons (Supplementary Fig. 4C), but did not appear altered upon mutant APP expression (data not shown), suggesting that the abnormal accumulation of amyloid-β₄₂ in patient induced neurons is not caused by increased activity of BACE1. Additionally, we examined the pharmacological response to inhibitors in these patient induced neurons. Eight days after doxycycline induction, patient induced neurons were treated by γ- and β-secretase inhibitors. Consistent with results in murine cultures, inhibition of γ- and β-secretase complexes significantly attenuated the number of amyloid-β₄₂ positive cells in the APP-expressing patient induced neurons (Fig. 3F, Supplementary Fig. 4D and E). Finally, we tested for susceptibility of APP-expressing patient induced neurons to hydrogen peroxide. Patient induced neurons were treated with 0–1.0 mM of hydrogen peroxide and survival was evaluated after 24 h. APOE ɛ3/4 patient induced neurons progressively degenerated within 24 h at 0.2–1.0 mM, and APP-overexpressing APOE ɛ3/4 patient induced neurons were more susceptible to peroxide-induced cell death (Fig. 3G). Taken together, these results suggest that APOE ɛ3/4 Alzheimer’s disease patient-derived induced neurons represent a valid model for studying sporadic Alzheimer’s disease pathogenesis, and provide a novel strategy of screening of candidate therapeutics.

Next, to gain insight into the molecular mechanism of APOE ɛ3/4-associated sporadic Alzheimer’s disease pathologies in induced neurons, we compared the global gene expression profiles of APOE ɛ3/3 healthy control and APOE ɛ3/4 patient-derived induced neurons in the presence and absence of APP expression. APOE ɛ3/4 sporadic Alzheimer’s disease patient induced neurons expressing mutant APP exhibited dramatic changes in global gene expression (Fig. 4A). We found that 1500 genes were upregulated in the APOE ɛ3/4 Alzheimer’s disease patient induced neurons, and 1298 genes were upregulated in APP-expressing APOE ɛ3/4 patient induced neurons compared to control APOE ɛ3/3 induced neurons (Fig. 4A and Supplementary Fig. 5A), whereas 1015 and 1383 genes were downregulated in the APOE ɛ3/4 patient induced neurons, and APP-expressing APOE ɛ3/4 patient induced neurons relative to control APOE ɛ3/3 induced neurons, respectively (Fig. 4A and Supplementary Fig. 5A). We also found that 314 upregulated and 196 downregulated genes were common between APOE ɛ 3/4 patient induced neurons and APP-expressing APOE ɛ3/4 patient induced neurons, indicating these are specific genes affected by the APOE ɛ3/4 genotypes (Fig. 4A and Supplementary Fig. 5A).

Figure 4

Global gene expression analysis of APOE ɛ3/4 patient (AG11414)-derived induced neurons and APOE ɛ3/3 control induced neurons. (A) Venn diagram showing the overlap of differentially expressed genes between sporadic Alzheimer’s disease (APOE ɛ3/4) induced neurons (iNs) per APOE ɛ3/3 induced neurons and sporadic Alzheimer’s disease (APOE ɛ3/4) induced neurons + mutAPP per APOE ɛ3/3 induced neurons + mutAPP. Number of 1.5 upregulated (top) and 1.5 downregulated (middle) gene are represented by specific group colour (bottom). (B) Heatmap showing microarray expression of differentially expressed genes. The column on the left represents specific group colour in A. (C) Scatter plot of the microarray expression data between APOE ɛ3/4 induced neurons and APOE ɛ3/3 induced neurons (left) and APOE ɛ3/4 induced neurons + mutAPP and APOE ɛ3/3 induced neurons + mutAPP (right). Up- and downregulated genes are presented as specific group colour based on A. Ellipses indicate 2-fold up- or 2-fold downregulated genes. (D) Gene set enrichment analysis of the microarray data from APOE ɛ3/4 induced neurons per APOE ɛ3/3 induced neurons. MSigDB C2 gene sets are used. (E) Gene set enrichment analysis of the microarray data from Alzheimer’s disease patient (APOE ɛ3/4 genotype) datasets in human post-central gyrus (GSE48350). Sets of the upregulated (fold change ≥ 2) genes based on APOE ɛ3/4 induced neurons per APOE ɛ3/3 induced neurons are used. The APOE ɛ3/4 patient post-central gyrus datasets were enriched in upregulated overlap genes between APOE ɛ3/4 induced neurons and APP-expressing APOE ɛ3/4 induced neurons. SAD = sporadic Alzheimer’s disease.

Additionally, when the distribution of expression changes of all genes was compared, we found that the expression patterns of human APOE ɛ3/4 induced neurons was significantly different from control APOE ɛ3/3 induced neurons, and that the changes in gene expression induced by mutant APP expression were not as significant as those conferred by the APOE ɛ3/4 genotype (Fig. 4B). Moreover, among the differentially expressed genes in APOE ɛ3/4 induced neuron cultures were genes known to be associated with APP processing including BACE2, CLU, DSG2, PLAU and MME (Fig. 4C), suggesting that the differential expression patterns of these genes may be associated with the development of APOE ɛ3/4-associated sporadic Alzheimer’s disease.

Next, to examine the extent to which APOE ɛ3/4 sporadic Alzheimer’s disease patient-derived induced neurons molecularly resemble primary patient neurons, we compared gene expression patterns between APOE ɛ3/4 Alzheimer’s disease induced neurons and APOE ɛ3/4 Alzheimer’s disease patient post-mortem brain (Curated gene sets in Molecular Signatures Database, MSigDB v5.1). Remarkably, gene set enrichment analysis indicates upregulated genes in APOE ɛ3/4 Alzheimer’s disease patient induced neurons relative to control APOE ɛ3/3 induced neurons were enriched in primary incipient APOE ɛ3/4 Alzheimer’s disease patient brain tissue (Fig. 4D, left) and APOE ɛ3/4 Alzheimer’s disease patient brain tissue gene sets (Fig. 4D, right). Moreover, genes differentially expressed in the APOE ɛ3/4 Alzheimer’s disease patient post-central gyrus were significantly enriched for differentially expressed genes from APOE ɛ3/4 induced neurons versus control APOE ɛ3/3 induced neurons (Fig. 4E, left) as well as for the genes whose differential expression was common between APOE ɛ3/4 induced neurons and APP-expressing APOE ɛ3/4 induced neurons (Fig. 4E, right), suggesting that APOE ɛ3/4 Alzheimer’s disease patient-derived induced neurons faithfully recapitulates the molecular pathology in APOE ɛ3/4-associated sporadic Alzheimer’s disease.

To gain insight into the molecular mechanism by which the APOE ɛ3/4 genotype promotes the development of sporadic Alzheimer’s disease, we combined the differentially expressed genes between APOE ɛ3/4 patient induced neurons and control APOE ɛ3/3 induced neurons with the human protein-protein interaction HTRIdb database and analysed networks with APOE ɛ3/4 specific genes (Fig. 5A and Supplementary Fig. 5B). The target nodes and edges of networks are depicted with the proteins encoded by differential expressed genes and their interactions. Surprisingly, we found that the major sub-network from the highest molecular complex detection analysis was a DSG2-associated complex (Fig. 5B, C and Supplementary Fig. 5C), which is a known Alzheimer’s disease genetic risk factor (Karch and Goate, 2015). This network analysis suggests that DSG2 may be functionally important in the development of APOE ɛ3/4-associated sporadic Alzheimer’s disease.

Figure 5

Protein–protein interaction networks in sporadic Alzheimer’s disease APOE ɛ3/4 patient-derived induced neurons. (A) Protein–protein interaction network. A subnetwork showing direct interaction with proteins encoded by upregulated overlap genes (DSG2, BACE2, CLU, CD2AP) and downregulated overlap genes (APOE, BIN1, PLAU, MME) in Fig. 4A and C. Node or edge size indicate degrees or edge-between, respectively. The ellipse-shaped nodes indicate proteins and rectangle shaped nodes represent proteins that have fold-change ≥ 2 in APOE ɛ3/4 induced neurons per APOE ɛ3/3 induced neurons and APOE ɛ3/4 induced neurons + mutAPP per APOE ɛ3/3 induced neurons + mutAPP. Node border colour represents specific group colour as shown in Fig. 4A. (B and C) Molecular Complex Detection analysis results from A showing 3.6 (B) and 3.5 (C) molecular complex detection score. (D) Validation of gene expression between microarray data and qRT-PCR. (E) Validation of APOE genotypes in human fibroblasts and Alzheimer’s disease patient fibroblasts. Mixture A includes genotype primers against Cys112 (558 bp) and Cys158 (451 bp). Mixture B includes genotype primers against Arg112 (588 bp) and Arg158 (451 bp). (F) qRT-PCR analysis of DSG2 expression in human fibroblasts, human induced neurons and APP expressing human induced neurons about control fibrobalst, AG05810, AG11414, AG05770, AG06840 and AG09908. hFb = human fibroblasts; FAD = familial Alzheier’s disease; iNs = induced neurons; SAD = sporadic Alzheimer’s disease.

To investigate the functional role of DSG2 in APOE ɛ3/4-associated sporadic Alzheimer’s disease further, we initially confirmed the differential expression of DSG2 using qRT-PCR and microarray-based transcriptome profiles. Remarkably, we found that DSG2 was the most upregulated gene in APOE ɛ3/4 patient induced neurons and APP-expressing APOE ɛ3/4 patient induced neurons compared to control induced neurons with APOE ɛ3/3 genotype (Fig. 5D). Next, we prepared additional human fibroblasts from a control with APOE ɛ3/3 genotypes, familial patients with Alzheimer’s disease with PSEN mutations, sporadic Alzheimer’s disease with APOE ɛ3/4 genotypes, and idiopathic, sporadic Alzheimer’s disease (Fig. 5E and Supplementary Fig. 5D). Interestingly, when human induced neurons were generated by ASCL1, BRN2, MYT1L and NEUROD1 factor expression, DSG2 was specifically upregulated in the induced neurons derived from sporadic Alzheimer’s disease fibroblasts with APOE ɛ3/4 genotypes, relative to induced neurons derived from familial PSEN mutant fibroblasts, idiopathic sporadic Alzheimer’s disease or controls with APOE ɛ3/3 genotype, while other genes were not affected (Fig. 5F and Supplementary Fig. 5E).

To test whether DSG2 functionally contributes to APOE ɛ3/4-induced sporadic Alzheimer’s disease phenotypes, we examined amyloid-β aggregation in APOE ɛ3/4 patient induced neurons upon DSG2 inhibition. Remarkably, we found that the increase in amyloid-β aggregation caused by APP overexpression in APOE ɛ3/4 patient induced neurons was effectively rescued by knockdown of DSG2 (Fig. 6A–C). Consistent with this result, the number of amyloid-β₄₂+ cells in the APP-expressing APOE ɛ3/4 induced neurons decreased upon DSG2 knockdown, while expression of neuronal genes was not affected (Fig. 6D and Supplementary Fig. 6A). Moreover, these phenotypes were specific to the APOE ɛ3/4-associated Alzheimer’s disease patient induced neurons and were absent from PSEN patient induced neurons (Supplementary Fig. 6B–D). We further demonstrated that accumulation of amyloid-β polymers in APP-expressing APOE ɛ3/4 patient induced neurons was significantly decreased by DSG2 inhibition (Fig. 5E and F). Consistent with this, overexpression of DSG2 in these APP-expressing APOE ɛ3/4 induced neurons increased amyloid-β aggregation (Supplementary Fig. 7A–E). Taken together, these data suggest that DSG2 mediates the APOE ɛ3/4-associated amyloid-β aggregation phenotypes in human induced neurons. Additionally, we found that the number of amyloid-β-positive puncta in APP-expressing APOE ɛ3/3 control induced neurons and APOE ɛ3/4 patient induced neurons are not affected by inhibition of β-secretase complexes (Supplementary Fig. 7A–E), suggesting that DSG2 is not their downstream target in APP processing.

Figure 6

Knockdown of DSG2 decreases amyloid-β peptides in APOE ɛ3/4 patient cell lines and APP expressing APOE ɛ4 mouse induced neurons. (A) Immunofluorescence of VGLUT1 and amyloid-β (6E10) in sporadic Alzheimer’s disease patient (APOE ɛ3/4 genotype) APP expressing induced neurons (iNs) and DSG2 knockdown. Scale bar = 10 µm. (B) Analysis of relative amyloid puncta (left) and fluorescence intensity (right) in APP expressing APOE ɛ3/4 patient induced neurons and APP expressing APOE ɛ3/4 patient induced neurons treated with DSG2-shRNA. Data represent mean ± SEM. Student’s t-test, *P < 0.05, **P < 0.01; n = 6 per sample. (C) Number of amyloid puncta per cell was decreased in APOE ɛ3/4 patient cells (AG05810 and AG11414), but APOE ɛ3/3 patient cells do not. Data represent mean ± SEM. Student’s t-test, *P < 0.05, **P < 0.01; n = 6 per each sample. (D) Analysis of amyloid-β₄₂/TUJ1 (top) and P-TAU/TUJ1 (bottom) -positive cells in patient induced neurons and APP expressing patient induced neurons treated with DSG2-shRNA about control human cell line and five patient cell lines. Data represent mean ± SEM. Student’s t-test, *P < 0.05, **P < 0.01; n = 6 per each sample. (E and F) Western blot analysis for APP and amyloid-β (6E10) shows reduction of amyloid-β oligomers in APP expressing sporadic Alzheimer’s disease patient (APOE ɛ3/4 genotype) induced neurons treated with DSG2-shRNA. Data represent mean ± SEM. Student’s t-test, *P < 0.05. (G) Immunostaining analysis of the intracellular accumulation of amyloid-β₄₂ in APP and human APOE (hAPOE) ɛ4 expressing mouse induced neurons. Scale bar = 100 µm. (H) The percentage of amyloid-β₄₂/TUJ1 positive APP expressing Apoe ɛ4 mouse induced neurons on 10 days after doxycycline induction. Data represent mean ± SEM. Student’s t-test, *P < 0.05, **P < 0.01 n = 6 per each sample. (I) ELISA analysis of the intracellular amyloid-β₄₂ secreted from APP expressing mouse induced neurons and APP expressing APOE ɛ4 mouse induced neurons treated with shDSG2. Data represent mean ± SEM. ANOVA test, *P < 0.05, **P < 0.01; n = 6 per each sample. SAD = sporadic Alzheimer’s disease.

Finally, we examined functional effects of DSG2 in the human APOE ɛ4- and APP-expressing mouse induced neurons. Consistent with the human data, elevated amyloid-β₄₂ secretion in Apoe ɛ4- and App-expressing mouse induced neurons was significantly decreased by Dsg2 knockdown (Fig. 6G–I). Moreover, we confirmed that transfection of Dsg2 knockdown plasmids with GFP tags into mouse induced neurons led to a reduction in amyloid-β₄₂ protein (Supplementary Fig. 8A and B), demonstrating a functional connection between APOE ɛ3/4 and DSG2, which may be critical for the control of APP processing in APOE ɛ3/4-associated sporadic Alzheimer’s disease.

Discussion

Direct lineage reprogramming of fibroblasts into neurons by overexpressing key transcription factors is a conceptually intriguing strategy for modelling neurological diseases, since this technique has proved to be a rapid, robust and reproducible way to generate functional neurons (Vierbuchen et al., 2010; Caiazzo et al., 2011; Kim et al., 2011; Wapinski et al., 2013). Particularly, the generation of human induced neurons from patient fibroblasts represents a powerful tool for understanding pathological features of neurological diseases and screening for therapeutic interventions (Iovino et al., 2014). However, the process of direct conversion of human fibroblasts into induced neurons by the transduction of transcription factors is highly inefficient, making it difficult to reproducibly and quantitatively characterize disease phenotypes in these cultures. Recently, a study reported the generation of chemically induced directly converted neurons from familial Alzheimer’s disease patient fibroblasts that developed Alzheimer’s disease pathological phenotypes (Hu et al., 2015). To date, none of these studies have reported direct conversion of human induced neurons from sporadic patients with Alzheimer’s disease. Thus, our study for the first time demonstrates the successful formation of sporadic Alzheimer’s disease patient induced neurons with high efficiency using nanopatterned substrates. We further demonstrate the application of this culture platform in the discovery of novel pathogenic mechanisms that contribute to idiopathic sporadic Alzheimer’s disease.

In the present study, we demonstrate that induced neurons derived from fibroblasts of mice, normal humans and patients with Alzheimer’s disease can be used to model Alzheimer’s disease. Murine induced neurons generated by the forced expression of Ascl1, Brn2, and Myt1l uniformly exhibit morphological and molecular features of glutamatergic forebrain neurons, and overexpression of mutant APP in these induced neurons recapitulates several pathological phenotypes associated with Alzheimer’s disease, including abnormal accumulation of amyloid-β₄₂ and altered tau protein phosphorylation. Notably, APP expressing human induced neurons recapitulated several abnormal Alzheimer’s disease phenotypes, and these pathologies were reproduced in both mutant APP-expressing APOE ɛ3/4 patient induced neurons, suggesting that these Alzheimer’s disease pathologies can be reproduced in directly converted human induced neurons. Consistent with this, these APOE ɛ3/4 patient induced neurons also exhibited vulnerability to peroxide-induced cell death. This is particularly interesting because the fibroblasts were derived from a patient with major Alzheimer’s disease risk factor, APOE ɛ3/4 genotype, and the Alzheimer’s disease pathological phenotype in APOE ɛ3/4 Alzheimer’s disease patient-derived induced neurons demonstrate that characteristics of Alzheimer’s disease are retained in the directly reprogrammed neurons. More importantly, a significant number of genes that were differentially expressed in APOE ɛ3/4 Alzheimer’s disease patient induced neurons was also enriched in the sporadic Alzheimer’s disease patient brain database, suggesting that this strategy can be used for identifying new disease interventions and for drug screening for Alzheimer’s disease.

Remarkably, we identify a direct functional connection between APOE ɛ3/4 and another Alzheimer’s disease susceptibility locus, which might contribute to the disease phenotypes in APOE ɛ3/4 sporadic patients with Alzheimer’s disease. Analysis of the subnetworks within differentially expressed genes from APOE ɛ3/4 patient-derived induced neurons revealed, that the DSG2 complex is closely related to the APOE ɛ3/4-induced Alzheimer’s disease phenotypes. While it has been recently reported that the polymorphisms in DSG2 are associated with sporadic Alzheimer’s disease through gene-wide association study, the functional connection between DSG2 and APOE ɛ3/4 was not previously known. We further demonstrated that this functional connection between APOE ɛ3/4 and DSG2 was specific for APOE ɛ3/4-associated sporadic Alzheimer’s disease by examining Alzheimer’s disease pathological phenotypes in healthy normal and other Alzheimer’s disease patient-derived induced neuron lines. This is the first report of a functional connection between genetic loci linked to sporadic Alzheimer’s disease. The identification of functional connections between Alzheimer’s disease disease-linked genes provides important insight into the pathophysiology of Alzheimer’s disease development and highlights cellular pathways upon which these gene products may converge. These pathways may represent promising therapeutic targets for the treatment of Alzheimer’s disease. Thus, our study elucidating the mechanisms affected by genetic variants identified by gene-wide association study using induced neurons represents a novel approach for the development of personalized medicine for more effective treatments of sporadic Alzheimer’s disease.

Funding

This work was supported by the National Research Foundation funded by the Korea government (NRF-2017M3A9C6029306, 2016R1A2B2014195, 2015R1A2A1A01003530, 2015M3A9C7030128), Korea Health Technology R&D Project, Ministry of Health & Welfare (HI16C1176), the Next-Generation BioGreen 21 Program, Rural Development Administration (PJ0110770) and the Ministry of Food and Drug Safety in 2017 (14172MFDS974).

Supplementary material

Supplementary material is available at Brain online.