Introduction

Alzheimer’s disease (AD) is one of the most common forms of neurodegeneration and accounts for approximately 70% of all cases of dementia. The key pathological hallmarks of AD are extracellular senile plaques composed of neurotoxic amyloid-beta (Aβ) peptide and intracellular neurofibrillary tangles (NFT) composed of hyperphosphorylated tau [1]. Clinical studies show that the most critical risk factors for AD are aging and a family history of the disease [2]. Less than 1% of AD cases occur in families where the disease is inherited in a fully penetrant, autosomal dominant manner, with early (<65) onset. Three early-onset familial AD (FAD) genes have been identified to date: the amyloid precursor protein (APP) gene, the presenilin 1 (PSEN1) gene, and the presenilin 2 (PSEN2) gene. Together, these genes can harbor >400 mutations (https://www.alzforum.org/mutations) that account for roughly 5–10% of autosomal dominant FAD.

Current models to study AD include cell cultures and animal models. Human diseases, however, are often poorly reproduced in animal models [3]. The development in the past decade of techniques to generate human-induced pluripotent stem cells (iPSCs) and differentiate them into different brain cell types, together with advances in genome editing, has provided a novel model to study AD and other neurodegenerative disorders [4, 5]. Neural cultures derived from patients or isogenic mutant iPSCs recapitulate at least some of the phenotypes observed in the AD brain [4,5,6]. However, some AD-related phenotypes cannot be reproduced in conventional two-dimensional (2D) cultures as they lack the brain three-dimensional (3D) structure. For instance, removing secreted Aβ with media changes most likely prevents the deposition of amyloid aggregates in 2D cultures [5]. To overcome these limitations, iPSC-derived 3D cultures have been generated to model AD (for review, see [5]). We developed a 3D bioengineered model of iPSC-derived neural tissue that combines a porous scaffold composed of silk fibroin protein with an intercalated collagen hydrogel to support the growth of neurons and glial cells into complex and functional networks for an extended time [7, 8]. The architecture of the scaffolds was optimized to meet the metabolic demand of high-density cultures in terms of free diffusion of nutrients and oxygen, a fundamental requisite for long-term cultures and aging-related studies.

In this study, we have optimized our protocol by seeding neural precursor cells (NPCs) instead of iPSCs. This improved protocol offers several advantages: NPCs can be expanded, frozen, banked, and subsequently differentiated. Having NPC stocks derived from multiple subjects allows synchronization of their differentiation and minimizes experimental variability. We generated NPC-derived cultures from iPSC lines obtained from two subjects carrying the FAD mutation APP V717I (also known as London), hFAD-1, and hFAD-2 (Clone A and B) [9]. We used two well-studied lines, YZ1 and BR24 [6, 9], as controls. Furthermore, we generated the isogenic control of the hFAD-2 Clone B line by correcting the APP V717I mutation to wild type (C-hFAD-2 Clone B). FAD and control cultures were analyzed at two time points (2 and 4.5 months).

An elevated Aβ42/40 ratio was detected in FAD cultures at both time points, as previously reported in 2D cultures derived from the same FAD lines [9]. Additionally, extracellular Aβ42 deposition and enhanced neuronal excitability were observed in FAD cultures at 4.5 months. Notably, neuronal hyperexcitability has been described in FAD and late-onset AD (LOAD) patients at the early stage of the disease [10]. Gene expression analysis was performed to identify the specific target genes which exhibit significantly increased or decreased expression in FAD lines compared to controls at both time points. Multiple gene sets were deregulated in FAD samples at 2 months. The difference was maintained for up to 4.5 months. Among these pathways, several are implicated in neurotransmitter release/reuptake and synthesis/storage, neural connectivity, structural stability of the cytoskeleton, axonal and dendrite structure, and vesicle trafficking regulation. Such alterations in gene expression were similar to those observed in human AD brains in a large study that performed a co-expression meta-analysis of harmonized Accelerating Medicines Partnership for Alzheimer’s Disease (AMP-AD) data across three independent cohorts [11]. These data provide compelling evidence that our 3D cultures reproduce some time-dependent phenotypes and transcriptomic features observed in AD patients. Thus, our bioengineered neural tissue represents a valuable model for studying AD-related pathomechanisms over time.

Methods

iPSC lines

hFAD-1, hFAD-2 Clone A and B iPSC lines [9] were generated in Dr. T. Young-Pearse’s laboratory (Brigham and Women’s Hospital, Boston, MA, USA). BR24 iPSC line (ROSMAP) [6, 12] was generated at the New York Stem Cell Foundation. YZ1 iPSC line was obtained from the University of Connecticut-Wesleyan Stem Cell Core (Farmington, CT, USA [9]). The manuscripts cited above describe the reprogramming methods and quality control assays for these lines.

hFAD-2 Clone B Corrected was generated at the University of Connecticut-Wesleyan Stem Cell Core. The guide RNA used for APP editing was designed using chop-chop (https://chopchop.cbu.uib.no/) for specificity to the loci of interest. The resulting guide 5′ GACAGTGATCATCATCACCT 3′ was identified to have high specificity to the desired loci. Potential off-targets were identified using the Cas-OFF finder (http://www.rgenome.net/cas-offinder/). The top eight identified sites were amplified from the clones, and Sanger sequenced to verify the status of these potential off-targets. None of the predicted sites showed evidence of off-targeting cutting. All iPSC lines have been confirmed by sequencing before starting any of the experiments listed in this manuscript.

ROSMAP study participants

iPSC lines were generated from autopsied participants in the Religious Orders Study or Rush Memory and Aging Project (ROSMAP) [12]. Both studies were approved by an Institutional Review Board of Rush University Medical Center; they enroll older persons without dementia who agree to annual clinical evaluation and brain donation at death. All participants signed informed consent, an Anatomical Gift Act, and a repository consent to repurpose their resources. Peripheral blood mononuclear cells were drawn and cryopreserved as part of the annual evaluation.

Karyotype analysis and mycoplasma testing

G-banded karyotyping was outsourced to WiCell (Madison, WI, USA). Karyotype analysis was performed every ten passages.

Mycoplasma testing was performed monthly for all the lines included in this study at any differentiation stage (LookOut Mycoplasma PCR Detection Kit, Sigma). All cells used in this study were consistently negative.

NPC differentiation

NPCs were differentiated using an EB-based protocol without SMADi supported by STEMdiff Neural Induction medium (StemCell Technologies) over ~20 days according to the StemCell Technologies method. To generate iPSC-derived EBs for neural induction, AggreWells 800 (StemCell Technologies) were used. Single-cell suspensions of NPCs were cryopreserved and banked for future usage using a Neural Progenitor Freezing medium (StemCell Technologies). At least two separate differentiations of NPCs have been generated from different iPSC stocks for all the lines included in this study and used to differentiate hNeurons.

Scaffold preparation, seeding, and differentiation into hNeurons

Silk protein was processed from Bombyx mori cocoons, and scaffolds were prepared as described previously [7, 13].

A strained (40 μm nylon mesh) single-cell suspension of (2 × 106/100 μL) NPCs between passages two and six was added to the coated scaffold (poly-L-ornithine 20 μg/mL, laminin 10 μg/mL, Sigma) in a 96-well plate and incubated for 24 h. The neural progenitor medium was changed daily for five days following seeding, moving the scaffolds to fresh wells with each change. After allowing five days of cellular expansion, the scaffolds were moved to a clean well and infused with a 100 μL solution of cold Collagen Type-1 Rat Tail (3.0 mg/mL, Corning) mixed with 10× PBS and 1N NaOH (ratio 88:10:2). The collagen-filled scaffolds were then incubated at 37 °C until the gelation occurred (~45 min). Once solidified, the scaffolds were transferred to a 24-well plate and flooded with 1.5 mL of BrainPhys media (StemCell Technologies) supplemented with SM1 (StemCell Technologies) and N2 (ThermoFisher) with media changes every four days and differentiated into hNeurons.

Immunofluorescence and imaging analysis

2D and 3D cultures were fixed with a 4% paraformaldehyde/4% sucrose solution in DPBS for 15 and 30 min, respectively, followed by membrane permeabilization with 0.3% Triton X-100 for 15 min. Non-specific binding sites were blocked for one (2D) or four (3D) hours at RT (5% BSA, 0.1% Triton X-100 in DPBS). Cultures were then incubated with primary antibodies (Table S1) diluted in 1% BSA, 0.1% Triton X-100 in DPBS overnight at 4 °C. After multiple washes in DPBS, sections were incubated with Alexa (1:1000 in 1% BSA in DPBS) fluorescent secondary antibodies against the corresponding host species for one (2D) or three (3D) hours at RT. Nuclei were stained with DAPI (Sigma). 2D cultures were mounted on superfrost slides (Fisher Scientific) using Fluoromount-G1 mounting media (Southern Biotech), while 3D cultures were imaged directly.

NeuroFluor™ NeuO (StemCell Technologies) was used to stain live cultures according to manufacturer instructions.

Z-stack fluorescence images were obtained using confocal microscopes (SP8 Falcon Leica, SP8 Leica, see figure legends for details), keeping constant imaging parameters. 3D reconstructions were generated with the LAS X Life Science Leica software. Image analysis was performed using Fiji ImageJ version: 2.0.0-rc-69/1.52p.

Aβ burden volumetric analysis was performed using the Fiji ImageJ plug-in 3D object counter. The sum of all extracellular Aβ deposition volumes was divided by the total volume of the region of interest (ROI) imaged to obtain the amyloid burden values. At least two sets of cultures per line and three images per set of cultures have been analyzed.

MSD assay and data analysis

Before harvest, BrainPhys media was changed, and fresh media was dispensed into each well. Twenty-four hours later, media was collected and stored at −80 °C. MSD (Rockville, MD, USA) assay for the detection of human Aβ38, Aβ40, and Aβ42 (V-PLEX Aβ Peptide Panel 1 (4G8) Kit) was performed in-house according to the manufacturer’s instructions on undiluted conditioned media. Technical duplicates were run for all the samples analyzed. The data were analyzed using MSD Workbench 4.0 software.

Aβ oligomers immunoprecipitation/western blot

Aβ oligomers (oAβs) were visualized using a highly sensitive immunoprecipitation/western blot (IP/WB) protocol that can readily detect as little as 200 pg of naturally secreted human Aβ [14]. Briefly, the culture media were conditioned for 24 h and cleared of cell debris by centrifugation. Protease and phosphatase inhibitors were then added to the supernatant. To minimize the detection of cross-reactive proteins that nonspecifically bind to protein A/G, samples were incubated with protein A/G agarose beads (Protein A/G PLUS-Agarose, Santa Cruz) for 1 h at 4 °C with gentle shaking. Pre-cleared supernatants were collected after centrifugation and incubated with the IP antibodies plus 45 μL of protein A/G for 1 h at room temperature with gentle shaking. Agarose beads were collected after centrifugation and washed with STEN buffer as previously described [14], then heated at 100 °C for 10 min in western blot loading buffer and centrifuged at 15,000 × g for 5 min. All retrievable liquid was collected and used for SDS-Polyacrylamide Gel Electrophoresis (SDS-PAGE) using 4–12% Bis-Tris mini gels with MES buffer (ThermoFisher). Proteins were transferred onto 0.2 mm nitrocellulose at 200 mA for 3 h, and the membrane was boiled in PBS for 10 min. Non-specific bindings were blocked in 5% non-fat dried milk in TBST. Primary antibodies (Table S1) were incubated overnight at 4 °C, detected with the species-specific HRP-conjugated secondary antibody, and then visualized by ECL. A chemiluminescent signal was captured on a LAS4000 Fuji Imager.

LFP recordings

Local field potentials (LFPs) were recorded from samples as described previously [8, 15]. Recordings were performed blind, always counterbalanced by condition, and replicated batches were measured within the same 12-h window. All data were normalized to DNA content quantified using Quant-iT PicoGreen dsDNA Assay kit (ThermoFisher). Data were also subjected to a log-10 transformation to stabilize and normalize expected trial-to-trial variation.

RNA isolation

At the time of desired analysis, scaffolds were snap-frozen using liquid nitrogen and stored at −80 °C. The frozen scaffolds were then homogenized using a liquid nitrogen-chilled Spectrum Bessman Tissue Pulverizer (Fisher Scientific). RNA from homogenized samples was extracted immediately after the homogenization using the RNeasy Plus Mini kit (Qiagen) following the manufacturer’s instructions.

Tau RT-PCR

After RNA isolation, cDNA was generated using the iScript Advanced cDNA Synthesis Kit (Biorad), and relative levels of 3-repeat (3R) and 4-repeat (4R) tau mRNAs were determined by RT-PCR using the following primers: forward 5′-AAGTCGCCGTCTTCCGCCAAG-3′; reverse 5′-GTCCAGGGACCCAATCTTCGA-3′ as previously described [16]. The PCR amplification was performed in a final volume of 25 μL under the following conditions: 95 °C for 15 min, and then 30 cycles at 94 °C for 30 s, 60 °C for 30 s, and 74 °C for 90 s with a final 10 min extension at 74 °C. RT-PCR products were analyzed on 2% agarose gel: 4R and 3R tau RT-PCR products were 381 and 288 bp, respectively.

NanoString neuropathology panel gene expression profiling

The NanoString Neuropathology panel was used for gene expression profiling of our RNA samples on the nCounter platform (NanoString Technologies®, Seattle, WA, USA). The neuropathology panel incorporated 770 human genes, including ten references (housekeeping genes) and 760 endogenous genes, to target all aspects of neurodegeneration. Data were analyzed by ROSALIND® (https://rosalind.onramp.bio), with a Hyperscale architecture developed by ROSALIND, Inc. (San Diego, CA, USA).

AMP-AD post-mortem brain cohorts and gene co-expression modules

The 30 human AMP-AD co-expression module data were obtained from the Synapse data repository [17,18,19] (https://www.synapse.org/#!Synapse:syn11932957/tables/;SynapseID: syn11932957).

iPSC-derived cultures-human brains expression comparison

The nSolver software-generated raw gene expression counts for the Neuropathology panel. Normalization was done by dividing counts within a lane by the geometric mean of the housekeeping genes from the same lane. Next, normalized count values were log-transformed for downstream analysis. To determine the effect of FAD mutation, we fitted a linear mixed-effect regression model using the lme4 function in R by considering the fixed effect of FAD mutation plus random variation in intercept among cell lines within controls and FAD mutation (formula ~ FAD + 1|FAD:Cell.Line).

To compare expression changes in FAD lines with those observed in human disease, we computed Pearson correlations between log fold change in transcript expression of human AD patients compared to control patients (Log2FC(AD/controls)) and the effect of FAD mutation estimated by the linear mixed-effect regression model [20, 21]. Correlation coefficients were computed using the cor.test function built-in R as: cor.test(log2FC(AD/Control,βFAD)).

LogFC values for human transcripts after correction for technical covariates were obtained via the AD Knowledge Portal (https://www.synapse.org/#!Synapse:syn14237651).

Statistical analysis

Information regarding statistical analyses can be found in the figure legends. No sample-size calculation was performed. Samples size was determined based on previous experience and consensus in the field. Data are expressed as the mean ± standard error of the mean (SEM), represented as error bars. All statistical tests were computed using GraphPad Prism 9 software unless otherwise noted. ROUT method (1%) was applied to identify outliers. In some cases, selected samples were excluded from specific analyses because of technical flaws during sample processing or data acquisition. A p-value of 0.05 was used as the significance threshold throughout this study. In all figures, p-values are illustrated for all tests used: n.s.p > 0.05, *p < 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. A minimum of two differentiations for each iPSC line were analyzed in independent experiments. Figure legends and graphs representing individual data points indicate the sample size (n = number of scaffolds).

Results

A 3D bioengineered silk-collagen neural model

NPCs were generated from various iPSC lines using an embryoid body (EB)-based protocol [9, 22,23,24] that seeks to recapitulate embryonic differentiation. In the absence of exogenous patterning factors, the default path of this protocol is to generate forebrain neurons of cortical fates [25]. For the success of this experimental strategy, quality control parameters are associated with the iPSC cultures [26]. iPSCs are susceptible to cell culture stresses and prone to genomic instability. Morphology is an obvious indicator of the iPSC health [26]. In our practice, karyotyping, potency, and mycoplasma testing have been routinely performed in all iPSC lines used for this study (Fig. 1A, B, Fig. S1).

Fig. 1: A 3D bioengineered silk-collagen neural model.
figure 1

A Quality tests performed on iPSCs used in this study. B Representative karyotype report. Reports for all the lines are in Fig. S1. C Schematic representation of the differentiation timeline to generate NPCs. (a) Representative iPSC colony stained for Oct4 (red) and DAPI (blue). Scale bar: 40 μm. Confocal z-stack. Leica SP8 HC PL APO CS2 20 × /0.75 IMM. 1.5 × Zoom. EB (b) and rosettes (c) phase-contrast images. Scale bar: 20 μm. Nikon TE300 CFI PL FL DL 4 × /0.13 and 10×/0.30. (d) NPC cultures stained for nestin (green) and DAPI (blue). Scale bar: 40 μm. Confocal z-stack. Leica SP8 HC PL APO CS2 20 × /0.75 IMM. 1× Zoom. D 2D 7-month-old hNeurons stained for GFAP (magenta), SYN (green), and MAP2B (red), and 10-month-old hNeurons stained for MAP2B (magenta), Tuj1 (green), and DAPI (blue). Scale bar: 50 μm. Confocal z-stack. Leica SP8 HC PL APO CS2 20 × /0.75 IMM. 0.75 × Zoom. E Schematic representation of the timeline to differentiate hNeurons in silk-collagen scaffolds from NPCs. Silk scaffold phase-contrast image. Scale bar: 500 μm. Z-stack. Stitched image. Zeiss CD7 PL APO 5 × /0.35. Magnification changer 0.5 ×. 3D reconstruction of hNeurons grown in a scaffold for 2.5 months and stained with NeuroFluor™ NeuO. Leica SP8 HC PL APO CS2 20 × /0.75 IMM. 0.75 × Zoom. F 3D reconstruction of hNeurons grown in a scaffold for 2 months (stained for MAP2B (magenta), NFH (green)), 4 months (stained for Tuj1 (magenta), GFAP (green)), and 4.5 months (stained for MAP2B (green)). For all the images DAPI and silk autofluorescent signals are in gray. Leica SP8 HC PL APO CS2 20 × /0.75 IMM. 0.75  × Zoom.

Full size image

Figure 1C illustrates the timeline schematic for the protocol, in which well-defined and compact iPSC colonies expressing pluripotency markers like the octamer-binding transcription factor 4 (Oct4) are differentiated into NPCs over approximately 20 days. The milestones of this approach are the formation of EBs, three-dimensional cell aggregates that mimic some structures of the developing embryo. They can differentiate into all three germ layers [27], followed by neural rosette maturation that can be identified morphologically by their characteristic appearance as round clusters of primitive neuroepithelial cells with apicobasal polarity [28] (Fig. 1C).

According to previous work, nestin-positive NPCs can be robustly passaged, frozen, thawed, and differentiated into functional neurons. NPCs could be typically passaged more than ten times in vitro without accumulating karyotype abnormalities [28]; however, in agreement with other studies [28, 29], we found that beyond ten passages, NPCs show increased cellular heterogeneity and lower differentiation efficiency. Thus, experiments were conducted on mycoplasma-free passage-matched NPC populations between passages two and six.

When differentiated in 2D on a poly-ornithine/laminin substrate using BrainPhys media supplemented with SM1 and N2, NPCs can efficiently differentiate into human neurons (hNeurons) expressing characteristic markers of a mature neuronal population such as the microtubule-associated protein 2B (MAP2B), the neuron-specific class III beta-tubulin (Tuj1), and synaptophysin (SYN). Glial fibrillary acidic protein (GFAP)-positive cells are also observed, indicating the presence of astrocytic-like cells (Fig. 1D). NPCs have already been described as having dual lineage potential and can differentiate into 70–80% neural populations of Tuj1-positive neurons and 20–30% GFAP-positive astrocytes [28].

For this study, NPCs have been seeded into silk scaffolds and cultured according to the timeline in Fig. 1E. Our data indicate that NPCs integrate into the silk-collagen scaffold model, expand (NeuroFluor NeuO live staining, Fig. 1E), and differentiate into healthy neurons and astrocytic-like cells, as indicated by IF imaging (Fig. 1F). Cultures express neuronal markers such as neurofilament heavy chain (NFH), MAP2B, and Tuj1 together with GFAP (Fig. 1F). They develop an intricated three-dimensional network extending deep into the silk pores and the supporting collagen, suggesting high connectivity among growing cells (Fig. 1E, F). Cultures generated with this approach supported by BrainPhys media supplemented with SM1 and N2 can survive in culture for extended periods.

Aβ42/40 ratio and extracellular Aβ42 deposition are elevated in FAD cultures

To explore whether our bioengineered cultures could be valuable for modeling AD, we generated cultures from extensively characterized iPSC lines [9] from a father (hFAD-1) and a daughter (hFAD-2), each harboring the FAD mutation APP V717I [9] (Fig. 2A). The following lines were selected as controls (Fig. 2A). YZ1 [9], derived from fetal lung fibroblasts, BR24 (Religious Order Study (ROS) or Rush Memory and Aging Project (MAP) [12]), a cognitively unimpaired woman at the time of biopsy, 76 years of age [6]. Participants in the ROS and MAP projects range from 50 to over 100 years and are free of dementia at the enrollment [6, 12]. Follow-up assessments indicated that the BR24 subject died at the age of 90, showing no signs of cognitive decline or AD pathology. We also generated an isogenic control for the hFAD-2 Clone B line (Fig. 2B) using genome-editing technology by reverting the APP V717I mutation to wild type (C-hFAD-2 Clone B). No karyotype clonal abnormalities were identified among the lines besides hFAD-1, which revealed a balanced (t(1;12)(q42.3;q21.2)) translocation in all cells analyzed (Fig. S1). The fibroblasts obtained from this subject displayed the same abnormal karyotype, suggesting a preexisting abnormality that did not arise during the reprogramming [9].

Fig. 2: Aβ42/40 ratio and extracellular Aβ42 deposition are elevated in FAD cultures.
figure 2

A Table reporting iPS lines used across the study. B Sequencing traces of hFAD-2 Clone B and its isogenic control after reversion of the APP V717I mutation (C-hFAD-2 Clone B). C nSolver Cell Enrichment analysis bar plot. D Aβ42/40 MSD analysis quantification in conditioned media. FAD cultures (hFAD-1, hFAD-2 Clone A, hFAD-2 Clone B) and controls (C-hFAD-2 Clone B, BR24, YZ1) plotted as groups are provided at 2 and 4.5 months. CTR 2 mo: n = 10, FAD 2 mo: n = 7, CTR 4.5 mo: n = 12, FAD 4.5 mo: n = 15. One-way ANOVA, Tukey’s post-hoc test. E Amyloid burden quantification (%). FAD cultures (hFAD-1, hFAD-2 Clone A, hFAD-2 Clone B) and controls (C-hFAD-2 Clone B, BR24, YZ1) plotted as groups are provided at 2 and 4.5 months. n = 6. One-way ANOVA, Tukey’s post-hoc test. F Representative 3D reconstructions of hNeuron culture (4.5 months) derived from C-hFAD-2 Clone B and hFAD-2 Clone B stained for MAP2B (magenta) and Aβ42 (green). DAPI and silk autofluorescent signals are in gray. Zoom-in view: green Aβ42 extracellular depositions (white arrows) interspersed among hNeurons (MAP2B, magenta). Leica Falcon SP8 HC PL APO CS2 20×/0.75 IMM. 3.5× Zoom.

Full size image

To gain a deeper understanding of our cultures and the reproducibility of the differentiation approach across various iPSC lines and clones, we utilized NanoString Technologies® to measure gene expression at 2 and 4.5 months. Data were analyzed by ROSALIND® (https://rosalind.onramp.bio), with a HyperScale architecture developed by ROSALIND®, Inc. (San Diego, CA, USA), and the nCounter® Advanced Analysis protocol.

We used the nCounter® Cell Type Profiling Module (as described in [30]) to assess the abundance of various cell populations (Fig. 2C). While markers for endothelial cells, oligodendrocytes, astrocytes, neurons, and microglia were screened, only scores for neurons and astrocytes reached a statistically significant threshold (p-value < 0.001 and <0.01, respectively, Fig. 2C), indicating that these two cell types are the main constituents of our cultures.

Exploiting NanoString normalized expression analysis, we compared the expression of specific groups of commonly expressed genes in iPSC-derived neural cultures (Fig. S2) based on previous studies [9, 22, 25, 31,32,33] between different lines, genotypes, time points, and biological replicates and in agreement with previous observations [9, 22], normalized expression analysis identified the enrichment of a vast array of glial (GFAP, ALDH1L1, S100B, etc.), pan-neuronal and synaptic markers (HOMER1, TBR1, STX1A, etc.) with most neurons exhibiting an excitatory profile, together with some progenitor markers (NES, CCND1) being expressed up to 4.5 months post-seeding. As expected, based on the differentiation method implemented, genomic analysis suggests that our culture acquired a cortical layer fate as indicated by the expression of genes (TBR1, DLX1, DLX2, etc.) involved in mammalian neocortical development and/or expressed in a layer-specific manner in the human neocortex (Fig. S2) [34]. Also, AD-related genes were expressed in our 3D cultures across all analyzed conditions. Differential expression analysis did not reveal any significant variance between lines and genotypes for this set of genes, suggesting that the cultures are homogeneously patterned and that the differentiation approach is highly reproducible.

To assess whether our cultures can recapitulate some AD-related phenotypes, soluble Aβ quantification (MSD: Meso Scale Discovery Assay, Aβ Peptide Panel 1-4G8) was performed on 24-h-conditioned media (Fig. 2D). In agreement with previous reports [9, 25, 35, 36], the Aβ42 over Aβ40 ratio was found to be increased in FAD cultures compared to control ones at both time points (CTR 2 mo: 0.040 ± 0.002 vs. FAD 2 mo: 0.089 ± 0.017, p = 0.0003; CTR 4.5 mo: 0.048 ± 0.004 vs. FAD 4.5 mo: 0.093 ± 0.005, p < 0.0001). The Aβ42/40 ratio calculated for the isogenic C-hFAD-2 Clone B control was comparable to that of other control lines, supporting the idea that the observed increase in the hFAD-2 Clone B depended on the APP V717I mutation.

Next, we performed IF using an antibody specific for Aβ42 (validated in 5XFAD:BACE1˗/˗ brain sections [37]) and calculated the amyloid burden for CTR and FAD cultures at both time points (Fig. 2E). Confocal imaging and 3D reconstruction revealed the presence of Aβ42-positive extracellular aggregates interspersed among FAD hNeurons (identified by MAP2B staining, Fig. 2F). Volumetric analysis of the amyloid burden (Fig. 2E) revealed a mild trend for increase starting at 2 months in FAD cultures compared to controls (CTR: 0.00053 ± 0.00022 vs. FAD: 0.00170 ± 0.00044, p = 0.9368) with a striking difference among conditions at 4.5 months (CTR: 0.00028 ± 0.00005 vs. FAD: 0.02572 ± 0.00281, p < 0.0001). No difference was identified between control groups at the two time points analyzed.

We aimed to fully explore how our cultures can be used to model AD by studying tau pathology. Tau proteins have six main isoforms that differ in the number of N-terminal inserts (0N, 1N, 2N) and C-terminal repeat domains (3R or 4R). These isoforms are expressed differently depending on the brain region and developmental stage. During neurogenesis, only the shortest tau isoform, 0N3R, is expressed, while in the adult brain, all six isoforms are present in roughly equal amounts of 3R and 4R isoforms [38,39,40]. RT-PCR for 3R and 4R tau isoforms (Fig. S3A) in FAD and control scaffolds showed that 2 months cultures already expressed 4R tau, while NPCs only expressed 3R tau. At this time point, the calculated ratio (Fig. S3B) between 3R and 4R was around 2.6/2.8 for both conditions. In contrast, at 4.5 months, the ratio was significantly reduced to 1.6/1.7 (CTR p = 0.0256, FAD p = 0.0072), approximating the one calculated for the human brain (1.2) and indicating that our 3D cultures recapitulate the developmental pattern of human brain tau, expressing mainly the shortest 3R tau initially and later both 3R and 4R to nearly equal amounts. No difference was identified between FAD and control cultures.

We further investigated the presence of hyperphosphorylated tau accumulation in our cultures by IF using TG3, a conformation-sensitive anti-tau antibody that recognizes tau phosphorylated at threonine-231 [41]. While no positive signal was detected in our cultures for up to 4.5 months, when tested in older cultures (12 months, Fig. S4), TG3 detected pathological tau in FAD scaffolds. No positivity was observed in the control lines.

Based on the data, it appears that our culture model is effective in replicating the expression of mature tau isoforms, which is essential for investigating tau pathology.

FAD cultures manifest enhanced neuronal excitability

To determine the electrophysiological properties of the cultures, local field potentials (LFPs) were recorded from single electrodes inserted into the superficial layers of the hydrogel next to silk fibers, adapting a protocol by Du and collaborators [15], where both spontaneous and evoked potentials (EPs) were collected. LFPs, indicative of cellular electrical activations [8], were quantified and compared across all conditions. The spontaneous and evoked activity was observed within all the scaffolds analyzed starting at 2 months, suggesting that our 3D model is populated by functional neurons capable of network activity (Fig. 3).

Fig. 3: FAD cultures manifest enhanced neuronal excitability.
figure 3

A Representative traces of spontaneous LFPs (mV) obtained from 3D cultures derived from a FAD line (hFAD-2 Clone B) and the isogenic control (C-hFAD-2 Clone B) at 2 months and 4.5 months. A 20-s reference is provided for traces. Quantification of normalized spontaneous and evoked LFP spikes/min (Log10) at 2 months (B) and 4.5 months (C). FAD cultures (hFAD-1, hFAD-2 Clone A, hFAD-2 Clone B) and controls (C-hFAD-2 Clone B, BR24, YZ1) plotted as groups are provided. B Spontaneous CTR 2 mo n = 26, FAD 2 mo n = 25. Evoked CTR 2 mo n = 26, FAD 2 mo n = 25. C Spontaneous CTR 4.5 mo n = 23, FAD 4.5 mo n = 21. Evoked CTR 4.5 mo n = 22, FAD 4.5 mo n = 21. Unpaired t-test.

Full size image

In our study, FAD cultures displayed comparable levels of spontaneous potentials (Log10 LFP spikes/min) relative to control samples at 2 months (CTR: 0.717 ± 0.130 vs. FAD: 0.872 ± 0.126, p = 0.3950); however, by 4.5 months, FAD samples displayed significantly more spontaneous potentials relative to controls (CTR: 0.515 ± 0.106 vs. FAD: 0.832 ± 0.114, p = 0.0475). Similarly, FAD samples showed more EPs relative to controls at 4.5 months post-seeding (CTR: 0.736 ± 0.079 vs. FAD: 1.050 ± 0.077, p = 0.0064), which was not apparent at 2 months (CTR: 1.040 ± 0.091 vs. FAD: 1.270 ± 0.112, p = 0.1069) (Fig. 3). The increased excitability in FAD cultures correlated with extracellular Aβ42 deposition but not with soluble Aβ42/40 ratio levels, as they were similar at both time points. These data suggest that extracellular Aβ deposition may trigger enhanced network activity in our model. Remarkably, neuronal hyperexcitability has been described in FAD and late-onset AD (LOAD) patients at the early stage of the disease [10].

Since numerous studies suggest that Aβ oligomeric species (oAβs) play a role in the neuronal hyperexcitability seen in AD [10], we isolated oAβs from the scaffold-conditioned media using an IP/WB protocol [14]. Immunoprecipitation was performed with R1282 [14] antibody followed by WB with 6E10 antibody. We also performed IP/WB with 4G8/6E10, respectively, as we previously reported [42] that this protocol can isolate Aβ monomers and other oligomeric species in conditioned media from cell cultures (Fig. S5). Soluble Aβ oligomeric species were detected using both 6E10 antibody and anti-oligomer antibody A11,. As expected, Aβ monomers were only observed following incubation with the 6E10 antibody (Fig. S5). Levels of oAβs were higher in FAD cultures than in controls at both time points analyzed (2 and 4.5 mo), ruling out that in our model, soluble oAβs represent a causal factor in the enhanced excitability only observed in the FAD cultures at the 4.5-month time point. Based on current evidence, the extracellular deposition of Aβ seems to have the strongest correlation with increased excitability over time. However, it cannot be ruled out that other factors may also play a role in causing this specific phenotype.

Differential expression and functional analysis of the NanoString panel in FAD cultures

Transcriptional profiling of NanoString data from our samples was performed by ROSALIND® platform (see method section for details). At 2 months, there were 133 differentially expressed genes (DEGs) (absolute fold change > 1.5 & FDR < 0.05) between FAD and control samples (Fig. 4A). Expression for 38 of those 133 genes was upregulated in FAD samples compared to controls and downregulated for 95 of 133 genes (Fig. 4A). Gene set analysis (GSA) identified increased expression of genes associated with Cytokines, Activated Microglia, and Neuronal Cytoskeleton (directed GSS > 0) (Fig. 4C).

Fig. 4: Differential expression and functional analysis of the NanoString panel in FAD cultures.
figure 4

DEGs (absolute fold change > 1.5, FDR < 0.05) in 2-month (A) and 4.5-month (B) CTR and FAD lines are shown in the heatmap. GSA of differentially expressed genes in iPSC-derived cultures (FAD vs. CTR, ROSALIND® platform, NanoString annotations) at 2 months (C) and 4.5 months (D). Upregulation: directed enrichment score > 0. Downregulation: directed enrichment score < 0. FAD 2 mo n = 8, CTR 2 mo n = 8, FAD 4.5 mo n = 9, CTR 4.5 mo n = 7.

Full size image

Based on the differentiation protocol used for this study, we expected our cultures to be populated exclusively by cells derived from the neuroectodermal lineage. So, we were surprised to observe an upregulation of the Activated Microglia pathway in FAD cultures. Follow-up immunofluorescence for TMEM119 (Fig. S6), a well-known microglia marker, followed by confocal microscopy, identified the presence of TMEM119-positive cells in the scaffolds interspersed among MAP2 and GFAP-positive cells, indicating the presence of microglia-like cells in our cultures. The cells show a ramified morphology and are intertwined with neuronal processes as described by others (Fig. S6) [43]. These data agree with previous studies that have identified mesoderm-derived progenitors and microglia-like cells in human iPSC-derived cerebral organoids [43, 44]. Quadrato et al. [44] performed single-cell RNAseq in cerebral organoids and found that while six of the seven clusters identified belonged to the neuroectodermal lineage, one expressed mesodermal markers. This suggests that although organoids are patterned to a neuroectodermal fate early on, they can still produce a minority of cells from another embryonic origin. Accordingly, Ormel et al. [43] also reported that microglia innately develop within cerebral organoids. It is worth mentioning that the protocol used by Lancaster et al. [45] and the method we used to generate NPCs in this manuscript did not involve dual-SMAD inhibition. This could have allowed mesodermal precursor cells to survive and differentiate into microglia-like cells. These findings suggest that microglia-like cells are not uncommon in iPSC-derived cultures and are unrelated to the silk-collagen scaffold structure. This could also explain why the Activated Microglia gene set was upregulated.

Further, GSA identified reduced expression of genes associated with multiple other neurodegeneration pathways such as Transmitter Release, Neural Connectivity, and Vesicle Trafficking (directed GSS < 0) (Fig. 4C).

At 4.5 months, there were 38 DEGs (absolute fold change > 1.5 and p < 0.05) between FAD and control samples (Fig. 4B). Expression for 5 of the 38 genes was upregulated in FAD samples compared to controls and downregulated for 33 of 38 genes (Fig. 4B). Gene set analysis identified increased expression of genes associated with Cytokines and reduced expression of genes associated with multiple pathways such as Transmitter Release, Neuronal Cytoskeleton, and Vesicle Trafficking (Fig. 4D). Although astrocytes and microglia secrete most cytokines in the brain, some evidence suggests that neurons can also produce cytokines under detrimental conditions [46,47,48]. So, we can’t exclude that neuronal secreted cytokines could partly mediate the upregulation of the Cytokines pathway in the FAD scaffolds in response to adverse stimuli.

FAD cultures reproduce human neurodegeneration signatures

To validate our cultures as an AD model in vitro, we applied a system biology approach developed to evaluate AD mouse models [20]. A correlation analysis was performed between gene expression changes in FAD cultures and changes in human AD cases in post-mortem co-expression modules. We used a recent human molecular disease catalog based on harmonized co-expression data from three independent post-mortem brain cohorts (ROSMAP, Mayo, Mount Sinai Brain bank) [17, 18, 49] and seven brain regions that define 30 human co-expression modules and five consensus clusters derived from the overlap of those modules [11]. FAD cultures at 2 months showed significant positive correlations (p < 0.05) with AD-associated changes in human co-expression modules in Consensus Cluster B, which include transcripts that are enriched for immune-related pathways in the cerebellum brain region (CBEturquoise) (Fig. 5A). Furthermore, FAD cultures at 2 months showed significant positive correlations (p < 0.05) with AD changes in multiple modules in Consensus Cluster C, which include transcripts that are enriched for neuronal-related pathways in the inferior frontal gyrus module (IFG), temporal cortex (TCX), and the parahippocampus gyrus (PHG) brain regions (Fig. 5A). Interestingly, FAD cultures at 2 months showed a significant positive correlation (Pearson Correlation = 0.27; p < 0.05) with AD ECM organization changes in module TCXblue in Consensus Cluster A but also showed a significant negative correlation (Pearson Correlation = −0.35; p < 0.05) with AD changes in the PHGyellow module enriched for similar biological process. However, these significant positive correlations diminished at 4.5 months (Fig. 5A). To understand this observation, we compared gene expression changes in control cultures at 4.5 months to control cultures at 2 months and correlated with human co-expression modules. We observed significant positive correlations (p < 0.05) with multiple AD-associated human co-expression modules in Consensus Cluster A (TCXblue), Consensus Cluster B (CBEturquoise and PHGturquoise), and Consensus Cluster C (IFGbrown, TCXgreen, and PHGbrown) (Fig. 5A). This suggests that by 4.5 months a strong aging signature correlated with the AD effects we observed for FAD cultures at 2 months. This implies that the correlation between human and FAD effects at 4.5 months was decreased because in vitro aging in the control cultures already mimicked human AD in the relevant modules. Similarly, gene expression in wild-type aged mice correlated with AD-associated expression modules [11], suggesting that the aging process may activate some of the gene expression observed in the AD human brain. Furthermore, after comparing gene expression changes in FAD cultures at 4.5 months relative to 2-month control cultures with the human co-expression modules, we found significant correlations with human AD effects in Consensus Cluster A (TCXblue), Consensus Cluster B (CBEturquoise, STGblue, and IFGturquoise) and Consensus Cluster C (IFGbrown and TCXgreen) (Fig. 5A). We also observed significant a 4.5-month, FAD-specific positive correlation (p < 0.05) with the immune-associated module IFGturquoise in Consensus Cluster B regardless of control (Fig. 5A). Finally, a common correlation with AD effects in the TCXblue module was observed for both FAD and aged cultures (Pearson Correlation > 0.4; p < 0.05) (Fig. 5A). These results suggest that AD-like gene expression and functional changes in extracellular matrix organization are driven by both cellular aging and FAD mutations.

Fig. 5: FAD cultures reproduce human neurodegeneration signatures.
figure 5

Correlation analysis between the gene expression profile of FAD cultures and human post-mortem co-expression modules. A 2-month-old FAD cultures exhibited a significant positive correlation (p < 0.05) with AD changes in immune-related modules in Consensus Cluster B and neuronal-related modules in Consensus Cluster C. Moreover, 4.5-month control and FAD cultures exhibited significant positive correlations (p < 0.05) with AD changes in immune-related modules in Consensus Cluster B and neuronal-related modules in Consensus Cluster C, suggesting common aging and FAD effects. Circles within a square correspond to significant (p < 0.05) positive (red) and negative (blue) Pearson correlation coefficients for gene expression changes in FAD lines associated with disease-associated changes in distinct human co-expression modules. Color intensity and size of the circles are proportional to the correlation. B Differentially upregulated genes in the 2-month-old FAD lines (FDR < 0.05, fold change > 1.5), showing increased expression in immune/inflammation-associated AMP-AD modules in Consensus Cluster B, are shown in the heatmap. C Differentially downregulated genes in the 2-month-old FAD lines (FDR < 0.05, fold change < 1.5), exhibiting reduced expression in neuronal-associated AMP-AD modules in Consensus Cluster C, are shown in the heatmap. FAD 2 mo n = 8, CTR 2 mo n = 8.

Full size image

Next, we examined the expression profile of differentially regulated genes in the FAD cultures identified by the ROSALIND® platform across human AMP-AD transcriptomic data. We determined that most of the differentially upregulated genes (fold change > 1.5) in 2-month-old FAD cultures showed increased expression in immune/inflammation-associated AMP-AD modules in Consensus Cluster B (Fig. 5B), suggesting increased expression of these genes in multiple human brain regions. Similarly, most of the differentially downregulated genes (fold change < 1.5) in 2-month-old FAD cultures showed reduced expression across numerous human brain regions except in the frontal pole and cerebellum brain regions of AD patients (Fig. 5C). Our 2-month-old FAD cultures exhibited expression profiles of neuronal genes remarkably similar to human AD for multiple brain regions.

Similarly, we examined the expression of differentially deregulated genes in 4.5-month-old FAD cultures across AMP-AD modules. There were only five genes that were differentially upregulated in the 4.5-month-old FAD cultures compared to controls, and these genes mainly exhibited reduced expression in some of the AMP-AD modules (Fig. S7A). Most of the differentially downregulated genes (fold change < 1.5) in 4.5-month-old FAD cultures showed reduced expression in neuronal-associated AMP-AD modules in Consensus Cluster C, except in CBEyellow and FPyellow modules (Fig. S7B). These results suggest that neurodegeneration-related genes were downregulated in our FAD cultures, similar to AD patients. Accordingly, staining for cleaved caspase-3 (Fig. S8), an established marker of apoptosis as one of the main executors of programmed cell death [50], revealed numerous positive cells across all the FAD lines at 4.5 months (Fig. S8). At the same time point, control scaffolds display minimal or no signal for the apoptotic marker, suggesting a higher degree of cell death in FAD cultures than controls.

Since iPSC reprogramming rejuvenates cells and erases donor aging signatures, concerns were raised about the ability to model age-related diseases with iPSC-derived neuronal cultures [51]. It is well-established that AD develops over decades [52]. Thus, genetic factors may produce endophenotypes early in life; therefore, they can be captured in iPSC-derived neural cultures.

Discussion

We developed a neural tissue model that supports the differentiation of iPSC-derived NPCs in a donut-shaped porous silk sponge with an optically clear central region.

This work offers convincing proof that our 3D bioengineered model, developed from patient-derived FAD iPSCs, exhibits phenotypes and transcriptomic features that progress over time, similar to what is observed in AD patients. Other iPSC-derived 3D models undergo similar changes [5]. However, as far as we know, none of these 3D cultures have been able to replicate multiple hallmarks associated with AD in a single model. While Aβ extracellular deposition, changes in the Aβ42/40 ratio, and elevated cleaved caspase-3 represent frequent findings in 3D models, the enhanced excitability and transcriptomic changes identified in our study have been difficult to recapitulate in these conditions. Importantly, in our scaffolds, these phenotypes have been observed at different time points, resembling some features that characterize AD progression. Aβ42/40 ratio and transcriptomic changes are already present at 2 months, while Aβ deposition appears only at 4.5 months when the neuronal hyperexcitability becomes evident, suggesting a temporal relation between these two events. Gahtak et al. [53] detected hyperexcitability in AD organoids accompanied by increased Aβ immunostaining and an increase in the ratio of secreted Aβ42/40 at 2 months; nevertheless, no other time points were analyzed, providing no information on the temporal relation between these phenotypes.

To date, no other studies have examined the transcriptomic features of these 3D models while comparing them to those of human AD brains in a large study that performed a co-expression meta-analysis of harmonized data from Accelerating Medicines Partnership for Alzheimer’s Disease (AMP-AD). These data are, in our opinion, key to understanding the pertinency of these models for studying AD-related pathomechanisms over time. This study demonstrated that our 3D cultures could recapitulate AD transcriptomic features identified in human disease. Interestingly, 2-month FAD cultures better correlate with AD transcriptomic changes than older cultures. Further analysis revealed that gene expression changes in 4.5-month control cultures compared to 2-month control samples correlated with multiple human AD modules similar to FAD cultures at 4.5 months compared to 2-month control lines. This observation suggests that by 4.5 months, the aging signature can partly obscure the FAD-mediated effects in our system. So, we should consider the possibility that our cultures would better model different aspects of the disease based on the time point analyzed (e.g., transcriptomic features at 2 months and Aβ extracellular deposition and hyperexcitability at 4.5 months).

One of the main strengths of this model is that every element was designed to be tunable and amenable to experimental modification. Although we did not incorporate a layered structure for this study, the 3D model can be designed using a concentric toroidal scaffold to more closely approximate the gross features of the layered brain tissue [54]. However, we need to consider that any modification to the structural design of the model could cause unexpected changes to the integrity and longevity of the cultures [8] and therefore require extensive and costly troubleshooting. We could also incorporate embedded electrodes for continuous, long-term recordings of electrophysiological activity [55, 56]. Moreover, mixed 3D neuronal tissue could be generated by co-culturing neurons, astrocytes, and microglia derived from different donors or isogenic iPSC lines carrying AD-linked genetic variants [57]. For example, mixed 3D cultures containing mutant neurons and wild-type astrocytes or microglia cells and vice-versa could be used to study the impact of cell-specific genetic variants on AD-like phenotypes. For all these reasons, our bioengineered neural tissue represents a unique tool to model AD-related pathomechanisms over time, with several advantages and potentialities compared to existing models.