Functional and Sustainable 3D Human Neural Network Models from Pluripotent Stem Cells

Abstract

Three-dimensional in vitro cell culture models, particularly for the central nervous system, allow for the exploration of mechanisms of organ development, cellular interactions, and disease progression within defined environments. Here we describe the development and characterization of three-dimensional tissue models that promote the differentiation and long-term survival of functional neural networks. These tissue cultures show diverse cell populations including neurons and glial cells (astrocytes) interacting in 3D with spontaneous neural activity confirmed through electrophysiological recordings and calcium imaging over at least 8 months. This approach allows for the direct integration of pluripotent stem cells into the 3D construct bypassing early neural differentiation steps (embryoid bodies and neural rosettes), which streamlines the process while also providing a system that can be manipulated to support a variety of experimental applications. This tissue model has been tested in stem cells derived from healthy individuals as well as Alzheimer’s and Parkinson’s disease patients, with similar growth and gene expression responses indicating potential use in the modeling of disease states related to neurodegenerative diseases.

Introduction

Understanding human neurodevelopment and degeneration are some of the most challenging areas of biological sciences due to the fragile and intricate nature of the human brain in vivo. With billions of neurons interconnected by trillions of synapses, parsing out accurate and specific information can prove difficult which results in data collection with limited resolution. The techniques often used include electroencephalogram (EEG), magnetoencephalography (MEG) or functional magnetic resonance imaging (fMRI), but are each limited. Both EEG and MEG record bulk electrical activity with a high level of temporal resolution, but limited spatial resolution, fMRI, which measures relative blood flow as a readout of activity, allows for better spatial resolution than either the EEG or MEG but is limited temporally due to the lag time between synaptic activity and changes in blood oxygen level-dependent (BOLD) contrast(1). Even with these methods in combination, attaining single-cell level of resolution is currently unattainable without more involved techniques, such as the implantation of neurotrophic electrodes(2, 3). Because of their invasive nature, these procedures are limited to severe cases such as patients suffering from “locked-in” syndrome(2). Due to the limitations inherently posed by attempting to access the human brain, alternative models for the study and experimentation of neural cell functions in the context of complex 3D brain-related niches are required to advance our collective understanding.

Much of the knowledge surrounding the human brain has been gained through the use of animal (in vivo) models. While animal studies continue to significantly help in understanding brain development and function, they do not fully mimic human conditions of complexity, constitution(4) or disease states(5). Using animals as the sole preclinical test model has also come under increasing scrutiny relating to their validity as a model for pharmacologic testing(6–8). While they may be the best available option for testing, a comprehensive survey of rates of clinical trial success for investigational drugs showed a 9.4% likelihood of acceptance with neurological drugs in phase 1(9) which does not account for the possible drug candidates discarded due to inefficacy observed within non-native neurological models. While there is clear value to their use, significant room for improvement is present.

An alternative to in vivo animal studies has been the use of tissue explants in the form of brain slices or hanging drop cultures. These culture methods have limitations including sample acquisition, damage to tissue during harvesting, limited time in culture and inaccessibility to experimental intervention as they are often closed systems(10). There have been improvements in culture methods to increase the survival of organotypic neural tissue, such as microfluidic devices(11), but even with these advances, human neurological tissues are rarely removed from a healthy patient and even diseased tissue is often only available post-mortem. Culturing primary neurons, regardless of the source is problematic due to their post-mitotic nature, limiting culture expansion. This problem was circumvented with pluripotent cells and the discovery that mature somatic cells can be induced back into a pluripotent state (induced pluripotent stem cells-iPSCs)(12, 13). These pluripotent cells can be exponentially expanded, and when under appropriate conditions can be directed to differentiate into specific cell types, including neurons. There have been multiple advances supporting the use of pluripotent cells to differentiate into functioning neurons in three dimensions, including human and murine cerebral organoids(14, 15) and human cortical spheroids(16). These approaches use the embryoid body stage of hiPSC differentitation to create developmental brain models(14–16). The spheres were grown in spinning bioreactors, free floating, under minimal direction and formed organized structures that mimicked those found in the human brain with varying levels of consistency. The structures developed by these cells in an undirected environment demonstrated an ingrained ability to self-organize. It has also been shown that stem cells can be directed towards different cell fates by matrix stiffness with substrates proving osteogenic, myogenic and neurogenic and the stiffness decreases(17).

Recent progress in tissue engineering has started bridging the gap between the accessibility of in vitro models and in vivo accuracy. In vitro, engineered models of human brain tissue can provide stable long-term systems to mimic brain tissue structure and function to support the study of neurological processes, brain development, degenerative disorders, drug interactions and related needs. In order to achieve this goal, it is necessary to generate consistent cultures that support the growth of biologically relevant cells. The goal of this study was to demonstrate the viability of a 3D tissue culture model for use in the differentiation of human induced pluripotent stem cells to functional neural networks utilizing neurogenic-supportive 3D culture systems that exploit the ability of the stem cells to self-organize, as observed in the organoid cultures, in combination with patient derived iPS cells, generating a platform that will allow for the study of the progression of numerous neurological diseases. The development of a complex model containing various neural subtypes and supporting astrocytes within this 3D tissue model provides a unique advantage capable generating a stable and functional long-term (months to years) cultures. Such systems should provide new and useful approaches for the study of neurodegenerative diseases, chronic drug effects and insight into the mechanisms of aging.

Materials and Methods

Cells

The hiPSC lines ND41866*C, GM24666*A, and ND35367*F were obtained from the Coriell Biorepository (Camden, NJ, USA), the YZ1 hiPSC line was obtained from Dr. Giusepena Tesco courtesy of the University of Connecticut-Wesleyan Stem Cell Core (UCSCC, Farmington, CT, USA).

hiPSC Maintenance

The hiPSCs were generated as described (ND41866(18), GM24666(19), ND35367(20), YZ1(21)) and were maintained in feeder free conditions as previously described(22). In brief, plates were coated using Matrigel hESC-qualified Matrix (Corning, Corning, NY, USA) and DMEM/F12 media (Thermo Fisher, Cambridge, MA, USA). The coated plates were stored at 4°C and were warmed to 37°C prior to use. The cells were maintained with daily media changes using mTeSR1 (Stem Cell Technologies, Vancouver, BC, Canada) and were passaged 1:7 every 5–7 days using ReLeSR passaging reagent. (Stem Cell Technologies, Vancouver, BC, Canada)

Scaffold Preparation

Silk protein was processed from Bombyx mori cocoons as described previously(23). In brief, silk cocoons were cut into fragments and boiled for 30 min in a 0.02M Na₂CO₃ solution. The fibroin was solubilized in a 9.3M LiBr solution which was then dialyzed away in deionized water. The silk solution was then diluted to 6% weight/volume. The solubilized silk was then used to generate the porous scaffolds via salt leaching, as explained previously(24) using 500–600um NaCl crystals to generate a sponge-like structure. Scaffolds were then coated with poly-L-ornithine (20ug/mL in 1x DPBS) (Sigma-Aldrich, St. Louis, MO, USA) for at least 2 hours at room temperature, and then washed 2x with DPBS, 1x with DMEM/F12, followed by a 2-hour (minimum) laminin coating (20ug/mL in DMEM/F12) (Roche, Basel, Switzerland). The coated sponges were stored under sterile conditions at 4°C until use.

Scaffold Seeding

Undifferentiated hiPSCs were treated with ReLeSR (Stem Cell Technologies, Vancouver, BC, Canada) to specifically remove undifferentiated portions of the colonies. These cell clusters were then added to the scaffold at approximately 5.0×10⁶ cells/mL and incubated for 24 hours. The media was changed daily for 5 days following seeding, moving the scaffolds to fresh wells with each change. After allowing 5 days of cellular expansion the scaffolds were then moved to a fresh well and filled with 100uL of a cold collagen-type I rat-tail solution (3.0mg/mL) mixed with 10x PBS and 1N NaOH (88:10:2). The collagen filled scaffolds were then incubated at 37°C until the gelation occurred (~30 minutes). Once solidified, the filled scaffolds were then transferred to a 24 well plate and flooded with 1mL of final neuro media (FNM) (Neurobasal media, Anti-Anti, Glutamax, with B-27 supplement) with media changes every four days.

Nucleic Acid Isolation

At the time of desired analysis scaffolds were snap frozen using liquid nitrogen and stored at −80°C. These frozen scaffolds were then homogenized using a liquid nitrogen-chilled Spectrum Bessman Tissue Pulverizer (Fisher Scientific, Hampton, New Hampshire, USA). The homogenized sample was run through a Qiashredder column and then purified using the AllPrep DNA/RNA/Protein Mini Kit (Qiagen, Hilden, Germany). DNA was quantified using Quant-iT PicoGreen dsDNA Assay kit (Thermo-Fisher, Cambridge, MA, USA). Genomic DNA was removed from the isolated RNA using TURBO DNA-free kit (Thermo-Fisher, Cambridge, MA, USA), which was then quantified using a NanoDrop 2000 Spectrophotometer (Thermo-Fisher, Cambridge, MA, USA)

rt-PCR

cDNA was generated using High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher, Cambridge, MA, USA). This cDNA was used for qPCR using the Taqman system (Fast Advanced Mastermix, supplemental figure for full list of probes) (Thermo Fisher, Cambridge, MA, USA). qPCR was conducted utilizing a BioRad CFX96 RT-PCR system (BioRad, Hercules, CA, USA) with the specific hiPSC maintenance cultures used as controls. All samples were run in technical triplicates. If a value never crossed the threshold the value of the last cycle run was used for ddCT calculations. The housekeeping gene used was 18S and the controls were always the hiPSC maintenance cultures for the specific hiPSC line

Immunocytochemistry

Cultures were fixed in 4% PFA (Santa Cruz, Dallas, TX, USA) for 10–20 minutes, washed with 1x PBS, and then blocked with blocking solution (Goat Serum (15mL), TritonX-100 (Sigma Aldrich, Natick, MA, USA), Sodium Azide (5mL), in 1x PBS (fill to 250mL)) for 1 hour. Primary antibodies (Beta-3-tubulin (B3T), glial fibrillary acidic protein (GFAP), neurofilament-heavy (NFH)) were diluted in the blocking solution at 1:500, and kept at 4°C overnight. Cultures were washed 3x with 1xPBS followed by treatment with secondary antibodies diluted 1:250 in the blocking solution and allowed to interact overnight at 4°C. The cultures were washed 3x with 1xPBS and then DAPI stained for 20 minutes.

Confocal Imaging and Analysis

A Leica SP8 Confocal microscope (Leica, Wetzlar, GER) was used to collect the source image stacks. ImageJ was used to measure beta-III-tubulin density and to perform intensity measurements for calcium imaging.

Functional Experiments

Extracellular Field Potential Recordings (EFP)

Scaffolds were recorded in a bath consisting of NaCl 140mM, KCl 2.8mM, CaCl₂ 2mM, MgCl₂ 2mM, HEPES 10mM, and D-glucose 10mM, with pH adjusted to 7.4 with NaOH. The sharp glass microelectrodes were filled with extracellular solution and had a resistance of 60–80 megawatts The recording electrode was placed near the edge of the central window. Fast field potential changes (spikes) were recorded with the NPI amplifier at a bandwidth of 0.3–10 kHz and were further amplified with an A-M Systems Differential AC amplifier (Model 1700) to a combined total gain of 10,000x. The signals were digitized at 10 KHz by a Molecular Devices digitizer (Digidata 1550) using a Dell Optiplex GX620 computer with pClamp 10 software (Molecular Devices). Electrical stimulations (40 V, 1 ms pulses) of the cultured tissues were generated via a Grass S44 Stimulator, passed through a Grass Stimulus Isolation Unit (SIU5), and delivered through a platinum parallel bipolar electrode (FHC PBSB0875) with a distance of 800 μm between two tips positioned near the recording electrode. Activity inhibitors were used to confirm neuronal activity by use of glutamate receptor blockers (6-cyano-7-nitroquinoxaline-2,3-dione; CNQX (50 μM), (2R)-amino-5-phosphonopentanoate; AP5 (250 μM), GABA antagonist (–) Bicuculline (100μM) acetylcholine antagonists to both nicotinic and muscarinic Mecamylamine hydrochloride (10μM) and Scopolamine hydrobromide (50μM) respectively, as well as the neuronal sodium channel blocker tetrodotoxin (TTX, 5uM). After recording responses with electrical stimulation (40V, 1 ms, at 0.1 Hz), the procedure was repeated in the presence of the specific receptor blockers and then a third time with addition of TTX. The responses of the post-TTX recordings were subtracted from the pre-TTX recordings to identify neural component of the stimulation response and to ascertain the presence of specific neural subtypes.

Calcium imaging

Fluo-4 (Invitrogen, Carlsbad, CA, USA) was used as described in the manufacturer’s manual. In brief the fluo-4 was dissolved in Puronic-F127 (Life Technologies, Carlsbad, CA, USA) at a volume of 1uL/ug. The resuspended fluo-4 was then diluted in artificial cerebrospinal fluid (NaCl: 140nM, KCl: 2.8nM, CaCl₂: 2nM, MgCl₂: 2nM, HEPES: 10mM, Glucose: 10mM, pH7.4,) 1:1000. The cells were incubated in the Fluo-4 solution for 60 min at 37°C. After the incubation the cells were washed once with PBS and supplied with fresh artificial cerebrospinal fluid and imaged for 3 minutes in a climate-controlled chamber at 5% CO₂ and 37°C on a Keyence BZ-X700 (Itasca, IL, USA). Some recordings were also taken using a Leica SP8 confocal microscope (Leica, Wetzlar, GER). Some samples were excited with MDNI-caged glutamate (Tocris, Bristol, UK) after a baseline recording was obtained, as previously described(25). The resulting recordings were analyzed using ImageJ to identify regions of interest and track fluorescent intensity over time. Drift was corrected using ImageJ Image Stabiilizer (45). dF/F was calculated as previously described(26, 27).This data was then graphed in Microsoft Excel.

Statistics

Prism 5 by GraphPad Software was used to perform statistics. To assess significance in DNA content within the cultures a two-way ANOVA (initial seeding density, time point) was performed with a Bonferroni post-test on group averages. This same analysis was used to analyze the qPCR data. Prism was also used to generate the box-and-whisker plot for percent of B3T coverage of max projections. P values are denoted as (*) P<0.05, (**) P<0.01, (***) P<0.001

Results

Attachment

The porous donut-like scaffolds were coated with poly-ornithine (PLO), followed by laminin (PLO/L) to maximize cell attachment while supporting neural growth (Fig. 1). The coated scaffolds were then seeded with clusters of hiPSCs derived from gentle colony dissociation using ReLeSR. The scaffolds were sized to fit within the wells of a 96 well plate and the media volume was kept minimal to maximize capillary action in the sponge scaffolds. The cell population within the scaffolds, as determined by DNA content, was observed to have the highest density, in the largest concentration condition, 2×10⁶ cells/mL, at the initial time point, day 1 (Fig. 2a). A significant decrease of DNA content was observed by the third day of growth in the higher density starting conditions (2×10⁷ and 1×10⁷ cells/mL) while scaffolds seeded with lower concentrations of cells did not appear to undergo a similar loss. In order to maximize cell content without significant loss, the scaffolds were seeded with 5×10⁵ cells, and attachment and expansion was allowed for 5 days with daily media changes. On the 5^th day the media was aspirated from the scaffolds, which were then filled with collagen-type I hydrogel, and transferred to 24 well plates and flooded with neural-supporting growth media.

Figure 1: Schematic of tissue model workflow.

A. Fibroin sponges are coated with poly-ornithine and then laminin, seeded with clusters of hiPSCs (5×10⁶ cells/mL) for 5 days. After the expansion period the scaffolds are filled with collagen to provide the three-dimensional growth substrate for the cells. The media was changed from a pluripotent maintenance media to a neuronal supporting media, allowing for long-term growth and differentiation of polarized neurons and supporting glial cells. B. Top and side view of scaffold prior to seeding

Figure 2: Determining optimal seeding density for hiPSC growth/differentiation.

A: Picogreen assay performed on scaffolds seeded with indicated cell densities, over time (n=3, error bars = standard deviation) ***P<0.001, *P<0.05 B: Image taken of scaffolds collagen filled on the indicated day after seeding with 5×10⁵ cells. C: Brightfield image showing projections growing into the central window of a scaffold seeded with 5×10⁵ cells at 8 weeks post-collagen fill (scale bar: 500um)

Differentiation

To explore the timeline of growth and differentiation we focused on the healthy ND41866 hiPSC derived scaffolds. After the scaffolds were transferred to the neural-supporting media, neuronal cells were observed as early as two weeks based on beta-III-tubulin (B3T) staining (Fig. 3a) and gene expression (Fig. 4, Supplemental Figure 1). Over time in culture the presence of B3T+ cells increased from a mean of 4.896% B3T coverage (range: 1.639%−8.469%) at week 2 to a mean of 41.22% coverage (range: 29.79%−54.82%) by week 12 (Figure 3b). The increase in B3T density demonstrated the continued growth of neurons within the tissue model over the 12-week period. qPCR was performed to determine relative gene expression of the differentiated cultures. When comparing the gene expression of our 3D cultures to similarly grown 2D cultures (Fig. 4) we observed significantly higher levels of the neural differentiation gene NEUROD6 (neuronal differentiation 6) at 2 weeks, and by 10 weeks the 3D cultures showed significantly higher expression of MAP2 (microtubule associated protein 2), ENO2 (neural specific enolase 2), and TUBB3 (beta-3-tubulin) when compared to the 2D cultures suggesting an increased rate of development and a longer culture maintenance (Fig. 6C) when grown in a stable, three-dimensional culture (Fig. 6D). The expression of the pluripotent gene markers (NANOG and POU5F1 – OCT4) decreased with differentiation, while KLF4 (Kruppel Like Factor 4) is upregulated (Supplemental Figure 2). KLF4 is one of the genes utilized by Yamanaka and Takahashi(12, 13) to generate iPSCs from somatic cells, but has also been shown to be involved in regulation of neural stem cells (28, 29), which could explain the expression levels observed. This possibility is supported by the absence of this increase in KLF4 expression when testing hiPSCs that are further differentiated to neural progenitor cells (Supplemental Figure 1). We then tested for germ-layer markers and observed an increase in both the endo- and ectodermal gene expression (GATA4 – GATA binding protein 4 and FGF5 – fibroblast growth factor 5, respectively) but saw substantial down-regulation of the mesodermal marker brachyury (T) (Supplemental Figure 2). GATA4, while commonly used as a genetic marker for endoderm has also been shown to be present in human neurons and a negative regulator of astrocytes (30, 31), which could explain the increased expression levels observed. As GATA4 is known for it role in cardiac development we explored the levels of TNNT2 (a cardiac specific troponin – T2), which was only up regulated in the 2D two week samples (Fig 4). The levels of neural markers including those necessary for the production of specific neurotransmitters (TH (tyrosine hydroxylase) – dopamine, ChAT (choline acetyltransferase) – acetylcholine, and TPH1 (tryptophan hydroxylase 1) – serotonin)), which were all substantially upregulated, as were markers for both NMDA (GRIN1 - Glutamate Ionotropic Receptor NMDA Type Subunit 1) and AMPA (GRIA1 - Glutamate Ionotropic Receptor AMPA Type Subunit 1) glutamatergic receptors when compared to the control, but only GRIN1 showed significant upregulation in the 3D vs the 2D cultures. The expression level of GABBR1 – (Gamma-Aminobutyric Acid Type B Receptor Subunit 1) GABAergic receptor gene is only slightly increased in the 3D cultures suggesting little to no presence of inhibitory neurons. (Fig 4, Supplemental Figure 2). We also looked for markers of non-neuronal supporting cells (GFAP - Glial Fibrillary Acidic Protein and SLC2A1 - Solute Carrier Family 2 Member 1) to determine the presence of astrocytes (GFAP), and glucose transporters (SLC2A1) found in the blood-brain-barrier. In order to confirm the utility of this model, the same approach was undertaken with 4 hiPSC lines derived from 4 individuals, including two healthy controls (ND41866 and YZ1), a patient diagnosed with sporadic Alzheimer’s Disease (GM24666), and a patient diagnosed with Parkinson’s (ND35367), compared at the 3 month time point. Gene expression indicating similar differentiation and neural subtype diversity was observed across the multiple hiPSC cultures with some variation (Figure 5E, Supplemental Figure 1), but the only gene expression observed to vary between up and down regulation compared to control was GATA4 (Supplemental Figure 1). We also observed similar growth of neurons and astrocytes in Alzheimer’s derived and healthy scaffolds(Fig. 6 A&B).

Figure 3: Timeline of beta-3-tubulin presence in 3D tissue model.

A. Representative images of scaffolds taken at indicated time points (green: B3T). Images are max projections of confocal z-stacks. (scale bar: 100 μm). B. Box and whisker plot of sum projections from 100-slice z-stacks to determine percent of the imaged area covered by B3T+ staining. (n=5 per time point). C. Example of 3D growth of neurons after 12 weeks of growth (scale bar: 100 μm)

Figure 4: Gene expression comparison of 2D vs. 3D cultures over time.

Gene expression of 2D and 3D samples displayed as ddCT. Genes involved in neural differentiation and maturation (NEUROD2, NEUROD6, NES, TUBB3, MAP2, ENO2); NMDA receptor subunit 1 (GRIN1); AMPA receptor subunit 1 (GRIA1); Cardiac Troponin 2 (TNNT2). 2D n=3; 3D n=4. Error bars are standard error of the mean. ***P<0.001, **P<0.01, *P<0.05

Figure 6: Both Healthy and Alzheimer’s Disease patient derived hiPSCs generate neurons and Astrocytes in 3D tissue model.

A. Growth of neurons differentiated from hiPSCs for 8 months (red: B3T, green: NFH, blue: DAPI). Shown as max projection of confocal z-stack. Scale bar 100μm. Top Left: scaffold seeded with hiPSCs derived from a healthy patient, image take from bulk of scaffold. Bottom left: scaffold seeded with hiPSCs derived from a healthy patient, image take from central window of scaffold. Top right: scaffold seeded with hiPSCs derived from an Alzheimer’s patient, image take from bulk of scaffold. Bottom right: scaffold seeded with hiPSCs derived from an Alzheimer’s patient, image take from central window of scaffold.n B. Max projection of z-stack (left: ND41866 – healthy, right: GM24666 – Alzheimer’s) showing presence of both astrocytes and neurons within the scaffold (red: GFAP, green: B3T). Scale bar: 100μm C. Max projection of z-stack from 10 month old culture from healthy hiPSCs (red: B3T, green: NFH, blue: DAPI). Scale bar: 100μm D. 3D nature of growth taken from central window of a healthy hiPSC derived culture. Scale bar: 100μm

Figure 5: Electrophysiological Activity in Scaffolds.

A and C. Quantification of synaptic potential from local field potential recordings pre and post addition of indicated receptor antagonists in cultures ranging from 4–8 months, normalized to post-TTX addition. A Healthy (ND41866) hiPSC derived cultures, C Alzheimer’s (GM24666) derived hiPSC cultures. B. Examples of recording used in A and C analysis. D. Calcium imaging response to glutamate stimulation in a 9month old healthy (ND41866) scaffold E. Comparison of gene expression levels across 4 hiPSC sources (ND41866 and YZ1 – Healthy derived, GM24666 – Alzheimer’s derived, ND35367 – Parkinson’s derived). GRIN1 – NMDA receptor subunit 1, GRIA1 - AMPA receptor subunit 1, GABBR1 – GABBA type B receptor subunit 1, TPH1 – tryptophan hydroxylase 1, ChAT – Choline acetyl transferase, TH – tyrosine hydroxylaser. Error bars = SEM

Function

In order to determine physiological activity within the 3D tissues and scaffolds, local field potential (LFP) was utilized. The presence of both spontaneous (Supplemental Figure 3) and inducible (Figure 5A) activity was observed within the scaffold by 10 weeks of growth indicates the presence of healthy, functioning neurons in our 3D network model, and is observable out to 9 months (and counting). The induced electrical response was able to be partially blocked using a combination of CNQX and AP-5(AMPA/kainate receptor and NMDA receptor antagonists respectively), indicating the presence of functional glutamatergic neurons within the model. Bicuculline, a GABA receptor antagonist, was also tested and showed a smaller response relative to the baseline recording, indicating a smaller involvement of GABAergic neurons in the network, which is supported by the lower gene expression of the GABA receptor (GABBR1) compared to the NMDA (GRIN1) or AMPA (GRIA1) receptor genes. The nicotinic and muscarinic acetylcholine receptor antagonists mecamylamine hydrochloride and scopolamine hydrobromide did not to have a strong effect on LFP recordings. After specific antagonists were administered activity was then blocked using the sodium channel blocker tetrodotoxin (TTX). The traces post-TTX administration were compared to the traces recorded prior to the TTX treatment and the differences were due to the presence of a synaptic-specific response. Calcium imaging using Fluo-4 showed similar results as the LFP data, where activity within the scaffolds was found as early as 8 weeks and was observed at 9 months of cultivation and still responsive to glutamate stimulation (Supplemental Video 1).

Discussion

The 3D tissue model developed here, derives from our previous work (32) which utilized rat primary cortical neurons to demonstrate the feasibility of this scaffold design to support neuronal growth over longer culture periods. By developing this tissue model for use with human iPS derived neurons we have a new tool to explore 3D neural growth in a human specific manner. In order to adapt this system to a human model growth conditions were optimized for use with human induced pluripotent stem cells. As both enzymatic(33, 34) and single-cell(35) passaging of stem cells have been shown to be linked to genetic abnormalities, therefore, for the seeding into the scaffold we utilized an enzyme-free passaging reagent which also maintained cells in multi-cell clusters. It was originally expected that a high density of cells would be needed to mimic the growth observed in organoid differentiation, but as shown in Figure 2, seeding at densities above 5×10⁵ cells resulted in a significant decrease in DNA content after the first day. This decrease was likely a result of a combination of poor attachment to the scaffolds and the limited metabolic support for the highly active stem cells provided from the small media volume utilized to exploit the capillary action in the sponge-like scaffolds. Optimal seeding density was determined to be 5×10⁶ cells/mL and after the 5 days of cell expansion, the sponges were filled with the collagen hydrogel to support the three-dimensional growth. These seeding outcomes were different than those established earlier with the rat neurons, which required a higher cell seeding density (1×10⁷ cells/mL) and due to the post-mitotic nature of primary neurons population expansion was not a consideration so collagen gel was added after an overnight period for cell attachment.

Once the scaffolds were filled with collagen and the media changed from pluripotency-maintaining media to a neural-supporting media, an early marker of neural identity (B3T) was evident at 2 weeks and increased to 12 weeks. These cells also exhibited spontaneous electrical activity as observed through local field potential recordings and calcium imaging, beginning at 8 weeks and increasing to 12 weeks of culture. In a parallel experiment of 2D differentiation, in which the enzyme-free dissociated hiPSC-clusters were also seeded similarly and allowed to grow under the same conditions were tested as in the 3D model, several non-neuronal morphologies were observed (Supplemental Figure 4) In addition, contractions were observed at 4 weeks in these cultures, as often seen from cardiac muscle (Supplemental Video 2). While these observations were present in the 2D systems, the 3D tissue models showed no indication of contractions, and the relative gene expression for both 2D and 3D for mesoderm marker T (brachyury) was down regulated in all but the two week 2D sample, which was only slightly up regulated. This led us to test the gene expression levels of TNNT2 (Troponin T2, Cardiac type), an indicator of cardiac muscle. The TNNT2 expression was only increased in the 2D 2-week samples, which mimics the T (brachyury) expression. While the absence of cardiac muscle in our 3D system is advantageous, the presence of various cell types is not surprising due to the relatively limited differentiation control when compared to other neural differentiation protocols. The undirected nature of this differentiation approach in combination with the differences in substrate stiffness between the 2D and 3D culture conditions could very well have played a role in cell differentiation as it has been shown that stem cells will differentiate to muscle or even bone as the stiffness increases(17). Other unexpected observations were the up regulation of the endoderm marker, GATA4 and the pluripotentcy marker KLF4. While GATA4 is a common marker for identifying endodermal lineage it has also been shown to be expressed in neurons and glial cells(30, 31, 36), and KLF4 to be involved in neural stem cell maintenance (29). The increased KLF4 expression could be related to the maintenance of a progenitor cell population as the increase in expression is not observed in neurons further differentiated along the neural pathway. The GATA4 expression appears to decrease in the disease derived (Parkinson’s-ND34367 and Alzheimer’s-GM24666) cultures (Supplemental Figure 1). This data is too preliminary to draw a conclusion as to the involvement of astrocyte regulation in the disease state, but could be an interesting avenue for continuing research. What we are able to identify is a complex interaction of various neuronal populations within the tissue model from both gene expression and electrical activity (Fig. 4,5, SF1–3). The expression of genes involved in neurotransmitter synthesis (ChAT, TPH1 and TH) and in neuronal receptor formation (GRIN1 and GRIA1) in combination with the neural differentiation, maturation and synaptic markers (ENO2, NEUROD2, NEUROD6, MAP2) demonstrated the growth of a diverse culture that is maturing and forming networks (Fig. 4).

There have been multiple models developed that mimic brain development and neurological disease states(15, 16, 37, 38) in engineered systems utilizing hiPSCs. Many of these approaches rely on the progression of neural differentiation from pluripotent cells to embryoid bodies and utilize the natural architecture developed in these spheres with minimal external direction to develop organoids that exhibit multiple characteristics of the human brain. Due to the dense nature and lack of active perfusion the size of these systems are limited. The tissue model described in this paper shows the direct differentiation of hiPSCs to neurons, bypassing spheroid and neural rosette stages, all while within a 3D scaffold supporting the formation of a functional three-dimensional neural network. While this model also lacks an active perfusion system the nature of the porous scaffolding helps alleviate the diffusion concerns found in other approaches(32, 39, 40). The spherical cultures provide excellent developmental models of the brain as an organ, but are not necessarily well suited for use in studies of network development and remodeling, as real-time experimental access to the network during maturation or dysfunction proves difficult. There are additional concerns with models involving dense spheroid cultures as the spheroid size increases beyond 400–600μm, the interior cells are exposed to a hypoxic environment and eventually can result in a necrotic core(41–43), a fact which is sometimes exploited to mimic the hypoxic conditions found in tumor environments(44).

When working with induced pluripotent stem cells, consistency of differentiation across multiple lines is necessary for model utility. Here, similar growth was demonstrated in 2 healthy hiPSC lines, as well as from a sporadic Alzheimer’s patient derived as well as Parkinson’s derived hiPSC line. This demonstration of neural network development across different patient disease states indicated the possibility of using these cultures to study early progression of chronic diseases in controlled 3D tissue cultures. One issue that could be optimized was the differing growth kinetics across hiPSC lines, which may necessitate variations in seeding and expansion prior to the collagen filling and cell differentiation process. This tissue model could also provide insight into early markers for diagnosis and potential therapeutics. The long-term in vitro stability of this model presents a possible option for investigating later stages of neurodegenerative diseases that require long culture periods in vitro prior to observable deficits. Additionally, varying the extracellular matrix included within the described culture could provide an avenue for matrix-related effects on disease or injury states as we have previously demonstrated the beneficial effects of the inclusion of brain-derived extracellular matrix on the growth of rat primary neurons (45). The ‘donut’ design allows for a window to visualize the 3D axonal growth and organization in real-time providing increased access for study. The ability of this tissue system to support the growth primary neurons (32, 40) as well as induced neural stem cells(46), and neurons derived directly from human iPS cells demonstrates its versatility for neurological uses.

While this model has numerous possibilities for future uses, there are some limitations to the system. One such limitation is the undirected nature of this differentiation procedure, which can even result in cardiac muscle contractions when not grown in an appropriate environment. While we did not observe these cardiac concerns in the 3D cultures, one side of the scaffold is in contact with the plate surface, which may affect growth. As the neurogenic nature of the tissue model generates a variety of neural subtypes as well as astrocytes (Fig. 6B) resulting in a more complex network organization, it does not generate a pure neural culture, which can convolute certain experimental interpretations. Another limitation, which is common across most three-dimensional models, is the difficulty in accessing cells within the bulk of the scaffold. This poses problems for activity recording (both electrophysiological and calcium based), as well as obtaining images of interior sections of the model beyond the limit of the confocal microscope. Local field recording is possible but not as straightforward as in other neural systems, and calcium imaging is limited. While the central window found in this model does provide easier access to the cells in the bulk it can also be used to generate a more compartmentalized system that maximizes axonal growth through the window supporting real-time observation of network development.

Conclusions

Our goal was to develop a 3D system that would provide the neurogenic conditions to support the self-organization of the human stem cell derived neural networks. We have developed a tissue model that allows for the direct integration of undifferentiated stem cells into three-dimensional cultures and differentiates into a functioning network consisting of interconnected neurons and astrocytes. The tissue model supports the growth and neuronal differentiation of both healthy and disease-derived hiPSCs, while eliminating the preliminary differentiation steps typically needed with other stem cell derived neural tissue models. This streamlined model, limits the reagents and growth factors required for successful differentiation, supports long term culture and provides utility in terms of higher throughput for various screening applications. In addition, the sustainable cultivation in vitro provides a suitable 3D tissue system for the study of neurodegenerative disease states and response to chronic conditions. This model could also be useful to study the development and plasticity of neural networks under various conditions, such as stimulation regiments or injury systems. It should be noted that this is not a developmental model, and as such does not model brain development in a physiologically relevant manner, but does provide access to 3D compartmentalization of neural processes and network maintenance and activity. In conclusion the described model provides an accessible system for studying neural growth from stem cells in three dimensions, its durable versatility supports use in a variety of applications, and its longevity allows for the exploration of neural network formation, malfunction or response to manipulation over time.