Neural layer self-assembly in geometrically confined rat and human 3D cultures

Abstract

Neurological disorders affect millions of Americans and this number is expected to rise with the aging population. Development of drugs to treat these disorders may be facilitated by improved in vitro models that faithfully reproduce salient features of the relevant brain regions. Current 3D culture methods face challenges with reliably reproducing microarchitectural features of brain morphology such as cortical or hippocampal layers. In this work, polydimethylsiloxane (PDMS) mini-wells were used to create low aspect ratio, adherent 3D constructs where neurons self-assemble into layers. Layer self-assembly was determined to depend on the size of the PDMS mini-well. Layer formation occurred in cultures composed of primary rat cortical neurons or human induced pluripotent stem cell - derived neurons and astrocytes and was robust and reproducible. Layered 3D constructs were found to have spontaneous neural activity characterized by long bursts similar to activity in the developing cortex. The responses of layered 3D cultures to drugs were more similar to in vivo data than those of 2D cultures. 3D constructs created with this method may be thus suitable as in vitro models for drug discovery for neurological disorders.

Introduction

Some of the most common neurological diseases that affect the brain, including Alzheimer disease, stroke, traumatic brain injury, migraine, epilepsy, and Parkinson disease, affect more than 65 million Americans and have a total annual cost exceeding half a trillion dollars [1]. Hundreds of medicines are under development by the biopharmaceutical industry in an effort to intervene or slow the progression in these disorders [2]. However, more than half of the drugs that emerge from pre-clinical development fail in clinical trials [3]. The rate of discovery of new treatments may be increased by improved experimental models that maintain experimental accessibility while faithfully reproducing salient features of the source brain region. The latter is particularly important to ensure that research findings from the screening model are translated successfully to the clinic.

Three dimensional (3D) cultures of brain cells have been developed several decades ago [4]. Since neurons in the brain are densely packed and are involved in a variety of cell to cell interactions, 3D cultures represent a more realistic in vitro model of the brain than 2D cultures. Traditional methods of creating 3D cultures, including explants of brain tissues [5] and neurospheroids [6], have been used to investigate a wide variety of biological questions in neuroscience including neurological disorders such as Alzheimer’s, Parkinson’s, epilepsy, and many others [7]. Recent advances in biomaterials and micro-scale engineering led to creation of 3D cultures with densities similar to those found in the brain [6] and with uniform and reproducible culture size [8–9]. 3D cultures have been reported to have spontaneous neural activity that comes closer to replicating activity in the brain than 2D cultures [10–11]. The advent of human induced pluripotent stem cell – derived neurons (hiPSC-neurons) led to the development of 3D cultures composed of human neurons in which disorders such as Alzheimer’s disease or microencephaly can be modeled (reviewed in [4,12]).

A significant challenge to modeling the brain in vitro lies with reproduction of its complex microarchitecture. Neurons in brain regions such as cortex and hippocampus are organized into well-defined layered structures that determine cell-to-cell interactions. The functional recapitulation of these brain regions, and the deficits induced by the disease state, are likely dependent on layered architecture and stereotyped connections between them. It is therefore important to reproduce cell layers in vitro to accurately brain function in health and disease. One approach to reproducing brain microarchitecture is to use explant cultures derived from cortical or hippocampal slices. These cultures are termed ‘organotypic’ since they preserve the structure of the originating brain region in vitro over several weeks [7]. However, only a relatively small number of organotypic cultures can be generated at a time, limiting their utility in high-throughput screens. An alternative approach is to start by dissociating a brain region into individual cells, and then form 3D structures by placing cells into a hydrogel [13] or a microwell [6,8–9,14–17]. Resulting 3D cultures are relatively homogeneous, but, in one effort, building blocks were created from 3D aggregates and used to create pre-designed, heterogeneous structures [15]. However, features in these structures were millimeter-scale compared to micro-scale architecture found in the brain. A third approach is to form 3D cultures composed of stem cell-derived neural progenitor cells, and then to differentiate the entire cell aggregate. It was found that this method produces microarchitectural features resembling layers of the cortex [18–20]. A significant disadvantage of this method is its potentially poor uniformity and reproducibility of resulting structures [4].

In this work, we set out to develop a method of creating 3D neural cultures that are suitable for drug discovery screens by satisfying the following requirements: (1) it robustly produces cultures of uniform size and density, (2) it works with hiPSC-derived neurons as well as rodent primary neurons, (3) cultures are substrate-adherent to ensure compatibility with connectivity control methods [21–22], and (4) cultures has similar microarchitectural features as the corresponding brain region. We discovered that neurons in cultures produced by our method spontaneously assemble into cortical or hippocampal – like layers: to the best of our knowledge, this is the first time that this type of self-assembly was reported for primary neurons and pre-differentiated hiPSC-derived neurons. We also discovered that layered 3D cultures have a drug response that is significantly different than that of 2D cultures but is similar to in vivo data. We also compared viability, morphology, and activity of 3D (PDMS mini-well) cultures with spheroid cultures.

Materials and Methods

PDMS micro-wells

A silicon wafer was spin coated with liquid Polydimethylsiloxane (PDMS) and cure mixture (1:10) (Sylgard 184 by Dow Corning). After overnight baking, a 100 μm thick flexible PDMS film was obtained. PDMS film was then removed from silicon wafer, and holes were punched from top with appropriately sized needles to create mini-wells: 25-gauge (530 μm), 20-gauge (910 μm) and 16-gauge (1630 μm) needle for 400 μm, 800 μm, and 1300μm bottom diameter, respectively. Resulting mini-wells were narrower at the bottom and wider at top. In this work, we referred to the bottom diameter of the mini-well by “mini-well diameter”. Devices with holes were then cut out, sterilized in Ethanol, and dried. Glass coverslips were cut with diamond tipped glass cutter into approximately 10 mm X 10 mm pieces. These glass coverslip pieces were coated with 200 μl of 1 mg/ml sterile solution of poly-D-lysine (PDL, Sigma) in borate buffer at pH 8.5 at 37 °C and 5% CO2 overnight. After incubation, they were washed in sterile distilled water and dried. Pre-cut, sterile and dry PDMS devices were then placed on PDL coated glass coverslip pieces (Figure 1). The clean PDMS devices spontaneously formed water-tight seal with the glass coverslip pieces and this seal remained intact for several weeks in culture medium (97.45% Neurobasal-A, 2% B27, 0.25% GlutaMAX™ (100x) and 0.3% 30 μg/mL gentamicin (Life Technologies)). Assembled devices were then placed in 35 mm culture dishes, submerged under culture medium and incubated at 37 °C overnight.

Figure 1.

Holes were punched out from cured PDMS films with appropriately sized needles. These devices were then cut out and placed on PDL coated glass coverslip before plating on DIV00. Droplet of dense cell solution (40 million/ml) was then pipetted carefully into these mini-wells (Top right panel). The cells in the droplet were then allowed to settle down for 15 minutes. The plate was then flooded with culture medium and incubated. 3D structure was visible on DIV 08.

Dissociated cortical neurons

We obtained cortices from neonatal rats (post-natal day 0–1 Sprague-Dawley rat pups, Charles River Laboratories) according to the protocol developed by Brewer et al [23]. Papain Dissociation System (PDS Kit, Papain Vial, Worthington Biochemical Corporation) was used for dissociation. All animal use protocols were approved by the Institution Animal Care and Use Committee (IACUC) at Lehigh University and were conducted in accordance with the United States Public Health Service Policy on Humane Care and Use of Laboratory Animals.

3D (PDMS mini-well) culture from dissociated cortical neurons

PDMS mini-wells were carefully filled with dense dissociated cell solution (40 million/ml) right after dissection of rat pup and incubated for 15 minutes in 37 °C to allow the cells to settle at the bottom. 3 μl, 1.5 μl and 0.5 μl of cell solutions were added to 1300 μm, 800 μm and 400 μm diameter mini-wells, respectively. After 15 minutes, the devices were submerged (1.5 ml of medium for each 35 mm dish) in 10% Fetal Bovine Serum (FBS, Thermo Fisher Scientific) in Neurobasal-A supplemented with 0.5 mmol/L GlutaMAX and 30 μg/mL gentamicin (Life Technologies) and incubated at 37 °C and 5% CO₂ for 1 hour. Medium was then removed and medium (1.5 ml per 35 mm dish) with adeno-associated virus (AAV) particles containing genetically encoded [Ca²⁺] indicator jRGECO1a [24] construct under Syn promoter (at titer ≥ 2×10¹⁰ vg/mL) was added. pAAV.Syn.NES-jRGECO1a.WPRE.SV40 was a gift from The Genetically Encoded Neuronal Indicator and Effector Project (GENIE) & Douglas Kim (Addgene plasmid # 100854). The culture was then incubated at 37 °C and 5% CO₂. Half of culture medium was replaced with fresh culture medium (97.45% Neurobasal-A, 2% B27, 0.25% 0.5 mmol/L GlutaMAX and 0.3% 30 μg/mL gentamicin (Life Technologies)) every 3 days. Cells spontaneously aggregated and created complex 3D structure over time. The final culture diameters for 1300 μm, 800 μm and 400 μm wells were 900 μm, 700 μm and 300 μm respectively after 8 days. Dynamic changes in jRGECO1a fluorescence due to activity could be observed from day in vitro (DIV) 08. For 3D cultures, ~120,000 cells were seeded in each mini-well of 1300 μm diameter. We also created dense 2D cultures with similar network size (number of neurons) as controls. For 2D cultures, ~120,000 neurons were seeded in 7 mm diameter wells. These 2D cultures were almost confluent. 2D cultures were also infected with jRGECO1a immediately after plating.

3D (PDMS mini-well) cultures from hiPSC neurons

We used hiPSC-derived neurons purchased from Cellular Dynamics International, Inc. We used a mixture of ≥90% pure population of primarily glutamatergic (excitatory) human neurons (iCell® GlutaNeurons), >95% pure population of primarily GABAergic (inhibitory) neurons (iCell® GABANeurons) and >95% pure population of human astrocytes (iCell® Astrocytes) to create our cultures. We followed manufacturer’s protocols to thaw the cells separately and then mixed them to our desired ratio. Cell culture surface was also prepared according the product website protocol [25]. We then added PDMS devices on prepared cell culture surfaces. The PDMS devices attached well to the surface and high density of hiPSC-derived cell mixture was seeded into PDMS devices. jRGECO1a infection was performed just after plating. The cultures were kept in complete BrainPhys medium consisting of BrainPhys Neuronal Medium, iCell Neural Supplement B, iCell Nervous System Supplement, N-2 supplement, laminin, and gentamicin. The culture was incubated at 37 °C and 5% CO₂. Half of culture medium was replaced with fresh culture medium every 3 days.

Spheroid Cultures

Intended numbers of rat cortical dissociated cells (50,000, 100,000 and 200,000) or hiPSC-derived neurons and astrocytes (100,000 and 200,000) were mixed well in 100 μl culture medium containing jRGECO1a at same titer as 2D and 3D cultures. Same culture medium was used for both 3D cultures and spheroid cultures in each case as described in the last two sections. This cell suspension was then put in a single well of Corning® 96-well Black/Clear Round Bottom Ultra-Low Attachment Spheroid Microplate. The cultures were then incubated at 37 °C and 5% CO₂. Half of the culture medium was replaced with fresh medium every other day from DIV 03.

Optical recording

Optical recordings were performed by placing cultures in a mini incubator (Bioscience Tools) kept at 37°C with constant supply of humidified blood gas (5% CO₂, 21% Oxygen, balanced Nitrogen, Airgas) on a stage of a fluorescent inverted microscope (Olympus). jRGECO1a induced fluorescence changes due to neural activity were observed via 4x objective at 600 msec exposure and camera frame rate of 1.59 seconds/frame. Optically recorded data were analyzed using ImageJ [26] and Matlab.

Optical data analysis

For 2D cultures, the active cells were identified from the recorded video frames by projecting each pixel’s standard deviation (z project>standard deviation in ImageJ [25]). Pixels with high standard deviation (changing fluorescence) were detected using analyze particle command of ImageJ software. This method detects mostly cell bodies with proximal processes. For 3D cultures and spheroids, a single region of interest around the three-dimension portion of the culture was used. Mean fluorescence in that region of interest was calculated for each time point (Supplementary Figure 1(g)). Mean fluorescence was then converted to signal to baseline ratio or ΔF/F [27] (Supplementary Figure 1(h)). If raw data is R and baseline is F, then

$$\frac{\mathrm{\Delta }F}{F}=\frac{R-F}{F}=y\left(t\right)$$ (1)

We then deconvoluted these values to obtain approximate firing rates at each frame. To do this, we investigated our MEA recorded electrophysiological data. After spike detection, these electrophysiological data are converted into raster plots where each vertical line means one neuronal spike (Supplementary Figure 1(A)). Then we modeled calcium spikes with a double exponential function (Supplementary Figure 1(B)):

$$h\left(t\right)=A\times \left( e^{-\frac{t}{{\tau }_d}}-e^{-\frac{t}{{\tau }_r}}\right)$$ (2)

where t is time and τ_d and τ_r are half-decay and half-rise times respectively.

From literature [24] value of A (change in ΔF/F for a single spike) for jRGECO1a was 0.28. τ_d and τ_r were set empirically by analyzing shortest spikes in recorded optical data. These constants were chosen to be τ_d = 1.47 sec and τ_r = 0.04 sec. We then estimated the ΔF/F for the electrically obtained raster plot by convoluting it with (2). If electrical data is spikes(t), then ΔF/F = z(t) is: (Supplementary figure 1(C)).

$$z\left(t\right)=h\left(t\right)*\mathit{\text{spikes}}\left(t\right)$$

Since, our optical recording frequency is lower (larger sampling interval) than the electrical recording frequency, we down-sampled z(t) to obtain y(t) with sampling frequency matching our optically recorded data. We assume that this simulated ΔF/F, y(t) is related to firing rate x(t) by the following simple equation.

$$y\left(t\right)=h\left(t\right)*x\left(t\right)$$ (3)

This is a system with a system response h(t). Upon performing Laplace transform,

$$y\left(t\right)\stackrel{\text{L}}{\to }Y\left(s\right)=\text{L}\left(h\left(t\right)*x\left(t\right)\right)=\text{L}\left(h\left(t\right)\right).\text{L}\left(x\left(t\right)\right)=H\left(s\right).X\left(s\right)$$

Here,

$$\begin{gathered} H\left(s\right)=A\times \text{L}\left(h\left(t\right)\right)=A\times \text{L(}{e}^{-\frac{t}{{\tau }_d}}-e^{-\frac{t}{{\tau }_r}}\text{)} \\ \text{= A×(L(}{e}^{-\frac{t}{{\tau }_d}})-\text{ L(}{e}^{-\frac{t}{{\tau }_r}}\text{))} \\ =\frac{A}{s+\frac{1}{{\tau }_d}}-\frac{A}{s+\frac{1}{{\tau }_r}}=\frac{A}{s+\frac{1}{1.4716}}-\frac{A}{s+\frac{1}{0.04}} \\ \frac{1}{H}=\frac{0.0411}{A}\times \left(0.0411\times s^2+1.0559\times s+0.6985\right) \end{gathered}$$

Performing inverse Laplace transform,

$$x\left(t\right)=F \times 0.147 \times \left(0.0411\times \frac{d^2\mathit{y}\left(\mathit{t}\right)}{d{t}^2}+1.0559\times \frac{d\mathit{y}\left(\mathit{t}\right)}{dt}+0.6985\times \mathit{y}\left(\mathit{t}\right)\right)$$ (4)

where F is a fitting parameter that allows the scaling of estimated activity index to match actual firing rate, and

$$\frac{d^2\mathit{y}\left(\mathit{t}\right)}{d{t}^2}=\frac{\left(y\left(t\right)-y\left(t-1\right)\right)-\left(y\left(t+1\right)-y\left(t\right)\right)}{d{t}^2}$$ (5)

$$\frac{d\mathit{y}\left(\mathit{t}\right)}{dt}=\frac{\left(y\left(t\right)-y\left(t-1\right)\right)}{dt}$$ (6)

The constant F = 15.3 used in equation (4) were obtained from fitting data from 3 different recordings. Equation (4) was then tested on 9 different electrical recordings which yielded a low (mean 0.75 spike/second error) (Supplementary Figure 1(F)).

ΔF/F data was deconvoluted using equations (4–6) and “activity index” corresponding to estimated firing rate x(t) was obtained (Supplementary Figure 1(I)). Activity was then detected (Red lines in Supplementary Figure 1(I)) using the method described below:

An activity episode was considered to start when the signal (activity index) crosses the start threshold. To determine the start threshold, the data range (difference between maximum and minimum data point for a recording) was calculated and the standard deviation of points below 20% of this data range was determined. The start threshold was set at 3 times this standard deviation and was calculated for each recording. The end threshold was chosen to be activity index = 0.2. When data crosses start threshold, an activity episode is detected, and this activity episode ends when data point is below end threshold.

Multi-electrode Array (MEA) Electrophysiology

Extracellular electrical recordings were performed using micro-electrode arrays (MEA) with 60 round titanium nitride (TiN) electrode of 30 μm diameter with glass ring and O-ring (60MEA 200/30 IR-TI, Multichannel systems). Clean, sterile and dry PDMS devices were attached on these MEAs so that the microwell covers the center 16 electrodes. The PDMS devices formed a stable water-tight seal with the glass substrate of MEA. The internal reference electrode was initially covered with a plain PDMS film of appropriate size so that cells cannot grow onto the reference electrode. The film was removed on DIV08 prior to recording. Recordings were performed in a mini-incubator at 37 °C with constant supply of humidified blood gas. The extracellular field potential signals from electrodes were fed into an amplifier (RZ2, Tucker Davis Technologies) fitted with high-impedance multiple-channel pre-amplifier stage (PZ2–64, Tucker Davis Technologies). 6kHz sampling rate and gain of 1000 were used. OpenEx (Tucker Davis Technologies) and Matlab (MathWorks) were used for signal processing and data analysis. A 100 Hz-3 kHz band pass filter was first applied to the raw data. Digital band stop filters of 60 Hz and its harmonics up to 3kHz was then used to denoise the bandpassed data. Spikes were detected using an automated thresholding method [28]. Recordings were performed every other day from DIV10 to DIV18. Recording time length was 20 min.

Whole Cell Recordings

Whole cell recordings were performed on DIV14. Upright microscope (Olympus BX51WI, Olympus Optics, Japan) equipped with infrared-differential interference contrast optics was used to locate neurons. The recordings were conducted at room temperature (~23 °C), and the resistance of the recording pipette (1.2 mm borosilicate glass, Warner Instruments Inc., Hamden, CT) was 4.5 to 7 MΩ. To record miniature excitatory postsynaptic currents (mini EPSCs), GABAergic transmission was blocked using picrotoxin (PTX; 50 μM, Sigma-Aldrich, St. Louis, MO) and spontaneous action potentials were blocked with tetrodotoxin (TTX; 1μM, Ascent Scientific, Princeton, NJ). The cultures were held in a perfusion chamber with constant flow of regular Ringer’s solution with 50 μM PTX and 1μM TTX. The recording cells were held at −70 mV. The recording pipette was loaded with Cs-based intracellular solution containing 120 mM Cs-gluconate, 5 mM lidocaine (QX-314), 6 mM CsCl₂, 1 mM ATP-Mg, 0.2 mM Na₂GTP, 10 mM Hepes, and 0.3% biocytin (pH 7.3, adjusted with CsOH). 8 three-dimensional and 8 two-dimensional cultures were recorded for mEPSC analysis and comparison. 3 cells were patched from each culture. The mEPSCs were recorded using MultiClamp 700B (Molecular Devices) and acquired at sampling frequency of 20 kHz through DigiData 1322A and pCLAMP 9.2 software (Molecular Devices). The data was analyzed in Matlab. 0.1 Hz high pass and an optimal 100 Hz low pass filters [29] were applied. A bandstop filter was used to eliminate 60 Hz noise, and mini EPSCs were detected using −3pA threshold.

Pharmacological tests

For tetrodotoxin (TTX) and kynurenic acid (KYNA) experiments, the cultures were recorded optically in normal culture medium first. The medium was then changed to 1 μM TTX or 3 mM KYNA in culture medium. Cultures were kept in inhibitor-containing medium for 30 minutes before 2^nd recording. The culture was then washed with fresh culture medium, incubated for2 hours,washed again with culture medium and incubated for 30 minutes. The cultures were then recorded optically for the 3^rd time.

For determining effects of anti-epileptic drugs (Phenytoin and Tiagabine), drugs were dissolved directly in culture medium or sterile distilled water to make stock solution. Cultures were first optically recorded in fresh culture medium. The medium was then changed to culture medium with lowest concentration of drug and kept in incubator at 37 °C and 5% CO2 for an hour. After an hour of incubation, the cultures were optically recorded. After recording, the medium was changed with medium with next higher concentration of drug. The cultures were incubated for an hour and then recorded again. This process was then repeated for all concentrations to be tested. Cultures were then washed as described above and recorded again.

Immunohistochemistry and cell counting

Cultures were fixed in 4% Paraformaldehyde on DIV14 for 1 hour. Cell permeabilization was done with Triton X-100 (Sigma-Aldrich) in PBS for 2 hours on a shaking platform. Cultures were then blocked using 10% goat serum in PBS for 1 hour. Primary antibodies (1:500 anti-NeuN (neuronal nuclei) (Millipore), 1:500 anti-GFAP (Thermo Fisher), 1:500 anti-SOX1 antibody (abcam) and 1:500 Anti-human Nestin (StemCell Technologies)) were then applied to the cultures on a shaker at +4°C for 48 hours, followed by secondary antibodies. Cultures were counterstained with DAPI. For non-adhered spheroid cultures, fixed and stained cultures were sandwiched between two glass coverslips in Fluoro-Gel (Electron Microscopy Sciences, Catalog # 17985–10) with spacers on the sides. Cultures were then imaged using a confocal microscope (Zeiss LSM 510 META, Germany) with 20× objective. Distance between optical slices was 1 μm, and cultures were imaged over their entire depth. Images were then processed in Fiji (ImageJ).

Cell counting was then performed using 3D watershed and 3D ROI manager plugin [30] in Fiji (ImageJ). Number of cells was determined by counting DAPI positive (DAPI⁺) 3d objects exceeding minimum volume. Cells with normal and abnormal (small, concentrated, bright) appearance were differentiated manually. Number of neurons was determined by counting NeuN-positive (NeuN⁺) objects exceeding a volume threshold. GFAP, SOX1 and Nestin positive cells were counted manually in cell counter plugin of Fiji (ImageJ).

Statistical methods

Two-sample Kolmogorov–Smirnov test was used for comparing the burst durations of 2D, 3D, and spheroid cultures of different diameters. Student’s t test was used for comparing effects of inhibitors and drugs on activity. Number of samples n refers to the number of bursts or the number of cultures as indicated.

Results and Discussion

To characterize morphology of 3D and spheroid cultures (Figure 2 (A–D)), we performed immunohistochemical staining and confocal imaging. DAPI was used to detect cell nuclei, anti-NeuN was used to detect neurons. As seen in Figure 2 below, for 1300 μm (Figure 2(I)) and 800 μm (Figure 2 (J)) diameter mini-wells, cultures adapt an almost cylindrical shape of around 100 μm height. Initially, a homogeneous mixture of cells was seeded into mini-well uniformly. As cultures matured, cells from center region migrated towards periphery. For 1300 μm mini-well diameter culture, a second thinner neuronal layer appears around the central vacant region which is not present in 800 μm mini-well diameter culture. For 400 μm mini-well diameter culture, the center vacant region is absent. Instead of being cylindrical, 400 μm cultures adapt a half-spheroid shape (Figure 2(K)). The bottom layer of all 3D cultures was consistently devoid of neurons. Cell counting revealed an increasing neuron to cell ratio with increasing mini-well diameter. This ratio was significantly different between 2D and 1300 μm diameter culture and between 400 μm and 1300 μm diameter culture. This finding was consistent with previous studies, where 3D cultures were found to provide a better neuronal viability than 2D cultures [31–32]. These structural characteristics were highly reproducible if the mini-well diameter and seeding density were kept same – images in Figure 2 are representative of 6 cultures per well diameter.

Figure 2.

Phase contrast image of A) 1300 μm, B) 800 μm, C) 400 μm mini-well diameter 3D and D) 100,000 cell spheroid culture. Projection of brightest points of confocal image stack of E) 1300 μm, F) 800 μm and G) 400 μm mini-well diameter 3D and H) 100,000 cell spheroid culture. Top panel shows DAPI stained cells, middle panel shows anti-NeuN stained neurons. Bottom panel shows merged image of top and middle panel. Cartoon representation of cross section through center of I) 1300 μm, J) 800 μm and K) 400 μm mini-well diameter 3D and L) 100,000 cell spheroid culture. (Not drawn to scale.) Red area indicates anti-NeuN positive neuron rich region. Blue area indicates region with DAPI positive cells and almost no NeuN positive cells or neurons. Bar plot of mean with standard deviation and data points of M) number of DAPI positive cells and Anti-NeuN positive neurons for different types of cultures (n=3 for all cases in panel M and N, n being the number of culture), N) Ratio of non-pyknotic cell count to total cell count for different types of 3D cultures and shell and core of spheroid culture. (In each case of panel N, 3 confocal planes form 3 different cultures were used, asterisks indicate the p-values of Kolmogorov–Smirnov KS test between the cultures indicated by lines. *=p<0.05, **=p<0.01, ***=p<0.001).

Spheroid cultures, on the other hand, showed a core-shell like morphology. After seeding, the cell suspension in ultra-low adherence concave wells started forming small clusters by DIV 03 (Supplementary Figure 2(A)). These small clusters then merged into larger clusters by DIV 05 (Supplementary Figure 2(B)). All the clusters finally merged into a single non-uniformly shaped culture by DIV 08 (Supplementary Figure 2(C)). We varied the number of seeded cells (50,000, 100,000 and 200,000) to create spheroids of varied sizes. In each case, the culture morphology was non-uniform (Supplementary Figure 2(D–F)). Immunohistochemistry revealed many bright but small DAPI positive objects in the core and a thin NeuN⁺ shell in these cultures (Figure 2 (H, L)). These bright but small DAPI⁺ objects are shrunk nuclei in cells undergoing pyknosis [33], an irreversible process of nuclear shrinkage and chromatin condensation which eventually leads to cell apoptosis and necrosis. The ratio of non-pyknotic (healthy) to total cell count in both core and shell of 100,000 cell spheroid cultures was significantly lower than that of 3D cultures of all sizes (Figure 2(N)). Although the 1300 μm 3D cultures are created using more cells (120,000) than 100,000 cell spheroids, they had a higher non-pyknotic to total cell ratio. Moreover, 3D cultures also yielded a higher neuron to cell ratio than spheroid cultures (Figure 2(M)).

To the best of our knowledge, this is the first report of neural layer self-assembly in 3D cultures by post-natal neurons. Earlier reports of neurospheroid microstructure showed relatively homogeneous distribution of neurons throughout 3D culture volume [6, 9, 15]. These spheroids were limited to diameters of 100 to 200 μm to avoid formation of a toxic core due to diffusion limitations of oxygen and other required molecules [34]. However, limiting culture diameter also limits the number of neurons in a spheroid, and prevents layer formation (our smallest 3D cultures also did not possess a layered structure (Figure 2 (G, K)). Increase in spheroid size results in increase in cell toxicity (Figure 2 (N)), consistent with previous reports [35]. Our method of producing 3D neural cultures is different from previous works in 2 key aspects: (1), height:width aspect ratio of our cultures ranges from 1:3.5 to 1:10 compared to neurospheroids’ ratio of ~1:1, and (2), our cultures are grown on an adhesive substrate at the bottom of the well. Thus, our 3D cultures are maintained in conditions that are more similar to organotypic slice cultures than neurospheroid cultures. Since the thickness of our 3D cultures was limited to 150 μm, diffusion limitations were avoided, and large cultures could be created without an accompanying increase in cell toxicity. It may be possible that same processes that lead to long-term maintenance of neural layers in organotypic hippocampal or cortical cultures also lead to formation of neural layers in low-aspect ratio, adherent 3D clusters in this work.

To examine the neuronal activity in our cultures, fluorescent changes in jRGECO1a infected cultures were recorded optically. This method provided a robust way to observe the whole culture’s neuronal activity simultaneously. As the cultures matured, rat cortical 2D, 3D and spheroid cultures started to exhibit sharp calcium spikes, including synchronized network-wide spikes (population spikes or ‘bursts’), consistent with previous studies [35–36]. These events could be observed from DIV 8 for 2D and 3D cultures. Spheroid cultures started to show synchronized activities from DIV11. After DIV 11, 2D cultures didn’t appreciably change their activity pattern. 3D and spheroid cultures, on the other hand, started showing complex and long repetitive population bursts. These findings can be seen in Figure 3. The optical activities were first converted into ΔF/F and then to activity index x(t) as described in data analysis subsection of methods section. This removed influence of slow calcium and jRGECO1a decay transients on burst duration analysis.

We then compared activity in confluent 2D cultures in 7 mm diameter wells with similar network size of a 1300 μm diameter 3D culture and 3D cultures in 400, 800, and 1300 μm mini-wells and spheroid cultures created with 50,000, 100,000 and 200,000 cells (Figure 3(B, C)). 2D cultures exhibited significantly lower burst durations than 3D and spheroid cultures (Figure 3 (A)). Cumulative distribution function of all detected bursts across all DIVs for different cultures shows strikingly short burst durations for 2D cultures compared to other culture types (Figure 3(D)). All cultures exhibited roughly consistent activity pattern from DIV 15 to 19. We also compared durations of all detected bursts during this period for different culture types and found that 2D culture bursts were consistently significantly lower than all types of 3D and spheroid cultures (Figure 3(F)). 3D culture and spheroid culture burst durations were not consistently different.

Figure 3.

A) Activity of 2D and 1300 μm mini-well 3D culture on different DIVs. Red lines indicate duration of detected activities B) Barplot with standard deviation of the mean normalized active time for cultures on different DIVs. n=4 for 1300 μm and 800 μm diameter culture. n=6 for 400 μm diameter culture, n=3 for all spheroid and 2D cultures. Here, n is the number of cultures recorded. C) Barplot with standard deviation of the mean of the longest 20% of burst durations of different types of cultures on different DIVs. Points indicate individual bursts (for 1300 μm diameter culture, n=2,3,4,2,6,4; for 800 μm diameter culture, n=1,3,14,4,5,4; for 400 μm diameter culture, n=3,3,20,6,7,10; for 50,000 cell spheroid cultures n= 0,6,10,3,9,11; for 100,000 cell spheroid culture n=0, 4, 6, 6, 19, 18; for 200,000 cell spheroid culture, n=0,7,11,5,8,7; and for 2D culture, n=2,4,5,7,15,22 for DIV 9,11,13,15,17, and 19 respectively. Here, n represents longest 20% of all detected bursts). D) Cumulative distribution function of all burst durations from all DIVs. Inset shows zoomed-in portion of CDF plot. Results of two-sample Kolmogorov-Smirnov test between the E) normalized total active time and F) detected burst durations of the culture types indicated on x-axis vs. culture types indicated on y-axis. Colored boxes reject the null hypothesis at 5% significance level. (For panel E and F, * = p<0.05, **=p<0.01 and ***=p<0.001. For Panel E, n=12,12,18,12,12,12 and 9 for 1300, 800 and 400 μm diameter 3D, 200,000, 100,000, 50,000 cell spheroid and 2D cultures respectively. Here, n is normalized total active time from DIV15–19. For panel F, n=56, 58, 109, 112, 208, 92 and 214 for 1300, 800 and 400 μm diameter 3D, 200,000, 100,000, 50,000 cell spheroid and 2D cultures respectively. Here, n is number of detected bursts from DIV15–19. In panel A, scalebars indicate 100 seconds and Activity index=2.5.

2D cultures did not consistently exhibit significantly different active time from DIV 15 to 19 (Figure 3(F)) when compared to 3D or spheroid cultures. This shows that in 3D or spheroid culture, the overall neuronal activity patterns were rearranged to yield long network bursts, but there was not an overall increase of activity.

Next, we recorded the electrical activities of 1300 μm diameter cultures with micro-electrode array (MEA) (Figure 4 (A)). MEA recording showed similar activity patterns as jRGECO1a imaging and confirmed the high firing rate and network-wide synchronization of neurons during long bursts in 3D cultures.

Figure 4.

A) Extracellular recording from 3D culture. The culture was grown on Micro-electrode array (MEA). 16 channels (Indicated by red circles with electrode number) were used to record extracellular potential across different points of the culture. (B-H) Raster plots derived from Extracellular recording recorded on different DIVs. Each vertical line here represents a neuronal spiking and each row represents recording from one electrode. X-axis represents time in seconds. I) Color plot of firing rate of a randomly selected culture on different DIV.

Spontaneous, synchronized Ca²⁺ activity is present in the developing cortex [37–40]. Analysis of Ca²⁺ transients in slices of neonatal rat cortex demonstrated presence of 1 – 20 sec long bursts occurring with 1–8 minute interburst intervals [38] – strikingly similar to spontaneous activity patterns we found in our 3D, but not 2D cultures. It is therefore possible that 3D cultures are closer to replicating the naturally occurring spontaneous activity in the developing cortex than 2D cultures. This may be due to the higher and more in vivo – like neural density of 3D cultures ((11+/− 3) x 10⁴ neurons/mm³ compared to average cortical neuronal density of 9.2 × 10⁴ neurons/mm³ [41]), leading to formation of more in vivo – like network.

Our analysis of mini EPSC amplitude confirms this hypothesis: the size of mini EPSCs in 3D cultures was close to that in cortical slices and significantly smaller than mini EPSCs in 2D cultures (Figure 5(B, D)). The probability of a synapse existing between neurons in a network is proportional to network density, while synaptic conductivity is inversely proportional to the number of synaptic connections [42]. 3D cultures have a significantly higher density of neurons than 2D cultures – therefore, the number of synaptic connections per neuron in 3D culture may be higher than corresponding number in 2D culture. Conductivity of individual synapses, on the other hand, would be lower in 3D culture than in 2D culture. We also found significantly higher interval between mini EPSCs of 3D cultures than 2D cultures (Figure 5(C, E)). As mentioned earlier, 3D cultures have a significantly lower mini-EPSC amplitude than 2D cultures; however, while analyzing data, we maintained a constant threshold (3 pA) for both types of cultures. This may have led to undercounting of some mini-EPSC spikes in 3D cultures, yielding a higher interval statistic than 2D cultures. These network differences may help to explain the differences we found in the patterns of spontaneous activity.

Figure 5.

A) Whole cell patch clamp recording of spontaneous mini EPSC in 2D and 3D culture. B) Cumulative frequency of detected mini EPSC amplitude in 2D and 3D cultures. Inset figure shows mean with standard deviation. C) Cumulative frequency of intervals between detected subsequent mini EPSCs in 2D and 3D cultures. Inset figure shows zoomed in portion of CDF trace (left) and mean with standard deviation (right). (*** = Kolmogorov–Smirnov KS test P < 0.001) E & F) Each dot represents one E) Amplitude of detected mini EPSC and F) interval between two subsequent mini EPSCs.

Application of tetrodotoxin (TTX) abolished all population [Ca²⁺] bursts (Figure 6 (A)), indicating that [Ca²⁺] bursts are dependent on voltage-gated sodium channels and thus neuronal firing [43]. Application of kynurenic acid (KYNA) also showed equivalent results (Figure 6 (B)). KYNA is an antagonist of AMPA, NMDA and kainate glutamate receptors. Suppression of bursts due to application of KYNA demonstrates that generation of spontaneous activity in our 3D cultures depends on excitatory synaptic activity.

Figure 6.

Activity of 3D culture before, during and after application of A) TTX and B) KYNA. 3 Cultures were tested for each case. Mean with standard deviation of C) normalized total activity duration and D) Normalized burst duration at different concentration of Phenytoin application. Mean with standard deviation of E) normalized total activity duration and F) Normalized burst duration at different concentration of Tiagabine application. Solid lines in panel C-F are fitted sigmoid curves. The curves were fitted using van Genuchten–Gupta model: y=1/(1+(c/k1) ^k2). C) k1=80, 90, k2= 4, 5 and Pearson’s correlation coefficient r²=0.9243 and 0.9309 for 3D and 2D curve-fitting respectively D) k1=60, 170, k2= 3, 12 and Pearson’s correlation coefficient r²=0.88 and 0.89 for 3D and 2D curve-fitting respectively.E) k1=3000, 18000, k2= 1.45, 1.9 and Pearson’s correlation coefficient r²=0.9182 and 0.8281 for 3D and 2D curve-fitting respectively F) k1=5500, 55000, k2= 5, 5 and Pearson’s correlation coefficient r²=0.9714 and 0.9376 for 3D and 2D curve-fitting respectively. 4 cultures were tested for each case in panel C-F.

Long, high frequency population firing exhibited by our 3D cultures resembles seizure-like activity in organotypic hippocampal cultures [44–45] as well as developmental activity described above. Organotypic hippocampal cultures are generated from postnatal day 7–8 animals where developmental spontaneous activity is minimal, as opposed to our 3D cultures which were generated from postnatal day 1–2 animals. Nevertheless, we hypothesized that since burst activity in 3D cultures requires voltage-gated sodium channels and excitatory synaptic transmission, bursts in 3D cultures may be sensitive to antiepileptic drugs similarly to seizure-like activity in organotypic hippocampal cultures [44, 46]. Application of phenytoin, an anti-epileptic drug (AED), revealed a sigmoid relationship between concentration of phenytoin and the ratio between activity time and total recording time (Figure 6 (C)). Phenytoin had a similar effect on total activity time for both 2D and 3D cultures. On the other hand, phenytoin decreased burst duration in 3D cultures at a lower concentration than in 2D cultures (Figure 6 (D)). One of the mechanisms for phenytoin’s anticonvulsant effect is use-dependent inactivation of fast sodium channels [47–48]. Same mechanism may be responsible for inhibition of bursts in our cultures.

Concentration response of another antiepileptic drug, Tiagabine, was significantly different in 2D and 3D cultures. Tiagabine is a GABA uptake inhibitor and works specifically at synapses [49] where it inhibits the uptake of GABA receptors hence facilitating inhibition. From our observations, it might be hypothesized that synapses in 3D cultures were more susceptible to Tiagabine than synapses in 2D cultures. EC₅₀ concentration for activity suppression in 3D cultures was 5.5 μM compared to EC₅₀ = 55 μM in 2D cultures. EC₅₀s for inhibition of convulsions in mice and rats were significantly closer to EC₅₀s in 3D cultures than 2D cultures [49]. This finding provides evidence that our 3D cultures with self-assembled neural layers had a more in-vivo like anticonvulsant dose response relationship than 2D cultures.

We then examined whether hiPSC-neurons 3D cultures in mini-wells develop self-assembled neural layers and long spontaneous bursts of activity similar to 3D cultures of neonatal rat cortical neurons. We found that hiPSC-neurons formed 3D clusters that were significantly smaller than the diameter of the PDMS well, but NeuN-positive cells formed a hollow cup-like structure (Figure 7 (B)). Cultures were approximately 120 μm high and had a similar neuron to cell ratio as rat cultures (46%). Changing the cellular composition by changing ratio of cell types allowed us to investigate effects of excitatory to inhibitory ratio on activity patterns (100% GlutaNeurons, 75% GlutaNeurons+25% GABANeurons, 50% GlutaNeurons + 50% GABANeurons and 100% GABANeurons, 3 cultures each). hiPSC-derived astrocytes were added to all of these cultures to make a Neuron:Astrocyte ratio of 85:15.

Figure 7.

Phase contrast image of A) hiPSC-neuron 3D cultures (75% iCell GlutaNeuron+25% iCell GABANeuron and 15% iCell astrocytes) and C) 200,000 cell spheroid culture at day in culture 34 (Top panel). Projection of brightest points of confocal image stack of hIPSC 3D cultures for DAPI and anti-NeuN staining. Bottom panel shows merged image of DAPI and Anti-NeuN staining. Internal structure of B) hiPSC 3D culture and D) spheroid culture (Not drawn to scale). E) Plot of mean with standard deviation of DAPI⁺ and NeuN⁺ cells in different regions of 3D cultures and 200,000 cell spheroids (both 100% iCell GlutaNeuron and 15% astrocytes) fixed on day 50. The counts were normalized with respect to the total cell count for the respective culture. F) Ratio of non-pyknotic to total nuclei in different regions of the 3D and spheroid cultures (n=3 optical slices per condition) G) Representative spontaneous activity of 3D culture (75% iCell GlutaNeuron+25% iCell GABANeuron and 15% iCell astrocytes) and 200,000 cell spheroid culture (100% iCell GlutaNeuron and 15% iCell astrocytes). H) Bar plot of mean with standard deviation of longest 20% of detected burst durations (for 100% GlutaNeuron 3D culture n=12, 5, 2, 1,4, 2, for 75% GlutaNeuron 3D culture n=5, 2, 6, 8, 3, 6, for 50% GlutaNeuron 3D culture n=7, 7, 5, 3, 4, 5 for 200,000 cell spheroid n=0, 0, 16, 12, 4, 9 and for 100,000 cell spheroid culture n=0, 0, 13, 12, 4, 4 for day 20, 24, 30, 34, 42 and 50 respectively) and of J) Normalized total activity duration (n=3, 2, 1, 3 and 3 for 100%, 75%, 50% GlutaNeuron 3D and 200,000 and 100,000 cell spheroid respectively for all days. Here n is the normalized total active time for one culture on a certain day). Bar plot of mean and standard deviation of I) all of longest 20% burst durations detected from day 30 – 50 (n=9, 23, 17, 41 and 33 for 100%, 75%, 50% GlutaNeuron 3D and 200,000 and 100,000 cell spheroids, respectively) and K) Normalized total active time (n=8, 8, 4, 12 and 12 for 100%, 75%, 50% GlutaNeuron 3D and 200,000 and 100,000 cell spheroids, respectively, here n is normalized total active time for a particular day) across day 30 to day 50. (Asterisks in panel F,I and K indicate results of two-sample KS test,, * p<0.05, **p<0.01 and ***p<0.001. Scale bars in (G) indicate 100s along x-axis and activity index = 2.5 along y-axis).

To compare our 3D cultures with existing methods, we created spheroid cultures from 100% GlutaNeurons with addition of astrocytes at 85:15 Neuron:Astrocyte ratio, in ultra-low-adherence concave wells. We varied the number of cells seeded (100,000 and 200,000) to vary the culture size. After seeding cells, similar to rat spheroids, the cells aggregated into small cell clusters and eventually merged into single highly irregularly-shaped spheroid (Figure 7(C)). Immunohistochemistry revealed a core-shell type structure similar to rat cortical spheroid with an almost hollow core and a NeuN⁺ shell (Figure 7(C, D)). The shell structure was not uniform and was discontinuous (Figure 7(C)). Once formed, these spheroids and 3D cultures did not appreciably change their shape. The cross-section area of these cultures shrank slightly with time in culture (Supplementary Figure 3), indicating the absence of proliferating cells.

We found substantial numbers of bright and small DAPI⁺ objects that were qualitatively different from relatively large and somewhat dimmer DAPI⁺ nuclei of NeuN⁺ cells. To determine whether these abnormal DAPI⁺ objects represented nuclei of non-neuronal or not fully differentiated cells, we stained our cultures with antibodies to GFAP, SOX1 and Nestin. Bright and small DAPI⁺ objects were negative for all these markers, while DAPI⁺ nuclei of GFAP⁺, SOX1⁺, or Nestin⁺ cells were larger and dimmer (Supplementary Figure 4). Thus, small and bright DAPI⁺ objects likely represent nuclei of pyknotic dead cells. To analyze relative levels of cell death in 3D cultures, we divided them into core and radial regions. The core region of 3D culture is the middle bottom neuron deprived area (Figure 7(B), bottom of the ‘cup’) and the radial region consists of the side walls of the ‘cup’. We divided the spheroid culture into two regions as well: core (mostly empty or pyknotic, devoid of NeuN⁺ cells) and shell (NeuN⁺ cell-rich outer region). We found significantly higher neuron to cell ratio in our 3D culture radial region compared to 3D culture core, spheroid core or spheroid shell region (Figure 7(E)). We found significantly lower relative numbers of pyknotic nuclei in both core and radial regions of our 3D cultures compared to both core and shell regions of spheroids (Figure 7(F)). The core-shell morphology in spheroid cultures is thus likely a by-product of cell death and is in line with previous observations in large hiPSC-derived organoids [50]. Anti-GFAP staining revealed the presence of astrocytes in both 3D and spheroid cultures, as expected (Supplementary Figure 5).

Next, we compared the activity patterns in our 3D cultures with varied cell composition and spheroid cultures with varied size. 100% GABANeuron cultures failed to exhibit any bursting activity. All other cultures exhibited long burst episodes similar to rat cortical cultures. 3D cultures started to show synchronous activity from around day 20 and spheroid cultures became active from day 30. Evolution of spontaneous activity of one 3D culture (75% GlutaNeurons+25% GABANeurons + iCell Astrocytes) and one 200,000 cell spheroid culture (100% GlutaNeuron + iCell Astrocytes) is shown in Figure 7 (G), with quantification of longest 20% burst duration in Figure 7 (H) and total activity duration in Figure 7 (J). We then compared the longest 20% burst duration from day 30 to 50, when all culture types were active. We found that addition of GABAneurons significantly decreased burst duration (Figure 7(I)). This result suggests that GABA is inhibitory in these cultures and can be used as an effective tool for burst duration modulation. We also observed that spheroid cultures have significantly lower burst duration than 100% and 75% GlutaNeuron 3D cultures (Figure 7(I)).

Apart from 100,000 cell spheroid, total active times for different culture type were not significantly different from each other (Figure 7(J, K)). This can indicate that our 3D culturing technique results in formation of network architecture that is more favorable to bursting than networks in spheroids.

We then applied TTX and KYNA to these 3D cultures (Figure 8 (A, B)). Application of both stopped all bursting activity, which resumed after wash. 3 cultures were tested in each case. This indicates that Ca²⁺ transients depended on neuronal firing and glutamatergic synapses. We also found that anticonvulsant phenytoin has a dose-dependent inhibitory effect on bursts with a lower EC₅₀ (35 μM) than in rat cortical 3D cultures (80 μM) (Figure 8 (C, D)).

Figure 8.

Before, during and after wash of A) TTX and B) KYNA on hiPSC -neuron 3D cultures. Mean with standard deviation of C) normalized total activity duration and D) Normalized burst duration at different concentration of phenytoin application to 3D cultures from hIPSC derived cells on day in culture 40. Solid line is sigmoid-fitted curve. The curves were fitted using van Genuchten–Gupta model: y=1/(1+(c/k₁) ^k2). G) k₁=35 and k₂= 8.5. Pearson’s correlation coefficient r²=0.9127 H) k₁=35, k₂= 3.5. Pearson’s correlation coefficient r²=0.9407. 3 (75% iCell GlutaNeuron+25% iCell GABANeuron and 15% iCell astrocytes) cultures were tested for each case for panel.

Recently, mini-well confined hiPSC-derived neurospheroids were used to model Alzheimer’s disease [17]. Interestingly, neuron distribution in spheroids reported in that work appeared to be uniform rather than layer-like in our 3D or spheroid cultures. The smaller culture size can be the main reason of this difference. The core-shell regions of spheroid culture in this work can be viewed as layered structure like the core-radial region of the 3D cultures. However, as described, the outer neuronal layer of spheroid culture comes at a cost of highly pyknotic core where the developed 3D culturing technique yields a more stable neuronal layer without the cost of excessive pyknosis. Similar to neurospheroids derived from rat neurons, the key difference appears to be the lower aspect ratio and substrate adherence of our 3D cultures. Our results from rat and human neurons thus suggest that layer self-assembly is induced by the engineered features of the culture substrate. A key advantage of using this method of layer self-assembly in 3D cultures may be its reproducibility compared to cortical-like layer formation in ‘mini-brain’ or ‘mini-cortex’ constructs. The latter are established by differentiation of 3D clusters of neural progenitor cells and are subject to variability induced by heterogeneity of cell populations at various stages of differentiation. In contrast, our method starts with differentiated neurons and astrocytes whose homogeneity may be ensured by quality control methods prior to 3D culture initiation. While the layers formed by 3D cultures in this work had simple cylinder-like shapes, complex and in vivo-like layers may be obtained in future work by using more complex well geometries.

Conclusion

We have developed a method for creating scaffold-free three-dimensional cultures with self-assembled neural layers from rat and hiPSC-derived neurons. We compared this method with spheroid culturing technique and observed that our 3D culturing method yields more stable and healthy culture with significantly longer population bursts. The power of this technique lies in its simplicity, reproducibility and amenability to further engineering of 3D geometry. 3D constructs created with this method have morphology that is closer to neural layers found in the cortex or hippocampus than what can be achieved by traditional neurospheroid 3D culture methods. Our 3D constructs may thus be a more accurate in vitro model of the brain. This is supported by our finding that spontaneous activity in 3D constructs is similar to activity in the developing cortex. We also demonstrated that anticonvulsants tested in 3D constructs have a more in vivo-like pharmacological profiles than 2D cultures. 3D constructs created with our method may thus be suitable as in vitro models for drug discovery for neurological disorders.