[1] 13w 3-D geometry and irregular connectivity dictate neuronal firing in frequency domain and synchronization

[1] 224w The replication of the complex structure and three dimensional (3-D) interconnectivity of neurons in the brain is a great challenge. A few 3-D neuronal patterning approaches have been developed to mimic the cell distribution in the brain but none have demonstrated the relationship between 3-D neuron patterning and network connectivity.Here, we used photolithographic crosslinking to fabricate in vitro 3-D neuronal structures with distinct sizes, shapes or interconnectivities, i.e., milli-blocks, micro-stripes, separated micro-blocks and connected micro-blocks, which have spatial confinement from "Z" dimension to "XYZ" dimension. During a 4-week culture period, the 3-D neuronal system has shown high cell viability, axonal, dendritic, synaptic growth and neural network activity of cortical neurons. We further studied the calcium oscillation of neurons in different 3-D patterns and used signal processing both in Fast Fourier Transform (FFT) and time domain (TD) to model the fluorescent signal variation.We observed that the firing frequency decreased as the spatial confinement in 3-D system increased. Besides, the neuronal synchronization significantly decreased by irregularly connecting micro-blocks, indicating that network connectivity can be adjusted by changing the linking conditions of 3-D gels. Earlier works showed the importance of 3-D culture over 2-D in terms of cell growth. Here, we showed that not only 3-D geometry over 2-D culture matters, but also the spatial organization of cells in 3-D dictates the neuronal firing frequency and synchronicity.

[1] 68w Degeneration, trauma and other pathological changes in neural system lead to Parkinson's disease, Alzheimer's disease and other neurological conditions and disorders, which lower the life expectancy and quality of life for patients. Reproducing in vitro brain surrogates can potentially unveil the relationship between structure and function of brain connectome, which could then be used as a high-throughput platform for real-time study of neurodegenerative diseases and drug screening [1].

[2] 127w Neuronal connectivity is the most fundamental characteristic of neuronal systems [2]. Intercellular connections of adjacent neurons are the foundation of signal transduction, material exchange, three-dimensional (3-D) network structures, and all major neuronal functions. Therefore, connecting neurons in in vitro cultured neural tissues is greatly desired as it incorporates the feature that healthy neural tissues rely on to work. In tissue engineered neural substitutes, connected neuron network allows re-establishment of communication between distal ends of wounded tissue upon implantation [3]. Signal transduction is also an important endpoint for drug testing on in vitro platforms [4]. However, the development of 3-D in vitro brain surrogates offers great challenge in terms of recapitulating the correct microenvironment, topological architecture and physiological function of the brain, not to mention the connectome study.

[3] 276w Several approaches have been developed to recapitulate the in vivo brain counterparts in vitro. Scaffolds [5,6], hydrogels [7][8][9], colloidal support [10], and organoid building blocks [11,12] are mainly used to fabricate 3-D neuronal tissue models. The advantage of hydrogels is that they can provide a suitable micro-environment, including the biological and mechanical support, for neuron growth with controllable architecture. For instance, collagen [13,14], fibrin [15], silk-protein [16] and gellan gum [17] hydrogels have been used to fabricate 3-D neuronal tissue with the layered brain cortex structure, but none have demonstrated the connection between patterned architecture and connectivity of neurons. Moreover, among the patterning strategies, the methods based on manual manipulation [17] and self-aggregation of cells (organoids) [18] lack precise controllability over architectures, which hinders them from being ideal models for connectivity studies. The methods based on specialized micro-fabrication techniques [13] or 3-D printing equipment [19] have limited feasibility for largescale application and rapid fabrication. Brain-on-a-chip models combined with microelectrodes arrays (MEAs) are widely developed methods for directed neural network analysis [20]. However, most of the studies on microfluidic chips and MEA substrates are confined to two-dimensional (2-D) neuron cultures. In a few studies that used 3-D hydrogels [21,22], the architecture of chip design were restricted to the mechanism of electrodes arrays, which may be not physiologically relevant. Moreover, brain-on-a-chip models usually employ the smallest possible number of neurons and volume of tissue. Therefore, the influence of size and geometry of in vitro 3-D cultures on neuron connectivity has not been well documented. This limited the ability to design pre-connected tissue engineering products and representatives of micro neural models for larger in vivo tissues of relevant dimensions.

[1] 113w We have previously shown that a simple photolithographic approach enabled precise digital modulation of 3-D architecture without requiring specialized micro-manipulation techniques [23]. We successfully altered primary neuronal growth inside the 3-D microstructure by tuning pattern sizes. In this study, in vitro 3-D neuron cultures in the form of millimeter scale squares, micrometer scale stripes, and micrometer scale squares (Figure 1) were fabricated by seeding fetal mouse primary neurons at similar densities in methacrylated gelatin (GelMA). GelMA was photo-crosslinked using photomasks (Figure S1) with low polymer concentration, high transparency and designed architecture. These geometries apply "1" dimensional, "2" dimensional and "3" dimensional (from "X" dimension to "XYZ" dimension) physical confinements to the neuron populations.

[2] 142w The neuronal connectivity and inter-population activity of individual neurons were characterized by calcium signal oscillation. Ca 2+ serves as a ubiquitous secondary messenger in neurons. Thus, the spatio-temporal features of intracellular and intercellular communications can be visualized by Ca 2+ transmission between the ion channels on neuronal membrane, that leads to the change in concentration of intracellular Ca 2+ [24] . Here, we designed 3-D neural tissue constructs with different neuronal connectivities to study the synchronized burst of non-planar network by tracing Ca 2+ fluorescent signal and analysis with Fast Fourier Transition (FFT) and time domain (TD) modeling. Our methodology for 3-D engineering of neuronal tissues was precisely controlled in both architecture and connectivity, providing a powerful platform for functional study of brain in vitro. The result of this study was expected to reveal the influence of physical confinement on neuronal interactions.

[3] 27w medium were seeded on to 24 well plates which were coated with 1 mg/ml poly-D-lysine for 1 h and wash with ultrapure water 2 × 5 min.

[4] 85w Preparation of 3-D neural tissues. The crosslinking of the 3-D hydrogel is based on photo-crosslking of the double bond of the Acryl on GelMA (Figure S1b). 15 mg of photoinitiator (PI) (2-Hydroxy-4′-(2-hydroxyethoxy)-2-methylpropiophenone, Irgacure 2959, Sigma-Aldrich) was dissolved in 5 mL PBS (Life Technologies) for 2 h at 55 °C. The 0.3% PI solution was added to freeze dried GelMA to generate a 2% (w/v) precursor solution, filtered through a 0.2 µm sterile filter (VWR) and kept at 37 °C typically within 1~2 hours until use.

[5] 122w Immediately after cortical neuron extraction, cells were centrifuged at 1200 rpm for 3 min. The supernatant was discarded and neurons were carefully resuspended in the 37 °C pre-warmed GelMA precursor solution at a concentration of 1 × 10 7 cells/mL. 100 µL of neuron-suspended GelMA precursor solution was pipetted onto a customized silicon base (Fullchance Industrial CO.) with a spacer of 100 µm thickness. The solution was carefully covered with a glass cover slip (VWR, 18 mm x 18 mm) to avoid air bubbles, and crosslinked by ultraviolet (UV) source (Omnicure® S2000 UV/Visible Spot Curing System) at an intensity of 2.9 mW/cm 2 for 25 s (Figure S2). Photomasks with different patterns were used to control the morphology of 3-D neural tissues.

[6] 261w In this research, we intended to classify the size scale in milli-scale and micro-scale, hence, we chose a milli-scale block pattern, a micro-scale stripe pattern and a micro-scale block pattern to study the influence of geometry confinement on neuronal firing in 3-D condition. 4 mm × 4 mm squares (Figure 1a), 200 µm wide stripes with 100 µm gaps (Figure 1b) and 500 × 500 µm squares with a distance of 250 µm in each direction (Figure 1c) were used as pattern templates, which served as represented patterns of each pattern category. To prepare the GelMA backfilled sample (Figure 1cii), cell loaded GelMA was cross-linked under a 500 × 500 µm squares photomask for 15 s and washed in pre-warmed PBS for 10 min, after the cells in uncrosslinked regions were washed away, the spare regions were filled with GelMA precursor solution without cells and UV crosslinked for 10s. After crosslinking and PBS washing steps, the coverslip with neuron encapsulated tissue hydrogel elements was carefully removed from the silicon base and placed into a petri dish containing 2 mL of 37 °C pre-warmed neural culture medium. Medium was changed every 3 days by replacing 50% with fresh medium. For the same batch of experiment, GelMA and primary neurons were from the same batch. In different batch of experiment, GelMA was from the same batch or different batches, primary neurons are from different isolation. Although the cells and GelMA were from different batches, we got similar results. "N = 2" for GelMA and "N = 3" for each type of cell experiments.

[7] 54w Live/Dead assay. The viability of cortical neurons in 3-D structure was assessed with a live/dead assay at day-invitro (DIV) 1, 3 and 7. 3-D neural tissues were incubated for 10 min at 37 °C in neural culture medium containing 4 µM calcein-AM (green, Fisher Scientific) solution and 2 µM ethidium homodimer-1 (Red, Fisher Scientific).

[8] 34w Samples were gently washed twice with pre-warmed culture medium and imaged under Zeiss AxioObserver Z1 microscope with FITC and TRITC filters. The number of live and dead cells were counted by ImageJ for each

[9] 251w Soft and transparent hydrogel was formed after UV crosslinking. The storage modulus of the GelMA hydrogel is ~212 Pa (Figure S3), which is in the same order of the brain modulus [28]. On day-in-vitro (DIV) 1, 3, and 7, we acquired neuron viability by fluorescence images of live/dead staining. From DIV 1 to DIV7 (Figure 2a), the viability of neurons remained constant at approximately 80 % of total cells encapsulated in the 3-D hydrogels (Figure 2b) without significant increase of cell death as culture time increased. This is in agreement with our previous studies on the viability of rat neurons grown in GelMA hydrogels [23]. Earlier studies by others on 3-D cultures have reported a decrease in viability of neurons in 3-D cultures within the first 5 days of culture [29]. We therefore focused our studies of neurons survival in 3-D GelMA hydrogels to these early time points. To study the cell composition in 3-D neuronal network, anti-β-tubulin/anti-GFAP co-staining was performed to identify the ratio between neurons and astrocytes (Figure 2c) and anti-CaKII/anti-GAD67 co-staining was performed to identify the ratio between excitatory neurons and inhibitory neurons (Figure 2d). There were 73.7% neurons and 4.5% astrocytes, considering astrocytes constituted 20% of the glial cell [30], the neuron-glial cell ratio was ~3, which is close to ratio in mouse cortex in literature (2.78) [31]. The majority of the neurons are excitatory neurons (81.5% of the total number of neurons), similar to the ratio of excitatory to inhibitory (4 to 1) in vivo [32].

[10] 88w In our study, the term 3-D refers to the environments that the cells are exposed to, which are significantly differ In order to study the mature neuronal network formation, neurons in 3-D tissues were analyzed according to their development over time. In particular, the length of axons and dendrites was analyzed in immunofluorescent images (Figure 3a). The length of axons (specified with anti-Tau-1 antibody, Figure 3b) and dendrites (specified by anti-Map2 antibody, Figure 3c) were chosen as a factor to determine the maturation of the 3-D neuronal tissue.

[11] 296w The quantification of presynaptic contacts (specified by anti-synapsin-1 antibody) was further chosen as an indication of neuronal network formation. On DIV 7, neuronal network formation was still at the beginning level with little axons and dendrites outgrowth, and few synaptic connections were visible. During DIV 7 to DIV 14, length of axons and dendrites as well as presynaptic density experienced a rapid increase, which saturated on DIV 21 to DIV 28. These are agreed with Zhou et al.'s results using neural stem cells in GelMA hydrogel, who also observed significant neurite growth in the beginning 2 weeks of culture [33]. Moreover, the length of axons were ~4 times that of dendrites in DIV 7, and ~2 times of that in DIV 14-28 (Figure 3b,c). Earlier study from literature shown that dendritic protrusions form at the early stage of neurite growth, yet once the break in morphological symmetry occurs, one of the neurites turns into axon and the length protrudes rapidly, with slower dendritic growth at the meantime [34]. During DIV 7-14, the fast increase of dendritic arborization and dendrite growth is an important multi-step biological process to create new synapses [35]. Accordingly, the number of synaptic contacts also increased significantly from 111 ± 40 in DIV 7 to 16148 ± 1046 in DIV 14 (Figure 3d). The formation of neurite network and synaptic connections are necessary to generate neural polarization and form functional neuronal networks [36]. During DIV 14 to 28, neurons continued to maintain fine 3-D networks with uniformly distributed neurons, which indicated the sufficient maintenance via the GelMA polymer network despite the long culture time. Previous reports suggested that methacrylamide crosslinks might increase the resistance to enzymatic degradation of GelMA [37], which may increase the life time of GelMA hydrogel used in this study.

[12] 163w Since synapsin-1 is a marker of presynaptic vesicles found only on axons, this allows for analysis of the connectivity of the neuronal network development over time [38]. The overlaid image of axon, dendrite and synapses at DIV 14 show that synaptic vesicles were distributed along the axons (Figure 3e). The number of synapsin-1 puncta per mm of axons increased significantly from 6 ±2 to 433± 28 between DIV7 and DIV14 and slightly decreased to around 350 over time after DIV21 (Figure 3f). In previous studies based on 2-D culture, the number of presynaptic vesicles reached 50~140/mm axon at around DIV7 [39,40], but after the axon network formation, the redistribution of presynaptic vesicles led to an increase of puncta size and a decrease of puncta density [41]. Comparing with 2-D culture, neural network formation in our 3-D architecture is slower, but the synaptic punctum density per mm axon reached a much higher value when saturated, thus indicating a mature connectivity of the neuronal network.

[13] 83w Recapitulating the functional activity of native tissues is important for tissue engineering, regenerative medicine, and drug screening applications. In this study of 3-D engineered cortical neural tissues, we used the chemical fluorescent indicator Fluo-4 of Ca 2+ to monitor changes in cellular calcium levels in response to Ca 2+ -mediated action potentials, an indicator of neural network activity. Oscillations in intracellular calcium ion concentrations ([Ca 2+ ]) serve as important second messenger signals to mediate the response of cells to external stimuli [42,43].

[14] 37w The frequency and amplitude of the Ca 2+ oscillation signal can vary largely among different tissues, brain areas and experimental conditions, which in turn triggers various downstream molecular events such as axon outgrowth and neuronal plasticity [44].

[15] 108w We chose to perform our functional neuronal analyses of 3-D neuronal tissues on DIV 21, when axons and dendrites formed intensive networks and Synapsin-1 puncta density reached a plateau value. In order to investigate the effects of increased neuronal activity and characterize its electrophysiological activity, three-weekold 3D neuronal tissues were treated with a non-competitive GABAA receptor antagonist, Picrotoxin (PTx), to block inhibition and increase network activity [45]. We observed a steep increase of fluorescent intensity of calcium signal upon PTx adding (Figure 4a and Video S2), and the spiking frequency and correlation coefficient also increased, however, there was no significant difference before and after PTx treatment (Figure 4b,c).

[16] 99w Glutamate is a neurotransmitter which widely exists in most excitatory neurons as well as the glial cells. Agreeing with previous reports [27], the addition of glutamate caused a sharp rise of fluorescent intensity of calcium signal (Figure 4a). A sustained rise in calcium influx was observed for neurons and a transient increase was observed for glial cells [46]. We did not observe many glia cells based on the cell response which is in agree with our anti-GFAP staining result in Figure 2c. As a result, for the rest of experiments, we did not consider the influence of glial cells.

[17] 165w The main purpose of this research is to study the inter-popular neuron connectivity in 3-D neuronal tissue with different size and distribution. For milli-scale 3-D block (4 mm × 4 mm × 100 µm squares), frequency spikes at similar time points all around the neuronal network over the observation period (Video S3), which was indicated by closed ∆F/F pulsating curves (Figure 4d) of several randomly chosen neurons (Figure S4). A polynomial function was apply to help to remove the artifacts introduced while acquiring the fluorescent images and leads to a more ordered data set to run the FFT analysis while converting to the frequency spectrum as shown in Figure S5. The pulsating curves were further turned into a heatmap with intensity varied over time for easier comparison of frequency spikes (Figure 4e). The dynamics consisting of network bursts lead to the conclusion that a functional and active neuronal network was formed when culturing neuronal cells in 3-D GelMA hydrogels over the course of three weeks.

[18] 190w For neurons in 200 µm wide GelMA stripes with 100 µm gap, neurite extension was mainly along the direction of stripe while a few cell soma (pointed by white arrows in Figure 5a and figure S6a) aggregated together to form clusters inside the 3-D hydrogel and neurites (pointed by yellow arrows in Figure S6a) crossed the gap between stripes forming connections between two or more stripes. In low magnification images of Figure 5a, the anti-Tau, anti-Map2, and Anti-synapsin-1 staining seemed to be co-localized as the neurites form bundle during growth, which can be distinguished in the high magnification image below. We can also distinguished axons, dendrites, and synaptic vesicles in the same image. Accordingly, when we picked 3 neurons in each stripe (Figure S6b), calcium oscillation bursts are different between stripe (Figure 5b,c) while the signal impulses are closely in sync among neurons in the same line (Video S4). We further quantitatively studied the correlation coefficient (CC) of neurons (Figure 5d) within the same stripe (CC=0.68) and between different stripes (CC=0.10), the much higher CC value between neurons in the same stripe statistically imply the synchronicity of their calcium oscillations.

[19] 22w The cells in L4 and L5 experienced correlated frequency spikes (CC=0.52), which indicated the cross connection of neurites between these two lines.

[20] 208w To further study the influence of neurite connection among spatially separated 3-D tissue micro-blocks. In separated 500 µm squares, neurite, axon and synaptic vesicle growth was confined inside the block and revolved alongside the block edge (Figure 6a,b,c). Calcium firing was synchronized within each separated microblocks, but not between blocks (Figure 6d), with CC=0.78 and 0.20 (Figure 6e), respectively. In connected microblocks, neurites and axons extended out from single blocks and through the backfilled GelMA because of the lower stiffness of the outer hydrogel (~43 Pa storage modulus), and formed 3-D neural networks by connecting to neurites originated from adjacent blocks (Figure 6f,g). However, the synaptic vesicles were not fully developed in the space outside the micro-blocks (Figure 6h). We compared the Ca 2+ oscillation of neurons in (1) separated 500 µm GelMA squares in an unconnected array (Video S5) with (2) an array of 500 µm GelMA squares connected by backfilled GelMA (Video S6). Two neurons were picked from each micro-block as shown in Figure S5 and S6.The neurons in the connected microblocks showed completely different signal cycles, with a CC=-0.05 (Figure 6i), indicating that they were not synchronized. In addition, the synchronicity of neurons cultured in different connected micro-block is also very low, with a CC=-0.02.

[21] 130w Statistically, the neurite length in different geometries were not significant different (Figure 7a), however, the number of synapses was significantly lower in the outer space of connected micro-blocks (p<0.05, Figure 7b). By analyzing the percentage of active neurons, average calcium spiking frequency and the synchronicity (Figure 7) of cells in milli-blocks, micro-stripes and micro blocks, we observed a significant decrease in the percentage of active neurons in stripes and connected micro-blocks, while cells in both the milli-and micro-scale block had an active neuron percentage around 90 % (Figure 7c). The average frequency of calcium spiking decreased from 99 mHz in GelMA milli-blocks to 69 mHz in micro-stripes and around 17 mHz in micro-scale blocks (Figure 7d), indicating that calcium spiking was dependent on connection patterns of neurons. For oscillation synchronicity,

[22] 32w milli-block, micro-stripe and separated micro-block had close correlation scores of around 0.8, and the connected micro-block had low score closer to zero, which agreed with the results in heat maps (Figure 7e).

[23] 116w We also compared the calcium oscillation on 2-D culture and in hydrogel with a thickness of 100 µm, 200 µm and 1000 µm. The 2-D culture system had a significant lower spiking frequency compared to 3-D cultures (figure 7f, Video S7), with a frequency at 34 mHz, which is closed to the citing literature (16-33 mHz) [27]. There was no significant difference in spiking behavior as the thickness increase, which support our hypothesis that 100 µm 3-D GelMA hydrogel provided sufficient 3-D microenvironment for 3-D neural network growth. In all culture conditions with different thickness the correlation coefficient is high and close to each other at ~0.7 (Figure 7g), indicating well connection of the neural network.

[1] 271w In the native micro-environment, cells are surrounded and physically supported by a complex network of extracellular matrix components with adaptive composition and organization of macromolecules. This extracellular matrix (ECM) has a strong influence on tissue development, migration, proliferation, and survival of cells that thrive in it [47]. Various proteins and polysaccharides were expressed in the ECM of neural system, including collagen, laminin, proteoglycans, tenascins and reelin [48]. Bioengineering approaches based on semisynthetic scaffolds including hydrogels can be used to replicate certain functions of the ECM, supporting complex cell-cell interactions in the 3-D microenvironment [15]. GelMA is derived from gelatin, a partial hydrolysis product of collagen [49], which is one of the main component of brain ECM despite the components of ECM varies along the developmental stages and the parts of the brain [50]. GelMA can provide adhesive ligands for cell attachment and the degradability of GelMA allows the neurite extension inside the 3-D gel [49]. The amount of methacrylate anhydrate and the exposure time to the photocrosslinker tailor the network structures and mechanical properties of the gel [51]. For instance, photocrosslinking of GelMA provides covalently crosslinked meshwork with less sensitivity to degradation and form a thermally and temporally stable hydrogel that is ideal for long term cultivation of functional neuronal networks [37]. Another advantage of photocrosslinkable GelMA is the flexibility in geometry design. In this study, we fabricated four 3-D neuronal structures with distinct sizes, shapes or interconnectivities. In future studies, we can achieve more complex structures by utilizing different photo-masks and design the neurite connecting network by connecting certain 3-D neuron blocks with GelMA hydrogel [23] to model neurite outgrowth.

[2] 98w Spontaneous network activity is an intrinsic property of neurons, which plays a pivotal role in neuronal maturation and function [52]. Synchronized bursting has been observed in in vivo [53], ex vivo [54] and in vitro [27] in different brain regions. Physical confinement has been reported to have a direct impact on neural network connectivity in 2-D culture. The number of functional cellular partners of a neuron was considered as an important factor for network activation. Cellular density which strongly influences cell-to-cell contact has been reported to be critical for spontaneous network spiking [55][56][57]. Besides, cellular distribution [58,59], and

[3] 148w neuronal network scale can also alter the number of partner and impact neuron functional spiking behaviors in 2-D [60,61]. Previous research on 2-D square patterns with varied sizes shown that although the cell density was the same, neurons in smaller micro-squares tend to have lower calcium spiking frequency and correlation coefficient than larger squares [62]. These 2-D studies suggested that the neuron number in the whole network significantly influenced neuron behavior even if cell densities are similar. The influence of partner neuron number could contribute to the synergistic effect of presynaptic neuron depolarization. Single neuron has a depolarization of ~ 2mV, while the threshold of membrane action potential difference is >20mV [62]. Neuron depolarization could be scaled up by the connecting neurons in the network, when the partner neuron number is small, there are less signal inputs to encourage neuron depolarization, which decrease the chance of network firing.

[4] 111w However, how neuronal firing in frequency domain and synchronization responses to 3-D geometry and irregular connectivity has not been reported. 3-D culture is believed to better recapitulate the natural cell environment and cell-cell interaction with more authentic, reliable, and physiological relevant tissue function. For instance, some important features of cancer cells can only be remodeled in 3-D cultures [63]; neuronal differentiation of embryonic stem cells was only supported in 3-D culture models [64]. In our research about 3-D neuronal networks activity, we found that the partner number is essential for neuron spiking, but the actual spatiotemporal patterns of spontaneous activity in neuronal networks show multiple differences compared to 2-D surface [59,61].

[5] 166w In 3-D neuronal network, as the confinement of dimension increased (from "1" dimensional confinement in milliblocks, "2" dimensional confinement in micro-stripes, to "3" dimensional confinement in micro-blocks), calcium spiking frequency significantly decreased. Even the percentage of active neuron and synchronicity correlation coefficient did not remarkably change, their standard deviation increased significantly, indicating the instability of the 3-D network. Moreover, when we connected the micro-blocks with backfilled GelMA, the neurite growth inside the micro-blocks were partially sacrificed to the neural network among blocks, which decreased the synchronicity notably. The poor development of synaptic vesicles along axons connecting different blocks also imply the inferior connectivity. This could be attributed to the dilution of cell density in the whole gel as neuronal density determines network connectivity and spontaneous activity of in vitro culture primary neurons [56]. The different cellular distributions of neurons in gel with different geometries changed the effectiveness and number of neuronal connection. This might be one of the reason leading to the differences of calcium oscillation.

[6] 125w Compared to ex vivo mouse brain cortex slice, with a simultaneous burst rate of ~1 Hz [54], our in vitro 3-D culture system has significantly lower spiking speed. This can be attributed to the less complex neural network in in vitro model, which also serves as a support of the necessity in building more complex 3-D neuronal network towards 2-D culture. On the other hand, our simplified 3-D neuronal culture structures are convenient for geometry and composition modulation, which is important for physiological mechanism study to eliminate other factors such as cell component. While spiking behavior of cultured brain slices varied depending on their location in the brain, for example, subthalamic nucleus neurons in rat brain slices had a burst frequency of 127 mHz [65],

[7] 43w it may be hard to acquire micro-scale brain tissue with controllable component for geometry effect study. Besides, the ex vivo tissues in suspension cultivation are easy to deform from their original shape in long term, which also hindered their application in geometry study.

[8] 30w In future study, we can integrate more biological and biostatistical study for the underling mechanism in geometry and irregular connectivity induced differences of neuronal firing in frequency domain and synchronization.

[9] 13w Thanks to the development and commercialization of various bioprinting and biofabrication technologies [66],

[10] 101w more complex brain mimicking architectures can be fabricated [67]. For example, acoustic [68][69][70][71] and magnetic [72] based biofabrication methods developed by our lab and others enable the fast organization of cells to form complex tissue structures; gradient [73,74] and oriented [75] 3-D scaffolds has been desired to direct neurite growth. By combining with our technologies with such technologies, we can study more complex 3D network with distinct geometries and connections as well as more cell types in neural system. We can also combine our technique with patch clamp to study the actual electrophysiological activity of different 3-D neuronal networks in real-time.

[1] 614w In summary, we have demonstrated the ability to culture primary neurons in 3-D hydrogels and form a functional neuronal network. By photo-crosslinking GelMA through a pre-designed photo-mask in situ, we generated well- M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT Figure 2 (a) Images and (b) quantification of neuronal cell viability in 3-D GelMA milli-scale block hydrogel using calcein-AM (green, live) and ethidium homodimer-1(red, dead). Standard deviation was calculated for n = 3 with a significance of p < 0.05. Samples were not significantly different according to the one way ANOVA analysis. (c) Neuro-astro-network of neuronal cell cultures in 3-D GelMA milli-scale block hydrogel after 3 weeks culture. "Red" indicates anti-beta-tubuling staining, "green" indicates anti-GFAP staining and "blue" indicates DAPI staining. (d) Immunostaining showing excitatory and inhibitory neurons in 3-D milli-block hydrogel. Anti-GAD67 indicates inhibitory neurons (green) and anti-CaMKII indicates excitatory neurons (red). M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT Figure 5 (a) Axo-dendritic-synapto in stripe-like 3-D neural pattern. Scale bars indicate 200 µm if not specifically labeled. White arrows indicate cell soma clusters in micro-strip 3-D gel. Ca 2+ intensity (b) traces (∆F/F 0 ) and (c) time-series heat map of (∆F/(F max -F min )) calculated from individual neurons in Video S4 and selected in Figure S6. 6 of GelMA 3-D stripes were studied and 3 neurons were selected from each lines. The amplitude of the calcium oscillation is indicated as a color intensity in (c), the lighter color indicate higher fluorescent intensity. Traces are displayed over the course of 240 seconds. (d) Quantification of correlation coefficient of calcium oscillation signals in the same stripe or between different stripes, which shows the synchronicity of neuronal cell firing. M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT Figure 6 (a,f) Neuro-, (b,g) axo-, and (c,h) synapto-network in (a,b,c) separated micro-scale blocks and (f,g,h) connected micro-scale blocks. Scale bars indicate 200 µm. (d,i) Ca 2+ intensity time-series heat map of (∆F/(F max -F min )) calculated from individual neurons in Video S5 and S6 and selected in Figure S7 and S8. 8 GelMA 3-D separated micro-scale blocks and 10 connected blocks were studied and 2 neurons in were selected from each blocks. The amplitude of the calcium oscillation is indicated as a color intensity, the lighter color indicate higher fluorescent intensity. Traces are displayed over the course of 240 seconds. (e,j) Quantification of correlation coefficient of calcium oscillation signals in the same block or between different blocks, which shows the synchronicity of neuronal cell firing in (e) separated and (j) connected micro-blocks. M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT Figure 7. Statistics of neurite network including (a) neurite length and (b) Synapse density as well as calcium oscillations including (c) the percentage of active neurons, (d) bursting frequency, and (e) synchronicity of neuronal cells in milli-block, stripe-like, and micro-block (separated (SP) or connected (CN)). At least 150 neurons from 3 independent samples were analyzed to calculate the percentage of active cells. The neuron bursting frequency and correlation coefficient were also statistically study in 3-D hydrogel with different thickness (including 2-D culture) and shown in (f) and (g), respectively. The synchronicity was indicated by correlation coefficient of Pearson correlation (-1 to 1), the closer to 1 indicating higher synchronicity. Standard deviation was calculated for "n = 3" with significance p < 0.05 according to one way ANOVA with a Tukey posthoc method. "*" indicates significant difference.

[1] 94w GelMA Synthesis. GelMA is obtained by chemical modification of gelatin with methacrylic anhydride (MA) at lysine side chains (Figure S1a). 2 g gelatin (Type A, 300 bloom from porcine skin, Sigma-Aldrich) was dissolved in 20 mL PBS (pH 7.4, Life Technologies) at 50 °C, After lowering the temperature of the solution to 40 °C while stirring the solution at 160 rpm for a total of 2h, 1.6 mL of methacrylic anhydride (94%, MA, Sigma-Aldrich) was added using a syringe pump (NE-1000, New Era Pump Systems) at a speed of 0.1 ml/ min for 2h.

[2] 32w The solution was dialyzed (Mw<14,000, Sigma-Aldrich) at 40 °C for 1 week against water. The dialyzed solution was freeze-dried for 5 days and stored in aliquots at -80 °C until further use.

[3] 132w Cortical neuron isolation and culture. All experiments were carried out according to animal care standards set forth by the National Institute of Health and were approved by the Institutional Animal Care and Use Committee (IACUC) at Stanford University. Primary cortical neurons were isolated from embryonic day 18 (E18) CD-1 mice provided by Charles River. Freshly micro-dissected whole mouse cortices were dissociated by trituration after 30 min of enzymatic digestion using Neuronal Isolation Enzyme (Thermo Scientific) in Hanks Balanced Salt Solution without Mg and Ca (HBSS(-), Life Technologies). The cells were then suspended in neurobasal medium with B27 supplement (Life Technologies) and 500 µM glutamine (Life Technologies). Culture conditions were maintained at 37 °C with a 95% relative humidity, and 5% CO 2 . For 2-D culture, 50,000 neurons in 0.5 ml culture

[1] 48w image where cells labeled with green were regarded as live cells, cells labeled with red as well as co-labeled with green and red were regarded as dead cells. More than 30 cells were analyzed for each of the three experimental repeats, totaling at least 90 neurons per condition.

[2] 156w Immunocytochemistry At DIV 7, 14, 21 and 28, GelMA tissues were fixed with 4% paraformaldehyde (Electron Microscopy Sciences) in PBS for 30 min at room temperature. Following 3× 5-min washes with PBS, 3-D tissues were permeabilized in PBS containing 0.25% Triton X-100 (Sigma-Aldrich) for 30 min at room temperature on a shaker. Following by three washing steps in PBS for 5 min, 3-D tissues were incubated in 5% (w/v) bovine serum albumin (Sigma-Aldrich) in PBS (PBSA) for 2 h at 37 °C. Then, samples were incubated on a shaker, overnight at 4 °C in 5% PBSA containing primary antibodies (details are shown in Table S1). The next day, samples were subjected to 3× 5min PBS washes followed by fluorophore-conjugated secondary antibodies staining in 5% PBSA were incubated overnight at 4 °C followed by 3× 1 h washes in PBS. 4',6-Diamidino-2-Phenylindole, Dihydrochloride (DAPI, 1: 1,000, Life Technologies ) was added into the staining solution for nuclear staining.

[3] 204w Samples were imaged on a Zeiss LSM 710 laser scanning confocal microscope using 20× dry (Plan-Apochromat 20x/0.8 M27, Zeiss) with Z-stack. Image analysis was performed with the Imaris imaging software (Bitplane A.G.) and ImageJ software (NIH). Three images were analyzed for each conditions. The details were described in supplemental information. Calcium imaging. At DIV 21 of culture, cells were washed 3 times with HBSS with MgCl 2 and CaCl 2 (HBSS(+)) and subsequently labeled for 1 h with 4 µM Fluo-4 AM calcium indicator dye (Life Technologies) at 37 °C in HBSS(+). After incubation, cells were rinsed four times with HBSS without MgCl 2 and CaCl 2 (HBSS(-)) and incubated for an additional 10 to 15 min at 37 °C to allow complete de-estherification of the Fluo-4 AM dye before imaging. Time-lapse imaging was performed with a Zeiss AxioObserver Z1 microscope (FLUAR 420130-9900, Zeiss) in a period of 4 min. Before recording, 1mg/ml Picrotoxin (PTx) in HBSS(-) was added to the solution as a to reach a final concentration of 10 µM. To trace the drug response with PTx and glutamate, calcium oscillation was traced for 13 min and PTx was added at 330 s and 30 µM glutamate was added at 720 s.

[4] 127w ImageJ software was used to analyze fluorescence intensity variation of selected neurons. Region of interest (ROI) was defined quantitatively in terms of variation in signal amplitude for the active cell compared to its surrounding. Active cells were defined as cells showing at least one peak of calcium burst in the total 240 s period. At least 150 cells in total from 3 independent samples were analyzed to calculate the percentage of active cells, 10-15 active neurons were picked from each experimental batch to calculate spiking frequency and correlation coefficient. For each selected neuron, a reference area in close proximity to the respective neuron was chosen in order to normalize the fluorescent intensity against the background. This was subtracted from the Fluo-4 fluorescence emanating from the cell bodies.

[5] 37w In order to normalize the signal, the fluorescent change of the single wavelength excitation/emission dye Fluo-4 AM was defined as a delta function. Delta (∆) was used to denote a change from an initial state. ∆F therefore

[6] 66w denotes the difference between initial fluorescence intensity at the resting state and fluorescence intensity after stimulation. Hence, ∆F/F calculates the relative change of the intensity to the original intensity before stimulation. This method allowed for the measurement of changes in fluorescence because it was insensitive to external factors such as bleaching of the fluorophore and was defined as follows [25]. The parameters are fitted such that

[7] 49w is minimized where A(t) denotes the signal recorded from the acquisition system and B(t) is the estimated polynomial function. After the trend subtraction performed in the above step in (using polynomial function), the time series data was transformed into the frequency domain to identify the frequency of the data.

[8] 59w To generate heat map of fluorescent variation over time, the normalized fluorescent traces (∆F/F) of each neuron was rescaled to a zero to one range in order to be able to compare fluorescent signals of different neurons. The fluorescent signal for each neuron displayed in all the heatmaps was calculated as follows, and heatmaps were generated using Rstudio software.

[9] 31w The heat maps show the relative calcium signal intensity in time-series of all selected cells, which qualitatively represent the oscillation of all cells and are not influenced by subsequent analysis method.

[10] 37w To estimate the mean frequency, Fast Fourier Transform (FFT) is applied to the time-domain data to perform the spectral analysis on the fluorescence data. The oscillatory nature of the signal can be mapped according to the equation:

[11] 18w where in ‫ݐ(݃‬ ) is the time-domain signal and ‫݂(ܩ‬ ) is the Fourier transformation for N points.

[12] 41w Due to the oscillatory patterns within the different neurons under varying GelMA geometries, a power spectra is calculated to identify the dominant frequency components. The power spectral density (PSD) is calculated for each frequency domain signal (݂ ) using the formula:

[13] 23w This method was used to analyze the signal contribution arising from different frequencies in neurons in the 4 different GelMA geometries investigated here.

[14] 25w Synchronicity is defined by the Pearson correlation coefficient [26,27]. For each pair of neurons (X and Y), the correlation coefficient R(x,y) can be calculated by:

[15] 59w where ݂ ௫ ‫)ݐ(‬ is the relative fluorescence intensity of cell x at time t and ݂ ௫ ഥ is the time averaged fluorescence intensity. R(x, y) can be ranged from -1 to 1, where R(x, y) = 1 indicates complete synchronicity, "-1" indicates reverse contrast, and "0" indicates no correlation between the oscillation of the pair of neurons.

[16] 44w Statistical analyses. Statistical difference of the data was analyzed by one-way ANOVA with a Tukey post-hoc method, and a significance level of p < 0.05 was used for all the tests. The error bars represent standard deviation (SD) value of each type of experiments.