Modeling controlled cortical impact injury in three-dimensional brain-like tissue cultures

Abstract

Traumatic brain injury (TBI) survivors suffer long-term from mental illness, neurodegeneration, and neuroinflammation. Studies of 3D tissue models have provided new insights into the pathobiology of many brain diseases. We here developed a 3D in vitro contusion model consisting of mouse cortical neurons grown on a silk scaffold embedded in collagen and used outcomes from an in vivo model for benchmarking. We characterized molecular, cellular and network events in response to controlled cortical impact (CCI). In this model, CCI induced degradation of neural network structure and function and release of glutamate, which was associated with the expression of programmed necrosis marker pMLKL. Neurodegeneration was observed first in the directly impacted area and it subsequently spread over time in 3D space. CCI reduced pAkt and GSK3β in neurons in vitro and in vivo, but discordant responses were observed in pS6 and pTau expression. In summary, our 3D brain-like culture system mimicked many aspects of in vivo responses to CCI, providing evidence that our model can be used to study the molecular, cellular and functional sequelae of TBI, opening up new possibilities for discovery of therapeutics.

1. Introduction

Worldwide, traumatic brain injury (TBI) is the leading cause of physical injuries and mental disability, as well as death among individuals before the age of 45.^[1] The diagnosis of TBI is defined as disrupted brain morphology or function after impact and/or head acceleration forces.^[1] TBI may be classified according to severity (Glasgow coma scale score), mechanism (e.g., diffuse, focal, blast) or pathoanatomically (contusion, concussion, white vs. gray matter damage, presence or absence of hemorrhage, etc.).

TBI is a heterogenous affliction that causes acute and chronic neurological symptoms associated with activation of specific molecular pathways, such as protein kinase B/mammalian target of rapamycin (Akt/mTOR), necroptosis, ferroptosis, nuclear factor κB (NFκB) and mitogen-activated protein kinases (MAPKs) to name a few.^[2–6] The anatomic and biochemical heterogeneity of TBI pathology makes TBI diagnosis^[7] and treatment challenging.^[8] Currently, specific therapies to reduce or prevent neurological sequelae of TBI are lacking, in large part due to limited understanding of secondary injury mechanisms that contribute to outcome.^[1,9] Examples of secondary injury mechanisms, initiated immediately after trauma, include (but are not limited to) excitotoxicity, edema, diffused axonal injury, energy failure and neuroinflammation.^[10,11]

Rodent in vivo models are widely used to study mechanisms involved in TBI,^[12,13] such as cellular and molecular responses to injury and behavioral changes (spatial learning and memory, motor learning, anxiety, fear, activity, etc.).^[12–15] In addition, two-dimensional (2D) in vitro cultures showed a distinction in neuroinflammatory responses between various mechanical perturbations, such as stretch injury, strain, and fluid percussion injury.^[16–19] These studies also described cell-type specific (e.g., neurons, astrocytes, and microglia) responses to injury.^[19–21] However, despite their utility, 2D models do not fully mimic the structural or functional complexity of the brain or the cellular microenvironment. Importantly, they do not replicate the mechanical aspects of an actual blast, concussion or contusion traumatic injury.

Bioengineered tissues grown in three-dimensional culture systems are emerging as powerful in vitro models of the brain.^[22,23] They closely resemble native brain anatomy and physiological responses,^[23–25] tissue stiffness, cell-cell interaction, extracellular matrix and heterocellular composition.^[24–27] In vitro three-dimensional (3D) brain tissue models have already yielded new insights into disease pathogenesis and potential treatments related to schizophrenia, Alzheimer’s and Parkinson’s disease.^{[23,25,26,28,29]} Here, we developed and characterized a new 3D in vitro brain injury model of controlled cortical impact (CCI) and used it to examine spatio-temporal aspects of damage propagation following TBI. CCI is a method commonly used to replicate cerebral contusion in rodents and large animals.^[30]

Our 3D tissue model consisted of two compartments: a silk scaffold, which serves as a neuronal growth substrate embedded in collagen type I, and an extracellular matrix that supports dense neuronal network growth. We validated the relevance of the in vitro TBI system by comparing experimental outcomes to CCI in living mice. Parameters tested included primary (direct) tissue injury and delayed secondary injury mechanisms leading to neurodegeneration. Spatio-temporal analyses of neuronal network structure identified focal damage shortly after CCI (within the first hour) that subsequently propagated throughout the 3D brain tissue within the 24 h timespan tested. Importantly, we observed that the loss of neural network structure was associated with necroptotic pathway activation and deactivation of the Akt/mTOR signaling pathway, a key regulator of cellular homeostasis in CCI.^[2–4,31] Necroptosis activation was tied to membrane permeabilization, release of glutamate and ultimately neuronal death. We conclude that our in vitro 3D brain injury model recapitulates many important aspects of CCI in mice and humans. Our 3D brain-tissue model is scalable, providing unique opportunities to gain insights in the molecular signaling pathways, cell-and structural and functional changes after TBI and for discovery of novel therapeutics.

2. Results

2.1. 3D in vitro brain injury tissue model fabrication and characterization

To establish a 3D in vitro model of traumatic brain injury, we employed embryonic day 16 (E16) mouse cortical neuron cultures, and compared the results to in vivo CCI experiments using the identical mouse strain (C57Bl/6J mice).^[32] Brain-like tissues were created by seeding 1 million cortical neurons onto 3-dimensional silk scaffolds. The 3D system, with a 6mm outer diameter, 2 mm diameter of internal window and a 1.5 mm height, was enveloped in a collagen type I hydrogel. Together with the porous nature of the scaffold (300–400μm diameter pores size), this allows for optimal physical support and exchange of nutrients and oxygen for long-term growth of cells (Figure 1A).^[22] Indeed, the 3D in vitro brain tissue model supported the growth and maturation of dense neural networks. Neurons expressed the early maturation markers β3-tubulin (Tuj1) and late markers of dendrites (microtubule associated protein 2 (MAP2)) and synapses (the synaptic scaffold protein Synapsin-1 (Figure 1B)). Cell mask, a commercially available dye used for plasma membranes labelling, further confirmed the formation of elaborate neuronal networks penetrating the depth and width of the scaffold (Figure 1C). Immunofluorescence analysis of postsynaptic density protein 95 (PSD95) and Gephyrin post-synaptic markers demonstrated the presence of excitatory and inhibitory synapses, respectively (Figure 1D, E). Cell-type specific analysis indicated that our 3D in vitro brain tissue model was composed predominantly of neurons, with a minor presence (less than 5%) of microglia and astrocytes (Figure 1F, G).

Figure 1.

Bioengineering of the 3D in vitro brain tissue model and characterization of 3D neural networks after 14 days in culture. A) Schematic of the 3D in vitro brain tissue fabrication process. B) Immunofluorescence analysis of neuronal markers (Tuj1, MAP2, Syn-1) (image width (w) = 232 μm; length (l) = 232 μm; depth (d) = 75 μm; image z step = 0.567μm) Scale bar: 50μm. C) Structure of neuronal network density evaluated with cell mask dye (w = 232 μm; l = 232 μm; d = 800 μm). Scale bar: 200μm. D) Immunofluorescence analysis of post-synaptic markers (Gephyrin, and PSD95) (w = 465 μm; h = 465 μm; d = 415 μm; z step = 0.567μm). Scale bar: 50μm. e Overlay image of PSD95 and Gephyrin markers with 3x magnification (w = 43 μm; h = 43 μm; d = 415 μm; z step = 0.567μm), Scale bar: 5 μm. F) Immunofluorescence analysis of glial markers (S100 calcium-binding protein B (S100B) for astrocytes, integrin alpha M (CD11b) for microglia) (w = 232 μm; h = 232 μm; d = 56 μm; z step = 0.567μm); Scale bar 50 μm. F) Quantification of astrocytes (S100B positive), microglia (CD11 positive and oligodendrocytes (oligodendrocyte transmembrane protein O4) in 3D in vitro brain model. Graph represents mean ± SEM of three independent experiments.

2.2. Controlled cortical impact injury characterization

Controlled cortical impact (CCI) of the 3D brain-like tissues was conducted after 14 days in culture using parameters identical to those we previously reported in vivo.^[32] The 3D scaffold was impacted with a 3mm diameter piston at a velocity of 6m/s, penetrating to a 0.6mm depth within the silk scaffold (Figure 2A). We monitored the appearance of well described structural, biochemical, and molecular markers of CCI at seven time points. In all experiments, sham injured 3D cultures were used as control for every time-point (Figure 2–8). We compared results from the in vitro models to published literature and to new in vivo experiments related to damage induced signaling pathways (Figure 8).

Figure 2.

Controlled cortical impact of 3D brain-like tissues triggered neuronal network degradation 24 hours post-impact without structural damage of collagen. A) Experimental set-up of controlled cortical impact (CCI) after 14 days in culture of the 3D brain-like tissue. Timeline showing data collection points after CCI. B) Multiphoton immunofluorescence microscopy analysis of Tuj1 positive neurites indicated neural network destruction. Second harmonic generation (SHG) analysis demonstrated absence of macroscopic changes in the collagen Type I fibrillary structure (w = 200μm; l = 200μm; d = 290μm; z step = 0.567μm). Scale bar: 50 μm. C, D): Quantitative analysis of Tuj1 and SHG image data represent mean ± SEM of three independent experiments with n= 3–6 3D tissues per group. * indicates significant differences (p < 0.05; t-test between control and experimental groups.

Figure 8.

Expression analysis of injury associated proteins in in vitro and in vivo CCI models. A) Western blot analysis of the levels of AKT, S6, GSK3β protein and phosphorylated AKT (Ser473 residue), S6 (Ser 235/236 residues) (Akt/mTOR pathway constituents). B) Western blot analysis of the levels of HMGB1 (cell death marker), LC3II (autophagy marker), phospho-Tau (residues Ser202 and Thr205, recognized by the AT8 antibody as a marker of neurodegeneration) and IL-1β (inflammation marker). Detergent lysates were prepared of 3D neuronal cultures and of neurons purified from injured mouse brain by immunopanning. The amounts of protein loaded were normalized based on the amount of β-actin present in the lysates. Three independent experiments were conducted for in vitro protein analysis, and lysates of 3 independent scaffolds were pooled per experiment. For the in vivo experiment, the neurons were isolated from the ipsilateral hemisphere and n=3–4 sample (3–4 pooled brains per sample) were analyzed for protein expression. The western blot data were analyzed through densitometry of representative western blots. The bar graphs present averaged data from three independent experiments. * indicates significant differences between the groups of p<0.05 (One-way ANOVA analysis for in vitro experiment and two-tailed t-test for in vivo experiments).

Multiphoton microscopy of Tuj1-positive neurites indicated complete disintegration of the neuronal network at 24 hours post-CCI at the focal side of injury (Figure 2B, C). Second harmonic generation images of the 3D extracellular matrix, analyzed to define the 3D orientation of collagen fibers and corresponding levels of alignment reported by the 3D variance, showed no changes in the collagen Type I fibrillary organization (Figure 2B, D), suggesting that neural network damage is not caused by a macroscopic breakdown of the extracellular matrix.

2.3. Spatio-temporal propagation of structural damage and functional response to CCI

A shared characteristic of cerebral contusion is progressive neurodegeneration from the directly impacted site to adjacent areas.^[33] We first examined the propagation of network damage as a function of time at the area injured directly below the impactor tip (Figure 3A). Fluorescence images of the 3D brain tissues network density and individual neurite length was quantified with our custom developed MATLAB (Matrix Laboratory) based code. One hour after CCI, the neuronal network was degraded approximately by half, reaching full disintegration by 24h (p < 0.05; Figure 3A, B; Supplementary Figure 1, 2). The number of Synapsin-1 punctae per image was significantly decreased 2h after CCI, further declining over time (p < 0.05; Figure 3C, D). In line with the changes in structure, by measuring spontaneous neuronal activity with local field potential (LFP), we observed a decline in function of the 3D brain-like tissues by 24 hours post-CCI (Supplementary Figure 3). In contrast, structure and the function of the neural network of sham treated samples remained intact at all timepoints.

Figure 3.

Temporal propagation of neuronal network disintegration directly under the impact tip after CCI. Integrity of the neuronal network structure was analyzed with immunofluorescence analysis of Tuj1 neurite marker at seven different time points. A) Representative images (maximum projection) of neuronal damage selected from the five time points that show the most prominent changes after CCI. B) Quantification of network density for all seven time points after CCI. Data are the percentage of voxels that exhibited fluorescence (w=230μm; l=230μm; d=85μm; z step=0.567μm). C) Representative images of synapse marker Syn-1 at the same time-points as in b. D) The number of Synapsin-1 punctae was quantified using Elements (Nikon) software. Data presented in b and d represent mean ± SEM of three independent experiments with n = 2–3 scaffolds per condition. * indicates significant differences (p < 0.05; two-way ANOVA (analysis of variance)) between control and experimental groups. Scale bar: 50μm.

We subsequently assessed secondary, spatial propagation of neurite damage in all three dimensions over time by examining a large volume of 3D brain tissue (1.5 mm x 1 mm x 42μm; indicated in Figure 2A as the boxed area highlighted as Figure 4). This encompassed a 400μm x 400μm section that extended outside the impact zone by approximately 1mm x 0.5mm. At one hours post-CCI, neurite structure damage was limited to the original, primary site of injury, while the area around this region continued to show an intact, dense neuronal network (Figure 4 A, B). Four hours after CCI, the structural damage had propagated approximately 600μm to 800μm outside the injury area. By 24 hours, all regions analyzed displayed significant neurite structure loss (Figure 4B; See Supplementary Figure 1 for high resolution images from all time points). The 3D brain-like tissue model thus replicates spatio-temporal spread of neurodegeneration from the site of primary injury to secondary areas that is observed in animal models in vivo and in human patients.^[34]

Figure 4.

Spatial-temporal propagation of neuronal network disintegration from the directly impacted area to adjacent sites within 24 hours post-CCI. A) Secondary damage propagation analyzed by confocal analysis of Tuj1 immunofluorescence at 1, 4 and 24 hours timepoints post-CCI. Schematic representation of locations imaged in the 3D tissue are shown in Figure2a; (w = 1340 μm; l = 810 μm; d = 42μm; z step = 0.567μm). Scale bar: 200μm. B) Quantification of neuronal network disintegration over time and in 3D space. The data is displayed from a representative experiment, showing the total length of all TuJ1 positive neurites present in each 3D image stack.

2.4. Membrane permeability and glutamate release of 3D brain-like tissues after CCI

To further investigate the response to CCI of neurons grown in the 3D environment, we assayed LDH (lactate dehydrogenase) and glutamate release into the media from damaged cells. We took advantage of the unique scalability of our in vitro brain model to examine the emergence of these two CCI biomarkers at multiple timepoints (Figure 5). Compared to sham, CCI cultures showed significantly higher release of LDH in the media, as early as one hour after injury. LDH levels doubled by four hours (p < 0.0001 vs sham) and tripled by 24 hours (p < 0.0001 vs sham). (Figure 5A). The CCI exposed 3D samples also responded with a nearly 1.8-fold glutamate increase by 4 h (p < 0.05 vs sham) and 2.1-fold by 24 h (p < 0.001 vs. sham; Figure 5B). The gradual increase in LDH after CCI, but not glutamate, was time dependent (comparing 1h with 4h and 4h with 24h timepoints; p < 0.001). The increase in LDH and glutamate release followed the kinetics of neuronal network degradation (Figure 3–4), suggesting that the processes are linked.

Figure 5.

CCI injury induced LDH and glutamate release by 3D brain-like tissues. A) Time dependent changes in LDH and B) glutamate levels in the 3D culture medium after CCI. Graphs represent mean ± SEM of three independent experiments, with n = 6 3D tissues per experiment. * indicates significant differences (p < 0.05; two-way ANOVA) between control and experimental groups.

2.5. CCI in vitro induces progressive neuronal cell death

Having observed significant neurite network structural damage, we next asked whether CCI resulted in death of neurons, like it does in vivo.^[32] 3D brain-like tissues were paraformaldehyde and ethanol fixed, and apoptotic and necrotic cells were identified using TUNEL (Terminal deoxynucleotidyl transferase dUTP nick end labeling) reagent, a commonly used assay for in vivo CCI studies.^[33] Cell nuclei were counterstained with propidium iodide (PI), which after the chemical fixation procedure labeled all neurons present. The ratio of TUNEL-positive (dead) cells over propidium-iodide (PI) positive cells (all neurons present) was determined as a marker of cell death. At one hour post-CCI, approximately 40% of the neurons were found to be TUNEL positive. By 4 hours this value reached 70%, and by 24 hours post-CCI nearly 85% (p < 0.05 vs sham and 1h CCI) Staurosporine treatment (1 μM), a positive control reagent used to trigger cellular death via caspase-dependent and independent pathways,^[3,35] induced approximately 60% cell death at 24 hours (Figure 6 A–C). The total number of PI positive cells that were imaged did not differ between the sham, CCI and Staurosporine groups, indicating that network degradation and cell death do not lead to complete loss of cell nuclei within the 24 hours window (Supplementary Figure 4).

Figure 6.

Controlled cortical impact injury induces death of neurons. A) Representative photomicrographs of TUNEL and PI fluorescence analysis at 1, 4 and 24 hours post-CCI (w = 232 μm; l = 232 μm; d = 56μm; z step = 0.567μm). B) Quantification of TUNEL and PI positive cells by Cell Counter plugin of ImageJ software. Graphs represent mean ± SEM of the ratio of TUNEL positive cells and the total cell number (PI positive). Data are the mean ± SEM of three independent experiments with n = 3–4 scaffolds per experiment. Scale bars: 50μm. C) Positive control samples were treated with 1μM Staurosporine for 24 hours to induce cellular death. * indicates significant differences (p < 0.05; two-way ANOVA) between control and experimental groups.

2.6. Necroptosis is a mechanism of neuronal death induced by CCI

We next sought to establish the potential mechanisms underlying the progressive neuronal network degradation induced by CCI. Necroptosis, a form of programmed necrosis induced by receptor interacting protein kinases (RIPKs), is a proposed cell death mechanism post-CCI.^[4,33] Phosphorylation of mixed lineage kinase domain-like protein (pMLKL), a substrate of RIPK3, was used as a marker to identify necroptotic cells in 3D neuronal tissues. Images were collected at three different areas to show propagation of pMLKL expression through the 3D brain-like tissue: injury (directly underneath the CCI piston), proximal (scaffold between impacted and edge area), and distal (at the edge of the scaffold) sites.

pMLKL was found to be expressed 15 minutes post-CCI in approximately 40% of cells and in 60% of cells by 4 h (p < 0.05 vs sham). Importantly, we observed pMLKL activation throughout the 3D in vitro brain tissue (injury area, proximal, and distal areas of the 3D scaffold) (Figure 7A, B). At every time point, the sham groups showed a significantly lower level of pMLKL expression when compared to the CCI injury group. We have found correlation between TUNEL and pMLKL positive cells in the control group (p=0.02) (Supplementary Figure 5). We conclude that, similar to in vivo models of CCI,^[4,36] neuronal death in bioengineered 3D tissue models involve at least in part, mechanisms related to necroptosis. The percentage of pMLKL positive cells (Figure 7) in the injured area is similar to the number of TUNEL positive cells (Figure 6), suggesting that necroptosis is one of the main pathways of neuronal death in vitro after CCI.

Figure 7.

Neurons express enhanced levels of phospho-MLKL marker of necroptosis within 15 minutes after CCI in vitro. A) Representative images of pMLKL acquired by immunofluorescence microscopy at 15 mins, 1, 4, and 24 hours post-CCI. Image sections were collected at three different areas (injury, proximal, distal) to show pMLKL expression throughout the 3D in vitro brain tissue. B) Quantification of pMLKL and DAPI markers using Cell Counter plugin of ImageJ software. Graphs represent mean ± SEM of three independent experiments with at least n = 3 tissue scaffolds per experiment. * indicates significant differences (p < 0.05; two-way ANOVA) between control and experimental groups. Scale bars: 50 μm.

2.7. Molecular signatures post-CCI in 3D tissue cultures model mimic CCI in vivo

To further assess the physiological relevance of the 3D in vitro brain tissue system as a model of CCI, we examined neuronal signaling and damage pathways implicated in brain injury, comparing in vitro and in vivo samples. We first focused on AKT/mTOR pathway constituents as they are critical regulators of cellular homeostasis, metabolism and survival.^[2–4] Akt phosphorylation inhibits the mTORC1 complex, thus inhibiting mTOR signaling pathway. Activation of mTOR pathway is linked to the regulation of synaptic signaling^[37] as well as necroptosis and brain cell survival and animal behavior in various experimental models of brain trauma.^[2,38,39]

We have previously shown in whole mouse brain tissue homogenates that CCI leads to an increase in Akt and its substrate glycogen synthase kinase 3 beta (GSK3β), and the mTOR target ribosomal S6 kinase (S6) protein expression.^[4] To directly compare the results from experiments on neurons grown in 3D tissues to the response in vivo, we isolated cortical neurons from mouse brains after CCI by immunopanning. We then performed quantitative Western Blot analysis of lysates of immunopanned neurons and compared them to homogenates from injured 3D neuronal cultures (95% pure neuronal culture, see Figure 1). When normalized to β-actin, the total amount of Akt and S6K protein expressed was not changed after CCI (Figure 8). The levels of neuronal phosphorylated Akt were decreased in both in vitro and in vivo models (p < 0.05) and, in line with these results, the levels of phospo-S6 kinase, a downstream target of Akt/mTOR signaling was decreased in vitro. Additionally, the expression of GSK3β was significantly downregulated in in vitro model and a similar decrease was observed in in vivo model. Interestingly, S6K phosphorylation was increased in neurons isolated from injured mice (p < 0.05), suggesting differences between in vitro and in vivo responses of neurons to CCI.

We next analyzed the expression of death, autophagy, inflammatory and neurodegeneration related proteins associated with CCI. HMGB1 (High mobility group box 1 protein), a cytosolic protein that is released from damaged neurons^[40] was downregulated in both in vitro and in vivo models post-CCI (Figure 8). Phospho-Tau protein expression (residues S202 and T205), a commonly used marker of neurodegeneration, was decreased in vitro, consistent with the observed neural network degradation (Figure 2–4) but remained unchanged in vivo 24 hours post-CCI (Figure 8). LC3II levels were decreased in vivo (p<0.05) and in vitro, while IL-1β (interleukin 1 beta) cytokine expression remained unchanged in both experimental systems. These results support the notion that neurons grown in isolation in 3D cultures respond to CCI in ways both similar and different than neurons exposed to CCI in vivo.

3. Discussion

The goal of the present study was to develop a scalable model of controlled cortical impact (CCI), benchmarking outcomes to in vivo mouse studies, so that in vitro findings can be translated to preclinical animal models as well as to the clinic. During the last decade, engineered 3D in vitro brain-like tissues have emerged as powerful tools to study brain biology and pathology.^{[23–25] [41]} 3D models have provided new insights into neurodegeneration, neuroinflammation, genetics, and development in various neuropsychiatric and neurodegenerative disorders.^[42]

In contrast to other neuroscience fields, a 3D brain like tissue model has not yet been reported to study the biology of CCI. The scalability of our 3D in vitro cultures will provide new opportunities for discovery of mechanisms of pathology and therapeutic drugs, either to support studies of in vivo models or to study aspects of biology that are specific to the human condition, such as genetics^[12–14] and longevity.^[43] To this point, our engineered 3D neural tissues derived from stem cells that were functional for up to one year in continuous cultivation in scaffolds^[43] may be useful to study the long-term consequences of traumatic brain injury.

We here developed an in vitro model of controlled cortical impact that mimics the damage experienced by living animals. We used established parameters for the study of CCI in vivo^[44,45] and adapted our protocols for the generation of 3D rat cortical cultures^[22] to create a new 3D mouse brain culture model. One distinct advantage of this new in vitro system is the use of a machine-driven impactor, which ensures high precision in terms of location, severity and reproducibility of injury, especially when compared to the weight drop model.^[22] Previous studies showed that weight drop and CCI inflict different behavioral, physiological and molecular responses in vivo.^[46] It may thus be possible to identify signaling mechanisms that are shared and distinct between the different TBI injuries using our two different in vitro systems. Second, mouse brain cells isolated from specific, genetically modified mice such as RIPK knockouts can be used to gain insights in specific molecular pathways involved in traumatic brain injury.^[47] Moreover, RNAi (ribonucleic acid interference), cDNA (complementary deoxyribonucleic acid) and CRISPR (Clustered regularly interspaced short palindromic repeats) reagents that target the mouse genome are abundantly available for related mechanistic studies.

To validate the relevance of our in vitro injury model, we correlated results from experiments done in 3D mouse cortical cultures to immunopanned brain cells from mice subjected to CCI using the same injury parameters as in vitro.^[4] We also made comparisons to the existing literature on CCI in rodents and human patients.^{[4,7,31,36,48]} We found many similarities between the two models, including the kinetics and spatial propagation of structural damage of neurites and synaptic networks, on the order of minutes to hours, the spread of injury from the primary damage site to the adjacent healthy cells, and functional network neurodegeneration.^{[4,7,31,36,48]} In addition, LDH, glutamate and HMGBI were released from the 3D cells similar to the response of damaged cortical tissues in vivo.^[40,45] IL1β expression, a hallmark of neuroinflammation, was not changed in bioengineered tissues or in neurons isolated from CCI exposed mice.

Several studies have supported the importance of Akt/mTOR signaling in inflammation, metabolism, cell cycle, survival, neuronal damage, and functional outcome post-CCI, suggesting that this pathway could be a promising target for therapeutic intervention.^[2–4,38] Akt phosphorylation leads to activation of the mTORC1 pathway, that in turn inhibits the autophagy (LC3II expressions) process.^[49] In our study we detected the downregulation of pAkt and its substrate GSK3β in both models, as also showed in numerous publications.^[2–4,38] However, the substrate of mTOR pathway, S6 kinase phosphorylation was significantly downregulated in in vitro model, and upregulated in in vivo, what also corresponded with the autophagy marker LC3II expression level (Fig.8). Other notable differences include the appearance of autophagy markers in neurons from CCI mice in vivo (but not in vitro) and the decrease in pTau (pS202 and pT205) in 3D tissues (but not in vivo). Several explanations may account for these differences, such as differences related to experimental conditions, and whether or not the observed changes in neuronal expression are cell autonomous in vivo. Related to the first point we cannot exclude the possibility that isolation of living cells from intact mouse brain after CCI induces stress and related signaling pathways in neurons. Moreover, neurons studied in 3D cultures differ from those examined in vivo in which cell bodies severed from their axons and dendrites are analyzed. These neurons are isolated from the entire injured hemisphere. Biochemical studies (signaling, LDH and glutamate assays) of the in vitro system are done on a population enriched for injured neurons (directly under the impact site and in adjacent tissue) with neurites and axons intact. This population of healthy and dying networks will change depending on the time-point chosen.

Secondly, biological differences among TBI models are known to influence Akt and mTOR signaling pathways, among others. Akt and S6 phosphorylation are downregulated after weight drop TBI in rats,^[38] while our group has shown the upregulation of phosphorylated S6 in glia after CCI.^[2] Thus Akt/mTOR signaling may be differentially regulated depending on the type of injury and the cell type (e.g., neuron vs glia) examined.

The age, cell composition and microenvironment of the engineered brain is different from cortical tissues found in living mice. Our 3D in vitro model is built of embryonic day 16 cortical neurons that were matured for 14 days in culture medium, whereas mice used in this study were 2–3 months of post-natal age. The 3D culture is lacking the native extracellular brain matrix (ECM) and dense cellular interactions with other cell types that make up the brain such as endothelial cells, monocytes, oligodendrocytes, and astrocytes and microglia were present in much lower quantities.^[50] Our tissue model is also smaller compared to the in vivo brain, it does not communicate with other organs such as the gut and lungs, and does not have a functional blood-brain-barrier.^[51] Notwithstanding the differences between our in vitro model and experiments in living mice, the similarities in the responses of 3D neuron cultures to CCI demonstrates the utility of our bioengineered 3D brain-like tissues to study brain trauma. Systematic additions of these cellular and biochemical components that are currently lacking will help accelerate the discovery of injury mechanisms.

To gain further insights in the mechanisms underlying the propagation of neural network destruction and cell death in vitro, we turned our attention to necroptosis, one of the well-documented mechanisms involved in traumatic cell death in vivo.^[32,36,52] In particular, several 2D in vitro and in vivo studies showed early phosphorylation (activation) of MLKL and subsequent neuronal death post injury.^[36,52] Consistent with these findings, we observed a rapid induction of pMLKL after injury, measured by immunofluorescence analysis at different times and locations in 3D cultures. Activation of MLKL, a biochemical marker specific for necroptosis, was associated with LDH release presumably caused by neuronal membrane permeabilization.^[53] At the plasma membrane, pMLKL is thought to sequester phospho-inosites PIP2 and PIP3 during assembly of a pore forming cytotoxic complex, precluding access to, and activation of, PI3-kinase, which may account for the reduced phosphorylation of pAkt, the substrate of PI3K.^[54] Whether the release of glutamate, an excitotoxic molecule,^[22] contributes to neurodegeneration in our in vitro model as it does in vivo^[55] remains to be established. Moreover, our data do not establish necroptosis as the major death mechanism as many other death mechanisms (e.g., apoptosis, parthanatos, autophagy, ferroptosis, others) may also contribute.

Significant advantages of the 3D in vitro model over in vivo studies of TBI include experiments to the role of cell autonomous responses and cell-type specific responses to injury. An integrative, human brain-like tissue culture model can be created from iPSC composed from specific cell-types, such as neural subtypes, microglia, astrocytes, endothelial cells for a more complete understanding of the pathogenesis of TBI in human. Overall, through systematic analysis of structure, function and molecular signaling we showed that our 3D in vitro brain tissue model replicates many of the features identified in vivo for TBI studies.

4. Materials and Methods

4.1. 3D silk scaffolds preparation

Scaffolds were prepared from silk fibroin solution extracted from Bombyx mori cocoons following our established protocols.^[56] In brief, silk cocoons were boiled for 30 minutes in sodium carbonate solution to separate sericin from fibroin. After the fibers were dried overnight in a chemical cabinet, they were dissolved in 9.3M solution of lithium bromide, followed by dialysis in deionized water for 3 days. A 6% (w/v) silk fibroin solution was prepared in a 10cm dish and incubated with sodium chloride particles with sizes varying from 300 μm to 425 μm diameter. After two days of incubation, the mixture is incubated for 1 hour at 65°C to trigger beta-sheet formation, forming a 3D sponge. The sponge is removed from the dish and subjected to two days of dialysis at room temperature to remove the sodium chloride. Scaffolds were cut with biopsy punches of different sizes (McMaster-Carr, Princeton, NJ, USA) to the desired size of 6 mm outside diameter and 2 mm and inside diameter, and trimmed with a razor blade to a height of 1,5 mm. The scaffolds were sterilized using an autoclave (20min) and stored for a maximum of 1 week at 4°C before usage.

4.2. Neuronal cell isolation

Brain cortices were collected from embryonic day 16 (E16) C57/Bl6 mice (Charles River Laboratories), following the established protocol.^[57] In brief, mouse cortices were freshly dissected and stored in HBSS solution for the duration of the dissection. Prior enzymatic digestion, cortices were washed twice with cold PBS. A cortical cell suspension was prepared by enzymatic digestion with a 2mL mixture of trypsin/EDTA (0.25%) (ThermoFisher) combined with DNase (final concentration 0.3mg/mL; Roche Applied Sciences). After 20min incubation at 37°C, Trypsin activity was inhibited by adding an equal volume of Neural Medium (NeuroBasal medium supplemented with 2% B27, 100U/mL penicillin, 100 ug/mL streptomycin, 2mM Glutamax (ThermoFisher)) and 10% of heat inactivated fetal bovine serum (hiFBS) (ThermoFisher)). The cell suspension was passed through a 100μm filter to remove cell aggregates and counted. Cells were then concentrated by centrifugation and resuspended in fresh neural medium at a cell density of 25 million cells per ml of Neural Medium. All animal procedures were approved by the Tufts University Institutional Animal Care and Use Committee and complies with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (Institutional Animal Care and Use Committee M2018–06).

4.3. 3D brain-like tissue model preparation

The conditions to generate the 3D mouse cortical tissue model were modified from our published 3D rat embryonic cortical neuron model.^[22] Silk scaffolds were coated overnight at 37°C with 0.1 mg/ml Poly-D-lysine (PDL) mixed with 50ug/mL laminin. The following day, scaffolds were washed 5 times with phosphate buffered saline (PBS) solution for 5 minutes. The scaffolds were conditioned with NM medium until the isolated neurons were ready for seeding.

Immediately before cell seeding, the scaffolds were placed in 96-well plates and a vacuum manifold was used to remove all liquid. Next, 40μl of a 25M/ml cell suspension was seeded onto the dried scaffolds and incubated for 30 minutes at 37°C to allow neurons to attach. Then 150 ul of Neural Media was added to each scaffold and the scaffolds were placed in a tissue culture incubator (37°C, 5% CO2 in a humidified atmosphere). The following day, cell seeded scaffolds were moved into new 96-well microplates, and a freshly prepared solution of collagen type I solution (Corning) was applied (100 ul; 3mg/mL; pH was adjusted to 7.0 with NaOH) to each scaffold. After an incubation for 30 minutes at 37°C to crosslink the collagen gel, 150 ul of Neural Media was added to all scaffolds and incubated for 24 hours at 37°C. The next day, the 3D brain-like tissues were moved into 48 well plates in 1 ml of NB media. Media was changed every fourth day until the day when the 3D tissues were subjected to CCI, 14 days after seeding.

4.4. Controlled Cortical Impact (CCI) model in vivo and in vitro

C57BL/6J mice (Stock 000664, The Jackson Laboratory) were subjected to 4.5% isoflurane (Anaquest) in 70% N2O and 30% O2 using a Flutec 3 vaporizer (Colonial Medical) to induce and maintain anesthesia. Anesthetized mice were placed in a stereotaxic frame and a midline scalp incision was made. A 5-mm craniotomy was performed using a drill (Harvard Apparatus) and a trephine (N/A) over the left temporoparietal cortex. The bone flap was removed, and the mice were subjected to CCI using a pneumatic cylinder with a 5-mm flat tip impactor at a velocity of 6 m/sec, penetration depth 0.6 mm, and a duration of 100 ms.^[58] After injury, the scalp was sutured closed (K831H, Ethicon) and the mice were placed in their home cage. Sham-injured mice underwent the same surgical procedure in the absence of CCI. All procedures were performed in accordance with the NIH Guide for Care and Use of Laboratory Animals and were approved by the MGH Institutional Animal Care and Use Committee.

In the in vitro model, 3D brain-like tissues were placed on a flat weight boat and subjected to CCI using the identical injury parameters and the same device as used for the in vivo CCI model. After injury, CCI and sham exposed 3D tissues were placed in a tissue culture incubator (37°C, 5% CO2, humidified atmosphere) for the indicated incubation times. The 3D tissues were then analyzed for LDH and glutamate release in the medium, cell death, neural network structure, western-blot and local field potential assays, all as outlined below.

4.5. Lactate Dehydrogenase Activity Assay

Cellular viability of the 3D brain-like tissues was measured with a lactate dehydrogenase assay (LDH) (MAK066–1KT, Sigma) following the manufacturer protocol. This assay is based on the ability of LDH to reduce nicotinamide adenine dinucleotide (NAD) to 1,4-dihydro nicotinamide adenine dinucleotide (NADH). In brief, culture media was collected at the indicated experimental time points, and frozen at −80°C until all samples were collected. 50μl of the supernatant was used per assay reaction, and amount of NADH was measured with SpectraMax M3 TECAN plate reader, at 450nm absorbance wavelength. LDH activity was calculated in two steps: initial absorbance read-outs were compared to the calibration curve to calculate the amount of NADH; after LDH activity was calculated based on the time of the reaction and the dilution applied. Calibration curves were prepared per every assay plate.

4.6. Glutamate Assay

Cellular glutamate release by 3D brain-like tissues was measured with a Glutamate assay kit (ab83389, Abcam) following the manufacturer protocol. The assay specifically detects free glutamate, but not glutamic acid, with glutamate enzyme mix. In brief, media was collected at the indicated experimental time points and frozen until all samples were collected. 50 ul of collected supernatant was used per each reaction. The reaction between enzyme and glutamate substrate was measured through absorbance at 450nm using SpectraMax M3 TECAN plate reader. Glutamate concentration was calculated based on the calibration curves, prepared for every plate.

4.7. In situ direct DNA fragmentation (TUNEL) assay

To measure cellular death in 3D brain-like tissues after CCI, a terminal deoxynucleotidyl transferase biotin-dUTP nick end labeling (TUNEL) assay was used (ab66108, Abcam) following the manufacturers protocol. As a positive control for cell death, 3D tissues were treated with 1uM staurosporine (Abcam) for 24 hours to induce apoptosis. At the indicated experimental time points, samples were fixed with ice-cold 1% paraformaldehyde (PFA) for 15 minutes and washed once with PBS at room-temperature. PBS was removed and ice-cold 70% ethanol was applied to all samples for 30 minutes at 4°C. Samples were subsequently stored at −20°C until all experimental groups were collected.

Samples collected at different timepoints within a single experiment were processed simultaneously. Ethanol was removed from the samples and replaced with wash buffer. The chemically fixed 3D brain-like tissues were incubated for 1 hour in 15 μl of the staining solution (Mix of reaction buffer, TdT enzyme, FITC-dUTP and ddH₂O) to mark cells with DNA fragments (i.e, dead cells). After washing twice with rinse buffer, 3D brain-like tissues were exposed to propidium iodide/RNase A solution for 30 minutes to label all cells present. Images of TUNEL and PI positive cells were collected with a Leica SP8 confocal microscope (Fluotar VISIR 25x/0.95 WATER objective, 2x optical zoom), Ex/Em=488/520 nm for FITC and 488/623 nm for PI. TUNEL/PI images were acquired with the same PMT gain settings, and laser power between 3 independent experiments and presented in Figure 6 as maximum intensity projection.

4.8. Immunofluorescence analysis

The 3D in vitro brain-like tissue models were fixed with 4% sucrose and 4% paraformaldehyde (PFA, Electron Microscopy Sciences) in Phosphate Buffered Saline (PBS) (Thermo Fisher) for 1 hour at room temperature. Next, the 3D tissues were washed with PBS and permeabilized for 1 hour with 0.2% TritonX-100 supplemented with 4% goat serum (permeabilization solution; ThermoFisher). Primary antibody incubation was performed overnight at 4°C in the permeabilization solution. Samples were washed five times with PBS while gently shaking. They were then incubated with secondary antibodies for 1 hour at room temperature, followed by 5 minutes incubation with DAPI (4’,6-Diamidino-2-Phenylindole). Unbound antibodies and DAPI was washed away by washing five times with PBS. Fluorescent image stacks of 3D brain tissues were acquired on a Leica SP8 FLIM confocal microscope (Leica Microsystems) (Fluotar VISIR 25x/0.95 WATER objective, 2x optical zoom). Images in Figures 1, 2, 3, and 4 represent maximum intensity projection, and were collected with the same PMT gain settings, and laser power between three independent experiments. Primary antibodies included: anti-alpha tubulin (ab78078 mouse; 1:1000); anti-MAP2 (ab5392 rabbit; 1:10000); anti-S-100 (ab52642 rabbit; 1:1000); anti-CD11b (ab8878 rat; 1:1000); anti-MLKL (phosphor S345) (ab196436 rabbit; 1:1000) purchased from Abcam; anti-GFAP (G3893 mouse; 1:1000); anti-Synapsin 1 (AB1543MI rabbit; 1:10000); anti-PSD95 (MAB1596, mouse IgG2a; 1:10000) obtained from Sigma; anti-gephyrin (147011, mouse IgG1) purchased from Synaptic Systems. Secondary antibodies: goat anti-mouse IgG, IgG1, IgG2 (PIA32723, A-21240, A-21135), anti-rabbit (A11037) or anti-chicken (A21449) Alexa 488 and 568, 647 (1:500; Thermo Fisher).

4.9. Electrophysiology: Local Field Potentials (LFPs)

Using micro-forceps, each sample was transferred from a well-plate containing culture media to a 35mm plastic petri dish (Corning) with 1.5 mL of warmed (37°C) extracellular solution, a formulation of artificial cerebrospinal fluid (ASCF). The solution was prepared with the following specifications (mM): 130 NaCl, 1.25 NaH₂PO₄, 1.8 MgSO₄, 1.6 CaCl₂, 3 KCl, 10 HEPES-NaOH, 5.5 glucose, pH 7.4. Once transferred, the dish was position over a WP-16 Warmed Platform (Warner Instruments) controlled by a DC-powered TC-134A Handheld Temperature Controller which maintained ACSF at a temperature of 37°C.

A silver-chloride (AgCl) reference electrode was placed in the ACSF at the outer perimeter of the dish, away from the sample at the center. Next, an adjustable rod was used to immobilize the sample by pressing it to the surface of the dish, stabilizing any unwanted movement. A single recording electrode was then lowered into the silk-region of the sample using micromanipulators. Each electrode consisted of an AgCl filament placed within a borosilicate glass pipette that was pulled with a Sutter P-97 (Novato, CA, USA) to achieve a resistance value of 40–80 MΩ and a tip diameter of approximately 2μm. The precise location of the recording electrode necessarily varied from sample to sample; however, we consistently satisfied the following criteria: the electrode should be within 50μm of a silk fiber at equidistance between the lumen of the central window and the outer perimeter of the silk scaffold.

Local field potentials (LFPs) were recorded continuously from the measurement site for 5 minutes using a sampling frequency of 2500Hz. Recurrent LFP spikes, brief (1ms) oscillations in potential difference (mV) indicative of spontaneous electrical activity, were evident throughout traces. An Axon Instruments analog-to-digital converter coupled to an Intan digital amplifier and HumBug 60Hz Noise Eliminator streamed LFP recordings to a desktop computer with a Windows XP operating system. Traces were represented in Clampex 10.7 (Axon Instruments) and eventually exported to Clampfit 10.7 to perform thresholding (±0.35 mV) which detects LFP spikes. Once detected, events were exported for further analysis and spontaneous electrical activity was computed by diving the total number of observed spikes over the observational window (LFPs/min).

4.10. Assessment of collagen structure after damage

Second harmonic generation images of the 3D extracellular matrix were acquired with Leica SP8 FLIM multiphoton microscope (Leica Microsystems). The 3D directional variance analyzed as previously described.^[59,60] Briefly, a simple intensity threshold was used to remove weakly fluorescent background signal and then a weighted vector summation was done to get the orientation of the fibers in terms of variance, where 0 corresponded to perfectly parallel fibers and 1 to random orientation.

4.11. Assessment of 3D neurite network length of 3D brain-like tissues

Image acquisition settings are specific to each experimental set-up and are indicated at the individual Figure captions. Z-stacks of images of Tuj-1 immuno-labeled 3D tissues were loaded into a 3-D analysis matrix in MATLAB. The fluorescence background signal of the silk and the signal emanating from the cell bodies were removed from the analysis matrices using Otsu’s thresholding method and eccentricity threshold (0.90) to remove more circular objects, leaving the more tube-like objects, respectively. Remaining neurite voxels were then grouped together into objects if neighboring pixels touched a voxel face or edge (18 total connection points; voxel size= 0.258 μm x 0.258 μm x 0.422 μm (x-y-z)). The smallest possible ellipsoid was then drawn around each object and the size (in voxel numbers of its largest principal axis was defined as the length of the neurite. The length of all objects was then calculated per image stack. The prevalence (or extent) of neural networks within an image stack was calculated as the percentage of voxels stained positive for Tuj-1 and included in the neurite length analysis compared to the total number of voxels in the stack.

4.12. Isolation of neurons after CCI of mice

C57BL/6J mice (Stock 000664, The Jackson Laboratory) were subjected to sham or CCI as specified in the above section. They were injected with Avertin intraperitoneally (250mg/kg) [LV1] at 24h after injury. The mice were transcardially perfused with PBS and the brains were dissociated using neural tissue dissociation kit (130-092-628, Miltenyi Biotec) according to manufacturer’s protocol. The myelin was removed using myelin removal beads (130-096-433, Miltenyi Biotec) and LS Columns (130-042-401, Miltenyi Biotec). The cell pellet was resuspended in PBS with 0.5% bovine serum albumin and the neurons were isolated using a neuron isolation kit (130-115-389, Miltenyi Biotec) and LS Columns. After the magnetic separation of neurons, the samples were centrifuged at 300 rcf for 10 minutes and the supernatant was discarded. The cell pellet was immediately processed for Western blot.

4.13. SDS PAGE and Western Blot

The 3D brain-like tissues were removed from the culture plate and placed on a dry Kim-Wipe for 10 seconds to remove excess media. The 3D tissues or the neurons isolated from mice after sham or CCI were placed in a 1.5 mL Eppendorf tube. 50 μL of RIPA lysis buffer (20–188, EMD Millipore) and protease/phosphatase inhibitor (1861281, Thermo Fisher) was added to lyse the cells. The samples were sonicated and centrifuged at 10,000 rcf for 20 minutes at 4°C. The supernatant was collected and Laemmli SDS-Sample Buffer (BP-110R, Boston Bioproducts) was added. The samples were denatured at 95°C for 5 minutes and run on a 4–20% polyacrylamide gel (456–1096, Bio-Rad) at 130V for an hour. The gel was transferred on to a PVDF membrane using iBlot2 (IB24001, Thermo Fisher) at 18V for 5 minutes. The membrane was blocked in 5% BSA in TBST for 1 hour and was incubated at 4°C overnight in primary antibodies (p-AKT, 3787, Cell Signaling Technology, 1:1000; AKT, 9272, Cell Signaling Technology, 1:1000; p-S6, 2211, Cell Signaling Technology, 1:1000, S6, 2217, Cell Signaling Technology, 1:1000; AT8, MN1020, Thermo Fisher, 1:1000; HMGB1, ab18256, Abcam, 1:1000; GSK3β, 9315, Cell Signaling Technology, 1:1000; Il-1β, Abcam, ab9722, 1:1000; LC3II, 3868, Cell Signaling Technology, 1:1000). Horseradish peroxidase-conjugated secondary antibody (Rabbit HRP – 7074, Mouse HRP – 7076, Cell Signaling Technology, 1:3000) was used for ECL (EMD Millipore) detection. The results were normalized to β-actin (5125, Cell Signaling Technology, 1:5000). Densitometry was performed using image analysis (Image J).

4.14. Data analysis

Statistical significance was established between experimental groups with GraphPad Prism 7 software. We used two-tailed t-tests to compare two experimental groups, and two-way ANOVA analysis of variance to compare multiple groups. Tukey’s post-hoc test was used to evaluate the significant difference between the experimental and control groups. Pearson’s correlation analysis was used to evaluate the correlation between TUNEL and pMLKL positive cells in the control groups. A p value less than 0.05 was considered statistically significant. All experiments were repeated three times, and technical replicas (2–6, depending on the experiment) were used for every assay. All quantified data presented are average and standard error of the mean of three independent experiments, with the exception of Figure 4, which is an analysis of a representative experiment. TUNEL and pMLKL image acquisition were performed blinded, where the regions of interests were selected through DAPI staining.