Design and evaluation of an in vitro mild traumatic brain injury modeling system using 3D printed mini impact device on the 3D cultured human iPSC derived neural progenitor cells

Abstract

Traumatic brain injury (TBI) is a severe public health burden affecting millions of people worldwide. Despite significant progress in understanding the neuro-pathological changes behind the biomechanical injury to the brain, promising preclinical therapeutics have seldom been translated into successful outcomes in clinical trials. One of the reasons is that preclinical models are mainly based on small animals, which have physiological and functional differences in the central nervous system (CNS) compared to humans. A more human relevant model is urgently required to develop new therapeutic strategies and to address the gap between preclinical studies and patient medical treatments. In this report, we have developed an in vitro mild TBI (mTBI) modeling system based on 3D cultured human induced pluripotent stem cells (iPSC) derived neural progenitor cells (iPSC-NPCs) to evaluate the consequence of single and repetitive mTBI by a 3D printed mini weight-drop impact device. Computational simulation was performed to understand the single and cumulative effects of weight-drop impact force on the NPC differentiated neurospheres. Experimental results demonstrated that the neurospheres showed reactive astrogliosis and glial scar formation after repetitive (10 hits) mild impacts, while no obvious astrocyte activation was found after one or two mild impacts. A 3D co-culture model of human microglia cells with the neurosphere was further developed. It was found that astrocyte response was promoted even after two mild impacts, which might be caused by the chronic neuroinflammation after microglia activation, characterized by the sustained increase of pro-inflammatory cytokines and chemokines. Our in vitro mTBI modeling system could recapitulate several hallmarks of the brain impact injury and might serve as a platform to better understand the disease mechanism, identify novel targets, and screen drug candidates.

1. Introduction

Traumatic brain injury (TBI) is a leading cause of disability for people less than 45 years old.^[1] Depending on the severity, TBI can be characterized by mild, moderate, or severe types.^[2] Mild TBI (mTBI) has the highest incidence among all new cases of TBI reported annually in the US.^[3] The risk of mTBI is found to be much higher for people playing contact sports, children being abused, and soldiers.^[4] TBI places a massive health and economic burden on the victims, and better treatment for TBI are in great need.^[5]

The pathophysiological change from TBI is caused by both primary and secondary injury mechanisms.^[6] Primary damage cause the shearing and/or stretching of axons, neurons, glia, and blood vessels.^[7] Secondary injury is responsible for neuronal degeneration and functional brain deficits and is mediated by various pathways including neuroinflammation,^{[8, 9]} which has been found to contribute progressively to the worsening outcome of TBI.^[10]

Neuroinflammation is composed of local and systematic immune activation and involves the participation and complex interaction of microglia, immune cells, cytokines, and chemokines.^[11] Microglia are major contributors in such neuroinflammatory response in the brain, as they can quickly respond to the injury by sensing the damage-associated molecular patterns (DAMPs).^[12] Signaling from DAMP receptors in microglia up-regulate the expression of various cytokines and chemokine, thus providing a local inflammatory environment.^[13] Microglia activation in the acute phase of TBI has some positive attributes,^[9] however, excessive microglia activation could disrupt the blood-brain barrier and produce neurotoxic chemicals and free radicals, exacerbating neuronal death.^[14] Astrocytes also play important roles in neuroinflammation by up-regulating the glial fibrillary acidic protein (GFAP) expression and releasing many pro-inflammatory cytokines and chemokines to propagate neuroinflammation.^[15] The significance of neuroinflammation in TBI has inspired an anti-inflammatory strategy to treat TBI,^[11] but there is yet to be a therapy that has shown efficacy in a large, multi-center Phase III clinical trial.

Due to the heterogeneity of clinical TBI patients, animal models (especially rodent models) have been developed to better understand the biomechanical mechanism, as well as molecular cascade, initiated by the impact and to test the therapeutic efficacy ^{[16, 17]}. Four animal models have been widely used in labs, which are comprised of controlled cortical impact (CCI) injury,^[18] fluid percussion injury (FPI),^[19] weight-drop impact acceleration injury,^[20] and blast injury.^[21] The detailed design, as well as strengths and limitations of each model, have been reviewed elsewhere.^[16] Besides the prevalence, many mTBI victims are at a high risk for multiple head concussions and susceptible to many mental disorders, however, the underlying disorder mechanism remains poorly understood.^{[22, 23]} Thus, animal models of repetitive mTBI have also been developed to mimic the clinical consequence of repeated mild injuries and to understand the cumulative effects of repetitive concussions. ^[24-26] Controlled cortical impact and weigh-drop impact approaches were most often used to create repetitive mTBI.^{[24, 27, 28]} Unfortunately, although promising treatment strategies based on those models have been discovered, they fail to improve the clinical outcome of TBI patients.^[16] One major challenge is the lack of preclinical testing of TBI therapy in large animals, especially primates, due to ethical and/or financial issues.^[16] The commonly used rodent models have innate physiological and functional differences in the central nervous system from humans, and such differences might cause distinct outcomes toward therapy.^{[16, 29, 30]} A more human relevant model is urgently required to develop new therapeutic strategies and to address the gap between preclinical studies and clinical treatments.

While a significant effort is being placed on improving preclinical testing in small rodent models,^[31] in vitro models can allow high-throughput drug screening and are more cost effective, and are thus popular alternatives in TBI studies.^{[32, 33]} However, current in vitro models still suffer from certain problems. Traditional 2D in vitro platforms using immortalized cell lines and dissociated primary culture could not recapitulate higher dimensional cell-cell interaction and the dynamic extracellular matrix (ECM) environment. ^[34] The favored 3D in vitro models based on organotypic culture or acute ex vivo preparations still rely on small animals’ brain tissues.^[33] The emergence of human induced pluripotent stem cell (iPSC) technology and progress in biomedical engineering have offered a strategy for the development of “brain-on-a-chip” system to study neurological diseases, which can solve both the 3D culture and human species issues and better mimic human diseases.^{[35, 36]} For example, a recently developed triculture microfluidic system consisting of iPSC derived neurons, astrocytes, and microglia has successfully modeled the neurodegeneration and neuroinflammation of Alzheimer’s disease (AD),^[37] which would aid the drug discovery for AD.^[36] However, it is still currently challenging to accurately model TBI or mTBI on a chip because of its complexity and limited understanding of its pathophysiology.

To facilitate the modeling of mTBI on a chip and basic mechanistic study of mTBI, we present our in vitro repetitive mTBI model using a 3D printed mini impact device on 3D cultured iPSC derived neural progenitor cells (NPCs).^[38] The mini impact device was fabricated by a 3D printer and fitted to the size to enable convenient, reproducible, and repeated weight-drop impact injury to the 3D cultured neurospheres formed by NPCs. The consequences of single and multiple mild impact injuries were both computational simulated and evaluated by experiments. We demonstrated that while a neurosphere subjected to single mild impact could recover from the injury with minimal and transient changes in pathology, repetitive mild impacts led to neuron loss, reactive astrogliosis, and glial scar formation, all are hallmarks of repetitive mTBI. To include neuroinflammation, a key component of mTBI, we co-cultured human microglial cells with the neurospheres and administered mild impact to the 3D co-culture system. The microglia were activated after impact and produced sustained release of more pro-inflammatory cytokines and chemokines, which promoted astrocyte response. This in vitro mTBI model is easily replicable and could be applied for high-throughput drug screening and serve as a good platform to study mTBI mechanisms and how it might be therapeutically targeted in the future.

2. Results

2.1. Development of 3D printed mini weight-drop impact device for neurospheres

The 3D printed mini weight-drop impact device contains three major parts: the guide construct, the neurosphere impactor and the neurosphere holder (Figure 1A&B). The guide construct (Figure 1C) was made to allow for the controlled free fall of the impactor (Figure 1E), while having it lined up with the spheroid holder (Figure 1D). The base is made of a supportive base section, with a slot to allow the spheroid holder to slide in and out, and two vertical pillars, with slots to allow the impactor to easily slide through them and stay in position. The spheroid holder was made with a small, half-sphere indentation in the center that allows for a spheroid (either alone or cultured in Matrigel or other matrix) to be placed in it and held in position. The holder slides into the slot of the base piece and has a handle for easy handling. The impactor is made with a half-sphere protrusion in the center, which matches the indentation in the spheroid holder. It also has two arms that fit into the slots in the pillars of the base piece. These allow the impactor to freefall while staying lined up with the spheroid holder. When in their respective slots, the impactor and spheroid holder are aligned to have the protrusion fit into the indentation.

Figure 1.

Mini-impact device design and detailed part parameters. Design (A) and appearance (B) of 3D printed mini weight-drop impact device; (C-E) detailed size parameters (in mm) for different parts of the impact device: C-the guide construct, D-the neurosphere holder, E-the neurosphere impactor.

2.2. Simulation results revealed accumulated cell damage after multiple impacts

The distribution of minimum principal strain in the neurosphere at the maximum indent depth, and impact induced cell damage for half impact, one impact and two impacts are shown in Figure 2A&E. It was observed that the neurosphere underwent a larger compression at the maximum indent depth during the impact load. The history of the minimum principal strain at the center of the neurospheres was shown in Figure 2C. The damage was quantified by equivalent plastic strain (PEEQ). As expected, PEEQ increased from half impact to one impact and two impact. The number of damaged cells caused by the impact was reflected by the area in the neurosphere with PEEQ value larger than 0.49 and it was found that the impact-induced more cell damage in the neurosphere, which increased from 0 to 12.4% and 19.6 % for half impact to one impact and two impacts (Figure 2F). The shear deformation was also investigated in our model (Figure 2B&D). The edge of the neurosphere experienced a larger shear strain than center and repetitive impact also induced a greater shear deformation (Figure 2B).

Figure 2.

Finite-element modeling of the neurospheres under different impact conditions using Abaqus software. A) Distribution of the compression strain; B) distribution of the shear strain; C) history of the minimum compression at the maximum indent depth; D) history of shear strain at the maximum indent depth; E) distribution of PEEQ in the cross-section area of neurosphere after various impact conditions; F&G) quantification of percentage areas with PEEQ >0.49 (maximal PEEQ value found in the neurosphere after half impact).

The simulation result of repetitive impacts with half height and full height shows that the injury in the neurosphere accumulated progressively. Compared with single impact, the ratio with area of PEEQ >0.49 after 10 impacts increased from 0 to 15.4% for dropping height of 15 mm, and from 19.6% to 35.4% for dropping height of 30 mm (Figure 2E& G).

2.3. Single mTBI did not affect proliferation and differentiation of 3D cultured neurospheres

Neurospheres formed by the iPSC-NPCs differentiated in neural differentiation media had axons radially growing out after they were embedded in Matrigel (Figure 3B). Prior studies indicated that our NPCs could successfully differentiate into cortical neurons and astrocytes in vitro, which has been used to mimic the early development stage of human brain.^[38-40] In the current study, after two-week differentiation, the neurospheres were impacted by our designed, 3D printed mini weight-drop device. After impact, the axon outgrowth was found with more damage in the one-impact group compared to the half-impact group (Figure 3B). The lactate dehydrogenase (LDH) released into the media was collected to monitor the neural damage directly related to the injury. A significant increase in LDH was found in Day 1 (D1) media but quickly decreased to a normal level and maintained normal levels during the rest of the study period in the half impact and one-impact groups (Figure 3C). The LDH level was, on average, 1.8 times higher in the one-impact group than in the half-impact group. IF staining of Caspase-3 on the D7 neurosphere samples showed that samples subjected to impact from 30 mm height had significantly more apoptosis than samples receiving impact from 15 mm height (Figure 4B). However, such impact did not affect the neurosphere proliferation compared to the control, as shown by Ki-67 staining (Figure 4B). The long-term (D27) neuronal maturation and astrocyte activation in the neurospheres, showed by neuronal nuclear protein (NeuN) and GFAP staining, also showed no significant difference among the impact and control groups (Figure 4B). All those results revealed that the single mild injury, based on current impact conditions (either at 15- or 30-mm height), only caused limited cell death without gross neuropathology or affecting the neuronal maturation and astrocyte activation.

Figure 3.

Development of an in vitro single mild TBI model. (A) Timeline and design of single mild impact study: neurospheres embedded in Matrigel were static cultured in neural differentiation media starting from -D14 to -D7 and then dynamic cultured from -D7 to D27; four groups were evaluated: sham control, half impact (initial impactor height of 15 mm), one impact (initial impactor height of 30 mm) and two impacts (two impacts at D0 and D3 with initial impactor height of 30 mm); (B) appearance change of neurospheres during the impact study: axons were growing out from the neurospheres after seeding in Matrigel; impact injury caused more radial axon damage in the one-impact group; (C) LDH release from D0 to D6 after initial impact in different groups: the identification of the second peak in LDH release from the two impacts group confirmed that the impact device could deliver reproducible impacts to the neurosphere. Scale bar =250 μm.

Figure 4.

IF staining of Caspase-3 and Ki-67 staining in sham control and mild impact groups at D7 and NeuN (top) and GFAP (bottom) staining in sham control and mild TBI groups at D27. (A) Representative IF staining of Caspase-3, Ki-67, NeuN and GFAP in different groups; (B) Quantification and statistical analysis of the IF staining: a higher impact height and two impacts caused more Caspase-3 positive cells, while no significant difference in Ki-67, NeuN, and GFAP positive cells was identified between impact groups and the control. Nuclei were stained by DRAQ-5 and shown in blue color. Scale bar (white) = 100 μm; scale bar (yellow) = 50 μm. **P < 0.01, *P < 0.05 (one-way ANOVA with post-hoc Tukey test). All the data are presented as mean ± SD and n = 3.

The initial attempt to evaluate the consequence of repetitive mild impact injury was conducted by impacting the neurosphere twice from a height of 30 mm (full height) at D0 and D3, respectively. The second impact caused a secondary increase in the LDH level in the conditioned media at D4 with about a similar amount as the first impact. The LDH level was again reduced to a normal level on the second day after the secondary injury (D5). Caspase-3 staining confirmed more apoptotic cells in the two-impact groups than the half impact and one-impact groups (Figure 4B). However, no significant difference was found for the relative ratios of cells with positive Ki-67, NeuN, and GFAP staining in the two-impact group, as compared to the control (Figure 4B). This might imply that the accumulation of only two mild injuries, based on the current impact magnitude, was still below the threshold to induce typical TBI consequence, particularly characterized by astrocyte activation. Quantitative PCR results also validated our finding. No significant difference was found in the NeuN and GFAP gene expressions between all of the mild impact groups and the control (Supplementary Figure 2).

2.4. Repetitive mTBI caused astrogliosis and glial scar formation.

The initial failure to induce the astrocyte activation after only two impacts led us to administer more mild impacts to the 3D cultured neurospheres to mimic the repetitive mTBI. A 72-h interval was present between every two injuries, with reference to many developed mTBI animal models. The consequence of the repetitive injury was evaluated at two different heights using the mini weight-drop device. IF staining was performed after the last injury at D27, and NeuN staining indicated significantly less mature neurons in the half height and full height impact groups (Figure 5B&D), which might be caused by the neuron loss after repetitive mild impact injury, a common finding reported in many mTBI animal studies.^[41] Another hallmark in repetitive mTBI is characterized by the upregulation of GFAP and was also found in both half height and full height impact groups (Figure 5E). Reactive astrocytes with prominent cell body hypertrophy were identified in both impact groups (Figure 5B). Besides, it was found that the peripheral region of the neurospheres seemed to be occupied by the reactive astrocyte, as indicated by having the most GFAP with little Tuj1 staining in the peripheral, as compared to the control (Figure 5C). The peripheral astrocyte formed a “scar like” region to encapsulate the inner remaining neurons, which is similar to a glial scar found in TBI patients and animal models.^{[42, 43]} Quantitative PCR results showed no significant change of NeuN gene expression but a substantial up-regulation of GFAP gene expression in both repetitive impact groups, as compared to the control (Figure 5F & Supplementary Figure 3). In summary, the above results demonstrated that the accumulation of, at most, ten times mild injury, based on the current impact magnitude, produced a representative pathology of repetitive mTBI: neuron loss, reactive astrogliosis, and glial scar found in animal and patients, and our system was able to mimic repetitive mTBI response in vitro.

Figure 5.

Development of an in vitro repetitive mild TBI model. (A) Timeline and design of repetitive mild impacts study: in the half-height group, neurospheres were impacted every three days until D27, with the impactor height set at 15 mm; in the full height group, neurospheres were impacted every three days until D27 with the initial impactor height set at 30 mm. (B) Representative IF staining of NeuN and GFAP in different groups at D27. (C) A glial like structure was formed in the periphery of the neurosphere in the repetitive mild impacts group (30 mm height). (D&E) Quantification and statistical analysis of IF staining revealed that repeated impacts from both 15 mm and 30 mm heights caused a significant decrease in NeuN positive cells and an increase of GFAP expression, compared to the control. (F) qPCR analysis confirmed the up-regulation of GFAP gene expression in the repeated impacts groups. Nuclei were stained by DRAQ-5 and shown in blue. Scale bar = 50 μm. **P < 0.01, *P < 0.05 (one-way ANOVA with post-hoc Tukey test). All the data are presented as mean ± SD and n = 3.

2.5. Microglia co-culture revealed important role of neuroinflammation in astrocyte activation after mTBI.

It is well known that microglia play important roles in TBI progression,^[44] so a better in vitro TBI modeling system should also include microglia for evaluation. In our study, we incorporated HMC-3, an established human microglia cell line into the Matrigel around the neurosphere and conducted some preliminary studies to validate our model system.

Microglia were introduced immediately after first impact by removing the original Matrigel matrix outside of the neurosphere, followed by embedding the neurosphere into fresh a Matrigel solution containing microglia. To prove the spatial distribution of the microglia and neurospheres, we labeled the NPCs with cell-tracker green and labeled the microglia with cell-tracker red before the incorporation. Neurospheres with the axon outgrowth were found surrounded by dispersed microglia (Figure 6B). The microglia continued to grow and formed spheroids inside the Matrigel. The microglia would not greatly interfere with the administered impact injury, as significant axon disruption was still recognized after the second impact (Figure 6C). The migration and infiltration of microglia into the neurosphere after impact injury was confirmed by staining the impact sample with neurofilament and ionized calcium-binding adapter molecule 1(Iba-1), and the overlay of neurofilament and Iba-1 staining could be identified in the peripheral region of the impacted neurospheres (Figure 6D), which was absent in the control group.

Figure 6.

Formation of a microglia and neurosphere 3D co-culture repetitive mTBI model. (A) Timeline and design of 3D co-culture impact study: microglia were incorporated into the neurosphere after the first impact at D0, and then the co-culture of microglia and NPCs was impacted again at D3. (B) the appearance of co-cultured microglia (red) and NPC (green): microglia were distributed around the neurosphere; (C) appearance difference between control and two mild impacts group at D6: substantial axon recession was found in the impact group; (D) Iba-1 and neurofilament staining on the sample (D7) from the two-impact group indicates the infiltration of microglia into the neurosphere, which is absent in control group. Nuclei were stained by DRAQ-5 and shown in blue. Scale bar (yellow, top) =500 μm, scale bar (white, top) =100 μm, scale bar (white, bottom) =50 μm.

IF staining of Caspase-3 confirmed that injury caused a significant increase of apoptosis in the two-impacts group (Figure 7B); however, no significant difference in Ki-67 staining was found in the neurospheres between control and impact groups (Figure 7B). NeuN and GFAP staining were also conducted in co-culture samples at D27. To our surprise, a significant decrease in NeuN positive cells and a substantial increase in GFAP expression (body hypertrophy) were detected in the two-impact groups (Figure 7B). The findings were quite different from the group without microglia subjected the same two impacts (shown in Figure 4). We speculated that the discrepancy was due to the increased release of pro-inflammatory cytokines by microglia after activation. A cytokine array assay was thus performed in conditioned media collected at D7 in several different groups. More pro-inflammatory species, such as macrophage inflammatory protein-1 (MIP-1)α/β, chemokine ligand 5 (CCL5), and interleukin 6 (IL-6), were identified in the control and impact groups with microglia (Figure 8A), suggesting their major origin to be from microglia instead of neurospheres. When comparing the cytokine level changes between control and impact groups, in the two groups without microglia, only macrophage migration inhibitory factor (MIF) level was slightly increased, while both plasminogen activator inhibitor-1 (PAI-1) and interleukin 18 (IL-18) showed evident decreases in the impact group; however, in the two groups with microglia, besides MIF, cytokines (chemokines), including MIP-1α/β, CCL5, and IL-18, were significantly increased, and IL-6 showed a slight increase on average in the impact group (Figure 8B). The increased release of pro-inflammatory cytokines and chemokines revealed the activation of microglia after impact injury. To determine how long the activated microglia last, we have collected and measured the cytokine levels in the control and impact groups with microglia at D7, D14, D21, and D27. Three major chemokines and cytokines: MIP-1α/β, IL-6, and IL-8 have shown sustained increases in the impact group compared to the control. MIP-1α/β showed the greatest difference at D7 but gradually decreased. Even until D27, the MIP-1α/β level was over two times higher in the impact group than the control. IL-6 gradually increased and displayed the largest difference at D21 and then decreased back to control level. IL-8 did not exhibit any difference at D7 but showed a marked increase at D14 (Supplementary Figure 4) and then gradually decreased. The CCL-5 and IL-18 change was not given, as its increase after impact was only found at D7 and not at the other tested time points.

Figure 7.

Caspase-3, Ki-67, NeuN, and GFAP IF staining in control and impact groups from the co-culture model at D7. (A) Representative IF staining images of Caspase-3 and Ki-67 in the control and two-impacts group at D7 and NeuN and GFAP in the control and two-impacts group at D 27; (B) Quantification and statistical analysis of the IF staining showed that: two-impacts injury caused significantly more Caspase-3 positive cells, reduced NeuN positive cells, and increased expression of GFAP, while no significant difference in Ki-67 positive cells was identified. Nuclei were stained by DRAQ-5 and shown in blue. Scale bar (white) = 100 μm and scale bar (yellow) = 50 μm. **P < 0.01, *P < 0.05 (one-way ANOVA with post-hoc Tukey test). All the data are presented as mean ± SD and n = 4.

Figure 8.

Cytokine array assay in conditioned media from different control and impact groups. (A) Representative cytokine array images in the conditioned media collected at D7 from four study groups: control without microglia, two impacts without microglia, control with microglia, and two impacts with microglia. (B) Semi-quantitative analysis of cytokine and chemokine release profiles in the conditioned media collected at D7 from four groups indicated that the incorporation of microglia into the neurospheres resulted in increased secretion of several pro-inflammatory cytokine and chemokine species (i.e. IL-18, MIP-1α/β, and CCL5) after impact injury. (C-E) Semi-quantitative analysis of three pro-inflammatory cytokines and chemokines (i.e. MIP-1α/β, IL-6, and IL-8) release profiles in the conditioned media collected at D7, D14, D21, and D27 from the control and impact groups of the 3D co-culture model. **P < 0.01, *P < 0.05 (Student’s t-test). All the data are presented as mean ± SD and n = 4.

3. Discussion

The necessity of developing in vitro TBI models arise from not only the ethical and economic issues of animal models but also from the biological mismatch between humans and frequently studied animals.^[30] Small animal models are important tools in TBI research, but they also show remarkable differences in the brain structure and function (e.g. brain geometry, white to grey matter ratio, astrocyte activity et al.) compared to humans and thus produce different responses to neural injuries, such as TBI.^[45] Besides, many animal models would require surgeries and use of anesthesia, and both have been found to interfere with the pathophysiology of mild impact injury.^[46] A convenient and human cells-based in vitro model that can not only mimic the concussion related primary damage but also replicate the dynamic change related to the secondary damage in TBI is highly encouraged.

Although significant progress has been made in TBI in vitro models based on organotypic culture or acute ex vivo preparations,^{[32, 33]} those models are still dependent on small animals’ brain tissues. Another rising option is based on 3D culture models of human neurons and astrocytes in various scaffolds.^[30] To introduce mechanical injury to the 3D cultured neurons and/or astrocytes, shear deformation devices, compression apparatus, and weight-drop approaches have been applied so far.^{[47, 48]} All those methods have produced impact severity dependent neuronal damage and certain TBI related responses similar to the animal models.^[30] However, those studies mainly focused on comparing the effects of injuries with different magnitude (or strain) after single impact. Few models have been targeted to the cumulative consequence of repetitive mild impacts or mild TBI, the most prevalent type of TBI.^[26] Besides, many of those reported models rely on certain devices that are not convenient to use or easily available. Although there is substantial evidence in animal models that repetitive mTBI leads to more severe neuropathological damage than single mTBI,^[49] the repetitive mTBI has rarely been modeled in vitro.

In our current system, we are more interested in creating a repetitive mTBI model. We have applied 3D printing to fabricate a mini weight-drop device, a device that is often used in animal models of TBI and mTBI.^{[27, 50]} The device contains three parts: the base, the spheroid holder, and the impactor. The fabrication process was facile and inexpensive, and the device is easy to set up. The device can be easily replicated in labs and allows controlled and reproducible impact to a variety of samples, including spheroids, organoids, and ex vivo cultured tissues. Human iPSCs are gaining increased attention for developing various in vitro models.^[51] Instead of using a co-culture of neurons and astrocytes, as previous models used,^{[52, 53]} we applied neurospheres formed by iPSC-NPCs. Neurospheres were 3D cultured in Matrigel and differentiated into neurons and astrocytes simultaneously. After 2-week differentiation, they were subjected to the weight-drop impact injury. The strain or magnitude of the impact was adjustable by altering the height of the impactor before its free-falling. The impactor weighs around 1 gram, one tenth of the weight used in a reported in vitro weight-drop TBI model,^[48] and our evaluated maximal height was only 30 mm, which we believe would limit the direct damage and only administer mild impact injury to the 3D cultured neurospheres. We are developing a prototype device with the impactor delivered by controlled pistons instead of free-falling, enabling more delicate control of the impact injury.

We first performed computational simulations to understand the mechanical damage to the neurosphere after single and multiple impacts. Finite element simulation has been applied to understand and predict TBI in animals.^[54] Here we applied it for an in vitro neurosphere impact injury. Our simulation results illustrated that the neurospheres were mainly subjected to compressive strains caused by the falling of the impactor, which led to the neurosphere deformation and damage. The 30 mm height impact (one impact) was calculated to generate a significantly larger strains than the 15 mm height impact (half impact) (Figure 2A&C), resulting more severe damage (Figure 2E&F). Multiple impacts from both half height and full height resulted in a cumulative and progressively damage to the neurospheres (Figure 2E-G). Shear strain was calculated to be greater in the edge of neurosphere enriched with axons (Figure 2B). Axons are known to be vulnerable to the shear injury, and the diffuse axonal injury (DAI) was supported by microscopic observation (Figure 3B and Figure 7) showing the apparent disruption of axon fiber growing out of neurosphere.

To validate the biological relevance of our mTBI model, we experimentally evaluated the effect of single and multiple-impact injury on the neurospheres. Clinical findings have shown that although victims of single mTBI easily recovered within a short period of time, ^[55] numerous victims of repeated concussions have suffered from neurological disorders, such as headaches, dizziness, sleeping disturbances, and cognitive deficits, which may be related to chronic neuroinflammation.^{[23, 56]} Many animal studies of mTBI have proved that repetitive mTBI caused more severe inflammation changes than a single mTBI, including increased and widespread astrogliosis (GFAP up-regulation) and microglia activation.^{[57, 58]} In our model system, neither half-impact, one impact, nor two impacts have induced substantial changes in the proliferation, neuronal maturation (NeuN+ cells), or astrocyte activation (GFAP+ cells). However, both the level of the transient LDH increase and the Caspase-3+ cell ratio correlated well with the impact magnitude and impact number, although the apoptotic cell ratio was less than 5%, even after two impacts. Previous studies have found for single mild brain injury, the apoptosis ratio could reach to around 10% maximal and declined quickly post-injury (after 24-72 h). While for single moderate brain injury, the apoptosis ratio could peak to approximately 20% maximal and the apoptosis ratio remained at high levels (> 10%) for much longer time (96 h to 1 week). Our results implied that each single-impact injury could be considered as “mild”, and the neurosphere was not significantly affected by single or even two mild impact injuries. After confirming each single impact as a mild injury, repetitive impacts were further evaluated. Instead of focusing on finding out the threshold number of impacts that would produce characteristic repetitive mTBI consequences, we evaluated an extreme condition, in which neurospheres were repeatedly impacted every 3 days until the end of the study (total 10 hits). The inter-injury interval was selected based on the work of many animal models.^[24] Emerging evidence is showing that the inter-injury interval is an important factor affecting the development of TBI, and the same amount of concussions with a short interval would increase the acute pathology (neuron death and glial activity) as compared to those with a long interval,^{[58, 59]} which could be further tested in our system. We found that the matured neuron ratio was significantly decreased in the half height and full height impact groups, which might be attributed to the progressive neuron death and loss after repetitive impact injuries, given no NeuN mRNA expression difference between impact and control groups. Also, neurons have been shown to be more vulnerable to high strain rate than astrocytes, as proved in another in vitro TBI model study.^[52] Astrocytes were activated by exhibiting cell body hypertrophy and increasing GFAP expression.^{[60, 61]} Under healthy conditions, astrocytes seem quiescent but play vital roles in supporting the integrity of neuronal function.^[62] In response to TBI, astrocytes undergo astrogliosis and switch into a reactive phenotype.^{[60, 63]} GFAP is the primary intermediate filament in astrocytes,^[64] and its significant increase is considered a hallmark change in reactive astrocytes after TBI.^[65] Astrogliosis is essential for the early support of neural tissue regeneration, as it acts to isolate the non-injured tissue from the lesion and control the spread of inflammation.^{[61, 66]} More severe injury is linked with higher GFAP and more hypertrophy, and the most prominent response of reactive astrocyte is the glial scar formation,^[63] as we have observed (Figure 6D). The glial scar is considered to have opposite roles for neurons: in one way it provides a favorable environment for the remaining neurons; in another way it hinders axonal regeneration and neuronal connectivity restoration.^{[60, 67]}

To further delineate the importance of neuroinflammation in mTBI, we generated a 3D co-culture model containing human microglial cells and neurospheres. Microglia are the primary guardian and play a critical role in immune surveillance in the CNS.^[68] After TBI, microglial activation occurred before astrogliosis, as Iba-1 upregulation was found at day 1 post injury, while upregulation of GFAP was only upregulated after 3 days post injury.^[69] Activated microglia were identified in animal models of repetitive mTBI and could last over months to a year.^[70] To better understand the neuroinflammatory cascade after TBI, for the first time, we have incorporated the microglia into an in vitro 3D culture model (in scaffolds) of TBI. Activation of microglia after mild impact was confirmed in our system by the infiltration of microglia into the neurosphere (Figure 7B), as well as elevated levels of secreted pro-inflammatory cytokines (Figure 8B&C). This may imply the easier activation of microglia than astrocytes in mTBI with axonal injury and is in line with reported studies.^[71] Many secreted chemokines and cytokines were up-regulated after impacts, including MIF, CCL5, IL-18, MIP-1α/β, IL-6, and IL-8. This is highly relevant to the clinical and animal results ^{[72, 73, 74]} and is quite a contrast to the reaction of the two-impacts group without microglia, which showed only slight MIF increase after impact (Figure 8A). It is noteworthy that rodents lack a direct homologue of IL-8, so the roles of IL-8 in TBI have not been fully characterized in animal models.^{[72, 75]} Our findings validated the value of including human microglia in in vitro TBI models. Three major cytokines (MIP-1α/β, IL-6, and IL-8) have exhibited persistent increases over the study period, creating a chronic inflammatory environment, and is in agreement with the long-lasting activation of microglia in animal models of mTBI.^[76] All three cytokines play critical roles in mediating neuroinflammation after TBI. Clinical studies have reported that levels of pro-inflammatory factors such as IL-6 and IL-8 correlated closely with the severity of brain trauma injury and complication rates.^[77] MIP-1α/β performs biological functions to recruit inflammatory/immune cells to the inflammation site and has been found to be induced easily by concussions.^[78] The matured neuron ratio was significantly decreased, even after two impacts in the co-culture mTBI model, which was either due to the neurotoxic effects of those continuously increased microglia-derived pro-inflammatory factors^{[69, 79]} or the neurotoxic phenotype of the astrocytes induced by activated microglia.^[80] Astrocytes showed increased GFAP expression, which was absent in the two-impacts group without microglia (Figure 7A vs Figure 4A). This might be attributed to combined effects of elevated microglia-derived pro-inflammatory factors, as many studies have documented critical roles of those factors in astrocyte activation.^{[74, 81]} Our result is also consistent with earlier studies showing enhanced astrocyte responses by the early microglia activation after injury through co-culture of microglia with astrocytes. ^{[69, 82]} The promoted astrocyte response underscored the roles of microglia and neuro-inflammation in mTBI. Although mild impact led to increased neuron apoptosis in the NPC differentiated neuron-astrocyte system, astrocyte was not easily directly activated by only one or two mild impact without the presence of microglia. This indicates that the astrocyte activation would require high-strain mechanical stress or multiple impacts. Thus, the production of various cytokines and chemokines were limited. The existence of certain up-stream pro-inflammatory cytokine may also contribute to the difference. One example is IL-6, which has been found to induce the secondary release of many other cytokines from both neurons and astrocytes.^[83] IL-6 was at undetectable level in the no-microglia group as astrocyte was probably not activated but was evident in the microglia group even without impact. The presence of IL-6 produced by microglia, could trigger the release of many more cytokines and resulted in more profound change of those cytokines.

Like other in vitro models, certain limitations may exist in our system. The first limitation is that the complexity of TBI could not be fully mimicked in vitro. Other than neurons, astrocytes, and microglia, many other immune cells, such as neutrophils and macrophages, and brain microvascular endothelial cells are also involved the disease progression.^[84] The blood-brain-barrier integrity is damaged and contributed to the infiltration of immune cells and the worsening outcome of the disease.^[85] The interactions between injured neurons/ astrocytes with those other components are quite challenging to be modelled at the same time. The second limitation is lack of behavior change evaluation by the in vitro models. Behavioral assessment is critical to evaluate both short-term and long-term effects on the cognitive, sensorimotor, and emotional function after TBI,^[24] but it is almost impossible to directly model them with in vitro models. The third limitation comes from intrinsic materials we used in our model. Matrigel has been successfully implemented in neural tissue differentiation,^[86] but it is derived from tumor tissues and has batch-to-batch differences.^[87] The advancement of neural tissue engineering and biomaterials should address this issue.^[88] Human iPSC-NPCs coming from different origins may respond diversely to the same impact injury. Cerebral organoids with structures and functions more similar to the brain may be a better choice for evaluation.^[89] The currently used microglial cell line only retains partial characteristics of the original human microglia,^[90] so its activation may also differ from the in vivo reaction. The emerging iPSC derived microglia may offer an alternative for future applications.^[91] In the future, an electro-physiologically mature neuronal network can be developed using the reported method ^[92] and used in our model system to probe post-traumatic epileptogenesis, as epilepsy is frequently found in victims of mTBI and could be associated with the glial scar.^{[43, 93]} We also plan to apply single-cell sequencing to evaluate the roles of microglia and astrocytes in the 3D co-culture model after mTBI.^[94] It is also envisioned that iPSCs derived from patient-specific fibroblasts may offer a better opportunity for studying the effects of specific gene defects or polymorphism on the individual TBI risk and potential therapy.^[95]

4. Conclusion

In summary, we have developed an in vitro mTBI model that reproduces many distinctive features of mTBI (e.g. axonal injury and gliosis) using a 3D printed mini weight-drop impact device on 3D cultured human iPSC-NPC based neurospheres. This model allows us to easily deliver reproducible mechanical forces to the 3D cultured neurospheres and study the cumulative consequences of repeated impact injuries and the role of chronic neuroinflammation in TBI. We demonstrated that, while a single mild impact induced minimal pathological changes, multiple and repetitive mild impacts could induce reactive astrogliosis and glial scar formation. We also showed that, although two mild impacts could not directly activate astrocytes by the impact injury, microglia were easily activated and could create sustained neuroinflammation and promoted astrocyte response. Our model may advance the development of “brain-on-a-chip” modeling of TBI responses and can be employed for mechanistical study, target identification, and high-throughput drug screening of mTBI to facilitate the clinical translation of mTBI treatment.

Experimental section

Design and fabrication of a mini weight-drop impact device by 3D printing

The mini-impact device was designed like a weight-drop device used in animal models of TBI and was 3D printed with three separate pieces, as shown in Figure 1. The computer-aided design (CAD) files for the parts were designed using SolidWorks. The three major pieces (neurosphere impactor, guide construct and neurosphere holder) of the mini-impact device were 3D printed by using a digital light processing (DLP) printer (EnvisionTEC Vida) and Clear Guide resin (EnvisionTEC). The printing resin is biocompatible and is used for dental applications. The detailed dimensional parameters of the device were shown in Figure. 1C-E. The device was sterilized with 70% ethanol and UV before use.

Finite element simulation of single and multiple-impact related injury on the 3D cultured neurospheres

The experimental setup was recaptured with finite element models to further delineate the dynamic behaviors of neurosphere with Abaqus software. Three-dimensional geometry of the Matrigel embedded with a cellular spheroid is shown in supplementary Figure 1. The neurosphere with a diameter of 0.5 mm, was covered by the Matrigel with a depth of 0.25 mm. The Matrigel had a semi-spherical shape with a radius of 1.5 mm. In total, the Matrigel and neurosphere were meshed with 88821 linear tetrahedral elements.

In this work, the Matrigel has a Young’s Modulus of 450 Pa and Poisson’s ratio of 0.45. ^[96] The nonlinear mechanical behavior of the neurosphere was described with a modified Mooney-Rivlin strain energy function W:

$$W=a_{10}(I_1-3)+a_{30}(I_1-3{)}^3$$

Where I₁ is the first invariant of the right Cauchy-Green strain tensor C with I₁=trace(C), a₁₀ and a₃₀ are the two material constants, which is 240 Pa and 3420 Pa, respectively.^[97] To quantify the impact-induced injury in the neurosphere, the irreversible deformation, i.e., plastic deformation, was considered with the yield strength of 20 kPa. The free-fall semi-spherical impactor, with a diameter of 3 mm, was modeled as a rigid body with the mass of 0.9 g.

Base on experimental setup, three scenarios of simulation were performed: single impact with dropping height of 15 mm and 30 mm (Half impact and One impact), two-impact with dropping height of 30 mm (Two impacts), and ten-impact with dropping height of 15 mm and 30 mm (Half height×10 and Full height×10), respectively. The duration of the dropping time was calculated according to the dropping height and gravity acceleration, which is less than 0.06 s for dropping height of 15 mm and less than 0.08 s for dropping height of 30 mm. So, we set the simulation time 0.06 s for dropping height of 15 mm and 0.08 s for dropping height of 30 mm in simulation. The bottom spherical surface of the gel was fixed. A general contact (multiple impact with half height and full height) enforced between the impactor and hydrogel with a frictionless contact behavior. The cell damage was quantified in terms of the equivalent plastic strain (PEEQ) and the damage induced cell death was considered with PEEQ value larger than 0.49, which was the maximum PEEQ observed in the neurosphere after half impact (we consider single one half impact caused almost none cell death). The percentage of area with PEEQ > 0.49 was quantified in different scenarios.

3D cell culture and differentiation of iPSC-NPCs

The iPSC-NPCs were prepared by following our reported protocol. The iPSCs were first reprogrammed from fibroblasts of a healthy individual, and then an embryoid body-based differentiation procedure was used for the differentiation of the iPSCs to NPCs.^{[39, 98]} The NPCs were plated onto 1:100 diluted Matrigel (Corning) coated 6-well plate and expanded in NPC growth medium, containing a mixture (1:1 v/v) of Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F12 from Cytiva) and neurobasal medium (Gibco), supplemented with 1× N2 (Gibco), 1× B27 (Gibco), 20 ng mL⁻¹ bFGF (Peprotech), and 1% Pen/Strep (Gibco) at 37 °C in a 5% CO₂ incubator. The cell-culture medium was changed every other day until the cells were confluent. The NPCs were harvested using TrypLE (Gibco), re-suspended in NPC growth medium, and seeded at 1.5×10⁴ cells per well in 150 μL of medium in a 96-well non-adherent U bottom plate (Corning) for neurosphere formation. The NPCs gradually formed spheroids, and the spheroids were allowed to grow for one week in the 96-well plate. The spheroids were then embedded in Matrigel (each spheroid in 30 μL) following a previously published protocol ^[99], and two spheroids (in Matrigel) were transferred into each well of a 6-well plate and cultured in neural differentiation medium, consisting of a mixture (1:1 v/v) of DMEM/F12 (Hyclone) and neurobasal medium, supplemented with 1× N2, 1× B27, 10 ng mL⁻¹ glial cell-derived neurotrophic factor (GDNF, Peprotech), 10 ng mL⁻¹ brain-derived neurotrophic factor (BDNF, Peprotech), 1 μM cAMP, 200 nM ascorbic acid, and 1% Pen/Strep. For the first week, the plates containing all the spheroids (in Matrigel) were placed in a stationary position and starting from the second week and for the rest of cell culture, the plates were placed on an orbital shaker placed in the incubator.

Development of in vitro single and repetitive mTBI model

After two weeks of neural differentiation, the Matrigel containing the differentiating neurosphere was placed in the indentation position in the spheroid holder of the 3D printed mini weight-drop device. The holder was then inserted into the base. The impactor was then lifted up using forceps and allowed to freefall to administer an impact to the neurospheres at two different heights: 15 mm and 30 mm. All above steps were conducted in sterile conditions in the bio-safety hood. The neurospheres receiving impact from 15 mm height were designated as the half-impact group, and those receiving impact from 30 mm height were designated as the one-impact group (Figure 3A). After each impact, the Matrigel with the impacted neurosphere on the holder was washed with 1 mL of fresh neural differentiation medium so as to be detached from the holder and was transferred into wells of a 6-well plate (2 spheroids per well), and the medium was continuously changed every day for the first 6 days, with the conditioned media collected and stored at −80°C. After 6–day culture, the media was changed every other day until the end. A sham control was performed by transferring the Matrigel with neurosphere to the holder and then placed into fresh media without impact. A group of the neurospheres receiving impact from 30 mm height were subjected to the impact again from 30 mm height three days after the initial impact and were designated as two impacts group (Figure 3A). The neurospheres were cultured for 4 weeks after the first impact.

To evaluate the cumulative effects of multiple, repetitive concussions, a group of neurospheres (embedded in Matrigel) were repeatedly subjected to impact from 15 mm height with an inter-injury interval of 72 h (total of 10 hits) and were designated as the half height group and another group of neurospheres (in Matrigel) were frequently receiving impact from 30 mm height with a 72 h inter-injury interval (total of 10 hits) and were designated as the full height group (Figure 5A). A sham control was also performed, and the media in all groups were changed every other day.

Formation of a microglia and neurosphere co-culture repetitive mTBI model

A human microglia cell line (HMC-3) was purchased from ATCC and cultured and expanded in Eagle's Minimum Essential Medium (EMEM) with 10% fetal bovine serum (FBS, Gibco) and 1% Pen/Strep at 37 °C in a 5% CO₂ incubator.^[90] The first mTBI was administered to the neurosphere embedded in Matrigel from a height of 30 mm. Immediately after the impact, the Matrigel component was carefully removed by pipet tips. Then HMC-3 cells were harvested and re-suspended in cold Matrigel solution at a density of 2×10⁶ mL⁻¹. The Matrigel, including HMC-3 microglia cells, was then used to embed those neurospheres with and without the initial impact. At D3, the impacted neurosphere was subjected to a second impact from a height of 30 mm (Figure 6A). The sham control and those receiving two impacts were then cultured in the neural differentiation medium until the end of study, with the media being changed every other day.

Celltracker labeling of microglia and neurospheres

To investigate the spatial location of the microglia and neurospheres in the 3D co-culture model, microglia and neurospheres were labeled by the Celltracker red and green (Invitrogen), respectively. The Celltracker agents were prepared in PBS buffer as working solutions, according to the manufacturer’s protocol. Microglia and neurospheres were suspended in the corresponding working solutions and incubated at 37°C for half an hour. Then microglia and neurospheres were centrifuged down and washed with warm media thrice before use.

Lactate Dehydrogenase (LDH) measurement

The LDH levels in the conditioned media from control and impact samples (50 μL each) were measured by the CytoTox 96^® NonRadioactive Cytotoxicity Assay kit (Promega) kit following a previously reported method.^[100] Absorbance (OD) at 490 nm wavelength was measured by a microplate reader. Each group has three replicates.

Quantitative real-time PCR study

Total messenger RNA was isolated from the matrigel embedded neurospheres using QIA-Shredder and RNeasy mini-kits (QIAgen). Total RNA was transcribed into complementary DNA (cDNA) using an iScript cDNA synthesis kit (BioRad Laboratories). Real-time polymerase chain reaction (PCR) analysis was performed in a StepOnePlus™ Real-Time PCR System (Thermo Scientific) using SYBR Green Supermix (Bio-Rad). The cDNA samples were analyzed for the target genes and for the housekeeping gene 18S rRNA. The primer sequences were summarized in Supplementary Table 1. The relative expression level of each gene of interest was calculated using the comparative Ct (2 ^−ΔΔCt) method.^[101] Each group has three to four specimens.

Immunofluorescent (IF) staining

For IF staining, the control and impacted neurospheres were rinsed with PBS and then fixed in 4% paraformaldehyde aqueous solution. After washing with PBS, the samples were placed in 30% sucrose solution overnight at 4 °C. Then, neurospheres were transferred into tissue embedding molds, followed by the addition of OCT compound. Samples in OCT were rapidly frozen and kept in a −80 °C freezer before sectioning. Cryosectioning was performed at 10 μm thickness by a Zeiss cryostat at −20 °C. The slides were fixed in cold methanol for 10 min then washed twice with PBS for 20 min and then blocked with 5% goat serum (SouthernBiotech) containing 0.4% Triton X-100 (Sigma) for 1 h. The primary antibody diluted in 5% goat serum, containing 0.1% Triton X-100, was added to the slides. The information regarding all the antibodies used were summarized in supplementary Table 1. After overnight incubation with the primary antibody solution at 4°C, the slides were washed with PBS thrice and then incubated in fluorescent dye labeled secondary antibodies for 2 h. The slides were washed again and finally stained by DRAQ5 fluorescent probe (Cell Signaling) to stain the nucleus. All the fluorescent images in the study were acquired using a Zeiss 880 confocal microscope. The Caspase-3, Ki-67 and NeuN expression were quantified by their percentage amount, and GFAP was quantified by the percentage area,^[102] using ImageJ software (NIH).

Human cytokine array

The relative level of secreted cytokines and chemokines in the conditioned media was detected by a human cytokine array kit (ARY005B from R&D Systems), following the manufacturer’s protocol. Generally, about 0.9 mL of conditioned media from control and impact samples were respectively mixed with the array buffer and a cocktail of biotinylated detection antibodies. The mixture was then added to the cytokine array membrane and incubated at 4 °C overnight. After washing off the unbound material, streptavidin-HRP solution and chemiluminescent detection reagents were added to the membrane in sequence. The developed films with all the positive signals were scanned and transformed into images. The pixel density was analyzed with ImageJ software.

Statistical analysis

In all cases, the impact group’s marker/cytokine expression was normalized to that of the control group (control group set as 100%). At least 3 or 4 specimens were quantified in each group. All quantitative data are shown as the mean ± standard deviation (SD). Statistical analysis was done using the GraphPad Prism 7.0 Software. Significant differences between means of two groups were determined using two-tailed Student's t-tests while differences among means of three groups were determined using one-way analysis of variance (ANOVA) with post-hoc Tukey tests. A P-value < 0.05 was considered statistically significant.