Over 50 years of research indicates that three characteristics – direction, duration, and type of head motion – are the primary determinants for how the brain moves and deforms within the skull during impact or non-impact (e.g., whiplash) conditions. The most prominent effect of acceleration direction is that it strongly influences brain regions affected during an injury and, thus, the resulting neurological sequelae. The timing or duration of acceleration also contributes to the points within the brain that are maximally affected during impact; generally, cortical regions are affected the most for extremely brief head motions, while the deeper brain structures are influenced more following longer duration accelerations. Third, and perhaps most importantly, the type of acceleration – rotational and/or translational – clearly contributes to the mechanics of loading within the intracerebral tissues.

The past two decades has shown a tremendous growth in the use of computational models to further dissect the role of these three primary loading characteristics. The development of these models, initially developed well after experimental data was collected to describe the in situ biomechanics of loading, has now quickly outpaced the rate of experimental data collection. With increasing frequency, the first exploration of new types of impact/injury loading conditions (e.g., traumatic blast loading) is done virtually with computational techniques (recent studies: [5–14]; reviews: [15–17]). These computational approaches are extremely critical because they provide a mechanical road map for the rational design of in vitro models of CNS trauma that, in turn, will provide critical biological response information at the millimeter and micron scale. In the next section, we review the salient features of the traumatic mechanical loading that need to be reproduced in any in vitro representation of CNS trauma.

The primary starting point for any in vitro model of traumatic brain injury is to establish some fidelity to the ‘real world’ scenario of injures that can span the mild (e.g., concussion) to the more complex and severe (depressed skull fracture and intracranial hematoma). Establishing this model fidelity may be more difficult than it first appears because nearly every clinical head injury occurs in a biomechanically unique situation. As a result, these in vitro models often reduce the complex input into a series of well-prescribed, controllable, and repeatable inputs at the cellular and tissue scale. A desirable trait of these models is to bring the complexity and variability of biology under the more prescribed conditions of an engineer: namely, modulating both stress and strain input independently and monitoring the resulting cellular, molecular and physiological changes to the circuit. Three main factors drive the design of nearly every model system interrogating the biomechanical input-output relationships for cellular and organotypic tissue systems.

The first decision faced by an investigator is the preparation used to interrogate this functional input-output relationship. An ideal and common platform is dissociated cell monolayers, since they offer the opportunity to directly constrain the input (mechanical) and immediately observe the output (biological). Cell lines derived from oncogenic cell populations were the first types of dissociated cells used over two decades ago, principally because there were no available techniques to isolate primary CNS cells from adult or embryonic animals. Beginning in the mid-1990s, though, the commercial availability of feeding media supplements made the isolation of a large number of diverse primary cell types possible, including primary neurons, microglia, astrocytes, and oligodendrocytes. These technologies have developed enough recently so that one may even prepare living slices of brain tissue (absent vascular perfusion), thereby preserving the architecture of the brain at the time of biomechanical testing. These preparations – termed organotypic cultures – represent the most fidelity to the living, in vivo environment. However, even these preparations lack an intact vasculature. Recent work showing how photopolymer and natural protein networks can provide an intact vasculature provides hope that even this technical hurdle of vascularized tissue in vitro is not too far away from solution. To date, though, no investigator has provided an in vitro model that contains an active vascular system to mimic the potential interaction of vascular perfusion on outcome of tissue injured in vitro.

The second major decision facing an investigator is defining the mechanical input. A very pragmatic part of selecting the mechanical input is driven by the output that is measured. For example, one may only want to measure the survival or degeneration of neurons, astrocytes and other cell types in the network for hours to days following the mechanical injury. If this is the only objective, the mechanical input device can be highly customized and the optical path for viewing the cell preparations during and immediately following the injury is not critical. Many questions, though, are critically reliant on how the cellular/tissue system responds immediately following the initial mechanical stimulation; examples include fluorescence-based ion indicators to indirectly assess ion flux immediately after injury, release of neurotransmitters, or changes in the membrane potential [18–21]. For these measures, testing systems often mount directly to a microscope [22–26]. Microscope mounting frequently means that the mechanical testing system is relatively simple, reliable and therefore may have less fidelity to the complex loading that occurs within the in vivo brain. However, the advantages of drawing relationships between these acute events and longer term assays of cellular and organelle viability make these simple models the most widely used. To date, the most versatile of the in vitro models provide the flexibility to challenge the in vitro preparations with more ‘real world’ mechanical inputs [27]. However, the practical constraints of exploring this parametric space – there are literally millions of combinations that one could explore – limit the utility of these more flexible architectures.

One often overlooked aspect is that the in vivo brain does not experience a uniform mechanical deformation during any traumatic brain injury. Several past experimental and computational studies show that, for the in vivo loading scenario, some brain regions are deformed far more than neighboring regions. Therefore, an additional design element in any in vitro model of traumatic brain injury is to select how much of the cellular population will be injured, and the mechanical uniformity of this injury. Many studies apply the injury to the entire cellular populations; a far smaller number apply the injury only to a subregion of the culture (see reviews: [1–3]). These in vitro preparations, since they contain several different cell types, are useful surrogates of the integrated response approximating small regions in the brain. Recently, the ability to mechanically stimulate single individual cells has appeared, opening up new opportunities to explore how single cells can influence the behavior of small neuronal ensembles. Therefore, a closer simulation of the in vivo environment can be targeted by studying how spatially defined patterns of mechanical stimulation can affect the broad behavior of cellular networks.

The high water content of brain tissue, combined with its relatively disorganized extracellular matrix, means that brain tissue is one of the softest tissues on the entire human body. Characterized by a high bulk modulus and low shear modulus, brain is often considered weakest under shear loading. Although impact and the subsequent acceleration of the head can produce transient pressures within the cranial vault, several studies have shown that the effects of these pressures that occur during impact are relatively modest. As such, most of the critical inputs that are re-created in in vitro models are linked to the primary shear deformation that occurs throughout the brain at the moment of injury. A common confusion appears, though, when designing how shear deformation can be represented with in vitro systems. This confusion often centers on how macroscopic shear exactly translates to deformations of cellular structures within the complex CNS tissue.

Examples of the tissue-cellular transformation in representative brain tissue samples can help illustrate how easily these concepts can be confused. In a simple shear cube deformation, a dendrite or axon that is oriented along the diagonal of this cube can either compress or extend, based on the initial orientation of the dendrite/axon. In comparison, if the dendrite/axons align along one edge of the cube, or is oriented in the plane of the cube, it would experience no deformation during simple shear.

The correct prediction of how bulk or macroscopic mechanical loading causes deformation of cellular structures – illustrated in the simple example of the previous paragraph – is an enormously significant problem in CNS mechanics. It is a problem that is highly influenced by the boundary conditions – i.e., how does the cellular, vascular, and extracellular structure couple together? It is also influenced greatly by the initial conditions – e.g., does the axon or dendrite traverse a normal, straight course through the tissue or does it have a meandering, tortuous path through the tissue? Indeed, a key knowledge gap now being filled is building structurally accurate models of CNS tissue that can be used to understand how an applied mechanical loading at the macroscopic scale – i.e., a sample of excised tissue, or a small volume of brain tissue in vivo- results in the deformation of cellular and subcellular structures [28–32]. Even simple models of composite material behavior show that cellular/subcellular deformation is very complex, depends critically on the mechanical coupling among components, and may vary substantially across different brain regions subject to the same macroscopic loading because of differences in tissue microstructure. Experimental platforms are also showing how subcellular structures are deformed uniquely in response to the applied mechanical loading [33].

With a continuing focus on the role that deformation plays in the subsequent injury sequelae to the brain, several technologies have emerged to study this phenomenon in vitro. In a crossover of technology from the vascular hemodynamics field, fluid flowing quickly past the model layer of cells will induce a deformation sufficiently high enough to activate glutamate receptors and, at higher levels, cause an immediate increase in plasma membrane permeability and changes in network dynamics [34, 35]. Alternatively, cells in two-dimensional culture can be moved rapidly and then stopped very suddenly [36]. In this model, the different densities of subcellular organelles and nuclear components will cause in intracellular deformation. In both of these techniques, it is difficult to establish the in vivo biomechanical fidelity of these models. Rather, the techniques are useful to study input/output responses not possible with other methods.

By far the most common method to study input-output relationships for cellular populations is stretching flexible substrates, where the cells of interest are adhered directly to the substrate [22–26]. Many studies evaluate the effects of deforming the cell population in two perpendicular directions simultaneously (termed ‘biaxial’), while a less frequently used method deforms the substrate primarily along one axis (‘uniaxial’). The uniaxial configuration is most readily comparable to shear deformation in vivo, as the principal directions of stretch are identical to a pure shear deformation. The biaxial method is more complex, because it involves a mixture of planar expansion and pure shear. Some differences appear when comparing these techniques directly: the uniaxial model of deformation requires higher peak deformations to cause immediate alterations in plasmalemma permeability, while biaxial deformation offers a symmetric deformation field that can apply a more homogeneous input to dissociated or organotypic culture networks.

Applying deformations to cells in vitro is a relatively straightforward technique, and has been the study of several past laboratories. However, the application of a defined stress, both the direction and magnitude of the stress, is a much more difficult technical challenge. One must first define the stress level that is appropriate. Next, one should consider how to focally apply these controlled stresses without affecting neighboring cells. Third, the material properties of the cellular and subcellular constituents may mean the loading regime will differ uniquely among individual cells, as well as across regions within individual cells. It is perhaps for this reason that models to accurately apply stresses at the micron and submicron scale are lacking still in the literature.

We define mechanosensors as the critical receptors/channels/organelles on neurons and glia that respond directly to applied mechanical stimulation. Like many sensors, the biological entities contain a ‘dynamic range’ – they often have a critical threshold for activation, and this threshold often changes when the force is applied dynamically, rather than statically. Much like how a typical instrumentation transducer will show a dynamic operating range over a range of frequencies, these mechanosensors are often equipped with a molecular architecture that makes them tuned for a particular part of the mechanical force spectrum. Models for the mechanically-influenced binding of ligands to receptors – a ‘mechanokinetics’ of receptor activation – have been proposed for some ligand-receptor interactions [37], but none have been applied to channels and receptors expressed in neurons and glia.

Analogous to developing measures of how an agonist binds to a receptor over a range of ligand concentrations, we expect mechanical sensors will possess a minimum force threshold (concentration) for their activation, and a saturating force (saturating agonist) for their maximal activation levels. Likewise, mechanosensors may respond differently to competing mechanical inputs-for example, some receptors may be sensitive only to the pressure applied, while other receptors will be sensitive to a certain type of deformation (i.e. tension), and still yet others may change their sensitivity to deformation based on the applied input pressure.

Despite the potential importance of mechanosensors, we know very little about the mechanoactivation profiles for most receptors expressed in the brain. Evidence shows that the NMDA receptor can respond or ‘sense’ mechanical forces that are applied during traumatic injury [38–40]. One important physiological consequence of this mechanical loading is the loss of magnesium block in the NMDA receptor, a feature that normally limits the ion flux across the receptor at normal resting membrane potentials [39]. Moreover, evidence shows that the NMDAR agonist affinity is altered after mechanical stretch. Linking the NMDAR to the cytoskeleton appears key for this mechanosensing property, and a single serine residue of the NR2B subunit of the receptor appears to substantially influence its mechanical force sensing ability [41]. With the role of the NR2B subunit of the NMDAR for neuronal death, the unique mechanosensing ability of the NR2B subunit appears uniquely positioned to control the survivability of neurons in the circuit after mechanical injury [42].

The α-subunit of the voltage gated sodium channel is a second mechanosensor [43–45]. Following activation of the sensor, changes in calcium homeostasis can appear and lead to the proteolytic cleavage of this channel, setting up a scenario where the activation of a channel may lead to its direct modulation. One direct consequence of altering sodium channel structure after injury is an effect on the network dynamics, suppressing the activity of the network until compensatory changes in the network can help restore a balance in activity. Intriguingly, this breakdown of the channel by proteases is influenced by a part of the cell cytoskeleton (microtubule), suggesting that the mechanical structure of the neuron confers some level of protection against degenerative changes that can be caused by mechanical trauma. This rapid change (<15 min) in channel integrity is contrasted with a later stage degradation phase (4–6 h) that precedes any evidence of neuronal death [46]. However, it is not yet clear if this later phase of degradation can be blocked to rescue neurons in the network, or if this later degradation is part of a larger irreversible sequence of events that will lead to the removal of the neuron from the injured network.

Other receptors are known to either activate rapidly and/or change their properties after mechanical injury, but there is no conclusive evidence yet to term these as primary mechanosensors. For example, AMPA receptors lose one important physiological property early after mechanical loading; they express a loss in the desensitization of the receptor [47]. The reason(s) for this loss in desensitization of the receptor are not completely known, but it may owe more to the upstream mechanoactivation of the NMDAR than the molecular structure of the receptor and any postsynaptic tethering to the cytoskeleton [48]. In this light, AMPA receptor modulation is a secondary effector for the primary mechanosensor (NMDAR). In addition, changes in GABA receptor currents appear after injury, but they also depend directly on NMDAR activation [49]. Similarly, focal mechanical stimulation in glial cells can elicit broad intercellular waves that are mediated by release of ATP and glutamate [50]. Although the mechanism(s) responsible for intercellular wave transmission is well known, specific mechanoactivated receptors in astrocytes are not well described for this phenomenon. Thus, although glial cells may have primary mechanosensors, this is inevitably complicated with the stimulated release of agonists, that can in turn activate other receptors.

As many candidate mechanosensors play important roles in cellular and neural circuit physiology, the activation of these mechanosensors can have broad impact on function. However, only scant evidence appears for mechanosensing properties of other glutamate receptors and channels in neurons. Future knowledge of sensors that are associated with specific neuronal subtypes is also needed, as well as knowing the differential role(s) for mechanosensors in adjusting the balance between inhibitory and excitatory circuitry in the brain after trauma. To this end, glial cell types may also provide a critical modulatory role that can influence the rebuilding of network function for days to weeks following initial injury.

Across length scales – from individual synapses, neurons, microcircuits and cell clusters to whole brain regions – mechanosensors very likely play a profound role in the recovery of circuits after injury. The most straightforward effect on function is when neurons and glia within an intact circuit degenerate, thereby degrading the information capacity of the circuit. Although the mechanical tolerance for neuronal survival exists for several testing systems, and these estimates span both dissociated and organotypic slice culture preparations [51–53], no published work extends how the function of the system is altered with this degeneration. Relatedly, new evidence shows the spontaneous electrical activity of the network can be impaired with mechanical stimulation [35, 54], yet there is no consensus yet on how these thresholds for activity impairments correspond to the thresholds for neuronal death. Several changes in the neural circuit state can now be traced back to mechanoactivation of the NMDA receptor, including a slow, neuronal depolarization that can lead to altered network excitability, the loss of AMPAR desensitization that can also change excitability, and the insertion of new glutamate receptor subtypes that appear over several hours (see [3]). At this time, though, little data exists on how the function of individual neurons within the network contributes to the overall network dynamics of the circuit.