Due to the complex nature of ischemic stroke, it is not possible to successfully model using a single in vitro or ex vivo system. However, use of these bench-side systems allows the investigation of biochemical and molecular mechanisms involved during stroke-like ischemic conditions and to examine the resultant cellular damage. These systems can be utilised to determine pathways and cellular triggers involved in both necrotic and apoptotic cell death, ¹² alongside the isolated investigation of cellular cascades that occur as a result of excitotoxicity. ¹³ Methods to induce stroke-like ischemic conditions in vitro may include chemical or enzymatic block of cellular metabolism. Metabolic inhibition is induced through application of chemicals such as 2-deoxyglucose, antimycin or sodium azide, ^{14- 16} which act to interact with the electron transport chain mimicking the energy depletion that occurs during cerebral stroke. Alternatively, a chemical-based model can use NMDA or glutamate receptor agonists to induce excitotoxic conditions via imitating the significant extracellular increase in glutamate that occurs during ischemia. ^{17, 18} Chemical or enzymatic bench-side models allow high-throughput testing, with ease of application and rapid responses, the method also allows isolated analysis of specific aspects of the molecular pathways involved. However, both approaches are reductionist in that they only attempt to mimic one aspect of the pathophysiological cascade and don't replicate the more complex interplay of mechanisms. A further disadvantage to the use of chemicals or enzymes to model ischemia is that these compounds may be difficult to wash out and therefore can disrupt the return to pre-insult conditions. The return to pre-insult conditions in vitro is undertaken to mimic the return of nutrients seen in vivo due to the return of blood flow to an area affected by stroke, known as reperfusion.

The most frequently used in vitro method to induce stroke-like ischemic conditions, is to remove all the available oxygen and glucose supply to the cells, known as oxygen-glucose deprivation (OGD). This is most often achieved by perfusing glucose-free media with a nitrogen/carbon dioxide mixture to displace oxygen, with subsequent experiments taking place inside a hypoxia chamber. Reperfusion conditions can be imitated through the reintroduction of glucose alongside a return to atmospheric oxygen. Induction of OGD leads to neuronal depolarisation within 10 minutes of onset, ¹⁹ with astrocytes showing immediate progressive depolarisation over the first 30 minutes of OGD. ²⁰ OGD with reperfusion shows continued neuronal degradation over several hours following a return to 'normal' culture conditions, which combined with large extracellular glutamate increases ²¹ are both consistent with in vivo observations. ²² Experiments undertaken in hypoxic-only conditions are less representative of ischemic stroke but may better represent cerebral hypoxia conditions such as carbon monoxide poisoning. ^{23, 24} Currently, many in vitro ischemia models mimic global ischemia as they induce an insult to the overall brain slice or culture preparation and therefore do not mimic the clinical situation of a focal insult. Recently, Richard et al. have demonstrated the development of focal OGD in ex vivo cortical brain slices using targeted OGD media stream perfusion, perfusing the tissue surrounding the target area with artificial cerebrospinal fluid (aCSF) solution. They reported rapid neuronal depolarisation within the core OGD targeted area with slower progressive depolarisation in the surrounding aCSF perfused area, such as is seen in the penumbra. ²⁵ The refinement of an ex vivo slice model of stroke to better represent the clinical presentation of cerebral ischemic events, may allow preclinical bench side investigations to occur in a more representative model.

To gain better relevance from bench side in vitro/ex vivo stroke models, the physiological micro-environment of the cells is critically important to gaining a true understanding of disease process and outcomes. The influence of oxygen, nutrients (including glucose), cell to cell contact and shear forces all need to be considered. With reference to the use of OGD as a key bench side ischemia-inducing model, oxygen levels should be considered when designing and interpreting data from these experiments. Physiological arterial blood oxygen concentrations differ significantly from external atmospheric levels. Maintenance at physiologically relevant oxygen levels has been shown to increase survival, cell proliferation and dopaminergic neuron differentiation in culture. ²⁶ Whereas, high oxygen levels affect not only basal functioning but also response to challenge, potentially providing resistance to stroke related oxidative stressors. ²⁷ In addition to high oxygen levels used within in vitro culture systems, glucose is also poorly matched to physiological concentrations. Cell culture glucose concentrations are often up to 8x higher than those reported within the brain. ^{12, 28, 29} Glucose is used in culture maintenance at concentrations proposed to be present within the brain during severe hyperglycaemia and shown to affect neuronal viability. ²⁸ A rethink of the use of long-standing outdated culture media maintenance and methods is required to improve physiological relevance.

One step towards improving physiological relevance of in vitro stroke models is the introduction and use of co-culture models which are based on multiple cell types, typically placed in repeated monolayers, allowing substrates and signalling molecules to pass between several cell types. This approach is common for modelling the blood brain barrier and has improved our understanding of neurovascular changes that occur during ischemia. Developments in 3D cell culture have the potential to improve the physiological relevance of bench side ischemia models even further. Cells cultured in 3D, on either scaffolds or a scaffold-free system, show natural cell shapes, prevalent cell to cell junctions, well differentiated cell types and a greater response to mechanical stimuli, overall improving physiological relevance compared to 2D culture models. ^{12, 30} Cells show lower mortality rates, resistance to nutrient deprivation and drug insults, this is alongside the presence of a gradient based availability of culture media nutrients more like that of a physiological state. Popularity and interest in the use of 3D cultures for in vitro modelling is increasing and they offer an additional way to investigate potential neuroprotective treatments prior to the need for in vivo approaches, reducing ethical impact. In addition to the use of bench side models of multiple cell types to increase physiological relevance, the use of microfluidic devices is also being explored to improve the relevance and applicability of bench side disease models, often termed organ-on-a-chip models. ³¹ Microfluidic devices, since the advent of lithography in 2001, ³² use a micro-engineered culture platform that can be utilised to mimic blood flow in health and disease models, alongside allowing the isolated investigation of upstream and downstream signalling within and between cells, due to microchannels that allow axonal growth along them. Since the early 2000s there have been multiple developments in the field of biomicrofluidics to try to overcome some of the limiting factors of this type of cell culture, as this is still constrained to the typical 2D rigid culture techniques discussed earlier. Improvements towards a brain-on-a-chip model are still required, but the use of microfluidics devices for axon-specific responses, cell-cell interaction and high-throughput screening are of great interest to the preclinical stroke research community, and with advancements in the field may prove to be another key step in the drug translation process that will help improve the positive potential of tested neuroprotective therapeutics within the clinic. Although in vitro research provides a platform to determine and understand cell-specific responses to stroke-like conditions the complexities of clinical stroke often require these methods to be combined with in vivo approaches.

The significant lack of neuroprotective therapeutics in the clinic has led to extensive work aimed at improving the reproducibility of preclinical stroke models and to ensure they reflect as closely as possible the clinical disease. ^{33- 36} To date, multiple preclinical stroke models have been developed to reproduce both global and focal ischemic stroke. Whereas global ischemia models more closely mimics the situation of cardiac arrest, focal ischemia models represent the typical clinical presentation of ischemic stroke. Various models are available for use in a variety of animal species, however the use of rodents, specifically rats and mice, is the most common. This is likely to be due to low acquisition and husbandry costs associated with these species, combined with simple effective monitoring methods and ease of tissue processing. ²⁴ To try and address the gap in clinical translation for stroke treatments the Stroke Therapy Academic Industry Roundtable (STAIR) published a number of recommendations including the need for potential therapeutics to be assessed across a number of species, initially rodents followed by studies using gyrencephalic species. ^{33, 37} The use of higher-order species is necessary to overcome the evolutionary differences between rodents and primates such as the corticospinal tract descending from the motor cortex. ³⁸ Larger species may also be suitable for investigations utilising endovascular techniques, such as clot retrieval and stenting, techniques used more frequently in the clinic. However, due to cost, availability and ethical considerations with larger higher-order species, rodents still play a key role in research to ensure experimental power in addition to testing the safety and efficacy of potential neuroprotective treatments.

The most frequently used in vivo model of focal ischemia is intraluminal suture middle cerebral artery occlusion (iMCAO), ³⁹ which has enhanced our knowledge of the pathophysiology of cerebral ischemia including penumbra development and functioning, blood brain barrier injury, cell death pathways and inflammatory processes related to cerebral ischemia. Developed initially for use in the rat, ⁴⁰ in which infarction success rates are 88-100%, ⁴¹ the method was later modified by Longa et al. ⁴² and subsequently adapted for use in mice ⁴³ where the model is increasingly being used particularly due to the availability of transgenic mouse strains. Induction of iMCAO is achieved by the insertion of a flexible monofilament into either the common carotid artery (CCA) ⁴⁰ or external carotid artery (ECA). ⁴² Access into the CCA requires the vessel to be ligated for incision. ⁴⁰ Similarly, for filament access, the ECA is typically transected in the modified iMCAO method. ⁴² Once in the vessel lumen, the filament is advanced into the internal carotid artery (ICA) and to the bifurcation of the middle cerebral artery (MCA). Once in situ, at the origin of the MCA, the filament head impedes blood flow into the MCA territory and is left in situ for the duration of ischemia. The utility of this model allows for accurate control of occlusion duration, allowing for either permanent ischemia where the filament remains in place, or transient ischemia where the filament is removed to allow reperfusion of blood flow to the MCA territory. Utilising the iMCAO model, post-stroke effects can be examined months following the initial insult. Vessel access is obtained without craniectomy and therefore avoids cranial damage associated with craniectomy which could impact on intracranial pressure changes and post-stroke outcomes such as reduced lesion volume. ^{44, 45} Use of iMCAO results in large infarct volumes encompassing both the striatum and cortex, ^{46, 47} however, longer durations of iMCAO can lead to hypothalamic damage which occurs rarely in humans and can lead to hyperthermic responses in rats and poor temperature control in mice. ^{48, 49} Following iMCAO, rodents can experience a range of side-effects that negatively impact the welfare of animals, including but not limited to, significant weight loss, abnormal or reduced motility, difficulty eating/drinking and mortality. These outcomes can be moderated through enhanced pre and post-surgery care; detailed recommendations to support this have been highlighted by the IMPROVE guidelines. ⁵⁰

The filament selection for use within iMCAO also plays a key role within the model, as unsuitable filament selection can lead to inadequate occlusion and filament-induced secondary subarachnoid haemorrhage due to arterial rupture. ⁵¹ This has led to an increase in the use of standardised silicone-coated filaments to attempt to improve reproducibility and the use of laser doppler flowmetry to confirm correct filament placement and monitor occlusion duration. ⁵² Although the iMCAO model is not appropriate to study the effect of thrombolysis treatment in conjunction with tested therapies it does recapitulate occlusion of the MCA in the clinic, which is the most common location of thromboembolic stroke in humans. ⁵³ However, placement of the intraluminal filament into the origin of the MCA is an all or nothing approach, which does not reflect the clinic where human stroke is frequently not caused by a complete occlusion, additionally the model does not replicate the event of partial spontaneous reperfusion that can occur in patients within 48 hours of stroke onset. ^{54, 55} Furthermore, the model does not profile the slow clot disruption that occurs following rtPA administration, instead showing surge reperfusion upon filament removal. ⁵⁶ More recently however, the model has become increasingly relevant due to the advent of interventional mechanical thrombectomy in the clinic. In 2015, five randomised controlled clinical trials reported beneficial effects of endovascular intervention therapy in treating patients with large vessel occlusions, with or without rtPA treatment. ⁵⁷ This beneficial effect was correlated with the abrupt recanalisation of the vessel and rapid reperfusion of the ischemic zone, corresponding to the mechanisms of the iMCAO model. The positive outcome of endovascular thrombectomy, as reported in the clinical trials, suggests this mechanical treatment will become the primary therapy for large vessel occlusion in the clinic; ⁵⁶ renewing relevance of the established iMCAO model as a model of endovascular thrombectomy.

In addition to iMCAO, focal ischemia can also be induced in vivo by direct occlusion of the vessel of interest, using a cranial window to either clamp, ligate or cauterise the vessel in situ. The most used cranial entry focal ischemia method was developed in 1981 by Tamura et al., using distal MCA occlusion, similar to the occlusion location obtained using iMCAO, inducing combined cortical and striatal lesions. ⁵⁸ Like the iMCAO method, craniotomy models can be used to induce either permanent or transient focal ischemia - although relying heavily on the experimenter's ability to reduce potential localised cerebral damage for both scenarios. Similarly, the return to perfusion following transient ischemia using this method results in a sudden prompt mechanical perfusion, again unlike the presentation of reperfusion in the clinic. Furthermore, access to the vessel of interest through the skull has been shown to induce cortical spreading depressions and inflammatory responses. ⁵⁹

Additional stroke models are available that utilise specialised mechanisms to induce ischemia, typically in rodents, including thromboembolic, endothelin-induced and photochemical models. Embolic models can be broadly categorised into thromboembolic and non-clot embolic models; the former utilising the induction of localised clots or the insertion of spontaneous/thrombin-induced clots and the latter utilising non-clot methods such as micro/macrospheres to occlude vessels. ^{52, 60, 61} Since most human strokes are caused by thromboembolism, thromboembolic stroke models may more closely represent the clinical disease pathology. However, there is a lack of a single standardized embolic stroke model leading to multiple occlusion induction methods, a further method that has shown increasing use is the localised injection of thrombin directly into a craniotomy exposed MCA bifurcation. ⁶² A significant advantage of this thromboembolic technique is the opportunity to test thrombolysis treatments alone or in combination with neuroprotective agents. ^{63, 64} Additionally, the use of rtPA to lyse the clot in thromboembolic models results in a reperfusion profile that is closer to the reperfusion profiles of rtPA treated patients, contrasting the sharp reperfusion profile following filament removal using iMCAO, increasing the relevance of clot emboli models. ^{56, 65} However, infarct location and volume can be highly variable using thromboembolic stroke models along with low seven-day survival rates, ⁶⁶ which impacts the reproducibility of thromboembolic stroke models and their utility for longitudinal studies. In addition to the difficulties experienced in controlling lesion location, the duration of the ischemic insult is difficult to control in these models with some animals showing spontaneous recanalisation, ^{62, 67, 68} a further disadvantage to the model is the potential for spontaneous clot formation following embolism disruption. ⁶⁹ The severity of this preclinical stroke model also has a high mortality rate of >30%, typically within the first 24 hours preventing longitudinal study. ^{50, 70} Non-clot embolic methods can include the injection of silicone or collagen into the ICA of rodents ^{71, 72} or, for example, the injection of micro/macrospheres resulting in microembolisations causing multifocal and heterogenous lesions. ^{73, 74} Administration of micro/microspheres results in an immediate reduction in cerebral blood flow (CBF) that becomes progressively stronger over the first 3-12 hours following sphere injection, this in turn leads to a slow infarct development, suggesting a longer therapeutic time window than other stroke models for instance iMCAO. ⁷⁴

Endothelin models of ischemia employ the peptide Endothelin-1, a long-lasting potent vasoconstrictor, to induce vessel occlusions. ^{75, 76} The peptide is typically applied using stereotaxic injection methods ^{77, 78} or craniotomy ^{79, 80} applied directly onto an exposed cerebral vessel causing a constriction of the vessel reducing CBF to the vessel territory for up to 22 hours post-injection ⁸¹ followed by gradual reperfusion. The severity and duration of an insult can be adjusted according to the concentration of the peptide at application. The sustained reduction in CBF and subsequent gradual reperfusion profile, alongside gradual lesion development resembles the evolution of clinical stroke. The topical application method of the peptide is a source of variability in this model, due to the difficulty in ensuring consistent diffusion. To reduce this variability intracortical injection of the peptide to sensorimotor areas has been developed. ^{82, 83} With this injection based method, care must be taken to avoid the compound entering the ventricles to avoid significant negative welfare outcomes such as barrel rolling or seizures. ⁶⁹ Another targeted approach of MCA occlusion induction is the photothrombotic stroke model, which induces localised permanent infarcts of the cortex by introducing a photosensitive dye (e.g. Rose Bengal) into the cardiovascular system and illuminating this through the skull using a specific wavelength of light. Targeting small vessels, this illumination activates the photosensitive dye resulting in endothelial damage due to oxygen species formation, causing platelet activation and aggregation. ^{84, 85} These together result in the formation of an occlusion and consequently rapid ischemia to the vessel territory, leading to the development of cortical lesion. ⁸⁶ This method produces reproducible and localised cortical lesions, with minimal variation in lesion volume, ⁸⁶ the small cortical lesions and low mortality associated with the model lend well to longitudinal study however, a lack of penumbral tissue within the lesion reduces the translational impact of the model, as the penumbra is the target tissue of interest for neuroprotective strategies. Photothrombotic induction requires the use of light sources, these can act as a heat source. The illumination can have a heating effect on the skull and subsequently the brain, therefore it is important that care over temperature must be taken and where possible cold light sources used, with exposure and distance from the skull/brain considered. ^{50, 82} Furthermore, due to the systemic nature of the dye the model is unsuitable for preclinical drug studies.