Two randomized, double-masked, placebo-controlled phase-3 trials investigated the effect of ruboxistaurin, a protein kinase C beta inhibitor, on slowing the development of diabetic macular edema, a complication of diabetic retinopathy, and its attendant visual loss (18). In the combined studies (N = 1,028), sustained moderate visual loss occurred in 4.4% of placebo- vs. 2.3% of ruboxistaurin-treated patients (P = 0.069) (18). The magnitude of effect of ruboxistaurin on vision loss was consistent with what was observed in two prior studies (approximately 50% reduction beyond standard care). However, event rates were low and statistical significance was not achieved (18). A challenge post-hoc is differentiating between anatomic success, e.g., prevention of edema, and functional success, e.g., slowing vision loss.

An open-labeled, non-randomized prospective pilot study investigated the use of topical brimonidine purite, an α-2 adrenoreceptor agonist, as a prophylaxis for second eye involvement in LHON, a maternally inherited, blinding, bilateral optic neuropathy (19). Treatment with brimonidine (0.15%) 4 times daily in the unaffected eye for up to 2 years was unsuccessful in preventing second eye involvement in recently monocularly-symptomatic LHON (19).

RESCUE was a multicenter, randomized, double-masked, sham-controlled, phase-3 clinical trial investigating the efficacy of a single intravitreal injection of rAAV2/2-ND4 (a gene therapy vector enabling allotopic expression and delivery of the wild-type ND4 protein to mitochondria within RGCs) in subjects with visual loss from LHON (20). At 96 weeks after unilateral injection of rAAV2/2-ND4, LHON subjects carrying the m.11778G > A mitochondrial DNA mutation treated within 6 months after vision loss achieved comparable visual outcomes in the injected and un-injected eyes (20). The primary endpoint of a −0.3 logMAR difference between rAAV2/2-ND4–treated and sham-treated eyes was not met (20). These trials were well-designed and enrolled appropriate patient participants, suggesting that the lack of sufficient efficacy was unrelated to structural trial issues.

Two randomized, double-masked, placebo-controlled, parallel-group, multicenter, 48-month studies were identically designed, initiated 1 year apart, and completed in 2006 (21). Their purpose was to evaluate the efficacy and safety of oral memantine, an N-methyl-D-aspartate receptor-channel inhibitor, as a potential neuroprotective agent in open-angle glaucoma (OAG) at risk for progression (21). These phase 3 trials, which lasted more than 5 years, did not reveal a significant benefit for memantine treatment in preventing the progression of visual field loss in patients with glaucoma (22). Overall, progression of glaucomatous findings assessed by standard automated perimetry (SAP), frequency doubling technology, or stereoscopic optic disc photographs did not reveal a consistent protective effect of memantine, and the primary endpoint was not met in either study or in a pooled analysis (21). Failures such as these have since raised awareness around issues of trial design, patient inclusion, and strength and consistency of preclinical data (23).

Clinical trials investigating the neuroprotective role of brimonidine have failed to translate the efficacy observed in animal models into similar efficacy in humans (24). A 3-month, double-masked, placebo-controlled, randomized European multicenter trial investigated the efficacy and tolerability of brimonidine tartrate (0.2%) for the treatment of NAION (25). A statistically significant benefit for visual acuity in patients receiving brimonidine tartrate was not observed (25). While better visual field results were observed in the brimonidine group, these were non-significant trends, both statistically and biologically (25).

A randomized, double-masked, multicenter clinical trial compared the effects of alpha2-adrenergic agonist brimonidine tartrate (0.2%) with those of the non-specific beta-adrenergic antagonist timolol maleate (0.5%) in preserving visual function in patients with low-pressure (“normotensive”) glaucoma (26). Statistically fewer brimonidine-treated patients (n = 9, 9.1%) had visual field progression than timolol-treated patients (n = 31, 39.2%, log-rank 12.4, P = 0.001) (26). However, more brimonidine-treated (n = 28, 28.3%) than timolol-treated (n = 9, 11.4%) patients discontinued study participation because of drug-related adverse events (P = 0.008) (26). While baseline characteristics were similar between dropout patients in both treatment groups, failure to obtain information from this subset of patients limited interpretation of the results (26). Thus, despite demonstrating some effect with brimonidine treatment, these and other caveats led to it not being widely accepted or translated into clinical practice (27).

DIAN-TU-001 was a phase-2/3 study that compared investigational therapies (gantenerumab and solanezumab; human immunoglobulin-1 antibodies directed at sequestering amyloid-β fibrils) with placebo to determine if either treatment could slow the rate of cognitive decline and improve disease-related biomarkers in patients with a genetic mutation for inherited AD (30). The study did not meet its primary endpoint in patients with an early-onset, inherited form of AD and did not show a significant slowing of the rate of cognitive decline in patients treated with gantenerumab compared with placebo (30).

A double-blind, randomized, multicenter, clinical trial with 24 months of treatment and follow-up investigated the efficacy of minocycline (a tetracycline antibiotic) treatment in modifying cognitive and functional decline in patients with mild AD (34). Minocycline treatment did not significantly delay progression of functional and cognitive impairment compared with placebo (34).

AMARANTH and DAYBREAK-ALZ were randomized, placebo-controlled, phase-2/3 and phase-3 trials investigating the potential of lanabecestat (an oral β-secretase cleaving enzyme inhibitor) to slow the progression of AD compared with placebo in patients with early AD (mild cognitive impairment) and mild AD dementia (35). Treatment with lanabecestat was well tolerated but did not slow cognitive or functional decline (35). Together these trials illuminated challenges in translation, including trial design, biomarker, and preclinical data translatability.

A phase-2b, multi-arm, parallel group, double-blind, randomized placebo-controlled trial conducted at 13 clinical neuroscience sites in the UK investigated the efficacy of 3 neuroprotective agents [amiloride 5 mg (a diuretic), fluoxetine 20 mg (a serotonin-uptake inhibitor), riluzole 50 mg (a tetrodotoxin-sensitive Na⁺-channel blocker with anti-anxiety and antidepressant activity)] in patients aged 25–65 years with secondary progressive multiple sclerosis (36). No difference was observed between any active treatment and placebo, and the authors suggested that the absence of evidence for neuroprotection in this adequately powered trial indicated that exclusively targeting these aspects of axonal pathobiology in patients with secondary progressive multiple sclerosis was insufficient to mitigate neuroaxonal loss (36). Whether targeting multiple pathways with a combination therapy may work more effectively remains to be tested.

A multicenter, randomized, parallel-group, double-blind, placebo-controlled trial investigated the effect of isradipine, a dihydropyridine calcium-channel blocker, on the rate of clinical progression of PD (37). Long-term treatment with immediate-release isradipine did not slow the clinical progression of early-stage PD in previously untreated patients (37).

EMPOWER was a randomized, double-blind, placebo-controlled phase-3 trial that investigated the efficacy and safety of dexpramipexole (R-enantiomer of a dopamine receptor agonist) in patients with familial or sporadic disease (with ALS onset 24 months or less before baseline) (38). Although dexpramipexole was generally well tolerated, its effects did not differ from placebo on any prespecified efficacy endpoint measurement (38).

A double-blind, randomized, placebo-controlled, multicenter trial of a duration of 18 months investigated the efficacy and safety of olesoxime (a cholesterol derivative used for the preservation of mitochondrial function) in patients with ALS treated with riluzole (39). Survival was not significantly different between treatment arms (P = 0.71); estimated overall survival was 67.5% (95% CI: 61.0–73.1%) in the placebo group and 69.4% (95% CI: 63.0–74.9%) in the olesoxime group (39).

The diagnosis of neurodegenerative diseases is often delayed and is generally reported years or even decades after commencement of neuronal destruction. By the time patients enter a clinical trial, their disease may be too advanced for a therapy to be effective (45). Neuroprotective therapies should be initially administered at the earliest stage of disease when they will be most effective in delaying progression (1, 13). However, patients are not typically seen in the clinic until mid- to late-stages of disease, when the chances of significantly halting neurodegeneration are minimal (1, 13).

A major challenge in translating neuroprotective strategies to the clinic is the need for sensitive tools, e.g., for retinal screening and disease monitoring (17). The quality of research findings is predominantly contingent on the quality of the input data and the methods for their processing and interpretation (40). To determine whether an agent is neuroprotective, it is imperative to have reliable methods to detect its efficacy (46). The improper use of statistical analysis methods and the misinterpretation or misuse of p-values can lead to inaccurate conclusions, thus adding to reproducibility challenges (40).

While mouse models have some value in helping researchers understand disease pathobiology and the mechanism of action of therapeutic candidates, they have mostly failed as a translational tool, and their predictive utility has been suboptimal (40). Even with a useful animal model, failures can occur due to an incomplete understanding of disease pathophysiology. Moreover, no available models mimic the full spectrum of neurodegenerative pathology (6–8, 46). There is no perfect model for any of the neurodegenerative diseases, as each one fails to capture the full spectrum of pathological components involved (6, 7, 11).

The methods used along the trajectory from preclinical to clinical research are associated with progressively larger amounts of variability (47). Differences in scale, timing, and choice of model can make it difficult to confidently extrapolate findings from animals to humans (47). Human research participants not only differ genetically and epigenetically, but also display heterogeneity in terms of how long they take to become symptomatic and present for treatment, how sensitive or susceptible to damage their neuroaxonal components are to the burgeoning insults they have to endure during disease initiation and progression, how well they comply with therapy, the degree of placebo effect, and other factors (6–8, 47).

Promising therapeutic strategies in recent years have failed to translate into effective treatments, and while studies may be well-intentioned, there are subtleties that may be missed when performing experiments or interpreting which can be consequential (40, 48). Methodological flaws and poor experimental design in preclinical studies can lead to systematic bias, thus leading to irreproducible and unreliable data as well as inaccurate conclusions (40). The culture of preclinical research is markedly different from that of clinical research (e.g., differences in the training of the participants, how hypotheses are generated and tested, and the emphasis on repeatability). These differences between cultures are reflected in how experiments are designed and results interpreted (47, 49–51). The lack of target relevance or disease mechanism can lead to high heterogeneity within a patient population (40). In addition, the lack of publications of many negative clinical trials can lead to inadvertent repetition of the same study rationale, thereby wasting valuable time and resources (45).

Diseases are multifactorial, involve multiple insults, and occur simultaneously in RGCs and/or axons. However, as noted above, experiments are typically performed using a model involving a single insult (6–8). In the case of neurological diseases such as ALS, where there are challenges related to complex genetic associations, patients with variable clinical subtypes are often pooled in the same study group, which can lead to significant statistical differences in the observations (45). Furthermore, neurodegenerative diseases progress very slowly, over many months and years (1, 2). Thus, neuroprotective studies must follow subjects for a sufficient length of time, so that measurable differences in outcomes can be ascertained between treatment groups. For most neurodegenerative diseases, without selecting for a “fast progressor” subset, this may require more than 5 years, which is longer than typical grant/funding cycles (41). Producing longitudinal data in diseases with extended lifetime risks requires decades, but funding agencies or pharmaceutical companies rarely operate on such timelines.

There are numerous specialized neuronal cell types and subtypes that give rise to a complex connectome and thereby a complex cognitive, sensory, or motor function. Due to the heterogeneity and complexity of mammalian neural cells, neuronal and glial classification has been challenging, and many subtypes have not yet been characterized (52). Despite superficial similarities between animal and human visual systems, there is divergence in terms of anatomy, physiology, genetics, and disease manifestation (53). As with neurons in the brain, RGCs are also highly diverse, and there is a knowledge gap in understanding which cell types are most relevant for preservation of visual capacity and whether different cell types display distinct susceptibilities to disease or trauma (52). For example, effective neuroprotection of RGCs will potentially require targeting of multiple dysfunctions contributing to neurodegeneration (54). For RGC replacement, the challenge has been converting stem cells to RGCs and establishing both structural and functional connections to the optic nerve and brain thalamic nuclei and ultimately the visual cortex (3, 53, 55).

In terms of glaucoma clinical trials, there is no example of a primary outcome measure for neuroprotection which has led to FDA approval. All approved drugs were based on IOP-lowering efficacy (54). However, this outcome is irrelevant for neuroprotection, for which an objective measure of visual loss, either functional and/or structural, is needed (54). The most studied structural biomarker for glaucoma is optical coherence tomography (OCT) of the retinal nerve fiber layer (RNFL). Despite many good prospective studies showing it correlates with visual fields, OCT of the RNFL is currently not recognized by the FDA as a primary outcome measure for glaucoma. Even with a good biomarker that is clearly related to the disease and whose variability is well-characterized, these surrogate biomarkers have not reached the requisite threshold. Thus, any other biomarker proposed would need to surpass that for a pivotal outcome (56). Furthermore, a therapy may have a biologically relevant impact in protecting a particular type of RGC, but may not have a clinical impact based on current FDA-approved endpoints.

There are many FDA-acceptable endpoints that are relevant in glaucoma (i.e., contrast sensitivity, patient-reported outcomes). However, there is insufficient data relating to some of these endpoints as compared to visual field in terms of testing reliability or measures of progression (56). In the case of glaucoma and many other diseases, there is probably no single “ideal” biomarker that will cover all aspects of clinical disease including early detection, severity prediction, progression, and response to treatment (57). The challenge with biomarkers, from a long-term perspective, is that what is needed are large-scale trials with gold-standard endpoints confirming that biomarkers correlate. These are expensive endeavors if trials are to be comprehensive (58).

To optimize research strategies, there is a need to differentiate between more robust neuronal cell subsets capable of enhanced survival and more vulnerable subsets that are more difficult to preserve (3, 60). While animal models have been useful in providing a greater understanding of disease pathology and drug mechanism of action, their predictive utility has fallen short of expectations (11, 40). Some questions will never be answered in rodents, but a better understanding could be gained using non-human primates. Non-human primates are probably best suited to mimic aspects of the human visual system because there are important anatomical similarities (51, 61, 62). However, questions remain as to whether the use of non-human primates in the study of neurodegenerative diseases could bring us closer to models that are more representative of human disease (61, 63, 64). Studies performed using monkeys have become prohibitive due to the costs of the animals and their husbandry, especially over a long-term basis needed to determine the neuroprotective efficacy of drug candidates (6–8).

There are challenges at the clinical trial as well. For example, for optic neuropathies there is a need to establish a clearer definition of effective neuroprotection (54). There are a number of questions that need to be discussed in optimizing clinical trial design (Table 1).

Novel strategies to identify new targets in neuroprotection are needed (76). For example, newer testing methods should be leveraged, such as protein array analyses, single cell RNA sequencing, functional imaging (e.g., OCT), visually evoked potential (with electrodes), biofluid-based measures/biomarkers, reflex physiological measures, and automated high-throughput testing. In addition, new pathways and mediators should be explored, such a pro- and anti-inflammatory immune pathways and mediators, mitochondria/energy conservation, apoptosis/cell death, and unfolded/misfolded protein response pathways (77–81). Greater therapeutic efficacy may be achieved by addressing multiple pathways if a single molecule and/or pathway is unlikely to elicit complete therapeutic benefits (6–8, 65). A greater focus on master regulator targets that can influence multiple downstream pathways could result in therapeutic effects that extend beyond those associated with a single arm of a pathway (82). The use of pharmacodynamic biomarkers is essential to confirm target engagement (11, 83). These may be static (indicative of disease state) or dynamic (indicative of disease progression) (67). Understanding what the variability of a measurement is for a given effect size will determine biomarker utility. Pharmacodynamic biomarkers are needed to demonstrate a biological effect or determine the reason for trial failure (83, 84).

Viable therapeutic options for the effective treatment of optic neuropathies are currently lacking (74). However, there are several promising options under investigation, ranging from therapies aimed at overcoming metabolic defects and compensating for mitochondrial dysfunction, to gene therapies, to stem cell-based approaches (15). Several novel therapies have been investigated as potential areas to explore for further consideration in neuroprotection (6, 8, 13, 16, 74, 85).

Neurotrophic factors are a family of growth factors that regulate the survival, development, and differentiation of neurons, and many have been shown to promote RGC survival in experimental models of injury (2, 16). For example, a biological proteinaceous secretome from human amniotic progenitor cells containing growth factors and cytokines elicited anti-inflammatory and neuroprotective effects in a mouse model of multiple sclerosis (86). A proof-of-concept experiment involving intranasal delivery of an amnion-derived secretome in mice demonstrated prevention of visual function loss and RGC loss, as well as protection of visual function in the optic nerve (87). Neurotrophic factors are currently in human trials in glaucoma and other eye diseases, leveraging standard and exploratory biomarkers (88).

Historically, inflammation has been viewed solely as a neurodestructive event, with inflammatory cells, antibodies, and cytokines mediating loss of neurons and their connectivity. In some cases, inflammation may directly mediate the disease, following which neuronal loss follows (89). In other cases, specific categories of inflammation within the nervous system can either facilitate neuroprotection or mediate physiological or developmental neuronal mechanisms, such as complement-mediated synaptic pruning relevant to AD (90, 91). The general topic of inflammation in neurodegeneration has been extensively reviewed recently, given that it is likely to be a part of future neuroprotective strategies (92–94).

Physiologic levels of electrical activity, whether provided through direct delivery of electrical current (95, 96) or through visual stimulation that the retina then converts into electrical stimulation (97), have also been identified as excellent candidates to promote neuronal survival and axon growth. The pathways by which electrical stimulation promote neuroprotection and regeneration include elevating specific second messenger pathways such as cyclic adenosine monophosphate (cAMP), enhancing neurotrophic factor responsiveness, and likely others (70, 96, 98).

Electrical stimulation has already demonstrated some levels of efficacy in human trials of patients with optic neuropathies, mainly using externally applied alternating current stimuli. For example, an average of 7–9% increase in visual field and a concomitant improvement in central visual acuity was reported in a large retrospective case series (99). Even more convincing, at least two masked, randomized controlled trials using high resolution perimetry and other visual field measures demonstrated significant improvement in scotomatous visual field areas and central acuity in the treated compared to sham control groups (100, 101). Equally encouraging have been the extended studies looking at concomitant enhancement of brain activity and circuity with electrical stimulation, e.g., using EEG (101–103). Measures such as alpha frequency on EEG may be both biomarkers of disease as well as endpoints for therapeutic efficacy (104). In most of these studies, patients were treated for a limited time, e.g., 2-week period; future work exploring whether longer courses of treatment could give larger, or more sustained efficacy will be important to explore.

The replacement of diseased or degenerated cells by stem cell-derived RGCs could be a potential strategy to treat optic neuropathies. Stem cells are capable of differentiating into multiple cell types, thus having the potential to repair tissue and restore function after the development of lesions (2, 74). Experiments involving the transplantation of RGCs into uninjured rat retinas in vivo by intravitreal injection have demonstrated the promising potential of this approach as a cell replacement strategy (55). Transplantation was successful in terms of host retention, response to light stimulation, and axon growth along the optic nerve to appropriate target areas in the brain (55).

Leveraging biomarkers is essential in the development of drugs and therapeutics in neuroprotection (11). A clear definition of the intended use of biomarkers could enable clarification of their role in therapy discovery and development (67). Identifying biomarkers that are translatable in parallel with the development of therapeutic candidates (bilateral biomarker development) is a sensible strategy. This approach would allow for characterization of drug-responsive signatures and provide greater confidence before advancing to human studies (11, 67, 105).

To expand opportunities to develop effective neuroprotective treatments, therapeutic candidates that target key pathways should be explored (13, 16, 68, 72), as well as addressing key criteria (65) (Table 3).

Given the complexity of neurodegenerative diseases, multiple therapeutic agents may be needed to produce a clinical benefit, with each individual agent having a contributing effect (6–8, 65). Novel cell-based therapies should be explored, such as a mixture of cellular factors as a sort of combination therapy (e.g., secretome/exosomal vesicles) (86, 87). Optic nerve regeneration studies in humans are needed pending achievement of regulatory safety standards. Lastly, there is a need to review experiences from other disease states for lessons learned and templates to follow (106). The cancer field, for example, has a well-established path that outlines a timeline for therapy development (105).

For data reproducibility that gives greater confidence in preclinical findings, there is a need to use the same animal model across different laboratories (6, 7, 11). Conversely, the use of different animal models would allow for determination of whether specific biology translates across etiology using consistent methodology (11). For example, it would be ideal to observe a drug function across models which induce optic nerve inflammation by distinct mechanisms. As with biomarkers, there could be different uses for different models depending on the research question. There is a need to establish a set of guidelines for models when testing a drug or a biomarker, taking into consideration the different tools and models currently available (66). In terms of optic neuropathies, essential experimental models should be identified (e.g., axotomy, optic nerve crush, ocular hypertension induced by different means) (110). Animal age should also be considered when selecting a model and study design (11). For example, progression of glaucoma extends over a much longer period in humans compared with preclinical models (51).

There is precedent for standardization of translational research, based on experiences with epilepsy in the National Institute of Neurological Disorders and Stroke Epilepsy Therapy Screening Program (111, 112). The neurology community recognized decades ago a need for more therapies for seizures, and funded a facility where models were established and companies could provide candidate drugs for testing (111, 112). The program had well-characterized models and met requirements for moving forward with a particular drug (standardized for FDA approval). The program operates to this day, and dozens of drugs for epilepsy have resulted from it (111, 112).

DARC is a potential surrogate marker in glaucoma (17, 114). DARC has demonstrated promise in early phase trials. The technology enables the direct visualization of individual apoptosing retinal cells in both healthy subjects and patients with glaucoma, with good safety and tolerability (114, 115). In contrast to current diagnostic techniques, which allow observation of structural or functional damage that has already occurred, DARC enables direct observation of cellular apoptotic events in real-time. Those detected events could later contribute to the changes seen in OCT parameters and visual field tests (114). DARC technology provides an opportunity to monitor retinal disease activity at the in vivo level and longitudinally at the individual cell level. Thus, it may prove to be a useful adjunct to clinical endpoints when assessing treatment efficacy (114). DARC has the potential to be an early biomarker of glaucoma and may be predictive of future disease and its progression (114, 115).

In the last decade, serum neurofilaments (NFs) have emerged as a non-specific but sensitive biomarker of neurodegeneration across many neurological diseases (116). These neuron-specific structural components of motor axons are readily detected by conventional antibody-based immunoassays and are released into cerebrospinal fluid (CSF) and peripheral blood in a variety of pathological states, including ALS (58). NF biology appears to be generally consistent between animals and humans, and several studies on NFs in CSF have confirmed their diagnostic biomarker capabilities in distinguishing ALS from healthy controls or ALS mimic disorders (116). A recent prospective, multicenter, longitudinal study demonstrated that serum NFs may be considered a clinically validated prognostic biomarker for ALS with potential utility as a pharmacodynamic biomarker of treatment effect (117).

With recent technological advances in imaging, genomic, metabolomic, lipidomic, and proteomic techniques, potential new biomarkers may be identified (67). New biomarkers will require validation in large patient populations covering different stages of disease to determine their appropriate context of use in terms of being either diagnostic, predictive, prognostic, or indicative of therapeutic response (67, 113, 118).

The eye is unusual in that the retina (which contains retinal neurons and the axons of the RGCs) can be directly imaged through the clear cornea and lens. The ability to image the retina has led to multiple recent or upcoming techniques that can be used as structural or functional biomarkers. Dynamic biomarkers of neuronal and retinal structure are straightforward to quantify, and include previously mentioned OCT of the RNFL, OCT of the combined ganglion cell layer and inner plexiform layer in the macula. The latter reflects loss of RGC somas, and not only axons as with the former, and is therefore more retinotopic.

Dynamic functional biomarkers of visual system disease can be used to study progressive diseases such as glaucoma and can go beyond standard visual acuity or visual field outcomes. These include multifocal visual evoked potentials, multifocal electroretinography, pattern electroretinography, OCT-angiography, laser speckle flowmetry, and retinal oximetry (119). The addition of eye tracking with microperimetry or in addition to standard visual field analysis are also effective biomarkers (120, 121).

Use of newer technologies, e.g., high resolution imaging using adaptive optics for structure and function, are likely to be available soon (122–124). The optoretinogram can be used to image photoreceptor function at high-resolution (125, 126), and should provide additional information about neurodegenerative processes in the eye. Imaging of intracellular calcium, reactive oxygen species, and other signaling biomarkers have been developed for animal imaging, but their use in humans depends on compounds that are safe for intravenous or intravitreal use (127, 128). An alternative is the use of narrow-band or multispectral imaging of retinal fluorescence, e.g., flavoprotein fluorescence as a marker of oxidative stress (129, 130).

Key considerations for the development and use of biomarkers in neuroprotection are outlined below. Many of these considerations and themes have also been described elsewhere (51, 56, 66, 67, 83, 113, 131). Foremost are a biomarker's reliability, reproducibility, variability, characteristics, and the factors that influence these parameters. It is also important to differentiate between biomarkers that reflect disease activity and those that reflect disease severity, especially with respect to determining inclusion criteria for stage of disease or rates of progression.

Key considerations for clinical trials in neuroprotection are outlined below. Many of these considerations and themes have also been described elsewhere (17, 51, 56, 65, 66, 83, 131).