Behavioral testing in animal models of spinal cord injury

Abstract

This review is based on a lecture presented at the Craig H. Neilsen Foundation sponsored Spinal Cord Injury Training Program at Ohio State University. We discuss the advantages and challenges of injury models in rodents and theory relation to various behavioral outcome measures. We offer strategies and advice on experimental design, behavioral testing, and on the challenges, one will encounter with animal testing. This review is designed to guide those entering the field of spinal cord injury and/or involved with in vivo animal testing.

1. Animal models in spinal cord injury (SCI) research

Using animal models in SCI research is challenging for many reasons. In order to deliver meaningful results, experiments need to be reliable and suitable for the specific scientific question that is being addressed. If they are considered to be at a preclinical stage, which is not well refined in the first place, they should mimic the clinical setting as close as possible. Research in the field of neurotrauma using animals is frequently questioned from an ethical and financial point of view, especially as the translation record and appropriateness to find treatments for humans has been meager (Howells et al., 2014; Kwon et al., 2010). Researchers thus have to ensure that animal models are optimized and chosen carefully. This choice is difficult and never a black and white decision, involving many considerations ranging from the injury type and level, to the animal species. As the mathematician Norbert Wiener once stated, “the best material model for a cat is another cat, or preferably the same cat.” We tend to agree that there is no ideal model for a human with SCI, a consideration that is worsened by the significant variability within the human population. Nevertheless, animal models are extremely valuable for understanding the complex events following SCI and developing treatments to improve the quality of life for those living with SCI. As such, they have contributed essential insight.

One can imagine various scenarios for animal experimentation in SCI: 1) basic research to either study anatomy, physiology, and pathology after injury, or to explore treatments, where a low variability in the outcome measure is preferred for increased power, and 2) so-called preclinical research, where a possible treatment is tested for its suitability for translation. In the latter case, the model should be adapted to mimic the clinical scenario (including variability). Realistically, most research considered preclinical finds itself somewhere in-between these two scenarios. Designing an animal experiment at par with the clinical scenario is extremely difficult. It requires the consideration of numerous factors when trying to mimic a human SCI trial. This includes the gender and ages of the individuals that are considered for a trial, the variability of injury causes, the injury locations, extent of injuries, and pre-existing health conditions among numerous other factors (Tuszynski et al., 2007). Another important decision is to choose a suitable time point when to initiate a treatment. This decision is not as straight forward as we would like it to be and should consider clinical feasibility (e.g., not starting immediately after an injury) and differences in injury progression between the animal model and individuals with a SCI. Human SCI patients also have access to different antibiotic treatments, rehabilitative training, and other recovery facilitators (depending on the clinic), (Tuszynski et al., 2007). These factors emphasize that every human SCI is individualized and thus, the recovery of patients is also highly variable and difficult to predict. To extend the complexity further, analysis of patient recovery has to be comprehensive (motor, pain, sensory, autonomics, immune, etc.) and most individuals living with SCI are at a chronic time point, therefore, chronic outcome measures should be considered during animal experimentation as well. When bringing all these factors into play, the preclinical experiment progressively becomes less feasible both in its labor intensity and economically. Thus, tradeoffs are a part of the preclinical world. To facilitate meaningful translation, reporting experimental details and all data, as described in MISACI (Minimum Information About a Spinal cord Injury experiment; Lemmon et al., 2014), into an open-data-commons (Fouad et al., 2019; http://odc-sci.org) will be important. Making this a standard practice within the pre-clinical SCI field will provide laboratories with worldwide access to SCI data, effectively promoting meta-analysis and research transparency.

In this review, our goal is to share our experience using animal SCI models, highlighting the problems we ran into and the conclusions we drew, to help others avoid our mistakes and better position pre-clinical models for break-through discoveries. We will first discuss injury models and behavioral testing with a focus on motor function for thoracic and cervical lesions. We will then address various topics that need to be considered when planning, executing, and analyzing experiments, including the challenge of discriminating true recovery from compensation and the need/usefulness of rehabilitative training as part of any treatment regimen.

2. Injury models

Selecting from the many SCI lesion models involves numerous factors including the type (contusion, compression, section), location (cord level and neurological region within the level), and injury severity. Selection is best made by aligning the SCI model with the experimental question being asked and the appropriate outcome measures to test that question. In our experience, this is the most important and the most difficult part of the experimental design. First deliberations include: whether the study has a physiological, neuroprotective or a neuroregenerative focus; whether the analysis is anatomical, behavioral or both; what tests/outcome measures are established in the laboratory; where the study falls on the translational spectrum; and lastly, the expected potency of the treatment (i.e., how far beyond spontaneous recovery can treatment effects be analyzed without reaching a ceiling effect).

There is a significant array of injury types in individuals with SCI. The majority of spinal injuries occur from a ventral point, often by an indirect trauma (e.g., vertebral bones pressing on the cord) resulting in a compression or bruising injury (Reier et al., 2012). This is difficult to recreate in an animal model as ventral access, especially with a mechanical device like an impactor is difficult and is likely best mimicked by the clip compression injury model (Joshi and Fehlings, 2002). Another challenge is to match both the velocity of an impact and the subsequent compression of the spinal cord following a trauma. Between these two factors, it has been reported that velocity might make the difference (Speidel et al., 2020). Although contusion injuries are usually applied with a dorsal approach, they can be graded and they can reliably produce injures that appear similar to those observed in the clinical setting, including the formation of a cavity (Basso et al., 1996; Bresnahan et al., 1991; Wrathall et al., 1985). Laceration injury models are generally considered the least clinically relevant as they directly cut the dura, blood vessels, and axons and do not normally produce cavitation at the injury site (Cheriyan et al., 2014; Krajacic et al., 2010; Onifer et al., 2005; Sharif-Alhoseini et al., 2017). Such sections are also rarely applied to the ventral cord (Brustein and Rossignol, 1998; Schucht et al., 2002), with the majority taking a dorsal approach. However, this is exactly the advantage of section lesion models. They are straight forward, can be executed with great precision and repeatability, and can be analyzed more directly when it comes to axonal regeneration, as compared to contusion or compression lesions (e.g., when a lesion primarily injures white matter, the effects of axonal regeneration and sprouting can be assessed more directly).

Another obvious consideration is the level of the SCI. In humans, SCI occurs at all levels of the spinal cord (cervical to sacral), with over 60% located at cervical levels (National Spinal Cord Injury Statistical Center, 2020). In contrast, most animal studies use thoracic injury models (81%) rather than cervical (12%) (Sharif-Alhoseini et al., 2017). This trend can likely be explained by the fact that thoracic injuries are surgically more straight forward and result in less severe impact to the animals (thus ethically preferable and allow for easier animal care). Furthermore, thoracic injuries are advantageous for studies requiring a severe or complete lesion such as grafting or regeneration promoting study (Chen et al., 1996; Ilha et al., 2011; Ramsey et al., 2010; Tetzlaff et al., 2010)). As gray matter damage at the thoracic level is less functionally relevant than at the cervical level, the success of treatments promoting regeneration may be more apparent and more easilyevaluated. Of course, if the focus of a study is on respiration or forelimb function, a cervical model must be chosen, which typically use hemisections (e.g., Alilain et al., 2011; Bezdudnaya et al., 2018) or unilateral contusions (e.g., Lee et al., 2012; Schmidt et al., 2020) and less frequently bilateral contusions (Lane et al., 2012).

When deciding on the level of the injury for an animal model, it is important to account for the differences in white to gray matter proportion across cord levels and the overall size of the cord. Gray matter is smallest at thoracic levels so a thoracic contusion affects less gray matter than a cervical or lumbar contusion (Magnuson et al., 1999; Reier et al., 2002). The overall size of the thoracic cord changes with the size of rodents. This is especially important in mice where small differences in body weight (4–5 g) can cause the severity of a contusion to vary even though the injury biomechanics are equivalent. The same contusion on a larger diameter cord produces a milder injury than on a smaller cord. Controlling for body weight should be considered in experimental design.

Another important issue that is closely linked to the relative size of gray matter is the differing importance of gray matter loss. A moderate cervical contusion using standard impactors (e.g., NYU or Infinite Horizon) does not typically ablate motoneuron pools beyond 1–2 segments in rats and mice (Lam et al., 2014; Lee et al., 2012). The difference in gray matter loss over such a distance will have a significant impact on functional outcome. This is due to the complexity of neuronal circuitry and the number of motoneurons involved in controlling the limbs, which is well reflected by the differences in size of the cervical and lumbar enlargements compared to the remainder of the spinal cord. For example, the loss of thoracic motor neurons would be measured by motor control of individual intercostal muscles or portions of trunk and abdominal muscles. However, these muscles are also widely innervated from cervical to sacral levels (van Hedel and Curt, 2006; Sienkiewicz and Dudek, 2010) and their sparing can mask the loss of function caused by the denervated thoracic regions (Collazos-Castro et al., 2006). The limited behavioral effects seen due to gray matter damage in thoracic injury models stands in harsh contrast to injuries sustained at the cervical and lumbar enlargements (Hadi et al., 2000; Magnuson et al., 1999; Wilcox et al., 2017). While there is considerable overlap of muscle innervation between segments within the enlargement, little compensatory innervation exists outside the enlargement. This pattern of motor neuron organization allows us to measure at-level changes in humans. Ablating a significant number of motor neurons, which directly innervate the muscles, or interneurons, which form essential premotor circuitry (e.g., pattern generating networks orchestrating locomotion) and mediate the communication between LMNs and UMNs, in the spinal enlargements may result in a complete loss of distal muscle function. This has important implications in experimentation focusing on white matter repair as without a critical number of inter- and motoneurons in the lumbar or cervical enlargement, no level of regeneration or sprouting of spinal tracts can promote functional recovery.

Following the consideration of where and how a spinal lesion is performed, the next consideration is lesion severity. This decision will be based on various factors including the experience in the laboratory regarding animal care and outcome measures, ethical considerations, and last, but definitely not least, the experimental questions. Unfortunately, the relationship between lesion severity and behavioral deficits is complex. In fact, studies that increased lesion severity (size) by systematically sectioning or contusing more of the spinal cord established that functional deficits did not follow a linear pattern (Dunham et al., 2010; Hurd et al., 2013; Kloos et al., 2005; Schucht et al., 2002). A small amount of tissue sparing can result in significant performance enhancement overall or greater function of specific movement features (i.e. stepping but not toe clearance in locomotion). Thus, it is very difficult to predict the resulting motor deficits of a particular lesion size, making it necessary to understand the neural systems affected and their role in the motor tests selected to measure recovery. Using locomotion as an example, sparing the tracts most significant to walking will support quite a bit of recovery regardless of the amount of spinal cord damage (Brustein and Rossignol, 1998; Schucht et al., 2002). Severe lesions with a small portion of white matter spared in the ventrolateral funiculus can result in surprising locomotor recovery while a similar amount of sparing in the dorsal funiculus does not enable walking. In the case of reaching and grasping ability after cervical lesions, the dorsolateral quadrant (DLQ) lesion is an ideal injury model (Hurd et al., 2013). The relevant tracts for the behavior are lesioned without incurring so much damage that recovery would be prevented or so little damage that small or no functional deficits occur, and recovery is unnecessary. General versus precise lesions both offer important strengths and limitations.

An important goal when deciding on lesion severity is the ability to follow recovery with outcome measures that prevent floor and ceiling effects. A ceiling effect occurs when the recovery in the control group is so close to normal performance that a treatment effect would not be detectable. Conversely, a floor effect occurs when the recovery is underestimated or not detectable due to insensitivity of the chosen outcome measures. Injury severity can be modulated to test biologically relevant forms of recovery. For locomotion, a key characteristic is moving from not stepping to stepping. Selecting an SCI severity that is below the threshold for stepping makes it feasible to use recovery of stepping as a main outcome measure. Similarly, titrating the injury severity to differentiate between recovery of reaching vs recovery of grasping produces a strong experimental design and greater likelihood for translation if successful.

Other considerations for the lesion model include details that might indirectly influence the outcome, for example, choosing to leave the dura open or closed likely has neuro-immunological implications. Furthermore, lesion severity and location can influence the degree of changes in autonomic function (Hou and Rabchevsky, 2014), gut microbiome (Kigerl et al., 2018; Schmidt et al., 2020), and infection risk due to reduced immune function (Prüss et al., 2017; Riegger et al., 2007). Overall, there are numerous factors at play when designing an animal model of SCI and at least for preclinical studies, these factors must be balanced to the complexity of human SCI in order to translate an experiment to the clinical level.

3. Behavioral testing

In the SCI field, the ultimate arbiter of a successful experiment is improvement in behavior without a worsening of function or condition. This pivotal position stems from the urgent demand by people living with SCI for greater function and higher quality of life. While a great deal of effort is focused experimentally on identifying new methods to promote regeneration or neuroprotection, the essential criterion rests on recovery of function; thereby making behavioral testing perhaps the most important element of any experimental design.

An important consideration when using behavioral outcome measures is to reduce variability between and within species, while increasing reproducibility, sensitivity to treatments, and transparency (to facilitate interpretation). Similar to choosing a lesion model, deciding which behavioral tests to use should be highly dependent on the experimental question being asked, the expected recovery, and the changes in the neural systems involved/targeted. In other words, behavioral tests have to match the neural substrate and the expected outcome has to match the sensitivity of the behavioral test. Gross behaviors should be measured by a general behavioral test while specialized tests analyze more complex behaviors and higher motor skills. Regardless of the scope of the behavioral test (general or precise), only those with good reliability, validity and sensitivity should be considered. Another consideration is that the behavioral test match but not surpass the functional capacity of the animals being tested. For example, applying a test that requires weight support like von frey monofilament testing or grid/ladder before it has recovered or in a severe injury will not test the experimental hypothesis. There is little to be learned from a finding that the animals can’t be tested. Along with choosing suitable behavioral tests, the entire recovery process should be approached with a holistic view, especially for translational research. Possible side effects of a treatment should be explored including effects on autonomic function, neuropathic pain, respiratory function, immune function and last but not least behavioral changes in cognitive function and mental health associated behaviors. For example, in human SCI, decreased respiratory function occurs even in lower thoracic injuries (ref) but respiratory rate is rarely considered as a relevant outcome measure for thoracic contusions and even in some cervical lesions (Yong et al., 2012). Unless testing for and reporting potential side effects occurs, the benefits of established pre-clinical animal models may go unrealized from a translational perspective. It is important for all findings to be reported regardless of outcome. Frequently only selected, positive findings get published, and used as a proof for the benefits of a treatment. Additionally, reporting the disposition of all animals assigned to the study should be adopted in order to provide full transparency and better prepare future studies for clinical translation. The process of reporting the best outcomes does not advance our understanding of complex SCI whereas a well-designed study with largely negative outcome measures will be beneficial in bringing the field forward.

3.1. Testing hindlimb function

The most common outcome measure following thoracic SCI in animal models is locomotion. The spinal circuitry that orchestrates locomotion (i.e., rhythmic and stereotypic leg movements), is substantially retained across species and has been extensively studied (reviewed in McCrea and Rybak, 2008; Steuer and Guertin, 2019). These features allow greater interpretation of recovery across species and hold possible application to human SCI. Locomotor circuitry, called central pattern generators (CPGs), is composed of feedback and feedforward loops which produce alternating stepping movements (Prochazka et al., 2002). This circuitry can adapt to environmental challenges, including changes in slope or obstacles, without supraspinal input. Proof of CPGs extend across animal species ranging from invertebrates to primates but has been challenging in humans because brain input cannot be ruled out definitively. Human CPGs have been inferred from studies of infant stepping (Yang et al., 1998) and from treadmill walking in individuals with severe spinal cord injury with or without electrical stimulation of the lumbar cord (Dietz, 2003; Dobkin et al., 1995; Maegele et al., 2002; Rejc et al., 2017). Sensory feedback plays an important role in orchestrating stepping. To initiate the swing phase, hip flexors muscles must stretch, and leg extensors muscles must be unloaded (Pearson, 2004). This feedback system is very powerful in humans (as tested in infants; Pang and Yang, 2000) and animals alike (Büschges, 2005; Hiebert and Pearson, 1999; Prochazka et al., 2002). Studies have demonstrated that stimulating relevant afferents to mimic loading kept the leg extended and prevented the leg from lifting up even though the treadmill stretched the hip flexors (Fouad and Pearson, 1997; Whelan et al., 1995). Sensory feedback forms the underlying mechanism to the efficacy of treadmill training; to promote locomotion following SCI by stimulating the appropriate afferents that trigger the spinal CPG to initiate walking. After SCI, assessing locomotion depends on whether animals can step (see Table 1 for common examples). For those that step, walking on a treadmill or runway can be quantified using kinematics, footprint analysis, and/or EMG (Ballermann et al., 2006; Hamers et al., 2001; Kaegi et al., 2002; Merkler et al., 2001). The added advantage of treadmill locomotion is that swing initiation is easily triggered via hip extension and comparable walking speeds can be achieved. EMG analysis is highly sensitive to recovery as it detects deficits not visible using kinematics (Ballermann et al., 2006; Kanagal and Muir, 2009; Neckel et al., 2020). Sensitive EMG parameters include co-contraction of joint agonist and antagonistic muscles as well as activation timing and amplitude. However, EMG analysis has a poor cost-to-benefit ratio as it requires significant expertise to implant electrodes in relevant muscles and the electrodes can lose their recording capacity over time. Another challenge is that a foreign body response may occur when electrodes are implanted, affecting the inflammatory response to SCI.

Table 1.

Behavioral outcome measures for assessing hindpaw function.

Behavior (Hindpaw)	Neural Substrate	Gross/Fine
Incline Plane	Rubrospinal	Gross
Grid/Ladder *	Corticospinal	Fine
Catwalk/Open Field/Activity Box *	Vestibulo-, Rubro-, Reticulospinal, CPG	Catwalk (Fine) Open Field (Gross) Activity Box (Intermediate)
Swimming	Similar to locomotion	Gross to Fine (Kinematics)
Plantar Placing	Lumbar Segmental Reflex	Gross to Fine (Kinematics)
Contact Placing	Supraspinal/Corticospinal	Gross to Fine
Air Righting	Vestibulospinal	Gross to Fine (Time)
Contact Righting	Vestibulo-, Reticulo-, Propriospinal	Gross to Fine (Time)
Von Frey Hair (VFH) *	Aβ Mechanoreceptors in Dermatome	Gross (Yes/No) Fine (Threshold)
Plantar Heat *	Aδ and C-fibers	Fine (Time)

Note: Serotonergic and other neuromodulatory fibers are important in regulating spinal excitability and thus influence all motor tasks.

^*These tasks can also be utilized for testing forelimb function.

Likely the most common analysis of locomotor recovery in rodents is measured in an open field and scored using the Basso, Beattie, Bresnahan (BBB) locomotor rating score in rats and the Basso Mouse Scale for locomotion (BMS) score in mice (Basso et al., 1995, 1996). The BBB score, introduced in 1995, has established transparency, predictive validity and sensitivity to measure locomotor recovery after incomplete thoracic contusion injury. A strength of these scores is that a single rating system can be used to compare normal locomotion to the most severely impaired; there are no ceiling nor floor effects. Another unique feature of the BBB and BMS, is that they are based on operational definitions which provide a universal language between researchers on the analysis of locomotor function. This effectively increased the transparency of SCI research, allowing one scientist to easily understand the scoring system of another. It is important to stick to a common language in order to enhance transparency rather than continuously designing new tests.

When examining mechanisms of locomotor recovery, it is important to realize that CPGs possess a memory and are capable of “learning”. This has been shown impressively by experiments using cats that received first a spinal hemisection, followed by training (or not) and then a complete SCI (Martinez et al., 2012). Those animals that were trained after the hemisection responded much better to training following the complete (second) lesion, indicating that changes in the circuitry below the lesion adapted due to the initial training. This has important implications because an experimental treatment may have the potential to promote recovery by directly affecting the CPG system rather than inducing reparative mechanisms of injured axons. This makes it difficult to interpret the mechanism of recovery. To better differentiate between the restoration of descending function compared to pure spinal function, tests that challenge the motor system and require higher level functioning (brain and brainstem) can be applied (e.g., the horizontal ladder (Bolton et al., 2006; Metz and Whishaw, 2002) grid walk (Ma et al., 2001; Prakriya et al., 1993). or narrow beam test (Hicks and D'Amato, 1975). These tests, however, strongly depended on lesion size and a minimum level of function given that severe injuries will not allow meaningful interpretation of the results (e.g., when animals drag their legs over a narrow beam or a horizontal ladder).

An obvious difference between animals and humans is quadrupedal locomotion which warrants further consideration (see Muir and Steeves, 1995 for exceptions). Rats, even with complete paraplegia, can reach significant walking speed using their forelimbs alone, which is not a similar option for humans. Additionally, quadrupedal animals with SCI constantly engage CPG circuitry due to sensory feedback when dragging their legs. Importantly, this contributes to their motor recovery (Caudle et al., 2011).

A less frequent approach to locomotor testing is observing animals walk in shallow water (Kuerzi et al., 2010). This is an easy and fairly inexpensive approach to explore the ability of the system to generate a locomotor pattern with partial weight support. Weight support can be adjusted by changing the water level, to a degree that swimming can also be tested. Both water stepping and swimming can be scored, and evaluated using kinematics (Magnuson et al., 2009; Schnell et al., 2011; Smith et al., 2006). The Rota-Rod, although well established in behavioral testing in other fields of research, is a less frequently applied locomotor test after SCI (e.g., Qian et al., 2018). A reason might be that rota-rod performance is based on various factors including motivation, cardiovascular function and motor function. Other tests to evaluate hindlimb function and their corresponding neural substrate are summarized in Table 1, including the incline plane test (Rivlin and Tator, 1977), the plantar placing test (Kunkel-Bagden et al., 1992), air righting (Altman and Sudarshan, 1975), and sensory tests (Detloff et al., 2012; Hargreaves et al., 1988; Lindsey et al., 2000).

3.2. Testing forelimb function

The desire to recover hand function is of high importance to individuals' living with SCI given that cervical injuries are the most common SCI (Anderson, 2004). These factors make forelimb testing a popular outcome measure in experimental SCI. Forelimb functions like reaching and grasping are largely controlled by the brain and brainstem rather than spinal circuitry (see locomotion). As an outcome measure, it can evaluate the restoration of connectivity between the brain and the spinal cord. Not surprisingly, testing forelimb function also comes with certain limitations and weaknesses. Firstly, most lesion models are unilateral (or at least affecting one side more than the other), as moderate and severe bilateral injuries can cause severe challenges for animal survival and well-being. This is, however, different between species as the recovery in primates following severe unilateral injuries is more robust than in rodents, likely due to the existence of crossing CST fibers (Friedli et al., 2015; Rosenzweig et al., 2010). Furthermore, similar to humans, rodents have a preferred forelimb, which needs to be identified for each animal before considering lesions and the testing approach. Unilateral lesions also make testing difficult as rats may compensate by using the limb with better function (Torres-Espín et al., 2018b; Whishaw et al., 1986). Lastly, various forelimb assessments/tasks require preinjury training and are fairly time-consuming. To reduce the workload, recent strategies from various laboratories have adopted automation of these tests (Ellens et al., 2016; Fenrich et al., 2014; Sindhurakar et al., 2017).

There is a huge variety of forelimb tests with varying sensitivity and specificity (see Table 2 for common examples). More simplistic and straight forward tests include: measurements of grip strength where rodents that hold on to a bar connected to a force sensor get pulled away until they release (Onifer et al., 1997), the cylinder or paw preference test where rodents are placed into a glass/or Plexiglas cylinder and the exploration of a wall is compared between the affected and the unaffected paw (Dunham et al., 2010; Schallert et al., 2000), grooming motion analysis (which requires video analysis; (Gensel et al., 2006; Lee et al., 2010), cereal manipulation (Irvine et al., 2010), a pasta breaking/eating tests (Ballermann et al., 2001), and the Montoya staircase test (Lee et al., 2010; Montoya et al., 1991). The latter has the advantage that it tests a more complex movement (i.e., retrieving pellets), it discriminates clearly between the left and right paw, and does not require complicated training as in the single pellet reaching task. The advantage of the single pellet reaching task is that it allows a quantification of reaching and retrieving success, and the qualitative analysis of the reaching movement (Whishaw et al., 2002). This is invaluable for identifying compensatory approaches. Another approach is to train animals in pulling a lever or rotating a knob with increasing force and distance (Butensky et al., 2017; Guo et al., 2014), which quantifies changes in task specific reaching behavior.

Table 2.

Behavioral outcome measures for assessing forpaw function.

Behavior (Forepaw)	Neural substrate	Gross/fine
Rodent Grip Strength Test	Peripheral and central sensory systems (dorsal funiculus of the spinal cord, dorsal spinal roots)	Gross
Cylinder	Reticulospinal, vestibulospinal	Gross
Montoya Staircase	Cortico-, Rubro-, Reticulospinal Tracts	Fine
Grooming Motion Analysis	Brainstem, cerebellum, striatum (for coordination, modulation and pattern generation)	Gross
	Cortex, hypothalamus and amygdala (for regulation)
Froot Loop Manipulation	Cortico-, Rubro-, Reticulospinal Tracts	Fine
Pasta Breaking/Eating
Single Pellet Grasping
Pulling a lever/Rotating Knob

Note: Serotonergic and other neuromodulatory fibers are important in regulating spinal excitability and thus influence all motor tasks.

It is important to keep in mind that all tests have strengths and weaknesses. Many of the tests we have mentioned can be adapted to the species being examined. A few examples include increasing/decreasing the distance between rungs for the horizontal ladder walk. For the catwalk, placing a thin layer of water on the surface so the light is attracted to where the water moves, not just the paw, generating a brighter signal. The height of the rotarod can also be adjusted to the size of each species. Food pellet sizes can be varied, and testing chambers adjusted. Notably some tests cannot discriminate between recovery and side effects or spasticity (e.g., grip strength even gets scored higher with spasticity, (Lu et al., 2012). Lastly, not every test is suitable for similar lesion severities and enables testing over a wide range.

4. Common Challenges

4.1. Variability

A common problem in behavioral testing is variability within experimental groups, between different experimental time points, and between different experiments. This variability introduces various problems, including the requirement for high animal numbers to increase statistical power and the occurrence of false positive or negative results. Thus, it is a main desire in hypothesis driven research to decrease variability, and to increase the reliability and repeatability of behavioral results. The challenge is the large number of factors potentially contributing to variability. Not surprisingly, one of the major sources of variability is the researcher (or rater) themselves. Sometimes subtle things can influence the outcome significantly, including the time testing is performed, the day of the week (and the associated attitude of the rater), the mood of the rater, a new cologne or shampoo, whether and how animals are encouraged to perform a task and many more. Sometimes it is external factors like the animal strain, a different breeder, the season of the year, the sex of the animal, stress, different drug batches, atypical vivarium conditions like lighting or noise, activity (e.g., different cage seizes) etc. It is important to consider as many of these variables as possible and to maintain the same testing standard. Kriegsfeld and colleagues (Kriegsfeld et al., 1999) showed clearly different outcomes in rodents of different genetic backgrounds when tested on a balance task, during the day or the night. We have anecdotal evidence that behavioral tests resulted in superior results when they were performed in the middle of the week rather than on Mondays or Fridays. One straightforward solution to conducting behavioral tests, is to keep the environment the same and to keep the rater for each test the same, which is especially important for sensory tests. It has been noted that for sensory testing, novice raters scored on average 92% lower than experienced raters. As well, experience with testing can affect response rate by as much as 33% for proprioceptive placing (Basso et al., 2015). With the high potential of variability within an experiment, it is important to confirm testing skill against a gold standard as a means of minimizing this variability. Another way to address variability is to evaluate recovery in a bigger picture by using subjective scores (e.g., BBB) and objective measures (e.g., ladder walk, kinematics, EMG, catwalk etc.). This allows us to evaluate a recovery pattern and not get lost in a single and potentially variable outcome measure.

Several solutions can be used to minimize the effects of variability in results. One of these includes a Latin square where the different tests being used are randomly balanced across testing sessions (essentially each test will have the chance to be first, last and in between; e.g., Meliska and Loke, 1984). The advantage of this system is that the animals have the chance at peak performance at least once. However, it is a complex design and may mask the effects of recovery over time. Fatigue effects will also lessen over time making the last test of the day less relevant acutely/subacuately after SCI. Another method of organizing tests includes performing the most significant test first and saving the least significant tasks for the end when the animal is most fatigued.

A very different approach to deal with variability is to explore its sources. This can be approached especially in larger data sets using principle component or cluster analysis, and comparing data between different experiments and laboratories (Ferguson et al., 2014; Nielson et al., 2014). Thus, making all data publicly available is an attractive approach to expand the use of experimental data, and it enhances research transparency. Lastly, when publishing all data (whether a paper is published or not), will greatly reduce the amount of dark data and data bias (Fouad et al., 2019; Ioannidis, 2005).

4.2. Animal housing, health and enrichment

Enrichment of the animal environment improves the mental health and quality of life of each animal. It is very important to provide animals with the opportunity to socialize and interact with new environments and objects regularly, encouraging the animals' curiosity, exploration, and movement. The amount of motor activity in the cage clearly influences activity (Berrocal et al., 2007; Fischer and Peduzzi, 2007). This has been dramatically shown when rats with thoracic SCI where restrained in a “wheelchair”, which compromised the normally observed spontaneous recovery (Caudle et al., 2011). Even untrained animals co-housed with trained animals have better locomotor abilities than untrained animals co-housed with other untrained animals (Risedal et al., 2002). A possible explanation is that untrained animals have to compete with trained animals for access to resources. An enriching environment also corresponds to better animal well-being. This is not only important for spontaneous recovery, but also participation in testing and training. In a grasping task based on self-motivation, animals will grasp with a high success rate prior to injury however, after injury, this success rate declines significantly and the motivation to grasp along with it (Fenrich et al., 2016; May et al., 2015). This raises the question if this decline occurs because the animal simply gives up or cannot perform the task in general.

Co-housing animals is important to maintain the “mental health” of animals. It provides the opportunity to socialize and reduces anxiety and depression like behaviors in healthy animals (Pinelli et al., 2017) and is especially important after SCI animals when anxiety and depression like behaviors are prevalent. These changes in “mental health” have recently been associated with inflammation and possibly changes in the gut microbiome post SCI (Maldonado-Bouchard et al., 2016; Schmidt et al., 2020; Wu et al., 2014). Co-housing may benefit animals because healthy animals may share their microbiome (being coprophagic) with unhealthy animals to improve their health. However, co-housing may also affect the outcome of the experiment if perhaps an antibiotic or a drug is used to manipulate an animal's microbiome but is then restored due to access to other animal's feces and other sources of bacteria.

4.3. Experimental design

After deciding on a lesion model, more questions regarding the design of the study need to be addressed. One has to ask whether the study is testing the effects of a drug to prove a hypothesis or whether it is trying to test a drug as treatment relevant for translation. Ideally, to be considered preclinical, one wants to ensure that the study mimics the clinical setting. This is challenging, however, considering our discussion that reducing variability is a key to producing reliable outcomes. In the clinical setting, variability is a given. To understand this, it is worth looking into the inclusion criteria of a typical trial, where both genders are considered, and a certain variability in age, lesion size, and location is accepted (Tuszynski et al., 2007). Specifically for lesion size, it is recommended for animal studies to relate the variability in behavior to the variability in lesion. Limiting variability of the lesion size minimizes variability in behavior between animals and defining a desired variability in lesion size beforehand allows the study to match the reality of a clinical trial. In preclinical research the addition of this variability would be essential to predict the translational success of a treatment, however considering the required animal numbers, and the related amount of work, this approach is frequently neglected. In the competitive research world productivity and clinical relevance seem to clash, and as researchers we need to find a compromise for the stage of the research projects.

Topics to enhance clinical relevance that may be easier to address include the choice of a relevant treatment time point (e.g., not immediately after injury), the addition of treatments standardly given in the clinic (antibiotics or rehabilitative training), using comprehensive outcome measures (including motor and sensory function, pain, autonomic function etc.). Lastly chronic treatments and measuring outcomes at chronic stages are frequently neglected despite their significant clinical relevance.

In conclusion, a study ranked higher/better on the translational spectrum will turn into a much more complex experiment. This potentially reduces the productivity of a lab and may introduce more opportunities for errors and mistakes. Realistically, here are only so many things that can be done and there is no ideal experimental design that will check all the boxes.

5. Compensation or recovery?

Defining recovery is not straight forward. Is recovery when an individual with a SCI gets better in an affected task or movement? What does better mean? To succeed in the task or succeed in performing it as it was done pre-injury? In the case of locomotion, compensation may involve using an assistive device to facilitate walking while recovery may be regaining the walking motion itself. It is very clear that after SCI, recovery is based on both the development of compensatory approaches and relearning of a motor pattern that is as close as possible to the original movement pattern; both involve neural plasticity (Curt et al., 2008; Hastings et al., 2012). In physical therapy, it is important to harness compensatory movements to then work towards restoring original function. Compensatory strategies can however also create a dead end for further recovery where individuals will not be able to progress without first reducing overall success to overcome compensatory strategies.

In animal studies, it is important to consider whether the animal is truly recovering or whether it is simply using methods to compensate for its deficits. Sometimes these strategies are not immediately obvious, for example contractions and stiffening limbs during standing and walking can assist in weight support (Ballermann et al., 2006; Harbeau et al., 2002; Neckel et al., 2020). Walking with a wider base and rotation of hind paws (see BBB score, Basso et al., 1994, 1995, 1996) is likely a mechanism to restore balance.

In tests involving pellet retrieval, compensation could be switching to the non-preferred paw, using the tongue to lick pellets, or most commonly scooping the pellets and dragging them towards the mouth. We recently show that training after SCI simply promotes the success rate in retrieving, however when success via scooping is prevented, the animals are as successful as untrained ones (Torres-Espín et al., 2018a, 2018b).

As discussed above, rehabilitative training following SCI is a standard for individuals with SCI, and currently the best option to promote recovery (Fu et al., 2016; Musselman et al., 2018; Sandrow-Feinberg and Houlé, 2015). Many studies have implemented training into preclinical animal research, and these studies found that training can be essential to translate plasticity promoting treatments to recovery (Torres-Espín et al., 2018a). In other words, it is highly recommended to include rehabilitative training in preclinical research, despite its time- consuming nature. Many studies inadvertently include training, as regular testing could be considered training. It has to be noted, however, that the effect of training following injury is time dependent (i.e., time after lesion) and does not always improve recovery. For example, treadmill training during a period of low inflammation will improve walking while training during high inflammation may worsen the ability to walk (Hansen et al., 2013). Whereas training in the chronic phase appears to be less effective than applied in the subacute phase and adding a mild inflammatory stimulus in the chronic setting can counteract this decline (Torres-Espín et al., 2018b).

6. Final suggestions and summary

• Choose lesion models according to tests & treatment effect

• Perform pilot experiments to help predict treatment efficacy and to adjust outcome measures or lesion size

• Avoid ceiling and floor effects

• Consider mixed methods design where allowing qualitative and quantitative outcome measures

• Reduce unnecessary variability or add clinically relevant variability (depending on stage of project on the translational spectrum)

• Do not use too many outcome measures, but use comprehensive ones

• Consider rehabilitative training

• Verify data that were collected in an automated manner

• Keep it simple, observe and look at the bigger picture.

• Report results from ALL outcome measures and all animals that fulfill predetermined exclusion criteria

7. Conclusion

When designing experiments and choosing which injury model to use or what behavioral and motor tests to deploy requires constant observation to obtain an accurate evaluation and make appropriate adjustments to the experiment. Research does not follow a cookbook. Ensure that the tests and injury models chosen pertain to the experimental question being asked. Both qualitative and quantitative measures are necessary to obtain a comprehensive picture of the SCI treatment outcomes. Keep the experiment simple and specific.