The protocol for this review was published on PROSPERO (an international prospective register of systematic reviews) on 3 December 2019 (CRD42019149671).

The inclusion and exclusion criteria used in this review are presented in full in the Supplementary Material.

To identify articles, a systematic search was performed of MEDLINE and Embase databases on December 14, 2020.

The search strategy was developed with the assistance of a medical librarian at the Cambridge University Medical Library. The terms used to search MEDLINE and Embase are provided in the Supplementary Material. No additional search limits were applied.

Duplicates were excluded in Mendeley (Elsevier, London, UK). The abstracts were then screened independently by 14 of the authors using Rayyan software. The abstracts were divided into 7 groups. Each group of abstracts was reviewed by a pair of authors. Disagreements were resolved through discussion between the reviewers until mutual agreement was reached.

The data extracted were: author, year of publication, country of experiments, study characteristics (e.g. number of experimental groups, level of evidence), sample characteristics e.g. size, number of groups, species, strain, age, sex, weight, and comorbidities), intervention (including injury model and the type, dose, frequency and route of drug), the methods and results of any neurobehavioural assessment, and the nature of any relevant statistical analysis performed. Data extraction was performed by one of the authors (FB).

Due to the diverse range of injury models, interventions and outcomes, meta-analysis was not possible and a narrative synthesis using the Synthesis Without Meta-analysis (SWiM) reporting guideline was conducted [28]. A checklist of adherence is provided in the Supplementary Material. Studies were primarily grouped based on outcome measures and secondarily the cannabinoid intervention used. The three broad outcome categories were motor function, pain and anxiolysis. The differences (mean/median/p values) between intervention and control groups reported by individual studies for each outcome measure was initially summarised in a table. Findings were subsequently summarised by transforming the difference measures into standardised metric-direction of effect (positive/negative/no effect) and synthesised in the form of harvest plots by vote-counting based on direction of effects. Combining p values and calculating summary statistics of intervention effect estimates was not possible. Heterogeneity in reported effects was investigated by structuring figures around the injury model, interventions, and outcomes assessed; these are provided in the Supplementary Material. The SYRCLE (Systematic Review Centre for Laboratory Animal Experimentation) tool was used to evaluate the risk of bias of the included studies [29]. Since risk of bias was determined to be ‘unclear’ for all the included studies, it could not be used to prioritise the reporting of certain findings over others, thus study findings were reported equally in the narrative synthesis.

The search generated 8714 results. A total of 2062 duplicates were removed using Mendeley (Elsevier, London, UK), resulting in 6652 unique studies. Following the abstract screening, 41 studies were found to meet the inclusion criteria (Supplementary Material). On full-text screening, 23 studies were excluded for the reasons outlined in the Supplementary Material. One additional relevant study was identified in the reference list of an included study. In total, 19 studies were included (Fig. 1).Fig. 1

Number of studies identified, screened, assessed for eligibility and included are shown.

Of the 19 included studies, 13 studies used rat models of SCI [30–42] and 6 used mice models [43–48]. Sprague-Dawley rats were used in 10 studies [30, 33–41] and Wistar rats in 3 studies [31, 32, 42]. With regards to strains of mice, CD1 mice were used by 3 studies [43, 45, 48] while CB57BL/6J [44], CB57BL/6 [46] and PPAR-αKO mice with litter-mate wildtype controls [47] were each used in 1 study. Male animals were used in 16 studies [30–45] whilst female animals were used in 1 [46]. 2 studies did not specify the sex of the animals used [47, 48]. The age of mice and rats were not specified in 11 studies [34–42, 46, 48]. Of the 8 studies which commented on animal age, 4 studies explicitly specified the age of the animals used [31, 33, 44, 47] and 4 studies stated adult animals were used [30, 32, 43, 45].

The different injury models used are summarised in Fig. 2 and Table 1. Compression models, involving microvascular clips (n = 9) [35–39, 43, 45, 47, 48] and silicon tube insertion into the vertebral canal [44], and contusion models involving spinal cord impaction devices (n = 4) [30, 31, 33, 46] were the most commonly used. Other models used included spinal cord ischaemia via aortic occlusion [40], cryogenic injury using liquid nitrogen [42], spinal cord hemisection [32], ischaemia-reperfusion injury via aortic occlusion [34] and ischaemia-reperfusion injury via aortic clamping [41]. The majority of SCI models used thoracic injury (n = 15) [30, 31, 33, 35–39, 42–48], whilst one study used cervical level injury (n = 1) [32].Fig. 2

Number of studies using each injury model and the reported effects on locomotor function (A) and pain (B) are shown.

Eight different cannabinoid receptor agonists were used as seen in Fig. 3. WIN 55,212-2 (n = 6) [30, 33, 35, 36, 38, 40] was the most commonly used agonist. Palmitoylethanolamide (PEA) [43, 47] and cannabidiol (CBD) [42, 46] were each used in 2 studies. CP 55,940 [39], JWH [32], N-(2-chloroethyl)−5Z, 8Z, 11Z, 14Z-eicosatetraenamide (ACEA) [44], oxazoline of Palmitoylethanolamide [45] and co-ultramicronised PEA and luteolin [48] were each used in 1 study. Two different inverse cannabinoid receptor agonists were used: rimonabant (n = 2) [38, 39] and hemopressin (n = 1) [38]. Five cannabinoid receptor antagonists were also used, with AM 630 (n = 8) [30, 31, 33–35, 37, 40, 41] and AM 251 (n = 7) [30, 33–35, 37, 40, 41] being the most commonly used. Other antagonists used included SR 144528 (SR 2) (n = 2) [32, 39] and AM 281(n = 1) [31]. One study used acetaminophen and naloxone [37]; both drugs have been reported to have effects on the endocannabinoid system. Initiation of cannabinoid administration ranged from 20 min [30] to 5 weeks [35] after SCI. The duration of cannabinoid administration ranged from single one-off doses to repeat dosing over 10 weeks [46]. A comprehensive summary of the drugs used in the included studies can be found in the Supplementary Material.Fig. 3

Number of studies using each cannabinoid receptor agonist and the reported effects on locomotor function (A) and pain (B) are shown.

Across the included studies, locomotor function was evaluated by 13 studies, pain perception was evaluated by 8 studies and 1 study explored anxiety. 8 different measures of locomotor function were used (summarised in Fig. 4A). The most commonly used measures of locomotor function were Basso Mouse Scale (BMS) (n = 5) [44–48], Basso, Beattie, Bresnahan (BBB) Locomotor score (n = 4) [31, 33, 42, 43] and 14-point motor deficit index (MDI) score (n = 2) [34, 40]. Beam-walking test [32], CatWalk [32], rodent rotarod) [44], spontaneous open-field locomotor activity [44] and Tarlov scoring system [41] were each used once. Two measures of pain were used (summarised in Fig. 4B), namely the von Frey filament test (n = 5) [31, 35–39] and hind paw withdrawal to thermal stimulus (n = 3) [30, 31, 46]. The elevated plus-maze test [44] was the sole measure of anxiety (n = 1). No studies assessed the sedative effects of cannabinoids. Table 2 summarises all of the scoring systems used to evaluate outcomes. Assessment time points ranged from 30 min [35] to 90 days [31] after cannabinoid administration. Table 3 groups the included studies by neurobehavioural outcome, and summarises the sample features, injury models, interventions, assessment tools and key findings. Figs 2 and 3 investigate the effects of specific cannabinoid interventions and injury model on locomotor and pain, and highlight the heterogeneity of the included studies. Fig. 4 visually represents the overall effects of cannabinoids on locomotor function and pain. Further information relating to individual studies can be found in the Supplementary Material.Fig. 4

Effect of cannabinoids on locomotor function (A) and pain (B) stratified by outcome measure is shown. Frequency of outcome measure use is also indicated.

Allocation sequence was adequately generated and applied in 12 out of 19 studies, the remaining studies may have been randomised but did not describe their allocation sequence. Four studies stated that group neurobehavioural characteristics were similar to baseline. No studies stated if the allocation was adequately concealed or if animals were randomly housed during the experiment. No studies stated if animals were selected at random for outcome assessment. The outcome assessor was blinded in 7 out of 19 studies; however, no studies explicitly stated that caregivers or investigators were blinded. One study stated one mouse in the vehicle group died but provided no further information regarding under what circumstances this occurred. A lack of reporting of items on the SYRCLE checklist across the 19 included studies means that accurately determining the risk of bias is challenging. This means that the overall risk of bias remains unclear and all results must be interpreted in the context of this. Full details of risk of bias assessment are provided in the Supplementary Material.

13 studies assessed locomotor outcomes of which five assessed BMS [44–48], four assessed BBB locomotor score [31, 33, 42, 43], two assessed 14-point MDI [34, 40]. Rodent rotarod [44], spontaneous open-field locomotor activity [44], beam walking [32], tarlov score [41] and catWalk assessment [32] were each used in one study.

Eight studies assessed pain outcomes of which seven used the von Frey filament test [31, 35–39, 46] and three assessed hind paw withdrawal thresholds [30, 31, 46].

Hong et al. used the elevated plus-maze test to investigate the anxiolytic effects of ACEA post-SCI. No anxiolytic effect of ACEA (3 mg/kg/day) was observed [44].

WIN 55,212-2 was observed to improve BBB locomotor score, hind paw withdrawal to thermal stimulus and von Frey filament test scores when administered i.p., s.c. and i.t. [30, 33, 35, 36, 38, 40]. However, improvement was only noted at one time point (30 min post administration) in one study [38].

The effects of WIN 55,212-2 were noted to be blocked by CB1 R antagonists in 2 two studies [30, 35] and CB2 R antagonists in two studies [33, 40]. The endo-CB system is thought to be modulated in two phases following SCI [11]. An initial, acute phase in the first week of SCI is characterised by increased levels of the CB1 receptor-specific endo-CB AEA and high levels of CB1 receptor expression on neurons and oligodendrocytes [11]. This acute phase is considered important for neuronal survival [11]. Two to three weeks after SCI, in the chronic phase post-SCI, levels of the non-specific CB receptor endo-CB agonist 2-AG and CB2 receptors in macrophages and astrocyte like cells increase [11]. Increased IL-10 release from CB2 receptor-expressing macrophages has been proposed as a mechanism for CB2 receptor-mediated analgesia post-SCI [30]. Thus, the effects of WIN 55,212-2 following SCI may be mediated by its actions on both the CB1 and CB2 receptors.

Following spinal cord ischaemia, Huo et al. showed WIN 55,212-2 improved motor function, as measured by the 14-point MDI score, reduced apoptosis and improved survival of neurons [40]. Mechanistically, both Huo et al. and Su et al. showed WIN 55,212-2 treatment blocked nuclear translocation of GAPDH, formation of GAPDH/Siah1complexes and iNOS expression in the spinal cord after spinal cord ischaemia or traumatic SCI [33, 40]. GAPDH/Siah1 activity is correlated with apoptosis after ischaemic injury, thus it is proposed that WIN 55,212-2 improves functional recovery following SCI via inhibition of GAPDH/Siah1 signalling cascades and reduction in iNOS expression [33, 40].

Importantly, Hama et al. found the antinociceptive efficacy of WIN 55,212-2 to be maintained over a twice-daily 7-day treatment regimen, whereas the efficacy of morphine decreased over the same period [36]. Tolerance is a major problem with existing SCI-pain medications, hence the sustained efficacy of WIN 55,212-2 suggests that cannabinoid receptor agonists may be useful in alleviating chronic pain after SCI [36].

Potent anti-inflammatory effects of PEA were demonstrated by Genovese et al. who found that following SCI, PEA reduced the degree of spinal cord damage, neutrophil infiltration, NF-kB activation, IkB-a degradation, nitrotyrosine formation, proinflammatory cytokines production, apoptosis, Bax and Bcl-2 expression and PPAR-a degradation [43]. Paterniti et al. provided further evidence supporting the role of PPARs in the mechanism of action of PEA, finding the anti-inflammatory effects of PEA were antagonised by administration of PPARy and PPARd antagonists, and abolished in PPARa KO mice [47]. Furthermore, another study by Paterniti et al. showed a co-ultramicronised composite of PEA and luteolin restored basal expressions of PPARa, β, δ and γ post SCI [48]. Collectively, these studies suggest a significant anti-inflammatory role of PEA post-SCI.

Impellizzeri et al. found PEA-OXA treatment significantly improved BMS scores and reduced histological alterations post SCI. Mechanistically, PEA-OXA was noted to reduce astrocyte activation and increased neurotrophic factors BDNF, GDNF and NT-3 suggesting PEA-OXA has neuroprotective properties. PEA-OXA, similar to PEA, was observed to have anti-inflammatory effects, reducing degradation of IkB-a (a regulatory protein of NF-kb), and reducing expression of iNOS and COX-2, as well as the release of the pro-inflammatory cytokines TNF-a and IL-1b [45].

ACEA treatment was found to improve both functional and histological outcomes post SCI [44]. Hong et al. identified that ACEA treatment decreased matrix metalloproteinase-9 (MMP-9). MMP-9 is known to be expressed in neurons, reactive astrocytes, infiltrating leucocytes and increased activity results in blood-spinal cord barrier disruption and decreased functional recovery following SCI [49].

Two studies implicated the endo-CB system as having a role in the protective effects of RIPC prior to ischaemia/reperfusion injury [41, 34]. Firstly, RIPC has been observed to increase AEA content in the spinal cord following ischaemia/reperfusion injury [34]. Secondly, both Jing et al. and Su et al. reported cannabinoid receptor antagonists reduced the protective effects of RIPC. Jing et al. found the protective effects of RIPC to be reduced at 4 h post-ischaemia/reperfusion injury by CB1, but not CB2, receptor antagonist pre-treatment [41]. At 24 h post-ischaemia/reperfusion injury pre-treatment with either CB1 or CB2 receptor antagonists abolished the neuroprotective effect of RIPC [41]. This implies involvement of both CB1 and CB2 receptors in the protective effects of RIPC. The findings of Su et al. differed to those of Jing et al.. Su et al. report that the protective effects of RIPC were only reversed by CB1 R antagonists at 24 and 48 h post ischaemia/reperfusion injury, implicating involvement of only the CB1 receptor [34]. Therefore, whilst these findings support the involvement of the endo-CB system in RIPC, it is not clear whether both of the cannabinoid receptors are involved.

A large degree of heterogeneity exists in the spinal cord injury models utilised by the included studies, as highlighted in Fig. 2. Each model induces different injuries with different pathophysiology. This is important when considering the potential translation of pre-clinical experiments into human studies. Each model has specific advantages and limitations (summarised in Table 1), and different types of preclinical SCI model may be required to address specific research questions. For example, transection models are a useful method of exploring neuronal regeneration and degeneration, but the pathophysiology of the injury is different to the contusive injury mechanisms more commonly seen in humans [50]. Contusion models are typically regarded as most representative of acute, traumatic SCI, whereas the more chronic injury produced by some compression models may be more representative of conditions such as degenerative cervical myelopathy [51]. The age of animals used provided another source of heterogeneity. Differences in animal age limit interpretation of the included studies as functional behaviours and regeneration after SCI may differ depending on animal age [52]. It is important that future pre-clinical work carefully considers which aspects of human pathology they aim to mimic through animal models, and choose a pre-clinical model and outcomes measures that appropriately reflect this.

There is also heterogeneity amongst the cannabinoid receptor agonists used in the included studies as seen in Fig. 3. WIN 55,212-2 was the most used compound and was noted to produce improvements in both locomotor and pain outcomes [30, 33, 35, 36, 38, 40]. Within those studies that used WIN, various routes of administration were used including intraperitoneal, subcutaneous, intrathecal and intracerebroventricular routes. Furthermore, dosing protocols varied between one off doses and repeat injections. These differences have important clinical implications as some routes of administration (e.g. subcutaneous injection may be easier than others, e.g. intrathecal injection or intracerebroventricular drug administration). The studies included in this review focus primarily on the pharmacodynamics of cannabinoid receptor agonists. However, before clinical trials can be considered, the pharmacokinetics and toxicity of these compounds must also be investigated. Thus, given that multiple compounds have shown improvements in locomotor function and pain scores, future work will need to determine which of these may be most amenable to human translation. This will require essential work investigating pharmacokinetics and distribution, safety and off-target effects, and logistical considerations such as the stability of compounds and shelf-life for use in the clinical environment. Future pre-clinical studies should aim to reach a consensus on lead compounds, explore suitable, clinically relevant dosing regimens, and determine acceptable clinical trade-offs such as the route and frequency of administration.

In our analysis, multiple items from the SYRCLE checklist were not commented on by all 19 studies, including whether animals were randomly selected for outcome assessment, or whether allocation was adequately concealed, and caregivers or investigators were blinded [29]. Due to this, risk of bias remains unclear for the included studies. This appears to be a common problem in the preclinical cannabinoid evidence base. The IASP Presidential Taskforce on Cannabis and Cannabinoid Analgesia identified similar challenges, including the unclear risk of bias due to lack of reporting of methodological criteria [25]. They posit this may be due to reporting of these terms not previously being required by journals for publication [25]. Future pre-clinical trials should be encouraged to follow the ‘Animal Research: Reporting of In Vivo Experiments’ (ARRIVE) guidelines to improve the quality of evidence generated [53].

The beneficial effects of cannabinoid agonists on outcomes assessing pain and locomotor function in animal models of SCI strengthens the argument that there may be scope for the endo-cannabinoid system to be harnessed in the treatment of motor and pain-related symptoms seen following human SCI. Despite many of the drugs discussed in this systematic review not being licensed for clinical use, there are now well-tolerated cannabinoid agonists used in clinical practice. These drugs may represent opportunities for translational clinical trials, bypassing the lengthy and costly process of licensing novel drugs. Sativex and Epidolex are two such drugs. Sativex, an oromucosal spray containing tetrahydrocannbinol and cannabidiol has been licensed in the United Kingdom since 2010 for the treatment of spasticity and other symptoms of multiple sclerosis [54]. Epidolex, a 99% pure oral CBD extract, was the first of its kind to be licensed by the FDA in June 2018 for the treatment of Lennox–Gastaut syndrome, Dravet syndrome and other severe forms of epilepsy [55]. A further three synthetic cannabis-related drug products have since been FDA approved, namely Marinol (dronabinol), Syndros (dronabinol), and Cesamet (nabilone). Existing clinical data from Sativex and Epidolex found diarrhoea, fatigue, somnolence, vomiting and pyrexia to be common adverse events but otherwise noted the drugs to be well tolerated [54, 56, 57]. This provides some insight into how these drugs might be tolerated if they were to be delivered following SCI as part of human clinical trials.

Hama et al. similarly investigated whether combinations of currently licensed drugs (i.e. drug repurposing) could be used to treat pain after SCI [37]. In particular combinations involving acetaminophen, which is proposed to act in part by blocking cellular uptake of anandamide [58, 59]. Combinations of acetaminophen with gabapentin or morphine displayed synergy which was attenuated using CB receptor antagonists implying involvement of the CB receptor activation [37]. The benefits of using such existing drugs include the avoidance of long periods of drug development and licensing as well as the ability to use lower doses to avoid side effects that may otherwise limit the use of such drugs.

Whilst these drugs represent promising opportunities, animal to human translation can be unpredictable and key research questions remain. Firstly, consensus must be reached regarding the optimal cannabinoid drug, dose, and determine what would constitute a clinically acceptable route of administration (particularly if repeated dosing is required). The current preclinical evidence base remains heterogenous, and it is difficult to reconcile inconsistencies between studies when variables such as the model of SCI, timing of drug administration, outcome assessment tool and timing of outcome assessment differ from study to study. Similar variation has been reported in a systematic review of the literature on the effects of cannabinoids in patients who have suffered SCI [23]. We echo their call for appropriately powered, randomised controlled studies with standardised outcome measures, which conform to Consolidated Standards of Reporting Trials, to increase the amount of good quality evidence on this topic. Secondly, there is currently little pre-clinical or clinical literature exploring the long-term effects of chronic cannabinoid use following SCI [23], and this would be worthy of further study. Of the studies included in this review, none explored the effects of administering cannabinoids beyond 10 weeks following SCI. Cannabinoids have been associated with addiction, cognitive decline, sedation, and psychotic disorders [60]. Given that SCI patients may need to use cannabinoids for a number of years, long-term longitudinal studies monitoring the incidence of such side-effects will be required if cannabinoids are to be considered in the management of these patients.