A systematic literature search was performed in the PubMed, Embase, Web of Science, Cochrane Library, China National Knowledge Infrastructure, WanFang data, Vip Journal Integration Platform, and SinoMed databases from their inception date to May 2023 (Sng et al., 2022; Zhou et al., 2022). Key words included “spinal cord injuries,” “spinal cord injury,” “spinal cord contusion,” “spinal cord compression,” “spinal cord trauma,” “astaxanthine,” “astaxanthin,” “E-astaxanthin’, “rats,” “rat,” and “murinae.” The bibliographies of the included studies were also screened to identify any additional relevant studies.

Two reviewers independently screened the title, abstracts, and full texts of the retrieved articles. Discrepancies were be resolved by consensus after discussion with a third reviewer. Studies that fulfilled all of the following criteria were included.

Two investigators independently extracted the information of the included studies. The following data was extracted: the name of the first author, publication year, animal strain and gender, animal age and weight, number of animals per group, method used to induce SCI, injury level, AST administration (including dosage, method, timing, and times), and measured outcomes. In studies with multiple intervention arms, only data from the AST and negative control groups were extracted for analysis. If the data were represented as a graph without a numerical value, GetData Graph Digitizer 2.26¹ was used to interpret graph data (Tian et al., 2020). If data were missing, we would contact the original authors and await a response for two weeks. In cases where there was no response, the related data would be included for qualitative analyses only.

The methodological quality of studies was evaluated using the systematic Review Center for Laboratory animal Experimentation’s Risk of Bias (SYRCLE’s RoB) tool (Hooijmans et al., 2014). This tool includes 10 items for the assessment of selection bias, performance bias, detection bias, attrition bias, reporting bias, and other biases. The items were assessed with a judgment of “yes,” “no,” or “unclear,” indicating a low risk, a high risk, or that the risk of bias was not clear, respectively.

The quality of evidence in the included studies was evaluated according to the previously described approach (Charalambous et al., 2014, 2016). In this tool, the evidence quality of experimental laboratory animal studies (ELAS) was graded at three levels. The blinded randomized ELAS (bRELAS) and non-blinded RELAS (nbRELAS) were considered to provide higher quality evidence, followed by non-randomized RELAS (nRELAS) and uncontrolled ELAS (UELAS), then case series and reports. Subsequently, for correcting the confidence rating for single studies in each group, three domains were assessed for indicating the strengths or weaknesses of each study, involving: (1) sample sizes; (2) subject enrolment quality; and (3) the overall risk of bias according to SYRCLE’s RoB tool. Collectively, bRELAS with large sample sizes, rigorous criteria in subject enrolment, and low overall risk of bias were considered to produce the highest quality of evidence.

Data from included studies were summarized and analyzed by the RevMan 5.3 software (provided by the Cochrane Collaboration). Two-sided p values of less than 0.05 were considered statistically significant. The mean, standard deviation, and the sample size of animals in each group were extracted for comparisons. In accordance with the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions, for studies with multiple intervention groups, those groups were combined to enable a single pair-wise comparison (Higgins and Green, 2011). For outcomes with the same unit, the results were described as the weighted mean difference (WMD); otherwise, standardized MD (SMD) was employed. The 95% confidence intervals (CIs) were calculated for both types of outcomes. Heterogeneity between studies was evaluated by p value in the chi-squared test and Cochrane’s I² (Higgins and Green, 2011). Heterogeneity is presumed in the event that the p value is less than 0.10, and is considered to be high when the I² value is more than 50%. The random effects model was used because it provides the expected average of all samples of individual true effect sizes. Linear graphs were constructed by GraphPad Prism (version 5.01) software to highlight the dynamic WMDs of BBB scores and dynamic BBB score improvements in both groups.

Repeated updates (sequential multiplicity) and sparse data increase the risk of random error. The trial sequential analysis (TSA) adjusts the random-error risk and provides a conservative estimation for the required number of study subjects in order to carry out a definitive meta-analysis. Thus, we also conducted the trial sequential analysis method based on the primary outcome to confirm the results of meta-analyses. TSA 0.9.5.10 beta software (downloaded from https://ctu.dk/tsa/downloads) was used to construct a cumulative Z-curve by including each study according to the order of their publication year, and to estimate the required information size. A firm conclusion could be drawn without further research in cases where the cumulative Z-curve either crossed the monitoring boundary or required information size. The TSA was performed with a desired power of 90% and an alpha error of 5%.

Among the 559 studies identified with the search strategy, 44 duplicates were eliminated and another 503 studies were excluded by performing title and abstract screening. Of the 12 papers selected for full-text screening, one reference was excluded due to data duplication (Ren et al., 2018), and another one did not report an outcome that met the inclusion criteria (Fakhri et al., 2022). Ten studies were, therefore, included in this systematic review and meta-analysis (Figure 1) (Chen, 2014; Masoudi et al., 2017; Fakhri et al., 2018, 2019; Ren et al., 2019; Mohaghegh et al., 2020; Li et al., 2021; Masoudi et al., 2021; Abbaszadeh et al., 2022, 2023).

Of the 10 publications, eight articles were presented in English and the remaining two in Chinese (Chen, 2014; Ren et al., 2019). Seven of them used Wistar rats; the other three selected Sprague–Dawley rats (Chen, 2014; Ren et al., 2019; Li et al., 2021). Six studies induced SCI using a weight-drop impactor, and the remaining four used an aneurysm clip compression method (Fakhri et al., 2018, 2019; Abbaszadeh et al., 2022, 2023). All studies established SCI at the T8–T12 level. Administration of AST by intrathecal injection was employed in seven studies, with a dose of 0.005 mg/kg. The other studies used the intragastrical administration route, with an intervention dose ranging from 35 to 75 mg/kg (Chen, 2014; Ren et al., 2019; Li et al., 2021). All studies administrated AST immediately post injury, and most studies except three (Chen, 2014; Ren et al., 2019; Li et al., 2021) employed a single dose of AST (Table 1).

According to the SYRCLE’s RoB tool, the mean number of reported items in the included studies was 4.1. Four studies adequately reported 50% of the details in the checklist (Masoudi et al., 2017; Fakhri et al., 2018; Ren et al., 2019; Li et al., 2021). All the included studies well described the items of “baseline characteristics” and “selective outcome reporting.” Six studies reported a design of outcome assessor blinding (Masoudi et al., 2017; Fakhri et al., 2018, 2019; Ren et al., 2019; Mohaghegh et al., 2020; Li et al., 2021), and three studies well addressed the item of incomplete outcome data (Masoudi et al., 2017; Fakhri et al., 2018; Masoudi et al., 2021). However, none or few studies reported the details of “sequence generation,” “allocation concealment,” “random housing,” “investigator blinding,” and “random outcome assessment.” Additionally, no other sources of bias such as the pooling of drugs, dropouts, unit of analysis errors, or design-specific bias were identified in the included studies (Table 2).

Overall, the evidence quality of the included studies was not high. Although all studies belonged to bRELAS or nbRELAS, three of them reported a design with a small sample size (Chen, 2014; Ren et al., 2019; Li et al., 2021) and another four did not adequately address the details of the subject enrolment quality (Fakhri et al., 2018, 2019; Abbaszadeh et al., 2022, 2023). Moreover, based on SYRCLE’s RoB tool, a potential risk of bias was widely presented in the included studies (Table 3).

A total of seven studies used the BBB scale to indicate the neurological function recovery in rat models of SCI (Chen, 2014; Masoudi et al., 2017; Fakhri et al., 2019; Ren et al., 2019; Mohaghegh et al., 2020; Li et al., 2021). The meta-analyses and dynamic BBB scores in both groups suggested that BBB scores were increased in rats treated with AST compared with controls, and WMDs between those two groups exhibited a gradual upward trend from days 7 (six studies, n = 88, WMD = 2.85, 95% CI = 1.83 to 3.87, p < 0.00001) to days 28 (five studies, n = 76, WMD = 6.42, 95% CI = 4.29 to 8.55, p < 0.00001) after treatment (Figures 2A–D and Table 4).

Subgroup analyses indicated that the improvement of BBB scores was associated with injury model, but did not differ between distinct administration routes (Figures 3A,B and Supplementary Table S1). Recent studies suggest that locomotor function and pathological changes may differ between rats based on gender, injury level, and timing of administration, and these factors were consistent in our included studies (Wilcox et al., 2017). Despite the different AST dosage and number of administrations among the studies, changes of those variables were highly attributed to the administration route (Chen, 2014; Ren et al., 2019; Li et al., 2021). Thus, subgroup analyses concerning those variables was not conducted in our study.

By using the luxol fast blue method, five studies described the spared white matter area in the form of residual tissue/cross-sectional area ratio (Masoudi et al., 2017; Fakhri et al., 2018, 2019; Mohaghegh et al., 2020; Masoudi et al., 2021). The pooled results of spared white matter area showed an increased percentage of spared white matter in AST-treated rats versus controls (four studies, n = 28, WMD = 0.08, 95% CI = 0.02 to 0.14, p = 0.007) at days 7 after SCI (Figure 4A and Table 4).

Meanwhile, five studies reported the number of motor neurons in the ventral horn area of the spinal cord by counting the Nissl positive cells (Masoudi et al., 2017; Fakhri et al., 2018, 2019; Mohaghegh et al., 2020; Masoudi et al., 2021). Meta-analysis regarding this outcome revealed an improved neuron loss in lesion areas in rats with AST intervention (Figure 4B and Table 4; four studies, n = 28, WMD = 7.96, 95% CI = 4.35 to 11.57, p < 0.0001).

Four studies reported the antioxidant role of AST through detecting the MDA or SOD levels in SCI rats (Chen, 2014; Li et al., 2021; Masoudi et al., 2021; Abbaszadeh et al., 2022). The units for each outcome were slightly different in those studies. Meta-analysis suggested an enhancement of SOD expression (four studies, n = 40, SMD = 5.29, 95% CI = 2.12 to 8.46, p = 0.001) and reduced MDA levels (four studies, n = 40, SMD = −4.20, 95% CI = −6.88 to-1.51, p = 0.002) in rats with AST treatment compared with control ones (Figure 5 and Table 4).

Data of BBB scores from days 14 to 28 after SCI were included in TSA to present the dynamic change of results. At days 14, the cumulative Z-curve crossed the monitoring boundaries, indicating although the cumulative sample size did not reach the expected value, a firm conclusion that AST could promote the recovery of locomotor function was obtained in advance (Figure. 6A). Interestingly, at the time points of days 21 and days 28, the cumulative Z-curve crossed both the conventional and monitoring boundaries, and the cumulative sample size reached the expected value (Figures 6B,C); thus, the present data firmly proved that AST can provide a beneficial role for the neurofunctional recovery of rats.

Sensitivity analyses were performed by excluding either studies without outcome assessor blinding, small-sample-sized studies, or all single studies. A moderate or high heterogeneity was remained after application of all exclusion strategies. Furthermore, the improvement of BBB scores with AST remained largely unchanged (Supplementary Table S2), indicating a robust result of the analyses of BBB scores.

The proposed mechanisms behind AST in the included studies are presented in Table 5. In general, AST exert roles of regulating oxidative stress, suppressing neuroinflammation, relieving neuron loss, and modulating autophagy in treating SCI. The N-methyl-D-aspartate receptors (NMDAR), protein kinase B (AKT), mitogen-activated protein kinase (MAPK), cleaved caspase-3, and light chain 3 (LC3) signaling may be involved in those effects of AST.

This review extracted data from 10 studies that compared AST with placebo controls. Although most of these studies were of moderate-to-high methodological quality, the quality of evidence in these studies was not high.

The meta-analyses showed an increase in BBB scores in rats with AST treatment, and the WMDs between AST and control groups displayed a gradual upward trend over time. There were no statistical differences in locomotor recovery with respect to administration route; however, there were differences related to the injury type. Importantly, TSA firmly proved that AST can promote locomotor recovery of rats with SCI.

Compared with controls, rats in the AST groups exhibited an increased percentage of spared white matter in lesion areas. The meta-analysis concerning the motor neuron numbers confirmed the role of AST on tissue sparing. Moreover, meta-analyses revealed that AST could enhance SOD levels and downregulated the MDA expression in rats subjected to SCI, indicating an antioxidant effects of AST for SCI.

As anticipated, we observed a substantial between-studies heterogeneity on treatment effects, which may reduce the credibility of the results from the analyses to a certain extent. The differences in animal characteristics, injury types, administration details, statistical methods, and methodological quality between studies may serve as potential contributors to the observed heterogeneity. Our subgroup analyses with respect to distinct injury types effectively reduced the heterogeneity in both subgroups, and the Cochrane’s I² in the subgroup of compression were stably maintained at 0% from days 7 to 28 after SCI. However, it is important to keep in mind that subgroup analysis will only be hypothesis-generating and serves as an initial project to shed light on this issue. Sensitive analysis is also beneficial for exploring the sources of heterogeneity, while our sensitivity analyses using the above exclusion strategies did not successfully reduce the treatment effect heterogeneity. Additionally, three included studies administrated AST by intragastrical route; the remaining studies used intrathecal injection. Following oral absorption, AST has been shown to gradually convert into diversified metabolites (Kistler et al., 2002). The different metabolites of AST between intragastrical and intrathecal administration routes may produce various pharmacologic roles, leading to the heterogeneity in treatment effects. Collectively, a number of variables may be involved in the high heterogeneity among studies; distinct injury types and varied administration routes appear to be important factors modifying the heterogeneity in our meta-analyses. Due to the high heterogeneity among studies, the results in this review need to be interpreted cautiously.

The injury mechanism is an important factor that may impact the efficacy of the treatment (Abdullahi et al., 2017). Our subgroup analyses indicated a superior locomotor recovery in compression SCI models, implying that AST may be more suitable for SCI caused by compression injury. The distinct pathophysiological mechanisms behind different injury types may account for the changed efficacy of AST treatment. Although compression and contusion SCI show a variety of similarities in the pathological changes, the compression model has its own features due to the prolonged compression of the spinal cord (Abdullahi et al., 2017). The prolonged compression can cause a decrease in blood flow, and a rapid drop in glucose level and oxygen supply in the compression site, leading to gradual ischemia damage of the spinal cord (Ueta et al., 1992; Okon et al., 2013). Subsequently, reperfusion events result in an enlarged area of damage. The ischemia and reperfusion courses rupture the oxygen or glucose supply at the injury site, and serve as typical events causing excessive oxidative stress (Ling et al., 2021). Thus, oxidative stress may act as a more prominent element in compression SCIs compared with contusion ones. As described above, AST stands out among its chemical family for its excellent antioxidant activities (Alugoju et al., 2022). Thus, we hypothesized that the more prominent oxidative stress in compression injuries may result in the superior applicability of AST on this injury type. However, this hypothesis needs further studies for verification.

AST is approved by the Food and Drug Administration and European Food Safety Authority as a food supplement. The adverse effects of AST are rare, involving the alteration of skin pigmentation, increased frequency in bowel movements, and coloration change of feces (Brendler and Williamson, 2019). Although excessive consumption of AST is associated with potential side effects, a large range between the median effective dose and the median lethal dose has been indicated in animal experiments. For instance, oral administration of AST at doses of 5–75 mg/kg all exerted neuroprotective roles in rats without death events caused by AST (Zhang et al., 2014; Liu H. et al., 2020); even at consumption levels of 20 g/kg AST, the majority of the treated animals exhibited a normal condition in clinical parameters (Niu et al., 2020). Meanwhile, only minor clinical differences in blood indexes and slight hepatotoxicity were observed in rats continuously treated by 1,000 mg/kg AST for two years (Edwards et al., 2016). In human trials, antioxidant and anti-inflammatory benefits were achieved with as little as 2 mg/day, and no serious adverse events were observed with a high dose of 45 mg/day for four weeks (Kajita et al., 2010; Park et al., 2010). A single dose of 100 mg AST was once applied in humans to investigate the pharmacokinetics of AST, but no serious events were presented in this study (Østerlie et al., 2000). Furthermore, the European Food Safety Authority concluded that a daily intake of 8 mg AST from food supplements is safe for adults even in combination with the high exposure estimate to AST from the background diet (Turck et al., 2020). However, the safety of AST for patients with SCI remains to be confirmed due to a lack of related human clinical trials.

Based on the proposed mechanisms in the included studies and related evidence, the potential mechanisms of AST are summarized as follows (Figure 7).

To our knowledge, this is the first meta-analysis quantitatively analyzing the efficacy of AST in SCI. This review primarily focused on the dynamic changes of BBB scores following AST intervention, and dealt with the tissue protective and antioxidant effects of AST in SCI. TSA confirmed the results from our meta-analyses and firmly supported the beneficial effects of AST. Given the significance of efficacy, safety, pharmacokinetics, and pharmacological mechanisms for effective clinical translation, we employed a combination of a systematic and traditional review to comprehensively analyze all of these parameters. The findings in this review suggest that AST is a promising natural neuroprotective agent for SCI treatment.

There are several limitations to our study. First, the included studies were still limited although an extensive search strategy was applied, which may challenge the results and meaning of our meta-analysis. However, analyzing the dynamic changes of BBB scores and using the multimodal assessments (assessor evaluation, biochemistry indicators, and histopathological examinations) were beneficial for indicating the actual effects of AST. Additionally, the TSA analyses further confirmed the results from our meta-analyses. Second, substantial heterogeneity between studies was presented in our review. The heterogeneity appears to be linked to the differences in injury type, administration route, and other above factors, and may lead to improper understanding of the efficacy of AST. However, the random effect model widely used in our analyses reduced the risk of producing incorrect results to a certain degree. Meanwhile, the robust results in sensitivity analyses and consistent result tendencies among the studies enhanced the strength of evidence regarding AST efficacy. Third, systematic subgroup analyses are important for investigating the efficacy of AST in distinct scenarios. Due to the variable distribution in studies and the limited number of included studies, subgroup analyses with respect to rat gender, injury level, timing, or dose of administration were not conducted. Fourth, the SYRCLE’s RoB tool showed a potential risk of bias within the included studies. It is a common problem for animal studies, but may have produced unreliable results in the included studies and our meta-analyses.