A comprehensive search was performed to identify experimental studies evaluating the effects of RRC for ischemia stroke from databases, including PubMed, embase, CBM, Web of Science, National Knowledge Infrastructure (CNKI), Wanfangdatabase and VIP information database. All searches were electronically searched from the inception up to Oct 2021. Studies about assessing the effectiveness of RRC for ischemic stroke in animals were identified. Our literature search strategy was as following: (Rhodiola OR Rhodiola rosea OR Roseroot OR Rhodioloside OR Salidroside) AND (Ischemic stroke OR Cerebral ischemic injury OR Cerebral infarction OR Brain infraction).

Experimental studies evaluating the effect of RRC for ischemic stroke were selected. Two authors independently screened the titles and/or abstracts according to the search strategy. Then, we assessed the full-text articles for eligibility. Studies were included if they met the following criteria: 1) Animal models were established for ischemic stroke; 2) RRC as monotherapy was administrated in the experimental group, regardless of its mode, dosage, and frequency. 3) The primary measured outcomes were neural functional deficit score (NFS), infarct volume (IV), brain water content, cell viability, apoptotic cells, terminal deoxynucleotidyl transferase (TdT)-mediated dUTP-biotin nick end labeling (TUNEL)-positive cells, B-cell lymphoma-2 (Bcl-2) level and tumor necrosis factor-α (TNF-α) level. The secondary outcome measures were mechanisms of RRC for ischemic stroke; and 4) The control group received vehicle or no adjunct intervention.

The prespecified exclusion criteria were as follows: 1) the targeting disease was not ischemic stroke; 2) RRC were used as combination; 3) the article was a clinical or in vitro study; 4) the study was a case report, clinical trial, review, abstract, comment, editorial, duplicate publication or in vitro study, and 5) lack of the control group.

Two independent reviewers assessed the articles and the following details were extracted: 1) the first author, publication year; 2) individual data from each study, including animal species, gender, samples for individual comparison, and weight; 3) type of animal model; 4) type of anesthetic; 5) intervention characteristics from both treatment and control groups, including drug, timing for initial treatment, dosage, mode, and frequency; 6) outcome measures and its corresponding pvalue. For each comparison, the mean value and standard deviation from each treatment and control group of every study were extracted. If the data were demonstrated graphically, we tried to contact the author for further information or digital ruler software was applied. Otherwise we only performed qualitative analysis. The data of highest dose was selected when the treatment group included various doses of the target drug. The result of the last time point was included when the data were expressed at different times.

Two authors independently assessed the methodological quality of the included articles according to the Collaborative Approach to Meta-Analysis and Review of Animal Data from Experimental Studies (CAMARADES) 10-item checklist (Sena et al., 2007): 1) peer-reviewed publication; 2) statements of temperature control; 3) randomization to treatment or control group; 4) blinded induction of model; 5) blinded assessment of outcome; 6) use of anesthetic without significant intrinsic neuroprotective activity; 7) appropriate animal model; 8) sample size calculation; 9) compliance with animal welfare regulations; and 10) declaration of potential conflict of interests. Each study was given an aggregate quality score based on one-point awarding for each item. Discrepancies were resolved by discussion or consultation with corresponding author.

The pooled analyses were performed using RevMan 5.3 software. All outcome measures were considered as continuous data. To estimate the effect of RRC on ischemic stroke, the random effects model and standard mean difference (SMD) with 95% confidence intervals (CIs) were calculated. Heterogeneity among individual studies was assessed via I ² statistics test. If probability value was less than 0.05, the difference was considered statistically significant.

We identified 1774 potentially relevant articles from seven databases. After removal of duplicates and irrelevant articles, 279 records remained. By reviewing titles and abstracts, 137 studies were excluded because they were case reports, abstracts, comments, clinical trials, editorials, letters and review articles. After going through the remaining full-text articles, 120 articles were excluded for at least one of following reasons: 1) the article was not a research about ischemic stroke; 2) not an in vivo study; 3) the intervention was a combination of RRC with potential effect on ischemic stroke; 4) the study did not access the effects of RRC on the animal model of ischemic stroke; 5) no control group. Finally, 15 eligible studies (Chen et al., 2009; Shi et al., 2012; Chen et al., 2014; Yu et al., 2014; Han et al., 2015; Atochin et al., 2016; Chen et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) involving 345 animals were identified ( Figure 1).

The basic characteristics of the eligible studies are summarized in Table 1. Fifteen studies included were published between 2009 and 2021. Among them, 13 studies were conducted in English and two studies (Chen et al., 2009; Chen et al., 2014) were Chinese paper. For animal species, 12 studies used rats including Sprague-Dawley (SD) rats (n = 144) and Wistar rats (n = 134) as animal models. Three studies used C57BL/6 mice (n = 67). The weight of rats ranged from 190 to 320 g, and the weight of mice ranged from 18 to 23 g. Cerebral ischemic injury in the included studies was induced by temporary middle cerebral artery occlusion (MCAO) in which ischemic time varied from 8 to 180 min (Chen et al., 2009; Shi et al., 2012; Chen et al., 2014; Yu et al., 2014; Han et al., 2015; Atochin et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021), and permanent MCAO (Chen et al., 2016). For anesthesia chosen in experiments, four studies (Shi et al., 2012; Han et al., 2015; Atochin et al., 2016; Zhang et al., 2020) used chloral hydrate, three studies (Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019) used isoflurane, four studies (Chen et al., 2016; Zhang et al., 2018; Li et al., 2020; Dong et al., 2021) used sodium pentobarbital, and the remaining four studies (Chen et al., 2009; Chen et al., 2014; Yu et al., 2014; Liu et al., 2018) did not report it. Ten studies utilized a dose gradient of RRC: one study (Atochin et al., 2016) used 5, 10 and 20 mg kg⁻¹ intravenously, one study (Chen et al., 2016) used 20, and 40 mg kg⁻¹ orally, one study (Han et al., 2015) used 15, and 30 mg kg⁻¹ intraperitoneally, one study (Liu et al., 2018) used 2.5, 5, 10 and 20 mg kg⁻¹ intravenously, two studies (Zhang et al., 2018; Li et al., 2020) used 25, 50 and 100 mg kg⁻¹ intraperitoneally, one study (Zhang et al., 2019) used 20, 50 and 100 mg kg⁻¹ intraperitoneally, one study (Zhong et al., 2019) used 20, 40, and 80 mg kg⁻¹ intraperitoneally, one study (Zuo et al., 2018) used 10, 20, and 40 mg kg⁻¹ intraperitoneally, and one study (Zhang et al., 2020) used 12 and 48 mg kg⁻¹ through gavage. Six studies (Shi et al., 2012; Chen et al., 2014; Atochin et al., 2016; Chen et al., 2016; Zuo et al., 2018; Zhong et al., 2019) administrated RRC before stroke; Seven studies (Shi et al., 2012; Yu et al., 2014; Chen et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zhang et al., 2020; Dong et al., 2021) administrated RRC after stroke; and two studies (Han et al., 2015; Li et al., 2020) administrated RRC before and after stroke. In the control group, twelve studies applied same volume of normal saline, one study (Liu et al., 2018) applied phosphate-buffered saline (PBS), one study (Zhang et al., 2020) applied distilled water and the remaining one study (Chen et al., 2015) did not clearly mentioned.

Nine studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Chen et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Li et al., 2020) adopted IV as outcome measurements; fourteen studies (Chen et al., 2009; Shi et al., 2012; Chen et al., 2014; Yu et al., 2014; Han et al., 2015; Atochin et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) used NFS as outcome measurements, among them eight studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Li et al., 2020) adopted both above two outcome measurements. However, the methods used to identify IV were different; eight studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Chen et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Li et al., 2020) used TTC staining and one study (Zhang et al., 2019) used MRI scan. The standards of NFS were diverse: seven studies (Chen et al., 2009; Shi et al., 2012; Chen et al., 2014; Han et al., 2015; Zhang et al., 2018; Zhang et al., 2019; Li et al., 2020) adopted Zea Longa (ZL) score; one study (Atochin et al., 2016) used McGraw scale; one studies (Yu et al., 2014) used modified neurological severity score (mNSS); 1study (Zuo et al., 2018) used Ludmila Belayev test, For grid test, Beam walk test, and Wire grip test; one study (Zhong et al., 2019) used Balance beam test, foot fault test and mNSS; one study (Dong et al., 2021) used rotarod tests, one study (Liu et al., 2018) used rotarod tests and mNSS, and one study (Zhang et al., 2020) used the method described by Brambrink et al. (2006). The included studies also reported TUNEL-positive cells, Caspase-3, Bcl-2, TNF-α, IL-1, IL-2, IL-6, IL-8, malondialdehyde (MDA), superoxide dismutase (SOD), glutathione (GSH), glutathione-S-transferase (GST), Evans blue content, MBP, MAP2, MAO, ROS, LDH, GLUT3, p53, GFAP, DA, HVA, DOPAC, cyclin D1, CDK4, p27Kip1, cell viability rate, and apoptotic cells.

The quality of the 15 included studies was evaluated and ranged from 1/10 to 7/10 with the average of 5.53 in Table 2. Of which, four studies (Zuo et al., 2018; Zhang et al., 2019; Zhang et al., 2020; Dong et al., 2021) obtained seven points, seven studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Chen et al., 2016; Liu et al., 2018; Zhong et al., 2019; Li et al., 2020) obtained six points, one study (Atochin et al., 2016) obtained five points, one studies (Zhang et al., 2018) obtained four points, one studies (Chen et al., 2009) obtained three points, and the remaining one study (Liu et al., 2018) obtained one point. All studies were published in peer-reviewed journals. Twelve studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Atochin et al., 2016; Chen et al., 2016; Liu et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) described control of the room temperature. Ten studies (Chen et al., 2009; Shi et al., 2012; Yu et al., 2014; Chen et al., 2016; Liu et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) declared that they had random allocation to treatment and control groups. Twelve studies (Chen et al., 2009; Shi et al., 2012; Han et al., 2015; Atochin et al., 2016; Chen et al., 2016; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) used anesthetic without significant intrinsic vascular protection activity. Thirteen studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Atochin et al., 2016; Chen et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) mentioned compliance with animal welfare regulations. Nine studies (Shi et al., 2012; Yu et al., 2014; Han et al., 2015; Liu et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Zhang et al., 2020; Dong et al., 2021) declared that the model establishment and outcome assessment were conducted in double-blind trial. Twelve studies (Yu et al., 2014; Han et al., 2015; Atochin et al., 2016; Chen et al., 2016; Liu et al., 2018; Zhang et al., 2018; Zuo et al., 2018; Zhang et al., 2019; Zhong et al., 2019; Li et al., 2020; Zhang et al., 2020; Dong et al., 2021) contained statements on potential conflict of interests. There was no study calculating sample size in the animal experiment and blinded assessment of outcome. No study used animals with relevant comorbidities.

To our knowledge, it is the first preclinical systematic review to assess the efficacy of RRC for cerebral ischemic stroke. In the present study, 15 studies with 345 animals showed that RRC significantly improved NFS and reduced IV in cerebral ischemia animal models. Thus, RRC exerted the potential neuroprotective function for ischemic stroke, mainly through anti-inflammatory, anti-apoptosis, and anti-oxidative and alleviating the pathological BBB damage. However, given methodological weaknesses, the overall available evidence from the present study should be interpreted cautiously. Thus, the conclusions in the present study should be partially treated with caution.

There are several limitations in the primary studies. Firstly, only Chinese and English literatures were searched, which may cause selection bias as studies published in other languages were absent (Zhang et al., 2019). Secondly, no study had used an animal with co-morbidities, such as hypertension, diabetes or hyperlipidemia (Heusch, 2017), which would be more relevant models for human pathology (Guyatt et al., 2011). Thirdly, the studies had methodological deficiencies. None of these studies reported the blindness of ischemia induction, allocation concealment, randomization to treatment group or control group and sample size calculation, which are the core criteria of study design. Thus the analysis may result in overestimated effect size (Ospina et al., 2005; Higgins and Green, 2012). Thereby, the results in the present study should be interpreted with caution.

The damage inflicted on the neuron during ischemic stroke is a complex process, involving multiple factors. The main mechanisms of injury are oxidative and nitrative stress, inflammation, apoptosis, ion imbalance, calcium overload, and energy depletion (Terasaki et al., 2014; Jayaraj et al., 2019), leading to neurovascular unit dysfunction and neurologic deficitse. Thus, neuroprotective drugs generally work through one or combined aspects of the above targets. The present study showed RRC could exert potential neuroprotective effects in experimental for ischemic stroke indicating that RRC are candidates for ischemic stroke treatment and can be used for further clinical trials. The possible mechanisms of RRC for cerebral ischemia injury are summarized as follows: 1) alleviating the pathological BBB damage; 2) repressing lipid peroxidation; 3) antioxidant through increasing the activity of SOD, HO-1, Nrf2, GSH-Px and GST and decreasing the concentration of MDA and ROS; 4) anti-inflammatory via decreasing the expression of proinflammatory cytokines such as TNF-α, IL-1β, IL-1, IL-2 and IL-6; 5) anti-apoptotic via increasing the levels of Bcl-2, decreasing the levels of Bax, caspase3, C-Fos, GFAP, p53, decreasing the activity of LDH and reducing TUNEL positive cells; 6) neuroprotective effect via regulating BDNK-mediated PI3K/Akt pathway, through calpain1/PKA/CREB pathway and through modulating monoamine metabolism; 7) inhibiting reactive astrogliosis and glial scar formation, probably through Akt/GSK-3β pathway. To summarize, the possible mechanisms of RRC for ischemic stroke are through antioxidant, lipid peroxidation, anti-apoptosis, anti-inflammatory, improving blood vessel endothelium differentiation, and cerebral metabolism. A recent review (Sun et al., 2020) illustrated that Rhodiola rosea L. and its components, particularly salidroside has strong antioxidant activity through regulating mitochondrial biogenesis, repressing ROS production, increasing the activity of the antioxidant enzymes (such as GSH-Px and SOD), and via various signaling pathways (AMPK, PI3K/Akt, Mitochondria-dependent, Nrf2). In addition, another review (Pu et al., 2020) showed that Rhodiola rosea L. and its compounds have immune-regulation effects through some inflammatory mediators, such as IL-6, TNFα, IL-1β, and NO, and signaling pathways, such as NF-κ B, AP-1, and STAT3. In the present study, the mechanisms are consistent with the evidences.

Preclinical animal research plays a critical role in human diseases understanding (Murphy and Murphy, 2010). However, original preclinical research is often conducted with a poor methodological quality, which is considered as a hindrance to the translation of animal research into effective preclinical drug treatments for human disease (Baginskait, 2012; Moher et al., 2015). The systematic review can dentify defects in study design, integrate preclinical evidence and guide potential clinical translation (Macleod et al., 2005; De Vries et al., 2014). In the present analysis, the average CAMARADES score of the included studies ranged from 1/10 to 7/10. The main flaws are lacking of sample size calculation, poor blinding in model induction and outcome assessment. Inadequate sample size can miss the real intervention effect in an experiment, while excessive sample size will result in wasting resources and raising animal ethical issues (Arifin and Zahiruddin, 2017; Chen et al., 2019). Poor blinding in outcome assessment could result in a 27% overestimation of the mean reported effect size (Holman et al., 2015) Additionally, all the animal experiments are conducted in healthy animals which lack the comorbidities, such as diabetes, hypertension and hyperlipidemia. Reporting guidelines set detailed predetermined standards to make biomedical research report more complete and transparent, and enhancing their value in scientific exploration and clinical practice. The Animal Research: Reporting of In Vivo Experiments (ARRIVE) (Kilkenny et al., 2012) is a reporting guidelines, which are organized into twenty sections, providing recommendations on Introduction, Methods, Results, and Discussion. The ARRIVE guidelines were recommended to be utilized when designing and reporting animal research on RRC for ischemic stroke which can provide guidance on the complete and transparent reporting of in vivo animal researches, helping to improve the quality of further researches (Karp et al., 2015). Thus, we suggest that further animal researches should follow up the reporting guidelines, increasing the value of clinical trials and further application. Furthermore, the following factors need to be considered: 1) method by which sample size was determined should be appropriately detailed; 2) experimental animals have relevant comorbidities, which are like human pathology under the clinical conditions; 3) primary outcome should be closer to clinical practice.