Preclinical Evidence of Berberine on Non-Alcoholic Fatty Liver Disease: A Systematic Review and Meta-Analysis of Animal Studies

As lifestyle and diet structure impact our health, non-alcoholic fatty liver disease (NAFLD) is prevalent all over the world. Some phytomedicines containing berberine (BBR) have been extensively used for centuries in Ayurvedic and traditional Chinese medicine. The goal of this systematic review is to investigate the preclinical evidence of BBR on NAFLD models. The following relevant databases, including Web of Science, PubMed, the Cochrane Library, and Embase, were retrieved from inception to May 2021. The content involved BBR on different animal models for the treatment of NAFLD. The SYstematic Review Center for Laboratory animal Experimentation (SYRCLE) Animal Experiment Bias Risk Assessment Tool was used to assess the methodological quality and RevMan 5.4 software was used to conduct the meta-analysis based on the Cochrane tool. A total of 31 studies involving 566 animals were included, of which five models and five animal breeds were reported. The results showed that TC, TG, ALT, AST, HDL-C, LDL-C, FBG, FINS, and FFA in the group treated with BBR were significantly restored compared with those in the model group. HOMA-IR had a significant downward trend, but the result was not significantly different (P = 0.08). The subgroup analysis of the different models and different animal breeds indicated that BBR could ameliorate the aforementioned indicator levels, although some results showed no significant difference. Finally, we summarized the molecular mechanisms by which berberine regulated NAFLD/NASH, mainly focusing on activating the AMPK pathway, improving insulin sensitivity and glucose metabolism, regulating mitochondrial function, reducing inflammation and oxidative stress, regulating cell death and ER stress, reducing DNA methylation, and regulating intestinal microenvironment and neurotoxicity. The preclinical evidence suggested that BBR might be an effective and promising drug for treating NAFLD/NASH. In addition, further studies with more well-designed researches are needed to confirm this conclusion.


INTRODUCTION
Obesity with extensive metabolic regulation disorders and excessive fat accumulation is a major risk factor for type 2 diabetes, various kinds of cancer, cardiovascular disease, and non-alcoholic fatty liver disease (NAFLD) . As the lifestyle and diet structure translate, obesity is prevalent all over the world. As high as 30 percent of the population in western developed countries suffer from NAFLD (Williams, 2006;Angulo, 2007) and the prevalence is on the rise in developing countries (close to 10%) (Fan et al., 2009). As the aggressive form of NAFLD, non-alcoholic steatohepatitis (NASH) can further develop into hepatic cirrhosis and even hepatocellular cancer without appropriate treatment and is increasingly recognized as the major reason for liver transplantation or end-stage liver disease (Zhang et al., 2016;Goldberg et al., 2017). At present, the pathogenesis of NAFLD/NASH is still not clear enough, but the "two-hit theory" and "multi-parallel hits theory" are widely accepted explanations (Day et al., 1998;Malaguarnera et al., 2009). Excessive hepatic fat accumulation, oxidative stress, insulin resistance or others play promoting roles in developing NAFLD/NASH (Day et al., 1998;Tilg et al., 2010). Therefore, treatment based on these theories may be extremely important in preventing NASH.
Due to the serious impact on the quality of human life and longevity and the high incidence, NAFLD has been receiving more and more attention worldwide. In fact, lifestyle change with exercise and diet intervention to ameliorate insulin sensitivity is the best treatment for NAFLD but limited by compliance and persistence difficulties, emphasizing the urgent need for pharmacotherapy (Scherer et al., 2016;Sumida et al., 2018). Unfortunately, no medication available for NAFLD/NASH is approved. At present, pioglitazone, metformin, vitamin E, and other pharmacotherapies are usually taken in the clinic to treat NAFLD, but they often had limited effects and critical side effects (Wong et al., 2014;Sumida et al., 2018).
Berberine (BBR, Figure 1), a protoberberine isoquinoline alkaloid isolated from many species of plants, such as Berberis aquifolium or Coptis chinensis, which has been extensively used for gut infections and diarrhea for centuries as a part of Ayurvedic and traditional Chinese medicine (Tillhon et al., 2012;Zhu et al., 2019). Over the past decades, it has been reported that BBR has various pharmacological effects of lowering blood glucose, lowering blood lipid, and improving insulin sensitivity and glucose tolerance, and anti-inflammatory, anti-oxidant, and anti-diabetic effects (Kong et al., 2004;Lee et al., 2006;Wang et al., 2011;Zhao et al., 2012;Mo et al., 2014;Sun et al., 2018).
At present, accumulated evidence from the clinical efficacy and pharmacological effect indicated that BBR may have great potential in treating NAFLD. Nevertheless, there are different types of NAFLD/NASH with various manifestations. In order to compensate for the disparity between the clinical efficacy and pharmacological effect or study BBR on a wide range of NAFLD/ NASH and its further mechanism, it is necessary to investigate the preclinical evidence of BBR on various NAFLD/NASH.

Literature Search and Review Strategy
We searched relevant databases, including Web of Science, PubMed, the Cochrane Library, and Embase, from inception to May 2021. The main search terms were combinations of "berberine" and "Non alcoholic fatty liver disease" or "Nonalcoholic fatty liver disease" or "Nonalcoholic fatty liver disease" or "NAFLD" or "non alcoholic steatohepatitis" or "non-alcoholic steatohepatitis" or "nonalcoholic steatohepatitis" or "NASH" or "metabolic associated fatty liver disease" or "MAFLD" in various databases. Moreover, the main goal is to gain documents that match most of the keywords. A preliminary screening of the retrieved literature was performed using Endnote software (Version X9.1, Thompson Reuters) to remove duplicate records. Subsequently, two researchers (Sichen Ren and Ying Wei) independently reviewed the title/abstract related to the topic. A full-text read of potential documents that met the eligibility criteria was conducted. Any disagreements between the two researchers were resolved through negotiation or thirdparty consensus. BBR only; 3) the included studies consist of a model group and BBR experimental group; 4) the primary endpoints were as follows: total cholesterol (TC), triglycerides (TG), alanine aminotransferase (ALT), aspartate aminotransferase (AST), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), fasting blood glucose (FBG), fasting insulin (FINS), homeostasis model assessment-insulin resistance (HOMA-IR), and free fatty acid (FFA) and the secondary endpoints were as follows: animal body weight, liver weight, liver index, NAFLD activity score (NAS), steatosis score, tumor necrosis factor-α (TNF-α), interleukin-6 (IL-6), interleukin-1β (IL-1β), glutathione (GSH), and thiobarbituric acid reactive substances (TBARS); 4) the language is limited to English.
The exclusion criteria of this study were as follows: 1) irrelevant or duplicate publications; 2) no animal experiment included; 3) experimental group without BBR and NAFLD/ NASH or other measures out of BBR and NAFLD/NASH; 4) review and/or meta-analysis; 5) insufficient primary and secondary outcome data; 6) RCT or clinical research; 6) book, thesis, or conference proceedings. Thus, the investigators screened the articles initially by title or abstract based on the inclusion criteria. For example, the relevant literature on BBR with curcumin for the treatment of NAFLD was not included.

Data Extraction
We independently assessed all the included studies and extracted the following data using a standardized data extraction form: 1) the name of the first author and the year of publication; 2) numbers of animals in the model group and BBR experimental group; 3) the model of NAFLD/NASH and BBR dosage; 4) the primary and second outcome measures. If the experimental group of animals in a certain study were measured several times at different time points after BBR administration, all data were extracted and the dominant dose of BBR was included. If details of the data are insufficient, the publishers were contacted for further information. Information of all studies is shown in Supplementary Table S1.

Assessment of Quality
The SYstematic Review Center for Laboratory animal Experimentation (SYRCLE) Animal Experiment Bias Risk Assessment Tool was used for assessing the methodological quality. The tool comprises 10 sections related to selection bias, performance bias, detection bias, attrition bias, reporting bias, and other biases. Moreover, the assessment score of each section was yes (low risk of bias), no (high risk of bias), and unclear (the risk of bias is insufficient to assess from the reported details). These ten terms are as follows: 1) the allocation sequence was fully generated or applied; 2) the groups were similar at baseline; 3) the allocation was fully concealed; 4) the animals were randomly placed during the experiment; 5) the investigators are blinded during the experiment; 6) the animals were selected at random for results; 7) the result assessor was blinded; 8) incomplete data were fully addressed; 9) the studies were free of selective results reporting; (10) the study was free of other problems that could result in high risk of bias (Peters et al., 2006;Hooijmans et al., 2014). Any disagreements between the two researchers during the assessment were resolved through negotiation or third-party consensus.

Statistical Statistics
All data were analyzed using the Review Manager (RevMan, version 5.4). Before determining the pooled effect, I-square (I 2 ) test and Q-test were used to assess heterogeneity between studies. Based on the results of the Q-test and I 2 test, a fixed-effects model with minor heterogeneity (I 2 ≤ 50% or p ≥ 0.1) or a randomeffects model with significant heterogeneity (I 2 > 50% or p < 0.1) were chosen. All outcomes were presented as the std. mean difference (SMD) with a 95% confidence interval (CI) and the significance of pooled effects was determined by the Z-test. p < 0.05 was considered statistically significant.

Study Identification and Selection
As shown in Figure 2, a flow diagram of the system evaluation was constructed. A total of 1,214 relevant studies were identified based on our search strategy. After excluding duplicate literature (1,071 records) and irrelevant documents (34 records), 109 articles were left for further full-text assessment. Ultimately, 31 studies were included for further quality assessment and metaanalysis according to our exclusion criteria.

Study Characteristics
In total, 566 animals (285 in the model group; 281 in the BBR group) in 31 studies were included in this meta-analysis. Moreover, Sprague Dawley (SD) rats, Wistar rats, albino rats, C57BL/6 mice, or C57BLKS/J-Lepr db /Lepr db (db/db) mice were reported in these included studies. The animal models included feeding a high-fat diet (HFD), or methionine-and cholinedeficient (MCD) diet, or db/db mice. Interestingly, there were also intraperitoneal injections of tunicamycin (TM) and CCl 4 models. The BBR group were mainly treated intragastrically with BBR at dosages ranging from 50 mg/kg to 300 mg/kg, while two studies have reported that animals were intraperitoneally injected with 5 mg/kg BBR. Meanwhile, one study has reported adding 1.4 g/kg BBR to animals' HFD. The duration of administration ranged from 3 days to 20 weeks. The main reason for this was due to the different modeling methods and BBR dosage. Moreover, the outcomes included main outcome measures such as TC, TG, ALT, AST, HDL-C, LDL-C, FFA, FBG, FINS, and HOMA-IR and other outcome measures such as animal body weight, liver weight, liver index, NAS, steatosis score, TNF-α, IL-6, IL-1β, GSH, and TBARS. Few adverse reactions of animals were reported in the included studies; see Supplementary Table S1 for details.

Methodological Quality
Nineteen studies were fully generated or applied the randomized allocation sequence. Twenty-five studies have reported that the groups were similar at baseline. Only one study described the animals that were selected at random for results. Seven studies described blinded result assessment. Only two studies described incomplete data and all studies were free of selective results reporting. However, the following aspects were not clear: fully concealed allocation, random housing of the animals, blinded investigators, and being free of other bias. All the details are shown in Figure 3 and Supplementary Figure S1.  Figure S2B). 24 studies with 426 animals have reported the blood TG level after BBR treatment for NAFLD/NASH. There was obvious    Figure S3A). 14 studies with 231 animals reported the liver tissue TG level after BBR treatment. Liver tissue TG was analyzed to have significant heterogeneity (p 0.01, I 2 53%). The random-effects model was used in the meta-analysis. Compared with the model group, BBR could remarkably reduce the liver tissue TG level [SMD 1.99, 95% CI (1.49, 2.49), p < 0.00001] ( Table 1; Supplementary Figure S3B).

Major Outcome Measures
12 studies with 219 animals have reported the HDL-C level after BBR treatment for NAFLD/NASH. There was outstanding heterogeneity in HDL-C level (p < 0.00001, I 2 83%). Thus, we used the random-effects model in the meta-analysis. BBR could observably increase the HDL-C level compared with the model group [SMD −1.04, 95% CI (−1.82, −0.26), p 0.0009] ( Table 1; Supplementary Figure S4A). 15 studies with 275 animals have reported the LDL-C level after BBR treatment for NAFLD/NASH. There was significant heterogeneity in LDL-C level (p < 0.00001, I 2 86%). The random-effects model was utilized for the metaanalysis. Compared with the model group, the LDL-C level was significantly different [SMD 2.56, 95% CI (1.69, 3.43), p < 0.00001] ( Table 1; Supplementary Figure S4B). Five studies with 82 animals have reported the FFA level after BBR treatment for NAFLD/NASH. FFA level was found remarkably heterogeneous (p < 0.00001, I 2 89%). Therefore, we took the random-effects model in the meta-analysis. BBR Figure S6A). Eight studies with 141 animals have reported the FINS level after BBR treatment for NAFLD/NASH. It was memorably heterogeneous in the FINS level (p < 0.00001, I 2 78%); thus, we adopted the random-effects model for the metaanalysis. As shown in Table 1 Figure  S6C). The pooled analysis of insulin resistance and glucose metabolism demonstrated that BBR can not only accelerate the recovery of insulin resistance but also improve glucose metabolism and alleviate liver damage.  Figure S7A). Eight studies with 122 animals have reported the animal liver weight after BBR treatment for NAFLD/NASH. The animal liver weight was demonstrated to be significantly heterogeneous (p 0.0005, I 2 73%) so as to select the random-effects model. BBR could evoke an apparent reduction in the liver weight compared with the control group [SMD 1.17, 95% CI (0.35, 2.00), p 0.005] (Supplementary Figure S7B). Five studies with 82 animals reported the liver index after BBR treatment for NAFLD/NASH. Obviously, there was remarkable heterogeneity in the liver index (p 0.0006, I 2 79%). Moreover, we used the random-effects model in the metaanalysis. However, although we could observe a decreasing trend of the liver index after BBR treatment, it has no significant difference compared with the model group [SMD 1.12, 95% CI (−0.01, 2.25), p 0.05] (Supplementary Figure  S7C). Six studies with 111 animals reported the NAS after BBR treatment for NAFLD/NASH. There was obvious heterogeneity in NAS (p 0.0008, I 2 76%), so we took the random-effects model in the meta-analysis. The results indicated that BBR could significantly alleviate the NAS compared with the control group [SMD 3.80, 95% CI (2.26, 5.23), p < 0.00001] (Supplementary Figure S7D). Three studies with 59 animals have reported the steatosis score after BBR treatment for NAFLD/NASH. It was not memorably heterogeneous in the steatosis score (p 0.76, I 2 0%); thus, we adopted the fixed-effects model for the metaanalysis. As shown in Supplementary Figure S7E, BBR could effectively reduce the steatosis score compared with the model group [SMD 1.34, 95% CI (0.76, 1.92), p < 0.00001].
Four studies with 88 animals have reported the TNF-α level after BBR treatment for NAFLD/NASH. There was obvious heterogeneity in TNF-α level (p < 0.00001, I 2 95%), so we took the random-effects model in the meta-analysis. The results indicated that BBR could significantly alleviate the TNF-α level compared with the control group [SMD 7.00, 95% CI (1.58, 12.41), p 0.01] (Supplementary Figure S8A). Three studies with 44 animals have reported the liver TNF-α level after BBR treatment for NAFLD/NASH. It was memorably heterogeneous in the liver TNF-α level (p 0.0002, I 2 88%); thus, we adopted the random-effects model for the meta-analysis. As shown in Supplementary Figure S8B, BBR could effectively reduce the level of liver TNF-α compared with the model group [SMD 3.31, 95% CI (0.39, 6.22), p 0.03].
Three studies with 68 animals have reported the IL-6 level after BBR treatment for NAFLD/NASH. It was memorably heterogeneous in the IL-6 level (p < 0.00001, I 2 96%); thus, we adopted the random-effects model for the meta-analysis. As shown in Supplementary Figure S8C, BBR could effectively reduce the level of IL-6 compared with the model group [SMD 7.08, 95% CI (1.02, 13.14), p 0.02]. Two studies with 28 animals have reported the IL-1β level after BBR treatment for NAFLD/NASH. It was not heterogeneous in the IL-1β level (p 0.23, I 2 30%); thus, we adopted the fixed-effects model for the meta-analysis. As shown in Supplementary Figure  S8D, BBR could effectively reduce the level of IL-1β level compared with the model group [SMD 1.83, 95% CI (0.89, 2.78), p 0.0001]. Three studies with 68 animals have reported the GSH level after BBR treatment for NAFLD/NASH. Obviously, there was remarkable heterogeneity in the GSH level (p 0.04, I 2 77%). Moreover, we used the random-effects model in the metaanalysis. The results indicated that BBR could significantly increase the GSH level compared with the control group [SMD −6.51, 95% CI (−10.15, −2.88), p 0.0004] (Supplementary Figure S8E). Three studies with 50 animals have reported the TBARS level after BBR treatment for NAFLD/NASH. Obviously, there was remarkable heterogeneity in the TBARS level (p 0.10, I 2 57%). Furthermore, we used the random-effects model in the metaanalysis. The results indicated that BBR could significantly alleviate the TBARS level compared with the control group [SMD 3.13, 95% CI (1.74, 4.52), p < 0.0001] (Supplementary Figure S8F). The pooled analysis results indicated that BBR can not only accelerate the recovery of inflammation but also alleviate oxidative damage.

Subgroup Analysis for Different Animal Breeds
13 studies with 236 SD rats have reported the blood TC level after BBR treatment for NAFLD/NASH. There was obvious heterogeneity in blood TC level (p < 0.00001, I 2 84%), so we took the random-effects model in the meta-analysis. The results indicated that BBR could significantly alleviate the blood TC level compared with the control group [SMD 3.08, 95% CI (2.08, 4.08), p < 0.00001] (Supplementary Figure S9A 2.1.1). Four studies with 71 C57BL/6 mice have reported the blood TC level after BBR treatment for NAFLD/NASH. It was memorably heterogeneous in the blood TC level (p 0.0002, I 2 85%); thus, we adopted the random-effects model for the meta-analysis. However, it was not significantly different compared with the model group [SMD 0.95, 95% CI (−0.47, 2.31), p 0.20], as shown in Supplementary Figure S9A 2.1.2. In addition, two studies with 40 Wistar rats have reported the blood TC level after BBR treatment for NAFLD/NASH. Obviously, there was remarkable heterogeneity in the blood TC level (p 0.0006, I 2 91%). Moreover, we used the random-effects model in the meta-analysis. However, although we could observe a decreasing trend of blood TC level after BBR treatment, it was not significantly different compared with the model group [SMD 0.56, 95% CI (−1.85, 2.97), p 0.65] (Supplementary Figure S9A  2.1.3). Four studies with 68 SD rats have reported the liver tissue TC level after BBR treatment for NAFLD/NASH. There was heterogeneity in liver tissue TC level (p 0.09, I 2 54%). Hence, we adopted the random-effects model in the meta-analysis. Compared with the model group, BBR caused a sharp decline in the liver tissue TC level [SMD 1.89, 95% CI (0.97, 2.80), p < 0.0001] (Supplementary Figure S9B 2.2.1). Two studies with 35 C57BL/6 mice have reported the liver tissue TC level after BBR treatment for NAFLD/NASH. The liver tissue TC level was demonstrated to be significantly heterogeneous (p 0.08, I 2 68%) so as to select the random-effects model. BBR could evoke an apparent reduction in the liver tissue TC level compared with the control group [SMD 1.63, 95% CI (0.51, 3.11), p 0.03] (Supplementary Figure S9B 2.2.2).
14 studies with 240 SD rats have reported the blood TG level after BBR treatment for NAFLD/NASH. Blood TG showed significant heterogeneity (p < 0.00001, I 2 92%). The random-effects model was used in the meta-analysis. BBR could observably reduce the blood TG level compared with the model group [SMD 3.64, 95% CI (2.21, 5.08), p < 0.00001] (Supplementary Figure S10A 2.3.1). Four studies with 67 C57BL/6 mice have reported the blood TG level after BBR treatment for NAFLD/NASH. Blood TG was found to have significant heterogeneity (p 0.03, I 2 65%). Therefore, we used the random-effects model in the meta-analysis. BBR could remarkably decrease the blood TG levels compared with the model group [SMD 1.70 Figure S10A 2.3.3). Five studies with 84 SD rats have reported the liver tissue TG level after BBR treatment. Liver tissue TG was analyzed to have significant heterogeneity (p 0.03, I 2 64%). The random-effects model was used in the meta-analysis. Compared with the model group, BBR could remarkably reduce the liver tissue TG level [SMD 1.88, 95% CI (0.96, 2.80), p < 0.0001] (Supplementary Figure S10B 2.4.1). Five studies with 83 C57BL/6 mice have reported the liver tissue TG level after BBR treatment. Liver tissue TG was analyzed to have significant heterogeneity (p 0.03, I 2 63%). The randomeffects model was used in the meta-analysis. Compared with the model group, BBR could remarkably reduce the liver tissue TG level [SMD 1.96, 95% CI (1.01, 2.92), p < 0.0001] (Supplementary Figure S10B 2.4.2). Two studies with 40 Wistar rats have reported the liver tissue TG level after BBR treatment. Liver tissue TG was analyzed to have significant heterogeneity (p 0.07, I 2 69%). The random-effects model was used in the meta-analysis. Compared with the model group, BBR could remarkably reduce the liver tissue TG level [SMD 2.30, 95% CI (0.75, 3.85), p 0.004] (Supplementary Figure  S10B 2.4.3). Two studies with 24 db/db mice have reported the liver tissue TG level after BBR treatment for NAFLD/NASH. There was minor heterogeneity in liver tissue TG level (p 0.30, I 2 8%). So, the fixed-effects model was chosen in the meta-analysis. Compared with the model group, the liver tissue TG level was remarkably different [SMD 2.20, 95% CI (1.07, 3.33, p 0.0001] (Supplementary Figure S10C 2.4.4).
Due to space constraints, the subgroup analysis for different animal breeds of additional related major indicators such as ALT, AST, HDL-C, LDL-C, and FBG is shown in Supplementary  Figures S11-14.

Subgroup Analysis for Different Models
In the HFD model, 19 studies with 362 animals have reported the blood TC level after BBR treatment for NAFLD/NASH. Blood TC showed significant heterogeneity (p < 0.00001, I 2 83%). The random-effects model was used in the meta-analysis. BBR could observably reduce the blood TC level compared with the model group [SMD 2.30, 95% CI (1.61, 2.98), p < 0.00001] (Supplementary Figure S15A 3.1.1). In the MCD model, two studies with 34 animals have reported the blood TC level after BBR treatment for NAFLD/NASH. Blood TC level was found to have apparent heterogeneity (p 0.03, I 2 78%). Therefore, we used the fixed-effects model in the meta-analysis. However, there were no significance compared with the model group [SMD 0.00, 95% CI (−1.51, 1.52), p 1.00] (Supplementary Figure  S15A 3.1.2).
In the HFD model, nine studies with 153 animals have reported the liver tissue TG level after BBR treatment for NAFLD/NASH. There was obvious heterogeneity in liver tissue TG level (p 0.009, I 2 61%). So, the random-effects model was chosen in the meta-analysis. Compared with the model group, the liver tissue TG level was clearly different [SMD 2.25, 95% CI (1.54, 2.96, p < 0.00001] (Supplementary Figure S15B). In the MCD model, two studies with 34 animals have reported the liver tissue TG level after BBR treatment. Liver tissue TG had minor heterogeneity (p 0.76, I 2 0%). The fixed-effects model was used in the metaanalysis. Compared with the model group, BBR could remarkably reduce the liver tissue TG level [SMD 1.55, 95% CI (0.75, 2.34), p 0.0001] (Supplementary Figure S15C 3.2.2). In the db/db model, two studies with 24 animals have reported the liver tissue TG level after BBR treatment for NAFLD/NASH. There was minor heterogeneity in liver tissue TG level (p 0.30, I 2 8%). Thus, we used the fixed-effects model in the meta-analysis. BBR could observably increase the liver tissue TG level compared with the model group [SMD 2.20, 95% CI (1.07, 3.33), p 0.0001] (Supplementary Figure S15C 3.2.3).
Due to space constraints, the subgroup analysis for different animal models of additional related major indicators such as ALT and AST is shown in Supplementary Figures S16-17.

Publication Bias
In order to explore the publication bias and heterogeneity indepth, we drew the funnel plots. As shown in Supplementary Figure S18, the funnel plots of main outcomes such as TC, TG, ALT, AST, HDL-C, LDL-C, FBG, FINS, HOMA-IR, and FFA were asymmetric, suggesting the potential publication bias and heterogeneity.

Exploration of Preclinical Mechanism
NAFLD is a chronic liver disease closely associated with obesity, type 2 diabetes, and hyperlipidemia and is becoming a serious health problem worldwide (Jahn et al., 2019). The major feature of NAFLD in the early stage is the slow accumulation of fat in hepatocytes (namely steatosis), and it can gradually develop into Frontiers in Pharmacology | www.frontiersin.org September 2021 | Volume 12 | Article 742465 NASH, which involves tissue injury, chronic liver inflammation, and fibrosis and eventually contributes to end-stage liver disorder, including liver cirrhosis and hepatoma (Zhang et al., 2016;Goldberg et al., 2017). Since no approved medication is currently available for NAFLD/NASH, this forces us to determine potential pharmacological targets and create future therapies. Moreover, it is universally acknowledged that no animal model can completely reproduce the pathological condition of human NASH. How to transform the obtained results into further clinical researches is an ongoing challenge. To this end, a basic and translational study is still indispensable; in the translational process, preclinical systematic evidence plays a crucial role. Different animal models with diverse mechanisms play various roles in the pathogenesis of NASH. Fortunately, the studies we considered identified the efficacy of BBR based on different models and mechanisms. The vast majority of included researches employed HFD to construct the animal model of NAFLD/NASH. The model presents similar characteristics of NASH pathology and metabolic syndromes such as hyperlipidemia and obesity. However, this model generally develops less severe hepatic inflammation and fibrosis, limiting its application for the study (Mow et al., 2004;Santhekadur et al., 2018). In this review, the results showed that BBR could improve HFD-fed animals' disease state and restore the lipids content, insulin resistance, and glucose metabolism to reduce liver damage.
Several included studies have reported the NASH model based on the MCD diet, a commonly used diet that contains moderate fat (10%) and considerable sucrose (40%) but is deficient in methionine and choline. It can generally produce the extensive liver histological phenotype of NASH within only a few weeks (Weltman et al., 1996;Anstee et al., 2006;Yamada et al., 2016). Moreover, the major reason for the rapid onset is that the impairment in hepatic very-low-density lipoprotein (VLDL) secretion causes disorder in hepatocyte lipid metabolism accompanied by oxidative stress injury (Weltman et al., 1996;Lee et al., 2015). Therefore, this model is more suitable for discussing the histology of advanced NASH and the mechanism of fibrosis and inflammation than the other models. However, the limitations of this model are that some metabolic changes such as significant body weight loss and lack of insulin resistance have considerable disparities with human NASH. The studies considered have demonstrated that BBR could effectively alleviate the lipids content of MCD-fed animals to relieve liver injury.
Several studies have used db/db mice to conduct the NASH model. This kind of mice carries a point mutation of point mutation, which leads to defective leptin signaling, resulting in defective leptin signaling. As leptin is in charge of controlling feeding behavior through enhancing satiety, these mice generally have abnormal hyperphagia and are obese, hyperglycemic, hyperinsulinemic, and diabetic (Hummel et al., 1966;Yang et al., 1997;Trak-Smayra et al., 2011;Ibrahim et al., 2016). The studies with db/db mice have indicated that BBR could significantly alleviate the accumulation of fat in hepatocytes and reduce liver tissue damage.
We can conclude that although different models have relative disparities with mechanisms and characteristics, they all represent some pathological phenotypes of NAFLD/NASH. The subgroup analyses of different animal models in this review have shown the breathtaking therapeutic effects of BBR, although it exerted different targets and mechanisms and displayed different efficacy strengths when faced with different models.

Discussion of the Subgroup Analyses on Different Rat Breeds
In this study, BBR was used to treat the NAFLD/NASH model caused by different breeds of animals, among which the therapeutic effects were significantly proved in most of the results. Due to the limited number of studies, some studies such as blood TG and TC in Wistar rats demonstrated to have a treatment trend, but the results were not significantly different, needing more researches to further identification. In addition, one study has revealed anti-fatty liver action of BBR in an HFD-induced larval zebrafish model besides rats or mice model , although we did not include it due to the lack of sample size. At present, more extensive animal breed experiments are still required to make up the gap.

Discussion of Heterogeneity on Effect of BBR
The emergence of heterogeneity not only deepens the content of this review but also leaves wider issues to be explored. The occurrence of heterogeneity is influenced by multi-factors, including experimental design, study subjects, intervention measures, and results combination (Thompson et al., 1997;Hooijmans et al., 2014;Shi et al., 2021). The great heterogeneity difference in this review may be due to BBR and animal diets coming from different laboratories. The dosage of BBR varies between different studies and so does the composition of animal diets. Another issue is the inconsistency of the outcomes in terms of TC, TG, HDL-C, and LDL-C. The measurement kits, animal feeding conditions, laboratory temperatures, and even operations by different operators can lead to potential heterogeneity. Due to sample limitations, we were unable to implement subgroup analysis on the factors mentioned above. In experimental design and intervention measures, we performed a subgroup analysis of different animal breeds and animal models. Through the aforementioned subgroup analysis, we hoped to control the heterogeneity as much as possible and broaden the comprehension of BBR in the treatment of NAFLD/NASH. Simultaneously, mechanisms and sensitivity to BBR vary from different models, influencing the administration time and dose. These all need further in-depth research.

Strength
The application of system reviews is conducive to preclinical design and clinical reliability. Furthermore, it can contribute to conducting experimental animal research and eliminating unnecessary experiments so as to explore the potential value of BBR in treating hepatic diseases in-depth, enhance the dependability of future clinical trials, and promote the extensive application of BBR. Additionally, traditional Chinese medicine containing berberine has a long history of application in China, and berberine has been used in several over-the-counter (OTC) drugs to treat gastroenteritis, indicating its relative safety and efficacy (Vivoli et al., 2016).
According to the results of this review, we can see that BBR was suitable for a range of NAFLD/NASH models. Based on the studies we included, we mapped the mechanism figure of berberine in the treatment of NAFLD/NASH. We could see the results from Figure 4 that BBR could activate the adenosine monophosphate-activated protein kinase (AMPK) pathway (Lee et al., 2006;Kim et al., 2009;Zhou et al., 2017), ameliorate insulin sensitivity and glucose metabolism (Lee et al., 2006;Chen et al., 2009;Kong et al., 2009), regulate mitochondrial function (Teodoro et al., 2013;Xu et al., 2014), alleviate inflammation and oxidative stress (Vivoli et al., 2016;Dinesh et al., 2017;Mahmoud et al., 2017;Zhou et al., 2017), modulate cell death and ER stress (He et al., 2016;Zhou et al., 2017;Mai et al., 2020), reduce DNA methylation (Chang et al., 2010), and regulate gut microenvironment (Zhang et al., 2012;Zhang Y. et al., 2015) and neurotoxicity (Kim et al., 2009;Ghareeb et al., 2015). Although in different models, the action mode of BBR is different, BBR could reduce the levels of TC, TG, ALT, AST, and so forth via the above mechanisms and in turn ameliorate the liver function.
Moreover, the effect of BBR varies among the different animal breeds and based on the administration time. Hence, in the experimental design, different species and administration times were applied according to different experimental purposes. Besides, more complicated issues, such as the dose range of BBR, are supposed to be given more focus. The overall results suggest that berberine plays a vital role in NAFLD/NASH. Further research on the related mechanism is warranted.

Limitations
We took into account the internal and external factors, including room temperature, laboratory equipment, BBR dosage, and different selection of kits in this previous review. Nevertheless, the researches with a detailed description in terms of their methodological procedures and transparency are not abundant. Thus, controlling the sources of the potential heterogeneity will be the focus of future studies. Furthermore, the asymmetric funnel plots suggested potential publication bias, which might result from the following causes. On the one hand, the researches were conducted in different laboratories, some of which were not GLP centers. On the other hand, the animal sizes varied from 4 to 20. The publication bias may be a possible interpretation for such a small-study effects bias. Therefore, the entire experiment process and limited animal sizes might contribute meaningfully to the increment of variability and potential bias.
In NAFLD basic research, many animal researchers are prone to regard blinding and randomization as redundant. In fact, some of the studies we included had a similar concept as the authors generally believed that their animals come from a sample group with homogeneous genetic background and environment. However, not only the animals but also the inducement of NAFLD may cause alteration, thereby weakening the rigor. Secondly, if each animal represents almost the same or unique sample, then repeating the experiment on such a sample will produce erroneous results. Considering that randomization is simple, practical, and cost-effective, it should be recommended to be an essential content of the experiment.
Additionally, no data in this study reported the calculation of the absorption dosage of BBR. That may be because the determination of BBR absorption is still technically difficult. Simultaneously, it is also challenging for us to unify the degree of absorption of different methods, such as gavage and intraperitoneal injection of BBR. Last but not least, no studies have taken into account gender and age differences in terms of efficacy, because all studies used adult animals aged 4-9 weeks without young or old animals and all studies except one (Ghareeb et al., 2015) have used male animals, which need pay more attention in further exploration. In conclusion, the results are supposed to be interpreted with reason and caution.