Flavonoids from traditional Chinese herbs for diabetes in rats: a network Meta-analysis

Abstract

OBJECTIVE

In this Meta-analysis, we evaluated the hypoglycemic effect of 5 flavonoids found in traditional Chinese herbs (naringenin, kaempferol, puerarin, baicalein, and luteolin) on diabetic rats.

METHODS

Four databases including PubMed, Web of Science, Embase, and Cochrane Library, were searched from inception to May 2020. Only studies using diabetes model rats were included in the analysis. Blood glucose data from the last measurement were collected and analyzed. Pair-wise Meta-analyses were conducted using STATA v14.0 software and a Meta-analysis was conducted using STATA v14.0, ADDIS v1.16.6, and R v3.6.1. The quality of included studies was assessed with the SYRCLE risk of bias tool for animal studies, and publication bias was evaluated with a comparison-adjusted funnel plot.

RESULTS

A total of 33 studies were included in the analysis, in which all 5 flavonoids showed a beneficial effect on blood glucose level of diabetic rats were included in the final analysis. The standardized mean differences (95% confidence intervals) were –4.92 (–6.67, –3.17) for naringenin, –12 (–18.74, –5.27) for kaempferol, –2.52 (–3.77, –1.26) for puerarin, –3.04 (–5.75, –0.34) for baicalein, and –1.94 (–2.95, –0.92) for luteolin. The network Meta-analysis showed no statistically significant differences between the effect sizes of the flavonoids.

CONCLUSION

The results of the Meta-analysis showed that naringenin, kaempferol, puerarin, baicalein, and luteolin all have clear hypoglycemic effects in rat diabetes models, highlighting their therapeutic potential for preventing and treating diabetes mellitus in clinical practice.

1. INTRODUCTION

Diabetes mellitus is a common chronic disease of the endocrine system characterized by hyperglycemia. Type 1 and type 2 diabetes account for the majority of cases.¹ The mechanisms underlYing each type differ, but oxidative stress and inflammation are critical pathologic changes in the progression of both and hyperglycemia is a common manifestation. The global prevalence of diabetes is increasing each year and is predicted to reach 10.9% by 2045,² posing a major challenge to public health systems. The currently used oral hypoglycemic drugs have various side effects including gastrointestinal dysfunction, hypoglycemia, edema, anemia, and weight gain.³ Therefore, alternative treatment approaches that target multiple physiologic processes and have fewer adverse effects such as traditional medicines, natural products, and dietary supplements are desired for more effective prevention and treatment of diabetes.^4,5

Traditional Chinese Medicine (TCM) has demonstrated efficacy in the treatment of diabetes, with many herbs having antidiabetic properties.^6,7 Diabetes, which is known as “Xiaoke” in TCM, is thought to be caused by Qi and Yin deficiency as well as internal heat according to TCM theory. TCM treats patients with diabetes with herbs that nourish Yin, benefit Qi, clear heat, and purge fire. Frequently used herbs include Huangqin (Radix Scutellariae), Gegen (Radix Puerariae), Huanglian (Rhizoma Coptidis), Tianhuafen (Radix Trichosanthis), etc. Flavonoids are active compounds in these herbs that have an aromatic ring-3-carbon bridge-aromatic ring (6C-3C-6C) structure and are among the most common plant metabolites in many herbs, fruits, and vegetables, and have physiologic benefits including antidiabetic effects.^8,9 Dietary intake of flavonoids has anticancer and antioxidant effects that are exerted through the suppression of inflammation, regulation of gastro-intestinal function and metabolism, and lowering of blood glucose level.^10-14

Flavonoids present in TCM herbs have known hypoglycemic effects. Naringenin is found in citrus fruit and herbs such as Foshou (Fructus Citri Sarcodactylis) and Chenpi (Pericarpium Citri Reticulatae);^15,16 kaempferol is widely distributed in fruits and vegetables as well as in Chinese herbs such as Baiguo (Semen Ginkgo), Honghua (Flos Carthami), and Gaoliangjiang (Rhizoma Alpiniae Officinarum);^17,18 puerarin is mainly found in the herb Gegen (Radix Puerariae Lobatae);¹⁹ baicalein is found in the herb Huangqin (Radix Scutellariae Baicalensis); and luteolin is present in carrots, celery, and broccoli²⁰ and is also distributed in Chinese herbs such as Pugongying (Herba Taraxaci Mongolici), Jinyinhua (Flos Lonicerae), and Juhua (Flos Chrysanthemi).²¹ These flavonoids exert a therapeutic effect in diabetes by targeting multiple factors and pathways.²² However, studies on these compounds have mostly been preclinical, whereas a systematic evidence-based analyses are needed to evaluate their clinical potential.

Meta-analyses assess all eligible published data to obtain a pooled estimate of the overall effects of an intervention, which is more reliable and robust than conclusions from a single study.²³ However, traditional pairwise Meta-analyses only include 2 parameters. In contrast, a network Meta-analysis involves direct comparisons between interventions in individual studies and indirect comparisons across studies based on a common comparator to obtain a final mixed-comparison results.²⁴ In the present study, we performed a network Meta-analysis in order to evaluate and compare the hypoglycemic effect of 5 flavonoids found in TCM herbs.

2. METHODS

This network Meta-analysis was conducted according to PRISMA Extension Statement for Reporting of Systematic Reviews Incorporating Network Meta-analyses of Health Care Interventions guidelines.²⁵

2.1. Literature search

Four databases including PubMed, Web of Science, Embase, and Cochrane Library were searched from inception to May 2020. The search strategy was as follows. (a) Flavonoids were searched in the title/abstract by Medical Subject Heading term and entry terms on the PubMed website. (b) Experiments performed on rats were identified using the terms “animal”, “rodent”, “rat”, and “rattus” in the title/abstract. (c) Titles and abstracts were searched for the terms “diabetes”, “insulin”, “glucose”, “obesity”, and “metabolic syndrome”. (d) Search results were combined using “AND”. No further restrictions were imposed with regard to language or study type/design.

2.2. Eligibility criteria

The eligibility criteria for studies were as follows. (a) Experiments were performed using a rat model of diabetes. (b) Flavonoids were orally administered to the animals and interventions were not combined with other chemicals or treatments. (c) The comparison group consisted of diabetic rats that were treated with placebo (solvent used for the flavonoid) or untreated. (d) Blood glucose levels were reported for the treatment and comparison groups at the end of study.

2.3. Study screening

Two authors independently screened the titles and abstracts of all retrieved articles, and those meeting the inclusion criteria were retained for further analysis. In the case of disagreement, a third author was consulted. The articles were reviewed and excluded if they met the following criteria: (a) no full-text article was available; (b) treatment duration was < 2 weeks; (c) blood glucose data were lacking; (d) the same data were published in more > 1 article; and (e) the results were not consistently represented in the literature.

2.4. Data extraction

The following information was extracted from each article by 2 authors (HY and LW): title; name of the first author; year of publication; country; rat strain, sex, initial age/weight, and number per group; method used to establish the diabetes model; flavonoid and positive control used in study; dosage and duration of treatment; and blood glucose levels of the experimental, positive control, and diabetes control groups (mean ± standard deviation). Only optimal treatment dosages were included in the Meta-analysis. If the results were presented in figures, Web-plot Digitizer software (https://automeris.io/WebPlotDigitizer/) was used to extract the data.

2.5. Assessing risk of bias

The SYRCLE risk of bias tool for animal studies²⁶ is the most commonly used tool for assessing the risk of bias in animal studies; it consists of 10 items reflecting different possible sources of bias. In our study, these results are presented in a figure and not as calculated summary scores according to the developer’s suggestion.

2.6. Meta-analysis

Pairwise Meta-analysis of each flavonoid was performed using STATA v14.0 software (StataCorp, College Station, TX, USA). Given the differences in study design, animal model, and intervention details, analyses were performed using a random-effects model and results are shown as standardized mean difference (SMD) with 95% confidence interval (95% CI). The heterogeneity across included studies was quantified with Higgins’ I ² heterogeneity index.²⁷ Meta-regression was performed to identify potential sources of heterogeneity, and trim-and-fill analysis was performed to verify the reliability of included studies.

A Bayesian network Meta-analysis, which uses posterior probability to rank all analyzed interventions, was performed using ADDIS v1.16.6 software²⁸ and R v3.6.1 software with rjags v4-10 and gemtc v0.8-2 packages. Inconsistency among studies was assessed by evaluating the concordance of indirect and direct comparisons. Node-splitting was performed to statistically evaluate the consistency, and inconsistency factors were used to assess inconsistency; if the 95% CI for an inconsistency factor contained a neutral value (0), the results were considered consistent. The following parameters were used to generate the model: number of chains = 4; tuning iterations = 20 000; simulation iterations = 50 000; and thinning interval = 10. The potential scale reduction factor (PSRF) was calculated by comparing within- and between-chain variance. PSRF reflects the convergence of the simulation, with a value close to 1 indicating good convergence. Probable relative rankings are presented as a rank probability plot.

Smaller studies can show different and larger treatment effects than large studies and negative results are less likely to be published, which can affect the reliability of results. Small study effects and publication bias were assessed with a comparison-adjusted funnel plot,²⁹ which displays the difference in effect size between individual studies and overall pooled estimate of each comparison on the x axis and standard error of the effect on the y axis. Asymmetry in the figure suggests the existence of small-study effects and publication bias.

3. RESULTS

3.1. Search results

A total of 3464 articles were selected from 4 databases. After removing 2181 duplicates, 93 articles met the inclusion criteria. The full text of these articles were screened, and 33 (naringenin, n = 13; kaempferol, n = 5; puerarin, n = 6; baicalein, n = 4; and luteolin, n = 5)^{19, 30-61} were included in the final Meta-analysis. The basic information and characteristics of the included studies are listed in Supplementary Table 1.

Table 1.

League table showing the results of the network Meta-analysis comparing the effect of flavonoid treatments on blood glucose level in diabetic rat modelsa

Baicalein
-2.63 (-133.87, 127.00)	Kaempferol
-29.63 (-163.37, 104.77)	-26.82 (-145.01, 90.10)	Luteolin
-27.54 (-156.63, 97.98)	-25.89 (-129.99, 78.58)	0.73 (-113.72, 117.55)		Metformin
-11.47 (-123.20, 104.36)	-7.83 (-105.10, 86.88)	18.68 (-85.70, 123.21)		17.32 (-68.07, 101.15)	Naringenin
-141.24 (-240.81, -36.43)	-138.37 (-219.65, -55.54)	-111.02 (-198.59, -22.41)		-112.20 (-185.78, -36.85)	-129.42 (-182.69, -77.07)	Placebo
-15.64 (-136.02, 108.60)	-12.77 (-118.18, 92.85)	14.23 (-97.88, 126.48)		13.54 (-78.60, 105.39)	-3.81 (-90.85, 80.55)	125.74 (54.89, 195.28)	Puerarin

Note: ^aNumbers below each treatment represent the mean difference (95% confidence interval) in blood glucose (mg/dL) between the row and column

3.2. Risk of bias of included studies

The risk of bias was assessed according to the SYRCLE checklist (Supplementary Table 2). The risk was unknown or high for 3 items — namely, allocation concealment, blinding (performance bias), and random outcome assessment. None of the included articles reported concealment in allocation or whether animals were randomly selected for outcome assessment. Although in several studies outcomes were measured in all animals, they were still regarded as having “unclear risk” for item 6 (“random outcome assessment”). One article³⁷ specifically mentioned the lack of blinding in performance and outcome measurements, and was thus deemed to be “high risk” on both items.

3.3. Results of pairwise Meta-analyses

The 5 flavonoids all showed a beneficial effect on blood glucose level in diabetic rats. The SMDs and 95% CIs were -4.92 (-6.67, -3.17) for naringenin, -12 (-18.74, -5.27) for kaempferol, -2.52 (-3.77, -1.26) for puerarin, -3.04 (-5.75, -0.34) for baicalein, and -1.94 (-2.95, -0.92) for luteolin (Supplementary Figures 1-5). There was a high degree of heterogeneity in each pairwise Meta-analysis and we used Meta-regression to identify the possible sources. Dose and duration of treatment were treated as variables and different approaches for establishing the diabetes model were treated as dummy (nonquantitative) variables. The approaches were classified into 4 types: (a) streptozotocin (STZ) injection; (b) STZ injection combined with nicotinamide injection; (c) dietary manipulation; and (d) STZ injection combined with dietary manipulation. The results of the analysis showed that the heterogeneity across studies was unlikely to be explained by the above parameters (data not shown).

A trim-and-fill analysis was performed to assess the reliability of each pairwise Meta-analysis (Supple-mentary Figure 6), but this did not affect the results.

3.4. Results of network Meta-analysis

The network diagram is shown in Figure 1. The contribution of each direct comparison to the final estimate (mixed comparison) was calculated and is shown in Supplementary Figure 7.

Figure 1. Network diagram of the network Meta-analysis .

The size of nodes corresponds to the total number of studies in each group, and line thickness represents the total number of studies between connected treatment groups.

All PSRF values reached 1 after multiple iterations, indicating convergence of the stimulations. The node-splitting approach was used to assess the inconsistency between direct and indirect evidence in specific comparisons.⁶¹ All node-split models had a P value > 0.05, indicating that the results from direct and indirect comparisons were statistically consistent. The results from the estimation of different effect sizes between included treatments are listed in Table 1, and the rank probabilities are presented in Figure 2.

Figure 2. Rank probability plot for network Meta-analysis.

The columns show the P values of each treatment for rankings from left to right (ie, from lowest to highest rank).

The risks of small-study effects and publication bias were assessed with a comparison-adjusted funnel-plot (Supplementary Figure 8). The asymmetry of the plot suggested that both types of bias existed in the results of the Meta-analysis.

4. DISCUSSION

In this study we performed a network Meta-analysis to investigate the hypoglycemic effects of 5 flavonoids derived from herbs used in TCM (naringenin, kaempferol, puerarin, baicalein, and luteolin) that have known benefits in the treatment of diabetes. A network Meta-analysis was performed within a Bayesian framework to compare their effects. To reduce the heterogeneity among included studies, only studies using rats that included data on optimal dosage were included.

The results of the Meta-analysis revealed clear antidiabetic effects for the 5 flavonoids (ie, they decreased blood glucose level compared to the placebo), with none of the 95% CIs of SMDs covering the null value (0). The rank probability plot of effect size of each treatment only showed the relative rank of the p value of each treatment, and did not indicate whether one treatment is superior to another. The SMDs of all interventions had the null value (0) in the 95% CI, indicating that the effect sizes did not differ significantly.

A Meta-regression analysis was performed to identify possible sources of heterogeneity between each pairwise Meta-analysis. The results showed that outcomes were unrelated to treatment dosage and duration and the method used to establish the diabetes model. However, different models manifest different aspects of diabetes, with only STZ dose adjustment contributing to a deviation in the final model.^63,64 Classifying the method of model establishment into 4 categories did not accurately reflect the models included in the present analysis. Additionally, other factors such as timing of and approach used to measure blood glucose, the age of animals, different diets used in the experiments, and different comparative approaches may have contributed to the heterogeneity across studies. A trim-and-fill analysis is typically used to assess the robustness of conclusions.⁶⁵ However, in the current study this had no effect on the outcomes, possibly because of the existence of other types of bias or extremely positive and highly weighted data.⁶⁶ Consistency in the network Meta-analysis was confirmed by the node-splitting approach and as only experiments with rats that reported optimal dosages and had a treatment duration > 2 weeks were included, we considered that the data did not violate the transitivity principle.

The pathogenesis of diabetes is not fully understood; β cell disruption by autoimmune mechanisms is considered as the primary cause of type 1 diabetes.⁶⁷ The main manifestations in these patients are inflammation in the pancreas and apoptosis of pancreatic β cells.^67,68 Insulin resistance is the major pathologic change occurring in type 2 diabetes; inhibiting the insulin signaling pathway results in impaired responses of target organs (eg, liver, adipose, muscle) to insulin.⁶⁹ Oxidative stress, inflammation, endoplasmic reticulum stress, and mitochondrial dysfunction all contribute to the onset and progression of insulin resistance⁷⁰ and diabetic complications.⁷¹

TCM has been effective in treating diabetes mellitus for over 1000 years. Herbs such as Gegen (Radix Puerariae Lobatae), Huangqin (Radix Scutellariae), and Honghua (Flos Carthami) are widely used in antidiabetic decoctions. The active compounds in these herbs and their mechanisms of action are only now starting to be known. Flavonoids have multiple biological activities, and dietary flavonoids benefit patients with chronic diseases.^72-74 For example, naringenin has been used to prevent and treat diabetes in animal models;⁵⁰ although there was no hypoglycemic effect when used in late-stage diabetes, it still functioned as an antioxidant and prevented vascular complications.⁷⁵

The hypoglycemic effects of the 5 flavonoids examined in this study may be attributed to various mechanisms. Oxidative stress promotes diabetes mainly through pathways related to glycolipid metabolism.⁷⁶ Most flavonoids have antioxidant activity that attenuates lipid peroxidation and oxidative stress-mediated injury in the development of diabetes. Flavonoid administration may increase level of enzymatic and non-enzymatic antioxidants in many organs. Antioxidant enzymes [eg, superoxide dismutase (SOD), catalase, glutathione peroxidase (GPx)] are upregulated in pancreatic tissues after naringenin intake; this is consistent with histopathologic observations, suggesting that naringenin protects against pancreas damage induced by STZ.³³ Enhanced antioxidant capacity has also been observed in the lenses of the eye, liver, testicles, and nerves of diabetic rats after treatment with naringenin,^{32,38,52,53,60,77} thereby inhibiting cataract formation, reducing hepatic and testicular injury, improving endothelial function,⁷⁸ and attenuating the pain caused by diabetic neuropathy.⁵² Kaempferol was shown to restore antioxidant levels to near-normal and inhibit lipid peroxidation,^31,79 which mitigated diabetic complications. Puerarin reduced oxidative stress damage in the retina of diabetic rats by stimulating SOD activity and decreasing malon-dialdehyde (MDA) level; ^80,81 baicalein was reported to alter the levels of SOD, GPx, and MDA in myocardial tissues and attenuate diabetic cardiomyopathy in rats; ⁸² and luteolin alleviated diabetic cardiomyopathy, diabetic encephalopathy, and retina and lens neurodegeneration at least in part through suppression of oxidative stress.^{35, 42, 43, 83, 84}

Inflammatory factors such as tumor necrosis factor (TNF)-α, monocyte chemotactic protein 1, and interleukins promote insulin resistance through activation of nuclear factor (NF)-κB or c-Jun N-terminal kinase signaling pathways.⁸⁵ Naringenin reduced the RNA expression of TNF-α and interleukins and decreased their concentrations in serum,^32,52,86 demonstrating that it improves glucose and lipid metabolism by downregulating inflammation. Kaem-pferol, puerarin, baicalein, and luteolin regulate the expression of inflammatory factors; baicalein was shown to attenuate STZ-induced diabetic retinopathy in rats⁵⁷ and luteolin improved lens neurodegeneration, while all 4 flavonoids were shown to modulate NF-κB signaling.^{30, 35, 43, 44, 87}

The 5 flavonoids may also exert antidiabetic effects through other pathways. For instance, insulin promotes glucose utilization (glucose uptake) and storage (glycogen synthesis) and regulates gluconeogenesis in different tissues; these functions are directly controlled by insulin receptor / phosphoinositide 3-kinase / AKT or 5'-adenosine monophosphate-activated protein kinase (AMPK) pathways.^69,88 Naringenin promotes glucose uptake in skeletal muscle by increasing glucose transporter 4 translocation and regulating the phosphorylation of AMPK.^49,89 Puerarin enhanced μ-opioid receptor expression and phosphorylation in skeletal muscle and activated α1 adrenoceptors expressed in the adrenal gland, thereby increasing β-endorphin secretion^90,91 and activating pathways that increased insulin resistance.⁹³ Diabetes is closely associated with obesity, dyslipidemia, and hypertension; collectively, these symptoms define metabolic syndrome.⁹³ The 5 flavonoids have all been reported to exert health benefits by improving lipid metabolism profiles.^{47, 94-98}

Glycemic control is critical for preventing and controlling diabetes mellitus. In type 1 diabetes, blood glucose should be controlled at a near-normal level, whereas in type 2 diabetes, greater caution is required as lowering HbA1c and blood glucose level can alleviate microvascular disease but may have no effect or even increase the risk of macrovascular disease.⁹⁹ Thus, in this study we investigated blood glucose level as the outcome measure of the effectiveness of flavonoid treatments.

Animal experiments provide insights that can serve as a basis for the design of later studies, including those in humans.¹⁰⁰ However, they are generally limited by a small sample size, and randomization and blinding as ways to prevent bias are not as widely accepted as in clinical trials.¹⁰¹ Systematic reviews and Meta-analyses allow researchers to draw more reliable and valid conclusions and precisely estimate effects for future human studies.¹⁰² Although there are limited clinical data on the efficacy of naringenin, kaempferol, puerarin, baicalein, and luteolin in the treatment of diabetes, they all show therapeutic potential because of their hypoglycemic effects, which were confirmed in the present Meta-analysis. The lack of difference in their effect sizes was likely attributable to insufficient data or bias in the analyzed studies. Moreover, antidiabetic effects may be reflected not just in blood glucose level but in insulin and HbA1c levels or oral glucose tolerance test results. Nonetheless, herbs, natural foods, and dietary supplements are alternatives to standard drugs for preventing and treating diabetes mellitus, and can have long-term benefits for patients with fewer side effects. Thus, additional experimental and evidence-based studies are needed to identify the most effective food supplements for managing diabetes.

This study had some limitations. The number of eligible studies and animals used in each study were limited, and the studies were not of very high quality and showed a certain degree of heterogeneity. We also did not evaluate the effect of flavonoids on diabetic indicators other than blood glucose level. Finally, publication bias should have been considered in greater detail. Additional studies are needed to draw more robust conclusions and elucidate the detailed antidiabetic mechanisms of TCM herbs.

In conclusion, the results of this Meta-analysis demonstrate that orally administered naringenin, kaempferol, puerarin, baicalein, and luteolin decrease blood glucose levels in diabetic rats, possibly through antioxidant and anti-inflammatory mechanisms and by improving insulin sensitivity. However, the effect sizes of the 5 flavonoids did not show statistically significant differences. This result may provide references for further studies and clinical trials. Furthermore, empirical evidence from clinical studies is needed to establish the optimal formulation for diabetes treatment.