In order to gather experimental data on the impact of tanshinones in animal models of AD, two researchers with the same education independently searched PubMed, Web of Science, the Cochrane Library, EMBASE, China National Knowledge Infrastructure, and Chinese Biomedicine Database. A third researcher resolved any disagreements. All searches were restricted to works of literature released before 8 September 2024. “Tanshinone” or “Salvia miltiorrhiza” and “Alzheimer’s” or “Alzheimer” or “dementia” or “cognition” or “cognitive” were the subject phrases. All searches and downloads were finished on 9 September 2024, to prevent any prejudice that might arise from later database changes. Supplement Material S1 contains a detailed search strategy.

After removing duplicates, two researchers (Shuwei Wang and Jinsha Yang) with the same training independently reviewed the titles and abstracts based on the inclusion and exclusion criteria for preliminary screening. The two researchers then read the entire text of the pertinent literature to determine whether or not to include it. If there were conflicting views, they were either discussed or sent to Dayong Zhong, the third researcher, for evaluation.

The following were the predetermined inclusion criteria: (1) Experimental AD model was induced in rats or mice, which served as the research subjects; (2) The AD treatment groups received only tanshinone; (3) The AD model control groups were blank or received saline; (4) The Morris water maze (MWM) was utilized to measure cognitive function; (5) Studies involving animal models of AD; (6) Studies written in either Chinese or English.

The following were the predetermined exclusion criteria (1) Other types of animals (e.g., sheep, cats, and dogs) were used; (2) In addition to tanshinones, the treatment group received another neuroprotective agent or Chinese traditional medicine; (3) Combination treatments that involved physical therapy or exercise; (4) No AD model control group was used; (5) The evaluation of treatment efficacy was limited to biochemical or physiological outcomes; (6) Reviews, conferences, case reports, and clinical trial studies; (7) Research data was either incomplete, incorrect, or impossible to extract; (8) Research data for the same population was published repeatedly.

Two researchers extracted information about the studies. The main contents of data extraction were followed: (1) General information, including the title of the study, the name of the first author, the year of publication, the type of AD model, animal species, and weight; (2) Intervention measures, including drug name, dosage, administration mode, and duration; (3) Data and indicators related to the outcome; (4) GetData Graph Digitizer software was used to extract data from graphs. For cases where the data were missing, the researchers contacted the authors and requested the additional information. If the required data were not available, then the study was excluded from the analysis.

In every study that was part of the analysis, MWM was employed to evaluate cognition. Only the final assessment, if cognition was evaluated multiple times during the study, was analyzed.

A checklist that was slightly altered from the Collaborative Approach to Meta-Analysis and Review of Animal Data from Experimental Studies (CAMARADES) was used to evaluate the studies’ methodological quality. In the altered CAMARADES, blinded induction of AD (allocation concealment) was replaced by randomization of participants into treatment groups (Sheng et al., 2015), since blinded induction is more suitable for clinical trial studies, and it is difficult to achieve blinding in animal experimental model induction and allocation. One score was tallied for written evidence of each of the following criteria: (1) peer-reviewed publication; (2) randomization of subjects into treatment groups; (3) assessment of dose-response relationship; (4) blind assessment of behavioral outcomes; (5) monitoring of physiological parameters, such as body temperature; (6) calculation of the sample size necessary to achieve sufficient power; (7) statement of compliance with animal welfare regulations; (8) avoidance of anesthetic agents with marked intrinsic neuroprotective properties (e.g., ketamine); (9) statement of potential conflicts of interest (10) use of a suitable animal model. Each study was given a quality score ranging from zero to ten. The greater the score, the higher the quality of the article.

The study quality also was assessed with secondary criteria as the following criteria described (Hooijmans et al., 2012). These criteria included study characteristics such as the age, species, and sex of the animals used; the duration of supplementation; and the dose(s) of tanshinone. These criteria also included an assessment of the internal validity of the study, i.e., performance bias (differences in care provided?); exclusion bias (differences in withdrawal from studies?); detection bias (differences in outcome measurements?); and selection bias (differences in allocation to comparison groups?), as well as an assessment of the external validity of the population, intervention, and outcome. Each study was given a quality score ranging from zero to twenty-one. The greater the score, the higher the quality of the article. Two researchers (Shuwei Wang and Jinsha Yang) were responsible for completing it.

All the data were analyzed using the Stata14.0 software. The standardized mean difference (SMD) with 95% confidence intervals (CI) was used for continuous variables to pool the results. The Higgins I² statistic was used to measure the inconsistency among the results of the included studies, making it possible to classify the heterogeneity as low (25%), moderate (50%), and high (75%) (Higgins et al., 2003). Either a fixed effect model or a random effect model was applied to synthesize the data, depending on the heterogeneity test. The fixed effect model was used when the heterogeneity test results in the included studies were P > 0.1 and I² < 50%; the random effect model was utilized when P < 0.1 and I² > 50%. Additionally, the sensitivity analysis was performed by eliminating each study separately in order to evaluate the results’ stability. To account for heterogeneity and determine whether treatment effects differ between subgroups based on the following criteria: (1) experimental animal models; (2) tanshinone type and dosage; (3) administration route; and (4) duration of the treatment. These subgroup analyses were selected based on previous main sources of heterogeneity in meta-analysis studies in animal experiments (Luís et al., 2023; Sheng et al., 2015).

Publication bias was assessed by funnel plot (Light et al., 1994) and Egger’s regression test (Egger et al., 1997), and the reliability of the results was tested by the trim and fill technique (Duval and Tweedie, 2000).

1,375 records have been identified in the database searches (Figure 1). 939 records remained after removing the duplicate records. 874 records that did not fulfill the inclusion criteria were eliminated after screening title and abstract; for example, most of them were removed because there were either no tanshinones or no animal model for AD. Subsequently, 14 studies met the inclusion criteria after the full-text analysis of the remaining 65 studies. Thus, this meta-analysis is based on 14 studies, comprising 26 platform crossing times comparisons, 28 time in target quadrant comparisons, and 37 escape latency comparisons. Half of the included researches were published in the last 10 years (2014–2024), with the earliest included study being published in 2009, suggesting that there has been a recent surge in interest in tanshinones’ impact on AD.

Of the total 14 included studies (Table 1) (He Y. Y. et al., 2020; Liao, 2020; Liao et al., 2019; Mei et al., 2010; Wan, 2022; Sheraz, 2019; Fang et al., 2021; Mei et al., 2009; Ma et al., 2022; Xiang et al., 2024; He Y. et al., 2020; Ding et al., 2020; Peng et al., 2022; Liu et al., 2016), half of them were published in academic journals in English, while 7 were published in Chinese (He Y. Y. et al., 2020; Liao et al., 2019; Mei et al., 2010; Fang et al., 2021; Liao, 2020; Wan, 2022; Sheraz, 2019), including 3 academic dissertations (Liao, 2020; Wan, 2022; Sheraz, 2019), which have been peer-reviewed and published. The most commonly utilized animal model of AD was the APP/PS1 double transgenic model (Mei et al., 2010; Wan, 2022; Mei et al., 2009; Ma et al., 2022; He Y. et al., 2020; Ding et al., 2020; Peng et al., 2022). Of the 14 studies, 3 used non-transgenic mice (He Y. Y. et al., 2020; Fang et al., 2021; Liu et al., 2016), 4 used Sprague Dawley rats (Liao, 2020; Liao et al., 2019; Sheraz, 2019; Xiang et al., 2024), and 7 used APP/PS1 double transgenic mice (Mei et al., 2010; Wan, 2022; Mei et al., 2009; Ma et al., 2022; He Y. et al., 2020; Ding et al., 2020; Peng et al., 2022). 11 studies used male animals only (He Y. Y. et al., 2020; Liao, 2020; Liao et al., 2019; Wan, 2022; Sheraz, 2019; Fang et al., 2021; Xiang et al., 2024; He Y. et al., 2020; Ding et al., 2020; Peng et al., 2022; Liu et al., 2016), occupying the vast majority. Equal numbers of male and female animals were utilized in one study (Mei et al., 2010). The sex of the animals utilized was not reported in the rest of studies (Mei et al., 2009; Ma et al., 2022). Regarding the route of administration, experimenters tended to utilize intragastric administration, representing 8 studies (He Y. Y. et al., 2020; Liao, 2020; Liao et al., 2019; Mei et al., 2010; Wan, 2022; Sheraz, 2019; Xiang et al., 2024; Liu et al., 2016). 5 studies (Fang et al., 2021; Ma et al., 2022; He Y. et al., 2020; Ding et al., 2020; Peng et al., 2022) used intraperitoneal injection, while only one study (Mei et al., 2009) used oral administration. Across all studies, as experimental treatments, 1, 5, 10, 15, 20, 25, 30, 40, 45, 50, 75, or 80 mg/kg/day of either tanshinone IIA or cryptotanshinone were given. The MWM was employed in every study to evaluate cognitive function.

The primary clinical sign of AD is cognitive impairment, and the most crucial way to track AD mouse models of cognitive function is to evaluate behavioral functions related to memory. MWM is a well-known experiment for assessing learning, memory, and cognitive function in animal models of AD (Lissner et al., 2021). It involves forcing experimental animals (mice and rats) to swim and learn to locate a hidden platform submerged in water. The escape delay in the spatial exploration experiment is a measure of acquisition memory, whereas platform crossing times and time in the target quadrant are used to assess the retention memory of mice, which shows their capacity to retain and retrieve memories (Othman et al., 2022).

Two tools were used to assess the quality of each study. Ten items were evaluated according to the CAMARADES checklist. With scores ranging from 5 to 7, the 14 included studies presented a quality score higher than 4, indicating that they were properly conducted. The 14 included studies are peer-reviewed publications and reported the presence of randomization of subjects into treatment groups, assessment of dose-response relationships, the statement of compliance with animal welfare regulations, and use of a suitable animal model. However, 2 studies (He Y. Y. et al., 2020; Fang et al., 2021) did not provide enough information to determine whether or not the avoidance of anaesthetic agents with marked intrinsic neuroprotective properties is fulfilled, and 5 studies (He Y. Y. et al., 2020; Liao et al., 2019; Mei et al., 2010; Fang et al., 2021; Mei et al., 2009) did not have the statements of potential conflict of interests. The median quality score indicated that 16 of the 21 secondary criteria had been reported fully. 3 studies (Liao, 2020; Liao et al., 2019; Xiang et al., 2024) did not provide the age of animals, and 2 studies (Mei et al., 2009; Ma et al., 2022) did not mention the sex of animals. The outcomes of the quality assessment are described in Tables 2, 3; Figure 2.

The methodology qualities of included studies were scores ranging from 5 to 7, indicating that some articles lacked rigorous animal experiment designs. Some methodological issues include: (1) Blinding: During surgical operations, the trial researchers were not blinded to the intervention that each animal received or to the physiological parameters being monitored. Double-blind research minimizes discrepancies in assessing the data and helps prevent the impact of the researcher’s subjective judgment on the findings. (2) Sample size: No study explained how to determine the sample size in order to verify that there was enough power. (3) Sequence generation: The precise sequence creation technique was not disclosed in any of the included investigations. In addition to preventing allocation bias, the creation of sequences in random assignment makes it easier to derive inferential inferences (Li et al., 2024).

As for each outcome, researchers considered all the experimental groups within each study instead of only a single representative group for each study, and the control group has been halved.

For reported escape latency, the results showed that tanshinones had a significantly beneficial effect on reduction of escape latency (SMD = −2.082, 95% CI = [−2.481, −1.683], p < 0.001), and the heterogeneity was significant (P < 0.001, I² = 58.9%). Thus, a random effect model was applied for this analysis. The forest plot is shown in Figure 3.

For platform crossing times, the results showed that tanshinones can increase the platform crossing times in AD animals [SMD = 1.464, 95% CI = (1.183, 1.744), p < 0.001]. A fixed effect model was applied for this analysis with low heterogeneity (P = 0.254, I² = 14.8%). The forest plot is shown in Figure 4.

For time in target quadrants, the global estimated effect of tanshinones was significantly beneficial [SMD = 2.703, 95% CI = (2.132, 3.275), p < 0.001], with significant heterogeneity among studies (P < 0.001, I² = 69.0%). Thus, a random effect model was applied for this analysis. The forest plot is shown in Figure 5.

Subgroup analysis was performed in order to determine the degree to which methodological differences between trials might have systematically influenced the variances in the principal treatment results. The stratifying characteristic is a substantial source of heterogeneity and could impact the treatment’s effectiveness, according to the general summary of each subgroup, which may then be analyzed for indicators of variance in the intervention’s effects (Amos et al., 2014). The findings of the subgroup analysis are described in Table 4.

To further investigate the sources of heterogeneity within studies, meta-regression analyses were performed for escape latency, platform crossing times, and time in target quadrants results. For time in target quadrants, type of tanshinones, dosage of tanshinones, duration of treatment contributed to the heterogeneity (p = 0.030, p = 0.045, and p = 0.040). For platform crossing times, the AD animal model contributed to the heterogeneity (p = 0.021) However, the above factors did not have a significant impact on escape latency results. Meta-regression analysis figures can be viewed in Supplementary Material S2 2.1–2.14.

Lastly, small-study effects were identified, which could be a factor in publication bias. Sensitivity analysis and funnel plots were performed. Sensitivity analysis were performed to explore how the results would alter if one or more studies were excluded in the meta-analysis. The findings demonstrated that excluding one or more studies did not significantly alter the effects of tanshinones on acquisition memory and retention memory (Supplementary Material S2 3.1–3.3). Funnel plots showed an asymmetry for escape latency (Figure 6A), platform crossing times (Figure 6B) and time in target quadrants (Figure 6C) data, which illustrates possible negative results in the included studies were not reported (Egger regression, p < 0.001, p < 0.001 and p < 0.001, respectively). Considering the presence of significant publication bias, the trim and fill method was applied to verify the impact on results. The results suggest that the results of the meta-analysis are stable [SMD = −1.965, 95% CI = (−2.340, −1.590), p < 0.001; SMD = 1.286, 95% CI = (0.943, 1.623), p < 0.001; and SMD = 2.466, 95% CI = (1.780, 3.152), p < 0.001] and that publication bias has little effect on the results of this study.

Every study that was included offered comprehensive details regarding potential tanshinones therapy mechanisms in animal models of AD. All of the suggested mechanisms are included in Table 5.

For thousands of years, Chinese people have been treated with herbal formulae. Currently organized alongside the traditional biomedical approaches of western medicine, traditional Chinese medicine is an essential component of the Chinese mainstream healthcare system. In traditional Chinese medicine theory, tanshinones effectively improve the stagnation of qi and blood in the patient’s body and restore the function of internal organs and meridians (Pharmacopoeia of the People’s Republic of China, 2020). But there is still not yet adequate evidence of tanshinones as a treatment program for AD. In order to assess the effectiveness of tanshinones treatment for AD and to comprehend the underlying mechanisms, we included 14 animal investigations. Tanshinones have neuroprotective properties that improve cognitive outcomes in AD, according to the findings of this systematic review and meta-analysis.

Our analysis further indicates that the design of the original research has an influence on the research results. Primarily, the effect values of transgenic mice were higher than those of non-transgenic mice and rats, indicating that animals of different modeling methods may have different responses to tanshinones. The possible reasons are as follows: Transgenic mice might be genetically engineered to express certain genes that mimic human diseases or conditions, such as AD. In the case of tanshinones, which are bioactive compounds derived from the Salvia miltiorrhiza plant, their efficacy may be heightened in transgenic models because these mice exhibit disease-specific pathways and biomarkers that are more similar to human conditions. This enhanced relevance of disease models makes transgenic mice more sensitive to treatments like tanshinones, which may have targeted effects on the molecular pathways affected by diseases like Alzheimer’s (He Y. et al., 2020). In contrast, non-transgenic mice or rats do not carry the genetic alterations associated with such diseases, which might lead to less pronounced effects when treated with tanshinones. Without these disease-specific markers, the compounds might not engage with the molecular pathways in the same way, thereby resulting in weaker or less significant effects (Li et al., 2016). Another factor that might contribute to the difference in effectiveness between transgenic and non-transgenic mice and rats is the pharmacokinetics of tanshinones. The absorption, distribution, metabolism, and excretion of a compound like tanshinone can vary between different species and even between genetically modified and non-modified animals. Transgenic mice might metabolize the compound in a way that enhances its bioavailability or its ability to reach the target sites in the brain. Non-transgenic animals, on the other hand, might not exhibit the same level of absorption or target site engagement, potentially leading to a lower therapeutic effect (Song et al., 2023).

Compared to Tanshinone IIA treatment, Cryptotanshinone treatment enhanced memory retention. While most of the studies used Tanshinone IIA, the effect values of Cryptotanshinone were probably affected by the experimental design since the studies were conducted by the same author. The lack of observation of the significance of the effect of Cryptotanshinone in retention memory probably resulted from the reduced number of studies using this treatment. It would be desirable to increase the number of studies using the Cryptotanshinone treatment; however, no more studies using this model were found that met the inclusion criteria.

Considering the dosage of tanshinones also affects the therapeutic effect. A 25 mg or lower dosage was associated with a greater beneficial effect on acquisition memory, while a dosage higher than 50 mg had a greater beneficial outcome for retention memory. The differential effects of tanshinones at varying dosages may be attributed to several pharmacological mechanisms. Lower dosages might facilitate the balance of neurotransmitters, such as acetylcholine, which is crucial for memory formation and retrieval. Higher doses, conversely, may lead to an oversaturation of these pathways, impeding optimal neurotransmitter function (Liao et al., 2020). Moreover, low doses of tanshinones can exhibit anti-inflammatory properties, which are essential for protecting neural tissues from the chronic inflammation typical in Alzheimer’s pathology. Higher doses, however, can trigger compensatory mechanisms that might exacerbate inflammation (Song et al., 2023).

Considering the administration route of tanshinones, intragastric administration produced a smaller effect than either intraperitoneal or oral administration of tanshinones. The possible reasons are as follows: The ability of tanshinones to reach the brain is crucial for their effects on cognitive performance, particularly in Alzheimer’s disease models where neurodegeneration and cognitive decline are key features. The blood-brain barrier (BBB) is a selective permeability barrier that controls the entry of substances into the brain. Certain administration routes, such as intraperitoneal and oral administration, may lead to a more efficient transport of tanshinones across the BBB, especially if the compounds are small, lipophilic, or have specific transport mechanisms that facilitate their passage. Intraperitoneal administration allows tanshinones to enter the bloodstream more rapidly and distribute throughout the body. The fast distribution increases the likelihood of tanshinones reaching the brain in higher concentrations, where they can exert their effects on neuronal function and cognitive processes. Although oral administration does involve absorption via the gastrointestinal tract, certain formulations can enhance the bioavailability and brain penetration of compounds. For instance, oral formulations like nanoparticles, liposomes, or modified-release systems can improve the absorption of tanshinones and increase their bioavailability, enhancing their effects on the brain. By contrast, intragastric administration may result in slower and less efficient absorption, meaning lower concentrations of tanshinones in the bloodstream and consequently, lower concentrations reaching the brain. This can result in diminished effects on cognitive performance (Zhang, 2019).

Studies with a 60-day duration or longer demonstrated a stronger protective impact on both acquisition and retention memory, according to our systematic review. The possible reasons are as follows: Long-term tanshinone treatment may increase the production of neurotrophic factors such as brain-derived neurotrophic factor (BDNF). These factors play a pivotal role in promoting neuronal survival, growth, and synaptic function. Sustained elevation of BDNF levels could enhance memory acquisition and retention over time. Another factor is synaptic plasticity: Prolonged exposure to tanshinones could facilitate enhanced synaptic plasticity, which is essential for memory processes. Long-term treatment may lead to structural changes in the brain, such as increased dendritic branching and synapse formation, both of which contribute to improved memory acquisition and retention in AD models (Meier et al., 2020).

Given the limited alternatives available for treating AD, the results of this meta-analysis are interesting since they demonstrate the considerable anti-cognitive impairment impact of tanshinones against AD in animal models, raising the possibility that treating AD patients with tanshinones may improve their cognitive impairment and emphasizing the need for more research, including human clinical trials.

The following were some of the limitations of the system review: (1) The meta-analysis’s limited selection of literature may have influenced its findings, which may have a certain bias in the conclusion. (2) We only assessed how tanshinones affected AD animals’ cognitive deficits. Due to a lack of data, we did not perform analyses specifically aimed at the impact on histology, such as plaques and tangles. (3) Due to language barriers, we failed to search for studies published in other languages, such as Japanese, and instead only searched databases for publications published in Chinese or English. We might have overlooked any pertinent papers because Japan is one of the major Danshen nations. (4) The literature lacked descriptions of experimental animals’ adverse reactions.

The number of preclinical experiments conducted annually continues to rise, and our understanding of the disease mechanism is improving. However, because of the translational paradigm’s shortcomings, the number of innovative treatments for AD that make it to the clinic keeps declining (Sena et al., 2014). Experts from a variety of scientific disciplines advocate the standardization of animal protocols and the systematic review of animal models that do not currently qualify as predictive modalities for human responses to drugs and disease (Greek and Menache, 2013). Therefore, these limited results may not be sufficient for the transition from animal experiments to human clinical trials. Consequently, high methodological reporting and quality control experimental trials are required to properly assess the impact of this intriguing pharmaceutical intervention before any clinical practice recommendations are performed.