Therapeutic Potential of Curcumin on the Cognitive Decline in Alzheimer’s Disease: A Study Integrated Meta-Analysis, Network Pharmacology, and Molecular Docking

Background: Curcumin, a polyphenol derived from the herb turmeric, has emerged as a promising potential therapy in the management of Alzheimer’s disease (AD). However, the e�cacy and potential therapeutic mechanisms remains largely unknown. Objective: To systematically meta-analysis the e ﬀ ect and to investigate the potential pharmacological mechanisms of curcumin on cognitive de�cits in AD. Methods: A systematic collection of curcumin studies was performed from MEDLINE’s database, PubMed, Web of Science, and Google Scholar until October 31th, 2020. Following quality assessment of study eligibility, strati�ed meta-analysis and meta-regression analyses were undertaken to recognize and control the heterogeneity in meta-analysis. An integrated network pharmacology and molecular docking approach were applied to decipher the potential pharmacological mechanisms of curcumin on AD. Results: A meta-analysis of 29 publications showed that curcumin exerts signi�cantly positive e ﬀ ects on cognitive performance. For acquisition, the global estimated effect of curcumin was -2.027 (95% CI: -2.435 to -1.619, p<0.001); For retention, the global estimated effect of curcumin was 1.606 (95% CI: 1.101 to 2.111, p<0.001). Strati�ed meta-analysis demonstrated that an increased effect size depended on various study characteristics. Network pharmacology analysis identi�ed 63 genes targets, and STAT3, CHEK1, AKT1, EGFR, MMP9, hsp90AA1, and EP300 were core target proteins. Molecular docking showed that curcumin can closely bind with these seven targets. Besides, 69 potential pathways of curcumin were identi�ed, like nitrogen metabolism. Conclusions: Our �ndings suggested that curcumin may reduce cognitive de�cits in AD through multi-target and multi-pathway mechanism, providing a scienti�c basis for further experimental and clinical application.


Introduction
Alzheimer's disease (AD), a neurodegenerative illness, is characterised by deterioration of cognitive functions and behavioural changes that eventually lead to dementia.According to the statistics, an estimated 30 million people have AD today globally, and the number of people living with AD are predicted to increase to 100 million by 2050 [1].The increasing prevalence of AD has resulted in substantial burdens at the personal, social, and economic levels.However, despite substantial efforts, there are no effective therapeutic strategies currently available for AD [2].
Curcumin, a polyphenol derived from the herb turmeric(Curcuma longa), is a multifunctional tissueprotecting agent that exerts anti-in ammatory, antioxidant, antiproliferation, chemo preventive, and chemotherapeutic effects [3].Indeed, curcumin demonstrated as a promising potential therapy in the treatment of AD.The main mechanism of action by which curcumin modi es AD pathology has been summarised as follows [4]: curcumin inhibits the formation and ameliorates the cytotoxicity of amyloid-β (Aβ) brils and attenuates the hyperphosphorylation of tau brils and enhances their clearance.However, a thorough review provided evidence that curcumin did not affect the Aβ aggregation in vitro and inhibit tau bril formation.According to the study authors, more than 120 clinical trials have not established a clear and direct causal effect of curry consumption on improving cognitive function [5].Not surprisingly, this manuscript contributed to the major controversy about curcumin in Nature.A supportive report by M. Baker de ned curcumin as a "chemical liar" under "gold armor" [6].Then, another short essay co-authored by multiple scientists noted that the researchers missed very important pharmacological or clinical data and that the pharmacological and medicinal values of curcumin were misinterpreted [7].
Systematic reviews and meta-analyses of animal experiments would offer a sensible and rational method for evaluating the translational potential of promising experimental interventions before decisions are made to proceed with clinical trials [8].Network pharmacology attempts to understand drug actions and interactions with multiple targets in a holistic manner.It helps in systematically explaining the role of a drug or drugs in disease treatment, and possess obvious advantages over conventional methods in the illustration of comprehensive mechanisms.In this study, a meta-analysis conducted here is expected to provide guidance for animal and clinical studies on curcumin intervention in AD.Furtherly, network pharmacology combined with molecular docking were applied to explore the potential molecule targets and mechanism of action, in order to have a deeper understanding of the intervention effect of curcumin in treating AD.

Search strategy
To identify animal studies describing the effect of curcumin in AD, a literature search was conducted based on databases, including MEDLINE's database, PubMed, Web of Science, and Google Scholar until October 31th, 2020.Search terms included (curcumin OR curcuminoids) AND (dementia OR cognitive impairment OR AD OR senile dementia).

Screening criteria
Inclusion criteria: (1) experimental AD was induced in rodents.(2) an AD treatment group was treated with curcumin, and a control group of animals was treated with placebo.(3) cognitive function was measured by the Morris water maze (MWM), and escape latency time was used to represent acquisition (learning) and/or time in the target quadrant in the retention test was used to represent retention (memory).
Exclusion criteria: (1) curcumin was not administered, or the treatment group was administered another neuroprotective agent in addition to curcumin.(2) only biochemical or physiological outcomes of treatment e cacy were assessed.(3) duplicate publications.

Data extraction
Two investigators extracted details of individual study characteristics, including reference details (publication year and name); recipient animal (rat strain, gender); AD model; main experimental groups and quantity; substances used as experimental and control treatments; method/dose/timing of curcumin administration; and time of cognitive outcome assessments.GetData software was used for obtaining data at the nal time point in the study of behavioural testing.

Methodological quality
The methodological quality of study was assessed using a 13-point project quality checklist as previously described with minor modi cations [9], which was based on the CAMARADES (Collaborative Approach to Meta Analysis and Review of Animal Data from Experimental Studies) quality checklist.

Statistical analysis
The global estimated effect of curcumin on cognitive outcome was determined using the standardised mean difference (SMD) and a 95% con dence interval (CI).Within-and between-study variation or heterogeneity was assessed using Cochran's Q-statistic, with a signi cant Q-statistic (p < 0.10) indicating heterogeneity among studies.Quanti cation of heterogeneity was calculated using the I 2 metric and was represented on a scale ranging from 0-100%.Funnel plots (visually assessing for asymmetry) and Egger's test (p < 0.05 indicating the presence of a small effect size) were selected for the assessment of potential publication bias.STATA version 15.0 was used for statistical analyses.

Network pharmacology to predict the potential mechanisms 2.2.1 Prediction of targets
The molecular target of curcumin was obtained from The Traditional Chinese medicine system pharmacology technology platform (TCMSP, http://tcmspw.com/tcmsp.php).AD targets obtained from the GeneCards (https://www.genecards.org/) and Online Mendelian Inheritance in Man (OMIM, http://www.omim.org/)databases.

Enrichment analysis
To explore the gene functions, Gene Ontology (GO) and Kyoto Gene and Genomic Encyclopaedia (KEGG) pathway were conducted and visualized using the R software (Version 3.26).Only functional annotations with an enrichment P-value less than 0.05 were chosen for further analysis.

Protein-protein interaction (PPI)
The predicted targets of curcumin toward AD were submitted to the STRING database (http://stringdb.org/cgi/input.pl)for PPI analysis.In this study, the obtained target was submitted to STRING, with the species limited to "Homo sapiens" and con dence scores limited to those > 0.4, and the PPI data and network were extracted.

Network Construction
The screened target and pathway les were input into Cytoscape 3.7.2 to construct the "componenttarget-pathway" network diagram of curcumin.In the network, component, targets and pathway are represented by nodes, and the interaction between two nodes is represented by edges.

Molecular docking analysis
The protein crystal structure was derived from the Protein Data Bank (PDB, http://www.rcsb.org/)and imported into the Molecular Operating Environment software for structure construction, as previously described [10].The speci c steps to construct the protein structure include removal of water, protonation, and energy minimisation, etc.Then, the protein structure was matched to the three-dimensional structure of the curcumin which downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/).

Study selection and quality evaluation
A total of 29 publications, including 40 comparisons of acquisition and 22 comparisons of retention, were assessed in the meta-analysis (Figure 1).Relevant characteristics of these trials, including recipient animal, AD model, main experimental groups and quantity, experimental and control treatments, method/dose/timing of curcumin administration, and time of cognitive outcome assessments are described in Table S1 [11-16] [17-21] [22-26] [27-30] [31-36] [37, 38].The quality evaluation of those studies was listed in Table S2.The median quality score was 8 items of 13 (range 5-10).

Meta-regression
Meta-regression is an extension of subgroup analysis that allows investigation of the effect of continuous as well as categorical characteristics.Hence, a meta-regression and subgroup analyse by animal model (species, strain, gender, model), study design (quality score), and intervention characteristics (delivery route) was performed.For acquisition, heterogeneities were independent of these factors (Table 1).For retention, species differences were signi cant sources of heterogeneity (p<0.05)(Table 2).Note *p value for subgroup differences; **p value for heterogeneity between subgroups with metaregression analysis

Publication bias
The funnel plots were applied to identify whether there was publication bias in these studies.As shown in Figure 3a and Figure 3b, asymmetry was identi ed in both funnel plots of acquisition and retention functions, suggesting publication bias existed.Furthermore, Egger's tests were created to statistically detect asymmetry, with a null hypothesis denying the existence of small-study effects.A p-value of <0.05 was found in both acquisition and retention data, evidently indicating to reject the null hypothesis in favour of the alternative (i.e.publication bias does exist in these studies).

Predicting the targets of curcumin
A total of 7945 AD-related genes were searched from the disease databases.After matching them with the predicted curcumin target, 63 anti-AD targets were obtained.The PPI network containing 63 nodes and 231 edges were building.In this network (Figure 4), the nodes, including AKT1, EGFR, MMP9, HSP90AA1, STAT3, CHEK1, and EP300 had higher degree.That's means, they had a closer relationship with the more surrounding genes (37 for AKT1, 33 for EGFR, 24 for MMP9, 21 for HSP90AA1, 21 for STAT3, 14 for CHEK1, 14 for EP300).The network results revealed that the above genes may play an important role in the mechanism of curcumin in the treatment of AD.

Function enrichment analysis and compound-target-pathway networks
The bar graph of GO enrichment terms for the identi ed genes showed that one-carbon metabolic process (GO:0006730), vesicle lumen (GO:0031983), and carbonate dehydratase activity (GO:0004089) were the most highly enriched biological processes, cell component, and molecular functions, respectively.The top twenty entries in each GO category were listed in Figure 5 (A-C).The bar graph of KEGG enrichment terms for the identi ed genes showed that nitrogen metabolism (hsa00910), prostate cancer (hsa05215), EGFR tyrosine kinase inhibitor resistance (hsa01521) were the most highly enriched signaling pathways.The top 20 pathways were listed in Figure 5D.

Compound-target-pathway network construction
Cytoscape software was used to construct "component-target-pathway" network model for exploring the potential mechanism by which curcumin treats AD.The relationship among the compound, targets, and pathways are shown in Figure 5.The network revealed that curcumin exert its therapeutic effect on AD through multiple targets and multiple pathways.
Curcumin formed an arene-H conjugation with Glu316 in STAT3 (Figure 8A).As shown in Figure 8B, curcumin formed one arene-H conjugation and one arene-cation conjugation with Leu15 and Lys132 in CHK1, respectively.As shown in Figure 8C, curcumin formed two hydrogen and arene-arene conjugation with Asp80, Trp80, and Ser205 in EGFR, respectively.As shown in Figure 8D, curcumin formed one hydrogen and two arene-H conjugation with Asp93, Asn51, and Leu107 in HSP90AA1, respectively.As shown in Figure 8E, curcumin formed one hydrogen and one arene-arene conjugation with Lys268 and Asp292 in AKT1, respectively.As shown in Figure 8F, curcumin formed two hydrogen, arene-arene, and arene-H conjugation with Leu188, ZN301, His226, and Tyr248 in MMP9, respectively.As shown in Figure 8G, curcumin formed one hydrogen and arene-H conjugation with Gln1455 and Tyr1414 in EP300, respectively.Energy docking scores, an indicator that re ects the binding ability of compound and proteins, were listed in Table 3.These results indicated that curcumin has a good ability to bind to target genes.

Discussion
natural polyphenol curcumin has been demonstrated as a promising potential therapy to treat AD.
However, there have been researches contributed to the major controversy about curcumin in related elds.In these studies, the meta-analysis results revealed that curcumin has a certain treatment effect on AD.Then, the network pharmacology strategy was applied to investigate the mechanisms of curcumin on AD treatment.Furtherly, molecular docking provides reliably molecular binding evidence for network pharmacology research results.The use of meta-analysis, network pharmacology methods in combination with molecular docking for the rst time enabled an investigation of the e cacy and potential underlying biological mechanisms of curcumin when being used in the treatment of AD.
In preclinical studies of AD, curcumin was shown to inhibit the formation of tau brils and Aβ brils and improve cognitive function in experimental treatment groups.Additionally, phase II clinical trials studying the tolerability and e cacy of oral curcumin were reported [39].This is the rst meta-analysis that systematically collected and evaluated the current preclinical evidence supporting the e cacy of curcumin in experimental AD in relation to improve learning and memory ability.Though small-study effects and statistical heterogeneity among studies exist, these ndings revealed a bene cial effect of curcumin in experimental AD and strengthened the evidentiality in terms of supporting curcumin-based therapy for experimental AD.
The strati ed analysis detected a signi cant influence of animal gender, animal strain, animal species, animal model, drug delivery route, and study quality.For example, we found that studies that did not use the most widely used animal models.The results above should be applied to provide a reference for future research when ascertaining the curcumin characteristics for successful outcomes.The major cause of heterogeneity, we found in the treatment effects between the study groups, is the potential bias of studies selected for analysis.The results of meta-regression showed acquisition and retention outcome heterogeneity was lower than would be expected by chance.Moreover, the methodological quality of each study were assessed [40].Studies with lower scores showed a trend towards better learning and memory outcomes.The global estimated effect of curcumin on cognition may be overstated in low-quality studies.
The network pharmacology methodology was then used to predict the molecular targets and the potential pathways modulated by curcumin during AD treatment.Speci c, key targets, including AKT1, EGFR, MMP9, HSP90AA1, STAT3, CHEK1, and EP300 were found to play potential pivotal roles in the anti-AD action of curcumin.Transcriptional regulation of Akt, related to the protein fragments involved in learning and memory, could directly has led to the impacts upon the central nervous system function, and highly correlated with AD [41].Moreover, Akt1 has been mapped to be a susceptibility gene on AD, and Aβ re ects an inhibition-activation effect toward AKT1 in neurons and then accelerate the progression of the disease [42].Meanwhile, it is revealed that curcumin possesses a protective effect on neurotoxicity and memory de cits and its mechanism is closely linked to the regulation of the JNK/NF-κB/Akt signalling pathway [43].Current studies suggested over expression of EGFR in astrocytes [44], and it is closely related to AD and aging nerve metabolic disorders [45].In addition, EGFR inhibitors play successful therapeutic effects in improving pathologic and behavioural conditions in neurodegenerative diseases, especially in AD [43].There is a signi cant higher level of MMP9 expression in plasma and brain tissue in AD patients [46].MMP9 augmentation in patients with AD may lead to a more vulnerable blood-brain barrier, inhibit the activity of nerve growth factor, and induce neuronal in ammatory reactions, which eventually cause nerve function defect in patients with AD [47].HSP90AA1 is involved in oxidative stress, cell apoptosis, and mitochondrial dysfunction [48].Upregulation of its expression can inhibit the aggregation and denaturation of protein [49], increase calcium current and neuronal excitability, inhibit amyloid plaque formation, and then exert an positive effect on spatial learning ability in AD mouse model [50].Besides, other three targets, especially STAT3, are also closely associated with abnormal synaptic plasticity, neurogenic oxidative damage, neurogenic in ammation, and other typical neurodegenerative diseases pathological features [51].Subsequently, the recognitions and interactions of curcumin with these seven crucial targets were studied via molecular docking method.The biological activity of curcumin was predicted reasonably well, and it was found that curcumin have a good binding a nity with gene targets (AKT1, EGFR, MMP9, HSP90AA1, STAT3, CHEK1, and EP300).Through network pharmacology and molecular docking, we can screen out the key genes which played an important role in the anti-AD action of curcumin.
The KEGG analysis identi ed the nitrogen metabolism pathway was the most signi cant in the potential anti-AD pathway of curcumin.The urea cycle, which converts excess systemic nitrogen from the decomposition of nitrogen-containing metabolites into urea, is strongly implicated in the pathogenesis of AD [52].According to the requirements of cells for urea cycle intermediates, the body expresses different urea cycle enzymes.Previous studies have con rmed that all the genes encoding for urea cycle enzymes are expressed in the brains of AD patients [53].Disorders in retrograde messenger nitric oxide (NO) are frequently observed in AD.Beta-amyloid protein can triggers induce a series of NO-mediated reactions, including excitotoxicity, neuroin ammation and oxidative stress [54].In this study, the GO analysis suggested that the biological processes of potential targets were mainly re ected in responding to amyloid-beta.Growing evidence suggests that NO synthases (NOSs) is closely related to the pathophysiology of AD.Two constitutive isoforms of NOSs are critically involved in the synaptic plasticity [55], and the pathological role of iNOS in AD immunoreactivity was known by the people [56].
Besides, beta-amyloid protein induce excitotoxicity through increases in intracellular calcium and subsequent activation of nNOS [57].NOSs substrates, including amino acids L-arginine and Lcitrulline, are important cellular metabolites.Their metabolic disorder may seriously affect the pathophysiology of AD [58].GO analysis results showed that the molecular functions were mainly re ected in the protein serine/threonine kinase activity, which are also associated with nitrogen metabolism to a certain extent.Thus, further pharmacological experiments for veri cation of the related pathway are needed.

Conclusions
The study combined meta-analysis, network pharmacology, and molecular docking to determine and elucidate the therapeutic mechanism of curcumin in the treatment of AD.The ndings suggest that curcumin has promising therapeutic potential on learning and memory recovery in animal models of AD.However, more well-designed and well-reported experimental animal studies are needed.The core genes targeted by curcumin toward AD were: AKT1, EGFR, MMP9, HSP90AA1, STAT3, CHEK1, and EP300.This nding has been proven by the molecular docking analysis.Likewise, curcumin was found to be involved in AD treatment by targeting nitrogen metabolism.Based on the results of network pharmacology and molecular docking, we will further reveal its mechanism of action through in vivo e cacy experiments.

Figures
Figures

Figure 1 Flow
Figure 1

Figure 4 Protein
Figure 4

Figure 5 Function
Figure 5

Table 2 .
Strati cation of estimated effect size for retention memory.

Table 3
The docking score between the curcumin and core targets