Inflammatory Proteins Mediate the Effect of Gut Microbiota on Graves’ Ophthalmopathy: A Mendelian Randomization Study

Abstract

Purpose: This study investigates the causal relationship between gut microbiota (GM) and Graves’ ophthalmopathy (GO) and explores the mediating role of circulating inflammatory proteins (IPs) in this association.

Methods: A two-step, two-sample Mendelian randomization (MR) analysis was performed using GM data from MiBioGen (N = 18,340), GO data from the FinnGen Research Project (691 cases, 411,490 controls), and IP data from a genome-wide association study (N = 14,824). The primary MR analysis utilized the inverse variance-weighted approach, supplemented by MR Egger, weighted median, maximum likelihood, and MR robust adjusted profile score methods. Mediation MR was used to assess the mediating role of IPs.

Results: We identified 26 GM taxa causally associated with GO. Notably, the genus Parabacteroides exhibited a protective effect on GO (odds ratio = 0.29; 95% confidence interval, 0.16–0.51; P < 0.001). Additionally, five circulating IPs demonstrated protective effects, while three IPs (CXCL10, CXCL11, EN-RAGE) were associated with increased GO risk. Mediation MR showed that CXCL10 mediated the pathway from Parabacteroides to GO (mediation effect = −0.07), accounting for 5.27% of the total effect.

Conclusions: This study supports a causal link between GM and GO, mediated by circulating IPs. These findings offer new insights into GO pathogenesis and potential targets for clinical intervention.

Translational Relevance: Our findings reveal how gut microbiota influences Graves’ ophthalmopathy through inflammatory proteins, providing potential therapeutic targets for disease prevention and treatment.

Introduction

Graves’ ophthalmopathy (GO), also known as thyroid eye disease, is an autoimmune disease involving the extraocular muscles (EOMs), orbital fat, and connective tissue.¹ The prevalence of GO varies by gender, race, and geographical location and shows an increasing trend worldwide due to population aging and environmental changes.^2,3 GO is strongly associated with thyroid autoimmune disorders, namely Graves’ disease (GD), with 80% to 90% of GO occurring in patients with hyperthyroidism.⁴ The typical natural course of GO involves an initial acute inflammatory phase followed by a chronic fibrotic phase, primarily affecting orbital soft tissues and EOMs.⁵ Ocular manifestations are intricate, encompassing eyelid retraction, proptosis, diplopia, and dysthyroid optic neuropathy.^6,7 GO is a disabling and disfiguring disease with facial deformities and visual impairments, which not only severely affects patients’ daily lives but also imposes a heavy burden on health care systems and society.^8,9 Patients often require long-term medication or repeated surgical procedures, but these therapies do not always achieve ideal outcomes and may be accompanied by severe side effects and complications.¹⁰ It is thus crucial to identify the causes and underlying mechanisms of GO to guide targeted and effective prevention and intervention efforts. 

The pathogenesis of GO remains incompletely elucidated, with current research focusing on immune modulation, cytokine signaling pathways, and genetic susceptibility.^3,11 Recently, increasing evidence suggests that gut microbiota (GM) may also play a role in the onset and development of GO.¹² GM, encompassing bacteria, fungi, and viruses, constitutes one of the richest microbial communities in the human body. It plays a crucial role in maintaining host physiological homeostasis and metabolism, promoting nutrient absorption, regulating immune function, and combating infections.¹³ Immune-mediated inflammatory responses play a central role in the pathophysiology of GO, primarily through autoantibody activation of orbital fibroblasts, triggering inflammatory infiltration, fibrosis, and adipogenesis. Both GM and inflammatory proteins (IPs) influence GO development, with GM modulating circulating IPs by altering intestinal barrier integrity, producing metabolites, and regulating immune cell function. Therefore, we hypothesize that IPs serve as mediating factors in the pathway from GM to GO.¹⁴ Dysbiosis of GM has been implicated in the occurrence and progression of autoimmune diseases, with studies showing a critical bidirectional signaling axis between GM and the thyroid.¹⁵ Cross-sectional studies have demonstrated that patients with GO exhibit significantly reduced GM diversity and distinct alterations in microbial composition compared to healthy controls.^16–19 Moreover, the incidence and severity of GO may be related to the specific compositions of GM.²⁰ These findings collectively suggest that GM may play a significant role in the pathogenesis of GO, although the exact mechanisms remain unclear. 

Current research on the association between GO and GM mainly focuses on investigating the composition and functional traits of GM in fecal samples of patients with GO, as well as the correlations between GM traits and GO clinical manifestations.^14,17,21 Additionally, researchers are delving into how GM modulates immune function and affects the development and progression of GO.^18,19 However, most evidence is based on observational study designs, which are subject to potential confounders and reverse causation.²² Furthermore, differences in GM do not always imply pathogenic mechanisms of GO, as these changes could be a consequence of GO. Although randomized controlled trials can help establish causality between GM and GO, they are challenging to conduct in humans due to practical constraints such as strain selection and low cytokine levels. 

To address these gaps, we conducted a Mendelian randomization (MR) analysis to elucidate the causal relationship between the GM and GO. MR is a genetic epidemiologic method using genetic variants as instrumental variables to investigate the causal relationship between exposure and outcome.²³ By utilizing the random distribution characteristics of genotypes and exposure factors existing in nature, MR can simulate the effects of randomization on study subjects, effectively reducing bias due to reverse causation and confounding factors.²⁴ In this study, we applied a two-step, two-sample MR analysis to explore the causal link between GM and GO. Additionally, we investigated whether such an association was mediated by IPs, which play a significant role in both GM and GO.^25,26 By analyzing large-scale genomic data and clinical information, we aimed to understand how GM influenced the onset and progression of GO while elucidating the mechanisms involving IPs. Our findings would yield novel insights into the pathogenesis of GO and inform new strategies for its prevention and treatment. 

Materials and Methods

Study Design

We employed a two-sample, two-step MR approach using genome-wide association study (GWAS) summary data to reveal the causal association between GM and GO and to ascertain whether circulating IPs could mediate this association. The study design adheres to the three fundamental assumptions of MR analysis²⁷ displayed in Figure 1: (1) correlation assumption: robust associations between instrumental variables (IVs) and the pertinent exposure; (2) independence assumption: absence of correlations between IVs and any potential confounders; and (3) exclusivity assumption: correlation of IVs with the outcome through exposure alone. This study did not require further ethical approval because it was based on a secondary analysis of published data. Furthermore, study-specific ethical approvals were provided in the original article. All included studies followed the guidelines in the 1975 Declaration of Helsinki, ensuring ethical compliance. Furthermore, this study strictly adhered to the Strengthening the Reporting of Observational Studies in Epidemiology using MR guidelines.²⁸ 

Figure 1.

 

A two-step Mendelian randomization study of GM on GO mediated by inflammatory proteins.

Data Sources

Exposure Data

Summary information on the data sources and sample sizes used in this study can be found in Supplementary Table S1. Genetic variants linked to GM composition have been elucidated through the international consortium MiBioGen.²⁹ This study, encompassing 18,340 individuals predominantly of European ancestry from 24 cohorts across 11 countries, represents the largest genome-wide meta-analysis published to date. The consortium coordinated 16S rRNA gene sequencing profiles and genotyping data to analyze microbial composition. Through microbiota quantitative trait loci mapping analysis, they identified host genetic variants associated with the abundance of gut bacterial taxa. This meticulous curation process yielded a data set of 211 taxa containing 122,110 variant sites (Ebi-a-GCST90016908 to Ebi-a-GCST90017118). After excluding 15 unclassified taxonomic units, the data set comprised 196 bacterial taxa across 9 phyla, 16 orders, 20 classes, 32 families, and 119 genera. 

Outcome Data

Summary data on GO were obtained from the GWAS, comprising 411,490 controls and 691 cases with a mean age at the first event of 51.35 years and a female-to-male ratio of 4.44. These data were sourced from the latest R10 release of the FinnGen database,³⁰ published in November 2022. Phenotypic selection in this investigation was based on the International Classification of Diseases, 10th revision. The GO GWAS data set specifically targeted individuals with the phenotype “Graves ophthalmopathy” identified by the phenocode “GRAVES_OPHT.” Further details on FinnGen can be accessed at https://r10.risteys.finngen.fi/endpoints/GRAVES_OPHT. 

Mediator Data

Data on genetic variables for 91 circulating IPs were obtained from a meta-analysis summary study, which examined genome variant associations in 14,824 participants of European ancestry across 11 independent cohorts. The original study thoroughly describes the experimental methods for measuring IPs.³¹ The GWAS summary statistics for each protein are available at https://www.phpc.cam.ac.uk/ceu/proteins and in the European Bioinformatics Institute GWAS Catalog under accession numbers GCST90274758 to GCST90274848. 

Instrumental Variable Selection

Using GO as the outcome, GM as the exposure, and circulating IPs as the mediators, we utilized the following criteria to select candidates for genetic IVs for MR analysis: (1) the significance threshold was set as P < 5 × 10^–8 for single-nucleotide polymorphisms (SNPs) associated with GO, P < 5 × 10^–6 for SNPs associated with IPs when there were limited IVs available, and 1 × 1 0^–5 for SNPs associated with GM to retain the largest explained variance on microbial features. (2) We retained SNPs with the most significant P values by calculating the linkage disequilibrium (clumping window size > 10,000 kb, r² < 0.001).²⁹ (3) We excluded SNPs with a minor allele frequency (MAF) of ≤0.01 and disregarded palindromic SNPs during the harmonization of exposure and outcome data. (4) We removed potential confounders and bypassed SNPs (e.g., smoking behavior, diabetes mellitus, other diseases) by using the phenoScanner tool.³² (5) The F statistic³³ was applied to screen for SNPs to mitigate the potential bias arising from weak IVs using a cutoff of 10. The formulae for F and R² are as follows³⁴: F = (N – K – 1) / K × R²/ (1 – R²), R² = 2 × MAF × (1 – MAF) × β². N is the sample size, K is the number of instruments, MAF is the frequency of the less common allele in a specific genetic variant within a population, and R² is the percentage of exposure variance explained by genetic variants. The MR analysis determines the estimated causal influence of an exposure variable on an outcome variable, which is represented by β. 

Statistical Analysis

All statistical analyses were carried out using the “TwoSampleMR” package³⁵ and “MRPRESSO” packages³⁶ in R statistical software (version 4.3.1; R Project for Statistical Computing, Vienna, Austria). The primary analysis chosen for this study was the inverse variance weighted (IVW),^37,38 which assumes that each IV is valid and assigns weights based on the inverse of the outcome variance to provide the most accurate estimation of causal effects. Additionally, we employed the following supplementary methods to assess causal relationships: MR-Egger, weighted median, maximum likelihood, and robust adjusted profile score (MR-RAPS). The P values of the IVW test were adjusted using a Benjamini–Hochberg false discovery rate (FDR), and the resulting q-value (FDR) threshold was set at 0.05.³⁹ A significant causal effect is determined by a q-value of <0.05 in the IVW test and further confirmed by the consistency in the directional estimates across the supplementary methods. The weighted median⁴⁰ method can offer robust and consistent estimates when up to 50% of the genetic instruments are invalid but with lower efficiency and precision than IVW. Maximum likelihood estimation⁴¹ based on probability theory maximizes the chance of detecting unknown parameters and is frequently used to evaluate the results of the IVW method. The MR-RAPS considers special pleiotropy and provides robust inferences for MR analyses of weak instrumental variables while minimizing collinearity and potential biases.⁴² 

We conducted the following sensitivity analyses to address potential pleiotropy issues in causal estimation: the MR-PRESSO test, Cochran's Q statistic, leave-one-out analysis, and the MR Steiger test. The MR-Egger regression^43,44 introduces an intercept term to detect horizontal pleiotropy based on the assumption of Instrument Strength Independent of Direct Effect (InSIDE). It can produce estimates accounting for horizontal pleiotropy, albeit with reduced precision. The MR-PRESSO test⁴⁵ provides estimates by identifying and removing possible outliers, thus correcting for horizontal pleiotropy. Cochran's Q statistic⁴⁶ is used for the heterogeneity test, with P < 0.05 and I² > 25% indicating potential heterogeneity across studies or populations. Leave-one-out analysis⁴⁷ determines if substantial connections are caused by an individual variant. Additionally, the MR Steiger test assesses the validity of the assumption that exposure causes the outcome.^48,49 

Results

Causal Effects of GM on GO

The study selected 2124 SNPs as IVs for 196 bacterial taxa and 7355 SNPs as IVs for 91 inflammatory proteins. The F statistics of all selected IVs were >10, indicating statistical potency and IVs’ reliability. A total of 26 GM (including 3 classes, 3 orders, 7 families, and 13 genera) were causally associated with GO (Supplementary Table S2). Among them, 10 bacteria were causally associated with GO in at least two MR methods (Table 1; Fig. 2). Specifically, the following six bacteria were protective factors against GO: class Deltaproteobacteria (odds ratio [OR] = 0.54; 95% confidence interval [CI], 0.36–0.81; P = 0.003), class Negativicutes (OR = 0.49; 95% CI, 0.25–0.96; P = 0.038), order Desulfovibrionales (OR = 0.56; 95% CI, 0.36–0.86; P = 0.008), order Selenomonadales (OR = 0.49; 95% CI, 0.25–0.96; P = 0.038), genus Parabacteroides (OR = 0.29; 95% CI, 0.16–0.51; P < 0.001), and genus Ruminococcaceae_UCG_011 (OR = 0.69; 95% CI, 0.51–0.94; P = 0.018). Conversely, the following four bacteria were identified as risk factors for GO: family Prevotellaceae (OR = 1.96; 95% CI, 1.18–3.26; P = 0.010), family Streptococcaceae (OR = 1.82; 95% CI, 1.11–2.98; P = 0.017), genus Lachnospiraceae_UCG_010 (OR = 1.82; 95% CI, 1.14–2.90; P = 0.012), and genus Tyzzerella3 (OR = 1.40; 95% CI, 1.13–1.75; P = 0.003). Post-FDR adjustment showed the protective effect of a higher abundance of the genus Parabacteroides on GO (P = 0.004). The reliability of the results was further confirmed by the consistency of weighted median and maximum likelihood methods with the IVW trend in MR analysis. Cochran's Q test showed no heterogeneity. MR-PRESSO indicated potential horizontal pleiotropy in class Negativicutes and order Selenomonadales, while MR-Egger found potential horizontal pleiotropy in the family Prevotellaceae. These findings suggest the need to carefully interpret research results to ensure their reliability and accuracy. The leave-one-out analysis showed that removing a particular SNP would not change the causal estimates (Supplementary Fig. S1). Steiger's test confirmed that the direction of causal association is correct. 

Table 1.

 

MR Analysis Showed the Causality of GM on GO Was Significant

 

Figure 2.

 

MR results of gut microbiota taxa on Graves’ ophthalmopathy. This circular diagram displays the MR analysis results examining the relationship between gut microbiota taxa and Graves’ ophthalmopathy. The visualization shows different bacterial taxa arranged in a circular pattern, with color intensity indicating their association with the disease. The inner rings represent results from different statistical methods, including inverse variance weighted, MR-Egger, penalized weighted median, maximum likelihood, and robust adjusted profile score approaches.

Effect of IPs on GO

We detected 91 IPs causally associated with GO using the IVW approach (Table 2, Supplementary Table S3). Eight associations surpassed the nominal P value significance level of 0.05. Specifically, the following five IPs were protective factors against GO: eukaryotic translation initiation factor 4E-binding protein 1 (4EBP1) (OR = 0.73; 95% CI, 0.53–1.00; P = 0.046), interleukin 24 (IL-24) (OR = 0.66; 95% CI, 0.46–0.96; P = 0.029), matrix metalloproteinase 1 (MMP-1) (OR = 0.72; 95% CI, 0.53–0.97; P = 0.033), TNF-related apoptosis-inducing ligand (TRAIL) (OR = 0.82; 95% CI, 0.69–0.98; P = 0.029), and TNF-related activation-induced cytokine (TRANCE) (OR = 0.81; 95% CI, 0.66–1.00; P = 0.049). In contrast, the following three factors were risk factors of GO: C-X-C motif chemokine 10 (CXCL10) (OR = 1.46; 95% CI, 1.04–2.04; P = 0.029), C-X-C motif chemokine 11 (CXCL11) (OR = 1.92; 95% CI, 1.25–2.95; P = 0.003), and protein S100-A12 (EN-RAGE) (OR = 1.78; 95% CI, 1.24–2.55; P = 0.002). Cochran's Q test showed no heterogeneity. The MR-Egger intercepts test indicated no horizontal pleiotropy. However, MR-PRESSO indicated potential horizontal pleiotropy in CXCL11. Notably, the MR-Egger analysis for CXCL11 showed a direction of effect opposite to that seen with other MR methods, which may be attributed to MR-Egger's lower statistical power and sensitivity to outlying SNPs. These methodological inconsistencies suggest that the causal relationship between CXCL11 and GO should be interpreted with caution, as the potential horizontal pleiotropy may violate key MR assumptions. The leave-one-out analysis showed that removing a particular SNP would not change the causal estimates (Supplementary Fig. S2). 

Table 2.

 

MR Analysis Showed Eight Inflammatory Proteins Had Causality on GO

 

The Mediating Effect of IPs in the Association Between GM and GO

We conducted a mediation analysis to explore the mediating effect of IPs in the causal association between genus Parabacteroides and GO (Table 3, Supplementary Table S4). Genus Parabacteroides was negatively and causally associated with GO, with a total effect of −1.25 (95% CI, −1.82 to −0.68). Using IVW as the baseline method, a negative causal relationship was observed between genus Parabacteroides and CXCL10 levels (β_direct effect A = −0.18; 95% CI, −0.35 to 0.00), while a positive causal relationship was observed between CXCL10 and GO (β_direct effect B = 0.38; 95% CI, 0.04–0.71). CXCL10 mediated the causal association between genus Parabacteroides and GO, with a mediation effect of −0.07 (95% CI, −0.18 to 0.00; P = 0.141), accounting for 5.27% of the total effect. 

Table 3.

 

Mediation Effect of the Genus Parabacteroides on Graves’ Ophthalmopathy Via CXCL10

 

Discussion

To our knowledge, this was the first two-sample, two-step MR study investigating the causal effect of GM on GO and exploring the mediating role of IPs. Based on the IVW results, we identified 26 GM taxa (including 3 classes, 3 orders, 7 families, and 13 genera) causally associated with GO. Among them, the genus Parabacteroides passed FDR correction and exhibited a protective effect on GO. Furthermore, we detected five circulating IPs with protective effects and three IPs with detrimental effects on GO. Finally, mediation MR analysis showed CXCL10 as a mediator in the causal pathway from genus Parabacteroides to GO, and the mediation effect accounted for 5.27% of the total effect. Our findings confirm the mediation effect of IPs in the causal association between GM and GO, which provides essential guidance for future prevention and intervention of GO. 

Using a genetic epidemiologic method, our study revealed that GM was causally associated with GO, which further supported the existing evidence of the association between GM and GO based on an observational study design. Cross-sectional studies have consistently shown significant changes in the structure and diversity of GM in patients with GO and mouse models.^16–18 Early intervention targeting GM has been shown to affect the onset and severity of GO in mice.⁵⁰ Recent studies have found that GM and its metabolites play an important role in the differentiation of human T cells and are involved in the progression of GO to some extent.⁵¹ GM may contribute to the development of GO through various pathways, including inducing T-cell differentiation,⁵² triggering cross-immune reactions with self-antigens,⁵³ and influencing the expression of host noncoding RNA (ncRNA).^54,55 Abnormal changes in the GM of patients with GO can induce abnormal activation of naive T lymphocytes, which affects the numbers and proportions of Th17 and/or Treg cells, thereby triggering a series of orbital inflammatory reactions and leading to GO.^52,56 Furthermore, dysbiosis of GM may lead to changes in intestinal permeability, allowing bacterial products to diffuse into blood vessels and tissues, triggering an immune response and leading to GO.⁵³ 

Our study showed that the genus Parabacteroides was protective against GO. Parabacteroides are members of the Bacteroidetes phylum and primarily participate in the digestion and absorption of carbohydrates, producing beneficial metabolites like short-chain fatty acids.⁵⁷ They can regulate the host mucosal immune system, alleviate inflammation, and maintain the integrity of the intestinal immunological barrier. There are currently around 15 species of Parabacteroides known to exist; the most well studied are Parabacteroides distasonis and Parabacteroides goldsteinii.⁵⁸ Sun et al.⁵⁹ found that the gavage administration of P. distasonis could alleviate the onset of rheumatoid arthritis (RA) in mice by promoting the production of succinate and secondary bile acids, facilitating intestinal gluconeogenesis, altering the Th17/Treg cell balance, neutralizing pathogenic autoantibodies, and improving gut permeability. Cuffaro et al.⁶⁰ discovered that P. distasonis could lessen intestinal inflammation and strengthen the intestinal epithelium, thus alleviating inflammatory bowel disease. There is currently no direct evidence associating the genus Parabacteroides with GO, demanding further investigation into the underlying mechanisms. 

Our investigation has revealed a significant causal relationship between the genus Parabacteroides and GO, mediated through inflammatory proteins, particularly CXCL10. Previous research establishes that outer membrane vesicles derived from P. distasonis upregulate CXCL10 expression, which in turn promotes CD8⁺ T-cell infiltration into affected tissues. Building upon this foundation, our current study demonstrates how this mechanism specifically contributes to GO pathogenesis.⁶¹ CXCL10, known as interferon gamma (INF-γ)–induced protein 10 (IP-10), is a significant chemokine implicated in inflammation and immune cell activation. CXCL10 can be secreted by various cell types under the action of INF-γ and tumor necrosis factor α (TNF-α).⁶² CXCL10 promotes the chemotactic activity of CXCR3⁺ cells, facilitates T-cell adhesion to endothelial cells, inhibits bone marrow colony formation and angiogenesis, and regulates cell growth and proliferation.⁶³ Therefore, high levels of sCXCL10 are markers of the host immune response, particularly Th1-mediated immune responses. Th1 lymphocytes recruited in tissues may be responsible for enhancing the production of INF-γ and TNF-α, which stimulate the secretion of CXCL10 by various cells, creating an amplified feedback loop that perpetuates the autoimmune process.⁶² 

Prior research has indicated a correlation between sCXCL10 levels and the active phase in patients with newly diagnosed and recurrent GD.⁶⁴ In patients with GO, CXCL10 levels are elevated in ocular tissues or serum, correlating with GO severity and activity.^65,66 Relative to baseline values in control groups and patients with GO, sCXCL9 and sCXCL10 exhibit notable reductions during corticosteroid and radiation therapy.⁶⁷ This suggests that the heightened concentrations of CXCL9 and CXCL10, to a certain degree, mirror the activity of orbital inflammation. Hence, these chemokines can guide treatment decisions in patients with GO. Moreover, inhibiting CXCL10 activity or blocking its receptor CXCR3 signaling may ameliorate patients’ ocular condition.^68,69 However, the basic pathophysiology of GO is complicated, and more research is needed to completely comprehend the molecular pathways that underpin the direct and strong relationship between gut microbiota and GO. 

Limitations

While MR analysis showed advantages over conventional epidemiological studies, several limitations need to be considered. First, the GWAS data primarily consist of European populations. Therefore, it remains unclear to what extent our findings apply to non-European populations. Second, the GM data come from the European MiBioGen, while the GO data come from the FinnGen research project. There may be significant differences in genetic background, environmental factors (such as diet, lifestyle), and disease characteristics between the two populations, which may lead to inconsistent effects of the instrumental variables (genetic variants) in the MR analysis, thereby affecting the accuracy of causal inference. Third, from a methodological perspective, different research may employ different experimental methods, sequencing technologies, and data analysis pipelines, which may lead to discrepancies in the definition and measurement of GM, GO, or GWAS outcomes, thus affecting the reliability of results. It is crucial to consider these differences when making cross-study comparisons and interpretations. Fourth, the use of GWAS aggregated data limited our ability to conduct stratified analyses by age and sex, while sequencing depth constraints prevented species-level microbiota analysis. Finally, the mediation analysis assumes linear relationships between GM, IPs, and GO, which may be more complex and nonlinear (e.g., curved, exponential, or polynomial) in the real world. Ignoring nonlinear relationships can lead to biased estimates and inaccurate understanding of the impact of the exposure on the outcome and the underlying mediation mechanism. Future research incorporating nonlinear MR methods (e.g., using nonlinear regression models or other statistical techniques), complementary epidemiological studies, and clinical trials may provide a nonbiased approach to elucidate the role of GM in GO pathogenesis. 

Conclusions

In conclusion, our study confirms the causal association between GM and GO, which is mediated by IPs. Our findings offer fresh insights into the pathogenesis of GO and provide important guidance to inform the prevention and treatment of GO. 

Acknowledgments

The authors thank all the participants and investigators involved in the MiBioGen and FinnGen study. 

Supported by the Foundation of Wenzhou Science & Technology Bureau (Grant Y20210997). 

All data sets utilized in this study were sourced from publicly available repositories, ensuring transparency and accessibility. Ethical clearance and approvals for each individual study included in our analysis are comprehensively documented in the corresponding primary articles. 

Disclosure: Y. Li, None; L. Chen, None; S. Lin, None; W. An, None; L. Miao, None; M. Wan, None; B. Zhang, None