Phenotypic insights into genetic risk factors for immune-related adverse events in cancer immunotherapy Authors: Haochuan Ma, Dili Song, Haibo Zhang, Taidong Li, Xing Jin Published: 2024 ABSTRACT AbstractBackgroundImmune-related adverse events (irAEs) pose substantial challenges in the realm of cancer immunotherapy, frequently affecting treatment efficacy and patient safety. To address the urgent need for identifying risk factors associated with irAEs, we conducted a comprehensive phenotype-wide Mendelian randomization analysis (MR-PheWAS).MethodsUtilizing publicly accessible genome-wide association study (GWAS) data, this investigation evaluated the impact of over 5000 exposure variables on susceptibility to irAEs using univariate Mendelian randomization (MR). We categorized these correlations and further explored potential mechanisms by which associated traits might influence irAEs through multivariate MR.ResultsMR-PheWAS identified numerous risk factors for irAEs, encompassing both previously documented and novel associations. Specifically, we identified 105 traits with probable causal relationships to all-grade irAEs and 119 traits with suggestive associations. For high-grade irAEs, we categorized 122 traits as probably associated and 141 as suggestively associated. Notably, multivariate MR analyses uncovered intricate interactions, particularly highlighting how diabetes impacts all-grade irAEs through mediators such as body mass index and sex hormone-binding globulin.ConclusionsThis study has not only identified new risk factors for irAEs but also confirmed several well-established ones. Further investigation is crucial to validate and assess these identified risk factors within clinical trials. A mechanistic understanding of these causal factors is essential for improving the management and prevention of irAEs.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00262-024-03854-8.Keywords: PheWAS, Immune-related adverse events, Mendelian randomization, Risk factor, Immunotherapy Introduction The advent of immune checkpoint inhibitors (ICIs) has revolutionized oncological treatment paradigms by harnessing the patient's immune system to combat tumors [1, 2]. Nonetheless, this broad activation of the immune system by ICIs is a double-edged sword, frequently resulting in immune-related adverse events (irAEs). These irAEs represent a range of conditions caused by nonspecific immune activation against host tissues, spanning from mild dermatological and gastrointestinal issues to severe endocrinopathies and pneumonitis [3, 4]. Studies indicate that up to 80% of patients receiving ICIs experience irAEs [5, 6]. Although ICIs are generally well-tolerated, they are not without significant risks. A systematic literature review reveals that severe (grade ≥ 3) irAEs occur in approximately 14% of patients treated with PD-(L)1 inhibitors, 34% of those treated with CTLA-4 inhibitors, and 55% of those receiving combination therapies [7]. These severe toxicities are major reasons for treatment discontinuation and can even pose life-threatening risks [8]. Thus, there is a critical need for a deeper understanding of the risk factors associated with irAEs. However, these risk factors have been relatively under-researched. For example, most meta-analyses and observational epidemiology studies have primarily assessed the relationship between risk factors and the onset of irAEs by using baseline data from clinical trial participants, with scant additional information on other potential risk factors for irAEs occurrence [9–13]. Mendelian randomization (MR), an innovative statistical approach, employs genetic variants as proxies to estimate the causal effects of potential risk factors on disease outcomes, thereby overcoming biases such as confounding and reverse causation inherent in traditional observational epidemiology [14, 15]. Analogies have been drawn between MR and randomized controlled trials, positioning these two methodologies as proximal in the hierarchy of evidence [16]. This similarity makes MR particularly adept at deciphering the complex genetic interactions that may predispose individuals to irAEs. While traditional two-sample MR has primarily focused on reassessing purported associations from previous observational studies, our study extends this approach by incorporating a phenome-wide association study (PheWAS) with MR, also known as MR-PheWAS [17, 18]. With this method, we combine the extensive coverage of PheWAS with the analytical rigor of MR to explore a broad range of previously unexplored irAEs risk factors, offering a new viewpoint on the genetic predispositions to irAEs. Methods Study design The study design, depicted in Fig. 1, follows a structured three-step procedure. Initially, we conducted a PheWAS. Next, we employed statistical significance and sensitivity analyses to categorize evidence levels for a causal relationship. Finally, we performed multivariable MR analysis (MVMR) to address potential pleiotropy among correlated risk factors. To ensure the robustness of our MR findings, the study adhered to three critical assumptions [19]: (1) genetic variations must be strongly associated with exposure factors (the relevance assumption); (2) genetic variations cannot have a direct association with the outcomes (the independence assumption); (3) genetic variants must not be linked with any confounding factors (the exclusion restriction assumption). Study overview. First, genetic variants were sourced to investigate exposure traits from MR-Base. This was followed by acquiring GWAS data for irAEs and conducting an MR-PheWAS. The identified relevant traits were then systematically categorized and stratified. Subsequent analyses involved MVMR on selected traits that may interact with each other. Furthermore, for MR to yield unbiased causal effect estimates, certain assumptions must be satisfied: a the genetic variant, used as an IV, has a strong association with the trait; b the genetic variant influences irAE risk solely through its relationship with the trait; and c the genetic variant is independent of any confounders, both measured and unmeasured, within the trait-irAEs relationship. Direct causal effects are depicted with solid lines, while potential bidirectional effects that could breach MR assumptions are shown with dashed lines. GWAS, genome-wide association study; irAEs, immune-related adverse events; MR, Mendelian randomization; MR-PheWAS, phenotype-wide Mendelian randomization analysis; MVMR, multivariate Mendelian randomization; IV, instrumental variable; ICIs, immune checkpoint inhibitors; SNP, single-nucleotide polymorphism Data sources Outcome data We obtained summary-level data on irAEs from a genome-wide association study (GWAS) conducted with the Dana-Farber Cancer Institute (DFCI) cohort, comprising 1751 cancer patients of European descent treated with ICIs across 12 cancer types from 2013 to 2020 [20]. About 90% of these patients received PD-1/PD-L1 inhibitors, while the remaining 10% were treated with a combination of CTLA-4 and PD-1/PD-L1 inhibitors. Among the cohort, 259 patients developed high-grade irAEs (grade ≥ 3), which were meticulously recorded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (NCI-CTCAE) version 5.0. Additionally, using algorithms to analyze autoimmune-like patterns in electronic health records, 339 patients were identified who experienced all-grade irAEs, primarily grade 2 or higher. For more comprehensive information, please refer to the original study. Exposure data We utilized both published and unpublished GWAS data for 18,115 traits from the IEU Open GWAS database (https://gwas.mrcieu.ac.uk/), accessed via the ‘TwoSampleMR’ package in R [21]. Our analysis included all European GWAS datasets, excluding those related to imaging phenotypes. Additional inclusion criteria were sample sizes exceeding 3000 and the exclusion of duplicate exposure traits. Instrumental selection Before performing MR-PheWAS, we applied rigorous quality control measures to identify appropriate single-nucleotide polymorphisms (SNPs) as instrumental variables (IVs). We included only SNPs with an effect allele frequency (EAF) greater than 0.01 and trait association significance of P < 5 × 10−⁸. Additionally, we excluded SNPs exhibiting linkage disequilibrium (LD) with an r2 above 0.01, selecting only those with the strongest effect. We then extracted SNPs from outcome statistics for harmonization, substituting specific SNPs with proxy SNPs (r2 > 0.8) sourced from the Ensembl server (https://www.ensembl.org) where the GWAS data lacked specific SNPs. Statistical analyses For two-sample MR analyses, we employed six methods: inverse-variance weighted (IVW), weighted median, MR Egger, simple mode, weighted mode, and Wald ratio. The choice of method was tailored to the traits under investigation and the availability of genetic instruments. When traits were influenced by multiple SNPs serving as IVs, we predominantly used the IVW method [22]. For traits where only one SNP was available, we relied on the Wald ratio for MR analysis. Recognizing that IVW results might be biased if any SNPs show horizontal pleiotropy, we incorporated four additional MR methods to improve robustness. The MR-Egger method, which estimates causal effects using the slope coefficient of Egger regression, provides a resilient estimate even with invalid IVs [23]. The weighted median method guards against up to 50% of invalid IVs. The weighted mode method is appropriate in scenarios where the relaxed assumption of IVs leads to reduced bias and a diminished type I error rate [24]. Finally, the simple mode method was used to further validate our results. All analyses were conducted using the R packages TwoSampleMR (version 0.4.10). To explore potential mechanisms, we used MVMR as a mediation analysis method, focusing specifically on the biologically relevant traits identified in our study. Our primary method was the robust IVW approach, leveraging the difference of coefficients technique. In this study, we conducted MVMR using the ‘mv_multiple() function’ from the TwoSampleMR package in R, curated on the MR-Base platform [25, 26]. The validation of MR-PheWAS and MVMR findings involved an extensive sensitivity analysis. We began by applying Cochran's Q test to assess heterogeneity among genetic instruments. Next, we employed MR-PRESSO, available at https://github.com/rondolab/MR-PRESSO/, to identify potential horizontal pleiotropy using a significance threshold of P < 0.05. Additionally, MR-Egger regression was used to detect directional pleiotropy bias. A significant non-zero intercept in the Egger regression (P < 0.05) indicates average pleiotropic effects across genetic variants, pointing to the presence of pleiotropy [27, 28]. Hierarchical classification of associations To handle the wide array of traits analyzed, we established a hierarchical classification for potential causal relationships between traits and irAEs, divided into four levels of statistical significance: robust, probable, suggestive, and no evidence (Fig. 2). Robust associations are those confirmed after false discovery rate (FDR) adjustment using the IVW method and verified through multiple sensitivity analyses, providing the highest level of statistical evidence due to consistent results across MR and validation tests. Probable associations, though not maintaining significance after FDR adjustment, meet the criteria for robust traits except for the FDR threshold, recognizing the wide range of traits evaluated and their isolated potential connections to irAEs. Suggestive traits, significant at P < 0.05, do not pass the sensitivity tests required for probable traits or rely on a single SNP as an IV, highlighting the uncertainties of using a Wald ratio test with a single IV. Lastly, traits with no evidence, which either lack statistical power or suggest that a moderate causal effect is unlikely, indicate no discernible causal relationship with irAEs. Hierarchical classification of associations. Potentially causal relationships between traits and irAEs were categorized into four levels of statistical significance: robust, probable, suggestive, and no evidence. Robust associations are those confirmed after adjusting for the FDR (PIVW-FDR < 0.05) using the IVW method and supported by multiple sensitivity analyses. Probable associations, while not maintaining significance post-FDR adjustment (PIVW-FDR > 0.05), meet the same criteria as robust associations except for the FDR threshold. Suggestive traits, significant at P < 0.05, fail the sensitivity tests required for probable traits or rely on a single SNP as an IV (PWald ratio < 0.05). Lastly, no evidence traits, which lack statistical power (PWald ratio > 0.05) to evaluate causal effect. irAEs, immune-related adverse events; MR, Mendelian randomization; IVW, inverse-variance weighted; FDR, false discovery rate; IV, instrumental variable In addition, to facilitate interpretation, we utilized the IEU Open GWAS classification system, categorizing traits into the following groups: anthropometric, behavioral, fitness, socioeconomic, dietary, hematological, immune system, family background, psychiatric/neurological, medication, metabolites, operation, cancer, disease, other, biomarker, and protein. Results Overview For the analysis of all-grade irAEs, we identified 5,930 traits as suitable, utilizing between 1 and 393 SNPs as genetic instruments for each trait (Supplementary Table S1). Similarly, 5,939 traits were appropriate for the analysis of high-grade irAEs, with the number of SNPs per trait also ranging from 1 to 393 (Supplementary Table S2). MR analysis from phenome-wide traits to all-grade irAEs Figure 3 and Supplementary Table S3 illustrate the inclusion of 2,936 traits in the IVW analysis for all-grade irAEs. The most represented categories were disease (n = 512), metabolites (n = 434), protein (n = 290), immune System (n = 188), and anthropometric measures (n = 174). According to the hierarchical classification of evidence levels, no traits were categorized as robust, 105 were classified as probable, 119 as suggestive, and the rest showed no evidence of association. Genetically predicted effects of phenome-wide traits on the risk of all-grade irAEs using IVW. The Y-axis displays the logarithmic transformation of the P value derived from the MR estimate, while the X-axis indicates the categories for the exposure traits. Symbols are color-coded according to trait category, and their sizes represent the beta coefficient of the MR estimate, calculated using the IVW method. A horizontal red line marks the threshold for the FDR-corrected P value of 0.05. MR, Mendelian randomization; IVW, inverse-variance weighted; FDR, false discovery rate. irAEs, immune-related adverse events Within the probable category, most traits were from the disease, hematological, immune system, and protein groups. Specific diseases, such as heart failure (HF, OR = 0.25, 95% CI 0.06–0.99, P = 0.04) and multiple sclerosis (MS, OR = 0.74, 95% CI 0.58–0.94, P = 0.01), were associated with a decreased risk of all-grade irAEs, whereas rheumatoid arthritis (RA, OR = 1.34, 95% CI 1.07–1.68, P = 0.01), type 1 diabetes (OR = 1.19, 95% CI 1.01–1.41, P = 0.04), and glaucoma (OR = 2.13, 95% CI 1.26–3.62, P = 0.01) were linked to an increased risk. Notable hematological markers, like total cholesterol (TC, OR = 0.45, 95% CI 0.21–0.98, P = 0.04), LDL cholesterol (OR = 0.53, 95% CI 0.32–0.90, P = 0.02), and total esterified cholesterol (Total CE, OR = 0.45, 95% CI 0.21–0.96, P = 0.04), were associated with a reduced risk of all-grade irAEs. Additionally, strong correlations with all-grade irAEs were found for biomarkers such as sex hormone-binding globulin (SHBG, OR = 0.43, 95% CI 0.22–0.85, P = 0.01), granzyme B (OR = 0.75, 95% CI 0.61–0.95, P = 0.02), and thiosulfate sulfotransferase (OR = 2.23, 95% CI 1.22–4.32, P = 0.01). Furthermore, we pinpointed several risk factors rarely addressed in prior research, including: (1) total body bone mineral density (OR = 2.23, 95% CI 1.06–4.69, P = 0.03), which was linked to an increased risk of all-grade irAEs, and (2) vitamin and mineral supplements (OR = 7.87 × 10–10, 95% CI 0.00–0.04, P = 0.02), which were associated with a decreased risk. The results of the sensitivity analysis are detailed in Supplementary Table S4–5. MR analysis from phenome-wide traits to high-grade irAEs Figure 4 and Supplementary Table S6 detail the inclusion of 2937 traits in the IVW analysis for high-grade irAEs. The trait categories were: disease (n = 541), metabolites (n = 443), protein (n = 279), hematological (n = 275), anthropometric (n = 243), and immune system (n = 177). In terms of evidence classification, none of the traits were classified as robust, 122 were classified as probable, 141 as suggestive, and the remaining showed no evidence. Genetically predicted effects of phenome-wide traits on the risk of high-grade irAEs using IVW. The Y-axis displays the logarithmic transformation of the P value derived from the MR estimate, while the X-axis indicates the categories for the exposure traits. Symbols are color-coded according to trait category, and their sizes represent the beta coefficient of the MR estimate, calculated using the IVW method. A horizontal red line marks the threshold for the FDR-corrected P value of 0.05. MR, Mendelian randomization; IVW, inverse-variance weighted; FDR, false discovery rate; irAEs, immune-related adverse events For probable traits, the most significant categories were disease, immune system, protein, and anthropometric. Within the disease category, specific conditions, such as HF (OR = 0.13, 95% CI 0.03–0.69, P = 0.02) and valvular heart disease (OR = 0.07, 95% CI 0.01–0.85, P = 0.04), were linked to a decreased risk of high-grade irAEs, while Alzheimer's disease (OR = 1.29, 95% CI 1.00–1.65, P = 0.04), type 1 diabetes (OR = 1.23, 95% CI 1.01–1.51, P = 0.04), and lacunar stroke (OR = 3.00, 95% CI 1.21–7.43, P = 0.02) were associated with an increased risk. In the protein category, levels of TNF-related apoptosis-inducing ligand receptor 2 (OR = 0.09, 95% CI 0.01–0.84, P = 0.03) and SHBG (OR = 0.14, 95% CI 0.03–0.59, P = 0.01) were potentially linked to a reduced risk of high-grade irAEs. Conversely, levels of signal-regulatory protein beta-1 (OR = 1.52, 95% CI 1.07–2.16, P = 0.02), transferrin receptor 2 (OR = 1.97, 95% CI 1.20–3.23, P = 0.01), and cerebellin-4 (OR = 3.04, 95% CI 1.27–7.30, P = 0.01) were associated with an increased risk. Anthropometric measures indicated that traits like trunk fat percentage (OR = 0.32, 95% CI 0.11–0.98, P = 0.04), as well as both systolic (OR = 0.84, 95% CI 0.75–0.94, P = 0.002) and diastolic blood pressures (OR = 0.91, 95% CI 0.84–0.99, P = 0.02) may offer protective effects against high-grade irAEs, whereas a higher body mass index (BMI) posed a risk (OR = 1.72, 95% CI 1.01–2.93, P = 0.04). Novel risk factors also emerged from the analysis, including (1) factors associated with a decreased risk of high-grade irAEs, such as maternal history of hypertension (OR = 1.96 × 10–9, 95% CI 0.00–0.41, P = 0.03), and serum copper (OR = 0.43, 95% CI 0.20–0.92, P = 0.03), and (2) factors associated with an increased risk, including mood swings (OR = 513.36, 95% CI 1.52–1.73 × 105, P = 0.04) and hot beverage consumption (OR = 141.41, 95% CI 1.68–1.19 × 104, P = 0.03). The sensitivity analysis results are presented in Supplementary Table S7–8. Multivariable MR of biologically related traits We conducted MVMR to explore potential interactions among selected biological traits in relation to irAEs. Specifically, we analyzed autoimmune diseases (AD, rheumatoid arthritis and multiple sclerosis), diabetes (diabetes and BMI, or diabetes, BMI, and SHBG), and cholesterol (LDL-C, Large HDL-FC, Small HDL-FC, and Total CE) in association with all-grade irAEs. Our findings revealed that BMI (PMVMR < 0.01; PUVMR = 0.04) and SHBG (PMVMR < 0.04; PUVMR = 0.03) maintained significant associations with all-grade irAEs across two models, while the association with diabetes (PMVMR > 0.05; PUVMR = 0.03) became nonsignificant, suggesting potential moderation effect through BMI or SHBG. However, autoimmune diseases (AD) and cholesterol-related phenotypes did not show independent effects on all-grade irAEs risk, hinting at alternative mediating pathways (Fig. 5 and Supplementary Table S9). Multivariable MR of biologically related traits for all-grade irAEs. The left side displays a forest plot of the MVMR results, while the right side presents a corresponding forest plot for UVMR outcomes, highlighting P values below 0.05 in red. MR, Mendelian randomization; irAEs, immune-related adverse events; MVMR, multivariate Mendelian randomization; UVMR, univariable Mendelian randomization; SNP, single-nucleotide polymorphism; BMI, body mass index; SHBG, sex hormone-binding globulin; LDL, low density lipoprotein; HDL, high density lipoprotein; FC, free cholesterol; CE, esterified cholesterol; OR, odds ratio; CI, confidence interval For high-grade irAEs, we separately analyzed diabetes (type 1 diabetes, BMI, and SHBG) and blood pressure (systolic, diastolic, and antihypertensive medications). Compared to all-grade irAEs, SHBG (PMVMR = 0.03; PUVMR = 0.01) and type 1 diabetes (PMVMR = 0.03; PUVMR = 0.04) remained significantly associated with high-grade irAEs, while BMI (PMVMR > 0.05; PUVMR = 0.04) lost significance, possibly indicating moderation through type 1 diabetes or SHBG. Blood pressure-related traits showed no independent effect on high-grade irAEs risk, suggesting the involvement of other mediating pathways (Fig. 6 and Supplementary Table S10). Multivariable MR of biologically related traits for high-grade irAEs. Blue squares denote P values less than 0.05, while red squares indicate P values greater than 0.05. MR, Mendelian randomization; irAEs, immune-related adverse events; MVMR, multivariate Mendelian randomization; UVMR, univariable Mendelian randomization; BMI, body mass index; SHBG, sex hormone-binding globulin; OR, odds ratio; CI, confidence interval Discussion In this study, we employed MR-PheWAS to elucidate the complex relationships between a wide array of traits and the risk factors for irAEs. By leveraging both established and novel genetic instruments, we identified numerous risk factors influencing irAEs across all and high grades, including both previously documented and newly discovered factors. These findings not only confirm known associations, offering fresh insights, but also reveal several previously unreported risk factors, thereby providing valuable evidence that could improve the prevention and management of irAEs. In line with existing research, our study confirms the significant roles of diseases, hematological profiles, proteins, and immune system markers as prevalent influences on irAEs. For instance, research has demonstrated that preexisting AD and a family history of these conditions may increase the risk of irAEs in patients receiving ICI therapy [29–31]. Furthermore, patients with cardiovascular risk factors like hypertension, coronary artery disease, and HF are more prone to cardiac-related irAEs [32]. In addition, blood cell counts, including white blood cells, neutrophils, and the neutrophil–lymphocyte ratio, are straightforward and objective indicators of irAEs risk. This correlation is supported by multiple clinical studies on immunotherapy [33–35]. However, our research has also revealed conditions less frequently linked with irAEs in prior studies, such as anorexia nervosa, glaucoma, and depression, suggesting a broader spectrum of potential risk factors than previously recognized. Additionally, identifying novel biomarkers and their associations with irAEs, including dihydrofolate reductase, TNF receptor superfamily member 1B, and sulfhydryl oxidase 1, provides new avenues for exploring the mechanisms of irAEs. These largely unexplored markers in the context of irAEs could shed light on the underlying mechanisms of immune modulation and its consequences. Notably, previous research into the role of the gut microbiota suggested a possible impact on irAE outcomes [36–39]. However, our current findings do not support a direct causal relationship for either all-grade or high-grade of irAEs. This discrepancy may result from differences in genetic instruments, study populations, or variations in the methodologies used across studies. Furthermore, we identified several risk factors that have been less explored across various categories, including anthropometric, dietary, psychiatric/neurological, socioeconomic, and others. For example, our findings that BMI, bone mineral density, and dietary patterns, like gluten-free diets, can influence irAE risks suggest that lifestyle and physical health metrics are as important as laboratory indicators in managing irAEs. Also, it is noteworthy that a range of brain structures, such as the thalamus, splenium of the corpus callosum, and anterior limb of internal capsule, were implicated in irAEs risk, providing a fresh perspective for exploring the potential mechanisms of irAEs. This observation underscores the intricate and fascinating links between psychiatric disorders, brain anatomy, and irAEs [40]. Lastly, our findings show differences in risk factors between high-grade and all-grade irAEs. In the 'Probable' category, high-grade irAEs were uniquely associated with specific anthropometric and disease-related traits like uterine fibroids and neurodegenerative conditions. These differences are crucial for further studies focused on understanding and mitigating the risks of high-grade irAEs [41]. The primary strength of this study lies in its hypothesis-free approach, which allows for a comprehensive and unbiased exploration of phenotypic traits associated with irAEs. Rigorous sensitivity analyses further validate our findings, and minimal overlap between exposure and outcome samples reduces potential biases. Moreover, this study is the first to distinguish risk factors for both high-grade and all-grades irAEs. However, the study does have limitations. Relying on available genetic instruments and focusing on traits mainly in individuals of European descent may limit the applicability of our findings to broader populations. Additionally, the complex interplay among traits like serum lipids and AD could confound our results, emphasizing the need for more detailed phenotypic data and larger, more diverse study populations to enhance the generalizability of our conclusions. Conclusion In conclusion, this study has not only identified new risk factors for irAEs but also confirmed several well-established ones. Further investigation is crucial to validate and assess these identified risk factors within clinical trials. A mechanistic understanding of these causal factors is essential for improving the management and prevention of irAEs. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (XLSX 2844 kb) Acknowledgements We would like to thank Professor Rong Zhang for her guidance on the methodological aspect of this study. We would also like to express our gratitude to the researchers who freely shared the GWAS data with us. Abbreviations Autoimmune diseases Esterified cholesterol Dana-Farber Cancer Institute Effect allele frequency False discovery rate Genome-wide association study Heart failure Immune checkpoint inhibitors Immune-related adverse events Instrumental variables Inverse-variance weighted Linkage disequilibrium Multiple sclerosis Mendelian randomization Multivariable MR analysis National Cancer Institute Common Terminology Criteria for Adverse Events Phenome-wide association study Rheumatoid arthritis Sex hormone-binding globulin Single-nucleotide polymorphisms Total cholesterol Univariable MR analysis Authors’ contribution XJ, TDL, and HBZ conceived and designed the experiments. HCM and DLS performed and analyzed the experiments. HCM and DLS were involved in collecting data. HCM and DLS wrote the manuscript. All authors read and approved the final manuscript. Funding This work was supported by the China Postdoctoral Science Foundation (2023M740779), Youth Foundation of the National Natural Science Foundation of China (82405509), and Postdoctoral Special Project of Guangdong Provincial Hospital of Chinese Medicine (2023-B5). Availability of data and materials No datasets were generated or analyzed during the current study. Declarations Ethics approval and consent to participate All data used in the study are from public databases, there was no need for a separate ethical approval for this study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Haochuan Ma and Dili Song contributed equally to this work. Contributor Information Taidong Li, Email: nkyyltd@163.com. Xing Jin, Email: jinxing0108@yzu.edu.cn. References Associated Data Supplementary Materials Supplementary file1 (XLSX 2844 kb) Data Availability Statement No datasets were generated or analyzed during the current study.