The study participants were TAO patients who had undergone orbital decompression surgery and normal subjects who had undergone blepharoplasty at the Eye, Ear, Nose and Throat (ENT) Hospital, Fudan University. TAO was diagnosed based on the Bartley criteria (19), and TAO activity was evaluated using Clinical Activity Score (CAS). A CAS ≥ 3 is indicative of active TAO, and a CAS ≤ 2 is indicative of inactive TAO (20). None of the TAO patients had any other inflammatory diseases, had undergone thyroidectomy or radioactive iodine therapy, or had received steroids or immunosuppressive agents in the previous six months. They had normal thyroid function after antithyroid treatment. We obtained orbital adipose/connective specimens removed during orbital decompression surgery for TAO patients. Normal orbital adipose/connective samples were collected during plastic surgeries. Fifteen TAO patients were selected for either metabolites analysis or western blot and quantitative real-time polymerase chain reaction (qRT-PCR) analysis. Clinical information of TAO patients and normal subjects enrolled for LC-MS was shown in Table 1 . No significant distinction in gender or BMI was found between the two groups ( Table 1 ). Detailed characteristics of all TAO patients included in this study were summarized in Table S3 . These samples were frozen in liquid nitrogen and stored at -80°C for analysis.

The study was approved by the Ethics Committee of Eye & ENT Hospital, Fudan University (identification number: 2021082), and conducted in compliance with the Helsinki Declaration.

Internal standard (2-chloro-l-phenylalanine in methanol, 0.3 mg/mL) (Shanghai Hengchuang Bio, Shanghai, China) and extraction solution with pre-chilled methanol in HPLC water (4/1, v/v) were added to each sample. Steel balls were added to the sample for homogenization. Then, sample was stored at -20°C for 30 min. The supernatant was transferred to a new tube after centrifugation and was then dried in a freeze concentration centrifugal dryer. Samples was reconstituted with methanol in HPLC-water (1/4, v/v) and then transferred to LC vials. Samples were stored at -80°C before LC-MS.

LC-MS was performed as previously described in detail (21, 22). A DionexUltimate 3000 RS UHPLC system coupled with mass spectrometer equipped with electrospray ionization (ESI) source (Thermo Fisher Scientific, Waltham, MA, USA) was utilized to detect the metabolic profiling. The ACQUITY UPLC HSS T3 column (100 × 2.1 mm, 1.8 μm) was applied in both positive and negative modes. The gradient elution system was consisted of (A) (water with 0.1% formic acid, v/v) and (B) (acetonitrile containing 0.1% formic acid, v/v). The gradient elution was 5% B in 0.01min, 5% B in 2min, 30% B in 4min, 50% B in 8min, 80% B in 10min, 100% B in 14min, 100% B in 15min, 5% B in 15.1min and 5% B in 14min. Progenesis QI was applied to analyze LC-MS raw data (Waters Corporation, Milford, USA).

qRT-PCR was done as previously described (23). Total RNAs were isolated and purified from orbital adipose/connective sample using the TRIzol reagent (Thermo Fisher Scientific, Waltham, MA, USA). Complementary DNA was synthesized utilizing the PrimeScript RT Reagent Kit (Takara, Otsu, Shiga, Japan) following the manufacturer’s instructions. Gene expression level was quantified by qRT-PCR using SYBR Premix Taq (Takara, Otsu, Shiga, Japan) and normalized to β-actin. All primer sequence information is listed in Table 2 .

The tissue was lysed by RIPA buffer containing proteinase inhibitors and phosphatase inhibitors (Roche, Basel, Switzerland). Protein concentration was evaluated by BCA protein assay kit (Vayme Biotechnology, Nanjing, China). The proteins were separated on 4-16% gradient SDS-PAGE, and then transferred to PVDF membrane (Millipore, Bedford, MA, USA). The detection of chemiluminescence was performed as previously described (24).Western blot results were quantified using ImageJ (http://rsb.info.nih.gov/ij/index.html). Primary antibodies involved in IRE1α (CST, cat.No: 3294, 1:1000), PERK (CST, cat.No: 5683, 1:1000), ATF6 (CST, cat.No: 65880, 1:1000), and β-Actin (CST, cat.No: 4967, 1:1000) were used for western blot.

Metabolomics data were analyzed using orthogonal partial least-squares discrimination analysis (OPLS-DA), variable importance in projection (VIP), heat map, and volcano plot (25). Pathway enrichment for the changed metabolites was analyzed using MBRole 2.0 (http://csbg.cnb.csic.es/mbrole2/) based on the Kyoto Encyclopedia of Genes and Genomes (KEGG). Interactions between metabolites and genes were analyzed using the MetScape plugin (v3.1.3) (26) in Cytoscape (3.7.1) (27)and networks generated based on known gene-metabolite interactions. Statistical analyses were conducted by prism (GraphPad, San Diego, CA, USA). The difference was determined by two-tailed Student’s t test. Data were shown as mean ± SEM. P < 0.05 was considered statistically significant.

We measured 3038 metabolites in samples from the TAO patients and the controls. OPLS-DA revealed a prominent distinction between the two groups in score plots, indicating that the metabolomes of orbital tissue from the patients with TAO were significantly changed( Figure 1A ). Then, cross validation test was performed to validate the OPLS-DA model. The Q2 of -0.613 is less than 0.05, indicating high predictive relevance of the OPLS-DA model( Figure 1B ).

All metabolite class annotation was analyzed, about 1/3 changed metabolites occurred in lipids and lipid-like molecules (1065) ( Figure 2A ). To differentiate the altered metabolites in TAO, we performed volcano plot analysis with a threshold of P < 0.05. Among the 3038 metabolites, 593 metabolites (367 increased metabolites and 226 decreased metabolites) were significantly changed in the TAO samples ( Figure 2B ). The changed metabolites were evaluated by calculating VIP score. High VIP score indicates significant group discrimination and 1 was selected as a cutoff for VIP score. Forty-two metabolites (VIP score > 1) were found. Notably, lidocaine, 5β-cholestane, monoethylglycylxylidide, triglyceride, pentadecylic acid, and 3’-deoxystreptomycin were the top altered metabolites (VIP score > 1.8) ( Table S1 ). To visualize the changed pattern of metabolites, we chose the 42 changed metabolites (VIP scores > 1, P < 0.05) for heat map analysis ( Figure 2C , Table S1 ).

To explore the potential mechanism of the pathogenesis of TAO, we identified the enriched pathways by analyzing 42 changed metabolites (VIP scores > 1, P < 0.05). The significantly altered metabolic pathways were cholesterol metabolism, choline metabolism in cancer, fat digestion and absorption, regulation of lipolysis in adipocytes, and insulin resistance ( Figure 3A ). Metabolites-pathway associated analysis was shown in Figure 3B . TG(20:0/20:1(11Z)/20:4(5Z,8Z,11Z,14Z)), 1D-myo-Inositol 3-phosphate, and phosphocholine were involved in the changed metabolic pathway ( Figure 3B ). These metabolomic findings demonstrated that orbital adipose/connective tissue in TAO has a unique metabolic characteristic.

TG(20:0/20:1(11Z)/20:4(5Z,8Z,11Z,14Z)), 1D-myo-Inositol 3-phosphate, and phosphocholine (VIP scores > 1, P < 0.05) were involved in predicted pathway ( Figure 4A ). To further elucidate the importance of these metabolites, receiver operating characteristic (ROC) curve was utilized to analyze these metabolites for sensitivity and specificity as biomarkers. High area under the curve (AUC) indicates good sensitivity and specificity. These metabolites had high AUC > 0.7 (TG(20:0/20:1(11Z)/20:4(5Z,8Z,11Z,14Z)): 0.7045; 1D-myo-Inositol 3-phosphate: 0.7273; phosphocholine: 0.7500) ( Figure 4B ).

Significantly changed expression profiles of mRNAs in TAO were found in our previous study (23). To determine the mechanism of the changed metabolites, we analyzed the differentially abundant metabolites and genes(> 2 fold change)based on known gene-metabolite interactions. Metscape network analysis were subsequently performed to integrate the transcriptomics and metabolomics data of TAO ( Figure 5A ). Cholesterol-related gene diacylglycerol acyltransferase 1 (DGAT1) was associated with TG (20:0/20:1(11Z)/20:4(5Z,8Z,11Z,14Z)) ( Figure 5A ). DGAT1 is a critical rate-limiting enzyme in the biosynthesis of triacylglycerol (TG) (28). TG synthesis by DGAT1 protects adipocyte from endoplasmic reticulum (ER) stress and adipose tissue inflammation, which is correlated with the etiology of TAO (29, 30). ER stress induces the response of unfolded protein, involving transmembrane receptors activation (ATF6, PERK and IRE1α) (31). Herein, an increased expression level of ATF6, PERK and IRE1α was found in orbital tissues of TAO examined by western blot indicating ER stress in TAO ( Figure 5B ). Additionally, statistically increased expression level of ATF6 was found in TAO sample ( Figure 5B ). qRT-PCR showed that DGAT1 was highly expressed in orbital tissues of TAO patients ( Figure 5C ). Taken together, these results indicated that dysfunction of cholesterol metabolism might be involved in TAO pathogenesis.

Herein, a significant change of triglyceride was found (P < 0.05; VIP > 2) in orbital fat tissue samples from the patients with TAO. Pathway analysis and metabolite-genes analysis showed cholesterol metabolism was potentially involved in TAO pathogenesis. Similarly, in our previous KEGG analysis of different circular RNAs of TAO orbital adipose/connective tissue indicates cholesterol metabolism pathway is changed (36). Disrupted cholesterol metabolism altered immunity in TAO. Cholesterol metabolism dysfunction promotes inflammatory responses, including the activation of Toll like receptor signaling (37), inflammasome (38), release of pro-inflammatory cytokines (39), and phagocytes production (40). In the other hand, impaired proinflammatory response in adipose cause abnormal lipid metabolism, which causes adipose remodeling and expansion (41). These changes lead to serious complications of TAO such as proptosis and vision loss (optic nerve compression) (42). The mechanism contributing to proinflammatory response of orbital adipose tissues in TAO patients is not entirely clarified. In our study, ER stress and increased DGAT1 expression were found in the orbital adipose/connective tissue of TAO. Adipocyte is protected from adipose inflammation and ER stress by DGAT1-mediated TG synthesis (29, 30). We speculate that disrupted cholesterol metabolism in TAO leads to ER stress. To protect from ER stress and adipose tissue inflammation, DGAT1 was upregulated to promote TG synthesis in TAO. Taken together, it is indicated that that TAO potentially reprograms cholesterol metabolism for its pro-inflammatory process.

According to the pathway analysis, insulin resistance is involved in TAO. In our previous study, insulin resistance is the top15 enriched hypomethylated differential methylated probes (DMPs) KEGG pathway (43). Insulin resistance is associated with the insulin like growth factor 1 receptor (IGF-1R) (44). IGF-1R is a member of insulin receptor (IR) family regulating development, growth, and cell transformation (45). IGF-1R expression level is elevated in TAO patients (46). The immunoglobulins in TAO patients that stimulate orbital fibroblasts might act through IGF-1R (47, 48). IGF-1R combines with TSH receptor (TSHR), the main autoantigen of TAO, to form a functional complex (49). Blocking the activity of IGF-1R suppresses the inflammatory reaction caused by IgG in TAO (50). Furthermore, IGF-1/IGF-1R regulates adipocyte differentiation (adipogenesis) (51, 52). Normal orbital adipose tissue maintains local physiological homeostasis. This homeostasis is disrupted by increased adipogenesis in TAO patients (8). Induced adipogenesis in adipocyte cause insulin resistance (53). Taken together, enhanced adipogenesis activated by IGF-1R leading to insulin resistance could be the potential molecular mechanism in TAO.

Considering that many factors affect metabolite changes in TAO, studies with larger sample sizes and various clinical stages are necessary to clarify the metabolite profiles in TAO. In addition, tracing these changed pathways with stable isotopes and exploring the role of the important enzymes are necessary to reveal the interactions of metabolic pathways.

In summary, A total of 593 metabolites differed significantly between TAO and control participates. Our study indicates a significant difference in the metabolic profiles of TAO, showing that TAO alters cellular metabolome, especially cholesterol metabolism. This study should shed light on the molecular pathogenesis of TAO and provide therapeutic benefits of targeting the changed metabolic pathways