Serum Bile Acid Proles Improve Clinical Prediction of Nonalcoholic Fatty Liver in T2DM patients

Background: The present study aimed to assess the ability of serum bile acid proles to predict the development of nonalcoholic fatty liver (NAFL) in type 2 diabetes mellitus (T2DM) patients. Methods: Using targeted ultraperformance liquid chromatography (UPLC) coupled with triple quadrupole mass spectrometry (TQ/MS), we compared serum bile acid levels in T2DM patients with NAFL (n=30) and age-gender matched T2DM patients without NAFL (n=36) at the rst time. Second, an independent cohort study of T2DM patients with NAFL (n=17) and age-gender matched T2DM patients without NAFL (n=20) was used to validate the results. The incremental benets of serum biomarkers, clinical variables alone or with biomarkers were then evaluated using receiver operating characteristic (ROC) curves and decision curve analysis. Area under the curve (AUC), integrated discrimination improvement (IDI) and net reclassication improvement (NRI) were used to evaluate the biomarker predictive abilities. Results: The serum bile acid proles in T2DM patients with NAFL were signicantly different from T2DM patients without NAFL, as characterized by the signicant elevation of LCA, TLCA, TUDCA, CDCA24G and TCDCA, which may be potential biomarkers for identication of NAFL in T2DM. Based on the improvement in AUC, IDI and NRI, addition of 5 bile acids to a model with clinical variables statistically improved its predictive value. Similar results were found in the validation cohort. Conclusions: These results highlight detected biomarkers may contribute to the progression of NAFL in T2DM patients, and these biomarkers particularly in combination may help diagnosis of NAFL and allow earlier intervention in T2DM patients. metabolites


Introduction
The prevalence of type 2 diabetes mellitus (T2DM) has increased in the world. Nonalcoholic fatty liver disease (NAFLD) has become the most frequent chronic liver disease in T2DM patients [1,2]. NAFLD includes simple steatosis, i.e. nonalcoholic fatty liver (NAFL), to nonalcoholic steatohepatitis (NASH), cirrhosis or nally hepatocellular carcinoma (HCC) [3]. Studies have shown that NAFLD may be present in up to 70% of patients with diabetes [4], however the prevalence of biopsy-proven NASH was 20% in asymptomatic T2DM patients with normal liver function [5].
The presence of both NAFLD and T2DM increases the possibilities of the development of complications of diabetes (including both macro-and microvascular complications) and raises the risk of more severe NAFLD, including cirrhosis and HCC [6]. The newest guidelines for the diagnosis and management of NAFLD from the American Association for the Study of Liver Diseases mention two points, that it is important to assess the possibility of NAFLD and NASH in T2DM patients and that the ratio of morbidity and mortality of cardiovascular disease (CVD) in NAFLD patients is very high [7]. Moreover, HCC happens more commonly in T2DM with NASH patients compared to those patients without T2DM [1]. Nagpal et al [8] found that serum cytokeratin-18 levels could be used to predict the development of T2DM in adult patients with NAFLD. However, there were no related studies describing biomarkers which can predict the development of NAFLD in T2DM patients, which could cause the high risk of occurrence of CVD or severe liver diseases. Therefore, it is urgent and important to nd simple, non-invasive diagnostic biomarkers that can help early detect and understand the mechanism involved in the formation of NAFLD in T2DM patients in order to prevent the occurrence of CVD and HCC.
Metabolomics is an emerging analytical technology which simultaneously analyzes many metabolites in biological samples. Bile acids (BAs) are amphipathic steroid molecules which are synthesized in the liver from cholesterol and excreted into the bile, and have been acknowledged as key signaling molecules that regulate glucose, lipid and energy metabolism due to their micelle-forming properties [9,10]. BAs are secreted by hepatocytes into the canaliculus actively and are nearly completely (95%) absorbed by an active uptake process after reaching the terminal ileum. It is well-known that BAs not only facilitate the intestinal absorption of lipids and lipid-soluble vitamins such as vitamins A, D, K, and E, but also mediate cellular and molecular signal transduction by activating the nuclear receptors, i.e., the farnesoid X receptor (FXR), pregnane X receptor (PXR), vitamin D receptor (VDR) and the G-protein-coupled bile acid receptor (GPBAR1 or TGR5) and sphingosin-1-phosphate receptor 2 (S1PR2) [11,12]. Caussy et al [13] found that there was no signi cant differences of the total serum BAs between NAFLD vs non-NAFLD and NAFL vs NASH, but serum BAs were signi cantly increased progressively as liver brosis increased.
Our former study based on untargeted metabolomics, has shown that serum differential metabolites, including bile acids, fatty acids, amino acids, lipids, carbohydrates, steroids, and tricarboxylic acids, were involved in the formation of T2DM complications according to a cohort study in 292 T2DM patients with different complications [14]. We recently summarized many different metabolomics studies on diabetic complications [15]. However, it is still unclear whether the concentrations of BAs are different in the T2DM patients with or without NAFL. We then hypothesized that BAs contribute to the formation of NAFL in T2DM patients and performed a serum targeted metabolomics study on T2DM patients with and without NAFL.
In the present study, we quanti ed the serum levels of bile acids in T2DM patients with or without NAFL using targeted metabolomics. The purpose of this study was to observe the metabolic differences between diabetes patients with different complications and to assess the performance of the new screening approach for the diagnosis of NAFL in patients with T2DM. We anticipate that this study could provide metabolomics support for the early intervention in diabetic complications, and offer new insights into the approach of targeting metabolites and pathways in the treatment of this serious global health problem.

Sample population
There were two study groups. We rst conducted a cross-sectional study of 30 T2DM patients with NAFL and 36 without NAFL. Second, an independent cohort composed of 17 T2DM patients with NAFL and 20 without NAFL was used to validate the former results. All patients were obtained at Longhua Hospital, Shanghai University of Traditional Chinese Medicine, China, from Sep 2015 to Dec 2016. According to the criteria by the World Health Organization in 1999 [16], T2DM was de ned as fasting plasma glucose (FPG) of ≥ 7.0 mmol/L, 2 h plasma glucose (2 h PG) of ≥ 11.1 mmol/L, or a history of oral hypoglycemic or insulin use, or both. NAFLD was de ned according to the guidelines for management of NAFLD by the Chinese Liver Disease Association in 2010 [17]. NAFLD can be diagnosed if the patient matches the following the criteria: 1) no history of alcohol consumption or alcohol equivalent to < 70 g per week (female), < 140 g per week (male); 2) fatty liver disease was not caused by other factors including viral hepatitis, drug-induced liver disease, total parenteral nutrition, Wilson's disease and autoimmune liver disease; 3) having obtained characteristic histological changes of fatty liver disease from histological examination. Considering that it is hard to achieve histological diagnosis, 3) could be substituted for 4) having obtained characteristic diffuse fatty liver imaging results; and/or 5) serum levels of transaminases, including alanine aminotransferase (ALT), and/or aspartate aminotransferase (AST) and γ-glutamyl-transpeptidase, have increased continuously for over half a year in patients with several components of metabolic syndrome, such as overweight and (or) visceral obesity, with elevated fasting blood glucose, dyslipidemia and hypertension. The imaging results of characteristic diffuse fatty liver are as below: if ultrasonography examinations have the following two items of three abdominal ultrasound manifestations 1) the near-eld echo of the liver is diffuse, enhanced and stronger than that of the kidney; 2) the far-eld echo of the liver gradually attenuates; 3) the structure of hepatic duct is unclear. Or computed tomography (CT) shows diffuse decreased liver density and the CT value of liver/spleen is < = 1. The diagnosis of NAFLD in the present study was based on ultrasonography examinations. Patients were excluded if they were pregnant or breast-feeding, or had intellectual dysfunctions or psychiatric disorders. Serum samples were stored at -80 °C until analysis. All patients who participated in the study provided signed informed consent. All methods were in accordance with the relevant guidelines and regulations. All the studies were approved by the Ethics Committee of Shanghai University of Traditional Chinese Medicine and have been performed in accordance with the ethical standards in the 1964 Declaration of Helsinki.

Sample preparation and bile acid detection
Serum BAs levels were measured according to the previously reported approach [18][19][20]. Brie y 50 µL of serum was used, and the nal extracts were reconstituted with 40 µL of acetonitrile and 40 µL of water. Next, 5 µL of supernatant was injected for analysis. Serum BAs were analyzed using the Waters ACQUITY ultraperformance liquid chromatography (UPLC) (BEH C18 1.7 µm 2.1 × 100 mm column) system coupled with Waters Xevo Triple Quadrupole mass spectrometry (TQ/MS). The raw data acquired from UPLC-TQ/MS were rst processed by TargetLynx software v 4.1 (Waters, USA), and then manually checked to ensure data quality. Total bile acids (TBA) concentrations were determined by calculating the molar sum of all detected BAs.

Statistical analysis
In the present study, we proposed a 3-step analytic approach, and several classi cation models were considered as follows. First, we used multivariate analysis to perform a global test to examine whether clinical characteristics (Model 1) or the entire set of all metabolites (Model 2) were related to a given outcome (NAFL development in T2DM). Second, if the metabolism panel was associated with the outcome, then as a second step, we further performed a feature selection procedure to identify the discriminative metabolites based on both univariate analysis (Kruskal-Wallis test) and multivariable analysis (orthogonal projections to latent structures discriminant analysis, OPLS-DA). The metabolites with p < 0.01 and variable in uence on projection (VIP) > 1.0 were considered as potential biomarkers in accounting for class discrimination. Subsequently, Model 3 was established based on these informative predictors. Finally, we investigated whether the addition of a panel of metabolites to clinical characteristics could improve the accuracy for classi cation. Here, we considered Model 1 as a baseline model, and Model 4 and Model 5 were established based on the combined predictors, respectively. The added discriminative power offered by the addition of metabolites to clinical characteristics was analyzed by comparison of model performance. All models were constructed based on exploration cohort, and their power of prediction was tested on validation cohort. All outliers have been included to analysis.
Model performances were compared in a pairwise manner by calculating the improvement in area under the curve (AUC), integrated discrimination improvement (IDI) and net reclassi cation improvement (NRI). To obtain the robust assessments, a 10000 bootstrap resamples were used to estimate the improvement in model performance assessments. IDI was used to determine the average increase in predicted probabilities [21], and NRI to assess model reclassi cation [22]. IDI > 0 and NRI > 0 represent the improvement of the prediction of outcome in the model.
We also evaluated the clinical usefulness and net bene t of the new predictive models using decision curve analysis according to the method described in JAMA Guide to Statistics and Methods [23]. Decision curve analysis is a method to evaluate the bene ts of a diagnostic test [23]. The X-axis shows threshold values for NAFL risk, and the Y-axis represents the net bene t for threshold probabilities. A decision curve that is far away from the "None" line (i.e., assume no NAFL case) and the "All" line (i.e. assume all NAFL cases) for prediction model can provide a high net bene t [24].
All values are shown as the mean ± SD for continuous variables and frequencies for categorical variables. Multivariate models were constructed using OPLS-DA. The number of orthogonal components was optimized using 10-fold cross-validation, and the models were assessed using cross-validation based metrics (R2X, R2Y, Q2Y values) and permutation analysis. The continuous variables were analyzed using non-parametric Kruskal-Wallis test, however the categorical variables were analyzed with Fisher's exact test. AUC from receiver operator characteristic (ROC) curves was performed to assess the incremental value of the metabolic markers for risk prediction of NAFL in T2DM patients. All calculations and multivariate analysis were performed using R software version 3.5.2 (www.r-project.org). A p-value of < 0.05 (two-tailed) was considered statistically signi cant. GraphPad Prism version 6.0 (GraphPad Software, San Diego, CA, USA) was used to generate histograms.

Clinical and metabolic characteristics of T2DM patients with or without NAFL
We recruited a total of 56 T2DM patients without NAFL and 47 T2DM patients with NAFL respectively.
According to the time of enrolment, the rst 30 T2DM patients with NAFL and 36 T2DM patients without NAFL were included in the exploration cohort, while the subsequent 17 with NAFL and 20 without NAFL were included in the validation cohort. The demographic, clinical and biochemical characteristics of these patients are shown in Table 1, including age, gender, body mass index (BMI), waist-hip ratio (WHR), fasting plasma glucose (FPG), post-load plasma glucose (2 h PG), ALT, triglyceride (TG), high-density lipoprotein (HDL) and low-density lipoprotein (LDL). Of note, important clinical characteristics such as age, gender and BMI, as well as serum levels of FPG, 2 h PG, HDL, LDL and TG, showed no signi cant differences between the two groups either in exploration cohort or in validation cohort. However, T2DM patients with NAFL had higher ALT levels than T2DM patients without NAFL both in exploration cohort and in validation cohort. The above information veri ed the comparability between T2DM patients with and without NAFL either in exploration cohort or in validation cohort.  Fig. S1, Table S1). BAs synthesized in the liver are termed primary BAs such as CA and CDCA, and those generated by bacteria are secondary BAs such as DCA and UDCA.

Predictive ability of the 5 NAFL-associated bile acids
In order to further test the predictive value of the above selected biomarkers, we then analyzed the AUC of biomarkers (Model 3  We also used decision curve analysis in order to assess predictors of NAFL in T2DM and the performance of the predictive models while combining biomarkers with clinical factors (Table 3, Fig. 3 and Supplementary Table S3). The decision curve analysis plot shows distinct net bene t of the biomarkers when added to clinical factors, which does justify their clinical use in decision-making (Fig. 3A).
In  (Table 3, Fig. 3B). These results further con rm that the addition of biomarkers improves the predictive value of the baseline clinical model, and the combined models achieve better performance in classi cation task that allows for the identi cation of NAFL in T2DM patients on the basis of bile acid characteristics ( Table 2, Supplementary Fig S2).

Discussion
In this study, we quantitatively pro led, for the rst time, bile acids in the serum of T2DM patients with or without NAFL based on targeted metabolomics, and combined traditional clinical characteristics to predict the development of NAFL, then further validated the results by another independent cohort study.
Finally, we de ned 5 serum bile acids extracted from the whole metabolites and combined traditional clinical characteristics to predict the NAFL status in T2DM patients. Interestingly, serum bile acid pro les in T2DM patients with NAFL were signi cantly different from T2DM patients without NAFL, as characterized by the signi cant elevation of LCA, TLCA, CDCA24G, TUDCA and TCDCA, which may be potential biomarkers for diagnosis of NAFL in T2DM based on UPLC-TQ/MS and OPLS-DA. To explore the potential ability of the biomarkers to discriminate NAFL status in T2DM patients, model performances were compared and it was shown that the addition of ve bile acid species (selected based on the exploration cohort, then validated in the validation cohort) to the clinical variables, improved the AUC, and categorical IRI, NRI for predicting the NAFL development in T2DM. Thus, the above evidences showed these biomarkers contributed to the improved model performance in predicting the development of NAFL in T2DM patients.
At the same time, these results suggest that the increase in bile acids may contribute to the formation of NAFL in T2DM patients, which could cause the high risk of occurrence of CVD and HCC. The increased ve bile acids may cause the progression of NAFL in T2DM patients with insulin resistance. Our motivation for performing this study was that biomarkers that improve discrimination of NAFL in T2DM patients will not only support the prediction of NAFL development but also promote to further acknowledging the roles of bile acids. However, the detailed mechanisms have been unclear until now.
It is shown that recent increase in the incidence of HCC is driven by NAFLD, especially in Western Countries [25]. Nevertheless, the morbidity and mortality of CVD in NAFLD patients are also very high [17]. It is therefore crucial to prevent the progression of metabolic diseases, including T2DM, dyslipidemia and NAFLD; moreover, elucidation of the pathogenesis of NAFLD in T2DM patients has become increasingly more important.
Arab et al [26] summarized current available data on the relationships of BAs to NAFLD. Cyrielle et al [13] found although serum total BAs were no signi cant differeces between NAFLD and non-NAFLD participants, the proportion of serum CA, CDCA conjugates were higher in NAFLD, while serum GHCA was lower than those non-NAFLD participants. Jiao et al [27] found that total serum bile acid levels in patients with NASH were about 3 times that of healthy controls. Total plasma bile acid levels were higher in obese and T2DM subjects than in healthy controls [28,29]. However, there is no systematical analysis about serum BAs between T2DM with or without NAFL. Puri et al [30] found that NAFLD was related to signi cantly changed circulating BA composition, unaffected by T2DM, and associated with histological features of NASH. These observations provided the foundation for future hypothesis-driven studies of speci c effects of BAs on speci c aspects of NASH.
In the present study we mainly found two classes of bile acids metabolites between the two groups including (A) CDCA and its metabolites (LCA, TLCA, CDCA24G and TCDCA) and (B) UDCA metabolite (TUDCA) as illustrated in Fig. 4. (A) CDCA and its metabolites CDCA plays an important role in controlling cholesterol homeostasis. Hydrophobic bile acids are known to induce hepatocyte injury experimentally, and the mechanisms are supposed as several different mechanisms, 1) mitochondrial and endoplasmic reticulum (ER) oxidative stress, 2) activate the death receptors FAS and tumor necrosis factor-related apoptosis-inducing ligand receptor 2 (TRAIL-R2) solubilization of the hepatocellular plasma membrane directly [31]. Here we found serum levels of CDCA metabolites such as LCA, TLCA, CDCA24G and TCDCA were higher in T2DM with NAFL than those without NAFL.
LCA increased in patients with liver disease as the most potent endogenous chemical causing liver toxicity [32]. LCA is converted from the primary BA-CDCA in the distal small intestine and colon after undergoing deconjugation and dehydroxylation [33]. LCA can be reabsorbed passively and constitute a portion of the total BA pool in the enterohepatic circulation [34]. Fickert et al [35] found that segmental bile duct obstruction, destructive cholangitis and periductal brosis in mice were caused by LCA feeding. LCA was reported to alter the levels of phospholipids, cholesterol, free fatty acids and TGs [36], and LCA disrupted phospholipid/sphingolipid homeostasis through TGF-β signaling [37]. Duboc et al [38] found that a decrease in serum LCA was implicated in the development of coronary atheromatous plaques.
However it was shown signi cant difference of LCA between the two groups only in exploration cohort, the possible reason may be the small sample size of the validation cohort.
TLCA is capable of impairing BA ow and inducing cholestasis [39]. It was shown that the contribution score of TLCA was high with high VIP score ( Fig. 2; Supplementary Table S1). However, serum concentration of TLCA was quite low compared to primary BA (CA, CDCA and their conjugates) ( Fig. 1;  Supplementary Table S1). These results suggested that LCA and TLCA are key products for developing NAFLD in patients with T2DM. It was shown that TLCA upregulated AP-1 proteins cFos and JunB in HepG2-Ntcp hepatoma cells [40]. Amonyingcharoen et al [41] showed that TLCA induced cholangiocarcinoma cell growth via the muscarinic acetylcholine receptor (mAChR) and epidermal growth factor receptor (EGFR) /extracellular signal-regulated kinase 1/2 (ERK1/2) signaling pathway. However, the detailed mechanisms of the increased serum TLCA level in T2DM patients with NAFL than those without NAFL were unclear. Marialena et al [42] reported that patients with NASH had increased fecal BA synthesis and high fecal levels of CA and CDCA, which suggested a possible role for BAs in the progression of NAFL to NASH. Li et al [43] found that children with NAFLD had higher levels of CDCAs, CDCA and CA, and lower levels of total DCA, DCA, TDCA, GDCA, GLCA, and total conjugated LCA (GLCA + TLCA) compared to subjects without NAFLD. Similar to the progression of NAFLD to NASH, the formation of NAFLD in T2DM patients likely involves not just one pathway but rather repeated injuries over time [44].
Trottie et al [45] concluded that UGT1A3 was the main UGT enzyme for the formation of hepatic CDCA-24G and glucuronidation inhibited the ability of CDCA to act as an FXR activator.
TCDCA as cytotoxic bile acids, changes in bile ow, biliary excretions of bile acids [46]. It was shown serum levels of ALT and AST were highly elevated in all rats given TCDCA [46], while another study showed that TCDCA as a signaling molecule showed obvious anti-in ammatory and immune regulation properties via protein kinase C (PKC)/Jun N-terminal kinase (JNK)-dependent pathway [47]. Song et al [48] quantitatively analyzed 15 biliary bile acids in cholangiocarcinoma (CCA) (n = 30), benign biliary disease (n = 57) and pancreatic cancer (n = 17) patients and discovered GCA and TCDCA as speci c CCA biomarkers. However, until now, the roles of CDCA-24G and TCDCA in the process of NAFLD in T2DM patients are unclear.
(B) UDCA metabolite UDCA, a hydrophilic bile acid, is used to treat a number of cholestatic disorders. There was a signi cant improvement in the levels of aminotransferases and steatosis in NAFLD patients treating with UDCA [49]. However, it was shown that UDCA did not offer a histological bene t compared with placebo in NASH patients [50]. Therefore, till now UDCA has not been recommended to treat NAFLD/NASH. Daniel et al [51] found that 24-norursodeoxycholic acid (NorUDCA), a side-chain shorted derivative of UDCA, may be a promising new approach in the treatment of cholestatic and metabolic liver diseases with anti-brotic, anti-in ammatory and anti-lipotoxic properties. The detailed mechanism of the increased serum level of TUDCA in T2DM patients with NAFL than those without NAFL remains unclear. Beuers et al [39] investigated that UDCA conjugates may improve the impaired secretory abilities of cholestatic hepatocytes.
It was reported that in the liver TUDCA improves insulin sensitivity and have cardiovascular protective effects by reducing ER stress [52]. It was reported that TUDCA attenuated hepatocarcinogenesis by suppressing carcinogen-induced ER stress-mediated cell death and in ammation without stimulating tumor progression [53]. However, until now there is no related research describing the effect of TUDCA on liver diseases.
There are still several limitations in our study to be addressed in the future studies. First, the sample size of groups may not be su cient to achieve more powerful data although we validated the results in an independent cohort study. Second, although fast serum samples were used for analysis, the information about diet, sedentary lifestyle was not available. Nevertheless, there is no information on other confounding factors that may in uence bile acid concentration such as treatments (e. g. statins) or the degree of insulin resistance [54,55]. Thus we could not entirely exclude the effect of those treatments on metabolism. Third, the diagnosis of NAFL in our study was based on ultrasonography but not con rmed by histological examination. The guidelines for NAFL in Europe, America and China all stipulate that liver histological examinations are only necessary in clinical trials of new drugs or clinical studies of NASH. In the present study, the patients are all NAFL patients, and it is not obligatory to perform histological examinations considering the risk and bene t of patients. However, it is better to perform histological examination than ultrasonography if condition permits. Nevertheless, it is better to further illustrate the mechanistic insight into the changes in the BA pro les in the pathogenesis of NAFL in T2DM from experimental studies using animal models and/or cell lines. In addition, an integrative omics method with other -omics, such as proteomics, transcriptomics and genomics, should be explored. We also intend to perform both in vivo and in vitro studies in the near future, to con rm the mechanisms of selected BAs, such as LCA, TLCA, TUDCA, CDCA24G and TCDCA, in the formation and development of NAFL in T2DM patients.

Conclusion
To the best of our knowledge, it is the rst study of this magnitude to provide a comprehensive association of bile acids and the progression of NAFL in T2DM patients. Our key nding is that the development and progression of liver disease in patients with T2DM (i.e., without NAFL→NAFL) are associated with metabolism disorders of serum bile acids. In summary, T2DM patients with NAFL have altered serum bile acid pro les compared to T2DM patients without NAFL, characterized by higher serum levels of LCA, TLCA, TUDCA, CDCA24G and TCDCA. Further understanding the role of bile acids in the pathogenesis of NAFL in T2DM patients may provide insights for the diagnosis of NAFL as well as potential therapeutic targets for treatment.   concentrations of BAs. * p<0.05, ** p<0.01, *** p<0.001, compared with T2DM patients without NAFL.