Metabolic Profile Reflects Stages of Fibrosis in Patients with Non-Alcoholic Fatty Liver Disease

Nonalcoholic fatty liver disease (NAFLD) is a leading cause of chronic liver disease worldwide, with fibrosis stage being the main predictor for clinical outcomes. Here, we present the metabolic profile of NAFLD patients with regards to fibrosis progression. We included all consecutive new referrals for NAFLD services between 2011 and 2019. Demographic, anthropometric and clinical features and noninvasive markers of fibrosis were recorded at baseline and at follow-up. Significant and advanced fibrosis were defined using liver stiffness measurement (LSM) as LSM ≥ 8.1 kPa and LSM ≥ 12.1 kPa, respectively. Cirrhosis was diagnosed either histologically or clinically. Fast progressors of fibrosis were defined as those with delta stiffness ≥ 1.03 kPa/year (25% upper quartile of delta stiffness distribution). Targeted and untargeted metabolic profiles were analysed on fasting serum samples using Proton nuclear magnetic resonance (1H NMR). A total of 189 patients were included in the study; 111 (58.7%) underwent liver biopsy. Overall, 11.1% patients were diagnosed with cirrhosis, while 23.8% were classified as fast progressors. A combination of metabolites and lipoproteins could identify the fast fibrosis progressors (AUROC 0.788, 95% CI: 0.703–0.874, p < 0.001) and performed better than noninvasive markers. Specific metabolic profiles predict fibrosis progression in patients with nonalcoholic fatty liver disease. Algorithms combining metabolites and lipids could be integrated in the risk-stratification of these patients.


Introduction
Nonalcoholic Fatty Liver Disease (NAFLD) affects a third of the general population in Western countries [1] and represents the most common cause of abnormal liver function tests. NAFLD includes a spectrum of pathological disorders, from simple steatosis to nonalcoholic steato-hepatitis (NASH) with inflammation and ballooning, and a certain degree of fibrosis up to cirrhosis [2]. The growing prevalence of NAFLD mirrors the epidemic of metabolic syndrome, mainly type-2 diabetes and obesity, to which it is closely associated [1]. Fibrosis stage represents the main predictor of clinical outcomes-liver-and nonliver-related-in this population [3]. As such, developing significant and advanced fibrosis marks a crucial point in the pathogenesis of the disease.
Currently, liver histology represents the gold standard for staging fibrosis in this population [4,5]. However, due to well-identified limitations of liver biopsy, such as the bleeding risk and the cost, it is unfeasible for all the patients to undergo such investigation. A plethora of noninvasive markers, such as ELF score, FIB-4, NAFLD fibrosis score and transient elastography, has therefore been developed over the last few years in an attempt to predict fibrosis stage [6], fast fibrosers and clinical events [7][8][9].
Over the last decade, metabolomic profiling has gained much popularity in the field of translational hepatology. There has been an increasing body of evidence hinting at a possible role of circulating aromatic amino acids as noninvasive markers for NAFLD severity. In a recent study, hepatocellular ballooning and inflammation, assessed by the NASH CRN scoring system, were associated with increased branched chain amino acids and aromatic amino acids, while fibrosis stages could be predicted by a combination of glutamate, serine and glycine [10]. Moreover, plasma branched chain amino acids correlated with NAFLD severity, more pronouncedly in women compared to men [11]. In another study, a combination of glycocholic acid, taurocholic acid, phenylalanine and branched chain amino acids could predict the presence of NASH accurately [12].
In this study, we analysed the metabolic profile of a well-phenotyped cohort of NAFLD patients and investigated its association with fast progressors of fibrosis.

Metabolic Profile in Fast Progressors vs. Nonfast Progressors
When comparing metabolites between fast progressors and nonfast progressors, 38 metabolites were significantly different ( Table 2). On multivariate analysis, only 14 metabolites were significantly associated with fibrosis progression (Table 3). Using binary logistic regression, a formula was generated to predict the progression of fibrosis, based on these metabolites: The metabolic profile generated from the above formula showed the ability to predict fibrosis progression in this population, with an AUROC of 0.788 (95% CI: 0.703-0.874, p < 0.001).
When comparing metabolites between fast progressors and nonfast progressors in those who had a biopsy, 37 metabolites were significantly different (Table 4).    When compared to those without NASH, only phenylalanine was significantly lower in those with NASH (0.07 mmol/L (0.06-0.08) vs. 0.08 mmol/L (0.06-0.09); p = 0.048). The AUROC of phenylalanine for predicting the presence of NASH was 0.381 (95% CI 0.269-0.493, p = 0.051), which is in keeping with poor diagnostic performance.
When comparing metabolites between fast progressors and nonfast progressors in those who had a biopsy, 37 metabolites were significantly different (Table 4).  On multivariate analysis, only one of these metabolites was significantly associated with fibrosis progression: H4CH, with an OR of 0.818 (95% CI: 0.722-0.927, p = 0.002). Using binary logistic regression and including baseline fibrosis stage, a formula was generated to predict the progression of fibrosis: 0.818 × H4CH (mg/dL) + 0.793 × Baseline Fibrosis Stage + 13.437 (2) The metabolic profile generated from the above formula showed the ability of prediction of fibrosis progression in this population, with an AUROC of 0.

Metabolic Profile in Cirrhotic vs. Noncirrhotic Patients
In this cohort, 21 (11.1%) patients had cirrhosis. When comparing to the noncirrhotics, 10 metabolites were significantly higher in the cirrhotic group (Table 5). Conversely, 10 different metabolites were significantly lower in the cirrhotic group (Table 5).

Discussion
NAFLD has become the most common cause of liver disease in the Western world and the fastest growing cause for liver transplantation [13]. Despite the big burden of the disease, identifying patients at risk of progression remains a challenge [14]. Of note, the recent advances in metabolomics and lipidomics may provide useful insights into the pathogenesis of the condition as well as new predictive tools for clinical outcomes in this population [15]. In this study, we analysed the metabolic profile of a well-phenotyped group of NAFLD patients from a tertiary care centre. We then evaluated their metabolic profile against fibrosis progression and severity.
Fibrosis stage represents the main predictor of clinical outcomes in patients with NAFLD [3]. Moreover, a faster progression of the liver disease translates into an earlier development of both hepatic and nonhepatic clinical outcomes [16]. As such, identifying those who are at higher risk for fibrosis progression may be clinically important in NAFLD patients. In this cohort, faster progressors presented a peculiar lipid profile, characterised by higher levels of very-low density lipoproteins (VLDL) (VLFC and V3PL) and triglycerides (H2TG) and low HDL (H4CH) ( Table 3), which were not observed in cirrhotics (Table 6). Overall, worsening insulin resistance is known to be characterised by elevated VLDL and triglycerides secondary to an impaired hepatic and systemic lipid metabolism [17]. On an opposite trend, sera from patients with cirrhosis were particularly enriched in low-density HDL1 lipoproteins (H1PL, H1CH) and apolipoprotein A2 (HDA2 and TPA2). Specifically, lower levels of VLDL and triglycerides may reflect reduced hepatic synthetic function, greater porto-systemic shunting and relative malnutrition [18]. Moreover, an impaired cholesterol efflux capacity [19], as well as lower lipoprotein scavenger activity [20], may be responsible for elevated HDL and apolipoproteins in those with cirrhosis. There was no difference in terms of statin treatment, suggesting that

Discussion
NAFLD has become the most common cause of liver disease in the Western world and the fastest growing cause for liver transplantation [13]. Despite the big burden of the disease, identifying patients at risk of progression remains a challenge [14]. Of note, the recent advances in metabolomics and lipidomics may provide useful insights into the pathogenesis of the condition as well as new predictive tools for clinical outcomes in this population [15]. In this study, we analysed the metabolic profile of a well-phenotyped group of NAFLD patients from a tertiary care centre. We then evaluated their metabolic profile against fibrosis progression and severity.
Fibrosis stage represents the main predictor of clinical outcomes in patients with NAFLD [3]. Moreover, a faster progression of the liver disease translates into an earlier development of both hepatic and nonhepatic clinical outcomes [16]. As such, identifying those who are at higher risk for fibrosis progression may be clinically important in NAFLD patients. In this cohort, faster progressors presented a peculiar lipid profile, characterised by higher levels of very-low density lipoproteins (VLDL) (VLFC and V3PL) and triglycerides (H2TG) and low HDL (H4CH) ( Table 3), which were not observed in cirrhotics (Table 6). Overall, worsening insulin resistance is known to be characterised by elevated VLDL and triglycerides secondary to an impaired hepatic and systemic lipid metabolism [17]. On an opposite trend, sera from patients with cirrhosis were particularly enriched in low-density HDL1 lipoproteins (H1PL, H1CH) and apolipoprotein A2 (HDA2 and TPA2). Specifically, lower levels of VLDL and triglycerides may reflect reduced hepatic synthetic function, greater porto-systemic shunting and relative malnutrition [18]. Moreover, an impaired cholesterol efflux capacity [19], as well as lower lipoprotein scavenger activity [20], may be responsible for elevated HDL and apolipoproteins in those with cirrhosis. There was no difference in terms of statin treatment, suggesting that changes in metabolic profile were due to primary disturbances. Hence, lipid profile may reflect, to some extent, the course of the progression of the liver disease, moving from a phase of florid metabolic dysfunction with fibrosis progression to a less atherogenic profile with established cirrhosis.
Among the metabolites, proline was independently associated with fibrosis progression in this population (Table 3). Proline and its derivate hydroxyproline represent a major player in the collagen synthesis of 30% of the body proteins [21]. Proline also acts as a stabilizer for the helical structure of collagen fibres in the liver [22]. Moreover, previous studies demonstrated that proline uptake increases in early stages of acute steatohepatitis and is proportional to collagenogenesis in animal models [23]. On a similar note, lysine, an essential amino acid mainly catabolised by the liver, was independently associated with the presence of cirrhosis. Previous studies have associated lower lysine levels with collagen disturbances as a result of overexpression of the enzyme lysil oxidases [24]. Under physiological conditions, lysil oxidases deaminate lysine residues for maintaining the structural integrity of the extra-cellular matrix. In pathological conditions such as fibrogenesis, such an enzyme is overexpressed and promotes collagen cross-linking and stabilisation against proteolytic degradation, maintaining hepatic stellate cells in an activated state [25]. Moreover, higher levels of pipecolic acid, one of lysine's catabolites, were previously described in patients with chronic liver disease and cirrhosis [26]. Taken together, these results suggest a potential role in measuring serum proline and lysine as a biomarker of hepatic collagen turnover in patients with NAFLD. Further prospective studies are required to explore their role as predictors of liver-related events in this population.
Finally, a combination of lipoproteins and metabolites gave an excellent prediction of fibrosis progression ( Figure 2) and presence of cirrhosis ( Figure 3). With regards to fibrosis progression, metabolic profile performed equal or better than FIB-4, NAFLD fibrosis score and liver functions tests, which are currently used to stratify patients at risk for more severe liver disease in clinical practice [4]. Moreover, while previous studies demonstrated that specific metabolic profiles may distinguish those with simple steatosis from those with NASH [27], this is the first study exploring the association of metabolomics with fibrosis progression and severity. Unfortunately, the small number of events in this population did not allow for an internal validation, while an external cohort was not available to test these findings at the time of the work. Future work should focus validating these results in longitudinal studies as well as external cohorts.
Further studies are required to validate these results in longitudinal studies as well as external cohorts.
In this study, we demonstrated that specific metabolic profile could predict fast fibrosis progression in a cohort of patients with nonalcoholic fatty liver disease. Metabolic profile performed better than traditional noninvasive markers of fibrosis. In an era of precision medicine, algorithms combining metabolites and lipids may provide comprehensive tools to stratify patients with NAFLD. Integrating clinical features and multiomics results may lead to a better understanding of the phenotypes of the patients and may allow for the capture of the complexity of the disease.

Study Population
This study included all consecutive new referrals to the specialist NAFLD clinic at St Mary's Hospital (Imperial College Healthcare NHS Trust, London, UK) between 2011 and 2019. Exclusion criteria were the use of steatogenic drugs, excess alcohol consumption (defined as alcohol consumption greater than 14 UI per week) as well as other concomitant liver diseases.
Demographic, anthropometric and biochemical data were collected at the time of the baseline fibroscan or at the time of the liver biopsy. Ethnicities were clustered into 6 groups (Table 1). If ethnicity was not specified by the patient, it was classified as Other. All comorbidities, such as hypertension, type-2 diabetes mellitus (T2DM) and hypercholesterolaemia, were recorded. Transient elastography (TE) was performed by an experienced physician after 4 h fasting and allowed for the assessment of liver stiffness measurement (LSM) and controlled attenuation parameter (CAP). A cut-off of LSM ≥ 8.1 kPa was considered to be significant fibrosis, while LSM ≥ 12.1 kPa was considered to be advanced fibrosis [4]. All patients were monitored every 6 months for more than one year, with clinical data documented at subsequent consultation. Liver biopsies were performed when clinically indicated. Liver biopsy specimens were formalin-fixed, paraffin-embedded, stained and scored by an expert liver pathologist as per the NASH CRN scoring system [2]. NASH was defined based on NAS score ≥ 5.
Cirrhosis was diagnosed either histologically or clinically, as a combination of biochemical, imaging and elastographic features. Delta stiffness was calculated as the difference in LSM over a set time period divided by number of months. Patients were divided into fast progressors and nonfast progressors when delta stiffness was more than 1.03 kPa/year or less than 1.03 kPa/year, respectively. This cut off was identified using the top 25% of our cohort, which is in line with previous literature [28].

Metabolic Profile
Fasting serum samples were collected for all patients included in the study. They were then centrifuged and stored at −80 • C in the Imperial Hepatology and Gastroenterology Biobank (Imperial College London, London, UK). Targeted and untargeted metabolomic profiles were carried out using proton nuclear magnetic resonance ( 1 H NMR). Overall, values for 27 small metabolites and 112 lipoproteins were obtained for each serum sample, as per published protocol [29].

Statistical Analysis
Distribution of the variables was identified using the Shapiro-Wilk normality test, which suggested a nonparametric distribution of the data, and therefore, nonparametric analyses were applied. Descriptive statistics were presented by the median and interquartile range for continuous variables or number and percentage for categorical variables. Difference between groups was measured using the Mann-Whitney U test and Kruskal-Wallis for continuous variables, while Pearson's chi-squared was used for categorical variables. A Bonferroni-corrected Dunn's test was used for pairwise analysis of variables within which there were multiple groups. Significant variables were carried forward to multivariate analysis to identify odds ratio (OR) of the variables associated with clinical outcomes. Binary logistic regression was then used to generate a formula collating the variables which were significantly associated with the clinical outcomes on multivariate analysis. ROC (receiver operating characteristic) curves were used to assess the diagnostic performance of the combination of variables identified with the analysis. Sensitivity, specificity and Youden indexes were estimated for a given cut-off. All tests were two-sided, and a p-value 0.05 was considered significant.

Ethics
This study was retrospective and included only fully anonymised data from investigations and assessments performed as per standard of care. As such, ethical approval was not required, as stated by the UK policy framework for health and social care. Liver tissue and plasma were stored at Imperial Hepatology Gastroenterology Biobank, which was fully REC approved by Oxford C Research and Ethics Committee under REC reference 16/SC/0021.