A branched‐chain amino acid‐based metabolic score can predict liver fat in children and adolescents with severe obesity

Summary Background Eighty percent of adolescents with severe obesity suffer from non‐alcoholic fatty liver disease (NAFLD). Non‐invasive prediction models have been tested in adults, however, they performed poorly in paediatric populations. Objective This study aimed to investigate novel biomarkers for NAFLD and to develop a score that predicts liver fat in youth with severe obesity. Methods From a population with a BMI >97th percentile aged 9‐19 years (n = 68), clinically thoroughly characterized including MRI‐derived proton density fat fraction (MRI‐PDFF), amino acids and acylcarnitines were measured by HPLC‐MS. Results In children with NAFLD, higher levels of plasma branched‐chain amino acids (BCAA) were determined. BCAAs correlated with MRI‐PDFF (R = 0.46, p < .01). We identified a linear regression model adjusted for age, sex and pubertal stage consisting of BCAAs, ALT, GGT, ferritin and insulin that predicted MRI‐PDFF (R = 0.75, p < .01). ROC analysis of this model revealed AUCs of 0.85, 0.85 and 0.92 for the detection of any, moderate and severe steatosis, respectively, thus markedly outperforming previously published scores. Conclusion BCAAs could be an important link between obesity and other metabolic pathways. A BCAA‐based metabolic score can predict steatosis grade in high‐risk children and adolescents and may provide a feasible alternative to sophisticated methods like MRI or biopsy in the future.

The need for characterization of a high-risk group for NAFLD and targeted approaches early in life was formally acknowledged by the Committee on prevention of obesity in children and youth of the U.S. Institute of Medicine. 10 Reliable quantitative determination of liver fat content, however, is based on either biopsy or sophisticated imaging techniques 11 only available in specialized centres. Thus, there is a vital need for reliable but simple biomarkers for quantification of liver fat content. To this objective, biochemical markers and routine measures have been evaluated. GGT and ALT, known measures of liver disease, have been proposed as independent predictors of NAFLD. 12 However, Wong et al. found that ALT levels show a high variability on repeated testing and do not reliably diagnose NAFLD nor correlate with histologic grading. 13 Therefore, metabolic risk factors were proposed as the basis of NAFLD evaluation, but most studies are limited by evaluating NAFLD solely as a dichotomous parameter. [14][15][16][17] Liver fat content seems to directly relate to the degree of metabolic disease and should therefore be quantitatively determined. 18,19 Metabolomics approaches showed that amino acid patterns are more strongly associated with metabolic health than traditional laboratory and also lipid markers. 20 Big cohorts identified circulating branched chain amino acids (BCAAs) to be chronically elevated in individuals with obesity. Compelling evidence derived from rodent studies highlights their causal connection to the risk of T2D and insulin resistance and cardiovascular disease. [21][22][23] Hence, elevated BCAAs are characteristic for deteriorated metabolic health 20,22,24,25 already early in life 16,18 and predict future disease risk. 16,22,26 BCAAs have been shown to promote intrahepatic fat accumulation in an animal model 27 and are elevated in human individuals with NAFLD, [14][15][16][17][18][19] suggesting a link between impaired amino acid metabolism and liver fat accumulation. 17 The aetiology and pathophysiological pathways of increased BCAA levels in obesity is still unclear, but may involve chronic low grade inflammation by inducing pro-inflammatory gene expression in adipose tissue, 28,29 thereby further deteriorating obesity effects on metabolic health, also in cardiovascular 24,25 and liver disease. 27 Substantial evidence that links BCAA dysmetabolism to a metabolically unhealthy phenotype with obesity including steatosis, hepatic injury or inflammation [14][15][16][17][18][19]27,30 has been published. Therefore, defining a BCAA-related metabolic signature that indicates the liver fat content could not only help to identify and monitor patients with NAFLD and increased cardiometabolic risk, 5,21,23-26 but also to elucidate the underlying pathomechanisms.
Early interference in paediatric patients to prevent progression of NAFLD and other obesity-related disorders is highly desirable. 2,10 Furthermore, the pathogenesis of NAFLD in children and adolescents is much less investigated and may significantly differ compared to adults. 31 Therefore, studies in paediatric patients are strongly needed. The only existing study with an accurate quantification of liver fat content by MRI analyzing amino acid levels in children and adolescents investigated a majority of non-Caucasian individuals. 18 Plasma concentrations of BCAAs were shown to be associated with intra-hepatic fat content independently of the degree of obesity and insulin resistance. 18 Although these results may not directly be applicable to European cohorts, 9,32 they provide a strong indication for BCAAs to be investigated in Caucasian paediatric patients in relation to NAFLD. Same accounts for acylcarnitines, which are not only linked to fatty acid metabolism but some of the shorter forms also to BCAA metabolism. Interestingly, the latter have been shown to be linked to insulin resistance. 15,21,26 To fill these gaps in research in the high-risk group of youths with severe obesity and to further contribute to elucidation of the complex mechanisms leading to paediatric NAFLD, we investigated whether circulating amino acid levels are associated with liver fat content as measured by MRI in children and adolescents with severe obesity.
Moreover, a plethora of amino acids, acylcarnitines and established clinical as well as experimental markers for liver function, metabolic state, and inflammation was considered for development of a simple and thus practicable score to predict liver fat content. The resulting score includes BCAAs, ALT, GGT, ferritin and insulin to predict liver fat content in paediatric patients with severe obesity with high accuracy.

| Amino acids and acylcarnitines
Plasma amino acid and acylcarnitine concentrations were determined on a Waters Acquity UPLC-coupled Xevo TQD mass spectrometer using non-derivatized a semi-quantitative kit for dried blood spots from Chromsystems (Gräfeling, Germany), which was modified by directly pipetting 1.3 μL plasma, sampled and stored as described above, into extraction buffer. For confirmation, amino acids were additionally quantified from fresh plasma of 30 of the patients using the EZ:faast kit from Phenomenex (Torrance, CA) on a Waters Q-Micro HPLC-coupled mass spectrometer with essentially same results (not shown). BCAA concentrations were calculated by addition of valine, leucine and isoleucine values.

| Statistics
In order to perform group-wise comparisons, steatosis was catego-

| Ethics
The study protocol was approved by the ethics committee of the

| RESULTS
Characteristics of the study population are shown in Table 1.
Sixty-eight patients with mean age of about 13 years completed MRI evaluation of the liver and were included in the study. All were of Caucasian ethnicity. Of the investigated parameters, HOMA-indices, liver transaminases, GGT, total cholesterol, insulin, triglycerides, ferritin, PCT, TNFα and CK-18 significantly correlated with liver fat content (p-value ≤ .02 for all, Table S1).
To assess the metabolic profile of the patients, 10 amino acids and 10 acylcarnitines were determined and exploratively analyzed for associations. In particular BCAAs and related acylcarnitine concentrations appeared to be linked to fatty liver: The correlation of BCAAs with liver fat content (p-value < .01, R = 0.46) also after adjustment for gender, age and pubertal stage is shown in Figure 1. Additionally, the ANOVA showed significant differences in BCAA levels between steatosis stages (p-value .03). Also concentrations of acylcarnitines C3 and C4 significantly differed between the groups (Table 1)

| Model evaluation
In the next step, we assessed how our model performed compared to previously published scores with c-statistic. Notably, our model had a higher accuracy than previously proposed scores for the detection of steatosis in children with severe obesity (Figure 4).
The ability of our model to detect patients with at least mild steatosis (>5.1% MRI-PDFF) was 91.7% (=sensitivity), while specificity to detect patients without the disease was 35%.
Among those who had positive tests for mild steatosis (predicted value >5.1%), the probability of having steatosis (=PPV) was 73.3%.
Among those who had negative Test (<5.1%), the probability of being disease free (=NPV) was 75%.
For the detection of at least moderate steatosis, the sensitivity was 71.4%, specificity 85.7%, PPV 75% and NPV 83.3%. We validated our model in an independent validation cohort of 32 patients (Table S2). Performance of the BCAA-based model was evaluated by calculating the AUC (Table S3)

| DISCUSSION
Here, we show that an incomplex model is able to accurately predict fatty liver. Since childhood obesity is a major health threat worldwide, as is concomitantly NAFLD, our score may be valuable to recognize keto acid dehydrogenase kinase may be a factor explaining chronically increased circulating BCAA levels. 30,47 As a consequence of enhanced BCAA levels, chronic activation of mTOR may be induced leading to increased oxidative stress (ROS) and suppressed autophagy 48,49 and thus may promote lipid accumulation and lipotoxic liver injury. 27,48 Therefore, the underlying pathways linking BCAA metabolism to metabolic disease could potentially be pharmacologically targeted to improve hepatic and overall insulin resistance as well as reduction of liver fat content, underlining the utmost importance of further investigations in this field. 30

| LIMITATIONS
One limitation of the current study is the relatively small sample size.
Hence, external validation of our proposed score in bigger cohorts is needed. Since we only included youth with severe obesity from our tertiary care centre in Vienna, Austria, the utility of our score in children with normal weight and of non-Caucasian ethnicity remains to be determined.
Strengths of this study include its prospective character providing a close correlation between blood sampling and liver MRI, which were all performed within a short time frame (8 weeks). Additionally, strict criteria of inclusion were followed: All patients were extensively tested and excluded from this study if autoimmune, infectious or drug-induced liver disorder was suspected providing a wellcharacterized homogenous cohort. Importantly, all tests were corrected for age, gender and pubertal stage.

CONFLICTS OF INTEREST
The authors declare no conflict of interest.