Herein, a subsample of 210 NAFLD cases and control individuals from a Greek case-control study was used and detailed methodology can be found elsewhere (12). Adult individuals with no self-declared concomitant liver injury at the time of recruitment were screened for NAFLD. Exclusion criteria included any liver disease, chronic viral hepatitis, hepatotoxic drugs exposure, excessive alcohol consumption, a life-threatening disease or psychiatric disorders impairing the patient’s ability to provide written informed consent, as well as pregnancy or lactation. All study subjects were informed about the aims of the study and signed a written consent. This study was approved by the Ethics Committee of Harokopio University of Athens (38074/13-07-2012), based on the Helsinki Declaration.

Serum samples collected from the individuals were thawed at room temperature and extracted according to Nagana Gowda et al. (16) protocol. In particular, 250 μL serum samples were extracted with 500 μL methanol (1:2 v/v), vortexed and placed at -20 °C for 20 min. Samples were centrifuged (11.000 rpm/4 °C) for 20 min to allow protein parts to precipitate. Supernatants were collected and evaporated to dryness. Samples were reconstituted to 400 μL D₂O and 150 μL phosphate buffer (0.2M, Na₂HPO₄ 2H₂O and NaH₂PO₄, pH=7.0) with internal standard trimethylsilyl propionic acid sodium salt (TSP) (2.75 mM) and were transferred to 5 mm NMR tubes for ¹H-NMR analysis.

¹H-NMR spectra were acquired using a Varian 600 MHz NMR spectrometer equipped with a 1H{13C/15N} 5mm PFG Automatable Triple Resonance probe at room temperature (25 °C). The CPMG pulse sequence with presaturation for water suppression was applied. In total, 128 transients were collected with 64 K data points, a relaxation delay of 5 s and an acquisition time of 4 s, Receiver gain was kept constant for all acquisitions. ¹H NMR spectra were referenced at TSP chemical shift (0.00 ppm) and processed with 0.3 exponential line broadening (17).

¹H-NMR spectra were preprocessed with MestreNova (v. 10.1, Santiago de Compostela, Spain) software. Manual phase correction, automatic baseline correction and sinc apodization were applied to improve spectra resolution. Binning of 0.001 ppm was selected. The water D₂O region (4.68 - 5.00 ppm) was excluded together with the peak at 3.36 ppm attributed to methanol residual. The spectra were normalized to the standardized area of the reference compound (TSP), manually aligned and converted into ASCII format using the software Mnova ver.10.1.

The circulatory metabolome provides a rational compartment to interpret metabolite variations, therefore the first step in our endeavor was to implement PCA on the host of serum blood samples.

In this direction, a PCA model with two components was computed explaining 48.0% of the data variation, featuring good predictability (Q²(cum) = 0.46) to provide an overview of the samples, highlighting a possible clustering and pinpointing strong outliers (Figure 1). Interestingly, of the several characteristics included in this PCA model, such as the absence or presence of NAFLD pathology of different grades and the presence of comorbidities, such as hyperlipidemia, MetS, DMII and hypertension, the observed trend along the second principal component probed to a differentiation based only on the sex of the subjects. In fact, the samples of male sex mainly localize in the first and second quadrants, while the samples of female sex assemble to a great extend in the third and fourth quadrants. This unsupervised overview highlights sex as a prominent factor for these samples’ differentiation, since the second principal component explains a high amount of variance with an eigenvalue (i.e., 10.5). This result further enhances the notion that liver is metabolically distinct for each sex (6), herein literature has noted that hepatic injury and inflammation are related to sex and diverse sex hormone levels. Therefore, in an attempt to alleviate the impact of this factor, each sex was further explored under the scope of OPLS-DA models.

The altered metabolites exhibiting AUROC>0.7 were submitted to the Pathway analysis module of MetaboAnalyst 5.0 to identify perturbed metabolic pathways in response to NAFLD (Figure 6).

The obtained results concerning the male sex serum substrate revealed the Phenylalanine, tyrosine and tryptophan biosynthesis pathway as statistically significant (p < 0.05) disturbed (Figure 6A). On the other hand, the statistically significant perturbed pathways (p < 0.05) for the samples of female sex samples included glycine, serine and threonine metabolism, tyrosine metabolism, phenylalanine, tyrosine and tryptophan biosynthesis and synthesis and degradation of ketone bodies with the former two pathways containing at least two metabolites (Figure 6B).

In order to further assess the underlying biological perturbation we also implemented Metabolite Set Enrichment Analysis (Figures 7A, B). In fact, for the male samples the altered metabolites were mainly involved in the Glycine and Serine Metabolism, Glutathione Metabolism, Selenoamino Acid Metabolism, Alanine Metabolism, Urea Cycle, Tryptophan Metabolism and Glutamate Metabolism. Whereas for the female samples the altered metabolites were mainly involved in the Glucose-Alanine Cycle, Sphingolipid Metabolism, Glycolysis, Galactose Metabolism, Gluconeogenesis, Lactose Synthesis, Lactose Degradation, Transfer of Acetyl Groups into Mitochondria, Warburg Effect, Glycine and Serine Metabolism, Catecholamine Biosynthesis, Thyroid hormone synthesis, Selenoamino Acid Metabolism, Urea Cycle, Tryptophan Metabolism, Tyrosine Metabolism, Phenylalanine and Tyrosine Metabolism, Threonine and 2-Oxobutanoate Degradation, Glutathione Metabolism, Alanine Metabolism and Glutamate Metabolism.

The revealed metabolites exhibiting AUROC>0.7 were uploaded to the Network Explorer analysis module of Metaboanalyst platform and a Metabolite-Disease Interaction Network was extracted identifying metabolite-disease connections that cross pathway boundaries. The exploration of disease-related metabolites for male and female sex samples is displayed in (Supplementary Figures S4A, B) respectively. Briefly, a link between liver steatosis and mental health was highlighted in both sexes.

Our study revealed higher serum concentration of the glucogenic and ketogenic amino acid tyrosine and of the glucogenic amino acids alanine and histidine, in both sexes with NAFLD (grade 2 & 3) and fluctuations of ketone bodies or acetic acid. In fact, NAFLD severity was positively correlated with the serum levels of tyrosine (AUROC=0.716 for male and AUROC=0.756 for female), alanine (AUROC=0.774 for male and AUROC=0.728 for female) and histidine (AUROC=0.684 for male and AUROC=0.712 for female) (Figures 3, 5).

Fluctuations in circulating aromatic amino acids are often detected in individuals with NAFLD. High tyrosine concentration in patients with NAFLD has been highlighted in a number of publications (8) possibly indicating a dysregulation in hepatic metabolism of this amino acid (30). Furthermore, increased levels of tyrosine has been observed related to IR (30, 31) and varying proportional to the severity of the fibrotic stage (31). Besides, phenylalanine is irreversibly converted to tyrosine mainly in the liver (30), thus the aromatic amino acid biosynthetic pathway emerges disturbed in both sexes (Figure 6).

Increased circulating alanine levels in NAFLD patients have also been linked with hepatic IR (8, 30), which is associated with the pathogenesis of NAFLD. Through the glucose–alanine cycle, alanine’s release from the muscle during starvation and delivery in the liver enables its conversion to pyruvic acid, which enters in the gluconeogenic pathway and regenerates glucose. Under this prism, the revealed imbalance in the glucose–alanine cycle and alanine metabolism (Figure 7) probes to a disturbed liver-skeletal muscle metabolic crosstalk (32). Recent findings point to both transcriptionally and non-transcriptionally regulation of amino acid metabolism by glucagon. In this line, elevated concentrations of alanine have been pinpointed in subjects with NAFLD and hyperglucagonemia (33).

The implemented biomarker analysis highlighted the significant increase of histidine along the progression of NAFLD for the female subjects and a respective increase for the male subjects. Few studies to date underline the contribution of histidine to the NAFLD pathogenesis (34). Of note, an animal study probed to the contribution of a histidine-excess diet in rats to the high occurrence of minor fibrosis and liver cell damage (35).

Higher serum concentrations of valine, aspartic acid and mannose were positively associated with the progression of NAFLD among the male subjects while a negative association was observed with the levels of creatine, phosphorylcholine and acetic acid (Figure 3).

Valine displayed significant increase (AUROC=0.717) in male NAFLD subjects. A sufficient body of evidence supports the positive correlation of high circulatory BCAAs with liver diseases including NAFLD occurrence (36). It has been observed that males with NAFLD have greater plasma levels of valine, leucine, and isoleucine than females while a sex-dependent relationship of plasma BCAAs with NAFLD severity has been reported (37). Animal studies have probed the expression of estrogen receptor ERa in the liver as key regulator of hepatic lipid metabolism highlighting also a differentiated metabolic function between the two sexes. Interestingly, the BCAA metabolism emerges as the most affected pathway due to the lack of hepatic ERa particularly in males (38).

Mannose has been also revealed as significantly increased in NAFLD male subjects (AUROC=0.703). Mannose is related to hepatic glycogen breakdown and has been associated together with Valine, Isoleucine and Glutamate among other metabolites with increased risk of developing T2D in a large prospective study (39).

Aspartic acid has emerged bearing sub-optimal statistical significance (AUROC=0.687). Interestingly, aspartic acid is involved in urea cycle, an essential hepatic metabolic function which is compromised in NAFLD and this may potentially link to the increased blood concentration of involved amino acids (40). Urea cycle dysregulation induces increased ammonia levels in liver and blood and could be the driving cause for hepatic inflammatory response and fibrosis, cognitive dysfunction, secondary sarcopenia, immune dysfunction and malignancy (41).

Creatine, phosphorylcholine and acetic acid were elucidated as metabolites related to a healthier liver, which are consistent with previous researches. Particularly, creatine which can be obtained from dietary sources but it is also de novo synthesized in the liver, has been observed in elevated concentration in the healthy male subjects (AUROC=0.715). Other in vivo and in vitro researches have portrayed that treatment with creatine, aids in the prevention of causes for fatty liver disease. In fact, the expression of transcription factors (PPARα and PPARγ) and of key genes related to fatty acid metabolism is modulated by creatine (42). Thus, PPAR agonists are intensively investigated in the treatment of NAFLD (43).

Furthermore, the healthy male subjects displayed higher levels of phosphorylcholine (AUROC=0.7), a precursor and a breakdown product of phosphatidylcholine. Low levels of phosphatidylcholine, which is vital for cell membranes and lipoproteins, have been shown to enhance fat accumulation and drive de novo lipogenesis (44). Hepatic phosphatidylcholine metabolism is linked to NAFLD (45), and variations in phospholipids subclasses have been documented in subjects with and without NASH. Interestingly, NAFLD/NASH patients have been found with a lower phosphatidylcholine-to-phosphatidylethanolamine ratio than healthy patients (46).

Acetic acid was also pinpointed as a potential marker for male healthy liver although with sub-optimal significance (AUROC=0.68). This metabolite deriving from the diet or gut microbiota fermentation has an important influence on fatty acid metabolism and may suppress NAFLD/NASH development by regulating hepatic lipid metabolism and insulin sensitivity via its receptor FFAR2 (free fatty acid receptor 2) in hepatocytes (47). In this direction, animal studies have highlighted acetic acid as an important metabolite in the hindrance of lipid accumulation as it advances lipolysis as well as fatty acid oxidation and inhibits fatty acid synthesis (48).

The metabolites glucose, glycine, acetoacetate, acetone and threonine were distinctly associated with the progression of NAFLD among the female subjects (Figure 5).

In the liver, insulin fails to stimulate glucose metabolism when in an insulin-resistant (IR) state and, as a result the production of endogenous glucose increases and insufficient glucose disposal occurs. Congruent with this supposition our study highlighted the increased concentration of serum glucose in NAFLD female subjects as compared to controls (AUROC=0.802). Previously, when the isotope tracer approach was applied, it pinpointed that NAFLD patients produced elevated glucose concentration (hepatic insulin resistance) and lower glucose uptake (peripheral insulin resistance) in spite of increased insulin levels in comparison to the healthy subjects (49). Estrogens acting mainly through ERα receptors suppress hepatic glucose production by reducing gluconeogenesis and increasing glycogen synthesis and storage (25). The higher circulating glucose level in Grade 2 & 3 NAFLD female subjects possibly mirrors the lack of estrogens due to menopause status as can be inferred by the age characteristics of the female subjects, (48.63 ± 10.84) years for moderate and (53.43 ± 6.80) for severe hepatic steatosis (Table S4). In accordance, MSEA revealed the disturbed Glycolysis and Gluconeogenesis pathways (Figure 7B).

Glycine serum concentration was revealed as significantly disturbed between controls and NAFLD cases (AUROC=0.702) (Figure 5). Besides its dietary source, glycine is also biosynthesized mainly from serine derived from diet or de novo synthesized from the 3P-glycerate intermediate of the glycolysis pathway mainly in the kidneys. Through one-carbon metabolism glycine is involved in the biosynthesis of purines and of the major intracellular antioxidant glutathione (50). In accordance, metabolite pathway analysis and MSEA revealed the disturbed Glycine and Serine Metabolism and the Glutathione metabolism (Figures 6B, 7B). The pathogenic role of oxidative stress on NAFLD is well established and decreased levels of glycine are reported in NAFLD patients (50). Furthermore, transcriptomics analysis in the livers of NASH patients have revealed suppression of glycine biosynthetic genes (51). Thus, the apparent increase of glycine concentration in NAFLD cases as revealed from our study is inconsistent with literature. This triggered our attention to explore further this important neurotransmitter’s variation through the different stages of hepatic steatosis. Analysis of variance (ANOVA) (p= 0.049076) through box–plots presentation (Supplementary Figure S3) revealed the trend in each hepatic steatosis stage. An initial decrease of glycine serum concentration in mild hepatic steatosis was followed by a recovery in moderate hepatic steatosis probably attributed to a transient response to oxidative stress or associated to microbial metabolism following a special dietary approach while the severe hepatic steatosis clearly featured a significant decay of glycine abundance in accordance with literature (Supplementary Figure S3). Perturbations in the levels of glycine have been associated to menopausal hormone therapy, albeit this warrants further investigation (29).

Our study elucidated fluctuations in the concentrations of ketone bodies, acetoacetate (AUROC=0.703) and acetone (AUROC=0.691), (Figure 5), the former bearing higher statistical significance. It has been postulated that ketogenesis holds a critical role in hepatic lipid disposal and is activated in states of increased fatty acids, decreased carbohydrates and/or low circulating insulin (52). Increased hepatic influx of fatty acids upregulates the mitochondrial β-oxidation pathway and the release of acetyl-CoA which can be directed towards ketogenesis or enter the TCA cycle and either catabolized to citrate and exported to the cytoplasm directed towards de-novo lipogenesis or oxidized and induce gluconeogenesis. Impairment in ketogenesis diverts acetyl-CoA in increased rates of gluconeogenesis and de-novo lipogenesis thus promoting NAFLD pathogenesis (52, 53). Evidence from human and animal studies supports the hypothesis that ketogenesis is initially increased in NAFLD in response to the increased liver lipid accumulation, followed by an impaired mechanism as NAFLD progresses reflecting in the reduction in circulatory ketone body levels (53). In a recent large prospective Dutch cohort study, circulating ketone bodies were upregulated in subjects suspected with NAFLD as determined from the calculated fatty liver index (FLI), while NAFLD and higher circulating ketone bodies were related to a high risk of all‐cause mortality (54). On the other hand, using stable isotope tracers, a study demonstrated that NAFLD subjects (24-hour-fasted) displayed resistance to ketosis associated with increased disposal of acetyl-CoA in the TCA cycle and increased hepatic glucose production (55).

Our study prompts also to an impaired ketogenic pathway mirrored in the significantly decreased levels of acetoacetate associated with the moderate and severe hepatic steatosis. Moreover, the higher circulating glucose level observed also in those subjects potentially advocate to a concomitant divergence of acetyl-CoA in the TCA cycle as an adaptive mechanism in agreement with literature. The complex interplay of the disturbed metabolic pathways is revealed in the metabolite pathway analysis and the MSEA (Figures 6B, 7B) including synthesis and degradation of ketone bodies and gluconeogenesis, glycolysis which generates pyruvate which either oxidates to acetyl-CoA (transfer of Acetyl Groups into Mitochondria) or converts to lactate (Warburg effect) which via inhibition of Histone Deacetylase (HDAC) stimulates de-novo lipogenesis pathway (56).

Interventions and therapeutics targeting ketogenic pathway activation holds promise for the treatment of NAFLD. Peroxisome proliferator-activated receptor α (PPARα) is a ligand activated transcription factor which mediates the expression of a number of genes involved in hepatic lipid metabolism including ketogenesis (57). Selective PPARa modulators are extensively investigated for NAFLD/NASH management (58).

Finally, we noted higher serum levels of threonine in healthy women as compared with NAFLD subjects (AUROC=0.701), (Figure 5). MSEA has prompt to a disturbed Threonine and 2-Oxobutanoate Degradation pathway among NAFLD female subjects (Figure 7B). It is reported that supplements of threonine for mice in a high-fat diet alleviated the onset of fat deposition (59). Of note, a community-based case-control study including senior Chinese aged over 65 years has showed that threonine, valine and lysine were negatively correlated with NAFLD risk and highlighted the potential beneficial effect of the increased consumption of food rich in those essential amino acids as eggs, milk and deep-sea fish (60).

The metabolic markers were used as a query in the Metabolite-Disease Interaction Network, in an effort to uncover potential associations of NAFLD with other medical conditions (Supplementary Figures S4A, B).

The obtained results in male subjects point towards a link between NAFLF and risk for lung cancer. This finding is consistent with recent research evidence stating that obesity and NAFLD might consist a risk factor for lung adenocarcinoma although the underlying causality is under investigation (61, 62). Furthermore, a large cohort prospective study in China demonstrated that men with NAFLD have a higher risk of developing all cancers in particular thyroid and lung cancer and revealed also a positive association between the latter and smoking (63).

The results from the Metabolite-Disease Interaction Network also indicated a potential relation between NAFLD and schizophrenia in both men and women. Interestingly, this link was established based on different metabolites sets in each sex. NAFLD/NASH is highly comorbid with mental illness and research efforts attempt to establish the underlying linking mechanisms at genetic, epigenetic, metabolic and inflammatory level, even though the exposure to antipsychotic drugs and unhealthy lifestyle behavior (64–66). In our study, only two participants were under antidepressant medication according to their statement thus antipsychotic drugs influence is unlikely. Perturbation of common metabolic pathways interrelating schizophrenia and NAFLD merits further investigation through longitudinal studies. A recent study overcame the potential bias due to the metabolic side effects of antipsychotic drugs by addressing first-episode psychosis (FEP) with the inclusion of drug-naïve patients. The study revealed the greater risk of FEP patients to present at least one altered MetS component as compared to control subjects (67).

Moreover, on females with NAFLD the metabolites alterations were linked with the occurrence of Alzheimer’s disease (AD). There is increasing evidence that NAFLD could be associated with moderate cerebral dysfunction and cognitive decline but again more longitudinal studies are needed to confirm if this association is linked to comorbidities as CVD and T2D or hepatic dysfunction and induced neuroinflammation and neurodegeneration (68–70).

Finally, disorders related to severe damage to the liver as tyrosinemia type I (71, 72) and pyridoxamine 5-phosphate oxidase (PNPO) deficiency (73–75), were also suggested for the female subjects with NAFLD. These uncommon metabolic disorders affect a different age range from our sample pool but their emergence in part explains why NAFLD is often associated with aspects of MetS.

The potential link of liver diseases with other medical conditions seems to be supported by multiple studies and should be considered in clinical practice.