Mechanistically linked serum miRNAs distinguish between drug induced and fatty liver disease of different grades

Hepatic steatosis is characterised by excessive triglyceride accumulation in the form of lipid droplets (LD); however, mechanisms differ in drug induced (DIS) and/or non-alcoholic fatty liver disease (NAFLD). Here we hypothesized distinct molecular circuits of microRNA/LD-associated target genes and searched for mechanistically linked serum and tissue biomarkers that would distinguish between DIS and human NAFLD of different grades. We analysed >800 rat hepatic whole genome data for 17 steatotic drugs and identified 157 distinct miRNAs targeting 77 DIS regulated genes. Subsequently, genomic data of N = 105 cases of human NAFLD and N = 32 healthy controls were compared to serum miRNA profiles of N = 167 NAFLD patients. This revealed N = 195 tissue-specific miRNAs being mechanistically linked to LD-coding genes and 24 and 9 miRNAs were commonly regulated in serum and tissue of advanced and mild NAFLD, respectively. The NASH serum regulated miRNAs informed on hepatic inflammation, adipocytokine and insulin signalling, ER-and caveolae associated activities and altered glycerolipid metabolism. Conversely, serum miRNAs associated with blunt steatosis specifically highlighted activity of FOXO1&HNF4α on CPT2, the lipid droplet and ER-lipid-raft associated PLIN3 and Erlin1. Altogether, serum miRNAs informed on the molecular pathophysiology of NAFLD and permitted differentiation between DIS and NAFLD of different grades.

Remarkably, the frequency of DIS is unknown and estimates of drug induced hepatic injury range considerably amongst different studies and countries with a crude incidence rate of 19.1/100,000 as was reported for the general population of Iceland 5 . Given that NAFLD has become an epidemic it would be highly desirable to have sensitive and specific serum markers for the rapid and cost effective screening of individuals at risk for NAFLD and/or DIS and to determine disease progression by minimal invasive or non-invasive means. Accordingly, therapeutic intervention strategies can be developed to prevent disease progression by actively screening patients with fatty liver disease. Although several risk factors for NAFLD have been identified and typically include type-2 diabetes & impaired glucose tolerance, clinical obesity, dyslipidaemia and eventually the metabolic syndrome, prolonged exposure of drugs can also be associated with DIS and NASH as reported for amidarone and other drugs 6 .
Importantly, drug induced hepatic steatosis is generally associated with more than one mechanism 7 and either involves micro-or macrovesicular steatosis (e.g. valproic acid) and can be associated with ballooning degeneration of hepatocytes. Particularly at prolonged exposure to drugs cellular stress, as a result of impaired detoxification, and mitochondrial function becomes compromised thereby aggravating the condition. The main mechanism by which drugs cause steatosis were recently summarized and involve (a) inhibition of mitochondrial fatty acid beta oxidation (b) sequestration of CoA and/or L-Carnitine (c) inhibition of the mitochondrial respiration and modulation of the mitochondrial membrane potential (d) impairment of mitochondrial DNA replication (e) impaired peroxisome proliferator activated receptor (PPARα ) transcriptional activity and other alterations in lipid homeostasis pathways 8 .
Due to the growing knowledge on the mechanisms of lipid droplet formation in hepatocytes and the possibilities to study mechanistically linked biomarkers we were particularly interested in studying molecular circuits and transcriptional responses of microRNA target genes in rat liver induced by steatotic drugs. The importance of non-coding RNAs (ncRNAs) in regulating lipid metabolism has just been unraveled with microRNAs (miRNAs) constituting a large family of small, approximately 21-nucleotide-long, non-coding RNAs that have emerged as key post-transcriptional regulators of gene expression 9 . By base pairing to mRNAs, miRNAs mediate translational repression or mRNA degradation 10 . In mammals, miRNAs are predicted to control the activity of approximately 30% of all protein-coding genes, and have been shown to participate in the regulation of almost every cellular process investigated so far 10 . Recent research highlighted the critical functions of miRNAs in regulating diverse biological processes including the immune system, cell proliferation, differentiation and development, cancer and cell cycle [11][12][13] and several miRNAs function in a tissue-specific manner 14,15 as determined for NAFLD 16 .
Altogether, the primary objective of our study was to define mechanistically linked serum miRNAs that would distinguish between DIS and NAFLD of different grades. We hypothesized distinct transcriptional regulations of co-expressed miRNA/target genes in rat liver induced by steatotic drugs and utilized our knowledge of about 200 genes being mechanistically involved in various pathophysiological processes during fatty liver disease as recently reported 17 . By interrogating the publically available Open TG-GATEs database the dose and time dependent effects of 17 DIS causing drugs could be investigated and by studying whole genome transcript expression data of human NAFLD co-expressed miRNA-target genes in DIS and NAFLD were ascertained. Importantly, the identification of mechanistically linked miRNA serum biomarkers for DIS and NAFLD of different grades provide opportunities for its early detection and the monitoring of disease progression and will aid drug selection in preclinical development programs. Data processing. A total of 808 Affymetrix GeneChip Rat Genome 230 2.0 raw data (CEL files) of 17 steatotic drugs and matched controls were retrieved from TG-GATE (http://toxico.nibio.go.jp/english/index.html). The data were analyzed using the robust multi-array average (RMA) methodology in the Bioconductor, R package for background-adjusted, normalized, and log-transformed perfect matched values of individual probes from the Array. (http://www.bioconductor.org/packages/release/bioc/html/BufferedMatrixMethods.html).
The fold change for each compound was calculated by comparing treatment data and matched controls; the student t test with multiple testing corrections was employed to calculate p values for each gene. Differential expressed genes (DEGs) were considered for further analysis by using the criterion fold change more than 1.5 and Benjamini & Hochberg adjusted p value less than 0.05.
Genes involved in lipid droplet formation of hepatocytes. Based on our previous works we considered approximately 200 steatosis related genes 17 which code for lipid transport & lipogenesis, lipid droplet associated proteins such as the perilipins, glucose & fatty acid metabolism and signalling events. Details are given in supplementary Table S1 and the gene symbols were mapped to Entrez Gene ID based on NCBI gene annotation (http://www.ncbi.nlm.nih.gov/gene/).

MiRNA-gene target predictions.
MiRNA targets were predicted using the miRTarBase repository http:// mirtar.mbc.nctu.edu.tw/human/index.php) 19 . The database defines experimentally proven miRNA-gene target relationships. DEGs for 17 steatotic drugs (see above) and data from N = 4 independent publically available genomic data sets of N = 105 cases of human liver NAFLD and N = 32 controls (see below "translational research") were mapped to miRTarBase. A summary of the miRNA-gene relationships is given in supplementary Table S2  with mild NAFLD, fibrosis stage 0-1 and N = 32 with advanced NAFLD, fibrosis stage 3-4 were retrieved from the GEO database (GSE49541). Importantly, the study of Moylan et al. 23 (GEO data set GSE49541) did not involve  healthy control cases 23 . Therefore, DEGs were calculated by considering whole genome hepatic gene expression data of N = 7 healthy controls regained from the López-Vicario et al. study 24 . In their study (GEO data set GSE37031) NASH patients and healthy controls were analyzed with the same microarray platform as utilized in the study of Moylan et al. 23 . In order to eliminate a batch effect, the CEL files from the two studies were processed together using the robust multi-array average (RMA) methodology and DEGs for mild NAFLD and advanced NAFLD were generated with the R limma package. Altogether, genomic data sets of N = 105 cases of human liver NAFLD and N = 32 controls were considered and DEGs were generated with the GEO2R (http://www.ncbi.nlm.nih.gov/geo/geo2r/) software with the filtering criteria fold change > 1.5 and Benjamini & Hochberg adjusted p value < 0.05. Tissue regulated miRNA were predicted from whole genome hepatic gene expression changes by querying the miRTarBase repository (see above). Additionally, the whole genome DEG data were filtered for N = 200 genes mechanistically involved in lipid droplet formation of hepatocytes (supplementary Table S1 and [18]) and both approaches were subsequently mapped to miRTarBase to identify associated miRNAs. Moreover, prediction of gene targets was based on miRNA consensus among 9 out of 12 different algorithms (see above).
The tissue regulated miRNA were validated by considering experimentally verified serum miRNA profiles obtained from N = 12 healthy controls and N = 167 NAFLD patients samples. Note, for all patients liver biopsies were obtained to permit an assessment of grades of fatty liver disease and included N = 66 cases of benign steatosis and N = 94 NASH cases of different grades 22 . A further independent data set of serum microRNA expression in human NAFLD (GEO Accession: GSE33857) encompassing N = 12 controls with normal liver function and N = 7 NASH cases was considered 25 .

Results
The study design and work flow is depicted in Fig. 1 and statistically significant differentially expressed genes (DEGs) were identified in liver tissue of rats after single and repeated treatment with 17 steatotic drugs for up to 28 days. Subsequently, the data were filtered for genes known to be mechanistically linked to lipid droplet formation in hepatocytes (see supplementary Table S1) and miRNAs targeting DIS regulated genes were obtained by interrogating the experimentally verified miRTar database. Thus, the data analysis focused on (1) an identification of miRNAs targeting LD associated genes in hepatic steatosis; (2) time/dose effects in drug induced steatosis (3) an identification of commonly regulated tissue and serum miRNAs for its utility as serum biomarker and (4) serum miRNAs to distinguish DIS and human NAFLD of different grades.
Prediction of tissue regulated miRNA in drug induced steatosis. Computational analysis of 17 steatotic drugs involved > 800 whole genome microarrays and the subsequent filtering of DEGs for 200 hepatic steatosis related genes (supplementary Table S2). This revealed 409 miRNA-gene associations and consisted of 157 distinct miRNAs and 77 genes. The mapping of significant DEGs to different processes in lipid droplet formation (Fig. 2) discovered > 20 genes to be regulated in the process of lipogenesis while KEGG pathways analysis highlighted significant enrichment for glycerol lipid and glycerophospholipid metabolism (Table 3) as exemplified by the significant regulation of DGAT1, i.e. a key enzyme catalyzing the terminal step in triacylglycerol synthesis as well as AGPAT isoforms involved in the metabolism of lysophosphatidic acid to phosphatidic acid (PA). The latter are important signalling molecules but also function as building block intermediates in lipid biosynthesis. Other  Table S3 and are mechanistically linked to fatty liver disease.
highly regulated genes include fatty acid synthase as well as lecithin cholesterol acyltransferase, the desaturase FADS1, FADS2, the transcription factors PPARγ , SREBF1 and the SREBP cleavage protein SCAP, the glucosylceramide synthase and various annexins to influence Ca2+ dependent phospholipid binding to modify cytoskeletal dynamics.
The frequency by which individual miRNAs target steatosis related genes was considered; Fig. 3a depicts hsa-mir-335-5p and has-miR-124-3p as top ranking. These are involved in the regulation of genes coding for lipogenesis (n = 6/3), signalling events (n = 3/2), lipid transport (n = 1/3), lipid droplet associated proteins (n = 1/5), fatty acid oxidation (n = 4/1) and glucose metabolism (1/2). The network of LD-associated gene regulations among the top 10 ranking miRNAs is given in Fig. 3b and Table 4 summarizes independent studies to confirm the top ranking miRNAs as significantly regulated in animal and human NAFLD.
Moreover, the drug induced dose and time dependent regulations were considered and as shown in Fig. 4 the higher dose majorly affected repression rather than induction of gene transcription after single and repeated drug treatment. In an effort to define commonly regulated genes (frequency > 10) the different drug, dose and time constellations were considered ( Fig. 5a-d) and compared by constructing a Venn diagram to obtain information    Continued on the overlap of genes among the various considerations (Fig. 5e). This defined the aryl hydrocarbon receptor (AhR) as commonly up-regulated after single and repeat dosing with 17 different drugs while the genes FADS1, PKLR, ACLY, GK, SREBF1, FADS2, FABP7, and ELOVL5 were commonly repressed in expression. Note, the molecular functions of the aryl hydrocarbon receptor and the liver X receptor in the regulation of metabolic homeostasis is the subject of intense research; evidence was obtained for the aryl hydrocarbon receptor to induce hepatic steatosis via the up-regulation of the fatty acid transporter CD36 18,26 . Additional miRNAs involved in the regulation of the AhR include has-mir-26a-5p, hsa-mir-130b-3p, has-mir-124-3p, has-miR-625-5p and has-miR-98-5p with proven experimental evidence for their participation in the regulation of genes coding for lipid transport most notable CD36, fatty acid binding proteins FABP1, FAB6, FAB7, low density lipoprotein receptor, RXRß and others based on miRTarBase data analysis and PubMed searches.
As depicted in Fig. 5e eight genes were commonly down regulated after single and repeated treatment of rats with 17 steatotic drugs and included FADS1, PKLR, ACLY, GK, SREBF1, FADS2, FABP7 and ELOVL5.
We also considered the chemical structure similarity amongst 17 steatotic drugs by calculating the Tanimoto coefficient and found it to be insignificant. Nonetheless, when the frequency of DIS associated miRNAs were considered hsa-miR-335 was most frequently regulated across a wide range of dissimilar drugs.
Lastly, when miRNAs obtained from consensus predictions were compared with results obtained from miR-TarBase queries a total of 35 miRNAs were in common (supplementary Fig. S1).

Molecular circuits of co-expressed miRNA-gene targets in DIS.
Based on the built miRNA-steatosis and drug-steatosis gene interactions, we considered drug, dose and time associations (supplementary Table S3) and subsequently constructed protein-protein-interaction networks by probing the STRING PPI database version 9.1. As a result, 128 and 162 PPIs corresponding to 48 and 50 genes were obtained after single and repeated drug treatment of rats with 17 steatotic drugs (supplementary Table S4). Furthermore, by utilizing the MCODE algorithm the endoplasmic reticulum localized fatty acid desaturases FADS1, FADS2 and the fatty acid elongases ELOVL2 and ELOVL5 were identified as commonly regulated after single and repeat drug treatment. Consistent with previous findings expression of these genes is repressed and confer functional diversity through changes in chain length and degree of unsaturation in the process of lipogenesis.
Alike, individual networks for top regulated miRNAs were considered and as an example Fig. 6 depicts hsa-miR-124-3p and hsa-miR-335 regulated DIS genes after single and repeated treatment of rats with 17 drugs for up to 28 days. Transcriptional regulation involved 12, 10, 9 and 11 repressed genes after single and repeated treatments, respectively and the LDL receptor was commonly targeted amongst these miRNAs irrespective of the treatment conditions. Note, up-regulation of liver tissue specific hsa-miR-335 was reported for murine models of NAFLD, e.g. the leptin deficient ob/ob, the leptin-receptor-deficient db/db and KKAy44 mice. It is of considerable importance that hsa-miR-335 is a major regulator for lipid accumulation and adipose tissue differentiation in mice 27 and the transcription factors FOXO1 and FOXA2, the insulin induced gene 2 and the low density lipoprotein receptor LDLR were equally regulated by miR-335. For its frequent regulation by different drugs and its important role in lipid homeostasis has-miR-335 should be considered as a DIS biomarker in drug safety evaluation. Similarly, among the miR-124 target genes was the Rho-associated, coiled-coil containing protein kinase 2. Recent evidence suggests repression of ROCK proteins to prevent hepatic steatosis by reducing lipid synthesis while pharmacological inhibition of ROCK prevented severe ischemia/reperfusion injury in steatotic fatty liver allografts 28 . The observed repression of ROCK transcripts can therefore be considered as an adaptive response to drug induced steatosis. Fig. 1 N = 4 independent studies were retrieved from the GEO database. Next to determining differentially expressed genes (DEGs) the data were mapped to miRTarBase and the results were compared with predicted DIS regulated miRNAs for 17 drugs after single and repeated treatment of rats for up to 28 days. Among 157 DIS regulated miRNAs 147 were regulated in common with human NAFLD (Fig. 7a). Alike, when human NAFLD data were filtered for 200 genes mechanistically linked to hepatic steatosis (Fig. 7b) 92 DIS regulated miRNAs were commonly regulated; vice versa 10 (see Fig. 7a) or 65 (see Fig. 7b) miRNA were predicted to be DIS tissue specific. Altogether, a total of 195 or 7 NASH specific miRNAs were identified by either considering whole genome human liver tissue DEGs or by filtering DEGs for 200 mechanistically linked genes. This analysis also revealed the lack of blunt steatosis specific tissue regulated miRNAs.

Translational research. As depicted in
Serum biomarker candidates. Based on whole genome gene expression profiling and by considering commonalities amongst independent publically available studies NASH and DIS specific miRNAs were analyzed. Their regulation was compared to experimentally verified serum miRNA profiling studies and as shown in Fig. 8 a total of 37 and 9 miRNAs were significantly changed in serum miRNA profiling studies of NASH and cases of blunt steatosis, respectively (Fig. 8 upper panel). Subsequently, DIS tissue specific miRNAs were compared  to blunt steatosis regulated serum miRNAs. This revealed none of the miRNA to be in common thus evidencing specificity of the performed analysis (Fig. 8a,b). Additionally, miRNAs commonly regulated in liver tissue amongst 4 independent NASH studies were compared with DIS tissue and serum regulated miRNAs. Such an analysis discovered 28 and 32 NASH serum specific miRNAs by either considering total DEGs (approach A) or DEGs filtered for 200 mechanistically linked hepatic steatosis genes (approach B). The Venn diagram shown in the lower panel of Fig. 8 informs on 24 common miRNAs to demonstrate that the hypothesis driven approach (see panel b) yielded 65% of experimentally verified serum regulated miRNAs in NAFLD. Our study highlights 8 common miRNAs when independent NASH tissue and serum profiling studies are compared and includes the miRNAs: hsa-let-7c-5p, hsa-let-7d-5p, hsa-let-7g-5p, hsa-miR-326, hsa-miR-150-5p, has-miR-885-5p, has-miR-486-5p and has-miR-126-5p. The network of regulated genes by these miRNAs is depicted in Fig. 9 and the results given in Fig. 10 evidence differentiation between NAFLD of different grades. Specifically, miRNAs regulated in blunt steatosis were analysed for target genes by mapping them to miRTar-Base; the resultant target genes were compared to DEGs determined in hepatic tissue of NAFLD patients and to the list of 200 genes mechanistically involved in fatty liver disease (supplementary Table S2). Accordingly, a network was constructed and visualised with Cytoscape and 4 miRNAs were found to influence transcriptional expression of 22 steatosis related genes. This included the lipid droplet associated protein PLIN3, the transcription factors FOXO1, HNF4a and PPARγ , the ER lipid raft associated ERLIN1 and 2, the lysophosphatidic acid acyltransferase, protein and mitogen activated kinases and SNAP23 involved in the lipid fusion process. Using the same approach NASH regulated serum miRNAs were analysed and a total of 7 miRNAs were identified to target 41 fatty liver disease regulated genes. Here, significant enrichment for adipocytokine (i.e. AKT1, IRS2, ACSL1, RXRB, PRKAG1, SLC2A1, PRKAB1, MAPK9, PRKAA2, IRS1, STAT3) and insulin signalling pathways (SREBF1, AKT1, MAPK1, IRS2, SOCS2, PRKAG1, SOCS1, FLOT1, PRKAB1, MAPK9, PRKAA2, IRS1), regulation of suppressor of cytokine (SOCS1 & SOCS2) and Toll-like receptor 4 signalling were important findings. Further notable findings were miRNAs involved in the regulation of the LDL receptor, various transcription factors (RXRB, STAT3), calveole and fusion associated proteins CAV1, SNAP23, Flotilin, the monoglyceride lipase 1 and the microtubule-activated ATPase dynein 1 which functions in vesicle transport. Additionally, a total of 17 genes were regulated in common when serum miRNAs regulated in blunt steatosis and/or NASH were compared (Fig. 10 lower panel). Taken collectively, circulating miRNAs provide opportunities for disease diagnostic that may be extended to drug therapeutic monitoring. Table 4 compiles independent experimental evidence for the top ten DIS regulated miRNAs in serum or tissue of animal and human NAFLD. For instance, the anti-fibrotic miR-335-5p as well as miR-30a-5p are known to repress lipid biosynthesis, lipoprotein secretion and autophagasome formation and were significantly regulated as was mir-21 and mir-335 which are mechanistically linked to hepatic triglyceride and cholesterol metabolism.

Discussion
A recent white paper by the American Association for the Study of Liver Diseases, the American College of Gastroenterology and the American Gastroenterological Association provided guidance for the diagnosis and the management of NAFLD 29 . In this report the current limitations in the diagnosis of NAFLD are summarized with serum biochemistries frequently failing to be informative as they can be within normal range or are insufficiently sensitive to reflect grades of disease; alike liver ultrasound imaging requires skilled healthcare professionals and might be obscured by gross morphologic changes other than NAFLD. Furthermore, ultrasound imaging does not inform on some important diagnostic and prognostic factors in NAFLD. Therefore, the diagnosis and assessment of progressive fatty liver disease remains challenging; nonetheless the discoveries of molecular disease markers carry the hope to replace liver biopsies by novel blood and/or urinary tests.
The significant role of miRNAs in regulating lipid metabolism was the subject of several reviews 30 ; however, only a few studies investigated the regulation of miRNAs in drug induced steatosis 31 . As of today it remains uncertain whether miRNAs known to regulate cholesterol and fatty acid metabolism and lipid droplet formation are the same in DIS and NAFLD. In the present study whole genome transcript expression changes were examined to investigate tissue regulated miRNAs and to permit construction of miRNA-gene networks specific for fatty liver disease and DIS. Based on diverse computational approaches distinct circuits of co-expressed microRNA target genes in NAFLD and DIS were postulated and validated by comparing tissue with serum regulated miRNAs. This enabled an identification of mechanistically linked serum biomarker candidates for DIS and NAFLD of different grades. Table 5 compiles a summary of regulated miRNAs in human NAFLD (Table 5A) and in animal models of NAFLD (Table 5B).
The present study highlights the importance of mechanistically linked lipid droplet associated gene regulations to inform on the pathophysiological state of fatty liver disease with serum miRNA profiling representing a rapid and cost effective means for an identification of diagnostic biomarkers.
Importantly, fatty liver disease is a clinico-pathologic feature of drug induced liver injury (DILI) and is typically seen with drugs such as amiodarone, perhexiline, carbamazepine amongst others. It results from an imbalance between xenobiotic defense and adaptation leading to ER/mitochondrial stress, unfolded protein response and liver injury. Recently, whole genome transcript profiling of drug induced steatosis was reported and comprehensive information on mechanistically linked gene expression changes was obtained for a wide range of drugs causing either steatosis and/or phospholipidosis while an identification of signature genes encourage their use in preclinical screening assays 18 .   supplementary Table S1) and mapped to the miRTar database. The data were compared with DIS associated miRNA regulations. Unfortunately, commonly used markers of liver injury lack specificity and do not distinguish between DIS and NAFLD of different grades but rely primarily on serum biochemistries, for instance by assaying serum alanine and aspartate aminotransferase activities as a means to examine the metabolic competence of the liver. In recent years molecular biomarkers of hepatotoxicity and their potential to improve understanding and management of DILI have been intensely researched as summarized in a recent research highlight 32,33 .
To overcome current limitations mechanistically based biomarkers have been advocated and await translational research to better predict prognosis and outcome than currently used clinical biochemistry parameters 34 . Serum miRNAs may therefore be employed to predict individuals at risk for developing fatty liver disease and in our previous research we identified serum acute phase reactants to define healthy individuals at risk for acetaminophen induced liver injury prior to drug treatment 35 , while the study of Yamaura and co-workers 31 identified miRNAs indicative of the pattern of liver injury that is steatosis versus NASH, cholestatic disease and acute versus chronic hepatocellular injury using a range of drugs and dietary animal models. Indeed, when the data of the Yamaura's study were compared with findings of the present study several miRNAs were regulated in common  Fig. 7b (i.e. the whole genome data was filtered for N = 200 mechanistically linked LD-associated gene regulations (see supplementary Table S1). Lower panel: Comparison of NASH and DIS tissue miRNAs with NASH serum regulated miRNAs. and included for blunt steatosis miR-10b and miR-183; similarly with NASH the miRNAs miR-17, miR148b-5p and miR-197 were commonly regulated thus providing independent evidence for their diagnostic utility in animal studies.
In the present study diverse genomic data sets were examined to define regulatory miRNA gene networks in DIS and NAFLD of different grades and unique miRNAs targeting LD-associated genes were identified to provide mechanistic information on the onset and progression (micro-versus macrovesicular steatosis) of fatty liver disease. The network analysis revealed 29 out of 56 DIS genes to be commonly regulated by the top 10 regulated miRNAs (Fig. 3a,b) and > 70% of all DIS regulated genes (56 over 77 DIS genes) are targeted by these miRNAs. Apart from common DIS regulated genes the transcription factors FOXO1 and FOXA2 and the insulin induced gene 2 are specific targets of miR-335-5p; note polymorphisms of INSIG2 are associated with severe obesity. Furthermore, the patatin-like phospholipase domain-containing protein 3 is a target of miR-335-5p and a single nucleotide polymorphism of this gene is strongly associated with NAFLD of different grades 36 . Other genes regulates by this miRNA are the UDP-Glucose Ceramide Glucosyltransferase (UCGC) and acyl-CoA synthetase long-chain family member 3. Besides, regulation of syntaxin 5 by miR-155-5p is of critical importance. This protein facilitates vesicle docking and its fusion and is part of the SNARE complex. Alike, miR-26b-5p influences translation of caveolin 2 and diacylglycerol O-acyltransferase 1, i.e. a key enzyme in hepatic steatosis in addition to lecithin-cholesterol acyltransferase which catalyzes its esterification for cholesterol transport. For its role in the regulation of the Ah-receptor miR-124-3p is of great importance. Specifically, AhR sustained activation results in induction of fatty acid transport proteins and enhanced lipid uptake via activity of the fatty acid transporter CD36 to eventually cause accumulation of triglycerides leading to hepatic steatosis. Likewise, it is of great significance that AhR suppresses fatty acid oxidation and export of triglycerides from the liver by hampering VLDL synthesis and secretion. Lastly, the AhR engages in a vicious cycle whereby serum triacylglycerides levels are increased through peripheral fat mobilization thus perpetuating hepatic lipid uptake from the systemic circulation 37 .
Regulation of 1-acylglycerol-3-phosphate O-acyltransferase 2 by miR-744-5p is also of key importance. This ER localized protein catalyzes the conversion of lysophosphatidic acid to phosphatidic acid, e.g. a building block in phospholipid synthesis and essential second messenger that mediates cellular functions through different modes of action. Noteworthy is also the regulation of the phosphatase and tensin homolog PTEN by miR21-5p and recent research identified Maf1 as a novel target of PTEN and PI3K signalling to negatively regulate carcinogenesis and lipid metabolism by repressing intracellular lipid accumulation and de novo lipogenesis 38 . Alike, this miRNA targets fatty acid synthase and the enzyme catalyze the synthesis of long-chain saturated fatty acids. Among the top 10 miRNAs is let-7b-5p known to influence translation of insulin receptor substrate 2 transcripts and this protein functions as an adaptor to link receptor tyrosine kinases with downstream effectors. Dynein is also regulated by this miRNA and this microtubule-activated ATPase is of uttermost importance in intracellular motility and trafficking of vesicles. A further example is the regulation of the liver receptor homolog 1 (NR5A2) by miR-1 and this nuclear receptor functions as a transcription factor and phospholipid binding protein and plays an essential role in hepatic metabolism. Repression of small heterodimer partner by NR5A2 affects VLDL synthesis and secretion and is deregulated in NAFLD as summarized in the seminal review of Bechmann and colleagues 39 . Lastly, the gene coding for the ER lipid raft associated protein 1 is a target of miR30a-5p and is part of the ERLIN1/ERLIN2 complex to assist ER-associated degradation of inositol 1, 4, 5-trisphosphate receptors (IP3Rs); note, alterations in ER lipid rafts have been shown to contribute to lipotoxicity 40 .
The chosen examples illustrate the benefit in construction molecular circuits based on co-expressed miRNA-gene targets and the knowledge gain by considering mechanistic information regarding common and exclusively regulated genes in DIS and NAFLD. In an effort to define commonalities amongst steatotic drugs four mechanistically linked genes (ELOVL2, ELOVL5, FADS1 and FADS2) were identified as consistently regulated  Continued after single and repeated treatments with N = 17 steatotic drugs with FADS1 and FADS2 catalyzing Δ5/Δ6 desaturation of fatty acids to influence arachidonic acid derived eicosanoid, prostaglandins and leukotriene production. The significant regulation of FADS1 and FADS2 will influence arachidonic acid homeostasis and associated herewith immune cell populations with a significant reduction in CD4+ /CD3+ T-cell and mononuclear cells as observed in FADS1 null mice and in murine models of NASH 41 . The observed repression of FADS1 and 2 may therefore be consider as an adaption to drug treatment to prevent overt toxicity associated with DIS induced inflammation. While deficiency in unsaturated fatty acids impacts lipid and energy metabolism, membrane structures and signalling pathways, variations in the expression of these enzymes will help to reduce production of inflammatory lipids 42 . Likewise, repression of the fatty acid elongases ELOV2 and 5 denotes coordinate responses in polyunsaturated fatty acid synthesis. Note, overexpression of ELOVL2 enhanced triacylglycerol synthesis in 3T3-L1 and F442A cells and caused induction of the lipogenic genes diacylglycerol acyltransferase-2 and fatty acid-binding protein-4 to support triacylglycerol synthesis and subsequent accumulation of lipid droplets 43 .
Modulation of the synthesis of very long-chain fatty acids by ELOV2 and 5 and subsequent desaturations by the Δ5/Δ6-desaturases FADS1 and 2 signifies the relevance of the identified DIS associated genes and assaying mechanistically relevant DIS genomic biomarkers can be easily adapted at the preclinical stage of drug development by merely comparing the hepatic gene expression signature of an experimental drug to DIS and DILI signatures along with appropriate controls. Testimony to the findings of the present study is also a recently published comprehensive lipidomics approach for an identification of classifier lipid metabolites in urine and plasma and transcriptomic data as identified by RNA-sequencing of liver tissue from a total of N = 31 healthy controls, N = 17 steatotic and N = 20 NASH patients. The transcriptomic data correlated with the clinical stage of disease and signatures of lipids were identified that can be employed as biomarkers for disease staging 44 .
In an effort to define mechanistically relevant serum biomarkers research of the present study identified 24 serum miRNAs as commonly regulated in NASH (Fig. 8). These miRNAs were examined for their role in targeting N = 200 lipid droplet associated genes (supplementary Table S1); a network was constructed whereby 7 of the 24 common regulated miRNAs were involved in the translational control of 41 fatty liver disease specific gene regulations (Fig. 10b). Once again the findings highlight the possibilities by which specific miRNAs instruct pathophysiological processes in fatty liver disease with miR-7-5p targeting suppressor of cytokine signalling SOCs and Downregulation of IRP1 was linked to an increased expression of microRNAs miR-200a and miR-223, which was negatively correlated with IRP1. The results of this study demonstrate that the interstrain variability in the extent of fibrogenesis was associated with a strain-dependent deregulation of hepatic iron homeostasis.
106 Table 5. miRNAs in human (A) and animal models (B) of NAFLD of different grades.
Scientific RepoRts | 6:23709 | DOI: 10.1038/srep23709 calveolin-1, i.e. an essential component of membrane lipid rafts with complex functions in cell signalling while the subnetwork of let-7b-5p involves translational control of the LDL receptor, hepatic lipoprotein lipase and SNAP23 that forms a complex with other vesicle-associated membrane protein and is part of the fusion machinery in macro-vesicular steatosis. Further evidence for a coordinate responses in fatty liver disease was obtained by considering regulation of the ER lipid raft associated 2, PPARγ and insulin induced gene 1 by miR-192-5p. The coded proteins are critically involved in lipid and cholesterol homeostasis while inflammatory signalling of the Toll-like receptor 4 and STAT3 pathways was influenced by let-7i-5p and miR-125b-5p, respectively. These and other examples mentioned above clearly demonstrate the benefit of considering mechanistically related miRNA gene target associations in obtaining pathophysiological relevant information by assaying specific serum miR-NAs. Indeed, ncRNAs function precisely to repress targets mRNAs and as a result influence expression of proteins and cellular phenotypes 45 . Although a number of miRNAs are widely expressed amongst diverse tissues, certain miRNAs appear to be organ specific. In addition, some miRNAs are conserved across species and hold promise for translational research. To date, over 1800 mature human miRNAs have been identified and recorded in the miRBase registry 46 . Lastly, the following caveats need to be considered. The study findings are based on the miRTar resource but were not obtained within the same liver tissue used for the DIS study. Nonetheless, the evidence within the miRTar repository is based on independent experimental research and for its high evolutionary conservation the results obtained can be exploited to search for transcriptional regulations of co-expressed microRNA target genes. Specifically, only miRNAs found to be expressed in liver tissue were considered and the research performed focused on the discovery of miRNA targeting genes mechanistically involved in lipid droplet formation and hepatic steatosis. Future research will involve the construction of feed-forward loops (FFLs) to decipher regulatory gene networks as we had demonstrated for the repurposing of drugs in the treatment of cystic fibrosis 47 .

Conclusions
New insight into co-expressed microRNA target genes in DIS and NAFLD was obtained. An identification of mechanistically linked safety biomarkers will aid drug development and contribute to an understanding of risk for drug induced steatosis and to distinguish amongst other causes of fatty liver disease. Our findings encourage independent validation of the proposed miRNA DIS and NAFLD biomarkers. J Borlak receives funding from the German Federal Ministry of Education and Research as part of the Virtual Liver Network initiative (grant number 031 6154). He is also a recipient of an ORISE stipend of the FDA which is gratefully acknowledged. The views presented in this article do not necessarily reflect current or future opinion or policy of the US Food and Drug Administration. Any mention of commercial products is for clarification and not intended as endorsement.