- Email: [email protected]
Relationship Between Methylome and Transcriptome in Patients With Nonalcoholic Fatty Liver Disease
Accepted Manuscript Relationship Between the Methylome and Transcriptome in Patients with NonAlcoholic Fatty Liver Disease Susan K. Murphy, Hyuna Yang, Cynthia A. Moylan, Herbert Pang, Andrew Dellinger, Manal F. Abdelmalek, Melanie E. Garrett, Allison Ashley-Koch, Ayako Suzuki, Hans L. Tillmann, Michael A. Hauser, Anna Mae Diehl PII: DOI: Reference:
S0016-5085(13)01137-2 10.1053/j.gastro.2013.07.047 YGAST 58588
To appear in: Gastroenterology Accepted Date: 26 July 2013 Please cite this article as: Murphy SK, Yang H, Moylan CA, Pang H, Dellinger A, Abdelmalek MF, Garrett ME, Ashley-Koch A, Suzuki A, Tillmann HL, Hauser MA, Diehl AM, Relationship Between the Methylome and Transcriptome in Patients with Non-Alcoholic Fatty Liver Disease, Gastroenterology (2013), doi: 10.1053/j.gastro.2013.07.047. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. All studies published in Gastroenterology are embargoed until 3PM ET of the day they are published as corrected proofs on-line. Studies cannot be publicized as accepted manuscripts or uncorrected proofs.
ACCEPTED MANUSCRIPT
Relationship Between the Methylome and Transcriptome in Patients with Non-Alcoholic Fatty Liver Disease (Functional methylation in NAFLD) Susan K. Murphy1, Hyuna Yang2, Cynthia A. Moylan3, Herbert Pang2, Andrew Dellinger , Manal F. Abdelmalek3, Melanie E. Garrett4, Allison Ashley-Koch4, Ayako Suzuki3, Hans L. Tillmann5, Michael A. Hauser4 and Anna Mae Diehl3
RI PT
2
SC
1 Department of Obstetrics and Gynecology, Duke University, Durham, NC 2 Department of Bioinformatics and Biostatistics, Duke University, Durham, NC 3 Division of Gastroenterology and Hepatology, Duke University, Durham, NC 4 Center for Human Genetics, Duke University, Durham, NC 5 Section of Medical Genetics, Department of Medicine, Duke University, Durham, NC
Grant Support: National Institutes of Health 5RC2-AA019399
TE D
M AN U
Abbreviations: ANOVA, Analysis of Variance; bp, base pairs; BMI, body mass index; CGI, CpG islands; CASP1, caspase 1; DM, differentially methylated; DUHS, Duke University Health System; ENCODE, Encylopedia of DNA elements; FGFR2, fibroblast growth factor receptor 2; kb, kilobase; gene expression, (GEx); HbA1, hemoglobin A1c; IPA, Ingenuity Pathway Analysis; M, methylation intensities; MAT1A, methionine adenosyl methyltransferase 1A; NAFLD, nonalcoholic fatty liver disease; NAS, NAFLD activity score; NASH, nonalcoholic steatohepatitis; NCBI, National Center for Biotechnology Information; SD, standard deviation; SNP, Simple Nucleotide Polymorphism; T2DM, type 2 diabetes mellitus; TSS, transcription start site; UCSC, University of California, Santa Cruz; UTR, untranslated region
AC C
EP
Correspondence: Anna Mae Diehl, MD Department of Gastroenterology, Duke University, Synderman Building (GSRB-1), 595 LaSalle Street, Suite 1073, Durham, NC 27710, USA, Phone Number: 919-684-4173 Fax Number: 919-684-4183 email: [email protected] Disclosures: The authors have no relevant disclosures. Author Contributions: Study concept and design: SKM, CAM, MFA, AMD Acquisition of data: SKM, CAM, HP, AD, MFA, DDH, AS, MAH, AMD Analysis and interpretation of data: HY, HP, SKM, CAM, MG, AAK, AMD Drafting of the manuscript: SKM, HY, AMD Critical revision of the manuscript: SKM, MFA, AMD, CAM Statistical analysis: HY, HP, AD Obtained funding: AMD Study supervision: SKM, CAM, AMD
ACCEPTED MANUSCRIPT
Abstract: Background & Aims: Cirrhosis and liver cancer are potential outcomes of advanced nonalcoholic fatty
RI PT
liver disease (NAFLD). It is not clear what factors determine whether patients will develop advanced or mild NAFLD, limiting non-invasive diagnosis and treatment before clinical sequelae emerge. We investigated whether DNA methylation profiles can distinguish patients with mild disease from those
SC
with advanced NAFLD, and how these patterns are functionally related to hepatic gene expression.
Methods: We collected frozen liver biopsies and clinical data from patients with biopsy-proven NAFLD
M AN U
(56 in the discovery cohort and 34 in the replication cohort). Samples were divided into groups based on histologic severity of fibrosis: F0–1 (mild) and F3–4 (advanced). DNA methylation profiles were determined and coupled with gene expression data from the same biopsies; differential methylation was validated in subsets of the discovery and replication cohorts. We then analyzed interactions between the
TE D
methylome and transcriptome.
Results: Clinical features did not differ between patients known to have mild or advanced fibrosis based
EP
on biopsy analysis. There were 69,247 differentially methylated CpG sites (76% hypomethylated, 24% hypermethylated) in patients with advanced vs mild NAFLD (P<.05). Methylation at FGFR2, MAT1A,
AC C
and CASP1 was validated by bisulfite pyrosequencing and the findings were reproduced in the replication cohort. Methylation correlated with gene transcript levels for 7% of differentially methylated CpG sites, indicating that differential methylation contributes to differences in expression. In samples with advanced NAFLD, many tissue repair genes were hypomethylated and overexpressed, whereas genes in certain metabolic pathways, including 1-carbon metabolism, were hypermethylated and underexpressed.
ACCEPTED MANUSCRIPT
Conclusions: Functionally relevant differences in methylation can distinguish patients with advanced vs mild NAFLD. Altered methylation of genes that regulate processes such as steatohepatitis, fibrosis, and
AC C
EP
TE D
M AN U
SC
Keywords: NAFLD; DNA methylation; gene expression; microarrays
RI PT
carcinogenesis indicate the role of DNA methylation in progression of NAFLD.
ACCEPTED MANUSCRIPT
Introduction Nonalcoholic fatty liver disease (NAFLD) is strongly associated with obesity, type 2-diabetes,
RI PT
and the metabolic syndrome. Like these other conditions, NAFLD is increasing in incidence and prevalence. It is now the most common cause of chronic liver disease in the United States and Western Europe 1. NAFLD encompasses a spectrum of liver pathology that is generally
SC
characterized by excessive accumulation of fat in hepatocytes (i.e., steatosis). Some individuals with fatty hepatocytes also have co-incident hepatic inflammation and increased liver cell death
M AN U
(i.e., nonalcoholic steatohepatitis (NASH)). The outcomes of steatosis and steatohepatitis are very different. Individuals with steatosis rarely develop liver fibrosis. Hence, they seldom progress to clinically significant liver disease and are considered to have mild NAFLD. In contrast, progressive liver fibrosis occurs in some individuals with NASH, significantly increasing their risk for liver cirrhosis, primary liver cancer, and resultant liver-related morbidity
TE D
and mortality 2. Because NASH may trigger a fibrogenic repair process that eventually culminates in cirrhosis and/or cancer, it is considered to be part of the spectrum of advanced
EP
NAFLD.
AC C
Currently, liver biopsy is the only way to reliably stage the severity of liver fibrosis and thereby distinguish individuals with mild NAFLD from those with advanced NAFLD before overt clinical sequelae of liver damage emerge 1. This limits population-based screening, delaying diagnosis of individuals with NAFLD who are at high risk for eventual liver-related morbidity and mortality. Thus, noninvasive biomarkers are needed. Once NASH has been diagnosed, interventions to prevent progression are necessary, but must have a low risk/benefit ratio given the high prevalence of NASH and the variable (but generally slow) rates of progression. Success
ACCEPTED MANUSCRIPT
in this endeavor mandates identification of risk factors for advanced NAFLD that are easily (and safely) modified.
RI PT
DNA methylation is an epigenetic form of gene regulation that is generally associated with transcriptional repression when present at the promoter regions of genes. It works in concert with histone modifications to regulate the activity of genes, and these regulatory mechanisms
SC
help guide levels of gene transcription in all tissues throughout the course of our lives. DNA methylation changes are known to modulate the susceptibility to obesity, a major risk factor for
M AN U
NAFLD 3, 4, as well as the outcomes of diet-induced liver injury in rodents. Methyl-depleted diets promote steatohepatitis, cirrhosis, and liver cancer in rats and mice 5-8, while replenishing methyl stores averts all of these outcomes 9. Thus, changes in the “methylome” are plausible in humans with NAFLD, might differentiate those with mild NAFLD from those with more
TE D
advanced NAFLD, and may provide novel therapeutic targets. Currently, however, almost nothing is known about the methylome in human NAFLD. To rectify this gap in knowledge, we generated comprehensive liver DNA methylation profiles in a large, well-characterized group of
EP
patients with either mild or advanced NAFLD, and correlated differences in liver DNA methylation with differences in liver gene expression (GEx). We discovered that there are
AC C
numerous, functionally-relevant methylation differences that distinguish mild from advanced NAFLD. These results are consistent with a prominent role for methylation in NAFLD progression in humans.
ACCEPTED MANUSCRIPT
Methods
RI PT
Detailed methods are available in the online supplement.
Study samples
The Duke University Health System NAFLD Clinical Database and Biorepository contains frozen liver biopsies and clinical data from patients with biopsy-proven NAFLD. The
SC
biorepository is approved by the Duke University Institutional Review Board. Demographic data
M AN U
and laboratory studies were obtained within 6 months of liver biopsy.
Generation of genomic data
Generation of GEx data for these samples has been described 10. The Illumina HumanMethylation450 beadchip platform was used for 33 mild (fibrosis stage, F0-F1) and 23
TE D
advanced (fibrosis stage, F3-F4) NAFLD specimens at Expression Analysis (Research Triangle Park, NC).
EP
DNA methylation data preprocessing
Data was processed in one batch (8 arrays with 12 samples per array). Samples were randomly
artifacts.
AC C
distributed across the arrays. Principle component analysis was used to identify potential sample
DNA methylation data analysis Modified t-test statistics identified differentially methylated (DM) CpG sites between advanced and mild NAFLD. Analysis of variance (ANOVA) model fitting was used to identify DM CpG
ACCEPTED MANUSCRIPT
sites between advanced and mild NAFLD after controlling for age and gender, with significance defined as q-value <.05. Data analyses were performed using R/Bioconductor statistical
RI PT
packages 11.
Integration of DNA methylation and expression data
Generation of Affymetrix Human Genome U133 Plus 2.0 GeneChip platform (Affymetrix, Santa
SC
Clara, CA) expression data has been described [10; NCBI Accession GSE31803]. GEx and methylation data were available for 45 of the 56 patients (27 mild and 18 advanced NAFLD).
M AN U
CpG island (CGI) and expression probe sets were coupled based on information in the UCSC genome browser (GRCh37/hg19)12 and correlations measured via the Spearman rank correlation coefficient.
TE D
Bioinformatics Analysis
The Ingenuity Pathways Analysis Tool (IPA, Ingenuity systems, Inc., Redwood City, CA, www.ingenuity.com) was used to determine potential functional significance of methylation-
AC C
Data Access
EP
expression relationships.
Human Methylation450 beadchip data will be made available from the Gene Expression Omnibus web site.
ACCEPTED MANUSCRIPT
Results Patient and sample characteristics
RI PT
Genome-scale DNA methylation profiles were obtained for liver tissue of 56 patients including 33 with mild and 23 with advanced NAFLD. Clinical and laboratory characteristics of patients with advanced NAFLD were not significantly different from patients with mild NAFLD in both the discovery and replication cohorts (Table 1). Histologic characteristics reflecting NAFLD
SC
severity differed among advanced and mild NAFLD patients however, with advanced NAFLD patients having significantly more portal inflammation and ballooning, and therefore
M AN U
significantly higher NAFLD activity scores (NAS), than those with mild NAFLD (Table 1). These findings confirmed that fibrosis stage provided a reliable surrogate for NAFLD disease severity.
TE D
All samples were processed in one batch, but we detected an apparent within-batch positional effect. Signal intensities differed for three of the arrays (24 samples) based on principal components analysis (Supplementary Fig 1). The samples were intentionally randomized across
EP
positions within the array such that no one group was solely influenced by this difference. We
effect.
AC C
used a two-step quantile normalization approach to reduce the artificial noise from this positional
DNA methylation varies across the genome in NAFLD To determine if NAFLD status influences the distribution of DNA methylation, we first examined methylation profiles according to gene structure. Consistent with the Encyclopedia of DNA elements (ENCODE) data showing that the vast majority of CGI and transcription start
ACCEPTED MANUSCRIPT sites (TSS) are normally unmethylated 13, we observed overall hypomethylation at transcription start sites (TSS) and most CGI in NAFLD tissue (Fig. 1A). Autosomal and sex chromosome methylation also differed, with relative hypermethylation of CGI on the X chromosomes, likely
RI PT
due to X-inactivation in females (64% and 78% of the mild and advanced NAFLD patients, respectively). This pattern was similar for both mild and advanced NAFLD.
SC
Differential methylation in cancer is reported as most evident at CGI ‘shores’ 14, 15, defined as regions located up to 2 kilobases (kb) upstream or downstream from a CGI 15. When we
M AN U
compared methylation profiles in mild to advanced NAFLD, we also found differential methylation was more pronounced at regions flanking TSSs or CGIs (Fig. 1B). Since many CGIs are located at the promoter regions of genes, very close to the TSS, it was not clear whether these findings were due to proximity to the CGI or the TSS. Thus, we restricted the analysis to CGI
TE D
and TSS positioned at least 1000 base pairs (bp) apart. Using this reduced dataset, we observed increased differential methylation between mild and advanced NAFLD at CGI shores rather than
EP
TSSs (Fig. 1C).
The Infinium HumanMethylation450 BeadChip data is obtained using two distinct chemical
AC C
assays (Infinium I and Infinium II probes) with divergent intensity distributions16 (Supplementary Fig. 2). We measured the proportion of differential methylation using the Type I or Type II probes independently, and reached the same conclusion regardless of the assay type. (Supplementary Fig. 3).
CpG methylation differs between mild and advanced NAFLD A regression model was used to test for differential methylation as the outcome due to NAFLD
ACCEPTED MANUSCRIPT
status after controlling for gender and age. The model identified 69,247 (14.3% out of the total CpG sites) DM CpG sites between advanced and mild NAFLD using a q-value threshold of .05. Among the DM CpG sites, advanced NAFLD showed relative hypomethylation at 52,830 CpG
RI PT
sites (11% of total), and hypermethylation at 16,417 sites (3% of total). These results suggest that advanced NAFLD has a greater loss of gene regulatory capacity compared to mild NAFLD.
SC
Functional methylation-expression relationships identified in NAFLD
There were distinct methylation profiles in NAFLD with regard to genomic position annotations,
M AN U
including intergenic, promoter, 5’ and 3’ untranslated region (UTR) and coding region CpG sites (Fig. 2). For example, promoters showed increased hypomethylated sites, while 3’UTRs showed increased hypermethylated sites. DNA methylation at gene promoter regions is often associated with transcriptional repression due to interactions between DNA methylation, methyl DNA
TE D
binding proteins, and histone deacetyltransferases, which contribute to the formation of heterochromatin. In contrast, promoter hypomethylation is often associated with a euchromatic state and transcriptional permissiveness. We used our pre-existing GEx data 10 for a subset of
EP
the NAFLD liver tissues to investigate how DNA methylation might affect GEx in this disease. For each transcript with a defined promoter, we took the median methylation intensities of all the
AC C
CpG sites located within the region 500 bp upstream to 25 bp downstream of the annotated TSS, and plotted this against the mean transcription levels of the hypomethylated (median methylation <10th quantile) and hypermethylated (median methylation >90th quantile) genes. As expected, expression levels from hypomethylated genes were higher than those from hypermethylated genes for all regions except the 3’ UTR, where there was little difference in expression regardless of methylation or NAFLD status (Fig. 2). We also confirmed that the general
ACCEPTED MANUSCRIPT
relationships between methylation and expression with regard to positioning relative to CGI and/or TSS were not altered by Infinium assay type (Supplementary Fig. 4).
RI PT
Advanced NAFLD exhibits generalized hypomethylation relative to mild NAFLD Next we investigated the number of DM genes between mild and advanced NAFLD.
Deregulation of methylation of multiple CpG sites within the same CGI often occurs. Thus to
SC
evaluate ‘gene’ level alterations, we tested each CGI rather than individual CpG sites using repeated measures ANOVA. Because the methylation status in promoter regions was more likely
M AN U
to directly affect gene expression, here we only studied CGIs in gene promoters, within 2 kb from the TSS. We used the UCSC genome browser to map the positions of the TSSs, and Wu 17 to define the CGIs. Out of 28,337 genes, 19,077 genes had at least one CGI at the promoter region (<2 kb from TSS). There were 1,741 hypomethylated and 624 hypermethylated genes in
TE D
advanced NAFLD (q-value <.05). For genes not having a promoter CGI, 1,514 have at least one CpG within 500 bp of the TSS, and of these, 110 were hypomethylated and 28 were hypermethylated in advanced versus mild NAFLD (q-value <.05). In total, we identified 2,503
EP
of 20,591 genes studied as DM; of these ,1,851 and 652 genes were hypo- and hypermethylated,
AC C
respectively in advanced NAFLD.
We used bisulfite pyrosequencing to validate the Infinium data for fibroblast growth factor receptor 2 (FGFR2), methionine adenosyl methyltransferase 1A (MAT1A) and caspase 1 (CASP1) by querying methylation of the same CpG dinucleotides significantly DM on the Infinium beadchip. Pyrosequencing assay performance was confirmed using defined proportions of unmethylated and methylated DNAs, with excellent correlation between input and measured methylation (R2>0.98; Figure 4A-C). For 24 of the same liver specimens with sufficient DNA
ACCEPTED MANUSCRIPT
remaining, there was good agreement between the Infinium beadchip and bisulfite pyrosequencing results for FGFR2 (r=.936, p<.0001) and MAT1A (r=.914, p<.0001) (Figure 4DE). For CASP1, despite a positive association between the two methods in detecting methylation,
RI PT
this relationship did not quite reach significance. The methylation levels on the two platforms for CASP1 differed substantially, with pyrosequencing showing <6% and the Infinium platform
SC
showing > 20% methylation for the same tissues (Figure 4F).
We validated methylation findings in an independent group of liver specimens classified as
M AN U
having mild (N=19) or advanced (N=15) NAFLD. Normal liver tissues (N=25) were included for comparison. We again found significant differences between mild and advanced NAFLD for FGFR2 (p=0.002) and CASP1 (p=.024), while MAT1A approached significance (p=.088) (Figure 5A-C). Significant differences in the same direction for CASP1 supports the validity of the
TE D
original findings. Methylation levels in normal livers compared to mild NAFLD showed no differences, while they differed markedly from advanced NAFLD (FGFR2: p<.0001; MAT1A:
EP
p=.008; CASP1: p=.001).
Livers with advanced NAFLD show increased expression of hypomethylated genes
AC C
Next we coupled GEx with methylation status of CpG sites, and identified genes that were both differentially methylated and expressed, and determined if the methylation and expression profiles for these genes went in the expected direction. To do this, we matched each CpG site to its corresponding GEx probe set and identified 76,833 unique CpG – gene expression pairs. We calculated the Spearman rank correlation between methylation of each CpG and the corresponding gene’s transcription level, and found 4,849 CpG sites with r<-.3 (corresponding to a p-value of .02), signifying a significant inverse relationship between the GEx and CpG site
ACCEPTED MANUSCRIPT
methylation. These 4,849 sites include hypermethylated downregulated genes as well as hypomethylated upregulated genes.
RI PT
Associations between expression and methylation for which only one or two CpG sites exhibit a correlative relationship may be due to chance. We therefore restricted focus to genes showing a significant inverse expression-methylation relationship with a unidirectional difference in
SC
methylation for ≥3 CpG sites. This conservative approach also protects against the possibility that SNPs within, or near to, the methylation probes might solely influence findings. One
M AN U
hundred forty three genes were hypermethylated and 365 genes were hypomethylated at ≥3 CpG sites with methylation levels showing a significant inverse relationship with expression in the NAFLD cases (total of 2,637 probes) (Supplementary Table 7).
TE D
Many tissue repair genes are hypomethylated and over-expressed in advanced NAFLD Often a group of genes rather than a single gene regulates a phenotype, and network based approaches have been used to identify sets of functionally related genes 18. We used the IPA tool
EP
to explore the functions and pathways enriched within the groups of hypomethylated and hypermethylated genes in advanced versus mild NAFLD. Genes with numerous CpG sites
AC C
showing differential methylation between mild and advanced NAFLD included transmembrane protein 204 (TMEM204; 24 CpG sites hypomethyated and upregulated in advanced NAFLD). This gene encodes a hypoxia-related adherans junction protein involved in cell adhesion and angiogenesis 19 and is aberrantly methylated in pancreatic cancer 20. FGFR2 is also hypomethylated (23 CpG sites) with increased expression in advanced NAFLD. FGFR2 functions as a receptor for keratinocyte growth factor, a protein made in chronic liver disease by stellate cells 21, 22. Other genes encoding matrix molecules and matrix remodeling factors were
ACCEPTED MANUSCRIPT
also hypomethylated with increased expression in advanced versus mild NAFLD, including those associated with fibrosis and cirrhosis (COL1A1, COL1A2, COL4A1, COL4A2, LAMA4, LAMB1, CTGF, and PDGFA), several chemokines (CCR7 and CCL5), and factors related to the
RI PT
inflammasome and pro-inflammatory immune response (STAT1, TNFAIP8 and CASP1).
“Cancer” was the most significant biological function category in pathway analysis with an
SC
overrepresentation of hypomethylated, upregulated genes in advanced versus mild NAFLD. Top functional annotations included neoplasia, tumorigenesis and gastrointestinal tract cancer
M AN U
(p<7.78e-14). Other relevant categories were “cellular movement” (cell movement and migration of cells; p<1.49e-15) and “cell death” (p=2.09e-13). Significant canonical pathways included “role of tissue factor in cancer” (5.93e00) and “hepatic fibrosis / hepatic stellate cell activation” (4.79e00) while significant enrichment of networks included “cellular movement,
TE D
cancer and cell morphology” (score=44), “cardiovascular system development and function, organismal development and cellular movement” (score=37) and “lipid metabolism, molecular transport, small molecule biochemistry” (score=36). Many of these genes, functions and
EP
pathways bear relevance to the biology underlying liver fibrosis, providing confidence that this approach has revealed logical methylation-expression relationships in this disease, while also
AC C
supporting the relevance for other findings that may not be as evident based on what is known about liver fibrosis. The complete lists of functions, canonical pathways and networks along with the corresponding hypomethylated and transcriptionally activated genes are provided in Supplementary Tables 1-3.
Certain metabolic genes are hypermethylated and down-regulated in advanced NAFLD Biological function categories enriched for the group of hypermethylated, transcriptionally
ACCEPTED MANUSCRIPT
repressed genes included “small molecule biochemistry” (uptake of thyroid hormone, 1.75e-05; oxidation of lipids; 2.85e-05). Canonical pathways enriched for hypermethylated genes were “LPS/IL-1 mediated inhibition of RXR function”(1.32e-06) and “fatty acid metabolism” (2.16e-
RI PT
05). Retinoid X Receptors are proteins important in the transmission of signals that regulate cholesterol metabolism 23, and a high cholesterol, high fat diet recapitulates many features of nonalcoholic steatohepatitis in rodent models 24. Specific hypermethylated and repressed genes
SC
within this pathway included APOC4 (lipid metabolism), CYP2C19 (cytochrome P450 family; hepatocellular carcinoma), SLCO1B3 (membrane transporter and role in apoptosis; liver cancer,
M AN U
known target of methylation mediated silencing, including HepG2 cells 25), SLC10A1 (membrane transport, role in liver cancer; demonstrated to be unmethylated in normal mouse liver tissues 26), ABCC2 (multidrug resistance, liver neoplasia), ALDH (cell migration, cancer stem cells self-renewal; targeted by methylation in rat hepatoma cells 27), MGMT (apoptosis, DNA damage response, known target of hypermethylation in liver cancer 28, 29), ACOX
TE D
(steatohepatitis, microvesicular hepatic steatosis) and ACSL5 (fatty acid anabolic pathways 30,
EP
colon cancer).
Genes involved in histidine, selenoamino acid, tryptophan, tyrosine and B-alanine metabolism
AC C
were also repressed by hypermethylation (all p<.004) as were genes involved in the “urea cycle and metabolism of amino groups” (p=.001). Networks that contain significant numbers of hypermethylated and repressed genes included “lipid metabolism, small molecule biochemistry, cellular compromise” (score=44) and “lipid metabolism, small molecule biochemistry, molecular transport” (score=42). The complete lists of functions, canonical pathways and networks along with the corresponding hypermethylated and repressed genes are provided in Supplementary Tables 4-6.
ACCEPTED MANUSCRIPT
Differential methylation and expression of genes that regulate methylation distinguishes advanced from mild NAFLD
RI PT
Of particular interest is the differential methylation and expression of genes involved in onecarbon metabolism, including methylenetetrahydrofolate dehydrogenase 2 (MTHFD2;
hypomethylated in advanced NAFLD), adenosyl homocysteine (AHCY; hypermethylated),
SC
MAT1A (hypermethylated) and aldehyde dehydrogenase 1 family, member L1 (ALDH1L1; hypermethylated) (see Figure 3). MTHFD2 functions in the mitochondria to generate formate,
31
M AN U
which enters the cytoplasm and is incorporated into tetrahydrofolates for one carbon metabolism . While MTHFD2 has not been implicated in fibrosis, its protein product promotes cell
proliferation 32. Expression of MTHFD2 has only been detected in transformed cells or during embryonic development 31. A role for AHCY in NAFLD has also not been reported, but Ahcy downregulation with increased methylation was documented in mice fed a methyl-deficient diet,
TE D
conditions that lead to NASH 33.
EP
The MAT1A protein generates S-adenosylmethionine, the universal methyl group donor. Mat1a deletion causes NAFLD in mice 34 and MAT1A is a target of promoter methylation in
AC C
hepatocarcinogenesis 35. Two of the three MAT1A CpG sites identified here are located upstream of the transcription start site (TSS), while the third is ~200 bp downstream from the TSS. This is consistent with the report that coding region methylation (+88 from the TSS) is strongly associated with transcriptional regulation of MAT1A in human liver cells 36.
ALDH1L1 encodes 10-formyltetrahydrofolate dehydrogenase (FTHFD), an enzyme involved in folate metabolism and the one carbon pathway for generation of methyl group donors. FTHFD
ACCEPTED MANUSCRIPT is abundant in normal hepatocytes, accounting for ~1% of the total cytosolic protein 37. This gene is also a known target of methylation in liver cells, being unmethylated in normal hepatocytes but highly methylated in hepatocellular carcinoma cells 38, with decreased
RI PT
expression an independent predictor of overall survival 39.
Discussion
SC
We report the first integrated analysis of genome-wide methylation and GEx data of diseased, but non-cancerous, human livers. Our results were generated from percutaneous liver biopsy
M AN U
samples of well-characterized patients with NAFLD, one of the most common types of chronic liver disease in adults 1. Liver histology, the current gold standard diagnostic test 1, was used to segregate these subjects into two groups that are predicted to have extremely different liver outcomes at 10 years based on differences in the severity of liver fibrosis at the time of tissue
TE D
acquisition 2. As expected, the subgroup with little-to-no fibrosis had mild liver damage (mild NAFLD), while the other subgroup with more severe fibrosis had worse liver damage (advanced NAFLD), although the two histologically-defined subgroups were clinically indistinguishable.
EP
We identified a large number of DM CpG sites between advanced and mild NAFLD, and independently confirmed methylation differences for FGFR2, MAT1A and CASP1 with
AC C
replication in an independent NAFLD cohort. FGFR2 was hypomethylated in advanced NAFLD and is involved in epithelial to mesenchymal transition. CASP1, hypomethylated in advanced NAFLD, is positively associated with liver inflammation and fibrosis 40, 41.
The study also revealed a large number of genes whose transcriptional status is significantly correlated with levels of DNA methylation, and showed that this relationship differs in mild versus advanced NAFLD. Furthermore, many of the differentially affected genes are interrelated
ACCEPTED MANUSCRIPT
not only by their roles in specific pathways and networks, but are highly relevant to the ultimate outcomes of liver injury, based on current understanding of cirrhosis and liver cancer
RI PT
pathogenesis.
Epigenetic processes are now widely recognized as playing prominent roles in many common diseases. While histone modifications and microRNAs are critically important to regulating
SC
gene expression, DNA methylation is a more tractable modification for analysis due to its
stability and the availability of technologies that allow for relatively rapid assessment of large
M AN U
fractions of the methylome. Combining methylation data with GEx microarray data provides an extremely powerful tool to identify genes whose expression levels correlate with changes in DNA methylation and thus may be directly influenced by factors that skew normal epigenetic regulation. We used this approach to determine if epigenome-transcriptome interactions differ in
TE D
individuals with mild NAFLD versus those with advanced NAFLD.
We found that livers with advanced NAFLD are generally hypomethylated relative to those with
EP
mild NAFLD. For example, of the genes showing significant inverse relationships between methylation and expression and with at least three DM CpG sites, there were more than twice as
AC C
many genes that were hypomethylated with increased expression in advanced NAFLD relative to mild NAFLD. Hypomethylation is also a characteristic of alcohol-related fatty liver injury. 6, 42 Consistent with these results, a prior study of mice fed a methyl-deficient diet to induce steatohepatitis also demonstrated relative hepatic hypomethylation. That study also noted significantly increased methylation and decreased expression of Ahcy in mice with steatohepatitis,33 as was observed in the advanced human NAFLD cases evaluated here. In addition, we found that MAT1A, a key gene responsible for generation of methyl groups used in
ACCEPTED MANUSCRIPT
DNA methylation reactions, was relatively hypermethylated and under-expressed in patients with advanced NAFLD. Knockdown of Mat1a in mice enhances vulnerability to diet-induced fatty liver, resulting in steatohepatitis, 5 altered lipid and carbohydrate metabolism and increased
RI PT
incidence of hepatocellular carcinoma 7. Like MAT1A, other genes that regulate one-carbon metabolism were DM and expressed in human NAFLD, with the resultant methylome-
transcriptome profiles predicting reduced net efficiency of methylation in advanced NAFLD.
M AN U
hypomethylated relative to those in livers with mild NAFLD.
SC
Consistent with that concept, genes in human livers with advanced NAFLD were generally
Notably, many genes that modulate diverse wound healing responses were hypomethylated and up-regulated in advanced NAFLD, and pathway analysis confirmed that processes involved in carcinogenesis and fibrogenesis were induced. The cumulative data, therefore, suggest that
TE D
methylome-transcriptome interactions that occur in advanced NAFLD may modulate NAFLD outcomes, including development of cancer and/or cirrhosis (Figure 6). Indeed, changes in DNA methylation at specific genes are well documented in cancer, and have a profound influence on
EP
cancer initiation and progression 8. Changes in methylation were recently reported to accompany steatohepatitis and precede emergence of hepatocellular carcinoma following infection with
AC C
hepatitis C virus 43. Altered DNA methylation has also been reported in liver cirrhosis, primarily in animal models 9 where it is associated with activation of stellate cells and development of fibrosis 3.
By identifying plausible methylation-sensitive mechanisms that influence NAFLD progression, our work may lead to the development of noninvasive biomarkers that will help identify NAFLD patients at high risk for bad liver outcomes. For example, hypermethioninemia is characterized
ACCEPTED MANUSCRIPT
by abnormally elevated methionine levels in blood and has been reported in patients with alcohol-induced cirrhosis 44. Known genetic causes include deficiencies of MAT1A, glycine Nmethyltransferase (GNMT) and AHCY 4. It is of interest here that transcription of all three of
RI PT
these hypermethioninemia-causative genes was repressed in the advanced NAFLD patients and this coincided with increased methylation of their respective promoters. Hence, epigenetic repression of transcription of these key genes in advanced NAFLD may mimic the genetic
SC
defects known to cause hypermethioninemia, providing a novel noninvasive biomarker for advanced NAFLD. Importantly, evidence for methylome-transcriptome interactions in human
M AN U
NAFLD has immediate therapeutic implications because a methyl donor-rich diet was able to reverse progression of fatty liver damage in rats fed a high-fat-sucrose diet 45, and others have shown that methyl group donor-rich diets directly affect DNA methylation changes in animal models of obesity 46. Thus, epigenetic modifications could provide malleable targets for future
TE D
interventions that aim to prevent/reverse advanced NAFLD. We acknowledge that our approach cannot determine causality and that the results are largely descriptive, indicating only a general pattern given that neither the expression or methylation platform used is an exhaustive
EP
representation. Also, future research is necessary to confirm that observed DM-related changes in gene expression translate into comparable differences in the respective liver proteins.
AC C
Nevertheless, the provocative data generated by this first-in-human genome-wide integrated methylome-transcriptome analysis of damaged livers provide compelling justification for further research that might alleviate an emerging burden to public health worldwide.
Acknowledgments We thank Daniel Hampton, Zhiqing Huang and Carole Grenier for technical assistance.
ACCEPTED MANUSCRIPT
Figure Legends
RI PT
Figure 1. Methylation patterns in NAFLD relative to transcription start sites (TSS) and CGI. (A) Median methylation for autosomes (solid lines) and the X chromosome in liver tissues from female NAFLD patients (dashed line) as a function of distance from TSSs (left) or from CGIs (right). (B) Probes showing significant differential methylation between mild and
SC
advanced NAFLD as a function of distance from TSSs and CGIs, including non-promoter and promoter CGI. Blue lines, p<.05; red line q<.05. (C) Median methylation for non-CGI probes
M AN U
with significant differential methylation between mild and advanced NAFLD relative to the distance from TSSs and promoter CGIs. Blue lines, p<.05; red line q<.05.
Figure 2. Methylation and gene expression patterns relative to genomic structure.
TE D
Methylation (left) and gene expression (right) level distributions among mild and advanced NAFLD at intergenic (A), ‘broad’ promoter regions (-2kbp to +25bp of TSS) (B), ‘narrow’ promoter regions (-500bp to +25bp of TSS) (C), the 5’ UTR (D), coding region (E) and 3’ UTR
EP
(F). Blue, mild NAFLD; red, advanced NAFLD. For gene expression: blue and red, low methylation (<10th quantile) in mild and advanced NAFLD, respectively; green and black, high
AC C
methylation (>90th quantile) in mild and advanced NAFLD, respectively.
Figure 3. Relationship between CpG methylation and gene expression. Data is shown for mild (open symbols) and advanced (filled symbols) NAFLD cases, with trendlines, for three representative genes: MTHFD2 (panel A; Spearman r=-.33, p=.03), AHCY (panel B; r=-.48, p=.0008) and MAT1A (panel C; r=-.49; p=.0006).
ACCEPTED MANUSCRIPT
Figure 4. Pyrosequencing validation of methylation detected by the Infinium HumanMethylation450 beadchip. Assay performance was evaluated using mixtures of
RI PT
unmethylated and methylated genomic DNAs for FGFR2 (A), MAT1A (B) and CASP1 (C). Pyrosequencing validation of the methylation levels for FGFR2 (D), MAT1A (E) and CASP1 (F)
SC
detected by the Infinium beadchip.
Figure 5. Independent confirmation of NAFLD differential methylation in a replication
M AN U
cohort. Pyrosequencing of FGFR2 (A), MAT1A (B) and CASP1 (C) in a replication cohort using the assays shown in Figure 4. DNA was insufficient for two of the 15 advanced NAFLD cases. Non-parametric student t tests p values are shown.
AC C
EP
TE D
Figure 6. Summary of research findings and potential implications.
ACCEPTED MANUSCRIPT
References Chalasani N, Younossi Z, Lavine JE, et al. The diagnosis and management of nonalcoholic fatty liver disease: practice guideline by the American Association for the Study of Liver Diseases, American College of Gastroenterology, and the American Gastroenterological Association. Hepatology 2012;55:2005-23.
2.
McCullough AJ. The clinical features, diagnosis and natural history of nonalcoholic fatty liver disease. Clin Liver Dis 2004;8:521-33.
3.
Bian EB, Huang C, Ma TT, et al. DNMT1-mediated PTEN hypermethylation confers hepatic stellate cell activation and liver fibrogenesis in rats. Toxicol Appl Pharmacol 2012;264:13-22.
4.
Mudd SH. Hypermethioninemias of genetic and non-genetic origin: A review. Am J Med Genet C Semin Med Genet 2011;157:3-32.
5.
Lu SC, Alvarez L, Huang ZZ, et al. Methionine adenosyltransferase 1A knockout mice are predisposed to liver injury and exhibit increased expression of genes involved in proliferation. Proc Natl Acad Sci U S A 2001;98:5560-5.
6.
Lu SC, Huang ZZ, Yang H, et al. Changes in methionine adenosyltransferase and Sadenosylmethionine homeostasis in alcoholic rat liver. Am J Physiol Gastrointest Liver Physiol 2000;279:G178-85.
7.
Martinez-Chantar ML, Corrales FJ, Martinez-Cruz LA, et al. Spontaneous oxidative stress and liver tumors in mice lacking methionine adenosyltransferase 1A. Faseb J 2002;16:1292-4.
8.
Robertson KD. DNA methylation and human disease. Nat Rev Genet 2005;6:597-610.
9.
Komatsu Y, Waku T, Iwasaki N, et al. Global analysis of DNA methylation in earlystage liver fibrosis. BMC Med Genomics 2012;5:5.
10.
Moylan CA, Pang H, Dellinger A, et al. Hepatic gene expression profiles differentiate pre-symptomatic NAFLD patients with high versus low risk for liver related mortality. Hepatology 2013;in press.
11.
Gentleman RC, Carey VJ, Bates DM, et al. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol 2004;5:R80.
12.
Sanborn JZ, Benz SC, Craft B, Szeto, C, et al. The UCSC Cancer Genomics Browser: update 2011. Nucleic Acids Res 2011;39:D951-9.
13.
Thurman RE, Rynes E, Humbert R, et al. The accessible chromatin landscape of the human genome. Nature 2012;489:75-82.
AC C
EP
TE D
M AN U
SC
RI PT
1.
ACCEPTED MANUSCRIPT
Doi A, Park IH, Wen B, et al. Differential methylation of tissue- and cancer-specific CpG island shores distinguishes human induced pluripotent stem cells, embryonic stem cells and fibroblasts. Nat Genet 2009.
15.
Irizarry RA, Ladd-Acosta C, Wen B, et al. The human colon cancer methylome shows similar hypo- and hypermethylation at conserved tissue-specific CpG island shores. Nature genetics 2009;41:178-86.
16.
Dedeurwaerder S, Defrance M, Calonne E, et al. Evaluation of the Infinium Methylation 450K technology. Epigenomics 2011;3:771-84.
17.
Wu H, Caffo B, Jaffee HA, et al. Redefining CpG islands using hidden Markov models. Biostatistics 2010;11:499-514.
18.
Matsumura N, Huang Z, Mori S, et al. Epigenetic suppression of the TGF-beta pathway revealed by transcriptome profiling in ovarian cancer. Genome Res 2011;21:74-82.
19.
Kearsey J, Petit S, De Oliveira C, et al. A novel four transmembrane spanning protein, CLP24. A hypoxically regulated cell junction protein. Eur J Biochem 2004;271:2584-92.
20.
Shimizu H, Horii A, Sunamura M, et al. Identification of epigenetically silenced genes in human pancreatic cancer by a novel method "microarray coupled with methyl-CpG targeted transcriptional activation" (MeTA-array). Biochem Biophys Res Commun 2011;411:162-7.
21.
Myers RP. Noninvasive diagnosis of nonalcoholic fatty liver disease. Ann Hepatol 2009;8 Suppl 1:S25-33.
22.
Baranova A, Birerdinc A, Estep M, et al. Pathogenesis of obesity-related chronic liver diseases as the study case for the systems biology. In: Choi S, ed. Systems Biology for Signaling Networks. New York: Springer, 2010:645-686.
23.
Zhao C, Dahlman-Wright K. Liver X receptor in cholesterol metabolism. J Endocrinol 2010;204:233-40.
24.
Savard C, Tartaglione EV, Kuver R, et al. Synergistic interaction of dietary cholesterol and dietary fat in inducing experimental steatohepatitis. Hepatology 2012;57:81-92.
25.
Ichihara S, Kikuchi R, Kusuhara H, et al. DNA methylation profiles of organic anion transporting polypeptide 1B3 in cancer cell lines. Pharm Res 2010;27:510-6.
26.
Imai S, Kikuchi R, Kusuhara H, et al. Analysis of DNA methylation and histone modification profiles of liver-specific transporters. Mol Pharmacol 2009;75:568-76.
27.
Lin KH, Brennan MD, Lindahl R. Expression of tumor-associated aldehyde dehydrogenase gene in rat hepatoma cell lines. Cancer Res 1988;48:7009-12.
AC C
EP
TE D
M AN U
SC
RI PT
14.
ACCEPTED MANUSCRIPT
Harder J, Opitz OG, Brabender J, et al. Quantitative promoter methylation analysis of hepatocellular carcinoma, cirrhotic and normal liver. International Journal of Cancer 2008;122:2800-4.
29.
Yu J, Zhang HY, Ma ZZ, et al. Methylation profiling of twenty four genes and the concordant methylation behaviours of nineteen genes that may contribute to hepatocellular carcinogenesis. Cell Res 2003;13:319-33.
30.
Bu SY, Mashek DG. Hepatic long-chain acyl-CoA synthetase 5 mediates fatty acid channeling between anabolic and catabolic pathways. J Lipid Res 2010;51:3270-80.
31.
Christensen KE, Mackenzie RE. Mitochondrial methylenetetrahydrofolate dehydrogenase, methenyltetrahydrofolate cyclohydrolase, and formyltetrahydrofolate synthetases. Vitam Horm 2008;79:393-410.
32.
Selcuklu SD, Donoghue MT, Rehmet K, et al. MicroRNA-9 Inhibition of Cell Proliferation and Identification of Novel miR-9 Targets by Transcriptome Profiling in Breast Cancer Cells. J Biol Chem 2012;287:29516-28.
33.
Tryndyak VP, Han T, Muskhelishvili L, et al. Coupling global methylation and gene expression profiles reveal key pathophysiological events in liver injury induced by a methyl-deficient diet. Mol Nutr Food Res 2011;55:411-8.
34.
Cano A, Buque X, Martinez-Una M, et al. Methionine adenosyltransferase 1A gene deletion disrupts hepatic very low-density lipoprotein assembly in mice. Hepatology 2011;54:1975-86.
35.
Frau M, Tomasi ML, Simile MM, et al. Role of transcriptional and posttranscriptional regulation of methionine adenosyltransferases in liver cancer progression. Hepatology 2012;56:165-75.
36.
Tomasi ML, Li TW, Li M, et al. Inhibition of human methionine adenosyltransferase 1A transcription by coding region methylation. J Cell Physiol 2012;227:1583-91.
37.
Kisliuk RL. Folate biochemistry in relation to antifolate selectivity. In: Jackman AL, ed. Antifolate Drugs in Cancer Therapy. Totowa: Humana Press, 1999:13-36.
38.
Oleinik NV, Krupenko NI, Krupenko SA. Epigenetic silencing of ALDH1L1, a metabolic regulator of cellular proliferation, in cancers. Genes Cancer 2011;2:130-9.
39.
Chen XQ, He JR, Wang HY. Decreased expression of ALDH1L1 is associated with a poor prognosis in hepatocellular carcinoma. Med Oncol 2011;29:1843-1849.
40.
Dixon LJ, Berk M, Thapaliya S, et al. Caspase-1-mediated regulation of fibrogenesis in diet-induced steatohepatitis. Laboratory investigation; a journal of technical methods and pathology 2012;92:713-23.
41.
Witek RP, Stone WC, Karaca FG, et al. Pan-caspase inhibitor VX-166 reduces fibrosis in an animal model of nonalcoholic steatohepatitis. Hepatology 2009;50:1421-30.
AC C
EP
TE D
M AN U
SC
RI PT
28.
ACCEPTED MANUSCRIPT
Lu SC, Tsukamoto H, Mato JM. Role of abnormal methionine metabolism in alcoholic liver injury. Alcohol 2002;27:155-62.
43.
Nishida N, Kudo M, Nagasaka T, et al. Characteristic patterns of altered DNA methylation predict emergence of human hepatocellular carcinoma. Hepatology 2012;56:994-1003.
44.
Russmann S, Junker E, Lauterburg BH. Remethylation and transsulfuration of methionine in cirrhosis: studies with L-[H3-methyl-1-C]methionine. Hepatology 2002;36:1190-6.
45.
Cordero P, Gomez-Uriz AM, Campion J, et al. Dietary supplementation with methyl donors reduces fatty liver and modifies the fatty acid synthase DNA methylation profile in rats fed an obesogenic diet. Genes Nutr 2012;8:105-113.
46.
Waterland RA, Jirtle RL. Transposable elements: targets for early nutritional effects on epigenetic gene regulation. Mol Cell Biol 2003;23:5293-300.
M AN U
SC
RI PT
42.
AC C
EP
TE D
Author names in bold designate shared co-first authorship.
ACCEPTED MANUSCRIPT
Table 1. Characteristics of the NAFLD methylation discovery and replication cohorts.
Laboratory measures, median (IQR) c Serum AST, U/L c Serum ALT, U/L c AST/ALT
33 (100) 0 (0)
21 (91.3) 2 (8.7)
32 (97) 21 (91.3) 1 (3) 2 (8.7) 32.8 (28.4 – 40.4) 33.8 (31.3 – 41.9) 14 (42.4) 15 (65.2) 6 (18.2) 9 (39.1) 5.8 (5.5 - 6.4) 6 (5.8 - 6.9) 7 (21.2) 4 (17.4) 6 (18.2) 5 (22.7)
42 (24 - 57) 47.5 (30 - 75) .8 (.64 - 1.03)
62 (46 - 85) 74.5 (46 - 101) .92 (.66 - 1.24)
14 (45.2) 6 (19.4) 3 (10.7) 17 (56.7) 7 (23.3)
14 (60.9) 9 (42.9) 12 (54.6) 22 (95.7) 12 (57.1)
.93 .09 .08 .34 .72 .72
.06 .45 .55
.25 .07 .0014 .0015 .014
EP
AC C
a
10 (52.6) 50.4 (3.2)
8 (53.3) 50.9 (3.1)
17 (89.5) 2 (10.5)
11 (73.3) 4 (26.7)
.56
Histologic characteristics, n (%) Steatosis (% > 34%) c Lobular inflammation (% > grade 2) b Portal inflammation (% > mild) d Ballooning (% any) e* NAFLD Activity Score (% ≥ 5)
.24 .97 .16
Comparison P .97 .74 .37
RI PT
5 (21.7) 51.7 (10.3)
Replication Cohort (n=34)
Controls (n=25)
18 (52.9) 50.6 (3.2)
13 (52) 50.4 (13.8)
28 (82.4) 6 (17.6)
13 (52) 11 (44)
.35
15 (78.9) 4 (21.1) 32 (30.5 – 33.6) 13 (68.4) 6 (31.6) 5.8 (5.6 – 6.2) 10 (52.6) g 1 (2.0)
14 (93.3) 1 (6.7) 32.8 (31 – 34.5) 13 (86.7) 7 (46.7) 5.7 (5.4 – 8.2) 8 (53.3) h 2 (33.3)
38 (30 - 58) 56 (38 – 99) .72 (.59 – .79)
10 (52.6) 1 (5.3) 6 (31.6) 16 (84.2) 5 (26.3)
*NAFLD Activity Score (range 0-8) is a sum of scores for steatosis, lobular inflammation, and ballooning. a missing data for 2 patients in discovery cohort b missing data for 6 patients in discovery cohort c missing data for 4 patients in discovery cohort d missing data for 3 patients in discovery cohort e missing data for 5 patients in discovery cohort f missing data for 12 patients in discovery cohort and 2 in replication cohort g missing data for 14 patients h missing data for 9 patients i missing data for 20 patients
P .94 .57 .03
.21
.38 .27 .81 .7 .96 1.0
29 (85.3) 5 (14.7) 32.3 (30.6 – 33.6) 26 (76.5) 13 (38.2) 5.8 (5.6 – 6.4) 18 (52.9) 3 (27)
18 (72) 7 (28) 38.9 (28.1 – 48.4) 15 (60) 7 (28) 5.7 (5.4 – 7) 9 (36) i 2 (4)
.06 .17 .41 .79 .19 1.0
61 (38 – 87) 61 (51 – 99) .84 (.66 – 1.15)
.07 .53 .07
46.5 (33 – 78) 59 (49 – 99) .77 (.62 – .88)
29 (25 – 34) 32 (22 – 39) .97 (.85 – 1.13)
<.001 <.001 .001
9 (60) 1 (6.7) 9 (60) 13 (86.7) 11 (73.3)
.67 .86 .1 .84 .02
19 (55.9) 2 (5.9) 15 (44.1) 29 (85.3) 16 (47)
-
-
SC
Hypertension, n (%) Diabetes Mellitus, n (%) f Hemoglobin A1c, median % (IQR) Hyperlipidemia, n (%) a Current Smoking, n (%)
12 (36.4) 51.5 (10.3)
P
Replication Cohort Mild NAFLD Advanced F0-F1 NAFLD F3-4 (n=19) (n=15)
M AN U
Male sex, n (%) Age in years at biopsy, mean (SD) Ethnicity, n (%) Non-Hispanic Unknown Race, n (%) White Black a BMI, median kg/m2 (IQR)
Discovery Cohort Mild NAFLD Advanced F0-F1 NAFLD F3-4 (n=33) (n=23)
TE D
Patient demographics and clinical characteristics
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
S0016-5085(13)01137-2 10.1053/j.gastro.2013.07.047 YGAST 58588
To appear in: Gastroenterology Accepted Date: 26 July 2013 Please cite this article as: Murphy SK, Yang H, Moylan CA, Pang H, Dellinger A, Abdelmalek MF, Garrett ME, Ashley-Koch A, Suzuki A, Tillmann HL, Hauser MA, Diehl AM, Relationship Between the Methylome and Transcriptome in Patients with Non-Alcoholic Fatty Liver Disease, Gastroenterology (2013), doi: 10.1053/j.gastro.2013.07.047. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. All studies published in Gastroenterology are embargoed until 3PM ET of the day they are published as corrected proofs on-line. Studies cannot be publicized as accepted manuscripts or uncorrected proofs.
ACCEPTED MANUSCRIPT
Relationship Between the Methylome and Transcriptome in Patients with Non-Alcoholic Fatty Liver Disease (Functional methylation in NAFLD) Susan K. Murphy1, Hyuna Yang2, Cynthia A. Moylan3, Herbert Pang2, Andrew Dellinger , Manal F. Abdelmalek3, Melanie E. Garrett4, Allison Ashley-Koch4, Ayako Suzuki3, Hans L. Tillmann5, Michael A. Hauser4 and Anna Mae Diehl3
RI PT
2
SC
1 Department of Obstetrics and Gynecology, Duke University, Durham, NC 2 Department of Bioinformatics and Biostatistics, Duke University, Durham, NC 3 Division of Gastroenterology and Hepatology, Duke University, Durham, NC 4 Center for Human Genetics, Duke University, Durham, NC 5 Section of Medical Genetics, Department of Medicine, Duke University, Durham, NC
Grant Support: National Institutes of Health 5RC2-AA019399
TE D
M AN U
Abbreviations: ANOVA, Analysis of Variance; bp, base pairs; BMI, body mass index; CGI, CpG islands; CASP1, caspase 1; DM, differentially methylated; DUHS, Duke University Health System; ENCODE, Encylopedia of DNA elements; FGFR2, fibroblast growth factor receptor 2; kb, kilobase; gene expression, (GEx); HbA1, hemoglobin A1c; IPA, Ingenuity Pathway Analysis; M, methylation intensities; MAT1A, methionine adenosyl methyltransferase 1A; NAFLD, nonalcoholic fatty liver disease; NAS, NAFLD activity score; NASH, nonalcoholic steatohepatitis; NCBI, National Center for Biotechnology Information; SD, standard deviation; SNP, Simple Nucleotide Polymorphism; T2DM, type 2 diabetes mellitus; TSS, transcription start site; UCSC, University of California, Santa Cruz; UTR, untranslated region
AC C
EP
Correspondence: Anna Mae Diehl, MD Department of Gastroenterology, Duke University, Synderman Building (GSRB-1), 595 LaSalle Street, Suite 1073, Durham, NC 27710, USA, Phone Number: 919-684-4173 Fax Number: 919-684-4183 email: [email protected] Disclosures: The authors have no relevant disclosures. Author Contributions: Study concept and design: SKM, CAM, MFA, AMD Acquisition of data: SKM, CAM, HP, AD, MFA, DDH, AS, MAH, AMD Analysis and interpretation of data: HY, HP, SKM, CAM, MG, AAK, AMD Drafting of the manuscript: SKM, HY, AMD Critical revision of the manuscript: SKM, MFA, AMD, CAM Statistical analysis: HY, HP, AD Obtained funding: AMD Study supervision: SKM, CAM, AMD
ACCEPTED MANUSCRIPT
Abstract: Background & Aims: Cirrhosis and liver cancer are potential outcomes of advanced nonalcoholic fatty
RI PT
liver disease (NAFLD). It is not clear what factors determine whether patients will develop advanced or mild NAFLD, limiting non-invasive diagnosis and treatment before clinical sequelae emerge. We investigated whether DNA methylation profiles can distinguish patients with mild disease from those
SC
with advanced NAFLD, and how these patterns are functionally related to hepatic gene expression.
Methods: We collected frozen liver biopsies and clinical data from patients with biopsy-proven NAFLD
M AN U
(56 in the discovery cohort and 34 in the replication cohort). Samples were divided into groups based on histologic severity of fibrosis: F0–1 (mild) and F3–4 (advanced). DNA methylation profiles were determined and coupled with gene expression data from the same biopsies; differential methylation was validated in subsets of the discovery and replication cohorts. We then analyzed interactions between the
TE D
methylome and transcriptome.
Results: Clinical features did not differ between patients known to have mild or advanced fibrosis based
EP
on biopsy analysis. There were 69,247 differentially methylated CpG sites (76% hypomethylated, 24% hypermethylated) in patients with advanced vs mild NAFLD (P<.05). Methylation at FGFR2, MAT1A,
AC C
and CASP1 was validated by bisulfite pyrosequencing and the findings were reproduced in the replication cohort. Methylation correlated with gene transcript levels for 7% of differentially methylated CpG sites, indicating that differential methylation contributes to differences in expression. In samples with advanced NAFLD, many tissue repair genes were hypomethylated and overexpressed, whereas genes in certain metabolic pathways, including 1-carbon metabolism, were hypermethylated and underexpressed.
ACCEPTED MANUSCRIPT
Conclusions: Functionally relevant differences in methylation can distinguish patients with advanced vs mild NAFLD. Altered methylation of genes that regulate processes such as steatohepatitis, fibrosis, and
AC C
EP
TE D
M AN U
SC
Keywords: NAFLD; DNA methylation; gene expression; microarrays
RI PT
carcinogenesis indicate the role of DNA methylation in progression of NAFLD.
ACCEPTED MANUSCRIPT
Introduction Nonalcoholic fatty liver disease (NAFLD) is strongly associated with obesity, type 2-diabetes,
RI PT
and the metabolic syndrome. Like these other conditions, NAFLD is increasing in incidence and prevalence. It is now the most common cause of chronic liver disease in the United States and Western Europe 1. NAFLD encompasses a spectrum of liver pathology that is generally
SC
characterized by excessive accumulation of fat in hepatocytes (i.e., steatosis). Some individuals with fatty hepatocytes also have co-incident hepatic inflammation and increased liver cell death
M AN U
(i.e., nonalcoholic steatohepatitis (NASH)). The outcomes of steatosis and steatohepatitis are very different. Individuals with steatosis rarely develop liver fibrosis. Hence, they seldom progress to clinically significant liver disease and are considered to have mild NAFLD. In contrast, progressive liver fibrosis occurs in some individuals with NASH, significantly increasing their risk for liver cirrhosis, primary liver cancer, and resultant liver-related morbidity
TE D
and mortality 2. Because NASH may trigger a fibrogenic repair process that eventually culminates in cirrhosis and/or cancer, it is considered to be part of the spectrum of advanced
EP
NAFLD.
AC C
Currently, liver biopsy is the only way to reliably stage the severity of liver fibrosis and thereby distinguish individuals with mild NAFLD from those with advanced NAFLD before overt clinical sequelae of liver damage emerge 1. This limits population-based screening, delaying diagnosis of individuals with NAFLD who are at high risk for eventual liver-related morbidity and mortality. Thus, noninvasive biomarkers are needed. Once NASH has been diagnosed, interventions to prevent progression are necessary, but must have a low risk/benefit ratio given the high prevalence of NASH and the variable (but generally slow) rates of progression. Success
ACCEPTED MANUSCRIPT
in this endeavor mandates identification of risk factors for advanced NAFLD that are easily (and safely) modified.
RI PT
DNA methylation is an epigenetic form of gene regulation that is generally associated with transcriptional repression when present at the promoter regions of genes. It works in concert with histone modifications to regulate the activity of genes, and these regulatory mechanisms
SC
help guide levels of gene transcription in all tissues throughout the course of our lives. DNA methylation changes are known to modulate the susceptibility to obesity, a major risk factor for
M AN U
NAFLD 3, 4, as well as the outcomes of diet-induced liver injury in rodents. Methyl-depleted diets promote steatohepatitis, cirrhosis, and liver cancer in rats and mice 5-8, while replenishing methyl stores averts all of these outcomes 9. Thus, changes in the “methylome” are plausible in humans with NAFLD, might differentiate those with mild NAFLD from those with more
TE D
advanced NAFLD, and may provide novel therapeutic targets. Currently, however, almost nothing is known about the methylome in human NAFLD. To rectify this gap in knowledge, we generated comprehensive liver DNA methylation profiles in a large, well-characterized group of
EP
patients with either mild or advanced NAFLD, and correlated differences in liver DNA methylation with differences in liver gene expression (GEx). We discovered that there are
AC C
numerous, functionally-relevant methylation differences that distinguish mild from advanced NAFLD. These results are consistent with a prominent role for methylation in NAFLD progression in humans.
ACCEPTED MANUSCRIPT
Methods
RI PT
Detailed methods are available in the online supplement.
Study samples
The Duke University Health System NAFLD Clinical Database and Biorepository contains frozen liver biopsies and clinical data from patients with biopsy-proven NAFLD. The
SC
biorepository is approved by the Duke University Institutional Review Board. Demographic data
M AN U
and laboratory studies were obtained within 6 months of liver biopsy.
Generation of genomic data
Generation of GEx data for these samples has been described 10. The Illumina HumanMethylation450 beadchip platform was used for 33 mild (fibrosis stage, F0-F1) and 23
TE D
advanced (fibrosis stage, F3-F4) NAFLD specimens at Expression Analysis (Research Triangle Park, NC).
EP
DNA methylation data preprocessing
Data was processed in one batch (8 arrays with 12 samples per array). Samples were randomly
artifacts.
AC C
distributed across the arrays. Principle component analysis was used to identify potential sample
DNA methylation data analysis Modified t-test statistics identified differentially methylated (DM) CpG sites between advanced and mild NAFLD. Analysis of variance (ANOVA) model fitting was used to identify DM CpG
ACCEPTED MANUSCRIPT
sites between advanced and mild NAFLD after controlling for age and gender, with significance defined as q-value <.05. Data analyses were performed using R/Bioconductor statistical
RI PT
packages 11.
Integration of DNA methylation and expression data
Generation of Affymetrix Human Genome U133 Plus 2.0 GeneChip platform (Affymetrix, Santa
SC
Clara, CA) expression data has been described [10; NCBI Accession GSE31803]. GEx and methylation data were available for 45 of the 56 patients (27 mild and 18 advanced NAFLD).
M AN U
CpG island (CGI) and expression probe sets were coupled based on information in the UCSC genome browser (GRCh37/hg19)12 and correlations measured via the Spearman rank correlation coefficient.
TE D
Bioinformatics Analysis
The Ingenuity Pathways Analysis Tool (IPA, Ingenuity systems, Inc., Redwood City, CA, www.ingenuity.com) was used to determine potential functional significance of methylation-
AC C
Data Access
EP
expression relationships.
Human Methylation450 beadchip data will be made available from the Gene Expression Omnibus web site.
ACCEPTED MANUSCRIPT
Results Patient and sample characteristics
RI PT
Genome-scale DNA methylation profiles were obtained for liver tissue of 56 patients including 33 with mild and 23 with advanced NAFLD. Clinical and laboratory characteristics of patients with advanced NAFLD were not significantly different from patients with mild NAFLD in both the discovery and replication cohorts (Table 1). Histologic characteristics reflecting NAFLD
SC
severity differed among advanced and mild NAFLD patients however, with advanced NAFLD patients having significantly more portal inflammation and ballooning, and therefore
M AN U
significantly higher NAFLD activity scores (NAS), than those with mild NAFLD (Table 1). These findings confirmed that fibrosis stage provided a reliable surrogate for NAFLD disease severity.
TE D
All samples were processed in one batch, but we detected an apparent within-batch positional effect. Signal intensities differed for three of the arrays (24 samples) based on principal components analysis (Supplementary Fig 1). The samples were intentionally randomized across
EP
positions within the array such that no one group was solely influenced by this difference. We
effect.
AC C
used a two-step quantile normalization approach to reduce the artificial noise from this positional
DNA methylation varies across the genome in NAFLD To determine if NAFLD status influences the distribution of DNA methylation, we first examined methylation profiles according to gene structure. Consistent with the Encyclopedia of DNA elements (ENCODE) data showing that the vast majority of CGI and transcription start
ACCEPTED MANUSCRIPT sites (TSS) are normally unmethylated 13, we observed overall hypomethylation at transcription start sites (TSS) and most CGI in NAFLD tissue (Fig. 1A). Autosomal and sex chromosome methylation also differed, with relative hypermethylation of CGI on the X chromosomes, likely
RI PT
due to X-inactivation in females (64% and 78% of the mild and advanced NAFLD patients, respectively). This pattern was similar for both mild and advanced NAFLD.
SC
Differential methylation in cancer is reported as most evident at CGI ‘shores’ 14, 15, defined as regions located up to 2 kilobases (kb) upstream or downstream from a CGI 15. When we
M AN U
compared methylation profiles in mild to advanced NAFLD, we also found differential methylation was more pronounced at regions flanking TSSs or CGIs (Fig. 1B). Since many CGIs are located at the promoter regions of genes, very close to the TSS, it was not clear whether these findings were due to proximity to the CGI or the TSS. Thus, we restricted the analysis to CGI
TE D
and TSS positioned at least 1000 base pairs (bp) apart. Using this reduced dataset, we observed increased differential methylation between mild and advanced NAFLD at CGI shores rather than
EP
TSSs (Fig. 1C).
The Infinium HumanMethylation450 BeadChip data is obtained using two distinct chemical
AC C
assays (Infinium I and Infinium II probes) with divergent intensity distributions16 (Supplementary Fig. 2). We measured the proportion of differential methylation using the Type I or Type II probes independently, and reached the same conclusion regardless of the assay type. (Supplementary Fig. 3).
CpG methylation differs between mild and advanced NAFLD A regression model was used to test for differential methylation as the outcome due to NAFLD
ACCEPTED MANUSCRIPT
status after controlling for gender and age. The model identified 69,247 (14.3% out of the total CpG sites) DM CpG sites between advanced and mild NAFLD using a q-value threshold of .05. Among the DM CpG sites, advanced NAFLD showed relative hypomethylation at 52,830 CpG
RI PT
sites (11% of total), and hypermethylation at 16,417 sites (3% of total). These results suggest that advanced NAFLD has a greater loss of gene regulatory capacity compared to mild NAFLD.
SC
Functional methylation-expression relationships identified in NAFLD
There were distinct methylation profiles in NAFLD with regard to genomic position annotations,
M AN U
including intergenic, promoter, 5’ and 3’ untranslated region (UTR) and coding region CpG sites (Fig. 2). For example, promoters showed increased hypomethylated sites, while 3’UTRs showed increased hypermethylated sites. DNA methylation at gene promoter regions is often associated with transcriptional repression due to interactions between DNA methylation, methyl DNA
TE D
binding proteins, and histone deacetyltransferases, which contribute to the formation of heterochromatin. In contrast, promoter hypomethylation is often associated with a euchromatic state and transcriptional permissiveness. We used our pre-existing GEx data 10 for a subset of
EP
the NAFLD liver tissues to investigate how DNA methylation might affect GEx in this disease. For each transcript with a defined promoter, we took the median methylation intensities of all the
AC C
CpG sites located within the region 500 bp upstream to 25 bp downstream of the annotated TSS, and plotted this against the mean transcription levels of the hypomethylated (median methylation <10th quantile) and hypermethylated (median methylation >90th quantile) genes. As expected, expression levels from hypomethylated genes were higher than those from hypermethylated genes for all regions except the 3’ UTR, where there was little difference in expression regardless of methylation or NAFLD status (Fig. 2). We also confirmed that the general
ACCEPTED MANUSCRIPT
relationships between methylation and expression with regard to positioning relative to CGI and/or TSS were not altered by Infinium assay type (Supplementary Fig. 4).
RI PT
Advanced NAFLD exhibits generalized hypomethylation relative to mild NAFLD Next we investigated the number of DM genes between mild and advanced NAFLD.
Deregulation of methylation of multiple CpG sites within the same CGI often occurs. Thus to
SC
evaluate ‘gene’ level alterations, we tested each CGI rather than individual CpG sites using repeated measures ANOVA. Because the methylation status in promoter regions was more likely
M AN U
to directly affect gene expression, here we only studied CGIs in gene promoters, within 2 kb from the TSS. We used the UCSC genome browser to map the positions of the TSSs, and Wu 17 to define the CGIs. Out of 28,337 genes, 19,077 genes had at least one CGI at the promoter region (<2 kb from TSS). There were 1,741 hypomethylated and 624 hypermethylated genes in
TE D
advanced NAFLD (q-value <.05). For genes not having a promoter CGI, 1,514 have at least one CpG within 500 bp of the TSS, and of these, 110 were hypomethylated and 28 were hypermethylated in advanced versus mild NAFLD (q-value <.05). In total, we identified 2,503
EP
of 20,591 genes studied as DM; of these ,1,851 and 652 genes were hypo- and hypermethylated,
AC C
respectively in advanced NAFLD.
We used bisulfite pyrosequencing to validate the Infinium data for fibroblast growth factor receptor 2 (FGFR2), methionine adenosyl methyltransferase 1A (MAT1A) and caspase 1 (CASP1) by querying methylation of the same CpG dinucleotides significantly DM on the Infinium beadchip. Pyrosequencing assay performance was confirmed using defined proportions of unmethylated and methylated DNAs, with excellent correlation between input and measured methylation (R2>0.98; Figure 4A-C). For 24 of the same liver specimens with sufficient DNA
ACCEPTED MANUSCRIPT
remaining, there was good agreement between the Infinium beadchip and bisulfite pyrosequencing results for FGFR2 (r=.936, p<.0001) and MAT1A (r=.914, p<.0001) (Figure 4DE). For CASP1, despite a positive association between the two methods in detecting methylation,
RI PT
this relationship did not quite reach significance. The methylation levels on the two platforms for CASP1 differed substantially, with pyrosequencing showing <6% and the Infinium platform
SC
showing > 20% methylation for the same tissues (Figure 4F).
We validated methylation findings in an independent group of liver specimens classified as
M AN U
having mild (N=19) or advanced (N=15) NAFLD. Normal liver tissues (N=25) were included for comparison. We again found significant differences between mild and advanced NAFLD for FGFR2 (p=0.002) and CASP1 (p=.024), while MAT1A approached significance (p=.088) (Figure 5A-C). Significant differences in the same direction for CASP1 supports the validity of the
TE D
original findings. Methylation levels in normal livers compared to mild NAFLD showed no differences, while they differed markedly from advanced NAFLD (FGFR2: p<.0001; MAT1A:
EP
p=.008; CASP1: p=.001).
Livers with advanced NAFLD show increased expression of hypomethylated genes
AC C
Next we coupled GEx with methylation status of CpG sites, and identified genes that were both differentially methylated and expressed, and determined if the methylation and expression profiles for these genes went in the expected direction. To do this, we matched each CpG site to its corresponding GEx probe set and identified 76,833 unique CpG – gene expression pairs. We calculated the Spearman rank correlation between methylation of each CpG and the corresponding gene’s transcription level, and found 4,849 CpG sites with r<-.3 (corresponding to a p-value of .02), signifying a significant inverse relationship between the GEx and CpG site
ACCEPTED MANUSCRIPT
methylation. These 4,849 sites include hypermethylated downregulated genes as well as hypomethylated upregulated genes.
RI PT
Associations between expression and methylation for which only one or two CpG sites exhibit a correlative relationship may be due to chance. We therefore restricted focus to genes showing a significant inverse expression-methylation relationship with a unidirectional difference in
SC
methylation for ≥3 CpG sites. This conservative approach also protects against the possibility that SNPs within, or near to, the methylation probes might solely influence findings. One
M AN U
hundred forty three genes were hypermethylated and 365 genes were hypomethylated at ≥3 CpG sites with methylation levels showing a significant inverse relationship with expression in the NAFLD cases (total of 2,637 probes) (Supplementary Table 7).
TE D
Many tissue repair genes are hypomethylated and over-expressed in advanced NAFLD Often a group of genes rather than a single gene regulates a phenotype, and network based approaches have been used to identify sets of functionally related genes 18. We used the IPA tool
EP
to explore the functions and pathways enriched within the groups of hypomethylated and hypermethylated genes in advanced versus mild NAFLD. Genes with numerous CpG sites
AC C
showing differential methylation between mild and advanced NAFLD included transmembrane protein 204 (TMEM204; 24 CpG sites hypomethyated and upregulated in advanced NAFLD). This gene encodes a hypoxia-related adherans junction protein involved in cell adhesion and angiogenesis 19 and is aberrantly methylated in pancreatic cancer 20. FGFR2 is also hypomethylated (23 CpG sites) with increased expression in advanced NAFLD. FGFR2 functions as a receptor for keratinocyte growth factor, a protein made in chronic liver disease by stellate cells 21, 22. Other genes encoding matrix molecules and matrix remodeling factors were
ACCEPTED MANUSCRIPT
also hypomethylated with increased expression in advanced versus mild NAFLD, including those associated with fibrosis and cirrhosis (COL1A1, COL1A2, COL4A1, COL4A2, LAMA4, LAMB1, CTGF, and PDGFA), several chemokines (CCR7 and CCL5), and factors related to the
RI PT
inflammasome and pro-inflammatory immune response (STAT1, TNFAIP8 and CASP1).
“Cancer” was the most significant biological function category in pathway analysis with an
SC
overrepresentation of hypomethylated, upregulated genes in advanced versus mild NAFLD. Top functional annotations included neoplasia, tumorigenesis and gastrointestinal tract cancer
M AN U
(p<7.78e-14). Other relevant categories were “cellular movement” (cell movement and migration of cells; p<1.49e-15) and “cell death” (p=2.09e-13). Significant canonical pathways included “role of tissue factor in cancer” (5.93e00) and “hepatic fibrosis / hepatic stellate cell activation” (4.79e00) while significant enrichment of networks included “cellular movement,
TE D
cancer and cell morphology” (score=44), “cardiovascular system development and function, organismal development and cellular movement” (score=37) and “lipid metabolism, molecular transport, small molecule biochemistry” (score=36). Many of these genes, functions and
EP
pathways bear relevance to the biology underlying liver fibrosis, providing confidence that this approach has revealed logical methylation-expression relationships in this disease, while also
AC C
supporting the relevance for other findings that may not be as evident based on what is known about liver fibrosis. The complete lists of functions, canonical pathways and networks along with the corresponding hypomethylated and transcriptionally activated genes are provided in Supplementary Tables 1-3.
Certain metabolic genes are hypermethylated and down-regulated in advanced NAFLD Biological function categories enriched for the group of hypermethylated, transcriptionally
ACCEPTED MANUSCRIPT
repressed genes included “small molecule biochemistry” (uptake of thyroid hormone, 1.75e-05; oxidation of lipids; 2.85e-05). Canonical pathways enriched for hypermethylated genes were “LPS/IL-1 mediated inhibition of RXR function”(1.32e-06) and “fatty acid metabolism” (2.16e-
RI PT
05). Retinoid X Receptors are proteins important in the transmission of signals that regulate cholesterol metabolism 23, and a high cholesterol, high fat diet recapitulates many features of nonalcoholic steatohepatitis in rodent models 24. Specific hypermethylated and repressed genes
SC
within this pathway included APOC4 (lipid metabolism), CYP2C19 (cytochrome P450 family; hepatocellular carcinoma), SLCO1B3 (membrane transporter and role in apoptosis; liver cancer,
M AN U
known target of methylation mediated silencing, including HepG2 cells 25), SLC10A1 (membrane transport, role in liver cancer; demonstrated to be unmethylated in normal mouse liver tissues 26), ABCC2 (multidrug resistance, liver neoplasia), ALDH (cell migration, cancer stem cells self-renewal; targeted by methylation in rat hepatoma cells 27), MGMT (apoptosis, DNA damage response, known target of hypermethylation in liver cancer 28, 29), ACOX
TE D
(steatohepatitis, microvesicular hepatic steatosis) and ACSL5 (fatty acid anabolic pathways 30,
EP
colon cancer).
Genes involved in histidine, selenoamino acid, tryptophan, tyrosine and B-alanine metabolism
AC C
were also repressed by hypermethylation (all p<.004) as were genes involved in the “urea cycle and metabolism of amino groups” (p=.001). Networks that contain significant numbers of hypermethylated and repressed genes included “lipid metabolism, small molecule biochemistry, cellular compromise” (score=44) and “lipid metabolism, small molecule biochemistry, molecular transport” (score=42). The complete lists of functions, canonical pathways and networks along with the corresponding hypermethylated and repressed genes are provided in Supplementary Tables 4-6.
ACCEPTED MANUSCRIPT
Differential methylation and expression of genes that regulate methylation distinguishes advanced from mild NAFLD
RI PT
Of particular interest is the differential methylation and expression of genes involved in onecarbon metabolism, including methylenetetrahydrofolate dehydrogenase 2 (MTHFD2;
hypomethylated in advanced NAFLD), adenosyl homocysteine (AHCY; hypermethylated),
SC
MAT1A (hypermethylated) and aldehyde dehydrogenase 1 family, member L1 (ALDH1L1; hypermethylated) (see Figure 3). MTHFD2 functions in the mitochondria to generate formate,
31
M AN U
which enters the cytoplasm and is incorporated into tetrahydrofolates for one carbon metabolism . While MTHFD2 has not been implicated in fibrosis, its protein product promotes cell
proliferation 32. Expression of MTHFD2 has only been detected in transformed cells or during embryonic development 31. A role for AHCY in NAFLD has also not been reported, but Ahcy downregulation with increased methylation was documented in mice fed a methyl-deficient diet,
TE D
conditions that lead to NASH 33.
EP
The MAT1A protein generates S-adenosylmethionine, the universal methyl group donor. Mat1a deletion causes NAFLD in mice 34 and MAT1A is a target of promoter methylation in
AC C
hepatocarcinogenesis 35. Two of the three MAT1A CpG sites identified here are located upstream of the transcription start site (TSS), while the third is ~200 bp downstream from the TSS. This is consistent with the report that coding region methylation (+88 from the TSS) is strongly associated with transcriptional regulation of MAT1A in human liver cells 36.
ALDH1L1 encodes 10-formyltetrahydrofolate dehydrogenase (FTHFD), an enzyme involved in folate metabolism and the one carbon pathway for generation of methyl group donors. FTHFD
ACCEPTED MANUSCRIPT is abundant in normal hepatocytes, accounting for ~1% of the total cytosolic protein 37. This gene is also a known target of methylation in liver cells, being unmethylated in normal hepatocytes but highly methylated in hepatocellular carcinoma cells 38, with decreased
RI PT
expression an independent predictor of overall survival 39.
Discussion
SC
We report the first integrated analysis of genome-wide methylation and GEx data of diseased, but non-cancerous, human livers. Our results were generated from percutaneous liver biopsy
M AN U
samples of well-characterized patients with NAFLD, one of the most common types of chronic liver disease in adults 1. Liver histology, the current gold standard diagnostic test 1, was used to segregate these subjects into two groups that are predicted to have extremely different liver outcomes at 10 years based on differences in the severity of liver fibrosis at the time of tissue
TE D
acquisition 2. As expected, the subgroup with little-to-no fibrosis had mild liver damage (mild NAFLD), while the other subgroup with more severe fibrosis had worse liver damage (advanced NAFLD), although the two histologically-defined subgroups were clinically indistinguishable.
EP
We identified a large number of DM CpG sites between advanced and mild NAFLD, and independently confirmed methylation differences for FGFR2, MAT1A and CASP1 with
AC C
replication in an independent NAFLD cohort. FGFR2 was hypomethylated in advanced NAFLD and is involved in epithelial to mesenchymal transition. CASP1, hypomethylated in advanced NAFLD, is positively associated with liver inflammation and fibrosis 40, 41.
The study also revealed a large number of genes whose transcriptional status is significantly correlated with levels of DNA methylation, and showed that this relationship differs in mild versus advanced NAFLD. Furthermore, many of the differentially affected genes are interrelated
ACCEPTED MANUSCRIPT
not only by their roles in specific pathways and networks, but are highly relevant to the ultimate outcomes of liver injury, based on current understanding of cirrhosis and liver cancer
RI PT
pathogenesis.
Epigenetic processes are now widely recognized as playing prominent roles in many common diseases. While histone modifications and microRNAs are critically important to regulating
SC
gene expression, DNA methylation is a more tractable modification for analysis due to its
stability and the availability of technologies that allow for relatively rapid assessment of large
M AN U
fractions of the methylome. Combining methylation data with GEx microarray data provides an extremely powerful tool to identify genes whose expression levels correlate with changes in DNA methylation and thus may be directly influenced by factors that skew normal epigenetic regulation. We used this approach to determine if epigenome-transcriptome interactions differ in
TE D
individuals with mild NAFLD versus those with advanced NAFLD.
We found that livers with advanced NAFLD are generally hypomethylated relative to those with
EP
mild NAFLD. For example, of the genes showing significant inverse relationships between methylation and expression and with at least three DM CpG sites, there were more than twice as
AC C
many genes that were hypomethylated with increased expression in advanced NAFLD relative to mild NAFLD. Hypomethylation is also a characteristic of alcohol-related fatty liver injury. 6, 42 Consistent with these results, a prior study of mice fed a methyl-deficient diet to induce steatohepatitis also demonstrated relative hepatic hypomethylation. That study also noted significantly increased methylation and decreased expression of Ahcy in mice with steatohepatitis,33 as was observed in the advanced human NAFLD cases evaluated here. In addition, we found that MAT1A, a key gene responsible for generation of methyl groups used in
ACCEPTED MANUSCRIPT
DNA methylation reactions, was relatively hypermethylated and under-expressed in patients with advanced NAFLD. Knockdown of Mat1a in mice enhances vulnerability to diet-induced fatty liver, resulting in steatohepatitis, 5 altered lipid and carbohydrate metabolism and increased
RI PT
incidence of hepatocellular carcinoma 7. Like MAT1A, other genes that regulate one-carbon metabolism were DM and expressed in human NAFLD, with the resultant methylome-
transcriptome profiles predicting reduced net efficiency of methylation in advanced NAFLD.
M AN U
hypomethylated relative to those in livers with mild NAFLD.
SC
Consistent with that concept, genes in human livers with advanced NAFLD were generally
Notably, many genes that modulate diverse wound healing responses were hypomethylated and up-regulated in advanced NAFLD, and pathway analysis confirmed that processes involved in carcinogenesis and fibrogenesis were induced. The cumulative data, therefore, suggest that
TE D
methylome-transcriptome interactions that occur in advanced NAFLD may modulate NAFLD outcomes, including development of cancer and/or cirrhosis (Figure 6). Indeed, changes in DNA methylation at specific genes are well documented in cancer, and have a profound influence on
EP
cancer initiation and progression 8. Changes in methylation were recently reported to accompany steatohepatitis and precede emergence of hepatocellular carcinoma following infection with
AC C
hepatitis C virus 43. Altered DNA methylation has also been reported in liver cirrhosis, primarily in animal models 9 where it is associated with activation of stellate cells and development of fibrosis 3.
By identifying plausible methylation-sensitive mechanisms that influence NAFLD progression, our work may lead to the development of noninvasive biomarkers that will help identify NAFLD patients at high risk for bad liver outcomes. For example, hypermethioninemia is characterized
ACCEPTED MANUSCRIPT
by abnormally elevated methionine levels in blood and has been reported in patients with alcohol-induced cirrhosis 44. Known genetic causes include deficiencies of MAT1A, glycine Nmethyltransferase (GNMT) and AHCY 4. It is of interest here that transcription of all three of
RI PT
these hypermethioninemia-causative genes was repressed in the advanced NAFLD patients and this coincided with increased methylation of their respective promoters. Hence, epigenetic repression of transcription of these key genes in advanced NAFLD may mimic the genetic
SC
defects known to cause hypermethioninemia, providing a novel noninvasive biomarker for advanced NAFLD. Importantly, evidence for methylome-transcriptome interactions in human
M AN U
NAFLD has immediate therapeutic implications because a methyl donor-rich diet was able to reverse progression of fatty liver damage in rats fed a high-fat-sucrose diet 45, and others have shown that methyl group donor-rich diets directly affect DNA methylation changes in animal models of obesity 46. Thus, epigenetic modifications could provide malleable targets for future
TE D
interventions that aim to prevent/reverse advanced NAFLD. We acknowledge that our approach cannot determine causality and that the results are largely descriptive, indicating only a general pattern given that neither the expression or methylation platform used is an exhaustive
EP
representation. Also, future research is necessary to confirm that observed DM-related changes in gene expression translate into comparable differences in the respective liver proteins.
AC C
Nevertheless, the provocative data generated by this first-in-human genome-wide integrated methylome-transcriptome analysis of damaged livers provide compelling justification for further research that might alleviate an emerging burden to public health worldwide.
Acknowledgments We thank Daniel Hampton, Zhiqing Huang and Carole Grenier for technical assistance.
ACCEPTED MANUSCRIPT
Figure Legends
RI PT
Figure 1. Methylation patterns in NAFLD relative to transcription start sites (TSS) and CGI. (A) Median methylation for autosomes (solid lines) and the X chromosome in liver tissues from female NAFLD patients (dashed line) as a function of distance from TSSs (left) or from CGIs (right). (B) Probes showing significant differential methylation between mild and
SC
advanced NAFLD as a function of distance from TSSs and CGIs, including non-promoter and promoter CGI. Blue lines, p<.05; red line q<.05. (C) Median methylation for non-CGI probes
M AN U
with significant differential methylation between mild and advanced NAFLD relative to the distance from TSSs and promoter CGIs. Blue lines, p<.05; red line q<.05.
Figure 2. Methylation and gene expression patterns relative to genomic structure.
TE D
Methylation (left) and gene expression (right) level distributions among mild and advanced NAFLD at intergenic (A), ‘broad’ promoter regions (-2kbp to +25bp of TSS) (B), ‘narrow’ promoter regions (-500bp to +25bp of TSS) (C), the 5’ UTR (D), coding region (E) and 3’ UTR
EP
(F). Blue, mild NAFLD; red, advanced NAFLD. For gene expression: blue and red, low methylation (<10th quantile) in mild and advanced NAFLD, respectively; green and black, high
AC C
methylation (>90th quantile) in mild and advanced NAFLD, respectively.
Figure 3. Relationship between CpG methylation and gene expression. Data is shown for mild (open symbols) and advanced (filled symbols) NAFLD cases, with trendlines, for three representative genes: MTHFD2 (panel A; Spearman r=-.33, p=.03), AHCY (panel B; r=-.48, p=.0008) and MAT1A (panel C; r=-.49; p=.0006).
ACCEPTED MANUSCRIPT
Figure 4. Pyrosequencing validation of methylation detected by the Infinium HumanMethylation450 beadchip. Assay performance was evaluated using mixtures of
RI PT
unmethylated and methylated genomic DNAs for FGFR2 (A), MAT1A (B) and CASP1 (C). Pyrosequencing validation of the methylation levels for FGFR2 (D), MAT1A (E) and CASP1 (F)
SC
detected by the Infinium beadchip.
Figure 5. Independent confirmation of NAFLD differential methylation in a replication
M AN U
cohort. Pyrosequencing of FGFR2 (A), MAT1A (B) and CASP1 (C) in a replication cohort using the assays shown in Figure 4. DNA was insufficient for two of the 15 advanced NAFLD cases. Non-parametric student t tests p values are shown.
AC C
EP
TE D
Figure 6. Summary of research findings and potential implications.
ACCEPTED MANUSCRIPT
References Chalasani N, Younossi Z, Lavine JE, et al. The diagnosis and management of nonalcoholic fatty liver disease: practice guideline by the American Association for the Study of Liver Diseases, American College of Gastroenterology, and the American Gastroenterological Association. Hepatology 2012;55:2005-23.
2.
McCullough AJ. The clinical features, diagnosis and natural history of nonalcoholic fatty liver disease. Clin Liver Dis 2004;8:521-33.
3.
Bian EB, Huang C, Ma TT, et al. DNMT1-mediated PTEN hypermethylation confers hepatic stellate cell activation and liver fibrogenesis in rats. Toxicol Appl Pharmacol 2012;264:13-22.
4.
Mudd SH. Hypermethioninemias of genetic and non-genetic origin: A review. Am J Med Genet C Semin Med Genet 2011;157:3-32.
5.
Lu SC, Alvarez L, Huang ZZ, et al. Methionine adenosyltransferase 1A knockout mice are predisposed to liver injury and exhibit increased expression of genes involved in proliferation. Proc Natl Acad Sci U S A 2001;98:5560-5.
6.
Lu SC, Huang ZZ, Yang H, et al. Changes in methionine adenosyltransferase and Sadenosylmethionine homeostasis in alcoholic rat liver. Am J Physiol Gastrointest Liver Physiol 2000;279:G178-85.
7.
Martinez-Chantar ML, Corrales FJ, Martinez-Cruz LA, et al. Spontaneous oxidative stress and liver tumors in mice lacking methionine adenosyltransferase 1A. Faseb J 2002;16:1292-4.
8.
Robertson KD. DNA methylation and human disease. Nat Rev Genet 2005;6:597-610.
9.
Komatsu Y, Waku T, Iwasaki N, et al. Global analysis of DNA methylation in earlystage liver fibrosis. BMC Med Genomics 2012;5:5.
10.
Moylan CA, Pang H, Dellinger A, et al. Hepatic gene expression profiles differentiate pre-symptomatic NAFLD patients with high versus low risk for liver related mortality. Hepatology 2013;in press.
11.
Gentleman RC, Carey VJ, Bates DM, et al. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol 2004;5:R80.
12.
Sanborn JZ, Benz SC, Craft B, Szeto, C, et al. The UCSC Cancer Genomics Browser: update 2011. Nucleic Acids Res 2011;39:D951-9.
13.
Thurman RE, Rynes E, Humbert R, et al. The accessible chromatin landscape of the human genome. Nature 2012;489:75-82.
AC C
EP
TE D
M AN U
SC
RI PT
1.
ACCEPTED MANUSCRIPT
Doi A, Park IH, Wen B, et al. Differential methylation of tissue- and cancer-specific CpG island shores distinguishes human induced pluripotent stem cells, embryonic stem cells and fibroblasts. Nat Genet 2009.
15.
Irizarry RA, Ladd-Acosta C, Wen B, et al. The human colon cancer methylome shows similar hypo- and hypermethylation at conserved tissue-specific CpG island shores. Nature genetics 2009;41:178-86.
16.
Dedeurwaerder S, Defrance M, Calonne E, et al. Evaluation of the Infinium Methylation 450K technology. Epigenomics 2011;3:771-84.
17.
Wu H, Caffo B, Jaffee HA, et al. Redefining CpG islands using hidden Markov models. Biostatistics 2010;11:499-514.
18.
Matsumura N, Huang Z, Mori S, et al. Epigenetic suppression of the TGF-beta pathway revealed by transcriptome profiling in ovarian cancer. Genome Res 2011;21:74-82.
19.
Kearsey J, Petit S, De Oliveira C, et al. A novel four transmembrane spanning protein, CLP24. A hypoxically regulated cell junction protein. Eur J Biochem 2004;271:2584-92.
20.
Shimizu H, Horii A, Sunamura M, et al. Identification of epigenetically silenced genes in human pancreatic cancer by a novel method "microarray coupled with methyl-CpG targeted transcriptional activation" (MeTA-array). Biochem Biophys Res Commun 2011;411:162-7.
21.
Myers RP. Noninvasive diagnosis of nonalcoholic fatty liver disease. Ann Hepatol 2009;8 Suppl 1:S25-33.
22.
Baranova A, Birerdinc A, Estep M, et al. Pathogenesis of obesity-related chronic liver diseases as the study case for the systems biology. In: Choi S, ed. Systems Biology for Signaling Networks. New York: Springer, 2010:645-686.
23.
Zhao C, Dahlman-Wright K. Liver X receptor in cholesterol metabolism. J Endocrinol 2010;204:233-40.
24.
Savard C, Tartaglione EV, Kuver R, et al. Synergistic interaction of dietary cholesterol and dietary fat in inducing experimental steatohepatitis. Hepatology 2012;57:81-92.
25.
Ichihara S, Kikuchi R, Kusuhara H, et al. DNA methylation profiles of organic anion transporting polypeptide 1B3 in cancer cell lines. Pharm Res 2010;27:510-6.
26.
Imai S, Kikuchi R, Kusuhara H, et al. Analysis of DNA methylation and histone modification profiles of liver-specific transporters. Mol Pharmacol 2009;75:568-76.
27.
Lin KH, Brennan MD, Lindahl R. Expression of tumor-associated aldehyde dehydrogenase gene in rat hepatoma cell lines. Cancer Res 1988;48:7009-12.
AC C
EP
TE D
M AN U
SC
RI PT
14.
ACCEPTED MANUSCRIPT
Harder J, Opitz OG, Brabender J, et al. Quantitative promoter methylation analysis of hepatocellular carcinoma, cirrhotic and normal liver. International Journal of Cancer 2008;122:2800-4.
29.
Yu J, Zhang HY, Ma ZZ, et al. Methylation profiling of twenty four genes and the concordant methylation behaviours of nineteen genes that may contribute to hepatocellular carcinogenesis. Cell Res 2003;13:319-33.
30.
Bu SY, Mashek DG. Hepatic long-chain acyl-CoA synthetase 5 mediates fatty acid channeling between anabolic and catabolic pathways. J Lipid Res 2010;51:3270-80.
31.
Christensen KE, Mackenzie RE. Mitochondrial methylenetetrahydrofolate dehydrogenase, methenyltetrahydrofolate cyclohydrolase, and formyltetrahydrofolate synthetases. Vitam Horm 2008;79:393-410.
32.
Selcuklu SD, Donoghue MT, Rehmet K, et al. MicroRNA-9 Inhibition of Cell Proliferation and Identification of Novel miR-9 Targets by Transcriptome Profiling in Breast Cancer Cells. J Biol Chem 2012;287:29516-28.
33.
Tryndyak VP, Han T, Muskhelishvili L, et al. Coupling global methylation and gene expression profiles reveal key pathophysiological events in liver injury induced by a methyl-deficient diet. Mol Nutr Food Res 2011;55:411-8.
34.
Cano A, Buque X, Martinez-Una M, et al. Methionine adenosyltransferase 1A gene deletion disrupts hepatic very low-density lipoprotein assembly in mice. Hepatology 2011;54:1975-86.
35.
Frau M, Tomasi ML, Simile MM, et al. Role of transcriptional and posttranscriptional regulation of methionine adenosyltransferases in liver cancer progression. Hepatology 2012;56:165-75.
36.
Tomasi ML, Li TW, Li M, et al. Inhibition of human methionine adenosyltransferase 1A transcription by coding region methylation. J Cell Physiol 2012;227:1583-91.
37.
Kisliuk RL. Folate biochemistry in relation to antifolate selectivity. In: Jackman AL, ed. Antifolate Drugs in Cancer Therapy. Totowa: Humana Press, 1999:13-36.
38.
Oleinik NV, Krupenko NI, Krupenko SA. Epigenetic silencing of ALDH1L1, a metabolic regulator of cellular proliferation, in cancers. Genes Cancer 2011;2:130-9.
39.
Chen XQ, He JR, Wang HY. Decreased expression of ALDH1L1 is associated with a poor prognosis in hepatocellular carcinoma. Med Oncol 2011;29:1843-1849.
40.
Dixon LJ, Berk M, Thapaliya S, et al. Caspase-1-mediated regulation of fibrogenesis in diet-induced steatohepatitis. Laboratory investigation; a journal of technical methods and pathology 2012;92:713-23.
41.
Witek RP, Stone WC, Karaca FG, et al. Pan-caspase inhibitor VX-166 reduces fibrosis in an animal model of nonalcoholic steatohepatitis. Hepatology 2009;50:1421-30.
AC C
EP
TE D
M AN U
SC
RI PT
28.
ACCEPTED MANUSCRIPT
Lu SC, Tsukamoto H, Mato JM. Role of abnormal methionine metabolism in alcoholic liver injury. Alcohol 2002;27:155-62.
43.
Nishida N, Kudo M, Nagasaka T, et al. Characteristic patterns of altered DNA methylation predict emergence of human hepatocellular carcinoma. Hepatology 2012;56:994-1003.
44.
Russmann S, Junker E, Lauterburg BH. Remethylation and transsulfuration of methionine in cirrhosis: studies with L-[H3-methyl-1-C]methionine. Hepatology 2002;36:1190-6.
45.
Cordero P, Gomez-Uriz AM, Campion J, et al. Dietary supplementation with methyl donors reduces fatty liver and modifies the fatty acid synthase DNA methylation profile in rats fed an obesogenic diet. Genes Nutr 2012;8:105-113.
46.
Waterland RA, Jirtle RL. Transposable elements: targets for early nutritional effects on epigenetic gene regulation. Mol Cell Biol 2003;23:5293-300.
M AN U
SC
RI PT
42.
AC C
EP
TE D
Author names in bold designate shared co-first authorship.
ACCEPTED MANUSCRIPT
Table 1. Characteristics of the NAFLD methylation discovery and replication cohorts.
Laboratory measures, median (IQR) c Serum AST, U/L c Serum ALT, U/L c AST/ALT
33 (100) 0 (0)
21 (91.3) 2 (8.7)
32 (97) 21 (91.3) 1 (3) 2 (8.7) 32.8 (28.4 – 40.4) 33.8 (31.3 – 41.9) 14 (42.4) 15 (65.2) 6 (18.2) 9 (39.1) 5.8 (5.5 - 6.4) 6 (5.8 - 6.9) 7 (21.2) 4 (17.4) 6 (18.2) 5 (22.7)
42 (24 - 57) 47.5 (30 - 75) .8 (.64 - 1.03)
62 (46 - 85) 74.5 (46 - 101) .92 (.66 - 1.24)
14 (45.2) 6 (19.4) 3 (10.7) 17 (56.7) 7 (23.3)
14 (60.9) 9 (42.9) 12 (54.6) 22 (95.7) 12 (57.1)
.93 .09 .08 .34 .72 .72
.06 .45 .55
.25 .07 .0014 .0015 .014
EP
AC C
a
10 (52.6) 50.4 (3.2)
8 (53.3) 50.9 (3.1)
17 (89.5) 2 (10.5)
11 (73.3) 4 (26.7)
.56
Histologic characteristics, n (%) Steatosis (% > 34%) c Lobular inflammation (% > grade 2) b Portal inflammation (% > mild) d Ballooning (% any) e* NAFLD Activity Score (% ≥ 5)
.24 .97 .16
Comparison P .97 .74 .37
RI PT
5 (21.7) 51.7 (10.3)
Replication Cohort (n=34)
Controls (n=25)
18 (52.9) 50.6 (3.2)
13 (52) 50.4 (13.8)
28 (82.4) 6 (17.6)
13 (52) 11 (44)
.35
15 (78.9) 4 (21.1) 32 (30.5 – 33.6) 13 (68.4) 6 (31.6) 5.8 (5.6 – 6.2) 10 (52.6) g 1 (2.0)
14 (93.3) 1 (6.7) 32.8 (31 – 34.5) 13 (86.7) 7 (46.7) 5.7 (5.4 – 8.2) 8 (53.3) h 2 (33.3)
38 (30 - 58) 56 (38 – 99) .72 (.59 – .79)
10 (52.6) 1 (5.3) 6 (31.6) 16 (84.2) 5 (26.3)
*NAFLD Activity Score (range 0-8) is a sum of scores for steatosis, lobular inflammation, and ballooning. a missing data for 2 patients in discovery cohort b missing data for 6 patients in discovery cohort c missing data for 4 patients in discovery cohort d missing data for 3 patients in discovery cohort e missing data for 5 patients in discovery cohort f missing data for 12 patients in discovery cohort and 2 in replication cohort g missing data for 14 patients h missing data for 9 patients i missing data for 20 patients
P .94 .57 .03
.21
.38 .27 .81 .7 .96 1.0
29 (85.3) 5 (14.7) 32.3 (30.6 – 33.6) 26 (76.5) 13 (38.2) 5.8 (5.6 – 6.4) 18 (52.9) 3 (27)
18 (72) 7 (28) 38.9 (28.1 – 48.4) 15 (60) 7 (28) 5.7 (5.4 – 7) 9 (36) i 2 (4)
.06 .17 .41 .79 .19 1.0
61 (38 – 87) 61 (51 – 99) .84 (.66 – 1.15)
.07 .53 .07
46.5 (33 – 78) 59 (49 – 99) .77 (.62 – .88)
29 (25 – 34) 32 (22 – 39) .97 (.85 – 1.13)
<.001 <.001 .001
9 (60) 1 (6.7) 9 (60) 13 (86.7) 11 (73.3)
.67 .86 .1 .84 .02
19 (55.9) 2 (5.9) 15 (44.1) 29 (85.3) 16 (47)
-
-
SC
Hypertension, n (%) Diabetes Mellitus, n (%) f Hemoglobin A1c, median % (IQR) Hyperlipidemia, n (%) a Current Smoking, n (%)
12 (36.4) 51.5 (10.3)
P
Replication Cohort Mild NAFLD Advanced F0-F1 NAFLD F3-4 (n=19) (n=15)
M AN U
Male sex, n (%) Age in years at biopsy, mean (SD) Ethnicity, n (%) Non-Hispanic Unknown Race, n (%) White Black a BMI, median kg/m2 (IQR)
Discovery Cohort Mild NAFLD Advanced F0-F1 NAFLD F3-4 (n=33) (n=23)
TE D
Patient demographics and clinical characteristics
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT
AC C
EP
TE
D
M AN U
SC
RI PT
ACCEPTED MANUSCRIPT