Fibrinogen production is enhanced in an in-vitro model of non-alcoholic fatty liver disease: an isolated risk factor for cardiovascular events?

Cell culture, treatment and sample collection

Hepatoblastoma C3A cells (ATCC® CRL-10741TM, LGC Standards, Teddington, UK) were cultured
as previously described 16]. Briefly, cell cultures were treated either with the combination of lactate, pyruvate,
octanoate and ammonia (LPON), oleate (OLE), or untreated controls. Both octanoate
and oleate readily diffuse into mitochondria to promote efficient ß-oxidation and
lipid accumulation, but while OLE treatment causes simple cellular steatosis, LPON
treatment induces both ROS formation and mitochondrial dysfunction, in addition to
steatosis, typically seen in NAFLD 16]. C3A cells were treated in three separate experiments either with LPON, OLE, or untreated
controls and processed for transcriptomic or proteomic analysis as described in the
following sections.

Sample preparation and transcriptomics

Cells were washed twice in cold PBS and transferred to cold RNALater® (Life Technologies,
Paisley, UK) for overnight incubation at 4 °C. Afterwards, RNA was isolated with an
RNAqueous®-4PCR kit (Life Technologies) and subsequently amplified and biotinylated
with an Illumina® TotalPrep RNA Amplification kit (Life Technologies), following the
manufacturer’s instructions. RNA expression was measured by hybridization to the Illumina®
Whole Human Genome BeadChip H12 Microarray (Illumina United Kingdom, Essex, UK). Data
were extracted through the GCOS software (Affymetrix UK Ltd., High Wycombe, UK). CELfiles
were used for additional data processing and imported into Bioconductor 17] to examine differences between LPON- and OLE-treated groups and untreated controls.
Data were normalized by robust multi-array average (RMA) in the Oligo module (http://www.bioconductor.org/packages/2.0/bioc/html/oligo.html). Gene ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment
analysis was performed with the DAVID tool 18], 19] on genes that were significantly differentially expressed. Data was statistically
analyzed with the bioconductor Limma package 20].

Sample preparation and proteomics

Protein extraction was performed as previously described 21]. Briefly, samples were denatured in 8 M urea, reduced by incubating with dithiothreithol
prior to cysteine alkylation with iodoacetamide and overnight digestion with 8 ?g
trypsin. Protein concentrations were estimated by Bradford protein assay (Thermo Scientific,
Rockford, IL, USA) on a 10 ?l sample, diluted to 2 M Urea and quantified against a
BSA standard curve. 4 ?g peptide samples were acidified (1 % formic acid), centrifuged
and cleaned using Stagetips 22], dried by SpeedVac, and stored at ?20 °C.

2 ?g peptide samples were analysed in a randomised sequence by capillary-HPLC- MSMS
as described previously 23], using an on-line system consisting of a micro-pump (1200 binary HPLC system, Agilent,
UK) coupled to a hybrid LTQ-Orbitrap XL instrument (Thermo-Fisher, Leicestershire,
UK). Acetonitrile and water were HPLC quality (Fisher). Formic acid was Suprapure
98-100 % (Merck, Darmstadt, Germany) and trifluoroacetic acid was 99 % purity sequencing
grade. LC-MSMS label-free quantification was performed using Progenesis 4.0 (Nonlinear
Dynamics, Newcastle upon Tyne, UK) as described previously 24]. Multicharged (2+,3+,4+) ion intensities were extracted from LC-MS files and MSMSdata
were searched using Mascot Version 2.4 (Matrix Science Ltd, London, UK) against the
Homo Sapiens subset of the NCBI protein database (12/01/2011; 34,281 sequences) using
a maximum missed-cut value of 2, variable oxidation (M), N-terminal protein acetylation
and fixed carbamidomethylation (C); precursor mass tolerance was 7 ppm and MSMS tolerance
0.4 Da. The significance threshold (p) was??0.05 (MudPIT scoring). A minimum peptide
cut off score of 20 was set, corresponding to 3 % global false discovery rate (FDR)
using a decoy database search. Proteins identified and quantified with two or more
peptide sequences were retained. A two-tailed t-test for independent samples or biological
triplicates was performed on arcsinh transformed, normally distributed intensity data.

Data mining and candidate gene identification

Changes in hepatoblastoma C3A gene transcription and protein expression in response
to cellular steatosis induced by nutrient overload, compared with OLE-treatment and
untreated controls, were analysed in a custom built bioinformatics data mining tool
established by the BHF Centre of Research Excellence Bioinformatics Team at the Queen’s
Medical Research Institute at the University of Edinburgh. This allows evaluation
of integrated transcriptomic and proteomic data, by matching differential gene transcription
with altered protein abundance, to identify gene products potentially involved in
pathogenic pathways. Genes of interest were analysed in a stepwise approach outlined
in Fig. 1 to identify genes showing a consistently greater than 2.0 fold increase in both gene
transcription and protein abundance. To focus on the specific effects of steatosis
with ROS formation and mitochondrial dysfunction and to eliminate candidates up-regulated
by simple steatosis, genes showing similar??2.0 fold increases in expression in both
OLE- and LPON-treated cells compared to untreated controls were excluded from the
analysis.

Fig. 1. Strategy for data mining and candidate gene identification. Step 1: C3A cell cultures
treated with LPON, OLE or untreated controls, as described under ‘Methods’, were subject
to transcriptomic or proteomic analysis and data screened for candidate genes specifically
up-regulated by nutrient overload using the screening criteria outlined in Box 1.
Step 2: Predicted functional partners of primary candidates showing 2.0 fold nutrient
overload-induced increases in gene transcripts and protein abundance were identified
by STRING network associations. Step 3: Primary gene candidates, identified in step
1, and their predicted functional partners, identified in step 2, were grouped by
enrichment analysis according to their functional annotations to identify representation
of common biological processes among the candidate genes. Further details of these
bioinformatics procedures are described under ‘Methods’.

STRING Network Associations

Predicted functional partners of candidate genes induced specifically in response
to nutrient overload were identified using the Search Tool for the Retrieval of Interacting
Genes/Proteins (STRING) v9.1 database (http://string-db.org). Only interactions in Homo sapiens with a probabilistic confidence score ?0.900,
corresponding to a “highest-confidence” network, were considered in this study. STRING
PFPs were cross-validated against the original transcriptomic and proteomic data for
differential expression and protein abundance, using similar criteria (2.0 fold increase
in LPON-treated cells, excluding PFPs with 2.0 fold increase in OLE-treated cells:
see Fig. 1) in order to identify candidate genes that had otherwise been excluded by the more
stringent primary search strategy.

Enrichment analysis

To identify dysregulated pathways and biological processes contributing potentially
to a pathogenic phenotype, candidate genes differentially expressed in response to
nutrient overload and their PFPs were grouped according to their functional annotations
with data from the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes
(KEGG) database via the Gene Ontology enRIchment anaLysis and visuaLizAtion (GOrilla)
tool (http://cbl-gorilla.cs.technion.ac.il) 25], 26]. Only dysregulations with p??0.05 were considered in this study, with p-values being
corrected for multiple testing using the Benjamini and Hochberg method 27]. Enrichment was based on gene ranking, which was indicated by the STRING analysis
confidence score.