Abstract

   The hepatopulmonary syndrome (HPS) is defined by the presence of
   pulmonary gas exchange abnormalities due to intrapulmonary vascular
   dilatations in patients with chronic liver disease. Changes in DNA
   methylation reflect the genomic variation. Since liver transplant (LT)
   reverts HPS we hypothesized that it may be associated with specific
   liver epigenetic changes. Thus, the aim of this study was to
   investigate the role of the liver epigenome in patients with HPS. We
   extracted DNA from paraffin embedded liver tissue samples from 10
   patients with HPS and 10 age-, sex- and MELD (Model for End-stage Liver
   Disease)-matched controls. DNA methylation was determined using the
   850K array (Illumina). Weighted Gene Co-expression Network Analysis
   (WGCNA) was used to identify modules related to defining physiologic
   characteristics of HPS. Only 12 out of the 20 liver biopsies (7 HPS and
   5 controls) had sufficient quality to be analyzed. None of the 802,688
   DNA probes analyzed in the case control comparison achieved a
   significant False Discovery Rate (FDR). WGCNA identified 5
   co-methylated gene-modules associated to HPS markers, mainly related to
   nervous and neuroendocrine system, apoptotic processes, gut bacterial
   translocation, angiogenesis and vascular remodeling ontologies. To
   conclude, HPS is associated with nervous/neuroendocrine system and
   vascular remodeling related liver epigenetic changes.

Introduction

   The hepatopulmonary syndrome (HPS) is defined by the presence of an
   arterial oxygenation defect caused by intrapulmonary vascular
   dilatations in patients with chronic liver disease, mostly liver
   cirrhosis (LC) [[36]1–[37]5]. The prevalence of the HPS in patients
   assessed for liver transplant (LT) ranges between 5–30% [[38]6–[39]8].
   Thirty years ago, we were the first to show that patients with LC
   [[40]9–[41]11] exhibit reduced or abolished pulmonary vascular tone,
   also distinctive of HPS. More recent studies have proposed a number of
   putative biological mechanisms [[42]12–[43]17], but the pathobiology of
   HPS still remains elusive to date. The progression of chronic liver
   damage to cirrhosis and HPS is highly variable between patients and the
   course disease is often clinically unpredictable [[44]18]. It is well
   established that there are several non-genetic factors that influence
   the progression of other chronic liver diseases, including diet,
   exercise, microbiome and alcohol consumption [[45]15]. These
   non-genetic factors have been associated with the modulation of organ
   functionality through changes in the epigenome [[46]18–[47]20]. DNA
   methylation is one of the main epigenetic mechanisms that regulates
   gene transcription. It involves the addition of a methyl group to a
   cytosine base adjacent to a guanine base (CpG site) in regulatory
   regions of DNA, and it is associated with decreased gene expression
   [[48]21]. We hypothesized that HPS could be related to specific
   epigenetic liver changes, as recently described in non-alcoholic fatty
   liver disease [[49]22], obesity and diabetes [[50]23]. Accordingly,
   this study seeks to: (1) compare the liver epigenome of patients
   undergoing LT, with and without HPS; and, (2) relate epigenome
   differences to the defining pathophysiological markers of HPS.

Material and methods

   The ethics committee of the Hospital Clinic de Barcelona reviewed and
   approved the study (HCB/2014/1045) and written consent was required.

Study design and ethics

   This is a controlled study performed in paraffin embedded liver
   biopsies obtained from patients with LC (with and without HPS) who
   required LT from 1996 to 2014 in our hospital. Samples were stored in
   the Biobank of our institution. Although the number of samples is
   limited, they are representative of the population of patients with LC
   with and without HPS as all the patients requiring LT in our hospital
   during 18 years have been included. For this study patients were
   recruited thought the revision of medical records of our hospital. All
   patients signed their informed consent form before LT. We confirm that
   all methods were carried out in accordance with relevant guidelines and
   regulations, and all experimental protocols were approved by the ethics
   committee of the Hospital Clinic (HCB/2014/1045).

Patients

   We identified 10 liver biopsies from transplanted LC patients with
   moderate-to-severe HPS (PaO[2] < 70 mmHg) [[51]2, [52]3, [53]5] and 10
   biopsies from transplanted LC patients with normal lung function and no
   HPS, matched for sex, age and severity of liver disease (MELD)
   [[54]24]. Patients with liver carcinoma and/or primary biliary
   cirrhosis were excluded.

Epigenetic analysis

DNA extraction

   Total DNA was extracted from paraffin preserved liver samples using the
   RecoverAll™ Total Nucleic Acid Isolation Kit for FFPE (Ambion, Life
   Technologies, US) following manufacturer instructions. DNA quantity was
   determined by Nanodrop (Thermo Scientific, Willmington, US) and quality
   by Qubit (Life Technologies, US). Before running the arrays, the
   suitability of DNA for methylation analysis was assessed using FFPE QC
   kit (Illumina, San Diego, US), as recommended by Illumina. Bisulfite
   conversion was performed with the EZ-96 DNA Methylation Kit (Zymo
   research, CA, US) following manufacturer instructions. The Infinium
   MethylationEPIC Beadchip [[55]25] (Illumina), which covers over 850,000
   CpG sites of the human genome, was used to assess the methylation
   levels at Hospital de la Fe, Valencia, Spain.

DNA methylation analysis

   The Illumina EPIC Array was used to determine the DNA methylation
   status of >800.000 CpG sites in liver samples, and the minfi package
   [[56]26] was used for quality check, normalization and processing of
   raw data. The minfi package also allowed us to identify methylated
   probes located in autosomal chromosomes, excluding those located in
   single nucleotide polymorphisms (SNP) positions. The Limma package (R
   platform) was then used on the filtered matrix of M values to calculate
   the differential methylation adjusting for age and sex in the linear
   model. CpGs were assigned to the nearest gene, so that probes with no
   gene were excluded from further analysis. Likewise, the R package
   GeneAnswers with the L5 of GeneOntolgy [[57]27] was used to identify
   functional enrichment in the CpGs differentially methylated in patients
   with and without HPS; significant enrichment was defined as a False
   Discovery Rate (FDR) <0.05. Finally, REVIGO [[58]28] was used to
   evaluate term overlap and to obtain a visual summary of functional
   enrichment results. Data is available from the GEO repository database
   ([59]GSE162984).

Weighted Gene Co-expression Network Analysis (WGCNA)

   To identify modules of co-methylated genes related to functional and
   clinical characteristics, the probes with nominal p value <0.05 in the
   case/control analysis were used to build the co-methylation network
   using the WGCNA R package [[60]27, [61]29]. To do so, probes were
   collapsed to genes. In the modules of interest, functional enrichment
   was calculated with GeneAnswers [[62]27], considering significant
   enrichment as FDR <0.05.

Statistical analysis

   Results are presented as number (proportion) or median [interquartile
   range]. Continuous and categorical variables were compared between
   patients and controls using Mann-Whitney test and Fisher’s Exact test,
   respectively. All statistics were computed with R.

Results

Patient characteristics

   Although we originally included in the study liver biopsies from 10
   patients with and 10 without HPS, only 12 of them (7 with and 5 without
   HPS) passed the minfi quality control of the 850K (EPIC) methylation
   array results (S1 Table in [63]S1 File) and could therefore be included
   in the analysis. [64]Table 1 compares the demographic and main clinical
   characteristics of these 12 patients. There were no significant
   differences between groups with respect to age, sex, etiology or
   severity of cirrhosis. Alcohol abuse and hepatitis C virus infection
   were the most frequent causes of LC. Spirometry was within normal
   limits in both groups but, by design, pulmonary gas exchange was
   severely impaired in patients with HPS but within the normal range in
   those without HPS (AaPO[2], 60±13 vs 8±5 mmHg; PaO[2], 54±6 vs 102±10
   mmHg; DL[CO], 54±19 vs 80±10% predicted, respectively; p<0.05 each).
   Liver biopsies stained with hematoxylin-eosin were blindly reviewed by
   an expert liver pathologist (MB). Neither the type of LC, inflammatory
   activity, presence of vascular proliferation within septa, sinusoidal
   dilatation, necrotic areas, cholangiolar proliferation, steatosis or
   cholestasis were different between patients with and without HPS (S2
   Table in [65]S1 File).

Table 1. Demographic and clinical characteristics of patients with LC without
and with HPS (mean ±SD).

   Without HPS With HPS p values
   (n = 5) (n = 7)
   Age, years old 53±7 52±9 NS
   Sex (female/male) 2/3 3/4 NS
   Body mass index, kg/m^2 26±3 26±5 NS
   Etiology Viral = 2 Viral = 3 NS
   Alcohol = 2 Alcohol = 3
   Viral+Alcohol = 1 Viral+Alcohol = 1
   CHILD-PLUG classification A, 0 A, 3 NS
   B, 4 B, 2
   C, 1 C, 2
   MELD Score 15±2 15±5 NS
   Smoker, n (packs-year) 4 (12±9) 5 (28±19) NS
   Forced Vital Capacity (FVC, % ref.) 100±11 95±12 NS
   Forced Expiratory Volume in 1 sec. (FEV[1], % ref) 100±6 97±8 NS
   FEV[1]/FVC (%) 0.79±0.06 0.81±0.07 NS
   DL[CO], (% ref) 80±10 54±19 <0.05
   AaPO[2], mmHg 8±5 60±13 <0.01
   PaO[2], mmHg 102±10 54±6 <0.01
   PaCO[2], mmHg 34±5 30±6 NS
   [66]Open in a new tab

   MELD, Model for End-stage Liver Disease; FVC, forced vital capacity;
   FEV[1], forced expiratory volume in 1 s; DL[CO], single breath lung
   diffusing capacity for carbon monoxide; AaPO[2], alveolar-arterial
   oxygen gradient; PaO[2], partial pressure of arterial oxygen; PaO[2],
   partial pressure of arterial carbon dioxide; NS, not significant.

DNA methylation analysis

   None of the 802,688 DNA probes contrasted between patients with and
   without HPS achieved a significant False Discovery Rate (FDR) lower
   than 0.05, likely in relation to the small sample size available (7 HPS
   vs. 5 non-HPS; n = 12). Yet, to explore potential differences between
   cases and controls, we decided to use less stringent statistical
   criteria (nominal p<0.05, not FDR corrected). Accordingly, our results
   have to be considered exploratory. By doing so, we identified 48,386
   differentially methylated probes between patients with and without HPS
   ([67]Fig 1).

Fig 1. Manhattan plot of differentially methylated genes in LC with and
without HPS.

   [68]Fig 1
   [69]Open in a new tab

   P values (-log10[p]) are plotted against their respective positions on
   each chromosome. Green dots indicate the 465 positions differentially
   methylated with p values <0.0009, FC>|0,4| and associated to a single
   gene. Any positions which do not follow these criteria are represented
   in black dots.

   S1 Table (in [70]S1 File) provides the list of the probes with the
   lowest p values (p<0.0009) and highest absolute fold change (FC>|0.4|),
   which represented the most variable probes among study groups (n =
   443). [71]Table 2 lists the top differentially methylated genes with a
   positive (i.e., more methylated in controls) or negative (i.e., more
   methylated in patients with HPS) FC, ordered by their respective p
   values.

Table 2. Top differentially methylated genes with a positive or negative Fold
Change (FC) ordered by |FC|.

   Positive FC: higher methylation in patients without HPS Negative FC:
   higher methylation in patients with HPS
   Gene Probe ID FC p values Gene Probe ID FC p values
   ORMDL3 cg09155575 1.21 6.47E-05 4 GBP cg00273257 -1.16 1.50E-05
   VAV3 cg19103456 0.99 2.44E-05 HIVEP2 cg05786041 -1.12 8.14E-05
   MYT1 cg11839596 0.94 4.75E-05 ANKRD28 cg22424370 -1.12 8.14E-05
   PAX8 cg23156509 0.91 5.57E-05 FAM168A cg16322681 -1.10 1.63E-05
   CNTNAP2 cg16508202 0.89 7.93E-06 SCAMP3 cg24034289 -1.05 9.95E-05
   CPNE1 cg15917394 0.87 9.20E-05 SPEN cg08892464 -1.03 6.32E-05
   TBC1D2B cg07690703 0.80 8.34E-05 VTI1A cg00592484 -1.01 5.20E-05
   C10orf142 cg19100384 0.79 4.44E-05 XPO1 cg26131881 -0.97 7.15E-06
   IL15 cg25690285 0.76 4.88E-05 RCHY1 cg13431998 -0.96 7.63E-05
   CRAMP1L cg15231222 0.75 5.09E-05 TMEM132D cg22386460 -0.87 5.15E-05
   GUCY2EP cg16855265 0.72 8.74E-05 DCAF5 cg16121759 -0.84 5.34E-05
   MYO18A cg02376256 0.70 4.31E-05 TCEA3 cg18167160 -0.78 1.72E-05
   ANKRD11 cg04200026 0.66 9.90E-05 KRAS cg18249781 -0.77 9.35E-05
   TTC39A cg24797045 0.64 3.46E-05 ANO10 cg19647107 -0.76 7.41E-05
   LAMA2 cg10671433 -0.63 6.68E-05
   [72]Open in a new tab

   To identify biological processes associated with these 443 most
   differently methylated probes, we computed the gene ontology (GO) and
   KEGG pathway enrichment analysis using the GeneAnswers R package. We
   identified 44 functionally GO enriched terms with a FDR<0.05 (S3 Table
   in [73]S1 File, summarized in [74]Fig 2A) which were related to neuron
   projection morphogenesis, regulation of synaptic vesicle clustering,
   peptidyl-lysine deacetylation, chemical synaptic transmission,
   phosphorylation and positive regulation of GTPase activity ([75]Fig
   2A).

Fig 2.

   [76]Fig 2
   [77]Open in a new tab

   Visual summary of GO terms (Treemap in REVIGO(46)) for: (A) 443 genes
   total list p<0.0009, (B) Negative FC (FC >0.4, FDR <0.05), (C) Brown
   WGCNA module, (D) Blue WGCNA module, (E) Darkgrey WGCNA module. Each
   box is a single ontology cluster, which are joined into super-clusters
   of loosely related terms, shown by the same color and named with a
   representative term. Larger boxes mean more significant enrichment.
   Letter C denotes control patients (without HPS).

   We then computed the Gene Ontology (GO) enrichment including only those
   genes with a positive FC (250) or those with a negative FC (188). This
   analysis identified three GO enriched terms in the former and 17 in the
   later (FDR <0.05). The 3 ontologies associated with a positive FC were
   related to pepdidyl-lysine deacetylation, histone H3 deacetylation and
   positive regulation of hydrolase activity (S4 Table in [78]S1 File).
   The 17 ontologies associated with a negative FC were related to cell
   morphogenesis involved in two main processes: neuron differentiation,
   chemical synaptic transmission and trans-synaptic signaling (S4 Table
   in [79]S1 File, summarized in [80]Fig 2B).

   Finally, we investigated if any of the genes listed in [81]Table 2
   (i.e., those with higher magnitude of change) were included in any of
   the gene ontologies identified for positive and negative FC, separately
   (S3 and S4 Tables in [82]S1 File), and found that no genes were
   included in the three ontologies more methylated in patients non-HPS
   patients (positive FC). However, two genes (LAMA2 and KRAS) were
   included in 7 of the 18 ontologies found to be more methylated in HPS
   patients (i.e., negative FC). These 7 ontologies were associated to
   cell morphogenesis involved in neuron differentiation, axonogenesis,
   neuron projection morphogenesis, gliogenesis and plasma membrane
   bounded cell projection morphogenesis (S4 Table in [83]S1 File).

Weighted Gene Co-expression Network Analysis (WGCNA)

   To identify groups of co-methylated genes related to clinical
   characteristics of interest, we performed WGCNA analysis with probes
   having a nominal p value <0.05. WGCNA identified 18 modules (labeled by
   a color name), each containing a minimum of 30 co-methylated genes
   ([84]S1A Fig).

   [85]Fig 3 illustrates a heat-map representing the potential
   relationship (positive [red], none [white] or negative [blue]) between
   these 18 modules of co-methylated genes and a number of relevant
   clinical characteristics (bottom axis). Differences were very clear
   between patients with and without HPS (so-called “groups”). No module
   was related to sex, body mass index, etiology of LC or spirometric
   variables (FEV[1], FVC or FEV[1]/FVC). By contrast, five modules (28%;
   blue [8947 genes], brown [3037 genes], cyan [128 genes], darkgrey [64
   genes] and salmon [131 genes]) were differentially associated with the
   four pulmonary gas exchange descriptors that best characterize HPS
   (high AaPO[2], low PaO[2], reduced DLco and positive contrast-enhanced
   echocardiography [ECHO]) ([86]Fig 3). Three out of these 5 modules
   (brown, blue and darkgrey) presented relevant biological processes and
   molecular function ontologies enriched with a FDR <0.05 (S4 Table in
   [87]S1 File, summarized in [88]Fig 2). The brown module (S5 Table in
   [89]S1 File, summarized in [90]Fig 2C) included ontologies related to
   cellular response to lipids, salivary gland morphogenesis, regulation
   of serine/threonine kinase activity, negative regulation of apoptotic
   process, regulation of establishment of protein localization to
   mitochondrion, negative regulation of cell size. The later included
   phosphorylation, epithelial cell differentiation and apoptotic process.
   The blue module (S5 Table in [91]S1 File, summarized in [92]Fig 2D)
   included ontologies related to regulation of neuron differentiation,
   dephosphorylation, chemical synaptic transmission, metal ion transport,
   positive regulation of GTPase activity, sphingolipid biosynthesis,
   cilium assembly, cellular hypotonic response. Interestingly, both the
   brown and blue modules included ontologies related to G1/S transition
   of mitotic cell cycle. Finally, the darkgrey module ([93]Fig 2E)
   included ontologies related to regulation of cellular macromolecule
   biosynthetic processes, phosphatidylinositol-mediated signaling, miRNA
   mediated inhibition of translation, regulation of estradiol secretion
   and microtubule sliding.

Fig 3. Heatmap of DNA methylation data based on unsupervised hierarchical
clustering that shows the correlation between the modules identified by WCGNA
and a number of clinically relevant variables.

   [94]Fig 3
   [95]Open in a new tab

   Red and blue cells indicate a positive or negative correlation
   coefficient (see scale). Only framed modules were further analyzed.
   Abbreviations: group: patients with or without HPS; ECHO,
   contrast-enhanced echocardiography; AaPO[2], alveolar-arterial oxygen
   gradient (mmHg); cumulative smoking exposure (pack-years); PaCO[2],
   partial pressure of arterial carbon dioxide (mmHg); DLco, single-breath
   diffusing capacity for carbon monoxide (% reference); PaO[2], partial
   pressure of arterial oxygen (mmHg); BMI, body mass index (kg/m^2); FVC,
   forced vital capacity (% reference); FEV[1], forced expiratory volume
   in the 1^st second (% reference).

Discussion

   The results of this pilot show, for the first time to our knowledge,
   that the key defining characteristics of the HPS, namely pulmonary gas
   exchange abnormalities and presence of intrapulmonary vascular
   dilatations [[96]2], are associated with specific liver epigenome
   changes related to nervous and neuroendocrine system, vascular
   remodeling and angiogenesis gene ontologies.

   We were among the first to define the gas exchange pathophysiologic
   hallmarks of HPS primarily characterized by moderate-to-severe
   ventilation-perfusion imbalance, and reduced oxygen diffusion transfer
   to the alveoli in the presence of intrapulmonary vascular dilatations
   [[97]17]. Subsequent studies proposed a number of putative biological
   mediators, including nitric oxide (NO) [[98]12, [99]13, [100]17] and
   endothelin-1 (ET-1) [[101]16], and identified a few genetic
   polymorphisms related to angiogenesis [[102]15] and inflammation
   [[103]14] in human and experimental HPS. To our knowledge, no previous
   study has explored potential liver epigenetic changes in patients with
   HPS. Likewise, no previous study has used WGCNA to get an integrated
   view of gene co-expression networks and to relate it with specific
   clinical and functional features of HPS. Although WGCNA does not
   provide causal information, it enables the identification of regulatory
   genes underlying various clinical phenotypes of interest [[104]27,
   [105]29].

   It is likely that due to the relatively small number of liver biopsies
   available for analysis (n = 12) in regards with the number of DNA
   probes analyzed (n = 802,688), no DNA probe achieved the minimum FDR
   level of statistical significance ([106]Fig 1); hence, our findings
   herein should be considered exploratory. With this caveat in mind,
   however, some observations reported here may provide novel information
   of potential interest.

   First, pulmonary angiogenesis and vascular remodeling are considered
   key pathogenic features of HPS [[107]3, [108]30], mostly involving the
   activation of vascular endothelial growth factor (VEGF)-dependent
   pathways [[109]14, [110]31–[111]33]. In this context, it is of note
   that some of the top differentially methylated genes identified in
   patients with HPS ([112]Table 2) were also related to angiogenesis,
   including VAV3 [[113]34], PAX 8 [[114]35], K-RAS [[115]36] and LAMA2
   [[116]37].

   Second, we also identified other significant ontologies associated with
   HPS, including neurological processes such as neurogenesis, neuron
   differentiation and gliogenesis. Interestingly, the role of the
   neuroendocrine and autonomic nervous system in the liver function has
   been well described, including glucose metabolism, regulation of blood
   flow and vasculature, hepatic regeneration and fibrosis [[117]38].
   Importantly, several neuropeptides and neurohormones, namely melatonin
   and A-calcitonin gene-related peptide (a-CGRP), have been shown to
   regulate angiogenesis, vasodilatation and inflammatory cytokines
   [[118]39].

   Finally, WGCNA identified five specific modules of co-methylated genes
   that were associated with key defining characteristics of HPS, i.e.
   increased AaPO[2], decreased PaO[2], low DL[CO] and intrapulmonary
   vascular dilations by echocardiography ([119]Fig 3).

   A more detailed analysis identified a number of gene ontologies with
   FDR<0.05 of 3 of these 5 modules ([120]Fig 2), including the regulation
   of cellular response to lipids, endothelium development and
   morphogenesis, regulation of apoptotic process, regulation of cell
   size/cycle, regulation of methylation and translation, and regulation
   of estradiol secretion. All of them are potentially relevant for a
   better understanding of the pathobiology of HPS. First, the regulation
   of the cellular response to lipids relates to the response to bacterial
   lipoproteins which, in turn, may relate to gut bacterial translocation,
   a potentially relevant pathogenic mechanism of HPS [[121]40]. Second,
   epithelial cell differentiation/migration and endothelial development
   ontologies can be related to abnormal angiogenesis, pulmonary vascular
   dilatations, endothelial dysfunction and vascular remodeling [[122]41],
   all key components of the vascular features of HPS. Third, the
   regulation of estradiol secretion includes ontologies related hormone
   secretion, thus neuroendocrine pathways. Interestingly, increased
   levels of progesterone and estradiol have been associated to
   intrapulmonary vascular dilatations and gas exchange abnormalities in
   patients with cirrhosis [[123]42]. In fact, progesterone and estradiol
   levels are increased in patients with HPS [[124]43]. Likewise, it is of
   note that previous studies have related hypothalamic-pituitary-adrenal
   (HPA) axis dysfunction with LC, portal hypertension and HPS [[125]44,
   [126]45]. The HPA axis regulates many liver functions through
   neuroendocrine forward signaling, such as glucogenesis, lipid
   metabolism, growth hormone signaling and inflammation [[127]46].
   Interestingly, these ontologies were also identified by WGCNA in the
   current patients with HPS. Finally, the interaction of proliferation
   and apoptosis of vascular endothelial cells is a pivotal mechanism of
   vascular remodeling [[128]34], and we observed that apoptotic processes
   (GO:0006915) were related to DL[CO] (a functional marker of abnormal
   pulmonary circulation). It is of note that Roberts et al. reported that
   CAV3 and TGFB1 (both included in this apoptotic ontology) were
   associated with HPS [[129]15]. Collectively, therefore, these
   observations support our working hypothesis, namely that HPS is
   associated with distinct liver epigenome changes. This can also
   contribute to explain the well-established clinical observations that,
   on the one hand, not all patients with LC develop HPS, and on the
   other, that HPS is not associated with the severity of LC, as reflected
   by the MELD score [[130]2, [131]47, [132]48]. Future research is needed
   to confirm our findings here and, eventually, to facilitate the
   identification of novel potential therapeutic targets for the HPS.

   The main strength of our study is that it investigates a previously
   unexplored hypothesis using state of the art technology in a small but
   carefully selected identified subset of patients with LC and HPS
   undergoing LT. We acknowledge, however, that the small number of liver
   biopsies (n = 12) of sufficiently high quality available for DNA
   methylation analysis is a limitation of our study. Difficulties in
   recruiting larger numbers of patients with HPS transplanted in our
   hospital for this type of complex translational approaches are similar
   to those noted in previous studies [[133]49–[134]51]. In any case, this
   has unfortunately restricted our ability to reach the required FDR
   value associated with the use of more than 800,000 DNA probes and
   implies that our findings need to be considered hypothesis generating
   and require validation by other studies. Likewise, we did not have
   access to frozen liver samples (which may have allowed us to explore
   transcriptomic differences) or serum samples (to contrast potential
   systemic biomarkers between the two groups of patients studied).
   Nonetheless, we hope that our observations will encourage other
   researchers in the field to confirm and expand our findings herein.

Conclusions

   This pilot study shows that the functional and structural defining
   characteristics of HPS are associated with specific liver epigenome
   changes, which are specially related to hepatic innervation,
   neuro-hormonal signaling, apoptotic processes, gut bacterial
   translocation and vascular remodeling. These findings point to the
   notion that future functional research on the role of methylation
   changes in the liver is needed to further address unanswered aspect.
   This may help to shed further light in the pathophysiological
   mechanisms underlying HPS and better guide its current frustrating
   therapy.

Supporting information

   S1 File

   (DOCX)
   [135]Click here for additional data file.^ (123.2KB, docx)
   S1 Fig

   (A) Network dendrogram of unsupervised hierarchical clustering of the
   co-methylation network. In the cluster each line represents a probe,
   which corresponds to a single gene. Probes under the same branch are
   co-methylated (top). Modules correspond to branches of the dendrogram
   and were assigned colors for visualization (bottom). (B) Number of
   single genes contained in each module.

   (TIF)
   [136]Click here for additional data file.^ (988.8KB, tif)
   S2 Fig

   (A)-(E) Scatterplots between module membership measure (X-axis) and the
   gene significance for DLco of the five selected modules: brown, blue,
   cyan, darkgrey and salmon.

   (TIF)
   [137]Click here for additional data file.^ (682.7KB, tif)

Acknowledgments