Lung Fibroblasts from Patients with Idiopathic Pulmonary Fibrosis Exhibit Genome-Wide Differences in DNA Methylation Compared to Fibroblasts from Nonfibrotic Lung

Excessive fibroproliferation is a central hallmark of idiopathic pulmonary fibrosis (IPF), a chronic, progressive disorder that results in impaired gas exchange and respiratory failure. Fibroblasts are the key effector cells in IPF, and aberrant expression of multiple genes contributes to their excessive fibroproliferative phenotype. DNA methylation changes are critical to the development of many diseases, but the DNA methylome of IPF fibroblasts has never been characterized. Here, we utilized the HumanMethylation 27 array, which assays the DNA methylation level of 27,568 CpG sites across the genome, to compare the DNA methylation patterns of IPF fibroblasts (n = 6) with those of nonfibrotic patient controls (n = 3) and commercially available normal lung fibroblast cell lines (n = 3). We found that multiple CpG sites across the genome are differentially methylated (as defined by P value less than 0.05 and fold change greater than 2) in IPF fibroblasts compared to fibroblasts from nonfibrotic controls. These methylation differences occurred both in genes recognized to be important in fibroproliferation and extracellular matrix generation, as well as in genes not previously recognized to participate in those processes (including organ morphogenesis and potassium ion channels). We used bisulfite sequencing to independently verify DNA methylation differences in 3 genes (CDKN2B, CARD10, and MGMT); these methylation changes corresponded with differences in gene expression at the mRNA and protein level. These differences in DNA methylation were stable throughout multiple cell passages. DNA methylation differences may thus help to explain a proportion of the differences in gene expression previously observed in studies of IPF fibroblasts. Moreover, significant variability in DNA methylation was observed among individual IPF cell lines, suggesting that differences in DNA methylation may contribute to fibroblast heterogeneity among patients with IPF. These results demonstrate that IPF fibroblasts exhibit global differences in DNA methylation that may contribute to the excessive fibroproliferation associated with this disease.


Introduction
Idiopathic pulmonary fibrosis (IPF) is a devastating and progressive scarring disease of the lung of unknown etiology that results in impaired gas exchange, respiratory failure, and eventually death [1]. Occurring predominantly in older individuals, the median survival of IPF patients is only 3-5 years from the time of diagnosis, with no known effective therapies [1,2].
The pathologic lesion in IPF is dominated by the presence of increased extracellular matrix proteins and accumulation of fibroblasts, the key effector cell of fibrosis. Fibroblasts from IPF patients have been shown to exhibit multiple differences in phenotype compared to cells isolated from nonfibrotic lung [3][4][5][6][7], including increased capacity to proliferate [5,6], generate extracellular matrix, and resist apoptosis [3,[8][9][10] and antifibrotic signals [4]; these differences were identified in in vitro studies, and have consistently been observed to be stable across numerous passages. These phenotypic differences in turn are thought to reflect the differential expression of a variety of genes [3,4,7,8,[11][12][13][14]. However, the mechanisms that account for such gene expression differences are unknown.
DNA methylation is a critical means of gene regulation and aberrant DNA methylation is important in the development of many diseases; however, explorations into its potential role in IPF have been limited. Studies utilizing whole tissue have identified global methylation differences between the lungs of IPF patients and individuals without fibrotic lung disease [15,16]. However, since the lung contains up to forty distinct cell types and since such studies cannot localize these methylation differences to individual cell types, the pathobiologic significance of such methylation differences remain uncertain. Regulatory regions of select genes have been shown to be differentially methylated in IPF fibroblasts [17][18][19], but a global analysis of methylation differences in IPF fibroblasts has not been reported.
Here, we compare the global DNA methylation profile between IPF fibroblasts and two groups of controls, and identify differences in DNA methylation at many loci. Selected methylation differences were independently validated by bisulfite sequencing, and were found to functionally contribute to altered gene expression. These methylation differences were found in genes enriched in specific gene ontology classes -some with recognized relevance to fibrosis and others that might be considered unexpected based on our current understanding of fibrogenesis. Our data thus provide the opportunity to identify novel genes that are differentially methylated and differentially expressed, which may contribute to the pathogenesis of IPF.

Methods
Cell culture IPF fibroblasts were outgrown from lung tissue obtained by surgical lung biopsy of patients with IPF, as previously described [20]. All biopsy specimens were confirmed histologically to show the characteristic histopathologic pattern termed usual interstitial pneumonia. Control fibroblasts were cultured from histologically normal regions of lung from age-matched patients undergoing resection for lung nodules. Because these nodules often proved to be cancer, we also used as an additional control commercially available primary fibroblast lines (CCL204, CCL190, and CCL210, all from American Type Culture Collection, Manassas, VA) grown from the lungs of adults with no preexisting lung disease, to rule out the possibility that any observed methylation differences might be the consequence of a ''field effect'' of cancer [21,22]. Patient-related data are shown in Table 1. Independent comparisons were made between IPF and these two nonfibrotic control groups. All fibroblasts were cultured in DMEM (Invitrogen, Carlsbad, California) supplemented with 10% fetal bovine serum (Hyclone, Logan, Utah) and 100 U/ml pencillin/streptomycin at 37uC with 5% CO 2 , and examined between passages 4-6.

DNA methylation analysis
To analyze global DNA methylation patterns, 1 mg of genomic DNA was subjected to bisulfite conversion using the EZ DNA Methylation Kit from Zymo Research (Irvine, California). Bisulfite-converted DNA was analyzed for methylation at 27,578 CpG sites using the Illumina (San Diego, California) Human-Methylation27 BeadChip Array according to the manufacturer's protocol. Signal intensity from methylated and unmethylated probes for all sites was scanned on the Illumina BeadArray Reader, and preprocessed using Illumina GenomeStudio software. The methylation status of individual CpG sites was verified by pyrosequencing. Bisulfite-modified DNA was amplified by PCR using biotin-labeled primers specific for the CDKN2B, MGMT, and CARD10 promoters. The primer sequences for CDKN2B and CARD10 are shown in Table S1; primers for MGMT (assays ASY514FS and FS1) were obtained from EpigenDx (Hopkinton, Massachusetts). The biotinylated PCR product was then bound to beads, washed through the Vacuum Prep Tool (Qiagen, Valencia, California), and mixed with sequence-specific primers before analysis on the Pyrosequencer (Qiagen).

Cell proliferation
Cells were plated at 2610 3 cells/well in 96-well plates and treated with either control siRNA or siRNA targeting CDKN2B (ON-TARGETplus SMARTpool, Thermo Scientific). After 72 h, cell proliferation was assayed by the CyQuant assay (Life Technologies) per manufacturer's protocol.

Data analysis
All methylation array data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) database under accession number GSE56074. For the HumanMethylation27 array data, signal intensities were corrected for red/green color balance, adjusted for background signal, and normalized across the set of arrays. M-values were calculated as the log2 ratio of the intensities of the methylated probe versus unmethylated probe as described by Du et al [24]. M-values were chosen over beta-values for differential methylation analysis because of their reported superior performance [24]. M-values were compared between IPF and both nonfibrotic control groups and statistically significant differences were identified using the limma algorithm in the Bioconductor R suite (http://www. bioconductor.org) (22). Significant differences in M values were defined using thresholds of p-value less than 0.05 and a fold change of greater than 2; this ''fold change ranking with a nonstringent p-value cutoff'' approach has been validated by other microarray studies [25][26][27] and was shown to perform well with M-values [24]. CpG loci with known single nucleotide polymorphism (SNP) annotations were identified from dbSNP 132, as suggested at http://rforge.net/IMA/. Gene enrichment analysis was performed using ConceptGen (http://conceptgen.ncibi.org) with Q,0.05 defined as statistically significant. Network analysis of gene concepts was performed using STITCH (http://stitch. embl.de/). All other data were analyzed on GraphPad Prism 5.0 (GraphPad Prism Software, San Diego, CA) using ANOVA or Student's t-test, as appropriate, with a p,0.05 defined as statistically significant. Data are expressed as mean 6 SEM.

Ethics statement
All patients provided written informed consent and protocols were approved by the University of Michigan Institutional Review Board (HUM00023700).

IPF fibroblasts exhibit global differences in DNA methylation compared to nonfibrotic control cells
Fibroblasts from the lungs of 6 IPF patients and 3 age-matched control patients without fibrotic lung disease were cultured and their DNA methylation profiles were analyzed on the Human-Methylation27 Bead-Chip array. The demographic characteristics of the patients are shown in Table 1. A second control group consisted of the commercial cell lines CCL190, CCL204, and CCL210, which are primary fibroblast lines grown from the lungs of individuals without lung cancer or fibrotic lung disease. Using the significance threshold criteria described in Methods, we identified 787 CpG loci that were differentially (either increased or decreased) methylated (listed in GEO database accession GSE56074) in IPF versus patient-derived control cells. The top 50 differentially methylated CpG loci, based on p-value and sorted by fold change, are shown in Table S2. When IPF cells were compared to the three commercial control cell lines, using the same selection criteria, 333 CpG loci were differentially methylated (GEO accession GSE56074). The top 50 CpG loci, based on p-value and sorted by fold change, are shown in Table S3. Of these two groups of differentially methylated CpG loci, 125 were identified to be shared (listed in Table S4), of which 72 were identified as concurrently hypomethylated and 45 were identified as concurrently hypermethylated in IPF cells when compared to either patient-derived or commercial cell line control cells ( Figure 1A). Table 2 lists those differentially methylated CpG loci, and their associated genes, that were found in common in the lists of 50 most-differentially methylated CpG loci between the two control groups.
Using the database of SNPs dbSNP 132, 3,690 of 27,578 CpG loci (13.4%) on the HumanMethylation27 array are annotated for known SNPs. Among the 787 CpG loci that were differentially methylated between IPF and patient-derived control cells, 116 (14.7%) were annotated for SNPs. Among the 333 CpG loci that were differentially methylated between IPF and commercial cell lines, 46 (13.8%) were annotated for SNPs. Since the proportion of annotated SNPs in the dataset is similar to the proportion of SNPs annotated over the entire array, the data do not suggest that SNPs represent a bias in the identification of differentially methylated CpG sites (Chi-Square p-values 0.29 and 0.82 for the two different control comparisons). CpG sites that have annotated SNPs are indicated in Tables 2, S2, S3, and S4.
The HumanMethylation27 array was designed to be biased towards analyzing CpG sites that are near gene promoters and located within CpG islands (72.5% of probes). Despite this bias, the majority (53%) of differentially methylated CpG fell outside of CpG islands ( Figure 1B).
To independently verify differences in DNA methylation between IPF and control cells, we performed bisulfite sequencing of three specific genes -CDKN2B, MGMT, and CARD10that were identified by the array as differentially methylated in IPF fibroblasts. CDKN2B and CARD10 were chosen for validation because they were the top two genes identified by the array as being hypermethylated in IPF cells compared to both control groups ( Table 2). MGMT was chosen because it was a hypomethylated gene in IPF cells that we had previously shown  was hypermethylated following treatment with an antifibrotic mediator, prostaglandin E 2 [28], whose endogenous biosynthesis is diminished in IPF [14]. Methylation analysis of 25 CpG sites within the CDKN2B gene, which included the site interrogated by the array, revealed that IPF fibroblasts were hypermethylated at nearly every CpG site sequenced ( Figure 2A). In contrast, the MGMT promoter was hypomethylated at 10 of 14 CpG sites analyzed in IPF fibroblasts compared to nonfibrotic control cells, which is consistent with the array results ( Figure 2B). The differential methylation of CDKN2B and MGMT in IPF fibroblasts was largely localized to CpG islands, and in regions where reference sequencing showed other cell types to also be differentially methylated (as indicated by MeDIP-Seq reference data [29]). Bisulfite sequencing of the CARD10 promoter revealed that like CDKN2B, it too was hypermethylated in IPF fibroblasts ( Figure 2C). The identified hypermethylated regions include the CpG site that was assayed by the array, as well as regions up-and downstream of the transcription start site. Interestingly, hypermethylation was identified in two separate regions, one in which MeDIP-Seq signaling was present, and one in which MeDIP-Seq signaling was absent in reference data [29].
Aberrant methylation of CDKN2B, CARD10, and MGMT in IPF fibroblasts is associated with altered gene expression and increased fibroblast proliferation DNA methylation, especially within gene promoters and CpG islands, is traditionally associated with decreased gene expression.
To determine the impact of altered DNA methylation on gene  expression, we compared the expression of CDKN2B, CARD10, and MGMT mRNA and protein in IPF fibroblasts with those of patient control cells. IPF cells exhibited decreased expression of CDKN2B and CARD10 and increased expression of MGMT compared to nonfibrotic control fibroblasts ( Figure 3A-D), concordant with the observations that CDKN2B and CARD10 were hypermethylated and MGMT was hypomethylated in IPF cells. Treating IPF fibroblasts with the DNA methylation inhibitor 5aza-29-deoxycytidine resulted in increased expression of CDKN2B and CARD10, but not MGMT ( Figure 3E), consistent with the hypothesis that increased methylation down-regulates expression of CKDN2B and CARD10. Treatment with 5-aza-29-deoxycytidine was associated with a decrease in CARD10 methylation ( Figure 3F), and a more modest decline in the methylation of CDKN2B (from 37.9% to 31%, based on the mean of the 28 CpG loci assayed). As expected, expression of MGMT did not change with methylation inhibitors, as MGMT was hypomethylated in IPF cells. CDKN2B is an inhibitor of cyclin-dependent kinases 4 and 6 and is a key regulator of cell cycle arrest [30]; thus, increased methylation followed by decreased CDKN2B expression might be expected to contribute to the reported increase in proliferation observed in IPF fibroblasts [5,6]. Consistent with this possibility, silencing CDKN2B expression by siRNA in normal fibroblasts resulted in increased cell proliferation (Figure 4).

Aberrant DNA methylation may account for the differential expression of other genes identified in IPF fibroblasts
Previous studies have shown that many genes are differentially expressed in IPF fibroblasts. We compared our set of genes that were differentially methylated between IPF fibroblasts and patientderived controls with those genes identified in a study by Lindahl et al (GEO accession GSE40839) as being differentially expressed in IPF cells [31], and identified 52 genes in common between the datasets (q-value 7.23610 25 ). These genes are shown in Table 3.

Genes aberrantly methylated in IPF fibroblasts are enriched in annotation for various cell functions
To determine if the 732 differentially methylated genes we identified in IPF fibroblasts share common biological functions, we used ConceptGen (conceptgen.ncibi.org) [32] to identify ''concepts'' -categorized by gene ontology (GO) classifications, medical subject headings (MeSH), and other publicly available annotative data sets. Statistically significant enriched GO concepts are shown in Table 4 and include ''extracellular matrix,'' ''organ morphogenesis,'' and ''potassium ion binding.'' Other concepts enriched by MeSH and miR database annotations are listed in Table S5. Because ''potassium ion binding'' is an enriched concept that was unexpected in our analysis and has no obvious relationship to fibrogenesis, we performed a network analysis of genes in this concept to examine their interrelationship, which is shown in Figure 5.

IPF fibroblasts exhibit heterogeneity in their methylation profiles
Although we identified significant differences in mean DNA methylation between IPF and nonfibrotic control cells, considerable variability in methylation was present among individual IPF cell lines. Hierarchical cluster analysis identified a group of IPF lines that exhibited similar overall DNA methylation profiles, with other lines exhibiting divergent profiles ( Figure 6A). This variability was evident even when the mean methylation of an individual gene, such as CARD10 was compared among IPF lines ( Figure 6B). Culture conditions have been suggested to influence global DNA methylation levels [33,34] and one might speculate whether in vitro culturing of fibroblasts is a source of variability in DNA methylation. All cell lines were analyzed at similar passage during our analysis. However, we found substantial consistency in the DNA methylation levels of CARD10 and MGMT even when cells were examined at different passage numbers (Fig. 7), consistent with the heritable nature of DNA methylation levels.

Discussion
Excessive fibroproliferation is a key hallmark of IPF, and fibroblasts are the major effector cell responsible for this process [35]. Work from numerous laboratories has demonstrated that fibroblasts from IPF patients exhibit an abnormally ''activated'' phenotype [3,[5][6][7], which in turn has been attributed to differential regulation of key genes [3,4,7,8,10,[12][13][14]. These differences in gene expression typically persist over time, through cell division, and despite changes in their microenvironment. Epigenetic mechanisms represent an attractive explanation for such stable differences in gene expression, but global methylation analysis of IPF fibroblasts has not previously been reported. Here, we used the HumanMethylation27 Bead-Chip array to compare the global DNA methylation patterns of IPF and nonfibrotic control fibroblasts. We identified multiple CpG loci and their associated genes that are differentially methylated; these methylation differences spanned the genome, were present in every chromosome, and were associated with differences in gene expression. Some of these differentially methylated genes have known effects on fibroblast biology, and their dysregulated expression can readily be envisioned to contribute to the pathogenesis of IPF. Other genes possessed annotations enriched in gene ontology classes such as potassium ion channels and organ morphogenesis that we found on first glance to be surprising. However, literature review of some of these genes in other organ systems suggests that these too, may be important to fibrosis. Additional studies will be needed to determine if these represent  previously unappreciated novel pathways involved in IPF pathogenesis. Taken together, our findings establish DNA methylation as a critical epigenetic mechanism that contributes to the altered phenotype of IPF fibroblasts. Although alterations in DNA methylation contribute to the pathogenesis of many diseases, the importance of this epigenetic modification in IPF is only beginning to be realized. Two independent studies have shown that tissue from the lungs of IPF patients exhibit global DNA methylation differences compared to nonfibrotic lungs [15,16]. However, numerous cell types comprise the lung parenchyma, and mixtures of different cell types present in the whole lung tissue utilized in these studies might have obscured important DNA methylation differences present in any given cell. This is especially problematic considering that certain cell types, such as epithelial cells and fibroblasts, are known to exhibit, dichotomous gene expression profiles. It is therefore not surprising that many of the methylation differences identified by these studies of whole lung were not observed in our studies of lung fibroblasts, despite use of the same array platform (i.e. Human-Methylation27 array) [16]. Studies of specific genes -namely, Thy-1 [19], PTGER2 [18], and p14 ARF [17] -have identified their differential methylation in IPF fibroblasts, but this is the first study of which we are aware that describes genome-wide differences in DNA methylation in these critical mesenchymal effector cells.
Bisulfite sequencing allowed us to validate the differential methylation of CDKN2B, CARD10, and MGMT, and we found that differences in methylation of these genes involved CpG sites not represented by the array. These genes exemplify both differential hypermethylation (in the case of CDKN2B and CARD10) and hypomethylation (in the case of MGMT), contributing to their differential gene expression in IPF fibroblasts. CDKN2B is an endogenous cell cycle inhibitor that binds to CDK4 and -6 [30]; its decreased expression in IPF fibroblasts may contribute to their increased proliferation. Indeed, silencing of CDKN2B in normal fibroblasts resulted in increased cell proliferation. CARD10 is a scaffold protein recognized to associate upstream G protein signals (such as that from the putative fibrogenic lipid mediator lysophosphatidic acid [LPA]) with downstream NF-kB activity [36,37]. MGMT is a DNA repair enzyme that regulates chromatin stability and susceptibility to apoptosis [38] and its increased expression in IPF cells may contribute to the well-recognized phenomenon of fibroblast resistance to apoptosis in IPF [3,9,10,12]. Further studies at the molecular and biochemical levels are needed to characterize the biological significance of each of these genes, as well as others identified in the array, in IPF. We compared our set of differentially methylated CpG loci with publicly available array data that compared gene expression between IPF and normal fibroblasts, and noted that 52 of the reported differentially expressed genes [31] overlapped with our dataset. This was not a result of mere chance, as the false discovery rate was 7.23610 25 . However, we also note that such in silico analysis has limitations due to the different platforms used, and variability in patient population between institutions and cell isolation techniques between different laboratories. Nonetheless, this suggests that some of the differences in gene expression of IPF fibroblasts may be attributable to differences in DNA methylation. It is noted that not all of the differences in gene expression were directionally opposite to the differences in methylation, suggesting that further studies would be needed to determine how differences in individual gene expression may be affected by either hyper-or hypomethylation. Although the patient selection criteria, biopsy techniques, and culture methods employed by Lindahl et al. were very similar to those utilized in our study, future experiments where DNA methylation and gene expression analysis are performed from identical samples in parallel may improve the robustness of matching DNA methylation and gene expression data.
Enrichment analyses revealed that certain genes that were differentially methylated were overrepresented in annotation for gene ontology classes such as ''extracellular matrix'' and ''extracellular space.'' The enrichment of these concepts is consistent with a disorder that is characterized by tissue remodeling due to excess deposition of matrix proteins such as collagens. Still, identifying the genes within these gene ontology classes that are differentially methylated may provide insight into the importance of each particular gene in driving fibrosis. The fact that these particular gene ontology classes were found overrepresented in annotation also validates the biological significance of the larger array findings and indicates that observed methylation differences occur in genes that are functionally relevant to IPF pathogenesis. We also identified unexpected gene ontology concepts such as ''organ morphogenesis'' and ''potassium ion binding'' that were enriched in our gene data set. Some genes within the ''organ morphogenesis'' gene ontology, such as PDGFRA, TWIST1, WNT7B, and SFPTB [39][40][41][42], have been implicated in IPF pathogenesis; other genes in the ''potassium ion binding'' gene ontology have been implicated in renal fibrosis [43] and further interrogation of the function of these genes could drive the discovery of novel mechanisms of IPF pathogenesis. Network analysis of these genes reveals that they share closely coordinated functions, based on their direct protein-protein interaction and on their ability to bind potassium ion and serve as a potassium ion channel. Future work may reveal a novel role for potassium signaling in IPF pathogenesis.
Up to 3,690 of 27,578 CpG loci (13.4%) on the HumanMethy-lation27 array are annotated for known SNPs. The inclusion of probes with SNPs has the potential to be problematic, as  differences in signals from these probes may be falsely attributed to methylation differences when in fact, they are due to SNP variations in the samples. However, since the proportion of CpG loci (,14%) that we identified as differentially methylated and annotated for SNPs is similar to the proportion of SNPs annotated over the entire array, the data do not suggest that SNPs represent a bias in the identification of differentially methylated CpG sites. Although probes with annotated SNPs could have been removed in the analysis, this has the potential for creating its own significant (and, to date, unknown) bias. The frequency of some minor SNP alleles may be quite low, such as 1% or even 0.1%, and removing these data points because they reside on SNPs of low frequency might not be justified. Furthermore, we confirmed differential methylation in the promoters of three genes (CDKN2B, CARD10, and MGMT) by bisulfite sequencing. Of these three, both CDKN2B and MGMT are annotated for known SNPs in the relevant sites, but both are validated as being differentially methylated. Finally, the presence of SNPs and their ability to affect DNA methylation levels might actually be important biologically. Thus, we did not a priori eliminate probes with known SNPs in our analysis, but instead annotated them in Tables 2, S2, S3, and S4. These annotated loci should not be ignored, but considered with caution in future follow-up studies.
IPF fibroblasts exhibited significant heterogeneity in their global DNA methylation patterns, as evident in hierarchical clustering analysis and in the DNA methylation analysis of individual genes. As compared to other fibrotic lung disorders, IPF is recognized to be clinically heterogeneous [49], and to exhibit pathologic heterogeneity (defined by areas of seemingly normal histology adjacent to areas of dense fibrosis) that distinguishes it from other types of interstitial lung disease [1,50]. One might thus speculate that even the process of culturing cells in vitro may produce populations of cells with distinct patterns of DNA methylation. Despite this, we were able to identify many differences in DNA methylation even among a small sample size. Systematic comparison of methylation levels of CARD10 and MGMT genes between different passages confirmed that DNA methylation patterns are stable through cell division. This is a central, albeit not often tested, tenet of epigenetics. It is presumed that heterogeneity could be reduced if a larger sample number was used or if specific subpopulations of fibroblasts, segregated by patient clusters or by the region of lung from which cells were isolated, were employed. In fact, methylation differences in THY1 [19], PTGER2 [18], and p14 ARF [17] were all previously shown to be heterogeneous among IPF fibroblasts, which may account for why these genes did not show up in our array analysis. This suggests that even cells of the same cell type, obtained from geographically similar regions of the lung, may exhibit heterogeneity in DNA methylation and gene expression. Heterogeneity also suggests that examining mean methylation differences between IPF and nonfibrotic control cells may obscure important differences that may only be evident if subpopulation of IPF samples, defined a priori by unsupervised hierarchical clustering of IPF patients, were performed. Despite the variability, we were still able to identify certain CpG loci that differed in methylation between IPF and nonfibrotic fibroblasts as a whole. Finally,the array assays on average only 1-2 CpG sites per gene and is biased towards sites within CpG islands. Future work with arrays that assay 450,000 CpG sites and nextgeneration sequencing approaches that provide greater unbiased coverage may reveal many more differentially methylated CpG loci that eluded our detection.
How DNA methylation differences arise in IPF is unknown. We previously reported that the antifibrotic mediator prostaglandin E 2 increases DNMT3a expression [28], and that expression of different DNA methyltransferases (DNMTs) vary in IPF fibroblasts [18]. However, alterations in DNMT expression is likely to only partially explain all of the DNA methylation differences observed in IPF cells, and does not account for how certain genomic regions are specifically targeted. How DNA methylation differences affect gene expression on a gene-by-gene basis is also unclear, especially since our data suggest DNA methylation differences in IPF are frequently found in non-CpG islands, and when compared to expression array data from other investigators, show varying levels of correlation between gene expression and methylation of specific CpG loci. Despite these unresolved issues, our data provide the first description of the global DNA methylation differences that are present in IPF fibroblasts, and offers a starting point for understanding the extent of such differences, and the importance of the genes affected by DNA methylation in IPF pathogenesis.

Author Contributions
Conceived and designed the experiments: SKH. Performed the experiments: SKH AMS. Analyzed the data: SKH RCM MPG. Contributed reagents/materials/analysis tools: ESW. Contributed to the writing of the manuscript: SKH RCM MPG.