Genome-Wide DNA Methylation Study Identifies Significant Epigenomic Changes in Osteoarthritic Subchondral Bone and Similarity to Overlying Cartilage

Abstract

Objective

To perform a genome-wide DNA methylation study to identify differential DNA methylation patterns in subchondral bone underlying eroded and intact cartilage from patients with hip osteoarthritis (OA) and to compare these with DNA methylation patterns in overlying cartilage.

Methods

Genome-wide DNA methylation profiling using Illumina HumanMethylation 450 arrays was performed on eroded and intact cartilage and subchondral bone from within the same joint of 12 patients undergoing hip arthroplasty. Genes with differentially methylated CpG sites were analyzed to identify shared pathways, upstream regulators, and overrepresented gene ontologies, and these patterns were compared with those of the overlying cartilage. Histopathology was graded by modified Mankin score and assessed for correlation with DNA methylation.

Results

We identified 7,316 differentially methylated CpG sites in subchondral bone underlying eroded cartilage, most of which (~75%) were hypomethylated, and 1,397 sites in overlying eroded cartilage, 126 of which were shared. Samples clustered into 3 groups with distinct histopathologic scores. We observed differential DNA methylation of genes including the RNA interference–processing gene AGO2, the growth factor TGFB3, the OA suppressor NFATC1, and the epigenetic effector HDAC4. Among known susceptibility genes in OA, 32 were differentially methylated in subchondral bone, 8 were differentially methylated in cartilage, and 5 were shared. Upstream regulator analysis using differentially methylated genes in OA subchondral bone showed a strong transforming growth factor β1 signature (P = 1 × 10⁻⁴⁰) and a tumor necrosis factor family signature (P = 3.2 × 10⁻²⁸), among others.

Conclusion

Our data suggest the presence of an epigenetic phenotype associated with eroded OA subchondral bone that is similar to that of overlying eroded OA cartilage.

Osteoarthritis (OA) is a chronic musculoskeletal disease characterized by progressive loss of articular joint function, leading to both significant pain and functional limitation. A truly unrecognized pandemic, OA is the leading cause of chronic disability in the US, affecting 1 in 5 adults, with an annual cost of nearly 1% of the gross domestic product. The rate of severe knee OA requiring joint replacement is increasing at a rate that exceeds expected increases due to obesity and aging (1).

The etiology of OA remains nebulous. Although a hallmark of OA is loss of articular cartilage, most investigations to date have focused only on the cartilaginous component of this larger joint “organ” disease. Nevertheless, several other articular tissue processes have been implicated in OA pathogenesis, including synovial inflammation and subchondral bone remodeling (2). Our group and others recently highlighted the contribution of an altered epigenome to the development of human OA. We reported significant alterations in DNA methylation patterns of erosion in OA cartilage using paired intact OA cartilage samples, associations between DNA methylation and histopathologic scores within these specimens, and epigenetic alterations of many OA susceptibility genes in eroded OA cartilage (3). Several other groups observed a dysregulated, albeit geographically distinct, epigenome among both human knee and hip OA cartilage compared with control cartilage obtained from patients with a femoral neck fracture, as well as differences along the spectrum of disease severity (4–6). These studies have been instrumental in demonstrating that human OA is, at least in part, driven by disordered epigenomics.

Given our knowledge of differential DNA methylation in OA cartilage and a particular interest in epigenetic differences within the joints of patients with OA, we sought to determine whether a similarly altered epigenome could be found in subchondral bone underlying eroded cartilage and how such an epigenome would compare with that of the overlying cartilage. To that end, we conducted a genome-wide DNA methylation study comparing subchondral bone obtained from underneath eroded OA hip cartilage with that underlying intact OA hip cartilage from within the same joint, and compared these findings with DNA methylation patterns in the overlying cartilage.

MATERIALS AND METHODS

Samples and nucleic acid isolation

Twelve femoral heads were obtained from patients undergoing hip arthroplasty for end-stage primary OA at the McBride Orthopedic Hospital in Oklahoma City. Demographic information about these patients is presented in Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://online-library.wiley.com/doi/10.1002/art.39555/abstract. The institutional review boards in all involved institutions approved this study. Full-thickness cartilage samples representing grossly intact and grossly eroded areas were obtained and histopathologically scored (see Supplementary Table 2, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract), DNA was isolated, and genome-wide DNA methylation was quantified (3), after which the femoral heads were stored in the vapor phase of liquid nitrogen.

For the current study, 0.5 cm–diameter, 1 cm–deep subchondral trabecular bone cores were obtained from areas immediately underlying the previously examined eroded and intact cartilage, using a surgical trephine. Residual cartilage tissue was removed. Bone was immediately flash frozen in liquid nitrogen and cryogenically ground using a liquid nitrogen–cooled grinder mill (Spex CertiPrep). DNA was isolated from each sample using a DNeasy kit (Qiagen), and 500 ng was subsequently treated with sodium bisulfite using an EZ DNA Methylation Kit (Zymo Research).

DNA methylation studies

Genome-wide DNA methylation was assessed using an Illumina HumanMethylation 450 BeadChip microarray, which analyzes the methylation status of >485,000 methylation sites throughout the genome, covering 99% of RefSeq genes at an average of 17 CpG sites per gene across the 5′-untranslated region (5′-UTR), gene promoter regions, first exon, gene body, and 3′-UTR, and covering 96% of University of California, Santa Cruz–defined CpG islands and their flanking regions. Following bisulfite treatment, DNA from each sample was loaded onto chips and processed at the Oklahoma Medical Research Foundation Genotyping Core Facility. This array includes a variety of both sample-independent and sample-dependent controls, which were evaluated in each chip.

Differential methylation analysis

The chips were imaged, and data were extracted using the GenomeStudio version 2011.1 methylation module (Illumina). The percent methylated cytosines at each CpG site (β) was calculated as the ratio of methylated probe signal to total locus signal intensity and defined within a range of 0.0 to 1.0 by GenomeStudio, and average β values were compared. Differentially methylated CpG sites were defined as those having a differential methylation score ≥|21|, corresponding to a P value of less than or equal to 0.01 after adjusting for multiple testing using the GenomeStudio built-in false discovery rate function, which employs a Benjamini and Hochberg procedure with α = 0.05. A second requirement was a mean methylation difference (Δβ) ≥0.15 (15%) between the 2 groups. CpG sites with a known single-nucleotide polymorphism located within 10 bp of the 3′-end of the probe were excluded. The locations of particular CpG sites within regulatory regions were defined by GenomeStudio, including CpG islands, North (5′) and South (3′) shores (N-Shores and S-Shores), and North (5′) and South (3′) Shelves (N-Shelves and S-Shelves), which represent ~2 kb surrounding CpG islands and shores, respectively. Location statistics were calculated using the Yates’ correction of a chi-square distribution P value of a 2 × 2 contingency table (see Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract).

Gene ontology analyses

Functional properties, networks, pathways, and upstream regulators enriched in differentially methylated genes were assessed using the Ingenuity Pathway Analysis (IPA) system (Ingenuity Systems) using the Ingenuity Knowledge Base reference set. Direct and indirect relationships were calculated, and experimentally observed relationships were included; P values less than or equal to 0.05 were considered significant. In addition to the standard upstream regulator analysis, microRNAs (miRNAs) that were overrepresented among differentially methylated genes were identified by IPA using experimentally demonstrated and predicted miRNA–messenger RNA interactions or binding sites from TarBase, miRecords, and TargetScan databases as well as peer-reviewed miRNA research articles as curated by Ingenuity Systems staff.

Histologic correlation

In the 11 matched samples with histopathologic scores, CpG methylation at each site was compared with the modified Mankin score. The coefficient of determination (R²) was calculated using a linear regression model in Excel. Correlations were deemed significant if R² was ≥0.60 and the range of methylation values across samples for the CpG of interest was at least 15% (Δβ_samples ≥0.15).

Comparison with previous cartilage differential methylation data

Differentially methylated CpG sites in paired samples of eroded and intact cartilage overlying the subchondral bone were determined using the above criteria and raw data from our previous experiment (3). Differentially methylated CpG sites in subchondral bone and cartilage were compared for overlap. The pathways and functional ontologies that were identified as being most overrepresented in both experiments were compared using the IPA Comparison Analysis tool, and the most highly significant pathways in both sets were reviewed.

Table 1. Top differentially hypomethylated and hypermethylated CpG sites in subchondral bone underlying eroded and intact osteoarthritic cartilage*

Table 1.

Top differentially hypomethylated and hypermethylated CpG sites in subchondral bone underlying eroded and intact osteoarthritic cartilage*

Illumina CpG site, ID no.	Associated gene symbol	Mean β, eroded	Mean β, intact	Δβ	Differential score, FDR corrected	P, FDR corrected	CpG location	†	Located in enhancer
Top hypomethylated sites
cg20918393	NA	0.19	0.52	−0.33	−346.99	2.1 × 10−38	–	–	–
cg23731089	EIF2C2	0.28	0.60	−0.32	−346.99	2.1 × 10−38	Gene body		–
cg10160614	NA	0.38	0.69	−0.31	−346.99	2.1 × 10−38	–	S-Shelf	Yes
cg10667102	NA	0.32	0.64	−0.31	−346.99	2.1 × 10−38	–	N-Shore	–
cg11006453	EIF2C2	0.25	0.56	−0.31	−346.99	2.1 × 10−38	Gene body		–
cg16345114	ERGIC1	0.35	0.66	−0.31	−346.99	2.1 × 10−38	Gene body		Yes
cg03475293	NA	0.34	0.64	−0.31	−346.99	2.1 × 10−38	–	N-Shore	–
cg07795766	C22orf9	0.25	0.56	−0.30	−346.99	2.1 × 10−38	TSS 200	–	Yes
cg02524983	LPP	0.35	0.66	−0.30	−346.99	2.1 × 10−38	5′-UTR	–
cg10113526	DENND3	0.36	0.66	−0.30	−346.99	2.1 × 10−38	Gene body	–	Yes
cg02029908	DUSP1	0.42	0.72	−0.29	−346.99	2.1 × 10−38	3′-UTR	N-Shore	–
cg07018389	DUSP1	0.28	0.57	−0.29	−346.99	2.1 × 10−38	3′-UTR	N-Shore	–
cg23972735	KDM2B	0.30	0.59	−0.29	−346.99	2.1 × 10−38	Gene body	Island	–
cg08554554	ARFRP1	0.27	0.56	−0.29	−346.99	2.1 × 10−38	3′-UTR	S-Shelf	Yes
cg27549186	TIMP2	0.30	0.58	−0.29	−346.99	2.1 × 10−38	Gene body	–	Yes
cg03211098	NA	0.35	0.63	−0.29	−346.99	2.1 × 10−38	–	–	Yes
cg19628600	FOXP1	0.26	0.55	−0.28	−346.99	2.1 × 10−38	Gene body	–	Yes
cg25291653	NA	0.32	0.60	−0.28	−346.99	2.1 × 10−38	–	–
cg00952054	NA	0.27	0.55	−0.28	−346.99	2.1 × 10−38	–	–	Yes
cg25951288	NFATC1	0.41	0.70	−0.28	−346.99	2.1 × 10−38	Gene body	Island
cg14599823	NA	0.39	0.67	−0.28	−346.99	2.1 × 10−38	–	–	Yes
Top hypermethylated sites
cg04541368	FLJ42709	0.67	0.38	0.30	349.65	1.1 × 10−38	Gene body	S-Shore	Yes
cg19147218	HOXA7	0.69	0.40	0.29	349.65	1.1 × 10−38	TSS 1,500	N-Shore	–
cg08075204	BIN2	0.76	0.48	0.28	349.65	1.1 × 10−38	TSS 200	–	–
cg26232412	PLCB2	0.74	0.46	0.28	349.65	1.1 × 10−38	TSS 1,500	–	Yes
cg02384857	HOXA7	0.73	0.46	0.27	349.65	1.1 × 10−38	TSS 1,500	N-Shore	–
cg14556909	BIN2	0.77	0.50	0.27	349.65	1.1 × 10−38	TSS 200	–	–
cg26427109	CD6	0.57	0.30	0.27	349.65	1.1 × 10−38	TSS 200	–	–
cg03649649	NA	0.64	0.37	0.27	349.65	1.1 × 10−38	–	–	–
cg14516100	SORBS2	0.69	0.43	0.26	349.65	1.1 × 10−38	Gene body	Island
cg03923561	HOXC4	0.61	0.36	0.26	349.65	1.1 × 10−38	TSS 1,500	N-Shore	Yes
cg18908499	C1orf150	0.75	0.49	0.25	349.65	1.1 × 10−38	TSS 1,500	–	–
cg14235846	NXN	0.72	0.46	0.25	349.65	1.1 × 10−38	Gene body	–	Yes
cg00172603	SSBP3	0.64	0.38	0.25	349.65	1.1 × 10−38	Gene body	Island	–
cg06911354	HOXA7	0.78	0.53	0.25	349.65	1.1 × 10−38	TSS 1,500	N-Shore	–
cg26255314	CD5	0.70	0.44	0.25	349.65	1.1 × 10−38	TSS 200	–	–
cg25045219	C17orf97	0.70	0.45	0.25	349.65	1.1 × 10−38	Gene body	Island
cg11648730	FLJ42709	0.64	0.39	0.25	349.65	1.1 × 10−38	TSS 1,500	Island	Yes
cg09450197	NA	0.69	0.45	0.25	349.65	1.1 × 10−38	–	–	–
cg18723409	LSP1	0.72	0.48	0.25	349.65	1.1 × 10−38	3′-UTR	–	–
cg22797031	NA	0.68	0.44	0.25	349.65	1.1 × 10−38	–	N-Shore	Yes
cg22309950	CD28	0.66	0.42	0.24	349.65	1.1 × 10−38	5′-UTR	–	–

^*Δβ= difference in methylation value between sample groups (β_eroded − β_intact); FDR = false discovery rate; NA = not applicable; S = South; N = North; TSS 200 = within 200 bp of transcription start site; 5′-UTR = 5′-untranslated region.

^†Location within or surrounding a University of California, Santa Cruz–defined CpG island.

To confirm the results from the DNA methylation array, we performed pyrosequencing (EpigenDx) on a subset of 12 matched subchondral bone samples (6 eroded and 6 intact) of the 4 most highly differentially methylated CpG sites: cg04541368 (FLJ42709 [hypermethylated in eroded tissue]), cg23731089 (EIF2C2 [hypomethylated]), cg11006453 (EIF2C2 [hypomethylated]), and cg19147218 (HOXA7 [hypermethylated]).

RESULTS

Significant differential CpG methylation in OA subchondral bone, clustering with overlying cartilage histology scores

A substantial minority of methylated CpG sites were shared between the 2 tissue types. We observed 7,316 differentially methylated CpG sites when comparing bone underlying eroded OA cartilage with bone underlying intact OA cartilage, representing 2,872 distinct genes and nearby genomic regions. Of these 7,316 sites, 5,477 (2,279 genes) were hypomethylated, and the remaining 1,839 sites (593 genes) were hypermethylated (Table 1) (for additional information see Supplementary Table 4, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract). Similar to our previous findings in OA cartilage, differentially methylated sites were concentrated in particular genomic locations, namely enhancers, which are segments of regulatory sequences that exert effects on gene expression from remote locations (hypomethylated = 2.6× the expected rate [P < 0.0001]; hypermethylated = 1.2× the expected rate [P = 0.0002]) and N-Shores of CpG islands (hypomethylated = 2.2× the expected rate [P < 0.0001]; hypermethylated = 4.4× the expected rate [P < 0.0001]) (see Supplementary Table 3, available at http://online-library.wiley.com/doi/10.1002/art.39555/abstract).

Supervised hierarchical clustering of differentially methylated sites segregated samples into 3 distinct clusters (Figure 1). Interestingly, the cartilage overlying each of these clusters had unique histopathologic scoring characteristics. Cluster 1 consisted of 7 samples underlying intact cartilage (mean ± SEM Mankin score 1.2 ± 0.4), cluster 2 consisted of 3 intact and 3 eroded samples (Mankin score 5.5 ± 0.9 [P = 0.002 versus cluster 1]), and cluster 3 consisted of 9 eroded and 2 intact samples (Mankin score 6.6 ± 3.5 [P = 0.003 versus cluster 1]).

Figure 1.

Supervised hierarchical clustering of 7,316 differentially methylated CpG sites in subchondral bone. Modified Manskin scores (sc) for overlying cartilage were not available for 2 specimens that were lost during preparation. I = intact overlying cartilage; E = eroded overlying cartilage; NA = not applicable.

Next, we compared subchondral bone CpG sites with the 1,397 sites we identified as being differentially methylated in overlying cartilage (950 hypomethylated and 447 hypermethylated, representing 762 total genes). One hundred twenty-six CpG sites (2% of subchondral sites, 9% of cartilage sites) were differentially methylated in both subchondral bone and overlying cartilage (Table 2). Three hundred thirty-nine genes (12% of subchondral genes, 44% of cartilage genes) had at least 1 differentially methylated site in both tissue types. Methylation alterations were concordant in 113 of 126 (90%) of shared CpG sites. The 4 most highly differentially methylated CpG sites were confirmed by pyrosequencing analysis (cg04541368, hypermethylated: Δβ = 20% [P = 0.006]; cg23731089, hypomethylated: Δβ = −37% [P = 0.002]; cg11006453, hypomethylated: Δβ = −37% [P= 0.002]; and cg19147218, hypermethylated: Δβ = 31% [P = 0.006]).

Table 2.

Differentially methylated CpG sites shared between OA subchondral bone and overlying cartilage^*

Illumina CpG ID no.	Δβ, subchondral bone	Δβ, cartilage	Associated gene	Illumina CpG ID no.	Δβ, subchondral bone	Δβ, cartilage	Associated gene
cg25588348	−0.27	0.15	TTLL5	cg11606195	−0.17	−0.18	GTF2H3
cg23903301	−0.26	−0.24	CD59	cg16651768	−0.17	−0.22	NA
cg13052638	−0.26	−0.21	NA	cg09800500	−0.17	−0.16	BCAT1
cg09670263	−0.25	−0.25	LRRC2	cg23902076	−0.17	−0.16	ASAP2
cg10992219	−0.23	−0.21	RB1CC1	cg14325112	−0.17	−0.19	GLIS3
cg13432945	−0.23	−0.21	NA	cg10968815	−0.16	−0.18	BPIL1
cg12192282	−0.23	−0.26	NA	cg08306614	−0.16	−0.21	NUDCD3
cg03532013	−0.23	−0.18	POU2F2	cg03321829	−0.16	−0.19	ERGIC1
cg26187031	−0.23	−0.16	DPYSL2	cg05529278	−0.21	−0.21	ADAMTS5
cg08381046	−0.23	−0.16	NA	cg13184448	−0.16	−0.22	LRRFIP1
cg14062643	−0.22	−0.21	PIP5KL1	cg03786920	−0.16	−0.17	STK35
cg18700744	−0.22	−0.24	NAA25	cg10995381	−0.16	−0.19	MTRR
cg01963754	−0.22	−0.17	C13orf16	cg03907570	−0.16	−0.15	PRKAR1B
cg11267810	−0.22	−0.16	NA	cg07110405	−0.16	−0.17	SHANK2
cg02927679	−0.21	−0.16	DUSP5	cg13305415	−0.16	0.16	SPRY4
cg08698943	−0.21	−0.22	NA	cg14986890	−0.16	−0.16	RARRES1
cg03221073	−0.21	−0.15	HMCN1	cg27050407	−0.16	−0.26	FAM65B
cg26647771	−0.21	−0.29	NPR3	cg04723343	−0.16	−0.17	LRRFIP1
cg25765315	−0.21	−0.18	NA	cg04757411	−0.16	−0.15	LMO7
cg00674995	−0.20	−0.24	DGKH	cg03503642	−0.16	0.19	COL12A1
cg24032304	−0.20	−0.17	PTPN11	cg06880930	−0.16	−0.20	CPNE2
cg22110158	−0.20	−0.18	ST14	cg22534374	−0.16	−0.28	NA
cg03896685	−0.20	−0.15	NA	cg14290576	−0.16	−0.17	NFIL3
cg05398095	−0.20	−0.18	YOD1	cg05218510	−0.16	−0.19	ITPRIPL2
cg06710195	−0.20	−0.15	VAV3	cg05588903	−0.16	−0.15	CRISPLD2
cg26400954	−0.20	−0.19	LMO7	cg19382019	−0.16	−0.15	NA
cg19763108	−0.20	−0.16	THRB	cg03097134	−0.16	−0.16	NA
cg16644023	−0.19	−0.17	HOXA3	cg18767735	−0.16	−0.17	GLIS1
cg16330359	−0.19	−0.32	NA	cg14552732	−0.16	−0.16	NA
cg18006379	−0.19	−0.17	TCERG1L	cg05467828	−0.16	−0.19	NA
cg08782899	−0.19	−0.18	ARL16	cg20228731	−0.16	−0.23	FLJ43663
cg14534803	−0.19	−0.18	NAV2	cg00148223	−0.16	−0.16	NA
cg00101629	−0.19	−0.21	KIAA1026	cg11401820	−0.16	−0.17	EFTUD1
cg19807286	−0.19	−0.18	NA	cg09643398	−0.15	−0.22	BCL6
cg09989996	−0.19	−0.17	NA	cg17435901	−0.15	−0.16	SLC39A7
cg12002139	−0.19	−0.19	SYNJ2	cg18499941	−0.15	−0.15	SMOC2
cg20216309	−0.19	−0.16	NA	cg21156057	−0.15	0.15	ABL1
cg23636802	−0.18	−0.17	TNXB	cg01656216	−0.15	−0.17	ZNF438
cg06480353	−0.18	−0.16	NA	cg24067911	−0.15	−0.18	ATXN1
cg14487111	−0.18	−0.16	NA	cg08216099	−0.15	−0.24	PXDN
cg07775417	−0.18	−0.26	AGPAT9	cg09243909	−0.15	0.22	FTO
cg16748008	−0.18	−0.22	HOXA3	cg01591152	−0.15	−0.21	TRIM29
cg05716270	−0.18	−0.19	C1QTNF8	cg03709468	−0.15	−0.18	TDO2
cg23138250	−0.18	−0.20	NA	cg20778199	−0.15	−0.19	NA
cg00245885	−0.18	−0.16	H2AFY	cg23508201	−0.15	−0.17	NA
cg18123043	−0.18	−0.17	NA	cg12048965	−0.15	−0.16	NA
cg00172872	−0.18	−0.17	NA	cg15891358	−0.15	−0.15	C10orf90
cg01424562	−0.18	0.20	ZFP36L1	cg21840806	−0.15	0.15	CREB5
cg26664528	−0.17	−0.24	MIR548H4	cg25699759	−0.15	−0.16	NTRK3
cg13306478	−0.17	−0.16	ALOX5AP	cg16388071	−0.15	−0.23	NA
cg09763439	−0.17	−0.15	NA	cg08243465	−0.15	−0.23	PMEPA1
cg16251016	−0.17	−0.23	NA	cg17235897	0.15	0.19	FGFR2
cg08564522	−0.17	−0.20	NA	cg15612947	0.15	−0.18	TRIO
cg16305094	−0.17	−0.17	SSU72	cg25825506	0.15	0.19	DLX6AS
cg02614661	−0.17	−0.21	SLC7A5	cg16738646	0.16	−0.18	SLC2A1
cg03478610	−0.17	−0.23	PPP2R3A	cg07249730	0.16	−0.18	MSI2
cg27588356	−0.17	−0.19	MAP3K4	cg11905061	0.16	−0.16	AGAP1
cg18484958	−0.17	−0.18	WWTR1	cg00551679	0.16	0.16	FOXF1
cg02131853	−0.17	−0.22	TMEM156	cg19092317	0.17	−0.19	NA
cg16932472	−0.17	−0.15	IPO5	cg03972071	0.20	0.18	ZADH2
cg10923036	−0.17	−0.22	IGFBP7	cg24937727	0.21	0.20	RGL3
cg20706315	−0.17	−0.17	NA	cg25574765	0.22	−0.15	PPFIA1
cg18995788	−0.17	−0.20	NA	cg07184578	0.24	0.19	NA

^*One hundred thirteen (90%) of the 126 shared CpG sites are concordantly hypomethylated or hypermethylated. Δβ = difference in methylation value between sample groups (β_eroded − β_intact). NA = not applicable.

Differential methylation in both novel and previously identified OA-associated genes

A full list of differentially methylated CpG sites in OA subchondral bone and their associated genes are shown in Supplementary Table 4 (available at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract). The most highly differentially methylated genes identified included the RNA-induced silencing complex (RISC)–associated gene EIF2C2 (also known as AGO2), represented by 8 hypomethylated and 3 hypermethylated CpG sites located within the body and S-Shore of a CpG island. We observed evidence for hypomethylation of 3 CpG sites within the N-Shelf of a CpG island of the growth factor gene TGFB3. Furthermore, we observed 5 significantly hypomethylated CpG sites within an island and S-Shore associated with NFATC1, a member of the NFAT family that functions as a key suppressor of OA (7). Finally, the epigenetic effector histone deacetylase gene HDAC4 demonstrated 4 hypomethylated and 1 hypermethylated site located within the body, 5′-UTR, and S-Shelf of a CpG island.

The most highly hypermethylated sites were associated with a large intergenic noncoding RNA that encodes the antisense RNA for NR2F1 and FLJ42709, with 10 sites within 2 CpG islands and their associated N-Shores and S-Shores. HOXA7 had 5 hypermethylated sites within the N-Shore of a CpG island in the promoter region. We identified 6 hypermethylated CpGs within CD6, a T cell costimulatory molecule with important functions in rheumatoid arthritis (but not previously described in OA) within the body, 5′-UTR, and promoter. Another costimulatory molecule, CD5, showed 2 hypermethylated sites within the 5′-UTR and promoter. Finally, HOXA2 demonstrated 6 hypermethylated CpG sites within the promoter.

Cytosine methylation was associated with histopathologic Mankin Scores in both subchondral bone and underlying cartilage, but no overlapping genes were identified

We next analyzed associations between CpG methylation and the Mankin score for overlying OA cartilage. Our samples represented a variety of histology scores, from 0 to 11. All eroded cartilage samples had a higher score than the corresponding intact area, confirming our gross categorizations (see Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract). We identified 54 CpG sites that met our criteria for significant correlation in subchondral bone and 113 in the overlying cartilage (see Supplementary Table 5, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract). None of these CpG sites or the genes that they represent were shared between the 2 tissue types. The methylation status of the vast majority, 96% in subchondral bone and 85% in cartilage, demonstrated an inverse relationship between histology scores and DNA methylation values.

Differentially methylated OA susceptibility genes in both OA subchondral bone and cartilage

We then compared differentially methylated genes identified in the current study and 127 genes previously shown to be associated with hip or knee OA concatenated by the Human Genome Epidemiology Network (HuGENet) (8). In subchondral bone, we identified differential methylation of 32 of these genes, and in cartilage, we identified differential methylation of 8 of these genes. Five of these genes (FTO, COL13A1, IGFBP7, LRP5, and NCOR2) were shared between subchondral bone and cartilage (Table 3).

Table 3.

Differentially methylated osteoarthritis susceptibility genes

Gene	Subchondral bone (n = 32)	Cartilage (n = 8)
ADAMTS14	X	–
ADAMTS3	X	–
ANP32A	X	–
AR	X	–
CALM2	X	–
CAMK2B	X	–
CD36	X	–
COL13A1*	X	X
COL1A1	X	–
COL9A1	X	–
EPAS1	X	–
ESR1	X	–
FTO*	X	X
GDF5	X	–
HFE	X	–
IGFBP7*	X	X
IL10	X	–
IL18	X	–
KL	X	–
LEPR	X	–
LRCH1	X	–
LRP5*	X	X
MCF2L	X	–
MEFV	X	–
MMP13	X	–
MMP3	X	–
NCOR2*	X	X
NFKB1	X	–
PITX1	X	–
PRKAR2B	X	–
RHOB	X	–
TNFRSF1B	X	–
ADAM12	–	X
PAPSS2	–	X
SMAD3	–	X

^*Shared between the 2 tissue types.

Ontologic analysis shows that several biologic pathways and functions were shared among differentially methylated genes, including a strong TGFβ and cytokine signature

Next, we performed IPA to determine associated networks, pathways, biologic functions, and upstream regulators associated with differentially methylated genes in subchondral bone. The most highly enriched canonical pathways included molecular mechanisms of cancer (94 genes; P = 4 × 10⁻¹²), axonal guidance signaling (102 genes; P = 9 × 10⁻¹¹), NF-AT (54 genes; P = 7 × 10⁻¹⁰), G protein–coupled receptor signaling (68 genes; P = 1 × 10⁻⁹), and ERK/MAPK signaling (54 genes; P = 2 × 10⁻⁹). The most highly overrepresented activated biologic functions included cellular differentiation (689 genes; P = 2 × 10⁻⁵⁸), cellular proliferation (1,010 genes; P = 2 × 10⁻⁵⁷), cellular movement (660 genes; P = 3 × 10⁻⁵⁷), and cellular migration (598 genes; P = 5 × 10⁻⁵³). A highly enriched network included several genes known to be involved in OA catabolism, including the RISC component AGO2, the signaling molecule CXCR4, and the stress-inducible protein NUPR1 (Figure 2).

Figure 2.

Enriched network containing known osteoarthritis effector genes. Genes in yellow are hypomethylated in eroded versus intact subchondral bone. Genes in blue are hypermethylated.

Upstream regulator analysis revealed the growth factor TGF&beta;1 to be the most highly associated upstream regulator in OA subchondral bone, with 360 target differentially methylated genes (P = 1 × 10⁻⁴⁰). The cytokine TNF family (320 genes; P = 3.2 × 10⁻²⁸), the transcription regulator p53 (280 genes; P = 9.4 × 10⁻²⁵), the kinase ErbB-2 (152 genes; P = 2 × 10⁻¹⁹), as well as 2 other inflammatory cytokines, interferon-γ (IFNγ) (232 genes; P = 5 × 10⁻¹⁵) and interleukin-4 (IL-4) (161 genes; P = 7 × 10⁻¹⁵), were also identified (Table 4).

Table 4.

Canonical pathways and upstream regulators shared by differentially methylated (DM) genes in OA subchondral bone and OA cartilage^*

Category	No. of DM genes, subchondral bone	P	No. of DM genes, cartilage	P
Top canonical pathways
ERK/MAPK signaling	54	2.0 × 10−9	19	1.6 × 10−5
NF-AT signaling	54	7.1 × 10−10	15	1.2 × 10−3
B cell receptor signaling	50	1.8 × 10−8	14	2.5 × 10−3
PI3K signaling in B lymphocytes	40	3.7 × 10−8	11	4.2 × 10−3
Leukocyte extravasation signaling	51	4.2 × 10−7	10	1.5 × 10−3
IGF-1 signaling	30	1.5 × 10−6	10	1.5 × 10−3
Top predicted upstream regulators
TGFβ1	–	9.7 × 10−39	–	4.1 × 10−13
Tumor necrosis factor	–	3.2 × 10−28	–	5.0 × 10−19
p53	–	9.4 × 10−25	–	5.6 × 10−8
ErbB-2	–	2.0 × 10−19	–	9.4 × 10−8
Early response gene	–	4.2 × 10−14	–	1.8 × 10−8
Interferon-γ	–	5.0 × 10−15	–	3.3 × 10−5
Interleukin-4	–	7.0 × 10−15	–	5.4 × 10−3
Interleukin-2	–	4.9 × 10−13	–	7.9 × 10−3

^*OA = osteoarthritis; PI3K = phosphatidylinositol 3-kinase; IGF-1 = insulin-like growth factor 1; TGFβ1 = transforming growth factor β1.

Significant overlap in functional ontology and network analysis among differentially methylated genes between subchondral bone and overlying OA cartilage

Next, we compared the ontologic analyses from subchondral bone and cartilage. We identified significant overlap among networks, particularly ERK/MAPK signaling (for subchondral bone, P = 2 × 10⁻⁹; for cartilage, P = 1.6 × 10⁻⁵), phosphatidylinositol 3-kinase signaling in B lymphocytes (for subchondral bone, P = 3.7 × 10⁻⁸; for cartilage, P = 4.2 × 10⁻³), NF-AT signaling (for subchondral bone, P = 7.1 × 10⁻¹⁰; for cartilage, P = 1.2 × 10⁻³), B cell receptor signaling (for subchondral bone, P = 1.8 × 10⁻⁸; for cartilage, P = 2.5 × 10⁻³), leukocyte extravasation signaling (for subchondral bone, P = 4.2 × 10⁻⁷; for cartilage, P = 1.5 × 10⁻³), and insulin-like growth factor 1 (IGF-1) signaling (for subchondral bone, P = 1.5 × 10⁻⁶; for cartilage, P = 1.5 × 10⁻³). Predicted upstream regulators that were shared included TGFβ1, TNF, p53, IL-2, IL-4, IFNγ, and the early response gene, among others (Table 4).

DISCUSSION

We performed the first genome-wide DNA methylation analysis of subchondral bone underlying eroded and intact cartilage in OA. Our experiments revealed several important findings that offer unique insight into the pathogenesis of OA. First, we observed a significant number of differentially methylated CpG sites when comparing subchondral bone underlying eroded cartilage with that underlying intact cartilage. A substantial number of genes were shared between subchondral bone and the overlying cartilage from the same joint areas. Several of these associations were novel, and many genes that were previously identified as being related to OA pathogenesis or differentially methylated in OA cartilage were confirmed. We identified 7,316 CpGs that met our criteria for differential methylation comparing eroded cartilage with intact subchondral bone (Table 1 and Figure 1) (for additional information, see Supplementary Table 4, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39555/abstract). Occurring in 2,872 unique genes, most of these CpGs were hypomethylated and enriched in enhancer regions and N-Shores of CpG islands. This represents nearly an order of magnitude increase in differential methylation compared with cartilage overlying these areas, where we identified 1,397 differentially methylated sites. Forty-four percent of cartilage genes were also differentially methylated in subchondral bone.

Hierarchical clustering of differentially methylated sites divided samples into 3 groups with distinct histopathologic (Mankin) scoring characteristics in the overlying cartilage (Figure 1). This supports the notion that epigenetic alterations may occur as OA progresses. Second, we observed differential methylation of OA susceptibility genes in both subchondral bone and cartilage. Third, network analysis identified pathways and upstream regulators that were enriched in differentially methylated genes; many of these were shared by differentially methylated genes in both subchondral bone and overlying cartilage (Table 4).

We observed novel differential methylation of several genes involved in OA pathogenesis. The most highly hypomethylated gene was EIF2C2, which was not previously associated with OA. This gene encodes argonaute 2 protein and is a core member of RISC, which processes small interfering RNAs (9). Given a plethora of recent data regarding differential expression of miRNAs involved in both chondrogenesis and OA (10–12), particularly the recent report of peripheral circulating let-7e as a potential biomarker of severe disease (10), epigenetic dysregulation of this RISC gene is quite interesting.

TGFB3, a multipurpose growth-modulating factor, was hypomethylated. The protein encoded by this gene plays an important role in both chondrogenesis and endochondral bone formation (13), along with mesenchymal cell proliferation and angiogenesis (14). Furthermore, it was demonstrated in a leporine model to be a potential therapeutic agent for the treatment of OA, through increased homing of endogenous cells to the damaged articular surface (15). NFATC1, a key suppressor of OA, was hypomethylated. This is in stark contrast to hypermethylation seen in cartilage, which is consistent with previous studies demonstrating reduced NFATC1 expression in OA cartilage (7). NF-ATs, in conjunction with osterix, induce bone formation by osteoblasts (16). Perhaps hypomethylation of NFATC1 represents a reactive, sclerotic response to overlying cartilaginous damage. Alternatively, it could represent T cell infiltration into subchondral bone. The hypomethylated gene HDAC4 has been previously associated with OA; specifically, decreased levels of HDAC4 with subsequent increases in expression of the key OA transcription factor RUNX-2 have been noted (17). Furthermore, expression of HDAC4 is negatively correlated with the Mankin score (18). Overexpression and knockdown experiments targeting HDAC4 in primary human chondrocyte cultures have further demonstrated that underexpression of HDAC4 increases a variety of OA-associated metalloproteinases (18).

The most highly hypermethylated gene was FLJ42709, which is a large intergenic antisense interfering RNA that disrupts transcription of NR2F1. The product of this gene forms a regulatory network with miR-140 (19), which was previously linked to OA (11). Hypermethylated HOXA7 is a regulator of extracellular matrix formation in monocytes (20) and dimerizes with PBX1, a gene that is involved in osteoblastogenesis (21); this is interesting in light of the sclerotic response seen in the subchondral bone of patients with OA. Hypermethylated CD6 has not previously been associated with OA. CD5 has been previously described as being overexpressed in the synovial fluid of patients with advanced OA (22) but has not been described previously in subchondral bone. Finally, overexpression of hypermethylated HOXA2 in murine models results in chondrodysplasia and delayed cartilage maturation (23); similarly, its expression can delay endochondral ossification (24).

Pathway analysis revealed several interesting findings. First, a striking TGFβ1 signature was shared among differentially methylated genes in both subchondral bone and overlying cartilage. Recent studies have highlighted the contributions of subchondral bone and TGFβ1 particularly to OA pathogenesis. Subchondral bone–derived mesenchymal stem cells (MSCs) are intimately linked to the formation of osteophytes (25). A 2013 study by Zhen et al demonstrated that blocking TGFβ signaling in subchondral bone MSCs can prevent the development of OA in a mouse model of anterior cruciate ligament transection (26); they also observed that TGFβ1 concentrations were elevated in the synovial fluid of patients with OA. They went on to report a transgenic mouse model overexpressing TGFβ1 in osteoblasts that develop spontaneous OA characterized by both subchondral bone sclerosis and cartilage degradation.

The TNF family has diverse roles in OA, activating catabolic matrix metalloproteinases and nitric oxide synthesis (27,28) as well as having angiogenic and pain-stimulation effects. TNF is known to stimulate RANKL-induced osteoclastogenesis (29), which may have importance in the subchondral cysts and bone marrow edema-like lesions associated with accelerated cartilage loss in OA patients (30). Furthermore, the presence of a variety of pain-related molecules, including TNF, has previously been demonstrated in the subchondral bone of patients with knee OA (31). TNF has been proposed as a genetic risk factor in knee OA (32), and RNA interference–induced silencing of TNF has been demonstrated as a therapeutic in a leporine model (33). However, the results of human trials with TNF monoclonal antibodies in the treatment of human OA have been mixed (34–36). We identified a strong demethylated TNF signature in both cartilage and subchondral bone, reiterating the importance of this pathway in both tissue types.

Differential methylation of genes involved in several pathways was shared by subchondral bone and cartilage. The ERK/MAPK pathway has been proposed as a possible therapeutic target in human OA. MEK/ERK inhibitors exhibit antimetalloproteinase and antiinflammatory activity themselves (37) and act synergistically when combined with hyaluronic acid (38). The IL-4 pathway was significantly associated with differentially methylated genes in both subchondral bone and cartilage. Elevated levels of soluble IL-4 receptor have been observed in the serum of OA patients (39), and an IL-4/IL-10 system was recently demonstrated in vivo to reduce expression of matrix metalloproteinases and induce proteoglycan synthesis (40). IGF-1 is an anabolic factor that previously was shown to be increased in OA cartilage and synovial fluid (41), although OA chondrocytes are hyporesponsive to its effects (42), due at least in part to localized oxidative stress and increases in reactive oxygen species (43). Our epigenetic findings are consistent with this result, because we identified the IGF-1 receptor, along with downstream targets AKT and MEK, to be hypomethylated. A previous study identified hypomethylation of IGF1R in patients with hip OA (4).

We found evidence of differential methylation of 4 susceptibility genes in both OA subchondral bone and OA cartilage. IGFBP7 has been previously reported to be differentially methylated in hip OA compared with femoral neck fractures and is a genetic susceptibility locus for both knee and hand OA (44). LRP5 was also differentially methylated in both OA cartilage and subchondral bone. This gene is differentially methylated in patients with hip OA compared with controls and is overexpressed in hip OA cartilage (4). Several studies have shown a genetic susceptibility locus in OA in close proximity to LRP5 (45). LRP5 acts as a co-receptor, which in combination with Frizzled transduces canonical Wnt/β-catenin signaling; alterations induced by either genetic mutation or differential DNA methylation may have deleterious effects on bone density in patients with OA. The genetic effect of FTO in patients with OA appears to be mediated through alterations in body mass index (46) and has been demonstrated in a number of tissues, including peripheral blood mononuclear cells (47). Finally, NCOR2 has been demonstrated to be a susceptibility gene in patients with knee OA (48) and is differentially methylated in knee OA cartilage (49).

The major strengths of our study include the precise matching of subchondral bone and overlying cartilage, utilizing femoral heads collected for our previous epigenetic analysis (3) and the identical analytic workflow used to identify differentially methylated genes and ontologies for both as well as the use of internal controls (intact sections within the same joint), reducing the possibility of false-positive results attributable to genetic variation or demographic mismatch.

Nonetheless, our study has a few weaknesses. First, we examined paired samples of eroded and intact cartilage from OA patients but did not compare these with subchondral bone from healthy controls, because we did not have access to this tissue. It would seem inappropriate to use the OA “standard” control samples obtained from patients with femoral neck fractures when comparing bone, because this tissue would no doubt have alterations related to underlying osteoporosis and possibly traumatic stress/healing responses. In the present study, the degree of osteoporosis in the comparison groups is identical, owing to our internal control design. In a 2012 study focused on differential methylation in osteoporotic subchondral bone, Delgado-Calle et al compared 27 patients with femoral neck fractures with 26 patients with hip OA and identified 241 differentially methylated sites associated with osteoporosis using the OA patients as controls (50). Upon close examination, we did not identify differential methylation in any of the 241 specific CpG sites nor other sites among the 228 genes they described, which was likely driven by the nature of the intraarticular experimental design that we used. Second, although overlying cartilage was completely excluded during sample acquisition, it is nonetheless possible that our findings were influenced by alterations in cellular populations within the underlying subchondral bone. At present, mechanisms to study individual cellular populations in these samples without the risk of ex vivo alteration of epigenetic signatures (i.e., explant culture) are limited. Nevertheless, our findings offer a first glimpse into differential DNA methylation patterns in the subchondral tissue of OA patients in toto.

Our experimental findings must be considered in light of the mixed subchondral cellular population we examined. The alterations in methylation we describe may be a result of the disease process itself, particularly inflammatory and compensatory microenvironmental reactions to overlying damage, rather than a principal driver of disease pathogenesis itself. Nonetheless, the epigenetic phenotype we describe will be useful in the planning of future experiments profiling subchondral bone in OA.

In summary, our present set of experiments represent the first genome-wide DNA methylation study in the subchondral bone of patients with hip OA. In this study, we identified differentially methylated CpG sites in bone underlying eroded cartilage compared with intact cartilage, and compared this with patterns observed in the overlying cartilage. We observed a number of differentially methylated CpG sites in both novel and previously associated genes. Ontologic analysis revealed differential methylation of several pathways and upstream regulators shared between cartilage and subchondral bone and reiterated the presence of an epigenetic phenotype of TGFβ and TNF with OA. Finally, we observed a number of OA susceptibility genes that were differentially methylated in both subchondral bone and overlying cartilage, with 4 genes being shared between the 2 tissue types. These findings offer further evidence that OA is a complex disease involving genomic–epigenomic interactions beyond the articular surface.