Changes in DNA methylation accompany changes in gene expression during chondrocyte hypertrophic differentiation in vitro.

Abstract

During osteoarthritis (OA), articular chondrocytes undergo phenotypic changes that resemble developmental patterns characteristic of growth plate chondrocytes. These phenotypic alterations lead to a hypertrophy-like phenotype characterized by altered production of extracellular matrix constituents and increased collagenase activity, which in turn results in cartilage destruction in OA disease. Recent studies have shown that the phenotypic instability and dysregulated gene expression in OA are associated with changes in DNA methylation patterns. Subsequent efforts have aimed to identify changes in DNA methylation with functional impact in OA disease, to potentially uncover therapeutic targets. Here, we paired an in vitro 3D/pellet culture system that mimics chondrocyte hypertrophy with RNA sequencing (RNA-Seq) and enhanced reduced representation of bisulfite sequencing (ERRBS) to identify transcriptomic and epigenomic changes in murine primary articular chondrocytes undergoing hypertrophy-like differentiation. We identified hypertrophy-associated changes in DNA methylation patterns in vitro. Integration of RNA-Seq and ERRBS datasets identified associations between changes in methylation and gene expression. Our integrative analyses showed that hypertrophic differentiation of articular chondrocytes is accompanied by transcriptomic and epigenomic changes in vitro. We believe that our integrative approaches have the potential to uncover new targets for therapeutic intervention.

Graphical Abstract

The phenotypic instability and dysregulated gene expression of articular chondrocytes in osteoarthritis (OA) are associated with changes in DNA methylation patterns. Here, we (A) used 3D/pellet culture systems to (B) induce hypertrophy in vitro and (C) perform RNA sequencing (RNA-Seq) and enhanced reduced representation of bisulfite sequencing (ERRBS) paired with (D) integrative bioinformatic analyses comparing non-hypertrophic and hypertrophic cells to identify and interpret transcriptomic and epigenomic changes associated with the hypertrophy-like differentiation of murine primary articular chondrocytes. Created with BioRender.com

Introduction

Osteoarthritis (OA) is the most common degenerative joint disorder and a leading cause of pain and disability worldwide, associated with an immense socioeconomic burden. Currently we lack efficacious nonsurgical treatments for OA in part because of our lack of understanding of disease mechanisms. OA is a multifaceted disease that involves and impacts all joint tissues and leads to the irreversible loss of structural and functional integrity of articular cartilage. Articular chondrocytes are the unique cellular component of articular cartilage, responsible for maintaining the structural and functional integrity of cartilage with a fine-tuned balance between anabolism and catabolism^{1, 2}.

In OA, in response to local and systemic injurious stimuli, the normally quiescent articular chondrocytes become activated and undergo phenotypic instability, which contributes to the irreversible damage of articular cartilage. OA chondrocytes acquire a hypertrophy-like phenotype characterized by the abnormal expression and activities of Mmp13, Runx2, and type X collagen, mimicking developmental events that lead to abnormal matrix remodeling. Therefore, a better understanding of the mechanisms that contribute to the phenotypic stability of articular chondrocytes and that, when dysregulated, lead to its hypertrophy-like conversion is required to understand OA disease^{2, 3}.

DNA methylation is one of the principal mechanisms by which cells maintain stable phenotypes^{4, 5}. Recent studies have highlighted the contribution of DNA methylation to OA^6-9. Our previous studies have linked changes in DNA methylation with the differential expression of chondrocyte-specific genes, matrix metalloproteinases (MMPs), and transcription factors in articular chondrocytes^10-13. Work with human cartilage identified subsets of OA patients defined by changes in DNA methylation patterns, uncovered the enrichment of pathways relevant to the development of the skeletal system in OA, and revealed the impact of methylation-dependent regulation of hypertrophy-related genes during OA disease^13-17. Furthermore, different studies highlighted the impact of DNA methylation and other epigenetic mechanisms during chondrogenesis and endochondral ossification processes^18-22. While these studies emphasize the relevance of DNA methylation to OA disease and chondrogenesis, how changes in DNA methylation patterns govern or impact hypertrophy-like changes in articular chondrocytes is not clearly understood.

We hypothesized that hypertrophy-like changes of articular chondrocytes are accompanied by changes in DNA methylation patterns. To test our hypothesis, we interrogated murine primary chondrocytes in an in vitro 3D/pellet system paired with transcriptomic and epigenomic analyses. Consistent with work in chondrogenic models, we found that the differentiation of murine articular chondrocytes towards hypertrophy is accompanied by changes in DNA methylation patterns. Integrating our transcriptomic and epigenomic datasets, we identified 143 genes that were differentially expressed and associated with differential methylation in hypertrophic chondrocytes along with 161 overlapping GO categories that includes unique and overlapping differentially methylated and expressed genes. Together, our findings provide further support for the contribution of DNA methylation to phenotypic alterations of articular chondrocytes.

Materials and methods

Primary mouse chondrocyte isolation and culture

Primary mouse chondrocytes were isolated from femoral head and tibial plateaus of 5- to 6-day old C57BL/6J mice by sequential digestion with collagenase D, as described²³. Murine cells were expanded in complete medium (DMEM/F12 + 10% FBS + 1% penicillin/streptomycin) and used for experiments between passages 2 and 3. For hypertrophic differentiation, cells were grown in pellet cultures for 1 and 3 weeks, as described^{24, 25}. Experiments with demethylating agents were conducted as described²⁶. Briefly, monolayer primary chondrocyte cultures were either incubated with vehicle or with a combination of 2μM 5-aza-2′-deoxycytidine (5-aza) and 40 nM trichostatin (TS) for 1 week. The concentrations of 5-aza and TS were selected in dose–response experiments (not shown). At 1 week after treatment, RNA was isolated for RT-qPCR analyses of selected genes.

Histology and immunohistochemical analyses

Safranin O and toluidine blue stains were performed on 5 μm frozen sections of 3-week pellet cultures, as indicated^{24, 25}. Immunofluorescence analyses were performed on frozen sections (5 μm) of 3-week pellet cultures using specific primary antibodies against type X collagen (Abcam), followed by incubation with Alexa Fluor®555-conjugated secondary antibodies (CST), as described²⁵. Sections were mounted with ProLong Gold antifade mounting medium containing DAPI (ThermoFisher). Adjacent sections were used for C1,2C immunostaining (IBEX Inc.), also as described²⁵. The resulting histological and immunohistochemical images were visualized with a Nikon Eclipse Ni-E motorized microscope.

RNA isolation

Total RNA was isolated using TRIzol followed by the RNeasy mini RNA isolation kit (Qiagen). Briefly, two cell pellets were pooled together to obtain one RNA sample. Pellets were homogenized in TRIzol using a motorized mortar and pestle. Homogenized cell pellets were mixed with chloroform: isoamyl alcohol (24:1), incubated on ice for 5 min, and centrifuged for 10 min for separation of phases. From this point onwards, RNA was purified following the RNeasy instructions. For total RNA isolation from monolayer cultures, cells were lysed in RLT buffer containing 2-mercaptoethanol, and total RNA was isolated using the RNeasy mini RNA isolation kit (Qiagen). RNA integrity was assessed at the Core Laboratories Center of Weill Cornell Medicine. For RNA-Seq and RT-qPCR analyses, we used total RNA with RIN >7 and OD_260/280 >1.8.

DNA isolation

Cell pellets were homogenized using a motorized mortar and pestle, and DNA was isolated using the Gentra Puregene DNA isolation kit (Qiagen) with minor modifications. Briefly, homogenized cells were lysed in lysis buffer supplemented with proteinase K. Lysed samples were incubated with RNase for 45 min at 37 °C, followed by protein precipitation. For DNA precipitation, samples were incubated in isopropanol at −20 °C for 90 minutes, centrifuged at 13,000 × g for 10 min, and washed 3 times in 70% ethanol. Finally, the DNA pellets were reconstituted in DNA hybridization buffer. DNA used for ERRBS had a OD_260/280 >1.8, and quality control steps were done at the Weill Cornell Medicine Epigenomics Core. Library preparation, sequencing, and post-processing of the raw data was performed at the Epigenomics Core at Weill Cornell Medicine.

Enhanced reduced representation bisulfite sequencing (ERRBS)

ERRBS library preparation and sequencing was performed at the Epigenomics Core at Weill Cornell Medical College as described previously²⁷. This is a modification of the reduced representation bisulfite sequencing approach, which generates single-base resolution DNA methylation bisulfite sequencing libraries that enrich for CpG-dense regions by methylation-insensitive restriction digestion²⁸. Briefly, DNA digested with the restriction enzyme MspI (New England Biolabs) was phenol extracted, ethanol precipitated, end repaired, A tailed, and ligated to methylated TruSeq adapters (Illumina Inc. San Diego, CA). Two size fragment-lengths of 150–250 bp and 250–400 bp were then gel isolated and subjected to overnight bisulfite conversion using the EZ DNA methylation kit (Zymo Research, Irvine CA), purified, and PCR amplified using TruSeq primers (Illumina Inc. San Diego, CA). The resulting libraries were normalized, pooled, multiplexed, and clustered on single read flow cells and sequenced for 50 cycles on an Illumina HiSeq 2500. Illumina’s CASAVA 2.17 software pipeline was used to perform image capture, base calling, and demultiplexing of raw reads to produce FASTQ files. The quality of the raw sequence reads was assessed with FastQC (version 0.10, Babraham Bioinformatics) and the FASTX toolkit (version 0.0.13, http://hannonlab.cshl.edu/fastx_toolkit/). Raw reads were then adapter trimmed, aligned (to the mm10/GRC38.p3 build of the mouse genome) and postprocessed to produce methylation calls at base-pair resolution using a previously described in-house pipeline²⁷.

Differential methylation analysis

CpG sites in the resulting ERRBS data were interrogated for methylation patterns and differential methylation (q value < 0.05 and methylation percentage difference of at least 25%) using methylKit²⁹. The differential methylation data was then queried for differentially methylated regions (DMRs) using eDMR³⁰. DMRs were annotated for proximal genes. Downstream statistical analyses and plots were generated using the R software environment for statistical computing (https://www.r-project.org/).

RNA sequencing (RNA-Seq)

RNA-Seq library preparations were done using established Illumina methods for mRNA-Seq (Part #RS-122-2001). Briefly, poly-A+ RNA was purified from 100 ng of total RNA with oligo-dT beads. Purified mRNA was fragmented with divalent cations at elevated temperature, to ~200bp. First-strand cDNA synthesis was performed with random hexamer priming and SuperScript II reverse transcriptase (Invitrogen, part # 18064-014). Second-strand cDNA synthesis was performed using RNAse H and DNA pol I. Following dscDNA synthesis, the double-stranded products were end repaired, followed by addition of a single A base and then ligation of the Illumina TruSeq adaptors. The resulting product was amplified with 15 cycles of PCR. Each library was made with a unique index sequence and libraries were pooled for sequencing. The pool was clustered at 6.5pM on a single-end-read flow cell and sequenced for 50 cycles on an Illumina HiSeq 2500. Base call files generated from the sequencer were demultiplexed and converted to FASTQ files using the Illumina CASAVA 2.17 pipeline. The quality of the raw sequence reads was assessed with FastQC (Version 0.10, Babraham Bioinformatics) and the FASTX toolkit (version 0.0.13, http://hannonlab.cshl.edu/fastx_toolkit/).

Differential expression analysis

RNA-Seq reads passing Illumina’s purity filter were adapter trimmed using FLEXBAR barcode and adapter processing tool³¹. The trimmed reads were aligned to the Gencode (https://www.gencodegenes.org/) mm10/GRC38.p3 build of the mouse genome using STAR aligner with default parameters³². Read counts for Ref-Seq (NCBI) transcripts were then quantified from the alignments using the featureCounts software package³³. Significant differential expression (adjusted P value < 0.05) was assessed between the read counts for the 1-week and 3-week sample groups using the DESeq2 R Bioconductor package³⁴.

Functional interpretation

Gene Ontology (GO) enrichment (pAdjustMethod = “BH”, pvalueCutoff = 0.01, qvalueCutoff = 0.05) analysis for biological processes (BP), cellular components (CC), and molecular functions (MF) was performed for differentially expressed genes and genes associated with differentially methylated regions using the clusterProfiler Biocondutor R package³⁵. Downstream statistical analyses and plots were generated using the R software environment for statistical computing (https://www.r-project.org/).

Comparison with human datasets

The biomaRt R interface for BioMart was used to interrogate genes with differential expression (DEGs) and/or associated differential methylation (DMRs) for human orthologs. DEGs and DMR-associated genes were grouped by gene symbol. Using human orthologs, DEGs and DMRs were compared with OA-relevant genes from the Human Genome Epidemiology Network (HuGENet) and selected published datasets^{6, 14, 36-39}. VennDiagram and UpSetR (R packages) were used to represent intersections with DEGs and DMRs between human OA data and datasets generated from this study.

Quantitative reverse transcription PCR (RT-qPCR)

Total RNA was reversed transcribed using the iscript2 reverse transcriptase kit (Bio-rad). Gene amplifications were carried out using SYBR Green I-based real-time PCR as described^{40, 41}, using specific primers against Cadm1, Col2a1, Col10a1, Mmp13, Mmp14, Pth1r, Runx2, and Sox9 (Table 1). The data were calculated as the ratio of each gene to Hprt1 using the 2^−ΔΔCt method for relative quantification. Eef1a1 was used as an additional housekeeping gene in control experiments, but not used as a normalizer in the final analyses.

Table 1.

Primer sequences

Gene	Forward (5'>3')	Reverse (5'>3')	Annealing temp. (°C)
Cadm1	GATCCCCACAGGTGATGGAC	TGATGGTTGCCACTTCTCCTT	65
Col10a1	ACGCATCTCCCAGCACCAGAATC	GGGGCTAGCAAGTGGGCCCT	60
Col2a1	CGATCACAGAAGACCTCCCG	GCGGTTGCAAAGTGTTTGGC	60
Eef1a1	AGTCGCCTTGGACGTTCTTT	TGGACTTGCCGGAATCTACG	60
Hprt1	TCCCAGCGTCGTGATTAGCGA	GGGCCACAATGTGATGGCCTCC	60
Mmp13	ATGGTCCAGGCGATGAAGACCCC	GTGCAGGCGCCAGAAGAATCTGT	60
Mmp14	AACCATGATGGCCATGAGGCGC	ACTCGCCCACCTTAGGGGTG	60
Pth1r	CAGGCGCAATGTGACAAGC	TTTCCCGGTGCCTTCTCTTTC	65
Runx2	TCCCCGGGAACCAAGAAGGCA	AGGGAGGGCCGTGGGTTCTG	60
Sox9	AAGCTCTGGAGGCTGCTGAACGAG	CGGCCTCCGCTTGTCCGTTCT	60

Statistical analysis

Statistical analyses of the RT-qPCR data were performed using GraphPad Prism 8 Software (GraphPad Software, Sand Diego, CA). Data are reported as means ± S.D. of at least three independent experiments. Statistical analysis was performed by Student’s t-test with P values of <0.05 considered significant.

Data availability

The data that supports the findings of this study are available from the corresponding author upon reasonable request. The RNA-Seq and ERRBS sequencing data have been deposited at the GEO database with accession code GSE154949.

Results

Hypertrophy-like differentiation in vitro

To analyze how transcriptomic and epigenomic changes correlate with hypertrophy-like differentiation of articular chondrocytes, we cultured murine primary chondrocytes for 1 and 3 weeks in 3D/pellet cultures using previously established protocols^{24, 25}. As expected, cells cultured in pellets produced a proteoglycan-rich, cartilage-like matrix, as shown by the intense and homogeneous safranin-O- and toluidine blue-rich staining (Fig. 1A,B). Immunohistochemical analyses of 3-week pellet cultures using antibodies against type X collagen revealed the production of a type X collagen–rich matrix characteristic of hypertrophic chondrocytes (Fig. 1C). Using the C1,2C antibody, which specifically recognizes the Col2 3/4C_short neoepitope, we confirmed the presence of collagenase activity in the 3-week pellet cultures (Fig. 1D). RT-qPCR analyses showed no differences in Col2a1 and Sox9 mRNA between 1 and 3 weeks, indicating that the 3-week cells retained expression of chondrocyte markers. Additional RT-qPCR analyses confirmed hypertrophy-like changes, with increased expression of Runx2, Col10a1 and Mmp13 mRNA over time (Fig. 1E). These observations were consistent with the expected time-dependent hypertrophy-like changes in pellet cultures of murine chondrocytes. Given these results, we used RNA isolated from 1- and 3-week pellet cultures (n = 3/each) for epigenomic and transcriptomic analyses.

Figure 1.

Hypertrophy-like changes in pellet cultures of murine chondrocytes in vitro. Representative (A) toluidine blue– and (B) safranin O–stained pellet cultures (4×). (C) Type X collagen (Col10) immunostaining (20×) confirming hypertrophy-like changes in 3-week pellets (inserts, isotype-stained negative controls). (D) C1,2C immunostaining (20×) showing matrix remodeling in 3-week pellet cultures (insert, isotype negative control). (E) RT-qPCR analyses of Col2a1, Sox9, Runx2, Col10a1, and Mmp13 mRNA expression in RNA isolated from 1- and 3-week pellet cultures (n = 3/each). *P < 0.05 and **P < 0.01 (by t-test). ns = not significant.

Transcriptomic changes associated with hypertrophic differentiation in vitro

We next evaluated transcriptomic changes in chondrocytes undergoing hypertrophic differentiation by comparing 3-week and 1-week pellet cultures. Our RNA-Seq analyses identified a total of 4500 differentially expressed genes (DEGs) in 3-week pellet cultures relative to 1-week controls (see Table S1 (online only) for a list of all DEGs). Consistent with the hypertrophy-like differentiation of 3-week pellets, we detected increased expression of Col1a1, Runx2, Ihh, Col10a1, Mmp13, Bmp2, and Alpl^42-53. See Table 2 for a summary of these genes. In addition to the enrichment of classic genes expressed by growth plate hypertrophic chondrocytes, we also identified changes in genes with potential relevance to OA disease and cartilage reparative responses, including Grem1⁵⁴, Il17b^55,56, and Trem1⁵⁷. See Figure 2A for a volcano plot representation of the comparison of differential expression, indicating DEGs (FDR <0.05) in 3-week pellets relative to 1-week controls (in red) and highlighting selected genes with known contribution to hypertrophy and potential significance to OA disease. We next performed functional enrichment analyses, represented in the cluster profile dot plot that depicts biological processes (BPs), cellular components (CCs), and molecular functions (MFs) for the 40 top gene ontology (GO) categories significantly enriched (Fig. 2B). Among the top BPs, we found small GTPase–mediated signal transduction, ossification, and regulation of mitotic cell cycle. CCs relevant to chondrocyte hypertrophy and ossification were also enriched in our dataset, including extracellular matrix (ECM), actin cytoskeleton, collagen-rich ECM and cell–cell junctions. Top MFs enriched in our dataset included actin- and signal transduction-related pathways. Together, our RNA-Seq data confirmed the utility of this 3D/pellet culture system to evaluate hypertrophy-like changes and OA-relevant genes in vitro.

Table 2.

Selected hypertrophy-related differentially expressed genes in the RNA-Seq dataset

Gene ref ID	Gene ID	Reference	Log fold change	Adjusted P value
NM_007431	Alpl	37, 42	4.282441	2.28E-238
NM_007553	Bmp2	43, 45	1.136359	9.84E-07
NM_007742	Col1a1	44, 76	2.134473	5.27E-13
NM_009925	Col10a1	39	4.004282	2.73E-40
NM_010544	Ihh	39, 46	4.65709	8.56E-56
NM_008607	Mmp13	38, 41, 47	1.998068	1.36E-62
NR_073392	Runx2	39, 40	1.726714	1.84E-39

Figure 2.

Transcriptomic changes in hypertrophic chondrocytes in vitro. (A). Volcano plot representation of differentially expressed genes (red, adjusted P value < 0.05) identified by RNA-Seq analyses comparing RNA isolated from 3-week pellets versus 1-week controls (n = 3/each). (B) Dot plot representation of the top 40 gene ontology (GO) categories enriched during differential expression, separated into biological processes (BPs), cellular components (CCs) and molecular functions (MFs).

Changes in DNA methylation associated with hypertrophic differentiation in vitro

To test the hypothesis that the hypertrophy-like changes in OA are part of the loss of phenotypic stability of chondrocytes in an attempt to recapitulate developmental and/or reparative processes ^{2, 58}, we next evaluated changes in 5-methylcytosine (5mC) in DNA isolated from pellet cultures maintained in hypertrophic medium for 3 weeks and compared these changes with week 1 samples (n = 3/each). Comparisons between week 3 and week 1 samples uncovered significant differences in hyper- and hypomethylation (Fig. S1, online only). Using at least a 25% methylation difference (and adjusted P value or q value <0.05) between week 3 and 1, we identified 10,929 significantly differentially methylated CpGs (DMCs). Figure 3A depicts the percentage distribution of these DMCs in the gene body and promoter regions. Next, we aimed to identify significantly differentially methylated regions (DMRs). We defined DMR as a genomic region with at least 3 CpGs within 100 bp, where at least 1 CpG is significantly differentially methylated (25% methylation difference and a q value <0.01) and the region has an overall average differential methylation of at least 20% across all the CpGs in the region. Comparing 3-week versus 1-week pellets, we identified 634 DMRs associated with 462 genes. A detailed list of all significant DMRs is provided in Table S2 (online only). We next performed functional enrichment analyses of gene-associated DMRs. The cluster profile plot represented in Figure 3B depicts BPs, CCs, and MFs for the top 40 enriched GO categories. Ion transmembrane transport, cell substrate adhesion, and regulation of supramolecular fiber organization were among the top BPs with known relevance to hypertrophy. Actin cytoskeleton, receptor complex, and transporter complex were among the top CCs with known relevance to chondrocyte hypertrophic differentiation. Only six MFs were identified in this analysis, including transcription activator activity, RNA polymerase II-specific, enzyme activator activity, phospholipid binding, cell adhesion molecule–binding, calcium channel activity, and intracellular ligand-gated ion channel activity.

Figure 3.

Changes in DNA methylation patterns are associated with hypertrophic changes in chondrocytes. (A) Pie chart showing percentage distribution of differentially methylated CpGs (DMCs) in gene bodies comparing 3-week versus 1-week pellet cultures (n = 3/each). (B) Dot plot for biological processes (BPs), cell components (CCs), and molecular functions (MFs) for the top 40 significantly enriched GO categories of gene-associated DMRs.

Taken together, our ERRBS analyses show changes in DNA methylation patterns accompanying chondrocyte hypertrophy in vitro and revealed enrichment of functional categories with known contribution to chondrocyte hypertrophy in the 3-week pellet cultures.

Integrative comparative analysis of transcriptomic and epigenomic datasets

We next used an integrative comparative approach of the RNA-Seq and ERRBS datasets, aiming to identify differentially expressed genes that also display associated changes in methylation in 3-week pellet cultures relative to the 1-week controls. The Venn diagram in Figure 4A summarizes unique and overlapping DEGs and DMRs at 3 weeks. Combined integration of DEGs and DMRs in GO categories that include BPs, MFs, and CCs revealed unique and overlapping GO categories in hypertrophic chondrocytes (3-week pellets), with 2624 categories unique to DEGs, 1859 categories unique to DMRs, and 161 categories that are overlapping between the DEGs and genes associated with DMRs (Figure 4B). The overlapping categories include ossification, cell matrix regulation, calcium ion transport, actin filament organization, and embryonic skeletal system development. Table 3 summarizes the hypertrophy-related GO categories along with the overlapping DEG and gene-associated DMRs genes for each of these categories. Additional correlative analyses of DEGs and DMRs revealed 143 genes with significant differential expression that had associated differentially methylated regions (Fig. 4C), including metalloproteinases (e.g., Mmp14), transcription factors (e.g., Sox8) and growth factors and receptors (e.g., Fgf7 and Fgr2) with known roles in developmental and cartilage degradative processes^59-62. Next, we evaluated the impact of the DNA demethylating agent 5-aza on the expression of selected genes with negative correlation between expression and methylation (Fig. 4C, red). RT-qPCR analyses showed that the expression of Mmp14 and Pth1r significantly increased at 1 week after 5-aza treatment, whereas the expression levels of Cadm1 did not change in 5-aza–treated cells (Fig. 5). These results suggest that changes in DNA methylation can impact gene expression in hypertrophy-related genes and highlight the complex context- and gene-specific interaction between methylation and other transcriptional mechanisms.

Figure 4.

Integration of RNA-Seq and ERRBS datasets. (A) Venn diagram depicting unique and overlapping differentially expressed genes (DEGs) and differentially methylated regions (DMRs) obtained from a comparison between 3-week versus 1-week pellet cultures (n = 3/each) by RNA-Seq and ERRBS analyses, respectively. (B) Venn diagram representing unique and overlapping significantly enriched gene ontology (GO) categories identified using DEGs and DMRs. (C) Representation of the correlation between expression and methylation of 143 genes with differential expression and methylation, indicating (i) hypomethylated genes with increased expression, (ii) hypermethylated genes with increased expression, (iii) hypomethylated genes with decreased expression, and (iv) hypermethylated genes with decreased expression. Positive mean differential methylation indicates hypermethylated, and negative mean differential methylation indicates hypomethylated. Genes selected for in vitro assays with 5-aza are highlighted in red. Selected genes overlapping with HuGEnet (yellow) and human datasets comparing OA versus healthy cartilage (blue) are also shown.

Table 3.

Summary of enriched gene ontology (GO) categories with overlapping differentially expressed genes (DEGs) associated with differentially methylated regions (DMRs)

GO ID	Description	DEGs	Gene DMRs	Overlapping DEG and DMR genes
GO:0001503	Ossification	169	18	Wwox, Sox8, Mmp14, Smad3,Gnas, Fgfr2, Foxc1
GO:0001667	Ameboidal-type cell migration	147	19	Ago2, Sox8, Pxn and Prkce
GO: 0007015	Actin filament organization	147	18	Prex1, Asap3, Lmod1, Pxn, Prkce, Smad3, Akap2, Avil
GO: 0007254	Calcium ion transport	124	18	Tpcn1, Slc24a3, Dysf, Prkce, Prkcb
GO:0005543	Phospholipid binding	121	19	Prex1, Snx18, Dysf, Tiam1, Sh3pxd2a
GO: 0010769	Reg. of cell morphog. involved in diff.	111	15	Prex1, Tiam1, Lrp1, Syne1, Mgll
GO: 0048705	Skeletal system morphogenesis	98	14	Wwox, Prrx2, Mmp14, Smad3, Foxc1, Gnas,Fgfr2
GO: 0050839	Cell adhesion molecule binding	92	14	Cadm1, Pxn, Mmp14, Ctnnd2, Cd9
GO:0003002	Regionalization	91	17	Disp1, Celsr2, Smad3, Fgfr2
GO: 0048762	Mesenchymal cell differentiation	86	12	Tiam1,Sox8, Smad3, Fgfr2, Foxc1
GO:0005769	Early endosome	85	12	Tmem184a, Rapgef1, Pmepa1, Dysf, Lrp1, App
GO: 0006817	JNK cascade	57	12	Rapgef1, Tiam1, Flt4, App
GO:0001952	Regulation of cell-matrix adhesion	43	8	Peak1, Mmp14, Smad3, Lrp1, Cdkn2a
GO: 0048706	Embryonic skeletal system development	41	11	Mmp14, Smad3, Gnas, Fgfr2, Prrx2
GO:0001942	Hair follicle development	35	10	Psen1, Gnas, Fgfr2, Fgf7

Figure 5.

DNA demethylation affects gene expression in articular chondrocytes. RT-qPCR analyses of (A) Cadm1, (B) Mmp14, and (C) Pth1r mRNA expression in RNA isolated from monolayer cultures treated with vehicle (ctrl) or a combination of 2μM 5-aza-2′-deoxycytidine and 40 nM trichostatin (5-aza) for 1 week (n = 3/each). **P < 0.01 and ***P < 0.001 (by t-test). ns = not significant.

Comparative analyses with human datasets

To identify changes with potential relevance to human OA, we compared our datasets with available human datasets using HuGENet. These comparisons identified 63 overlapping mouse genes (with 61 human orthologs) out of 168 knee OA–associated human genes in HuGENet (Fig. 6A), including 3 genes with gene-associated DMRs (Smad3, Lrp1, and Pitx1). Smad3 and Lrp1 were also differentially expressed in our dataset (Fig. 4C). See Table S3 (online only) for the complete gene lists. Additional comparisons with published human OA versus non-OA data^{14, 36} uncovered 24 DEGs and 9 DMRs overlapping with our data, including Best3, Disp1, Runx1, Grb10, Dysf, Laptm5, Nfatc1, Rora, and Ulk4 (Fig. 6B). Four genes with associated DMRs (Disp1, Grb10, Dysf, and Laptm5) also displayed differential expression (Fig. 4C). Comparisons with eroded versus intact OA cartilage^{6, 37-39} uncovered 917 overlapping DEGs and 135 DMRs, with 52 DMRs associated with differential expression (Fig. 6C). See Figure S3 (online only) to visualize the intersections of these datasets.

Figure 6.

Comparative analyses with human datasets. Venn diagrams representing unique and overlapping differentially expressed genes (DEGs) and genes containing differentially methylated regions (DMRs) with (A) knee OA–associated genes in HuGENet, (B) OA versus non-OA human datasets^{14, 36}, and (C) eroded versus intact cartilage human datasets^{6, 37-39}.

Together, our analyses show that the transcriptomic changes concomitant with hypertrophic differentiation of articular chondrocytes in vitro are accompanied by changes in DNA methylation patterns.

Discussion

Chondrocyte hypertrophy is a necessary, transient step in growth plate chondrocytes during endochondral ossification processes, which leads to increased matrix remodeling and calcification⁶³. In OA disease, articular chondrocytes display phenotypic changes that mimic hypertrophy and are known to contribute to the progression of the disease^{2, 3}. Here, we used an in vitro model of chondrocyte hypertrophy paired with transcriptomic and epigenomic analyses to identify changes in DNA methylation patterns that accompany the hypertrophy-like conversion of articular chondrocytes. Our analyses uncovered 634 DMRs associated with 462 differentially expressed genes in hypertrophy-like articular chondrocytes. Integration of DEG and DMR GO analyses revealed 161 GO categories that overlapped or were enriched as a result of differential expression and differential methylation. Together, our multimodal analyses show that hypertrophy-like changes in articular chondrocytes are accompanied by epigenomic and concomitant transcriptional changes.

Growth plate chondrocyte hypertrophy is characterized by drastic phenotypic changes that involve decreased proliferation, increased cell size, modulation of mineralization, and apoptosis⁶³. Articular chondrocytes acquire a hypertrophy-like phenotype in OA, which is known to contribute to cartilage damage as reflected by the decreased disease progression that follows the pharmacological or genetic inhibition of Runx2^{64, 65}, Sik3⁶⁶, Chuk/Ikkα,⁴⁰ and Hif2α^{67, 68}. While growth plate hypertrophy is a perfectly organized process, hypertrophy-like changes in OA occur in an asynchronous and disorganized manner, with increased transcription of cartilage-specific anabolic genes often overlapping with increased expression of markers of hypertrophy^{2, 3, 58}. Our in vitro data (Figs. 1 and 2) agree with this observation and confirm the increased expression of hypertrophic (e.g., Col10a1, Mmp13, and Alpl) and pre-hypertrophic genes (e.g., Runx2 and Ihh) without detecting the decreased expression of Col2a1 or Sox9 mRNA that is often observed in hypertrophic growth plate chondrocytes. While this result likely reflects the presence of heterogeneous cells populations and an asynchronous differentiation process in the pellet cultures, it is also consistent with expression signatures reported in articular cartilage in vivo, where anabolic and hypertrophic genes are simultaneously expressed (reviewed in Ref. 2). This was also reported in recent single-cell RNA-Seq analyses of human OA primary chondrocytes⁶⁹ that uncovered the differential expression of genes included in our dataset, like Col10a1, Alpl, Tnfrsf12a, Il17b, and Pappa. Taken together, the transcriptomic analyses of our murine pellet cultures confirmed that this is a valid system to study hypertrophy-like changes in articular chondrocytes, with validity to establish parallels with gene signatures observed in OA disease.

Several studies have reported epigenetic changes in OA cartilage and the impact of epigenetic modulation on OA disease (reviewed in Refs. 2 and 8). However, relatively little is known about epigenetic changes that accompany or regulate chondrocyte hypertrophy in OA. Chondrogenesis in vitro is associated with changes in DNA methylation¹⁸, and a number of genes that participate in the chondrogenic and hypertrophic program and have known contributions to OA have been reported to be epigenetically regulated, including Runx2¹³ and Mmp13^{10, 70}. Furthermore, ablation of the DNA methyltransferase gene Dnmt3b results in catabolism and OA progression in mice⁹, and inhibition of the downstream target of Dnmt3b, 4-aminobutyrate aminotransferase, protects against surgical induction of OA⁷¹. These results represent strong evidence of the contribution of DNA methylation and epigenetic changes to chondrocyte hypertrophy and OA.

In our study, we identified 143 differentially expressed genes in hypertrophic cells with associated changes in DNA methylation patterns (Fig. 4C). Using DNA demethylating agents, we found that DNA demethylation impacted gene expression in a gene-dependent context (Fig. 5), likely interacting with other transcriptional mechanisms. Additional comparisons (Fig. 6, Table S3, and Fig. S3) uncovered overlapping changes in gene expression and DNA methylation between our data and OA-relevant human datasets. The relatively little overlap that we observed between DEGs and DMRs within our dataset could be partly explained by the use of articular chondrocytes and not chondrogenic cells, or by the asynchronous differentiation of primary chondrocytes in pellet cultures in vitro. Follow up studies that identify zonal changes in DNA methylation associated with transcriptional and phenotypic changes in growth plate chondrocytes undergoing hypertrophy could help us define the relationship between methylation and hypertrophy-like transcriptional changes in articular chondrocytes. Integration of our RNA-Seq and ERRBS datasets uncovered overlapping GO categories that comprise the NF-κB, Wnt, TGFβ, and Hedgehog signaling pathways, which are known to modulate hypertrophy and OA (for reviews, see Refs. 2, 58, and 72). The Wnt signaling downstream effectors Lef1 and Axin1, with well-documented impact in OA^{73, 74}, are differentially expressed and methylated in 3-weeks hypertrophic chondrocytes in our datasets. Mmp14, known to contribute to hypertrophy and OA^{60, 75}, also displayed differential expression and methylation in the 3-week hypertrophic murine pellet cultures. Similarly, the expression of Smad3 in hypertrophic chondrocytes is associated with DNA methylation. This could have implications for the balanced TGFβ signaling in chondrocytes, known to rely on SMAD2/3 and SMAD1/5/8 and to contribute to hypertrophy in OA cartilage⁷⁶.

Our results suggest that integration of transcriptomic and epigenomic analyses obtained using relevant in vitro systems can provide insight into the mechanisms that contribute to OA and can uncover novel therapeutic cues. Interestingly, Gene Set Enrichment Analyses (GSEA) performed across all the MsigDB gene sets using the clusterProfiler Bioconductor R package for the RNA-Seq data identified 33 enriched gene sets (Fig. S2, online only). Two of the enriched gene sets that we identified (MEISSNER_BRAIN_HCP_WITH_H3K4ME3_AND_H3K27ME3 and MIKKELSEN_MCV6_HCP_WITH_H3K27ME3) were of particular interest to us. Previously, it has been reported that the activities of the ESET and EZH2 histone methyltransferases that catalyze methylation of histone 3 at lysine 9 (H3K9) and trimethylation of histone 3 lysine 27 (H3K27me3), respectively, are required for optimal chondrocyte hypertrophy and skeletal development^{2, 77, 78}. We found that 150 genes from our data set overlap with a gene set reported by Meissner and colleagues, who subjected DNA isolated from embryonic stem cells and embryonic stem cell-differentiated neural precursor to a RRBS pipeline⁷⁹. Similarly, we identified 50 overlapping genes with a dataset reported by Mikkelsen et al.⁸⁰ that was obtained using mouse embryonic fibroblasts trapped in the differentiated state. These parallels further highlight the notion that lessons from developmental models can provide essential information to understand the mechanisms underlying the phenotypic dysregulation of OA articular chondrocytes^{2, 58}.

The integration of our RNA-Seq and ERRBS datasets and the comparisons with available human and murine data showed that DNA methylation can contribute to the dysregulated expression of hypertrophy-related genes and functional pathways in OA disease. Additional studies that dissect the functional impact of specific changes in DNA methylation patterns might help us uncover novel therapeutic approaches to prevent the loss of cartilage functional and structural integrity associated with chondrocyte hypertrophy in OA.

Supplementary Material

Table S2

Table S2. DMRs identified by ERRBS analyses in 3-week pellet cultures relative to 1-week controls.

Table S3

Table S3. Summary of the comparative analyses with human datasets.

Table S1

Table S1. DEGs identified by RNA-Seq analyses in 3-week pellet cultures relative to 1-week controls.

Figs S1-S3

Figure S1. Representation of the differential hyper- and hypo-methylation (%) per chromosome.

Figure S2. Representation of Gene Set Enrichment Analyses (GSEA) performed across all the MsigDB gene sets using cluster Profiler Bioconductor R package for the RNA-Seq obtained in pellet cultures.

Figure S3. UpSet plot representing the intersections of our dataset (pellet cultures of murine primary chondrocytes, 1 week and 3 weeks; DEG, differentially expressed genes; and DMR: gene-associated differentially methylated regions) with selected human datasets.