RNA-Seq analysis of chondrocyte transcriptome reveals genetic heterogeneity in LG/J and SM/J murine strains

Abstract

Objective

To investigate the transcriptomic differences in chondrocytes from LG/J (large, healer) and SM/J (small, non-healer) murine strains in an attempt to discern the molecular pathways implicated in cartilage regeneration and susceptibility to osteoarthritis (OA).

Design

We performed RNA-sequencing on chondrocytes derived from LG/J (n=16) and SM/J (n=16) mice. We validated the expression of candidate genes and compared SNPs between the two mouse strains. We also examined gene expression of positional candidates for ear pinna regeneration and long bone length QTLs that display differences in cartilaginous expression.

Results

We observed a distinct genetic heterogeneity between cells derived from LG/J and SM/J mouse strains. We found that gene ontologies representing cell development, cartilage condensation, and regulation of cell differentiation were enriched in LG/J chondrocytes. In contrast, gene ontologies enriched in the SM/J chondrocytes were mainly related to inflammation and degeneration. Moreover, we found that multiple validated genes vary in sequence between LG/J and SM/J in coding and highly conserved noncoding regions. Finally, we showed that most QTLs have 20–30% of their positional candidates displaying differential expression between the two strains.

Conclusions

While the enrichment of pathways related to cell differentiation, cartilage development and cartilage condensation infers superior healing potential of LG/J strain, the enrichment of pathways related to cytokine production, immune cell activation and inflammation entails greater susceptibility of SM/J strain to OA. These data provide novel insights into chondrocyte transcriptome and aid in identification of the quantitative trait genes and molecular differences underlying the phenotypic differences associated with individual QTLs.

INTRODUCTION

Articular cartilage is a highly specialized tissue with unique characteristics for weight-bearing and frictionless motion. Articular cartilage is avascular, aneural and hypocellular. These characteristics contribute to the limited self-renewal ability of cartilage. Moreover, traumatic and degenerative lesions of the cartilage can lead to degenerative joint diseases, such as osteoarthritis (OA). OA is the most common musculoskeletal disorder, affecting 10–12% of the global population[1]. The prevalence of OA in the U.S. alone is projected to be nearly 67 million by 2030 (25% of the adult population)[2]. Currently, treatment options are limited and no disease-modifying OA drugs are currently available.

Although primarily considered a disease of age-related wear and tear of the cartilage, increasing evidence suggests that OA has a strong hereditary component. Heritability and twin studies indicate a genetic influence in OA ranging from 39–65%[3]. Furthermore, less than 50% of individuals who experience cartilage damage go on to develop post-traumatic OA[4]. Thus, understanding the genetic differences that may lead to a greater intrinsic ability to regenerate cartilage or less susceptibility to injury and/or inflammation is essential to understand the etiology of OA and may help devise potential novel treatment options.

A variety of factors have limited the genetic studies of OA in humans. For example, lack of self-renewal potential of cartilage, lack of adequate controls, as well as unreliable injury or disease history. In contrast, numerous animal studies of full-thickness cartilage healing have been reported in mice[5] and other species[6]. Among all, genomic studies have highlighted the striking genetic homologies between mouse and human, which facilitate mouse as a model for human biology and disease[7]. Therefore, we focused on two mouse strains with distinct healing abilities to gain insights into transcript-level differences in their chondrocytes.

The LG/J (healer) and SM/J (non-healer) inbred mouse strains selected respectively for large and small body size[8], possess diverse metabolic and skeletal phenotypes, as well as varying abilities to repair damaged tissues. Compared with SM/J, LG/J mice can regenerate a 2 mm hole perforated in their ear-pinna[9, 10], full-thickness knee cartilage lesions[5], as well as exhibit protection from post-traumatic OA[11]. The process of tissue regeneration is characterized by spontaneous initiation of several genes and biological pathways[12]. Several studies have compared the gene expression profile in ear wound tissues between the healer and non-healer strains[13–16] and revealed about 600 candidate genes within the mapped quantitative trait loci affecting healing[16]. Analysis of the gene expression pattern in these strains through branched-chain DNA technology revealed 4 genes significantly correlated with both auricular and articular cartilage healing[17]. However, the material analyzed was prepared from multiple joint tissues instead of just from the articular cartilage. Thus, stronger evidence is essential to establish the strain differences in gene expression from the cartilage or chondrocytes.

Here, we performed RNA-sequencing (RNA-seq) to investigate transcript-level differences to discern the molecular pathways related to cartilage regeneration and OA susceptibility in LG/J and SM/J mouse strains.

MATERIALS AND METHODS

Mice

The Institutional Animal Care and Use Committee of Washington University approved this study. LG/J and SM/J mice were purchased from The Jackson Laboratory and were housed in groups separated by sex in individually ventilated cages under pathogen-free conditions with ad libitum access to standard mouse chow and water and a 12:12 h light:dark cycle. As some sexual dimorphism has been reported for these strains for some phenotypes[10], both male and female mice were used in this study.

Murine primary chondrocyte isolation and characterization

Primary chondrocytes were isolated from cartilage fragments from 3-week (±2 days) old LG/J (n=16) and SM/J (n=16) mice as described before with some modification[18]. Detailed description is provided in Supplementary Text. Chondrocytes were seeded in a 24-well plate at a density of 1.0×10⁵ cells/well and cultured with 10% fetal bovine serum (FBS; Gibco) in high glucose Dulbecco’s Modified Eagle’s Medium (DMEM; Gibco) plus 1% penicillin and streptomycin (10000 U/mL, Gibco) at 37°C and 5% CO₂. Afterwards, chondrocytes were collected without passaging following 4 days of culture. Images were taken prior to cell collection to visualize characteristic chondrocyte phenotype.

RNA preparation

Total RNA was extracted from chondrocytes using the RNeasy Mini Kit (Qiagen) according to supplied protocol. RNA samples with an RNA integrity number >9.5 were used for RNA-seq.

Assessment of cartilage specific genes in chondrocytes

We used real-time PCR to assess the expression of Col2a1, Acan and Sox9 in order to confirm that chondrocytes from both mouse strain exhibit similar gene expression profile. Real-time PCR was performed as described before[19] using custom-designed gene-specific primers (Table 1). The relative expression level of each gene was normalized to Gapdh and calculated using the 2^−ΔΔCt method.

Table 1:

Characteristics of primers used for real-time PCR

Gene symbol	Gene name	Accession #	Forward primer (5’–3’)	Location		Reverse primer (5’–3’)	Location		Amplicon size (bp)
				from	to		from	to
Rflna	refilin A	NM_028443.3	catcgaggtggacccagaac	366	385	ccgccactatggtctcactg	505	486	140
Efemp1	EGF containing fibulin extracellular matrix protein 1	NM_146015.2	gggtatcagaagcgaggtga	739	758	ctgcactggcagtagaaggaa	848	828	110
Bmp6	bone morphogenetic protein 6	NM_007556.3	agtcttgcaggagcatcagc	1042	1061	ggtgtcaccacccacagatt	1170	1151	129
Ptafr	platelet-activating factor receptor	NM_001081211.2	gactccaccaacctagtgcc	720	739	acgaggaagaagcagaaggc	847	828	128
Fcgr3	Fc receptor, IgG, low affinity III	NM_010188.5	tgtcaaatggagcagacccg	360	379	cgtgatggtttccccttcca	470	451	111
Cd300ld	CD300 molecule like family member d	NM_145437.2	cgaaaggtggacctgatccc	408	427	gtgttctcagcgcttggttg	528	509	121
Mmp12	matrix metallopeptidase 12	NM_008605.3	gtgcccgatgtacagcatct	343	362	atcctcacgcttcatgtccg	450	431	108
Hyou1	hypoxia up-regulated 1	NM_021395.4	tggcgtgctcagtttagaca	1757	1776	cacctccaaacaggctggat	1875	1856	119
Wisp2	WNT1 inducible signaling pathway protein 2	NM_016873.2	cccaggagaatacaggtgcc	751	770	caggagtgacaagggcagaa	874	855	124
Bmf	BCL2 modifying factor	NM_138313.4	caggaagacaaggccactca	475	494	aaactggcagggagaggaag	611	592	137
Tgfa	transforming growth factor alpha	NM_031199.4	agccagaagaagcaagccat	556	575	caccactcacagtgtttgcg	674	655	118
Tnfrsf23	tumor necrosis factor receptor superfamily, member 23	NM_024290.4	gttgcaagacctgtccctca	165	184	ccctgtgaatgttcctgggt	265	246	101
Mybphl	myosin binding protein H-like	NM_026831.1	actgcactacagtcactggc	895	914	gtacttggggttgccttgga	1010	991	116
Lefty1	left right determination factor 1	NM_010094.4	aggactcagtatgtggccct	260	279	actagcaggtgagtggaggt	387	368	128
Rps26	ribosomal protein S26	NM_013765.2	ctgaagcaagcgtcttcgac	189	208	cgcgggatcgattcctaaca	296	277	108
Ifit3	interferon-induced protein with tetratricopeptide repeats 3	NM_010501.2	acagcatggaatgcccagaa	463	482	cactctgggtcctttggctt	586	567	124
Car2	carbonic anhydrase 2	NM_009801.5	tcaacaacggccactccttt	644	663	ccatcagatgagccccagtg	770	751	127
Tescl	tescalcin-like	NM_001163810.1	agacctggaagagcctcgaa	684	703	cttctccagtgtgggtgtgg	794	775	111
Rpl14	ribosomal protein L14	NM_025974.2	gggctttagtggatggaccc	126	145	ttcggacatacttctggcgg	245	226	120
Lyz1	lysozyme 1	NM_013590.4	aggatgacatcactgcagcc	341	360	cagactccgcagttccgaat	476	457	136
Gapdh	glyceraldehyde-3-phosphate dehydrogenase	NM_001289726.1	aggtcggtgtgaacggatttg	100	120	tgtagaccatgtagttgaggtca	222	200	123
Col2a1	collagen, type II, alpha 1	NM_031163.3	gggtatgacgagaaggctgg	787	806	caggttcaccaggattgcct	922	903	136
Acan	aggrecan	NM_001361500.1	cctgctacttcatcgacccc	539	558	agatgctgttgactcgaacct	688	668	150
Sox9	SRY (sex determining region Y)-box 9	NM_011448.4	cggctccagcaagaacaag	663	681	gcgcccacaccatgaag	724	708	62

PCR = polymerase chain reaction; bp = base pair

RNA-seq library preparation and analysis

10 ng RNA samples were prepared with the Clontech SMARTer system according to manufacturer protocol and sequenced across three 50-bp single-end Illumina HiSeq 3000 lanes. The resulting reads were aligned, quantified, filtered, and analyzed for differential expression as noted in the Supplementary Text and as done previously[19].

For each contrast extracted with Limma, global perturbations in known GO terms and KEGG pathways were detected using respectively the R/Bioconductor package GAGE (generally applicable gene set enrichment) and Pathview as detailed in Supplementary Text. RNA-seq data were deposited in the Gene Expression Omnibus (GEO) and is accessible through the accession number GSE128490 at https://www.ncbi.nlm.nih.gov/geo/.

Target gene selection and validation by real-time PCR

We used real-time PCR to validate the expression of a subset of genes differentially expressed between LG/J (n=8) and SM/J (n=8). Twenty target genes were selected based on their magnitude of expression and biological function in cartilage healing and degeneration. Real-time PCR was performed as described before[19] using custom-designed gene-specific primers (Table 1). The relative expression level of each gene was normalized to Gapdh and calculated using the 2^−ΔΔCt method.

Immunostaining

To further confirm the expression of two candidate genes namely Tnfrsf23 and Car2, we obtained articular cartilage from the femoral head of LG/J and SM/J mice (n=3 each). The samples were prepared for histology using a standard process. Paraffin sections were deparaffinized with xylene and rehydrated with gradual ethanol. Proteinase K (Abcam, 10 μg/mL for 15 minutes at 37°C) was applied to the sections for antigen retrieval. After washing with PBS and blocking with 10% normal goat serum (NGS), slides were incubated with the primary antibodies of Tnfrsf23 (1:100, Abcam), Car2 (1:100, Abcam), and in-house type-II collagen (Col2, 1:200) in 2% NGS overnight at 4°C. Then slides were washed and incubated with the corresponding Alexa 488 or Alexa 594 conjugated secondary antibodies in 2% NGS for 1 hour at room temperature and counterstained with Fluoro-Gel II with DAPI (Electron-Microscopy-Sciences). At least 3 sections from each mouse were stained. Images were visualized on a Confocal Laser Scanning Microscope (Leica). For quantification of immunofluorescence signals (represented as fluorescent intensity per cell), at least two fields were selected for each section and 30.79±3.82 chondrocytes per field were used for intensity calculation.

Single nucleotide polymorphism (SNP) analysis

For upregulated genes validated by real-time PCR, we compared SNPs between LG/J and SM/J strains, using UCSC genome browser sequence annotations (NCBI37/mm9 assembly). For each gene, we considered SNPs occurring within 2500 bp upstream and downstream of transcription start and stop sites, as well as each exon, intron, and untranslated region. Evolutionary conservation scores were obtained from the PhastCons 30Placental table. Nonsynonymous amino acid changes were analyzed for probability of functional significance using Polyphen-2[20]. Scores less than 0.9 were considered benign as opposed to possibly damaging.

Gene expression differences for genes located in quantitative trait loci (QTLs)

We have previously mapped QTLs for pinna-based wound healing and long bone lengths in the F₃₄ generation of the LG,SM:G34 Advanced Intercross line[10, 21]. We examined expression in the genes within one-LOD drop QTL-support intervals of the most probable locations for the QTL. For each of these QTL sets we determined the number and intensity of positional candidate genes with significant differences in expression as reported here.

The ear pinna involves several tissues, but its volume is dominated by elastic cartilage while the tissue examined in this paper is hyaline articular cartilage. Even so, there is a strong genetic correlation between ear wound healing and articular cartilage healing in the LG/J by SM/J intercross[5]. We identified 19 genomic regions with significant effects on ear wound healing. QTL support intervals had a median size of 2 Mb containing, on average, 19 genes each. There were 510 positional candidate genes within the 19 QTL support intervals. Function enrichment analysis of positional candidate genes in QTLs for regenerating pinna tissue, discovered strain differences in cell cycle/DNA damage repair, cytoskeleton and motility, and developmentally related genes[10].

We also compared genes in long bone length (humerus, ulna, femur, tibia lengths) QTLs from the LG,SM-G34 Advanced Intercross for expression differences in cartilage[21, 22]. Hyaline growth plate cartilage is the major tissue responsible for long bone length growth. We discovered 30 pleiotropic QTLs affecting adult long bone lengths. There were 807 positional candidate genes in the QTLs, about 27 genes per QTL. The median QTL support interval length was 2.38 Mb.

Statistical analysis

Real-time PCR and immunofluorescence data were analyzed using non-parametric Mann-Whitney U test (GraphPad Prims v7.03). Results were considered significant at P<0.05.

RESULTS

Chondrocytes from LG/J and SM/J mouse strains exhibit similar characteristics

We observed that chondrocytes from both mouse strains appeared similar and exhibited characteristic chondrocyte phenotype (Fig. 1A). Assessment of cartilage specific genes (Col2a1, Acan, Sox9) revealed that chondrocytes from both mouse strains were not significantly different in the expression profile of the aforementioned genes (Fig. 1B). Moreover, these genes were not significantly different between the strains as assessed by RNA-seq.

Figure 1: Chondrocyte characterization.

A) Microscopic images of chondrocytes showing typical chondrocyte morphology with no obvious differences in chondrocyte phenotypes between SM/J and LG/J mouse strains. B) Real-time PCR analysis revealed no significant differences in the expression of cartilage-specific genes (Col2a1, Acan, Sox9) in chondrocytes derived from LG/J and SM/J mouse strains.

Descriptive RNA-seq data reveals distinct clustering of sample from LG/J and SM/J mice

Samples were clustered into two distinct clusters based on principal component analysis: one cluster exclusively had samples from LG/J mice and the other cluster had samples from SM/J mice (Fig. 2A). Samples were clustered by mouse strain based on gene expression signatures on hierarchical clustering heatmaps (Fig. 2B). Expression fold change and averaged expression level of differentially expressed transcripts are shown as MA plot (Fig. 2C). Moreover, volcano plots (Fig. 2D) displayed differentially expressed transcripts to indicate the trend of transcription expression in magnitude (fold change) and significance (P value). MA and Volcano plots serve to exemplify the relationship of the statistical uncertainty and the observed changes in expression between the two mouse strains.

Figure 2: Descriptive RNA-seq data.

A) Principal component analysis of 13 LG/J and 16 SM/J samples demonstrated a clear distinction between LG/J and SM/J chondrocyte gene expression profiles. Each dot represents one sample (pooled from 3–5 mice). B) Normalized gene expression level of differentially expressed transcripts between LG/J and SM/J mice were used to generate heatmaps showing that LG/J and SM/J were distinctly separated. C) Expression fold change and averaged expression level of differentially expressed transcripts between LG/J and SM/J. D) Volcano plot of all genes expressed greater than 1 count-per-million in at least 12 samples where the observed log₂ fold change is on the x-axis and the unadjusted P value converted to the −log₁₀ scale is on the y-axis. All genes with Benjamini-Hochberg adjusted P ≤ 0.05 are highlighted in red.

Chondrocytes derived from LG/J and SM/J mice exhibited distinct genetic heterogeneity

Our data revealed that 5,161 genes were significantly differentially expressed between LG/J and SM/J with Benjamini-Hochberg FDR adjusted P≤0.05 (Fig. 2C, Table 2, Supplementary Table 1). Of these, 386 genes (363 protein-coding) were up-regulated and 378 (359 protein-coding) were down-regulated in LG/J compared to SM/J at ≥2 fold with multiple gene biotypes represented as differentially expressed (Supplementary Text). Most of these transcripts were protein-coding genes (363/386 up-regulated in LG/J and 359/378 down-regulated in LG/J). Remaining were long intergenic non-coding RNAs (lincRNAs).

Table 2:

Gene transcripts differentially expressed between LG/J and SM/J^*

Gene ID	Gene name	Fold change	Adjusted P value
Transcripts up-regulated in LG/J compared to SM/J
Tnfrsf23	tumor necrosis factor receptor superfamily, member 23	116.10	<0.001
Mybphl	myosin binding protein H-like	87.41	<0.001
Lefty1	left right determination factor 1	51.40	<0.001
Rps26	ribosomal protein S26	49.36	<0.001
Ndufs5	NADH dehydrogenase (ubiquinone) Fe-S protein 5	39.02	<0.001
Gm10320	predicted pseudogene 10320	31.18	<0.001
H2-Ab1	histocompatibility 2, class II antigen A, beta 1	25.84	<0.001
Tnfrsf26	tumor necrosis factor receptor superfamily, member 26	19.26	<0.001
Tagap	T cell activation Rho GTPase activating protein	17.97	<0.001
Tmie	transmembrane inner ear	17.06	<0.001
Gm8909	predicted gene 8909	15.65	<0.001
Fam151b	family with sequence similarity 151, member B	14.69	<0.001
C1rb	complement component 1, r subcomponent B	13.83	<0.001
Rab5b	RAB5B, member RAS oncogene family	13.29	<0.001
H2-M2	histocompatibility 2, M region locus	12.13	<0.001
Fxyd3	FXYD domain-containing ion transport regulator 3	12.08	<0.001
Cramp1l	cramped chromatin regulator homolog 1	11.22	<0.001
C1s2	complement component 1, s subcomponent 2	10.93	<0.001
Hist1h4d	histone cluster 1, H4d	10.37	<0.001
Lrp4	low density lipoprotein receptor-related protein 4	9.88	<0.001
Transcripts up-regulated in SM/J compared to LG/J
Lyz1	lysozyme	−132.98	<0.001
Rpl14	ribosomal protein L14	−49.39	<0.001
Zfp979	zinc finger protein 979	−47.12	<0.001
Zfp979	zinc finger protein 979	−47.12	<0.001
Tpm3-rs7	tropomyosin 3, related sequence	−45.93	<0.001
Tescl	tescalcin-like	−34.02	<0.001
Car2	carbonic anhydrase	−22.90	<0.001
H2-T24	histocompatibility 2, T region locus 2	−18.90	<0.001
Ifit3	interferon-induced protein with tetratricopeptide repeats 3	−17.25	<0.001
Rpl26	ribosomal protein L26	−16.47	<0.001
Gm49339	predicted gene, 49339	−16.07	<0.001
Zfp429	zinc finger protein 429	−15.08	<0.001
Traf3ip1	TRAF3 interacting protein 1	−14.94	<0.001
Mmp12	matrix metallopeptidase 1	−14.43	<0.001
Rpl35a	ribosomal protein L35A	−12.68	<0.001
Gsta4	glutathione S-transferase, alpha 4	−12.09	<0.001
Tmem87a	transmembrane protein 87A	−11.83	<0.001
Gdf9	growth differentiation factor	−11.09	<0.001
Eif3j2	eukaryotic translation initiation factor 3, subunit J2	−10.87	<0.001
Pdlim1	PDZ and LIM domain 1 (elfin)	−10.55	<0.001

^*top 20 differentially expressed gene transcripts are shown for each comparison

GO analysis revealed a number of distinct GO biological processes between LG/J and SM/J (Fig. 3). Interestingly, we found that the top GO biological processes elevated in LG/J were related to chondrocyte development and cartilage condensation, which also included regulation of cartilage development (GO:00161035), cartilage development (GO:0051216), regulation of chondrocyte differentiation (GO:0032330), chondrocyte differentiation (GO:0002062) and cartilage condensation (GO:0001502). In contrast, GO biological processes elevated in the SM/J mice were mostly inflammation related, specifically, tumor necrosis factor (TNF) related pathways, such as TNF superfamily cytokine production (GO:0071706), TNF production (GO:0032640) and regulation of TNF production (GO:0032680).

Figure 3: Go terms enriched for differentially expressed genes in LG/J and SM/J.

Bar plot of statistically significant –log₁₀(P value) as identified by the R/Bioconductor package GAGE. The top 25 GO terms for processes elevated in SM/J (A) and LG/J (B) with color intensity depending on mean log fold change (FC) are shown.

KEGG pathway analysis identified no significant KEGG terms based on genes significantly up-regulated in LG/J. In contrast, a number of KEGG terms were significant for genes elevated in SM/J. The most prominent pathways among these included TNF signaling pathway, osteoclast differentiation, apoptosis, cell cycle and cytokine-cytokine interaction (Supplementary Text).

Real-time PCR data confirmed RNA-seq results

The gene expression patterns (i.e. direction and magnitude of expression) of all but three (Rps26, Rpl14, and Lefty1) transcripts out of 20 tested by real-time PCR were very concordant with the RNA-seq data (Table 3). The expression of two genes, namely Hyou1 and Wisp2, was concordant between the RNA-seq and real-time PCR results, however, they did not achieve formal statistical significance.

Table 3:

Comparison between RNA-seq and real-time PCR data

Gene symbol	Gene name	RNA-seq		Real-time PCR
		Fold change	Adjusted P value	Fold change	P value
Tnfrsf23	tumor necrosis factor receptor superfamily, member 23	116.10	<0.0001	42.47	<0.0001
Mybphl	myosin binding protein H-like	87.41	<0.0001	374.90	<0.0001
Lefty1	left right determination factor 1	51.40	<0.0001	1.16	0.97
Rps26	ribosomal protein S26	49.36	<0.0001	−1.06	0.97
Rflna	refilin A	7.42	<0.0001	12.47	<0.0001
Efemp1	epidermal growth factor-containing fibulin-like extracellular matrix protein 1	3.98	<0.0001	4.68	<0.0001
Bmf	BCL2 modifying factor	3.84	<0.0001	7.49	<0.0001
Bmp6	bone morphogenetic protein 6	3.28	<0.0001	4.79	<0.0001
Col10a1	collagen, type X, alpha 1	2.83	0.002	4.47	<0.0001
Hyou1	hypoxia up-regulated 1	2.76	<0.0001	1.43	0.17
Tgfa	transforming growth factor alpha	2.49	<0.0001	2.34	<0.0001
Wisp2	WNT1 inducible signaling pathway protein 2	−2.27	<0.0001	−1.77	0.12
Cd300ld	CD300 molecule like family member d	−3.15	0.014	−21.51	0.01
Fcgr3	Fc receptor, IgG, low affinity III	−4.18	0.016	−133.04	<0.0001
Ptafr	platelet-activating factor receptor	−4.78	<0.0001	−7.31	0.01
Mmp12	matrix metallopeptidase 12	−14.43	<0.0001	−30.29	<0.0001
Car2	carbonic anhydrase 2	−22.90	<0.0001	−28.46	<0.0001
Tescl	tescalcin-like	−34.02	<0.0001	−54.80	<0.0001
Rpl14	ribosomal protein L14	−49.39	<0.0001	1.53	0.12
Lyz1	lysozyme 1	−132.98	<0.0001	−13.61	<0.0001

Immunostaining of Tnfrsf23 and Car2 showed distinct intensities in LG/J and SM/J cartilage

We found that two candidates namely Tnfrsf23 and Car2 deserve further validation as the former was highly expressed in LG/J chondrocytes and has been implicated in chondrocyte biology, the latter was significantly highly expressed in SM/J and has been shown to play an important role in chondrocyte metabolism and homeostasis. Immunostaining analysis performed on these candidates confirmed the results obtained by RNA-seq and real-time PCR. Briefly, as expected Tnfrsf23 expression was higher in LG/J mice while that of Car2 was lower compared to SM/J mice although it did not reach formal statistical significance (Fig. 4).

Figure 4: Immunofluorescent analysis of the hip cartilages from LG/J and SM/J mice to confirm the RNA-seq result.

The cartilages from femoral head were sectioned and stained with Tnfrsf23 (A) or Car2 (C) antibodies (green). Immunofluorescent staining for collagen type II (Col2, red) revealed the morphology of each cartilage sample. Dapi (blue) was used to show cell nuclei. Tnfrsf23 was much more in the LG/J mice, especially in the deeper zone of the cartilage (A). Correspondingly, the intensity of Car2 staining is much higher in the SM/J mice in both superficial and deeper zone of the cartilage (C). Scale bars = 100 μm. Quantification of Tnfrsf23 (B) and Car2 (D) staining intensities as measured by ZEN V2.3 software showed that Tnfrsf23 had higher signals in LG/J mice while Car2 had higher signals in SM/J mice, although these differences did not reach a formal statistical significance.

Multiple validated genes vary in sequence between LG/J and SM/J in coding and highly conserved noncoding regions

In addition to differences in level of expression, several of the PCR-validated genes demonstrate allelic variations between LG/J and SM/J strains that potentially mediate between strain differences in gene regulation and function underlying cartilage homeostasis and susceptibility to OA (Table 4). Between strains, Tnfrsf23 differs in 19 bases, eight occurring in the upstream 2500 bases and 11 in the introns. With respect to Car2, LG/J and SM/J differ in 76 bases. One of two coding SNPs is nonsynonymous, encoding glutamine in LG/J but histidine in SM/J at the thirty-ninth amino acid (Q:39:H). Of 66 intronic SNPs, three occur at highly conserved bases. Five and three are located in the upstream and downstream regions of the gene, respectively. In the Lefty1 locus, LG/J encodes a threonine but SM/J an alanine at the fifty-ninth amino acid (T:59:A), an evolutionarily conserved position. An additional fifty-five SNPs occur in non-coding regions. The two strains differ at 67 base positions in Wisp2, one a highly conserved nonsynonymous polymorphism (Q:165:R). TGF-α varies at 240 bases, including three highly conserved intronic positions. Lyz1 differs at 45 non-conserved bases, most of which occur in the 3’UTR and downstream region. Alleles for Mybphl differ at nine nonconserved intronic SNPs. LG/J and SM/J possess identical sequences for Mmp12, suggesting its contribution to phenotypic variation is driven by variations in genes that regulate its expression level and timing. Using the Polyphen-2 prediction tool, all nonsynonymous amino acid changes were considered benign and not likely to radically alter protein function.

Table 4:

SNPs in candidate genes differentially expressed between LG/J and SM/J

Gene	RefGeneID	Total SNPs		Coding SNPs						Noncoding SNPs												Nsyn coding changes	Polyphen2 score
		SNPs	HC	Total coding	HC	nsyn	HC	syn	HC	Total non-coding	HC	2500 bp up stream	HC	5’ UTR	HC	Introns	HC	3’ UTR	HC	2500 bp down stream	HC
Bmf	NM_138313	124	6	2	1	0	0	2	1	122	5	11	0	0	0	82	5	14	0	15	0
Bmp6	NM_007556	12	0	0	0	0	0	0	0	12	0	0	0	0	0	12	0	0	0	0	0
Car2	NM_009801	76	3	2	0	1	0	1	0	74	3	5	0	0	0	66	3	0	0	3	0	Q:39:H	0.008
Cd300ld	NM_145437	29	0	0	0	0	0	0	0	29	0	0	0	0	0	2	0	10	0	17	0
Col10a1	NM_009925	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
Efemp1	NM_146015	45	1	0	0	0	0	0	0	45	1	3	0	0	0	42	1	0	0	0	0
Fcgr3	NM_010188	31	0	0	0	0	0	0	0	31	0	8	0	0	0	8	0	2	0	13	0
Hyou1	NM_021395	23	1	1	1	0	0	1	1	22	0	1	0	0	0	18	0	0	0	3	0
Lefty1	NM_010094	56	1	1	1	1	1	0	0	55	0	8	0	2	0	13	0	9	0	23	0	T:59:A	0.001
Lyz1	NMv013590	45	0	0	0	0	0	0	0	45	0	13	0	0	0	0	0	1	0	31	0
Mmp12	NM_008605	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
Ptafr	NM_001081211	199	6	7	6	2	2	5	4	192	0	10	0	0	0	133	0	10	0	39	0	D:82:N, N:87:S	0.013, 0.001
Rpl14	NM_025974	50	2	2	2	0	0	2	2	48	0	8	0	0	0	22	0	4	0	14	0
Rps26	NM_013765	41	1	1	1	0	0	1	1	40	0	14	0	0	0	11	0	1	0	14	0
Tgfa	NM_031199	240	3	1	0	0	0	1	0	239	3	1	0	0	0	232	3	4	0	2	0
Tnfrsf23	NM_024290	19	0	0	0	0	0	0	0	19	0	8	0	0	0	11	0	0	0	0	0
Wisp2	NM_016873	67	2	3	2	1	0	2	2	64	0	4	0	1	0	44	0	14	0	1	0	Q:165:R	0.006
Rflna	NM_028443	19	0	0	0	0	0	0	0	19	0	4	0	0	0	13	0	0	0	2	0
Tescl	NM_001163810	36	1	1	0	0	0	1	0	35	1	27	0	0	0	1	1	0	0	7	0
Mybphl	NM_026831	9	0	0	0	0	0	0	0	9	0	0	0	0	0	9	0	0	0	0	0

SNP = single nucleotide polymorphism; HC = highly conserved; nsyn = nonsynonymous; syn = synonymous; bp = base pair UTR = untranslated region.

Most QTLs have 20–30% of their positional candidates displaying differential expression between LG/J and SM/J mouse strains

We also examined gene expression of positional candidate genes for ear pinna regeneration and long bone length QTLs that display differences in cartilaginous expression. Genes within the QTL confidence regions for ear pinna healing showed differential expression between the parent strains 23.5% of the time (120/510), compared to 30.4% in the genome as a whole indicating that positional candidate genes show expression differences less often than the genome as a whole (X²=11.07, 1 df, Prob=0.001). Restricting ourselves to genes having a 2-fold or greater expression difference between strains, 4.7% (24/510) of positional candidate genes are differentially expressed, compared to 4.5% in the genome as a whole (X²=0.053, 1 df, Prob=0.82). The percentage of genes with differential expression in individual QTLs ranges from 0.0% (Q11.2) with none of the positional candidate genes differentially expressed (0/28) to 44.4% (Q7.3) with 12 of 27 genes being differentially expressed.

Individual positional candidate genes with known effects on cartilaginous structures include Ptprc (protein tyrosine phosphatase, receptor type, C) with effects on osteoblast, osteoclast, and long bone metaphyseal morphology in KO mice[23]; Med9 (mediator complex subunit 9) with effects on craniofacial morphology[24]; and Rnf7 (Ring finger protein 7) with effects on cell cycle progression[25]. Rnf7 expression is 2.6-fold lower in LG/J than in SM/J in cartilage and in regenerating ear-pinna tissue. Rnf7 controls the degradation of Hif1a so that Hif1a shows a 1.5-fold increase in LG/J cartilage compared to SM/J. Absence of Hif1a results in abnormal cartilage and growth plate development[26, 27].

Twenty-five percent of the 807 positional candidate genes in long bone length QTLs are differentially expressed in LG/J and SM/J cartilage. This is low given that 30.4% of genes are differentially expressed across the genome (X²=7.59, 1 df, Prob=0.006). Restricting ourselves to genes having a 2-fold or greater expression difference between strains, 3.3% (27/807) of positional candidate genes are differentially expressed, compared to 4.5% in the genome as a whole (X²=1.994, 1 df, Prob=0.16). Three of the QTLs had no differentially expressed genes (LB1.1, LB17.1, LB17.2) but also had very few genes within the QTL support intervals. This is especially true for LB1.1 and LB17.2 which only contained four and one gene, respectively. The maximum percentage of differentially expressed genes within QTL support intervals is 75% for (LB4.1), with three out of four genes showing differential expression.

Individual positional candidate genes with greater than 2-fold expression differences and known effects on cartilaginous structures include: Ccn3 (cellular communication network factor 3) that shows abnormal cartilage and bone morphology when knocked out[28]; Rcan2 (regulator of calcineurin 2) shows postnatal growth retardation and a short adult tibia[29]; Rnf7 has been described above[25]. Fermt3 (fermitin family member 3) shows osteopetrosis[23] and Aass (aminoadipate-semialdehyde synthase) has effects on sternal morphology and chondrocyte differentiation[30].

DISCUSSION

RNA-seq analysis revealed numerous distinct genes and pathways between LG/J and SM/J mice. These data, for the first time, provide important insights into the process of cartilage regeneration and the pathogenesis of OA. Since the most significant differences between the healer and non-healer mice were related to chondrocyte biology and inflammation, our findings corroborate the phenotypic differences reported between the two mouse strains[5, 31]. These findings are likely the most convincing molecular targets for understanding regeneration and degeneration of articular cartilage.

We identified potential new target genes. Tumor necrosis factor receptor superfamily member 23 (Tnfrsf23) showed the highest fold change in the LG/J vs. SM/J mice (116-fold). Tnfrsf23 encodes a member of the TNF superfamily of proteins and the encoded receptor binds to the ligand TRAIL (tumor necrosis factor-related apoptosis-inducing ligand). TRAIL has been shown to be expressed in normal human tissues and is known to induce rapid apoptosis[32]. However, the Tnfrsf23 encoded receptor neither induces apoptosis nor triggers NF-κB activation[33]. Instead, it serves as a murine decoy receptor for TRAIL. Higher mRNA expression of Tnfrsf23 in LG/J chondrocytes may be associated with higher resistance of chondrocytes to injury. It is difficult to relate whether similar pathways exist in human disease as a prior study has reported that Tnfrsf genes are degenerated or deleted or relocated in primates, and in humans only a single short homologous sequence is observed[34].

Another promising candidate identified was Car2 (carbonic anhydrase 2), which was 23-fold higher in SM/J than LG/J. It encodes an enzyme that catalyzes the interconversion between CO₂ and bicarbonate. Although it is vital to osteoclast function[35], macrophages also highly express Car2, which plays a significant role in demineralization of ectopic calcification[36]. We have previously shown that SM/J mice are protected from injury-induced ectopic calcification while LG/J mice are not[37]. Therefore, a higher expression of Car2 in SM/J mice might help explain why they are resistant to ectopic calcification. Besides the decalcification activity, Car2 has also been shown to be expressed in articular cartilage where it appears to be necessary for acid base buffering due to the production of lactate[38]. It is plausible and even likely that there is an increased accumulation of cytosolic lactate in SM/J chondrocytes as we expect in OA[39, 40]. This notion is favored by the fact that Car2 is highly expressed in osteoarthritic cartilage[41] further suggesting that SM/J mice are more susceptible to OA.

Other genes upregulated in LG/J included Mybphl and Lefty1. Mybphl, l is a cardiac myofilament protein enriched in the atria and is related to human arrhythmias and cardiomyopathy[42]. Its function in the musculoskeletal system remains to be determined. Lefty1 encodes a secreted ligand of the TGF-beta superfamily of proteins[43]. A gene expression comparison between the congenic BALB.D1–1 mice and the parental BALB/c^−/− strain revealed that Lefty1 was down-regulated in the congenic strain in joint tissue, which is relatively resistant to spontaneous arthritis[44].

The most up-regulated genes in SM/J mice were related to the immune system (e.g. Lyz1) and inflammation (e.g. Ifit3). The gene with the highest fold-difference, Lyz1, encodes the protein lysozyme-1, which functions in innate immune response[45]. No study to-date specifically explored its role in cartilage. Ifit3 is implicated in the inflammatory cytokine interferon α signaling pathway in the human degenerated annulus fibrosus, but its function in inflammatory responses is complex and far from fully understood.

Investigations into some genes, with a known role in chondrocyte biology and OA, yielded interesting information. Mmp12 and Wisp2 were highly expressed in the SM/J mice. MMP-12 has been considered as a matrix degrading enzyme and has been shown to enhance the development of inflammatory arthritis in rabbits[46]. Wisp2 encodes proteins of the WNT1 inducible signaling pathway and has been found to be closely related to the pathogenesis of inflammatory arthritis and the modulation of bone turnover[47]. Hyou1 and Tgfa were upregulated in the LG/J strain. Hyou1 exhibits anti-apoptotic function[48] while higher levels of TGF-α are consistent with its involvement in chondrogenesis[49]. Some studies suggest that TGF-α is a catabolic factor in rodent OA[50].

Gene ontology categories of genes differentially expressed in LG/J (healer) and SM/J (non-healer) strains were related to multiple cell functions. Condensation or aggregation of chondroprogenitor mesenchymal cells by cell-cell and cell-matrix interactions is associated with an increase in cell adhesion[51]. Cell adhesion is an important process that facilitates adherence of cells with other cells or matrix. Cell adhesion to matrix regulates cell shape, proliferation, intracellular signaling and differentiation, and is therefore a key process in maintaining normal cartilaginous tissue function[52] and a prerequisite step in the chondrogenic pathway[53]. The process of chondrogenesis also occurs as a result of mesenchymal cell condensation. Cellular condensation is the initial marker of differentiation which occurs during chondrogenesis and the formation of skeletal elements[54]. Mesenchymal cell condensation is the initiating event in endochondral bone formation. Cell condensation is followed by differentiation into chondrocytes, which is accompanied by induction of chondrogenic gene expression[55]. In addition to cellular condensation, processes related to cell differentiation were also enriched in chondrocytes from LG/J mice. Chondrocyte proliferation is required to maintain chondrocyte phenotype[56]. Taken together, significant enrichment of cellular condensation and proliferation processes in LG/J mice indicate increased potential of their chondrocytes for adhesion and chondrogenesis, as well as maintenance of their characteristic phenotypes, than chondrocytes derived from SM/J mice. These findings are also in line with other literature on LG/J and SM/J strain comparisons; for example, LG/J has faster growth plate dynamics, which are responsible for their longer limbs than SM/J[22].

The pathways that were enriched in chondrocytes derived from SM/J mice included cytokine production and regulation, immune cell activation and inflammation. Cytokines, particularly pro-inflammatory, are biochemical biomarkers of OA[57]. The latest theories of OA pathogenesis implicate the interplay between cytokine production and inflammation[58]. The role of inflammation[59] is well established in OA as inflammation is considered a major factor associated with the risk of both progressions of articular cartilage lesions and symptoms of the OA disease process. Under normal conditions, there is a balance between anabolic and catabolic processes in the joint. However, a number of factors such as age, obesity and trauma may trigger a response in the joint leading to disruption of this homeostasis. This imbalance between the production of anabolic and catabolic cytokines may account for the pattern of immune response. Enrichment of processes related to inflammation and cytokine production is SM/J together with immune response processes as well as a number of related KEGG pathways suggest that the chondrocytes derived from SM/J mice exhibit an OA prone phenotype. These findings are in line with the previous reports that SM/J is very susceptible to obesity on a high fat diet. Inflammation is an important process in obesity and SM/J respond more strongly to the high fat diet than LG/J[60].

The gene expression study reported here also has consequences for identifying cartilage-related positional candidate genes in QTLs for cartilage-related traits, such as long bone lengths and ear pinna regeneration. Most QTLs have 20%–30% of their positional candidate genes showing differential expression between LG/J and SM/J chondrocyte cultures. This will aid in identifying the quantitative trait genes and molecular differences responsible for the phenotypic differences associated with individual QTLs.

An important study limitation to recognize is the young age of the mice at which chondrocytes were collected, 3 weeks postpartum, as this age is not reflective of skeletal maturity. Isolation of pure chondrocyte population from skeletally mature mice is not technically feasible, as it increases the likelihood of contamination with bone cells. Therefore, we propose that the immature articular chondrocytes may be potent tools for mechanistic and physiological studies of cartilage aimed at identifying the genetic differences in our mouse strains, since these were able to form confluent monolayers within a few days and exhibited characteristics of fully differentiated chondrocytes. Moreover, validation of two candidates in skeletally mature joints by immunofluorescence confirmed that gene expression differences at early age are maintained into adulthood. Additional studies on these chondrocytes in terms of cytokine treatment and gene transfer will further establish the authenticity of these cells. Another limitation is that the chondrocytes were enzymatically isolated and were cultured in plastic dishes for a few days prior to RNA preparation. While the ex-vivo culture and plastic adherence may have induced changes in gene expression, isolation and culture steps were necessary to obtain a high number of adherent cells, and to remove any non-adherent dead cells. Short term cultures have been extensively used to characterize and compare gene expressions in articular chondrocytes from mice with genetic differences. It is well established that in short term culture murine chondrocytes maintain their phenotype. Also, we expect that ex-vivo culture conditions would have exerted similar effects, if any, on chondrocytes from both strains. Lastly, despite obvious differences, the quantification of immunofluorescence signaling intensity for Tnfrsf23 and Car2 did not achieve a formal level of significance that could be attributed to small sample size.

In summary, our data provide novel insights into chondrocyte transcriptomics from two distinct mouse strains. While the enrichment of genes and pathways related to inflammation in mice susceptible to OA provides clues for higher cartilage degeneration in these mice, enrichment of genes and pathways related to cartilage matrix suggests a role in cartilage regeneration in mice with superior healing potential. Furthermore, the differential expression of multiple functional pathways represents the genetic heterogeneity that leads to the dissimilar healing/injury resistance abilities of cartilage of these two genetic mouse strains.