Available via license: CC BY 4.0
Content may be subject to copyright.
UNCORRECTED MANUSCRIPT
1
Genetic risk of osteoarthritis operates during human
1
skeletogenesis
2
3
Sarah J. Rice1,*, Abby Brumwell1, Julia Falk1, Yulia S. Kehayova1, John
4
Casement2, Eleanor Parker1, Ines M.J. Hofer1, Colin Shepherd1 and John
5
Loughlin1
6
7
1Biosciences Institute, International Centre for Life, Newcastle University, Newcastle upon
8
Tyne, NE1 3BZ, UK
9
2Bioinformatics Support Unit, Faculty of Medical Sciences, Newcastle University,
10
Framlington Place, Newcastle upon Tyne NE2 4HH, UK
11
12
13
14
15
16
17
18
19
20
21
22
© The Author(s) 2022. Published by Oxford University Press.
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
2
23
24
25
26
27
28
29
30
31
32
33
34
35
*To whom correspondence should be addressed at: Biosciences Institute, International Centre
36
for Life, Newcastle University, Newcastle upon Tyne, NE1 3BZ, UK. Tel: +441912418850;
37
Fax: +441912418666; Email: sarah.rice@newcastle.ac.uk
38
Abstract
39
Osteoarthritis (OA) is a polygenic disease of older people resulting in the breakdown of
40
cartilage within articular joints. Although a leading cause of disability, there are no disease-
41
modifying therapies. Evidence is emerging to support the origins of OA in skeletogenesis.
42
Whilst methylation QTLs (mQTLs) co-localizing with OA GWAS signals have been identified
43
in aged human cartilage and used to identify effector genes and variants, such analyses have
44
never been conducted during human development.
45
Here, for the first time, we have investigated the developmental origins of OA genetic
46
risk at seven well-characterized OA risk loci, comprising 39 OA-mQTL CpGs, in human foetal
47
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
3
limb (FL) and cartilage (FC) tissues using a range of molecular genetic techniques. We
48
compared our results to aged cartilage samples (AC) and identified significant OA-mQTLs at
49
14 CpGs and 29 CpGs in FL and FC tissues, respectively. Differential methylation was
50
observed at 26 sites between foetal and aged cartilage, with the majority becoming actively
51
hypermethylated in old age. Notably, 6/9 OA effector genes showed allelic expression
52
imbalances during foetal development. Finally, we conducted ATAC-sequencing in cartilage
53
from the developing and aged hip and knee to identify accessible chromatin regions, and found
54
enrichment for transcription factor binding motifs including SOX9 and FOS/JUN.
55
For the first time, we have demonstrated the activity of OA-mQTLs and expression
56
imbalance of OA effector genes during skeletogenesis. We show striking differences in the
57
spatiotemporal function of these loci, contributing to our understanding of OA aetiology, with
58
implications for the timing and strategy of pharmacological interventions.
59
Graphical abstract
60
61
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
4
Introduction
62
Osteoarthritis (OA) is a complex disease of the articulating joints, typically impacting those
63
over the age of 45, with the risk of developing progressive disease increasing with age (1). OA
64
impacts the lives of over 500 million individuals worldwide, a figure which increased by 48%
65
between 1990 and 2019, and which continues to rise in ageing populations (2). Despite this,
66
there are currently no disease modifying OA drugs (DMOADs) available.
67
In the healthy joint, the articular cartilage lubricates joint surfaces and absorbs
68
mechanical impact. In OA, this cartilaginous surface roughens and breaks down, eventually
69
exposing the underlying bone. This leads to chronic pain, disability, and premature death from
70
secondary co-morbidities (3,4). The aetiology of OA is complex and multifactorial, yet genetic
71
variation accounts for up to 50% of total disease risk, placing it as one of the highest single risk
72
factors for disease (1,5).
73
The developmental origins of OA have long been debated due to a plethora of factors
74
including the association of disease with abnormal joint morphology, the extent of disease
75
heritability (6-8), and the reversion to a developmental phenotype within hypertrophic OA
76
articular chondrocytes (9,10). The articular chondrocyte, the sole cell type in cartilage,
77
undergoes very little proliferation once adulthood is reached, with the non-dividing cells
78
remaining metabolically active. This raises the question of whether defects in cartilage integrity
79
conferred by genetic variation are established during embryonic and foetal development, which
80
manifest in later life (10). Such evidence would have enormous implications upon the
81
successful development of DMOADs, and the determination of a “window of opportunity” for
82
pharmacological intervention in early-stage disease (11). As a result, OA research has heavily
83
utilized small animal developmental models, along with stem cell models of chondrogenesis
84
(12,13), and, more recently, human foetal samples (8).
85
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
5
Over 100 common genetic variants, namely single nucleotide variants (SNVs), have
86
been reported, which reproducibly associate with OA genetic risk (14). Identified through
87
genome wide association studies (GWAS), the reported SNVs are overwhelmingly enriched
88
within non-coding regions of the genome, and often fall in regions of high linkage
89
disequilibrium (LD), whereby variant alleles are co-inherited, making it difficult to identify
90
causal variants and effector genes at the loci.
91
Across the field of complex disease research, the identification of methylation and
92
expression quantitative trait loci (mQTLs and eQTLs, respectively) is increasingly being
93
applied to prioritize candidate variants and genes at disease risk loci (15-17). Furthermore, in
94
studies of neuropsychiatric disorders, disease associated mQTLs have been identified in
95
developing foetal brain tissue, predisposing individuals to disease from the start of life (18). In
96
OA, the identification of mQTLs and eQTLs (the latter often discovered by allelic expression
97
imbalance analysis) has been applied to prioritize causal SNVs, effector genes, and regulatory
98
elements in aged joint tissues (14,19-23). Successful follow-up studies using targeted
99
epigenome editing have demonstrated a causal effect, and functional roles for DNA
100
methylation (DNAm) in altering expression of effector genes (22,23). However, such QTL
101
analyses have not previously been applied to relevant human foetal tissues to investigate
102
developmental origins of musculoskeletal disease.
103
In this study, we investigated seven OA genetic risk loci distributed across the human
104
genome, at which mQTLs at 39 CpGs have previously been identified and replicated in aged
105
human cartilage (19-22,24-26). We hypothesized that the epigenetic mechanisms of gene
106
regulation found in aged human cartilage may be active during foetal development, impacting
107
cartilage integrity from the beginning of life. We quantified DNAm at these CpGs in pooled
108
limb tissues from individual human foetuses and foetal joint cartilage. Further, we conducted
109
a series of molecular genetics analyses to investigate allele specific gene expression in
110
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
6
developmental samples. Finally, we conducted assay for transposase accessible chromatin
111
sequencing (ATAC-seq) in developmental and aged hip and knee cartilage to compare
112
chromatin accessibility between foetal limb development and in OA.
113
114
Results
115
DNAm profiling of foetal limb tissues at OA risk loci
116
We extracted DNA from two human developmental tissue types: whole limb tissues, which
117
had a mean developmental age of 56 days (E56), and isolated limb cartilage samples, which
118
were significantly older (E86; P=5.4x10-10) (Supplementary Material, Fig. S1A and B). For
119
cartilage analysis, we used the cartilaginous (non-ossified) tissue from the ends of the
120
developing long bones (Supplementary Material, Fig. S1C and D), the size of which decreased
121
as the developing bone ossified, and which showed significantly higher expression of the
122
cartilage marker genes COL2A1, ACAN, and MATN3 than the pooled limb samples
123
(Supplementary Material, Fig. S1E). There was a significant increase in the expression of
124
COL2A1 and MATN3 (both P=0.01) with developmental age (Supplementary Material, Fig.
125
S1F).
126
DNAm was quantified in both developmental tissue types at 39 OA-CpGs: sites of, or
127
adjacent to, OA-mQTLs that have been characterized in aged cartilage. The CpGs span 7
128
genomic loci, which harbour 9 OA risk genes: COLGALT2 (Locus 1), GNL3 and SPCS1 (Locus
129
2), SUPT3H and RUNX2 (Locus 3), PLEC (Locus 4), ALDH1A2 (Locus 5), GDF5 (Locus 6),
130
and RWDD2B (Locus 7) (Table 1, Fig. 1A). We first compared mean DNAm at each of the
131
CpGs in the two tissue types, foetal limb (FL, n=15-19) and foetal cartilage (FC, n=54-75), to
132
DNAm levels which we had previously measured and reported in aged articular cartilage
133
(19,21,22,24-26) (AC, n=71-139) (Fig. 1B, Supplementary Material, Table S1). DNAm was
134
significantly different between FL and FC tissues at 17/39 CpGs (adjusted P value, Padj<0.014),
135
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
7
highlighting the tissue-specificity of DNAm patterns during musculoskeletal development
136
(Fig. 1B, Supplementary Material, Table S2). This was particularly striking at Locus 4,
137
harbouring PLEC, where DNA hypermethylation (mean >80%) was observed in 10 of the 12
138
investigated CpGs in FL samples (Fig. 1B); yet in FC samples, mean DNAm within the tissue
139
was up to 28.8% lower (CpG 7, Supplementary Material, Table S2). When comparing the FC
140
and AC samples, differentially methylated sites (DMS) were identified at 26/39 CpGs
141
(Supplementary Material, Table S2), the majority (69%) demonstrating increased methylation
142
in the aged samples. The largest differences in mean DNAm were again observed at Locus 4
143
(-44.36% [-47.81, -40.90]) and additionally at Locus 1, harbouring COLGALT2 (24.88%
144
[21.44, 28.32]).
145
These data show the tissue-specificity of DNAm in distinct limb tissues during skeletal
146
development. Further, our data show a trend towards active methylation of disease associated
147
CpGs in chondrocytes during the ageing process.
148
149
OA-associated mQTLs operate during skeletogenesis
150
To investigate the presence of OA risk mQTLs in the developing human skeleton, DNAm
151
measured in FL and FC tissues at each of the 39 CpGs was stratified by genotype at the
152
association SNV for each respective locus (Supplementary Material, Fig. S2). Significant
153
mQTLs were identified at 14/39 CpGs in FL (n=15-19, Padj=0.047-3.7x10-5) at the following
154
loci: Locus 1 (4/8 CpGs), Locus 2 (1/3), Locus 3 (2/6), Locus 4 (2/12), Locus 6 (1/2), and
155
Locus 7 (4/6) (Supplementary Material, Fig. S2, Supplementary Material, Table S3). In FC
156
samples, a more comparable tissue to the AC samples in which the mQTLs were originally
157
discovered, significant correlations were identified at 29/39 CpGs (n=54-75, Padj=0.042-
158
4.4x10-19; Supplementary Material, Fig. S2, Supplementary Material, Table S3).
159
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
8
Interestingly, at Locus 1 (harbouring COLGALT2), significant FC mQTLs
160
(Padj=2.00x10-4-6.60x10-11) were identified at 7/8 CpGs across an enhancer marked by
161
cg18131582 (CpG7 at the locus), with a mean genotypic effect (GE) of 33.9% across the
162
investigated region (Fig. 1C). No mQTL was detected at CpG1 in any of the investigated
163
tissues, and this CpG was consistently hypermethylated (mean DNAm FL, 93.4%; FC, 84.3%;
164
AC, 83.84%). In AC, mQTLs were identified at only 5/8 CpGs across the enhancer (Padj<0.016)
165
with a mean GE of 13.8% across the region (Fig. 1C). This indicated an OA risk locus with a
166
stronger and physically broader mQTL effect during development. Conversely, at Locus 7
167
(harbouring RWDD2B), a stronger mean GE was detected in FL (42.0%) tissues than in either
168
FC (7.5%) or AC (11.2%) (Fig. 1C). This indicated that joint tissues other than cartilage could
169
be the site of the functional effect contributing to OA genetic risk at this locus. Furthermore,
170
at Locus 4 (harbouring PLEC), where we identified a strong contribution of genotype to DNAm
171
in both developmental and aged cartilage (GE, 38.7% and 59.2%, respectively; Fig. 1C), FL
172
samples were hypermethylated (mean DNAm 76.1-98.3%), and no significant mQTLs were
173
observed for 10/12 of the CpGs (Padj=0.82-0.07; GE=7.6%), pointing towards a cartilage-
174
specific effect.
175
These data demonstrate the presence of mQTLs associated with OA in the developing
176
human skeleton for the first time, indicating that their functional impact could be exerted from
177
the beginning of life, despite the disease manifesting in older age.
178
179
DNAm is co-regulated between CpGs at individual loci
180
Correlation matrices of DNAm at each of the 39 CpGs were created across all samples (Fig. 2)
181
and in the distinct tissue types (Supplementary Material, Fig. S3). DNAm at CpGs generally
182
clustered by locus, confirming a co-regulation of methylation levels by the association signal
183
(Fig. 2). Notably, at Locus 4 and Locus 7, where multiple mQTL effects have been identified
184
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
9
at the individual loci, the CpGs formed distinct clusters. At Locus 4 (harbouring PLEC), two
185
distinct clusters were identified, with significant negative correlations between the two in FC
186
and AC samples, indicating a single regulatory mechanism with opposing effects upon the two
187
CpG clusters (Supplementary Material, Fig. S3B and C). However, at Locus 7 (harbouring
188
RWDD2B), where 6 CpGs were investigated, the three hypomethylated CpGs in physical
189
proximity (CpGs 2-4, located in the RWDD2B promoter, Fig. 1A) formed a single cluster
190
(Supplementary Material, Fig. S3), whilst DNAm at the three individual CpGs did not cluster
191
with others at the same locus. This indicated a single genetic signal exerting independent
192
regulatory effect upon multiple CpGs around the RWDD2B gene.
193
Here we have shown that the DNAm levels are co-regulated in cartilage by SNVs at
194
each locus throughout the human life course.
195
196
DNAm correlates with allelic expression of OA risk genes in the developing human
197
skeleton
198
We had previously identified allelic expression imbalance (AEI) in aged cartilage samples at
199
nine genes across the seven investigated loci, identifying them as OA effector genes (21,22,24-
200
28) (Fig. 3 and Supplementary Material, Fig. S4, orange plots). Transcriptome-wide analyses
201
of OA cartilage have also reported AEI at most of these genes (29,30). We first confirmed
202
expression of all nine genes in the developmental tissues (Supplementary Material, Fig. S5)
203
and then investigated the presence of AEI during skeletogenesis. In FL samples, significant
204
AEI (P=0.008-0.016) was identified for RWDD2B and PLEC (Fig. 3). The relatively small
205
sample size of heterozygous FL tissues at the loci (n=4-11) may have precluded the
206
identification of significant AEI for other genes, namely SUPT3H (P=0.063, n=5) and SPCS1
207
(P=0.074, n=9) (Supplementary Material, Fig. S4, purple). In FC, significant AEI (P<0.016)
208
was identified at 6/9 genes, including COLGALT2 (mean allelic ratio=1.17, P=0.016, n=17),
209
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
10
SPCS1 (0.93, P<0.0001, n=32), SUPT3H (1.08, P=0.006, n=15), PLEC (0.89, P<0.0001,
210
n=27), ALDH1A2 (0.81, P=0.008, n=8), and RWDD2B (0.73, P<0.0001, n=18). Individual data
211
points for DNA and cDNA ratios are displayed in Supplementary Material, Fig. S4. When
212
significant AEI was identified, the direction of the imbalance occurred in the same direction in
213
all investigated tissues.
214
We next searched for methylation-expression QTLs (meQTLs), whereby DNAm at the
215
CpGs correlated with measured AEI ratios. Methylation M-values were regressed by AEI ratios
216
at each of the nine investigated transcripts (Supplementary Material, Fig. S6). Correlations
217
were identified in all three investigated tissue types and are displayed as a heatmap in Fig. 4.
218
In FL samples, significant meQTLs were detectable at Locus 2 (GNL3, P=0.022); and Locus 4
219
(PLEC, P=0.028-0.008). In FC, significant meQTLs were detectable at Locus 7 (RWDD2B,
220
P=0.005-0.002). Other loci exhibited correlative trends, namely Locus 3 (SUPT3H, r2=0.29-
221
0.69). However, sample size was often limited for meQTL analysis due to multiple factors
222
including low minor allele frequencies (which limited the number of available heterozygous
223
samples) and low gene expression providing limited mRNA template for amplification (which
224
increased the dropout rate of samples following QC). No correlations were identified between
225
the genotypic effect (strength of mQTL) and meQTL r2 value (impact upon allelic imbalance),
226
indicating that the mQTL effect size is not a predictor of functional impact.
227
These data show that OA effector genes are differentially expressed at the allelic level
228
in human cartilage and limb tissues from the start of life, and that the mechanisms leading to
229
cartilage loss in older age are present throughout life. Significant correlations between
230
expression ratios and DNAm indicate a functional role for the observed mQTLs in a
231
spatiotemporal context.
232
233
Identification of open chromatin regions in developmental and aged cartilage
234
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
11
We next investigated chromatin accessibility in the context of OA genetic risk during
235
development and older age. To do so, we conducted assay for transposase accessible chromatin
236
sequencing (ATAC-seq) in developmental and aged cartilage and primarily analysed this
237
dataset on an epigenome wide scale, before integrating this information with our targeted
238
analysis of OA risk loci. For this investigation, we separately isolated chondrocytes from the
239
human foetal proximal and distal femur (representative of the developing human hip and knee,
240
respectively), along with human aged hip and knee articular cartilage. We identified 78,219
241
open chromatin regions that were common across all investigated tissues (Fig. 5A). A total of
242
49,922 open regions were unique to the foetal samples, and 63,058 unique to the aged samples.
243
Between foetal and aged hip cartilage, 113,887 differentially accessible regions (DAR) were
244
identified (FDR<0.05, Supplementary Material, Table S4). Similarly, 121,050 DAR were
245
identified between foetal and aged knee cartilage (Supplementary Material, Table S5).
246
Gene ontology (GO) analysis of the transcripts mapping to the DAR was performed
247
(Fig. 5B), which showed significant enrichment of cellular component morphogenesis
248
(P<6.0x10-10), cell projection morphogenesis (P<1.1x10-12), and cell morphogenesis involved
249
in neuron differentiation (P<8.6x10-9) in the developing foetal cartilage, whereas in the aged
250
cartilage samples there was a significant enrichment of terms including regulation of catalytic
251
activity (P<4.9x10-13) and positive regulation of signal transduction (P<5.3x10-12, Fig. 5B).
252
Differentially accessible peaks were annotated using ROADMAP chromatin state data
253
generated in mesenchymal stem cell (MSC) differentiated cultured chondrocytes (E049) (Fig.
254
5C). DAR were significantly overrepresented in gene enhancers yet underrepresented in gene
255
promoters (P < 0.0001) confirming that the changes in gene expression between development
256
and older age are predominantly driven by altered enhancer activity. Further analysis of
257
sequence motifs within the DAR identified significant enrichment for transcription factor
258
binding motifs including SOX9 (E < 2.9x10-12) and JUN (E < 1.2x10-7) within foetal
259
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
12
chondrocytes (Fig. 5D, Supplementary Material, Tables S6-S9). In aged chondrocytes,
260
significant motif enrichment was also identified for transcription factors (TFs) including
261
ZN264 (E < 8.7x10-68), SP4 and MAZ (E = 1.4x10-52) (Fig. 5E, Supplementary Material, Tables
262
S6-S9). Interestingly, the 5'-TGAGTCA-3' motif, common to transcription regulators including
263
JUN and FOS was enriched in both up- and down-regulated regions in foetal chondrocytes,
264
indicating that the regulation of binding regions for these TFs is integral to chondrocyte
265
homeostasis during both developmental and ageing processes (Fig. 5D and E).
266
These data identify open (and potentially functional) chromatin regions in
267
developmental and aged cartilage. We identified that the majority of differentially accessible
268
regions are annotated as enhancers, which are driving the changes in gene expression
269
throughout the life course via altered transcription factor binding.
270
271
Epigenetic risk of OA and chromatin state in human chondrocytes
272
Intersection of 1,005 SNVs in high LD (EUR r2>0.8) with the investigated association signals
273
with the open chromatin regions prioritized 77 functional variants across the 7 investigated loci
274
(Supplementary Material, Table S10). Interestingly, only 32/77 prioritized SNVs intersected
275
with open chromatin regions in all four tissue types, indicating that, if functional, the effects
276
could be exerted at different stages of the life course. Twenty variants intersected with regions
277
uniquely accessible in aged cartilage, and 16 were unique to developmental samples
278
(Supplementary Material, Fig. S7, Supplementary Material, Table S10).
279
Finally, we intersected the physical location of the 39 investigated CpGs with open
280
chromatin regions to further indicate likely functional timepoints of the mQTLs at the
281
investigated loci (Table 2). Investigated CpGs fell within open chromatin regions at Locus 1
282
(COLGATL2), 2 (GNL3/SPCS1), and 7 (RWDD2B). At these genomic positions, peaks were
283
identified across both foetal and aged cartilage samples, however all three regions were
284
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
13
significantly differentially accessible between foetal and aged chondrocytes (Table 2). At
285
Locus 1 and 2, chromatin was significantly more accessible in foetal knee when compared to
286
aged knee cartilage (log2 fold change=0.89, FDR=0.015; log2 fold change=0.57, FDR=0.022,
287
respectively) (Fig. 6A and B). Conversely, at Locus 7, the promoter region of RWDD2B was
288
significantly less accessible in both foetal hip (-1.07-fold, FDR=1.4x10-6) and foetal knee (-
289
0.51-fold, FDR=0.012) (Fig. 6C, Supplementary Material, Tables S4 and S5).
290
These data have prioritized causal variants falling in open chromatin regions at each of
291
the investigated loci. Finally, we were able to identify in which cartilage tissues our
292
investigated CpGs fell in open regions to provide further evidence for potential timepoints of
293
functional DNAm impact.
294
295
Discussion
296
DNA methylation is the most comprehensively investigated human epigenetic mark, and
297
mQTLs have been widely investigated in disease-relevant adult tissues to fine map causal
298
variants and effector genes at genetic risk loci in complex diseases. To date, the investigation
299
of mQTLs during human development has been limited (18,31-33), and studies have
300
predominantly focused upon cerebral tissues and neuropsychiatric disorders. Such analyses
301
have not been previously conducted in the developing human skeleton. In this report, we used
302
a total of 100 human embryonic and foetal skeletal tissue samples to identify differences in
303
DNAm between developmental and aged limbs and investigate the existence of OA genetic
304
risk mechanisms during development.
305
We investigated 39 CpGs: sites of, or adjacent to, well-characterized OA-mQTLs in
306
aged cartilage. We identified significant DMS at 67% of CpGs when comparing developmental
307
and aged cartilage, the majority of which became hypermethylated in aged cartilage. This is
308
noteworthy, as in ageing tissues a trend towards global hypomethylation has been observed,
309
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
14
known as the epigenetic drift. The relationship between DNAm and age is complex, however
310
the consensus is that DNAm increases across tissues in the early years of life, and then
311
gradually declines with age due to a loss in maintenance stringency (34,35). Furthermore, the
312
co-methylation of proximal CpGs has also been observed to decline with older age (36). We
313
identified co-methylation clustering of CpGs at the seven investigated loci. At Locus 4 (PLEC)
314
we investigated two distal (7.5kb) clusters of CpGs (21) at which DNAm was co-regulated,
315
with strong negative correlations between the CpGs, reflecting the 3D architecture of DNA
316
(35). Conversely, at Locus 7, we investigated 6 mQTL CpGs, only three of which showed a
317
significant co-regulation from the SNV. Importantly, the co-methylation did not appear to be
318
lost with age at any of the loci, suggesting that the epigenetic effects captured by this targeted
319
study are directly relevant to CpGs that are tightly regulated by SNVs and involved in
320
pathogenesis of tissue-specific diseases.
321
We identified OA-mQTLs at 28% of investigated CpGs in FL tissues, and at 85% in
322
FC. We attribute the relative depletion of mQTLs in the FL samples to both a relatively small
323
sample size (n=19), coupled with mQTL tissue specificity. Whilst this is difficult to disentangle
324
at loci where smaller effect sizes were observed, the absence of FL mQTLs at Locus 4,
325
harbouring PLEC, can be attributed to a cartilage-specific effect (CpG 9 GE 71.9% in FC, 6.7%
326
in FL).
327
One of the most striking observations that we made during this study, was the
328
inconsistency in the pattern of observations between the loci, which indicates that the inter-
329
locus mechanisms behind the observed epigenetic changes are distinct. From this we
330
hypothesize that the functional impacts of these risk loci are exerted at different spatiotemporal
331
points during the human life course. An epigenome wide study would potentially allow the
332
identification of clusters of loci that show comparable effects and share regulatory timepoints.
333
Yet, in the current targeted study, this is not possible to determine.
334
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
15
Analyses of transcript expression at the nine OA effector genes identified significant
335
AEI at six genes (P<0.016) in FC, confirming the presence of differential transcript expression.
336
This verifies that the functional genetic risk of OA is occurring at the majority of the
337
investigated loci during skeletal development. We investigated meQTLs to determine whether
338
this was being driven by the differential DNAm and identified significant correlations at 15/39
339
CpGs: FL (Locus 2 and 4), FC (Locus 7), and AC (Locus 1, 2, 4). Our inability to widely detect
340
linear correlations between DNAm and target gene expression is unsurprising. The question of
341
DNAm as a cause vs. consequence has long been debated, and the answer has been hindered
342
due to the complex involvement of multiple cis-regulatory elements (CREs) in the finetuning
343
of gene expression. A recent study used single-cell technologies in mouse embryonic stem cells
344
to investigate the co-occurrence of DNAm with chromatin accessibility, and TF occupancy
345
(37). The authors demonstrated that at most enhancers, DNAm does not antagonise chromatin
346
accessibility, nor the binding of TFs. However, they identified a subset of cell-type specific
347
enhancers at which DNAm directly regulates TF binding, and target gene expression (37). We
348
postulate that such a subset of methylation-sensitive enhancers would be enriched in targeted
349
mQTL investigations such as ours, where the CREs in which they fall have prior association
350
with disease. This is supported by our recent research using dCas9 epigenome modulators,
351
which have shown a direct causality between DNAm and gene expression at OA loci
352
(22,23,25). The identification of tissue specific methylation sensitive enhancers in OA would
353
be bolstered by the expansion of molecular epigenetic investigations to include additional
354
tissues of the articulating joint, rather than solely focusing on cartilage; the inclusion of joint
355
tissues from young adults, although their acquisition in sufficient numbers will be challenging;
356
and the adoption of modern single-cell technologies to look for subpopulations of cells driving
357
disease aetiology within the joint.
358
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
16
Finally, we conducted ATAC sequencing and identified open chromatin regions in
359
developmental and aged samples. The identified accessible regions in the foetal cartilage were
360
enriched for TF binding motifs including SOX9, the master regulatory TF essential for
361
chondrogenesis (38), and FOS/JUN, TFs that can form homo- and heterodimers essential in
362
cartilage development (39-41). Intriguingly, this motif was also enriched in the aged cartilage
363
accessible regions. The FOS/JUN transcription factors have also been shown to play a role in
364
OA pathogenesis (42), and our data further substantiates the notion that stringent regulation of
365
the binding of these TFs to their motifs is integral to cartilage health and function throughout
366
the human life course. We were further able to prioritise causal SNVs across the seven loci,
367
and our data identified three loci at which the investigated CpGs fell into open chromatin
368
regions.
369
At Locus 1, 8 CpGs were investigated across a COLGALT2 enhancer (22). Generally,
370
levels of DNAm were significantly lower and the genotypic effect stronger in the
371
developmental cartilage, consistent with our observation of increased chromatin accessibility
372
during development. Yet, whilst all 7 mQTL CpGs were differentially methylated between FC
373
and AC, the difference in mean DNAm at CpG6 was just 4.5%. Additionally, CpG6 was
374
relatively hypomethylated in the AC samples, when compared to the rest of the investigated
375
sites, and this was the site of the highest measured GE at the locus (55.7%). Furthermore, the
376
only significant meQTLs at this locus were identified in AC samples (CpGs6-7), and whilst
377
AEI for COLGALT2 was present in both FC and AC samples, the differential expression was
378
greater in AC (mean ratio=1.95) than in FC (mean ratio=1.17). We therefore conclude that the
379
biological mechanism underlying OA risk at this locus is established during development,
380
however, the functional impact at this locus is acquired in later life. Notably, the effect size of
381
mQTLs alone does not appear to be a strong predictor of biological function at this locus.
382
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
17
Conversely, Locus 7 appears to be functionally active during foetal development. Here,
383
3/6 CpGs cluster within the RWDD2B promoter and are hypomethylated both in development
384
and older age, consistent with the open chromatin state observed across all tissues. Significant
385
meQTLs were observed in FC samples, indicating that the observed DNAm changes are
386
driving a decrease in RWDD2B expression from the beginning of life. RWDD2B encodes the
387
RWD domain–containing protein 2B, about which very little is currently known (25). Proteins
388
containing RWD domains are capable of binding to enzymes including ubiquitin ligases (43).
389
This lack of knowledge regarding the biological function of RWDD2B makes it a worthy target
390
of future functional studies in the context of musculoskeletal disease.
391
GDF5 (Locus 6) and its regulation have been extensively studied in the context of
392
musculoskeletal development and OA (28,44-48). Identified as a risk gene for OA by the
393
association SNVs rs143383 and rs143384 (49) within the GDF5 5'UTR, early studies into the
394
mechanism behind the regulation revealed AEI for the gene in human aged cartilage and cell
395
models, with the OA risk allele, T, correlating with lower levels of gene expression (28,44,47).
396
More recent, in-depth studies into this locus have utilized murine models, and in vitro reporter
397
assays to identify a Gdf5 enhancer, GROW1, which contains a common derived SNV
398
(rs4911178, G>A) at an otherwise highly conserved position (45). However, to date, the
399
molecular basis of GDF5 expression has not been investigated in primary human foetal
400
chondrocytes. Surprisingly, we did not identify significant AEI for GDF5 in FL or FC samples,
401
however, we did observe a significant imbalance in AC samples, consistent with previous
402
studies (28,44). Therefore, we postulate that, in humans, the differential expression of GDF5
403
becomes more pronounced throughout the life course, rendering Locus 6 one at which the
404
functional risk of OA is conferred in ageing, with the decreased levels of GDF5 hindering the
405
ability of chondrocytes to maintain and repair the cartilage tissue in older age (50).
406
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
18
Whilst decades of research have been dedicated to understanding the genetic aetiology
407
of OA, the clinical exploitation of OA genetic discoveries is still out of reach (51). In this
408
report, we undertook a detailed molecular genetic analysis of well-characterized OA risk loci,
409
using primary human musculoskeletal foetal tissues for the first time. Our data indicate that the
410
functional timepoints of OA risk loci can differ, raising the question of whether causal variants
411
could be classified by whether they confer a developmental or age-acquired risk. Our study
412
further confirms that leveraging tissue specific DNAm data can prioritise effector genes and
413
their regulatory elements yet highlights that the effect size of mQTLs alone is not an indicator
414
of function. The translation of genetic discoveries in the OA field requires a deep understanding
415
of the molecular mechanisms by which risk-conferring alleles impact their target genes, and
416
the appropriate timepoint for therapeutic intervention, prior to macroscopic structural damage
417
occurring in the joint (11). This is the first study to identify the presence of mQTLs in human
418
foetal cartilage and limb tissues, and our report demonstrates that the functional genetic risk of
419
OA can be laid down during human skeletogenesis. Strides are being made within the field,
420
with the first recent reports of polygenic risk scores (PRS) for the disease (52,53), yet the
421
clinical utility of such systems is still lacking (54). We would encourage the integration of
422
epigenetic data at the loci, along with clinical and biochemical parameters to further advance
423
these tools for patient benefit.
424
425
Materials and Methods
426
Human foetal sample collection and processing
427
Human embryonic and foetal tissues were obtained from the MRC and Wellcome Trust funded
428
Human Developmental Biology Resource (HDBR) at Newcastle University
429
(http://www.hdbr.org, project number 200363), with appropriate maternal written consent and
430
approval from the Newcastle and North Tyneside NHS Health Authority Joint Ethics
431
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
19
Committee. HDBR is regulated by the UK Human Tissue Authority (HTA; www.hta.gov.uk)
432
and operates in accordance with the relevant HTA Codes of Practice.
433
The cartilaginous tissue from the ends of developing long bones was specifically
434
isolated and nucleic acids were extracted by the HDBR. Briefly, 20-30mg of tissue was placed
435
in RLT buffer (Qiagen, 1053393) and homogenized with Precelleys lysing kit, CKmix
436
(P000918-LYSko-A) using Precelleys 24 tissue homogenizer (Bertin Technologies). Total
437
RNA and DNA was extracted using an AllPrep DNA/RNA/miRNA Universal Kit (Qiagen,
438
80224), according to the manufacturer’s instructions on a QIAGEN QIAcube Automated
439
DNA, RNA isolation machine. The RNA integrity number (RIN) and concentration for each
440
sample was assessed using a 2100 Bioanalyzer (Agilent). Pooled limb tissues were isolated
441
using the same methodology. Summary statistics of the samples are included in Table 3, with
442
full sample details available in Supplementary Material, Figure S1A and B, and Supplementary
443
Material, Table S11.
444
445
Adult patient sample collection and processing
446
Human articular cartilage samples were obtained from patients undergoing hip or knee joint
447
replacement surgery due to end-stage OA. The tissue collected was macroscopically intact
448
cartilage, distal from the site of the lesion. Arthroplasty was conducted at the Newcastle upon
449
Tyne NHS Foundation Trust hospitals. The Newcastle and North Tyneside Research Ethics
450
Committee granted ethical approval for the collection, with each donor providing verbal and
451
written informed consent (REC reference number 14/NE/1212). Further details of the patient
452
samples used in this project are provided in Supplementary Material, Table S12 and
453
summarised in Table 3.
454
RNA was extracted from cartilage by TRIzol-chloroform (Life Technologies)
455
separation, following which the RNA was purified from the aqueous phase using the RNeasy
456
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
20
Mini Kit (Qiagen). DNA was extracted from cartilage using the EZNA DNA Isolation kit
457
(Omega Bio-Tek). For genotyping, 50ng genomic DNA was amplified by PCR as described
458
below. For methylation analysis, 500ng DNA was deaminated using sodium bisulfite with the
459
EZ DNA methylation kit (Zymo Research).
460
461
Gene expression analysis
462
Expression of each of the 9 genes of interest was measured across the 7 loci in the three
463
investigated tissue types. Complementary DNA (cDNA) was reverse transcribed from total
464
RNA using the Superscript IV standard protocol (Invitrogen) after an initial 15-minute
465
treatment with 1 unit of amplification grade DNaseI (Invitrogen). Gene expression was
466
measured by reverse transcription quantitative polymerase chain reaction (RT‐qPCR) using
467
pre-designed TaqMan assays (Integrated DNA Technologies, Supplementary Material, Table
468
S13). Gene expression was quantified using TaqMan chemistry, normalized to housekeeping
469
genes 18S, HPRT1 and GAPDH and expressed as 2-Δct as described previously (22).
470
471
Pyrosequencing
472
PyroMark Q24 Advanced (Qiagen) was used to genotype all patient DNA samples as
473
previously described (24). At locus 4 and 5, restriction fragment length polymorphism assays
474
were used as previously described (21,24). Pyrosequencing was also used to quantify DNAm
475
at the 39 CpGs investigated in this study following bisulfite conversion of DNA (EZ DNA
476
Methylation Kit, Zymo). Each sample was amplified in duplicate. Samples were excluded from
477
the analysis if the replicates differed by >5%. Assays were designed using PyroMark assay
478
design software 2.0. All primer sequences used for genotyping, methylation and allelic
479
quantification are listed in Supplementary Material, Table S14.
480
481
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
21
Allelic expression imbalance analysis
482
Allelic expression imbalance (AEI) analysis was also performed by pyrosequencing as
483
previously described (27). Briefly, samples that were heterozygous for OA association SNVs
484
were genotyped at SNVs falling within one of the nine transcripts of interest. For COLGALT2,
485
SUPT3H, and ALDH1A2, the association SNVs fell within transcribed regions and were used
486
directly for AEI analysis. All other transcript SNVs were in high LD with the association SNV
487
(r2>0.82 in European ancestry cohorts (EUR), Supplementary Material, Table S14), except for
488
RUNX2, where AEI analysis was performed using a variant in linkage equilibrium as
489
previously described in detail (26). Allelic quantification at the transcript variants was
490
performed in triplicate in DNA and cDNA from each heterozygous sample. The ratio measured
491
in the cDNA was then normalized to the DNA ratio. Samples with triplicate values which
492
differed by >5% were excluded from analysis.
493
494
Assay for transposase accessible chromatin sequencing (ATAC-seq)
495
Fresh cartilage tissue from proximal (hip, n=6) and distal (knee, n=6) human foetal femurs at
496
12 post conception weeks, was dissected and provided by the HDBR. Adult articular cartilage
497
was dissected from hip (n=5) and knee (n=5) joint arthroplasty obtained as described above.
498
For chondrocyte isolation, the cartilage was digested in 1% collagenase solution (Sigma
499
Aldrich) for 3h (foetal) or 16h (arthroplasty) at 37˚C. Cells were washed with 1xPBS, re-
500
suspended in 5%FBS-DMEM and counted. Nuclei were isolated from 50,000 cells by 3min
501
incubation on ice in lysis buffer (5M NaCl, 1M MgCl2, 1M Tris-HCl (pH 7.5), 10% w/v NP-
502
40 (Roche), 10% w/v Tween 20 (Roche) and 1% w/v Digitonin (Promega) (55). Isolated nuclei
503
were re-suspended in a transposase mix (25μl 2xTD Buffer, 2.5μl TDE1 enzyme (Nextera Tn5
504
transposase, Illumina), 0.5μl 1% w/v Digitonin, 0.5μl 10% w/v Tween 20, 16.5μl 1xPBS and
505
5μl nuclease-free H2O) and incubated for 30min at 37˚C, 1000rpm. DNA was purified using
506
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
22
MinElute PCR Purification kit (Qiagen) according to manufacturer’s instructions and eluted in
507
10μl H2O. The purified DNA was partially amplified by PCR in a 50μl reaction using the NEB
508
Next High-Fidelity PCR Master Mix (New England Biolabs, Hitchin, UK) and primer
509
Ad1_noMX in combination with barcoded primers (Supplementary Material, Table S15). A
510
5μl aliquot of the reaction was amplified by quantitative PCR (qPCR) and the remaining 45μl
511
was amplified by PCR for an additional number of cycles, calculated for each sample based on
512
the qPCR readings, to avoid amplification to saturation. The libraries were purified using
513
AMPure XP magnetic beads (Beckman Coulter). The DNA libraries were sequenced by
514
Newcastle University’s Genomics Core Facility on a NextSeq S1 generating paired-end 100
515
bp reads.
516
517
Bioinformatic and Statistical analyses
518
Pre-processing: Quality of FASTQ files was assessed with FastQC Version 11.8 (56). Adapter
519
sequences were removed using Trimmomatic Version 0.36 (57). Paired-end reads were aligned
520
to GRCh38 using Bowtie2 Version 2.3.4.2 (58). Multi-mapped reads, duplicate reads, and
521
reads aligning to the mitochondrial genome were removed using Picard Tools Version 2.2.4
522
and SAMtools Version 1.9 (59). Peaks were called using MACS2 Version 2.1.1 and ENCODE
523
blacklisted regions were removed from the peak calls using BEDTools Version 2.27.1 (60).
524
The R package DiffBind (Version 2.16.2) was used to derive consensus peak sets for
525
further analysis. Consensus peak sets consisted of peaks identified in at least 4/6 of the foetal
526
samples, and at least 3/5 of the aged samples. Consensus peak sets were lifted to hg19 using
527
the UCSC liftOver tool and then annotated using the R package ChIPseeker (Version 1.26.2)
528
(61). The Diffbind package was used to determine sites that are differentially accessible
529
between sample groups.
530
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
23
Motif discovery in the differentially accessible sites was performed using MEME-Chip
531
from the MEME Suite (version 5.4.1) (62). Enriched GO terms were determined using the R
532
package GOstats (version 2.56.0) (63). Intersections with CpG sites and ROADMAP
533
chromatin state maps were identified using BEDTools (60).
534
535
Data Availability
536
All data associated with this study are present in the paper or supplementary materials. The
537
ATAC-sequencing data is available via the Gene Expression Omnibus (GEO, accession
538
number GSE214394).
539
540
Acknowledgements
541
The authors would like to thank Dr David Wilkinson for his discussions on the use of cartilage
542
markers in the foetal samples. The human embryonic and foetal material was provided by the
543
Joint MRC/Wellcome Trust (grant# MR/R006237/1) Human Developmental Biology
544
Resource (http://hdbr.org). We would like to thank Steven N. Lisgo and Lynne M. Overman
545
for their support in supplying these HDBR samples. Additionally, we are grateful to the support
546
of the Genomics Core Facility at Newcastle University. The graphical abstract and schematic
547
images of human foetal samples were created using BioRender (agreement number
548
JD24FABNKB). This work was supported by Versus Arthritis (grants 20771 to J.L. and 22615
549
to S.J.R.), by the Medical Research Council and Versus Arthritis Centre for Integrated
550
Research into Musculoskeletal Ageing (CIMA, grant references MR/P020941/1 and
551
MR/R502182/1 to J.L.), and by the Ruth & Lionel Jacobson Charitable Trust (to J.L.).
552
553
Conflict of Interest Statement
554
The authors declare no conflict of interests.
555
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
24
556
References
557
1. Prieto-Alhambra, D., Judge, A., Javaid, M.K., Cooper, C., Diez-Perez, A. and Arden,
558
N. (2014) Incidence and risk factors for clinically diagnosed knee, hip and hand
559
osteoarthritis: influences of age, gender and osteoarthritis affecting other joints. Ann.
560
Rheum. Dis., 73, 1659-1664.
561
2. Hunter, D. J., March, L. and Chew, M. (2020) Osteoarthritis in 2020 and beyond: a
562
Lancet Commission. Lancet, 396, 1711-1712.
563
3. Kendzerska, T., Jüni, P., King, L. K., Croxford, R., Stanaitis, I. and Hawker G. A.
564
(2017) The longitudinal relationship between hand, hip and knee osteoarthritis and
565
cardiovascular events: a population-based cohort study. Osteoarthritis Cartilage, 25,
566
1771-1780.
567
4. Wang, H., Bai, J., He, B., Hu, X. and Liu, D. (2016) Osteoarthritis and the risk of
568
cardiovascular disease: a meta-analysis of observational studies. Sci. Rep., 6, 39672.
569
5. Palazzo, C., Nguyen, C., Lefevre-Colau, M. M., Rannou, F. and Poiraudeau, S. (2016)
570
Risk factors and burden of osteoarthritis. Ann. Phys. Rehabil. Med., 59, 134-138.
571
6. Richard, D., Liu, Z., Cao, J., Kiapour, A. M., Willen, J., Yarlagadda, S., Jagoda, E.,
572
Kolachalama, V. B., Sieker, J. T., Chang, G. H., et al. (2020) Evolutionary selection
573
and constraint on human knee chondrocyte regulation impacts osteoarthritis risk. Cell,
574
181, 362-381.
575
7. Loughlin, J. (2015) Genetic contribution to osteoarthritis development: current state of
576
evidence. Curr. Opin. Rheumatol., 27, 284–288.
577
8. Muthuirulan, P., Zhao, D., Young, M., Richard, D., Liu, Z., Emami, A., Portilla, G.,
578
Hosseinzadeh, S., Cao, J., Maridas, D., et al. (2021) Joint disease-specificity at the
579
regulatory base-pair level. Nat. Commun., 12, 4161.
580
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
25
9. Pitsillides, A. A. and Beier, F. (2011) Cartilage biology in osteoarthritis - lessons from
581
developmental biology. Nat. Rev. Rheumatol., 7, 654-663.
582
10. Qi, Y., Li, B., Wen, Y., Yang, X., Chen, B., He, Z., Zhao, Z., Magdalou, J., Wang, H.
583
and Chen, L. (2021) H3K9ac of TGFβRI in human umbilical cord: a potential
584
biomarker for evaluating cartilage differentiation and susceptibility to osteoarthritis via
585
a two-step strategy. Stem Cell Res. Ther., 12, 163.
586
11. Mahmoudian, A., Lohmander, L. S., Mobasheri, A., Englund, M. and Luyten, F. P.
587
(2021) Early-stage symptomatic osteoarthritis of the knee - time for action. Nat. Rev.
588
Rheumatol., 17, 621-632.
589
12. Swingler, T. E., Wheeler, G., Carmont, V., Elliott, H. R., Barter, M. J., Abu-Elmagd,
590
M., Donell, S. T., Boot-Handford, R. P., Hajihosseini, M. K., Münsterberg, A., et al.
591
(2012) The expression and function of microRNAs in chondrogenesis and
592
osteoarthritis. Arthritis Rheumatol., 64, 1909-1919.
593
13. Farhang, N., Brunger, J. M., Stover, J. D., Thakore, P. I., Lawrence, B., Guilak, F.,
594
Gersbach, C. A., Setton, L. A. and Bowles, R. D. (2017) CRISPR-based epigenome
595
editing of cytokine receptors for the promotion of cell survival and tissue deposition in
596
inflammatory environments. Tissue Eng. Part A., 23, 738-749.
597
14. Aubourg, G., Rice, S. J., Bruce-Wootton, P. and Loughlin, J. (2022) Genetics of
598
osteoarthritis. Osteoarthritis Cartilage, 30, 636-649.
599
15. Perzel Mandell, K. A., Eagles, N. J., Wilton, R., Price, A. J., Semick, S. A., Collado-
600
Torres, L., Ulrich, W. S., Tao, R., Han, S., Szalay, A. S., et al. (2021) Genome-wide
601
sequencing-based identification of methylation quantitative trait loci and their role in
602
schizophrenia risk. Nat. Commun., 12, 5251.
603
16. Villicaña, S. and Bell, J. T. (2021) Genetic impacts on DNA methylation: research
604
findings and future perspectives. Genome Biol., 22, 127.
605
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
26
17. Zhang, T., Choi, J., Dilshat, R., Einarsdóttir, B. Ó., Kovacs, M. A., Xu, M., Malasky,
606
M., Chowdhury, S., Jones, K., Bishop, D. T., et al. (2021) Cell-type-specific meQTLs
607
extend melanoma GWAS annotation beyond eQTLs and inform melanocyte gene-
608
regulatory mechanisms. Am. J. Hum. Genet., 108, 1631-1646.
609
18. Jaffe, A. E., Gao, Y., Deep-Soboslay, A., Tao, R., Hyde, T. M., Weinberger, D. R. and
610
Kleinman, J. E. (2016) Mapping DNA methylation across development, genotype and
611
schizophrenia in the human frontal cortex. Nat. Neurosci., 19, 40-47.
612
19. Rushton, M. D., Reynard, L. N., Young, D. A., Shepherd, C., Aubourg, G., Gee, F.,
613
Darlay, R., Deehan, D., Cordell, H. J. and Loughlin, J. (2015) Methylation quantitative
614
trait locus analysis of osteoarthritis links epigenetics with genetic risk. Hum. Mol.
615
Genet., 24, 7432-7444.
616
20. Rice, S. J., Cheung, K., Reynard, L. N. and Loughlin, J. (2019) Discovery and analysis
617
of methylation quantitative trait loci (mQTLs) mapping to novel osteoarthritis genetic
618
risk signals. Osteoarthritis Cartilage, 27, 1545-1556.
619
21. Rice, S. J., Tselepi, M., Sorial, A. K., Aubourg, G., Shepherd, C., Almarza, D., Skelton,
620
A. J., Pangou, I., Deehan, D., Reynard, L. N., et al. (2019) Prioritization of PLEC and
621
GRINA as osteoarthritis risk genes through the identification and characterization of
622
novel methylation quantitative trait loci. Arthritis Rheumatol., 71, 1285-1296.
623
22. Kehayova, Y. S., Watson, E., Wilkinson, J. M., Loughlin, J. and Rice, S. J. (2021)
624
Genetic and epigenetic interplay within a COLGALT2 enhancer associated with
625
osteoarthritis. Arthritis Rheumatol., 73, 1856-1865.
626
23. Rice, S. J., Roberts, J. B., Tselepi, M., Brumwell, A., Falk, J., Steven, C. and Loughlin,
627
J. (2021) Genetic and epigenetic fine-tuning of TGFB1 expression within the human
628
osteoarthritic joint. Arthritis Rheumatol., 73, 1866-1877.
629
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
27
24. Shepherd, C., Zhu, D., Skelton, A. J., Combe, J., Threadgold, H., Zhu, L., Vincent, T.
630
L., Stuart, P., Reynard, L. N. and Loughlin, J. (2018) Functional characterization of the
631
osteoarthritis genetic risk residing at ALDH1A2 identifies rs12915901 as a key target
632
variant. Arthritis Rheumatol., 70, 1577-1587.
633
25. Parker, E., Hofer, I., Rice, S. J., Earl, L., Anjum, S. A., Deehan, D. J. and Loughlin, J.
634
(2021) Multi-tissue epigenetic and gene expression analysis combined with epigenome
635
modulation identifies RWDD2B as a target of osteoarthritis susceptibility. Arthritis
636
Rheumatol., 73, 100-109.
637
26. Rice, S. J., Aubourg, G., Sorial, A. K., Almarza, D., Tselepi, M., Deehan, D. J.,
638
Reynard, L. N. and Loughlin, J. (2018) Identification of a novel, methylation
639
dependent, RUNX2 regulatory region associated with osteoarthritis risk. Hum. Mol.
640
Genet., 27, 3464-3474.
641
27. Gee, F., Clubbs, C. F., Raine, E. V. A., Reynard, L. N. and Loughlin, J (2014) Allelic
642
expression analysis of the osteoarthritis susceptibility locus that maps to chromosome
643
3p21 reveals cis-acting eQTLs at GNL3 and SPCS1. BMC Med. Genet., 15, 53.
644
28. Southam, L., Rodriguez-Lopez, J., Wilkins, J. M., Pombo-Suarez, M., Snelling, S.,
645
Gomez-Reino, J. J., Chapman, K., Gonzalez, A. and Loughlin, J. (2007) A SNP in the
646
5'-UTR of GDF5 is associated with osteoarthritis susceptibility in Europeans and with
647
in vivo differences in allelic expression in articular cartilage. Hum. Mol. Genet., 16,
648
2226-2232.
649
29. den Hollander, W., Pulyakhina, I., Boer, C., Bomer, N., van der Breggen, R.,
650
Arindrarto, W., Coutinho de Almeida, R., Lakenberg, N., Sentner, T., Laros, J.F.J., et
651
al. (2019) Annotating transcriptional effects of genetic variants in disease-relevant
652
tissue: transcriptome-wide allelic imbalance in osteoarthritic cartilage. Arthritis
653
Rheumatol., 71, 561-570.
654
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
28
30. Coutinho de Almeida, R., Tuerlings, M., Ramos, Y., den Hollander, W., Suchiman, E.,
655
Lakenberg, N., Nelissen, R.G.H.H., Mei, H. and Meulenbelt, I. (2022) Allelic
656
expression imbalance in articular cartilage and subchondral bone refined genome-wide
657
association signals in osteoarthritis. Rheumatology (in press).
658
31. Hoffmann, A., Ziller, M. and Spengler, D. (2016) The future is the past: methylation
659
QTLs in schizophrenia. Genes (Basel), 7, 104.
660
32. Andrews, S. V., Ellis, S. E., Bakulski, K. M., Sheppard, B., Croen, L. A., Hertz-
661
Picciotto, I., Newschaffer, C. J., Feinberg, A. P., Arking, D. E., Ladd-Acosta, C., et al.
662
(2017) Cross-tissue integration of genetic and epigenetic data offers insight into autism
663
spectrum disorder. Nat. Commun., 8, 1011.
664
33. Bonder, M. J., Kasela, S., Kals, M., Tamm, R., Lokk, K., Barragan, I., Buurman, W.
665
A., Deelen, P., Greve, J. W., Ivanov, M. et al. (2014) Genetic and epigenetic regulation
666
of gene expression in foetal and adult human livers. BMC Genomics, 15, 860.
667
34. Jones, M. J., Goodman, S. J. and Kobor, M. S. (2015) DNA methylation and healthy
668
human aging. Aging Cell, 14, 924-932.
669
35. Seale, K., Horvath, S., Teschendorff, A., Eynon, N. and Voisin, S. (2022) Making sense
670
of the ageing methylome. Nat. Rev. Genet., 23, 585-605.
671
36. Heyn, H., Li, N., Ferreira, H. J., Moran, S., Pisano, D. G., Gomez, A., Diez, J., Sanchez-
672
Mut, J. V., Setien, F., Carmona, F. J., et al. (2012) Distinct DNA methylomes of
673
newborns and centenarians. Proc. Natl. Acad. Sci. USA, 109, 10522-10527.
674
37. Kreibich, E., Kleinendorst, R., Barzaghi, G., Kaspar, S. and Krebs, A. R. (2022) Single
675
molecule multi-omics reveals context-dependent regulation of enhancers by DNA
676
methylation. bioRxiv 2022.05.19.492653. doi:10.1101/2022.05.19.492653.
677
38. Lefebvre, V., Angelozzi, M. and Haseeb, A. (2019) SOX9 in cartilage development
678
and disease. Curr. Opin. Cell Biol., 61, 39-47.
679
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
29
39. Papachristou, D., Pirttiniemi, P., Kantomaa, T., Agnantis, N. and Basdra, E. K. (2006)
680
Fos- and Jun-related transcription factors are involved in the signal transduction
681
pathway of mechanical loading in condylar chondrocytes. Eur. J. Orthod., 28, 20-26.
682
40. Mechta-Grigoriou, F., Gerald, D. and Yaniv, M. (2001) The mammalian Jun proteins:
683
redundancy and specificity. Oncogene, 20, 2378-2389.
684
41. Karreth, F., Hoebertz, A., Scheuch, H., Eferl, R. and Wagner, E. F. (2004) The AP1
685
transcription factor Fra2 is required for efficient cartilage development. Development,
686
131, 5717-5725.
687
42. Neefjes, M., van Caam, A. P. M. and van der Kraan, P. M. (2020) Transcription factors
688
in cartilage homeostasis and osteoarthritis. Biology (Basel), 9, 290.
689
43. Alontaga, A. Y., Ambaye, N. D., Li, Y. J., Vega, R., Chen, C. H., Bzymek, K. P.,
690
Williams, J. C., Hu, W. and Chen, Y. (2015) RWD domain as an E2 (Ubc9)-interaction
691
module. J. Biol. Chem., 290, 16550-16559.
692
44. Egli, R. J., Southam, L., Wilkins, J. M., Lorenzen, I., Pombo-Suarez, M., Gonzalez, A.,
693
Carr, A., Chapman, K. and Loughlin, J. (2009) Functional analysis of the osteoarthritis
694
susceptibility-associated GDF5 regulatory polymorphism. Arthritis Rheumatol., 60,
695
2055-2064.
696
45. Capellini, T. D., Chen, H., Cao, J., Doxey, A. C., Kiapour, A. M., Schoor, M. and
697
Kingsley, D. M. (2017) Ancient selection for derived alleles at a GDF5 enhancer
698
influencing human growth and osteoarthritis risk. Nat. Genet., 49, 1202-1210.
699
46. Chen, H., Capellini, T. D., Schoor, M., Mortlock, D. P., Reddi, A. H. and Kingsley, D.
700
M. (2016) Heads, shoulders, elbows, knees, and toes: modular Gdf5 enhancers control
701
different joints in the vertebrate skeleton. PLoS Genet., 12, e1006454.
702
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
30
47. Syddall, C. M., Reynard, L. N., Young, D. A. and Loughlin, J (2013) The identification
703
of trans-acting factors that regulate the expression of GDF5 via the osteoarthritis
704
susceptibility SNP rs143383. PLoS Genet., 9, e1003557.
705
48. Reynard, L. N., Bui, C., Syddall, C. M. and Loughlin, J. (2014) CpG methylation
706
regulates allelic expression of GDF5 by modulating binding of SP1 and SP3 repressor
707
proteins to the osteoarthritis susceptibility SNP rs143383. Hum. Genet., 133, 1059-
708
1073.
709
49. Miyamoto, Y., Mabuchi, A., Shi, D., Kubo, T., Takatori, Y., Saito, S., Fujioka, M.,
710
Sudo, A., Uchida, A., Yamamoto, S., et al. (2007) A functional polymorphism in the 5'
711
UTR of GDF5 is associated with susceptibility to osteoarthritis. Nat. Genet., 39, 529-
712
533.
713
50. Kania, K., Colella, F., Riemen, A., Wang, H., Howard, K. A., Aigner, T., Dell'Accio,
714
F., Capellini, T. D., Roelofs, A. J. and De Bari, C. (2020) Regulation of Gdf5 expression
715
in joint remodelling, repair and osteoarthritis. Sci. Rep., 10, 157.
716
51. Loughlin, J. (2022) Translating osteoarthritis genetics research: challenging times
717
ahead. Trends Mol. Med., 28, 176-182.
718
52. Lacaze, P., Wang, Y., Polekhina, G., Bakshi, A., Riaz, M., Owen, A., Franks, A., Abidi,
719
J., Tiller, J., McNeil, J., et al. (2022) Genomic risk score for advanced osteoarthritis in
720
older adults. Arthritis Rheumatol., 74, 1480-1487.
721
53. Sedaghati-Khayat, B., Boer, C. G., Runhaar, J., Bierma-Zeinstra, S., Broer, L., Ikram,
722
M. A., Zeggini, E., Uitterlinden, A. G., van Rooij, J. and van Meurs, J. (2022) Risk
723
assessment for hip and knee osteoarthritis using polygenic risk scores. Arthritis
724
Rheumatol., 74, 1488-1496.
725
54. Yau, M. S. and Loughlin, J. (2022) Towards precision medicine - is genetic risk
726
prediction ready for prime time in osteoarthritis? Arthritis Rheumatol, 74, 1477-1479.
727
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
31
55. Buenrostro, J. D., Wu, B., Chang, H. Y. and Greenleaf, W. J. (2015) ATAC-seq: a
728
method for assaying chromatin accessibility genome-wide. Curr. Protoc. Mol. Biol.,
729
109, 21.29.1–21.29.9.
730
56. Andrews, S. (2010) FastQC: a quality control tool for high throughput sequence data
731
available online at http://www.bioinformatics.babraham.ac.uk/projects/fastqc/.
732
57. Langmead, B., Trapnell, C., Pop, M. and Salzberg, S. L. (2009) Ultrafast and memory-
733
efficient alignment of short DNA sequences to the human genome. Genome Biol., 10,
734
R25.
735
58. Danecek, P., Bonfield, J. K., Liddle, J., Marshall, J., Ohan, V., Pollard, M. O.,
736
Whitwham, A., Keane, T., McCarthy, S. A., Davies, R. M., et al. (2021) Twelve years
737
of SAMtools and BCFtools. Gigascience, 10, giab008.
738
59. Zhang, Y., Liu, T., Meyer, C. A., Eeckhoute, J., Johnson, D. S., Bernstein, B. E.,
739
Nusbaum, C., Myers, R. M., Brown, M., Li, W., et al. (2008) Model-based analysis of
740
ChIP-Seq (MACS). Genome Biol., 9, R137.
741
60. Quinlan, A. R. and Hall, I. M. (2010) BEDTools: a flexible suite of utilities for
742
comparing genomic features. Bioinformatics, 26, 841-842.
743
61. Yu, G., Wang, L. G. and He, Q. Y. (2015) ChIPseeker: an R/Bioconductor package for
744
ChIP peak annotation, comparison and visualization. Bioinformatics, 31, 2382-2383.
745
62. Machanick, P. and Bailey, T. L. (2011) MEME-ChIP: motif analysis of large DNA
746
datasets. Bioinformatics, 27, 1696-1697.
747
63. Falcon, S. and Gentleman, R. (2007) Using GOstats to test gene lists for GO term
748
association. Bioinformatics, 23, 257-258.
749
750
Legends to Figures
751
752
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
32
753
Figure 1. Measurements of DNA methylation at the 39 investigated CpGs, spanning 7 OA risk
754
loci. (A) Schematic diagram of the 7 investigated loci showing the genomic position of each
755
of the 39 CpGs and the OA association SNVs marking the regions. Gene transcripts are shown
756
in dark blue. The chromatin state, determined in mesenchymal stem cell (MSC)-derived
757
cultured chondrocytes (E049) by ChIP-seq in the ROADMAP epigenomics database, is
758
indicated at the foot of each panel. TSS, transcription start site; UTR, untranslated region. (B)
759
Left-hand panel, Heatmap indicating the mean DNAm levels in each of the three investigated
760
human tissue types: FL, foetal limb; FC, foetal cartilage; AC, aged cartilage. Right-hand
761
panel, Heatmap indicating the genotypic effect, the amount by which observed changed in
762
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
33
DNAm can be explained by the carriage of one or more risk allele, at each CpG. Both heatmaps
763
in B-C are clustered by locus and CpG.
764
765
766
Figure 2. Co-regulation of mQTL CpGs by single SNVs at the loci. Correlation matrix of
767
DNAm values at the 39 CpGs. The colour of the circles represents the r2 correlation (red, strong
768
positive correlation (1.0); white, no correlation (0.0); blue, strong negative correlation (-1.0).
769
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
34
The areas of circles show the absolute value of corresponding correlation coefficients and
770
CpGs are ordered using the hierarchical clustering method, the results of which are summarized
771
in the dendrogram above the matrix.
772
773
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
35
774
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
36
775
Figure 3. Allelic expression imbalance of OA risk genes is present in human developmental
776
limb tissues. Nine genes were investigated across the seven loci. Violin plots represent the
777
mean allelic ratio in foetal limb (purple), foetal cartilage (green) and aged cartilage (orange)
778
tissues for COLGALT2 (n= FL, 6; FC, 17; AC, 25), GNL3 (n= FL, 11; FC, 34; AC, 20), SPCS1
779
(n= FL, 9; FC, 32; AC, 22), SUPT3H (n= FL, 5; FC, 15; AC, 18), RUNX2 (n= FL, 7; FC, 19;
780
AC, 16), PLEC (n= FL, 7; FC, 27; AC, 13), ALDH1A2 (n= FL, 4; FC, 8; AC, 14), GDF5 (n=
781
FL, 6; FC, 22; AC, 23), and RWDD2B (n= FL, 8; FC, 18; AC, 20). Individual data points are
782
shown, and horizontal bars represent the median and interquartile range. For RUNX2, the
783
transcript variant was in linkage equilibrium (r2=0.0) and so the cDNA allelic ratios were
784
stratified by samples that were homozygous for either the major (A) or minor (G) allele at the
785
association SNV, rs10948172 (black triangles) and those that were heterozygous (coloured
786
triangles).
787
788
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
37
789
Figure 4. Correlations between DNAm and allelic expression are present in the developing
790
and aged human skeleton. Heatmap showing r2 values between the Log2 AEI ratio for
791
expression of the nine investigated genes and methylation M-values at each CpG in foetal limb
792
(FL), foetal cartilage (FC), and aged cartilage (AC). Negative slopes are coloured from white
793
(r2=0.0, no correlation) to blue (r2=1.0, perfect correlation) and positive slopes are coloured
794
from white (r2=0.0, no correlation) to red (r2=1.0, perfect correlation). P-values were calculated
795
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
38
using simple linear regression and corrected for multiple testing using the method of
796
Bonferroni. **, Padj < 0.01; *, Padj < 0.05.
797
798
799
Figure 5. Differential chromatin accessibility between developing and aged hip and knee
800
cartilage. (A) Venn diagram summarizing the number of differentially accessible regions
801
(DAR) between foetal hip (blue), foetal knee (pink), aged hip (orange), and aged knee (green)
802
cartilage. (B) Gene ontology (GO) analysis of the transcripts mapping to the DAR between
803
foetal and aged cartilage. The size of the circle represents the gene count mapping to the
804
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
39
respective GO term and the colour represents the P-value. Horizonal black bars connect the
805
terms that are upregulated in both foetal hip and foetal knee, or down regulated in both foetal
806
hip and foetal knee. (C) Enrichment of peaks mapping to ROADMAP chromatin states in E049
807
MSC-cultured chondrocytes. The left-hand plot shows the distribution of peaks across the
808
genome. The middle three plots show the peak enrichment in foetal hip (FH) and aged hip
809
(AH) samples, along with the enrichment in the differentially accessible hip peaks (DAR(H)).
810
The right-hand plots show the peak enrichment in foetal knee (FK) and aged knee (AK)
811
samples, along with the enrichment in the differentially accessible knee peaks (DAR(K)). The
812
percentages of peaks mapping to enhancer (yellow) or promoter (red) regions are annotated.
813
Quies, quiescent; Rep, repressed polycomb; Enh, enhancer; Tx, transcribed region; TSS, active
814
transcription start site. (D) The transcription factor motifs most highly enriched in accessible
815
regions in foetal cartilage. (E) The transcription factor motifs most highly enriched in
816
accessible regions in aged cartilage.
817
818
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
40
819
Figure 6. OA-mQTL CpGs fall within differentially accessible regions in foetal and aged hip
820
and knee cartilage. (A) The 8 CpGs investigated at Locus 1 fall within an intronic COLGALT2
821
enhancer. (B) The three intronic CpGs investigated at Locus 2 reside on the edge of an open
822
chromatin region. (C) The Locus 7 CpGs that fall within the RWDD2B promoter. For A-C, the
823
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
41
regions marked as differentially accessible (DAR) in hip and knee cartilage are marked with
824
blue bars. Blue peaks show the average accessibility in each of the four investigated tissue
825
types. The ROADMAP chromatin state in MSC-derived cultured chondrocytes (E049) is
826
summarized at the foot of each diagram.
827
Table 1. Summary of the loci included in this investigation
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
42
Lo
cus
GWA
S SNP
NEA
/EA
E
A
F
SNP
Chrom
osome
SNP
Positio
n
Locus
CpG
Discovery
CpG(s)
CpG
Position
(hg19)
Discovery
P-value
Gene*
Regio
n
Annot
ation
Refer
ences
1
rs1158
3641
T/C
0.
83
1
183,90
6,245
1
183,911,97
0
COLG
ALT2
Introni
c;
Enhan
cer
20,22
2
183,911,99
9
3
183,912,01
4
4
183,912,02
0
5
183,912,15
4
6
183,912,18
7
7
cg181315
82
183,912,30
5
0.003
8
183,912,39
7
2
rs6976
C/T
0.
41
3
52,728,
804
1
cg180994
08
52,552,593
3.73x10-6
GNL3
SPCS1
Exoni
c
19,27
2
52,552,598
3
52,552,602
3
rs1094
8172
A/G
0.
20
6
44,777,
691
1
cg139797
08
44,695,318
6.2x10-5
SUPT
3H
RUNX
2
Interg
enic
19,26
2
cg192547
93
44,695,348
9.00x10-03
3
cg209137
47
44,695,427
4.9x10-12
4
cg185512
25
44,695,536
1.12x10-10
5
44,695,543
6
44,695,547
4
rs1178
0978
G/A
0.
23
8
145,03
4,852
1
145,001,36
1
PLEC
Gene
body;
Transc
ribed
21
2
145,001,37
8
3
145,001,38
4
4
145,001,40
6
5
cg194051
77
145,001,42
8
3.33x10-17
6
145,001,44
4
7
145,001,46
4
8
145,001,48
5
9
cg145988
46
145,008,90
9
2.72x10-19
10
145,008,91
8
11
145,008,92
7
12
145,008,93
0
5
rs3204
689
C/G
0.
26
15
58,246,
802
1
cg120319
62
58,353,849
2.0x10-8
ALDH
1A2
Introni
c;
Repre
ssed
19,24
2
58,353,861
6
rs1433
84
rs1433
83
T/C
0.
47
20
34,025,
756
1
cg147522
27
34,000,481
0.010
GDF5
Introni
c
19
C/T
0.
45
34,025,
983
2
34,000,519
7
rs6516
886
T/A
0.
41
21
30,393,
664
1
cg000653
02
30,366,250
0.040
RWD
D2B
Interg
enic
21,25
2
cg054680
28
30,391,383
0.006
Introni
c;
Promo
ter
3
30,391,385
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
43
*Putative effector genes at the loci, which have previously shown allelic imbalance in investigations using OA tissues. NEA, non-effect
allele; EA, effect allele. Regions were annotated using ROADMAP chondrocyte data (E049).
828
Table 2. OA-mQTL CpGs fall in open chromatin regions in developmental and aged hip and
knee cartilage
Locus
Locus
CpG
Discovery
CpG(s)
CpG Position
(hg19)
Gene
Region Annotation
Open Chromatin?
DAR (fold)
FH
FK
AH
AK
H
K
1
1
183,911,970
COLGALT2
Intronic; Enhancer
2
183,911,999
3
183,912,014
4
183,912,020
5
183,912,154
6
183,912,187
0.89
7
cg18131582
183,912,305
8
183,912,397
2
1
cg18099408
52,552,593
GNL3
SPCS1
Exonic
2
52,552,598
0.57
3
52,552,602
3
1
cg13979708
44,695,318
SUPT3H
RUNX2
Intergenic
2
cg19254793
44,695,348
3
cg20913747
44,695,427
4
cg18551225
44,695,536
5
44,695,543
6
44,695,547
4
1
145,001,361
PLEC
Gene body;
Transcribed
2
145,001,378
3
145,001,384
4
145,001,406
5
cg19405177
145,001,428
6
145,001,444
7
145,001,464
8
145,001,485
9
cg14598846
145,008,909
10
145,008,918
11
145,008,927
12
145,008,930
5
1
cg12031962
58,353,849
ALDH1A2
Intronic; Repressed
2
58,353,861
6
1
cg14752227
34,000,481
GDF5
Intronic
2
34,000,519
7
1
cg00065302
30,366,250
RWDD2B
Intergenic
2
cg05468028
30,391,383
Intronic; Promoter
-1.1
-0.51
3
30,391,385
4
cg18001427
30,391,784
Intergenic; Promoter
5
cg20220242
30,392,188
6
cg16140273
30,455,616
Intronic; Transcribed
The shaded cells indicate an intersection of the CpG position and open chromatin regions. Differentially accessible regions (DAR) in hip
(H) and knee (K) are also shaded and the mean log2 fold change in peak height is summarised. FH, foetal hip; FK, foetal knee; AH, aged
hip; AK, aged knee.
Table 3. Summary statistics of the human samples used for molecular genetic analyses
Tissue
Number of samples
Mean developmental age (days)
Age (years)
Sex
Foetal limb
19
56.3 (32-84)
n/a
M, 42.1%; F, 57.9%
Foetal cartilage
75
85.5 (56-119)
n/a
M, 52.0%; F, 48.0%
Adult cartilage
292
n/a
66.9 (25-91)
M, 39.8%; F, 60.2%
4
cg180014
27
30,391,784
0.010
Interg
enic;
Promo
ter
5
cg202202
42
30,392,188
2.40x10-6
6
cg161402
73
30,455,616
0.010
Introni
c;
Transc
ribed
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
44
Ages are expressed as the mean, with the range in parentheses. M, male; F, female.
Abbreviations
AC; articular cartilage
adj; adjusted
AEI; allelic expression imbalance
AH; aged hip
AK; aged knee
ATAC-sequencing; assay for transposase-accessible chromatin with sequencing
CpG; cytosine-guanine dinucleotide
CRE; cis-regulatory element
DAR; differentially accessible regions
DMOAD; disease modifying OA drug
DMS; differentially methylated site
DNAm; DNA methylation
E56; embryonic stage 56
E86; embryonic stage 86
eQTL; expression quantitative trait locus
EUR; European ancestry cohorts
FC; foetal cartilage
FDR; false discovery rate
FH; foetal hip
FK; foetal knee
FL; foetal limb
GE; genotypic effect
GEO; Gene Expression Omnibus
GO; gene ontology
GWAS; genome-wide association study
h; hour
HDBR; Human Developmental Biology Resource
HTA; Human Tissue Authority
kb; kilobase
LD; linkage disequilibrium
mg; milligram
mQTL; methylation quantitative trait locus
meQTL; methylation-expression quantitative trait locus
MRC; Medical Research Council
MSC; mesenchymal stem cell
ng; nanogram
NHS; National Health Service
OA; osteoarthritis
PCR; polymerase chain reaction
PRS; polygenic risk score
QC; quality control
qPCR; quantitative PCR
REC; research ethics committee
RIN; RNA integrity number
RT-qPCR; reverse transcription quantitative PCR
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
UNCORRECTED MANUSCRIPT
45
SNV; single nucleotide variant
TF; transcription factor
TSS; transcription start site
l; microliter
UTR; untranslated region
Downloaded from https://academic.oup.com/hmg/advance-article/doi/10.1093/hmg/ddac251/6754371 by guest on 10 October 2022
