Osteoporosis- and obesity-risk interrelationships: an epigenetic analysis of GWAS-derived SNPs at the developmental gene TBX15

Figures & data

Figure 1. Obesity-trait or osteoporosis-related GWAS Tier-1 SNPs overlap tissue-specific regulatory chromatin in TBX15. (a) The two TBX15 RefSeq gene isoforms, a truncated ENSEMBL isoform, and the upstream noncoding RNA gene (chr1:119,412,120–119,554,007, hg19); the broken arrow at TBX15 shows the main TSS. (b) The location of two nonsynonymous coding SNPs (double arrows) and eight Tier-1 (best regulatory candidates) SNPs; ‘3,’ a cluster of three Tier-1 SNPs. (c) Roadmap-derived chromatin state segmentation of strong (str), weak (wk), or bivalent (biv; poised) promoter or enhancer chromatin or repressed (repr) chromatin; SAT, subcutaneous adipose tissue; ostb, osteoblasts; Sk muscle, skeletal muscle; PBMC, peripheral blood mononuclear cells; fib, fibroblasts; myob, myoblasts; myot, myotubes; LCL, lymphoblastoid cell line; HMEC, mammary epithelial cells; NHEK, embryonic kidney cells; HUVEC, umbilical vein endothelial cells; ESC, embryonic stem cells; dotted horizontal lines, super-enhancers. (d) ENCODE strand-specific total RNA-seq. (e) RNA-seq TPM (log₁₀) from GTEx; the median TPM is given for some of the samples; VAT, visceral adipose tissue. All tracks were visualized in the UCSC Genome Browser and are aligned in this figure and Figures 2–5, and purple labels denote TBX15-expressing samples; grey, non-expressing samples; brown labels, samples expressing predominantly the truncated TBX15 isoform.

Figure 2. Many obesity-trait or eBMD risk-associated SNPs, EnhPro SNPs, and cis-eQTLs are found in the gene neighbourhood of TBX15. (a) The gene neighbourhood of TBX15 (chr1:119,134,901–119,884,900). (b) Obesity-trait or eBMD GWAS SNPs are designated as EnhPro SNPs if they overlap enhancer or promoter chromatin preferentially in SAT or ostb. (c) All index SNPs, proxy SNPs (r² ≥ 0.8, EUR) derived from them, and imputed SNPs in this region from obesity-trait or eBMD GWAS. (d) eQTLs for TBX15 or WARS2 in SAT or VAT. (e) Placental mammalian conserved elements (from phastCons track at the UCSC Genome Browser). (f) Chromatin state segmentation as in Figure 1. (g) CTCF binding determined by Roadmap ChIP-seq. Panels B-D show custom tracks derived from Supplementary Tables 4–6.

Table 1. Expression of TBX15 is significantly associated with that of the immediately upstream ncRNA gene, LOC105378933, but not with the further upstream WARS2.^a

Pairwise gene comparisons	Corr. coeff. (rs)	P-value
TBX15~ WARS2	0.13	0.46
TBX15~ LOC105378933^b	0.39	0.02
TBX15~ LOC101929147	0.08	0.63
TBX15~ WARS2-IT1	0.36	0.03
WARS2~ LOC105378933	0.32	0.06
WARS2~ LOC101929147	0.71	0.000001
WARS2~ WARS2-IT1	0.72	0.000001
LOC105378933~ LOC101929147	0.34	0.041

^aUsing GTEx RNA-seq data, the correlation coefficients and p-values were determined for Spearman’s rank-order correlation for the genes shown in Figure 2

^bLOC105378933 is also known as RP4-712E4.1 and LOC101929147 as RP11-418J17.1.

Table 2. Eight best candidates for transcription-regulatory obesity- or eBMD-causal SNPs (Tier-1 SNPs) in the TBX15 neighbourhood.

Tier-1 SNP for obesity (ob) or BMD assocn.^b	Index/proxy or imputed SNP	P-value for imputed assocn with WHR or eBMD	SNP’s region within TBX15	SNP’s distance to the main TBX15 TSS (kb)	Alleles (Ref/Alt)	Minor allele frequency	Trait-increasing allele (frequency)^c	Chromatin state at the SNP in SAT and osteoblasts^d	Predicted TF binding^e		Type of DHS at the SNP in SAT/adipocytes and ostb^d	LD thinned^f SNP group (r^2 > 0.2) for obesity	LD thinned^f SNP group (r^2 > 0.2) for BMD
									Ref allele	Alt allele
rs10923711	Imputed (ob)	6.5E-9 (WHR)	Intron 1	27.6	T/C	0.20	Alt (0.20)	Str enh	ETS1, ETV5 ELK1	None	Medium	IV	NA
rs984222	Index/proxy (ob) Imputed (ob & BMD)	3.8E-83 (WHR) 7.1E-18 (BMD)	Intron 1	26.6	C/G	0.39	Alt (0.61)	Str enh	MZF1, ZNF281	None	Small	I	I
rs9659323	Index/proxy & imputed (ob & BMD)	1.4E-48 (WHR) 1.5E-33 (BMD)	Intron 1	26.1	A/G	0.41	Alt (0.41)	Str enh	AP-1 family, MAF/NFE2	None	Small	I	I
rs1409157	Index/proxy (ob) Imputed (ob & BMD)	3.7E-71 (WHR) 1.2E-17 (BMD)	Intron 1	26.0	G/T	0.39	Alt (0.61)	Str enh	GATA2 (SAT), GATA6 (ostb)	None	Shoulder	I	I
rs61806235	Imputed (BMD)	5.6E-19 (BMD)	Intron 1	14.0	A/G	0.10	Alt (0.10)	Wk enh	None	AHR	Large	NA	II
rs12742627	Imputed (ob & BMD)	4.2E-11 (WHR) 2.4E-10 (BMD)	Intron 1	4.7	G/A	0.06	Ref (0.94)	Str pro	TFDP1, E2F1, E2F4, E2F6 (UniBind)	None	Medium	V	III
rs961470	Imputed (BMD)	1.0E-10 (BMD)	Intron 1	0.3	A/G	0.06	Ref (0.94)	Str pro &/or enh	None	NFIC, SMAD3	Medium	NA	III
rs1106529	Index/proxy (ob) Imputed (ob & BMD)	1.2E-80 (WHR) 3.4E-17 (BMD)	TSS −1 kb	−1.0	G/A	0.27	Alt (0.73)	Str pro	None	FOXO1, FOXC2 SOX4	Small	I	I

^aRegulatory SNPs in the neighbourhood of TBX15 were imputed SNPs for WHR_adjBMI from Pulit et al. (37) or for eBMD from Morris et al. (26) GWAS or were index plus proxy SNPs from BMI, WHR_adjBMI and WC_adjBMI or eBMD (see Supplementary Table 1). Not included in this table are two exonic missense variants with the respective trait-increasing alleles rs10494217-T (obesity and eBMD GWAS; Alt allele) and rs61730011-A (obesity GWAS; Ref allele); all p-value and LD data are for the EUR population (75); NA, not applicable

^bTier-1 SNPs are associated with allele-specific predictions of TF binding sites and overlap enhancer (enh) or promoter (pro) chromatin preferentially in subcutaneous adipose tissue (SAT)/adipocytes or osteoblasts (ostb); rs984222 and rs1409157 are in perfect LD, and rs12742627 and rs961470 are in almost perfect LD (r² = 0.98)

^cThe eBMD increasing allele and the obesity-trait increasing allele was the same for both types of traits for alleles associated with obesity as well as eBMD

^dChromatin state and size of DNaseI hypersensitive sites (DHS) from Roadmap for SAT/adipocytes or ostb; str, stong; wk, weak

^ePredictions of allele-specific TFBS by hand curation of data from four programs (see Supplementary Table 7) and assessment of TF expression in the target tissue

^fLD thinning by LDLink SNPClip program (53); see Figure 6.

Figure 6. The genetic architecture of the Tier-1 SNPs at TBX15 suggests that there are multiple causal regulatory SNPs for obesity risk and for osteoporosis risk. (a) The SNPclip program at LDlink [53] was used to trim the Tier-1 SNPs into groups with r² > 0.2 (EUR); the SNP members of the resulting groups are shown in pie charts. (b) The haplotypes for trimmed Group I are shown with designations of trait-increasing (trait ↑) or trait-decreasing (trait ↓) alleles; the trait-increasing alleles from obesity GWAS are the risk alleles but for eBMD GWAS they are the non-risk alleles because of the osteoporosis-protective effects of high BMD.

Figure 3. The 5ʹ end of TBX15 contains all eight Tier-1 SNPs. (a) TBX15’s alternative first exons, part of intron 1 (chr1:119,484,631–119,535,630), and its main TSS (broken arrow). (b) WHR_adjBMI and eBMD imputed SNPs that are EnhPro SNPs. (c) The eight EnhPro SNPs that qualify as Tier-1 SNPs; ‘2’ indicates two adjacent Tier-1 SNPs. (d) Chromatin state segmentation as in previous figures; Panc isl, pancreatic islets; Astroc, cultured astrocytes; CpG islands, CpG-rich regions as defined at the UCSC Genome Browser. (e) DNaseI hypersensitivity, vertical viewing range 0–10. (f) H3K27ac signal; vertical viewing range 0–4. (g) Binding of miscellaneous TFs to this region in a variety of cell cultures but not including ostb or adipocytes (see Supplementary Table 8). Coloured tick marks indicate positions of Tier-1 SNPs.

Figure 5. Zoomed in view of four Tier-1 SNPs in the middle of TBX15 intron 1. (a) The common SNPs in this subregion of intron 1 of TBX15 (chr1:119,500,183–119,508,182). (b-e) EnhPro SNPs and the subset of EnhPro SNPs that are designated Tier-1 SNPs, chromatin state segmentation, DNaseI hypersensitivity, and histone modifications as in Figure 4 except that the CpG island track is not shown due to there being no CpG islands in this region. In panel B, lollipops denote the two TBX15 eQTLs for SAT.

Figure 4. Zoomed in view of three Tier-1 SNPs near the main TSS of TBX15. (a) The common SNPs (version 151) at the 5ʹ end of TBX15 (chr1:119,523,915–119,533,914). (b) The eBMD-and-WHR_adjBMI or eBMD-only Tier-1 SNPs are shown in red or purple, respectively; EnhPro SNPs that did not meet all the criteria for Tier-1 SNPs are denoted by short grey bars. (c), (d), and (e) as in Figure 3 except that the vertical viewing range for DNaseI hypersensitivity was 0–7 and H3K4me3 and H3K4me1 are shown in addition to H3K27ac. Highlighting denotes the positions of Tier-1 SNPs.

Machiela MJ, Chanock SJ. LDlink: a web-based application for exploring population-specific haplotype structure and linking correlated alleles of possible functional variants. Bioinformatics. 2015 Nov 1;31(21):3555–3557. PubMed PMID: 26139635; PubMed Central PMCID: PMCPMC4626747.

 PubMed Web of Science ®Google Scholar