Reply

To the Editor:

Staley et al point out that, in our recent article, ¹ the strength of evidence for the causal effect of sclerostin lowering on increased myocardial infarction (MI) risk we reported was relatively weak and below the Bonferroni adjusted evidence threshold, raising the possibility of a spurious finding. However, absence of evidence does not imply evidence of absence. Contrary to their assertion, triangulation of evidence from other sources points to a similar conclusion. In terms of experimental data, up‐regulation of sclerostin inhibited the development of atherosclerosis in a mouse model, ² and SOST knockout mice developed more extensive vascular calcification. ³ Furthermore, two previous randomized controlled trials (RCTs) identified an increased risk of cardiovascular disease (CVD) events following romosozumab, ⁴ , ⁵ and a meta‐analysis of five trials found evidence of increased risk of a composite CVD outcome. ⁶ However, such trials suggest that any increased MI risk from sclerostin lowering is relatively modest and substantially smaller than the gain in bone mineral density (BMD) underpinning its therapeutic action. This has implications for the statistical power of Mendelian randomization (MR) analyses, which is further compounded by the fact that common SOST variants only have modest effects on circulating sclerostin (no genome‐wide significant signals in the SOST gene encoding sclerostin were identified in our initial genome‐wide association study (GWAS) based on approximately 10,000 individuals ⁷ ).

Data from SomaLogic and O‐link proteomic platforms, which both include sclerostin, have recently been made publicly available from DeCode ⁸ and UK Biobank, ⁹ respectively, providing an opportunity to repeat our analyses based on 100,482 individuals, giving considerably greater statistical power (N_Zheng = 31,491; N_{UK_Biobank} = 33,628; and N_Decode = 35,361). A large‐scale analysis combining these additional data with those in our recent article ¹ identified two conditionally independent signals in the cis SOST region. The top hit rs66838809 was the same single‐nucleotide polymorphism (SNP) as we reported in our largest published GWAS, ¹ and this showed a stronger signal in the updated meta‐analysis (Table 1). The secondary hit rs1107748 also showed robust association with sclerostin (P = 4.9 × 10⁻²⁰; Table 1) and is in perfect linkage disequilibrium (LD) with one of five cis‐protein quantitative trait loci (pQTL) instruments (rs1107747) we used in our paper. ¹ Using these two SNPs as instruments, we conducted an MR analysis of protein levels of sclerostin on 15 atherosclerosis‐related outcomes by applying generalized inverse variance weighted, which accounts for weak LD between the two instruments (LD r² = 0.13 in the 1000 Genome Europeans). We found robust MR evidence that lower sclerostin levels increase the risk of coronary artery disease ¹⁰ (CAD; odds ratio 1.50, 95% confidence interval [CI] 1.15−1.93, P = 0.002) (Figure 1A). The CAD result withstands the Bonferroni corrected threshold (P = 0.0052) after accounting for correlations among the 15 outcomes. CAD is closely related to MI but more inclusive because it also includes individuals undergoing related procedures such as angioplasty. In addition, marginal predicted effects of lower sclerostin levels on type 2 diabetes mellitus ¹¹ and high‐density lipoprotein cholesterol ¹² were observed (Figure 1B). The point estimate for an effect on myocardial infarction ¹³ was very similar to that previously reported ¹ ; however, in this instance, the lower 95% CI included unity (Figure 1A).

Table 1.

Genetic signals of the two conditional independent SNPs associated with plasma protein levels of sclerostin in the cis‐SOST region^*

Phenotype	SNP	Effect allele	Other allele	Effect allele freq	Beta	SE	P value	N	Note
Sclerostin	rs66838809	A	G	0.087	−0.080	0.008	1.7 × 10−24	100,482	first signal
Sclerostin	rs1107748	T	C	0.615	0.041	0.004	4.9 × 10−20	100,482	second signal
Sclerostin	rs9910625 a	T	C	0.352	−0.042	0.005	7.0 × 10−20	100,482	second LD proxy

^*freq, frequency; LD, linkage disequilibrium; SNP, single‐nucleotide polymorphism.

^a rs9910625 is the LD proxy of rs1107748 with LD r² = 0.9 with rs1107748.

Figure 1.

Effect of lower sclerostin on 15 atherosclerosis related outcomes. (A) Effect of lower sclerostin on disease outcomes; (B) effect of lower sclerostin on continuous outcomes; (C) single SNP effect of lower sclerostin on coronary artery disease and type 2 diabetes. *rs1107748 was proxied by rs9910625 (LD r² = 0.9) due to missing SNPs in the CAD GWAS. (D) Locus comparison plot of sclerostin versus CAD for the top hit rs66838809 in the SOST region. (E) Locus comparison plot of sclerostin versus CAD for the secondary hit rs9910625 in the SOST region. The three subplots for panels D and E are −log₁₀(P) value of sclerostin versus the outcome, as well as the regional plot of sclerostin and outcomes in the cis‐SOST region. CAD, coronary artery disease; CI, confidence interval; GWAS, genome‐wide association study; HDL, high‐density lipoprotein; IVW, inverse variance weighted; LD, linkage disequilibrium; LDL, low‐density lipoprotein; SNP, single‐nucleotide polymorphism.

In terms of why a robust effect of sclerostin on CAD was now evident, the new secondary pQTL signal, rs1107748, showed a higher point estimate for an effect on CAD risk compared with the top hit (Figure 1C). The locus plot for CAD associations in the SOST region was also consistent with two distinct cis‐pQTL signals for sclerostin, with the secondary pQTL signal associated with a higher CAD risk estimate compared with the top pQTL (Figure 1D and 1E). The evidence provided by colocalization remains weak, however, with SharePro ¹⁴ (an approach that can be applied when a region contains multiple signals) yielding a 13.2% probability that the second sclerostin signal co‐localizes with CAD. Power for this test in the context of the modest effect size compared to that on BMD is low, however.

In summary, these additional analyses identified multiple cis‐pQTL signals in the SOST region, which together show a clear albeit weak effect of sclerostin lowering on CAD risk. Taken together with evidence from other domains, these findings run contrary to the assertion by Staley et al that this region has “no” association with any other phenotypes beyond BMD. In terms of the clinical significance of our findings, more data are required as to the propensity of sclerostin inhibition to increase cardiovascular risk, including findings from RCTs with other sclerostin inhibitors.

Supported by the National Key Research and Development Program of China (grant 2022YFC2505203). Dr. Smith works within the Medical Research Council (MRC) Integrative Epidemiology Unit at the University of Bristol, which is supported by the MRC (MC_UU_00011/1).