Hyperthyroidism and bone mineral density: Dissecting the causal association with Mendelian randomization analysis

Untreated hyperthyroidism is associated with accelerated bone turnover, low bone mineral density (BMD) and increased susceptibility to fragility fractures. Although treatment appears to improve or even reverse some of these adverse skeletal effects, there is limited guidance on routine BMD assessment in hyperthyroid patients following treatment. By using Mendelian randomization (MR) analysis, we aimed to assess the causal association of hyperthyroid thyroid states with BMD and fractures using the UK Biobank.


| INTRODUC TI ON
Hyperthyroidism, characterized by excess thyroid hormone synthesis and secretion from the thyroid gland, affects approximately 1 in 100 individuals. [1][2][3] Graves' disease and multinodular goitre are the most common causes of hyperthyroidism in young/middle-aged and older individuals, respectively. 2,3 Untreated overt hyperthyroidism has well-documented skeletal consequences including low bone mineral density (BMD), secondary osteoporosis and increased fragility fracture risk, which result from high and more frequent initiation of bone turnover (i.e. simultaneous increases in bone formation and resorption). [4][5][6][7][8][9][10][11] Impaired bone health has also been shown among patients with subclinical hyperthyroidism (TSH concentrations below the normal reference range and free T3 and T4 levels within the normal reference range) [12][13][14] and euthyroid individuals with relatively low TSH and relatively high free T4 levels, albeit within the reference range. 15,16 The treatment options for hyperthyroidism commonly include antithyroid drugs, radioactive iodine therapy and near-total or total thyroidectomy. 1,17 Although some studies have shown increases in bone formation, improvements in BMD and reduction in fracture risk, 5,[17][18][19][20] others suggest persistent, low BMD even after treatment, especially in populations at risk of osteoporosis such as postmenopausal women. 17,21,22 Current guidelines for the diagnosis and management of osteoporosis (National Institute for Clinical Excellence NICE) clinical guideline CG146, ESCEO and IOF, 23 Scientific Advisory council of osteoporosis Canada, 24 27 This law postulates that alleles segregate randomly from parents to offspring 28 and thus, offspring genotypes are randomly distributed in the population and are unlikely to be associated with confounders. An assessment of the association of excess genetic risk for hyperthyroid states such as Graves' with BMD and fractures will help in understanding if these states have a direct causal role in lowering of the BMD and increasing risk of fractures. If such an association exists, it will support careful monitoring of BMD and periodic basement of fracture susceptibility in these patients. However, if there is no such association, it would support the current NICE guidelines.
The aim of the present investigation was to assess the effects of genetic susceptibility to hyperthyroidism on BMD and history of fragility fractures using the UK Biobank data and data from the GEnetic Factors for OSteoporosis (GEFOS) Consortium. 29

| Power calculation
The power calculation was carried out with publicly available power calculator for Mendelian randomization study available at https://shiny.cnsge nomics.com/mRnd/. The power calculation was performed for a definitive outcome, all-cause fractures in the UK Biobank study. A sample size of 473 818 with around 10% prevalence of all-cause fractures (at the level of significance 0.05 and with 1% variance explained for the association between allele score and exposure variable) gives more than 50% power to detect an odds ratio of 1.1 and more than 90% power to detect an odds ratio of 1.2.

| Replication cohort
We used publicly available data from GWAS for BMD from the GEFOS consortium, which identified novel loci for BMD at the femoral neck, lumbar spine and forearm; sites of common osteoporotic fractures. Forearm BMD data were not used in the present study because of the relatively low number of participants (n = 8143). We did not include the femoral neck BMD as the discordance rate between the femoral neck, and lumbar spine BMD was as high as 33%, and in older patients with osteoporosis assessment of lumbar spine, BMD is preferred. 29 The Mendelian randomization (MR) analysis conducted on lumbar spine BMD comprised 28,498 participants from eight cohorts of European ancestry. The mean age of the participants in the available cohorts ranged from 17.7 to 80.2 years, and 34% of the participants in the meta-analysis were men.

| SNP Instrument selection
We used the NGHRI-EBI catalogue of GWAS studies to identify uncorrelated SNPs associated with Graves' disease, hyperthyroidism and autoimmune thyroiditis with P-value < 10 −5 . As of September 2018, the Catalogue contains 5687 GWAS comprising 71 673 variant-trait associations from 3567 publications. 31 We identified 19 independent loci for Graves' disease, 32-35 9 independent loci for hyperthyroidism 36 and 11 independent loci for autoimmune thyroid disease 32 (Table S1). We used genetic instruments for Graves' disease, patients with hyperthyroidism due to other causes, and patients with autoimmune thyroiditis (including Hashimoto's thyroiditis which in some cases is characterized by transient hyperthyroidism or hashitoxicosis), to cover most causes of hyperthyroidism.
Risk score-based MR analysis to assess the effect of genetic predisposition to Graves' disease, hyperthyroidism and hypothyroidism on BMD we constructed a beta-weighted (wGRS) genetic risk score using 19 independent SNPs for Graves' disease, 9 independent SNPs for hyperthyroidism and 11 independent SNPs for autoimmune thyroiditis. The wGRS was calculated by multiplying each known β-coefficient (from NHGRI catalogue) for each phenotype by the number of corresponding risk alleles using Plink. The weighted genetic risk score wGRS was adjusted for age, gender, BMI and top five principal components. Sensitivity analyses was performed in the self-reported Caucasian population in the UKbiobank database. In order to correct for multiple testing in the wGRS analysis, Bonferroni correction was applied a P-value of (0.05/12) .004 was considered statistically significant.

| Replication and sensitivity analyses for Mendelian randomization
To explore potential pleiotropic effects, we carried out three sensitivity meta-analyses: simple and weighted median and Mendelian randomization-Egger regression methods using the r Program Mendelian randomization. 27,37 Simple and weighted median MR analysis provides estimations that are robust to the inclusion of up to 50% invalid instruments in a Mendelian randomization analysis. 38

| RE SULTS
The MR analysis was performed on 473 818 participants from the UK Biobank with 54% females, with a median age of 58.00

| Genetic risk score-based MR analysis for BMD and fractures
We performed a genetic risk score-based MR analysis for BMD and fractures in UK Biobank population. To account for multiple testing, a Bonferroni corrected P-value of .004 was considered statistically significant.
MR analysis using beta-weighted genetic risk score adjusted

| Sensitivity analysis in Caucasian population
We performed a sensitivity analysis with a genetic risk score-based MR in self-reported Caucasian population (n = 409 633) with GWAS data. MR analysis in the caucasian population using beta-weighted genetic risk score adjusted for age, gender and BMI and top 5 principle component showed no association (P < .004) of genetic risk for Graves' disease (Beta = −0.01 P-value = .02), autoimmune thyroiditis (Beta = −0.01 P-value = .34) and hyperthyroidism (Beta = −0.007 P-value = .34) with BMD. Also, the MR analysis using beta-weighted genetic risk score showed no association of genetic risk for Graves' disease (OR = 1.01, P-value = .6), autoimmune thyroiditis (OR = 1.02, P-value = .87) and hyperthyroidism (OR = 0.98, P-value = .31) with self-reported fractures in the last 5 years.  (Table 1).

| Replication analysis in GEFOS consortium using Mendelian randomization
A meta-analysis of the MR-Egger Estimates from UK Biobank and GEFOS consortium using fixed-effect models showed no association of genetic risk for Graves' disease, hyperthyroidism and autoimmune thyroid disease with bone mineral density (Table S5).

| D ISCUSS I ON
In this MR study, we demonstrated that excess genetic risk for Graves' disease, hyperthyroidism and autoimmune thyroiditis is not associated with low BMD or history of fractures in the UK

Biobank.
Several epidemiological studies in adults have shown an association between a hyperthyroid status and low BMD. 6,14-16 For example, a meta-analysis, which assessed alterations in BMD and F I G U R E 1 Figure 1A 1B and 1C shows the results of causal effects of hyperthyroid states on BMD estimated using each SNP separately in the UK Biobank data and the MR-Egger and Inverse-variance weighted MR estimates [Colour figure can be viewed at wileyonlinelibrary.com] fractures risk in patients with hyperthyroidism, demonstrated significant BMD reductions and increased fracture risk in untreated patients. 6 Subclinical hyperthyroidism, individuals with TSH at the lower end and free T4 at the upper end of the normal euthyroid range, have also been associated with unfavourable bone outcomes including more significant bone loss and fracture risk. [14][15][16] The results of these observational studies do not establish a causal association as they are constrained by confounders such as duration of hyperthyroidism, variations in treatment and response to treatment, and other comorbidities. Evidence from observational and interventional studies on the effects of treatment for hyperthyroidism on bone outcomes has yielded mixed results. Although some studies have shown full recovery of BMD, 5,17-20 some others suggest the only partial restoration of bone mass and lower BMD. 17,21,22 Although a causal association between thyroid states and BMD is more likely to be established with MR studies, these remain limited. In a recent study, van Vliet et al looked at 20 genetic variants that were previously identified for circulating TSH levels and found no evidence that a genetically determined circulating TSH concentration was associated with femoral neck or lumbar spine BMD. 39 They also found no association of variants in the TSHR gene and BMD. In contrast, two earlier genetic studies which explored the association between TSH and BMD found that the Asp727Glu polymorphism in the TSHR gene (rs1991517) was associated with higher BMD. 40,41 This finding was, however, no longer significant after adjustments for BMI, 40,41 and available studies have failed to identify any further common TSHR genetic variants in association with BMD.
To the best of our knowledge, this is the first MR study looking at the effect of genetic predisposition to treated Graves' hyperthyroidism and autoimmune thyroiditis on BMD. The MR estimates obtained using the MR-Egger, and weighted median analysis were consistent and do not support a causal association between treated hyperthyroid conditions and BMD. Further, we did not show an association between hyperthyroid states and fractures, with these findings extending our understanding on the importance of thyroid function in the skeleton in middle-aged and older adults. The lack of association between hyperthyroid states and skeletal fractures may partially reflect the lack of association of thyroid function with BMD. An alternative explanation may be that treated hyperthyroidism is not associated with other skeletal parameters including bone geometry, microstructure and material, or even extra-skeletal characteristics including physical performance or falls. 42,43 Our study is strengthened by the employment of the MR approach, which lessens systematic biases such as confounders and reverse causality, commonly affecting the results of conventional observational studies. 27 Another strength of our study is that we used data from large-scale genetic databases, which enabled us to explore the associations of hyperthyroidism, BMD and fracture risk in a precise way. Our study also has several limitations. UK Biobank does not have information about the treatment received for hyperthyroid states; hence we are unable to adjust for this in the analysis. We performed the replication analysis in the publicly available GEFOS consortium using the TA B L E 1 MR estimates from each method of the causal effect of the Graves' disease, hyperthyroidism and autoimmune thyroid disease on BMD

| CON CLUS ION
In summary, our study does not support a causal association between increased genetic risk for hyperthyroidism and the risk for low BMD. Our results support the current guidelines, which do not recommend routine long-term screening for osteoporosis following treatment for hyperthyroidism. However, further MR studies in diverse populations are needed to confirm these findings.

ACK N OWLED G EM ENT
This research has been conducted using the UK Biobank Resource under Application Number 44242.

CO N FLI C T O F I NTE R E S T
None of the authors have any disclosures related to the content of this manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in the UK Biobank data.