Association of Bone Mineral Density and Bone Turnover Markers with the Risk of Diabetes: Hong Kong Osteoporosis Study and Mendelian Randomization Open Access

ABSTRACT

Preclinical studies demonstrated that bone plays a central role in energy metabolism. However, how bone metabolism is related to the risk of diabetes in humans is unknown. We investigated the association of bone health (bone mineral density [BMD] and bone turnover markers) with incident type‐2 diabetes mellitus (T2DM) based on the Hong Kong Osteoporosis Study (HKOS). A total of 993 and 7160 participants from the HKOS were studied for the cross‐sectional and prospective analyses, respectively. The cross‐sectional study evaluated the association of BMD and bone biomarkers with fasting glucose and glycated hemoglobin (Hb_A1c) levels, whereas the prospective study examined the associations between BMD at study sites and the risk of T2DM by following subjects a median of 16.8 years. Body mass index (BMI) was adjusted in all full models. Mendelian randomization (MR) was conducted for causal inference. In the cross‐sectional analysis, lower levels of circulating bone turnover markers and higher BMD were significantly associated with increased fasting glucose and Hb_A1c levels. In the prospective analysis, higher BMD (0.1 g/cm²) at the femoral neck and total hip was associated with increased risk of T2DM with hazard ratios (HRs) of 1.10 (95% confidence interval [CI], 1.03 to 1.18) and 1.14 (95% CI, 1.08 to 1.21), respectively. The presence of osteoporosis was associated with a 30% reduction in risk of T2DM compared to those with normal BMD (HR = 0.70; 95% CI, 0.55 to 0.90). The MR results indicate a robust genetic causal association of estimated BMD (eBMD) with 2‐h glucose level after an oral glucose challenge test (estimate = 0.043; 95% CI, 0.007 to 0.079) and T2DM (odds ratio = 1.064; 95% CI, 1.036 to 1.093). Higher BMD and lower levels of circulating bone biomarkers were cross‐sectionally associated with poor glycemic control. Moreover, higher BMD was associated with a higher risk of incident T2DM and the association is probably causal. © 2023 The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR).

Introduction

There is a complex interplay between type 2 diabetes mellitus (T2DM) and bone health. It has been shown that T2DM can affect bone metabolism.^[1,] However, a paradox is usually observed among patients with T2DM—individuals tend to have a higher bone mineral density (BMD)^[2,3,] but also an increased risk of fragility fracture.^[4,] The increased risk of fragility fracture can be also attributed to the deteriorated bone microarchitecture,^[1] but the common higher BMD among T2DM patients remains unexplained.

The ability of bone to regulate whole‐body energy metabolism was shown before.^[5,] Osteocalcin, one of the osteoblast‐secreted proteins, was first demonstrated as a bone‐specific osteokine affecting whole‐body energy metabolism,^[6,] in which insulin signaling is also involved.^[7,] Since then, a number of association studies have been conducted to evaluate the relationship of osteocalcin with glycemic traits. Meta‐analyses have also shown that increased osteocalcin is associated with reduced fasting glucose and glycated hemoglobin (Hb_A1c).^[8,9,] Besides osteocalcin, bone‐specific lipocalin‐2 was also found able to affect energy metabolism.^[10,11] These animal studies all imply that bone metabolism may play an important role in the whole‐energy metabolism. However, the number of human studies on the relationship of BMD with the risk of diabetes is limited; most only focused on the impact of glucose metabolism on bone health instead of the other way around.

In this study, we aimed to evaluate the relationship of BMD and bone biomarkers with the risk of diabetes using the Hong Kong Osteoporosis Study (HKOS), via prospective and cross‐sectional cohort study design, respectively. We further conducted a Mendelian randomization (MR) study to determine if BMD plays a causal role in such a relationship.

Subjects and Methods

Data source

The HKOS

The HKOS is a prospective cohort study investigating the epidemiology of osteoporosis and related comorbidities. The baseline examination was conducted from 1995 to 2010, and 9449 participants were recruited. A detailed description of this cohort has been described before.^[12] In brief, baseline information including their anthropometric measurements, socioeconomic status, education, and medical history was obtained using structured questionnaires, and serum samples were collected for further analyses.

Since 2015, an in‐person follow‐up study (HKOS‐FU) was conducted with 1386 participants.^[12] In this follow‐up study, bone turnover markers and markers related to metabolic syndrome (e.g., fasting glucose, Hb_A1c, lipids, blood pressure, etc.) were also measured. After excluding patients with missing demographic information, exposure, outcome, or covariates measurements, 993 participants were included in the analysis.

For evaluating the temporal association of baseline BMD with incident diabetes, the HKOS baseline cohort (HKOS‐B) was linked to a population‐based electronic health record, The Clinical Data Analysis and Reporting System (CDARS). The CDARS is maintained by the Hospital Authority of Hong Kong, the statutory body responsible for managing all public hospitals and clinics in Hong Kong. CDARS is a centralized electronic health record database with data on demographics information, diagnosis, laboratory tests, procedures, medication prescriptions, and hospital admission and discharge information. Among 9449 participants, after excluding participants with a prior history of diabetes or prescription of anti‐diabetic medication, or missing records of demographics, baseline BMD, smoking and alcohol consumption, and physical activity status, 7160 participants were included in the final analysis.

Measurements of bone parameters in the HKOS

BMD was measured using a Hologic dual‐energy X‐ray absorptiometry (DXA) system (Hologic, Waltham, MA, USA) in the study. Serum N‐terminal of propeptides of type 1 collagen (P1NP), osteocalcin, and C‐terminal of telopeptides of type 1 collagen (CTx) were measured using a Cobas e411 analyzer (Roche Diagnostics, Mannheim, Germany). Serum N‐terminal of telopeptides of type 1 collagen (NTx) was measured using Osteomark assay (Abbott Laboratories, Abbott Park, IL, USA). All bone turnover markers were measured from the blood taken from participants fasting for at least 8 hours and in the morning between 9 a.m. and 11 a.m. to minimize the impact of nocturnal variation on the measurements.

Covariates in the HKOS

In the HKOS, major factors associated with both BMD and diabetes were adjusted; these included age, sex, body mass index (BMI), smoking, drinking and physical activity status, total cholesterols (in HKOS‐FU only), triglycerides (in HKOS‐FU only), and presence of diabetes (in HKOS‐FU only). Participants were asked about their smoking status of tobacco products (including cigarettes, pipes, or cigars). They were classified as current smokers, former smokers and nonsmokers. Based on their alcohol consumption, the participants were categorized as current drinkers, former drinkers, and non‐drinkers. Based on their physical activity level, participants were also categorized as active and inactive.^[13,] Serum lipids (including total cholesterol and triglycerides) measurements in the HKOS‐FU study were made using Ortho‐Vitros Fusion 5.1 (Johnson & Johnson, New Brunswick, NJ, USA).^[14]

Measurements of glycemic parameters and ascertainment of diabetes

Two glycemic traits were measured in the HKOS‐FU cohort, namely fasting glucose and Hb_A1c. Fasting glucose (fasting ≥8 h) and Hb_A1c were measured using Ortho‐Vitros Fusion 5.1 (Johnson & Johnson) and high‐performance liquid chromatography (HPLC) with Hemoglobin A1c Program Reorder Pack (Bio‐Rad D‐10; Bio‐Rad Laboratories, Hercules, CA, USA), respectively. T2DM status was ascertained through the linked CDARS records using the International Classification of Diseases, Ninth Revision(ICD‐9) code of 250.XX or prescription of anti‐diabetic medication as in our previous study,^[15,] which has been validated.^[16] Patients were followed from the date of baseline visit to the first diagnosis of diabetes, date of death, or study end (December 31, 2020), whichever came first.

Statistical analysis

Baseline characteristics are presented as mean (standard deviation [SD]) and frequency (%) for continuous and categorical variables, respectively.

In the cross‐sectional HKOS‐FU cohort study, the associations of bone turnover markers and BMD with glycemic parameters were assessed based on the coefficient estimates (β) of linear regression models. All units of the bone turnover markers and glycemic parameters were standardized, and the unit of BMD was 0.1 g/cm². Age and sex were adjusted in the simple model, while BMI, total cholesterol, triglyceride, drinking and drinking status, physical activity, and the presence of diabetes were further adjusted in the full model. In the prospective HKOS‐B cohort study, Cox proportional hazards regression models were used to estimate the hazard ratio (HR) and 95% confidence intervals (CIs) of T2DM per 0.1 g/cm² increase in the baseline BMD. Age and sex were adjusted in the simple model, whereas BMI, smoking, drinking, and physical activity status were further adjusted in the full model. In addition, because the participants of HKOS‐FU and HKOS‐B are overlapping, we did a sensitivity analysis to evaluate the relationship between BMD and the risk of T2DM after excluding those in both HKOS‐FU and HKOS‐B. As the most recent update in the diagnosis criteria for diabetes was on April 21, 2006, we did a sensitivity analysis after excluding the cases identified before the criteria change. Moreover, we conducted the association analysis further adjusting for serum calcium (mmol/L), serum vitamin D (nmol/L), and daily intake of calcium tablets (mg) on a subset of participants with the information available.

MR analysis

We used the MR approach to estimate the genetic causal relationship of BMD on diabetes and 2‐h glucose level after an oral glucose challenge test (2hGlu) because 2hGlu is considered the gold standard in diagnosing diabetes. MR is a method that uses genetic variants as instrumental variables (IVs) to estimate causal inference under three assumptions: the IVs are robustly associated with the exposure, there are no confounders of the IVs and the outcome, and the IVs affect the outcome only through the exposure. Estimated BMD (eBMD) was used as the exposure in this study. eBMD is the BMD estimated at the heel using quantitative ultrasound and was shown to have a phenotypic and genetic correlation with the DXA‐measured BMD. Furthermore, the genomewide association study (GWAS) of eBMD^[17] has a much larger sample size compared to DXA‐measured BMD; therefore, using eBMD as the exposure provided ample power in the current study. Genetic variants associated with eBMD were utilized as IVs to infer causality with 2hGlu and diabetes.

SNPs showing a genomewide significant association (p < 5 × 10⁻⁸) to eBMD were selected from the largest GWAS on eBMD, which included 426,824 participants of European ancestry from the UK Biobank.^[17,] The single‐nucleotide polymorphisms (SNPs) were then clumped using PLINK^[18,] v1.90b6.21 with an r² threshold of 0.001 and a clumping window of 10 Mb, resulting in 609 independent SNPs to be used as IVs for the analysis. Proxies were used to replace instruments that were not available from the outcome datasets or were palindromic with allele frequency >0.42. Proxies were selected if they were in high linkage disequilibrium (LD) with the original instruments (r² ≥ 0.8) based on the LD pattern in the European population of the 1000 Genomes Project. The reference dataset used to calculate LD was obtained from the IEU OpenGWAS API^[19,] which contains only biallelic autosomal variants with a minor allele frequency (MAF) >0.01. Only proxies with a genomewide significant association with eBMD (p < 5 × 10⁻⁸) were kept. BMI, fat‐related traits, and osteocalcin are plausible confounders of the relationship between higher BMD and increased risk of diabetes (Table S1). To avoid violation of the independence assumption, SNPs that had a significant association, or were in high LD (r² ≥ 0.8) with an SNP showing significant association, (p < 5 × 10⁻⁸) with BMI, waist circumference, waist‐hip‐ratio (WHR), low‐density lipoprotein‐cholesterol (LDL‐C), coronary heart disease, and/or osteocalcin, as revealed by the web‐interface PhenoScanner,^[20,] were excluded from subsequent analyses. The outcomes of interest of the MR study were 2hGlu and the incidence of T2DM. For 2hGlu, summary statistics were obtained from the European‐specific GWAS meta‐analysis conducted by the Meta‐Analyses of Glucose and Insulin‐related traits Consortium (MAGIC consortium)^[21,] comprising 63,396 individuals after excluding those with type 1 diabetes, T2DM reported use of diabetes‐relevant medication, or had a fasting glucose ≥7 mmol/L, 2hGlu ≥ 11.1 mmol/L, or Hb_A1c ≥ 6.5%. In the GWAS meta‐analysis, 2hGlu measures were obtained 120 min after a glucose challenge in an oral glucose tolerance test. For T2DM, the summary‐level data were obtained from a GWAS meta‐analysis^[22] involving a total of 74,124 T2DM cases and 824,006 controls of European ancestry. The diagnostic criteria for T2DM varied between the meta‐analyzed studies, including ascertainment from diagnostic codes in electronic health records, prescription of anti‐diabetic medicine, self‐reporting or levels of fasting glucose, 2‐h plasma glucose, and Hb_A1c. All contributing studies were conducted with informed consent from all participants and approved by the relevant ethical committees.

Radial MR analysis with default settings (i.e., alpha level of 0.05 and modified second‐order weights) was also performed to remove outliers that could lead to a violation of MR assumptions. The inverse‐variance weighted (IVW) method was adopted in the primary analysis of our MR study. The weighted median, MR‐Egger regression, and Contamination Mixture (ConMix) methods were also performed as sensitivity analyses. Potential pleiotropy was examined using the intercept test in the MR‐Egger method and heterogeneity was evaluated by Cochran's Q test. The effect measures were the standardized beta for 2hGlu, whereas for T2DM, the effect measure was the odds ratio (OR), which should be interpreted as the relative chance of outcome risk per SD increase in eBMD.

The power and the F‐statistics for each MR analysis (Table S2) were calculated using two online calculators, respectively (Power: https://sb452.shinyapps.io/power/^[23,] and F‐statistics: http://sb452.shinyapps.io/overlap^[24,]). Because the GWAS for both eBMD and T2DM included UK Biobank participants, the bias and type 1 error rate incurred by the potential overlapping sample were also calculated for the T2DM MR analysis using the latter online tool.^[24] Assuming the bias of the observational estimate was 1.05 (OR of T2DM per SD increase in eBMD), the bias and Type I error rate arising from the overlapping sample were estimated to be 0.002 and 0.06, respectively. Because the GWAS for eBMD and 2hGlu were from different samples, there was no overlap. However, both were conducted with participants from the European population.

All statistical analyses were conducted using R version 4.2.0. (R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/). Packages TwoSampleMR,^[25,] MendelianRandomization, and RadialMR^[26] were used for MR analyses. All statistical tests were two‐sided and p value <0.05 was considered statistically significant.

All participants from the HKOS cohort gave informed consent. The study protocol of the HKOS study was approved by the Institutional Review Board of the University of Hong Kong and the Hospital Authority Hong Kong West Cluster Hospitals. (IRB numbers: UW 03–140 and UW 15–236).

Results

Baseline characteristics of the HKOS‐FU (N = 993) and HKOS‐B (N = 7,160) are shown in Table 1. The majority of the cohort participants are female, and the average age is over 50 years.

We first evaluated the cross‐sectional relationship of the markers of bone metabolism with glycemic traits in the HKOS‐FU cohort. In the simple model adjusted for age and sex, both bone formation and resorption markers showed a significant inverse association with fasting glucose and Hb_A1c, except for the association of NTx with Hb_A1c (Table 2). In the fully adjusted model, similar inverse associations were observed. CTx (β_CTx = −0.049; 95% CI, −0.093 to −0.005), NTx (β_NTx = −0.057; 95% CI, −0.100 to −0.013), and OC (β_OC = −0.052; 95% CI, −0.097 to −0.008) are significantly associated with the levels of fasting glucose, whereas the inverse association with P1NP is marginally significant (β_P1NP = −0.042; 95% CI, −0.085 to 0.002). For bone turnover markers and Hb_A1c, all coefficient estimates for the associations were inverse but insignificant (Table 2). On the other hand, BMD at the total hip, femoral neck, and lumbar spine showed a significant positive association with fasting glucose (β_{total hip} = 0.057, 95% CI: 0.018 to 0.096; β_{femoral neck} = 0.056, 95% CI: 0.018 to 0.095; β_{lumbar spine} = 0.030, 95% CI: 0.001 to 0.059) and Hb_A1c (β_{total hip} = 0.048, 95% CI: 0.020 to 0.076; β_{femoral neck} = 0.051, 95% CI 0.023 to 0.078; β_{lumbar spine} = 0.026, 95% CI: 0.006 to 0.047), with smaller estimates being observed for BMD at the lumbar spine (Table 3).

Then, we evaluated whether baseline BMD could predict the risk of incident diabetes in the prospective HKOS‐B cohort. With a median follow‐up of 16.8 years and a total of 111,626.7 person‐years, 12.0% (n = 857) of the subjects developed diabetes resulting in an incidence rate of 7.68 per 1000 person‐years. In the Cox regression model, BMD at all sites measured were associated with an increased risk of diabetes in the simple model adjusted for age and sex only. In the full model after further adjusted for BMI, smoking, drinking, and physical activity status, the association of BMD at the femoral neck and total hip remained significantly associated with a higher risk of T2DM (Table 4; total hip BMD per 0.1 g/cm²: HR = 1.143, 95% CI 1.077 to 1.214; femoral neck BMD per 0.1 g/cm²: HR = 1.103, 95% CI 1.034 to 1.178). The presence of osteoporosis at any skeletal site was associated with a 30% reduction in risk of diabetes when compared to those participants who had normal BMD at all sites (Table 5; HR = 0.704, 95% CI: 0.552 to 0.897). After excluding the overlapping participants in both HKOS‐FU and HKOS‐B (n = 6,201), quite similar results were observed (Table S3). There was no significant interaction with sex and age detected (data not shown). Thus, no subgroup analyses were performed. Similar results were observed if excluding the diagnosis of T2DM based on the old criteria (Table S4) and if further adjusting for serum calcium and vitamin D levels and daily intake of calcium tablets (Table S5).

Given that the observational study design is known to be subject to unmeasured confounding, MR was used to evaluate the potential causality of BMD with T2DM (Table 6). A significant genetic robust causal relationship of genetically determined eBMD with T2DM was observed in the IVW analysis (OR_IVW = 1.064, 95% CI: 1.036 to 1.093), similar significant associations were observed in the weighted median (OR_MW = 1.085, 95% CI: 1.039 to 1.134), MR‐Egger regression (OR_Egger = 1.063, 95% CI: 1.012 to 1.116), and ConMix (OR_ConMix = 1.062, 95% CI: 1.020 to 1.128) analyses per SD increase in eBMD. The one SD increase in eBMD was also associated with an increased risk of 2hGlu in the IVW analysis (standardized β_IVW = 0.043, 95% CI: 0.007 to 0.079) and the ConMix analysis (standardized β_ConMix = 0.060, 95% CI: 0.020 to 0.100). The IVs used in the analysis and their associations with the exposure and outcome are shown in Supplementary Data D1 and D2 for T2DM and 2hGlu, respectively. In all analyses, the F‐statistic was greater than 10, indicating a low risk of weak instrument bias, no pleiotropy was detected in the MR‐Egger intercept test, and Cochran's Q test did not detect heterogeneity (Table 6).

Discussion

The relationship of bone health on T2DM was barely investigated in previous cohort studies. In this study, we evaluated the association of bone biomarkers and BMD with important glycemic parameters and the risk of T2DM incidence. We observed that lower circulating levels of bone formation and resorption markers (and hence lower bone turnover) and higher BMD were cross‐sectionally associated with higher fasting glucose and Hb_A1c. In addition, higher BMD is associated with poor glycemic control and a higher risk of T2DM. MR results further demonstrated that a higher eBMD was causally associated with higher 2hGlu and risk of T2DM.

To our knowledge, we are the first to investigate whether BMD can be a risk factor for T2DM incidence, although it was commonly observed that BMD is higher among T2DM patients than those without T2DM.^[2,27,] The positive association is in line with a study showing high BMD is associated with higher prediabetic risk and glycemic deterioration.^[28] In the prospective HKOS‐B cohort, we observed that BMD at the femoral neck and total hip but not the lumbar spine was significantly associated with an increased risk of T2DM in the fully adjusted model. This is in agreement with the observation in the cross‐sectional analysis that the association of BMD at the spine with fasting glucose and Hb_A1c had a smaller estimate than the BMD at the hip (Table 3).

The cross‐sectional relationship between bone turnover markers and glucose metabolism was examined before.^[29‐33,] In general, inverse associations of bone turnover markers with several glycemic traits were reported, such as Homeostatic Model Assessment for Insulin Resistance (HOMA‐IR),^[29,] Hb_A1c,^[32,] fasting glucose,^[31,] amplitude of glycemic excursions and rise in dawn glucose,^[30,] whereas bone turnover markers were also reported to be positively associated with HOMA‐B,^[29,] despite positive association reported for CTx with Hb_A1c.^[33,] Nevertheless, most of these studies only evaluated one single bone formation and one single bone resorption marker at most,^[30‐33,] or conducted the study in people with abnormal glycemic control.^[29,30,] Conversely, we used multiple markers representing bone formation and resorption and showed that lower bone turnover is associated with higher fasting glucose and Hb_A1c. This could be a potential mechanism underlying the relationship of BMD with the risk of diabetes, as low bone turnover is associated with higher BMD (Table S6). One cross‐sectional study also evaluated the relationship of BMD with glycemic traits in people without dysglycemia, showing that people with osteoporosis (defined by BMD at the spine) had higher fasting glucose and postprandial glucose.^[34] However, that was a simple correlation analysis without adjustment for confounders. Conversely, our study showed that higher BMD was indeed associated with higher fasting glucose and Hb_A1c, even after adjustment for important confounders.

The interplay between bone metabolism and glycemic control is complex. Obesity is the potential contributor to the inverse relationship, but adjustment for BMI in the model did not attenuate the significant relationship. Osteocalcin, a bone‐secreted protein, could play a role in the relationship between high BMD and dysglycemia. A previous study also observed women with a higher risk of prior gestational diabetes mellitus tend to have higher BMD and lower osteocalcin. As osteocalcin declines over time, the glycemic status gradually changes from normoglycemia to diabetes.^[35,] We showed the inverse cross‐sectional relationship between osteocalcin and fasting glucose and Hb_A1c in HKOS‐FU. However, the role of osteocalcin in humans remains controversial due to inconsistent findings.^[9,36,] The lack of standardization in osteocalcin measurement, such as whether to measure uncarboxylated osteocalcin or total osteocalcin, the choice of the assay, and the timing of measurement considering the diurnal rhythmicity in circulating bone turnover markers, all contribute to the difficulty in comparing results across studies. Furthermore, it even remains unknown whether the active form of osteocalcin is fully uncarboxylated. In contrast, BMD might serve as a better alternative to osteocalcin given its standardized measurement and stable nature. Moreover, the causal role of osteocalcin in human energy metabolism has yet to be confirmed. It is possible that the previously observed association of osteocalcin with energy metabolism in humans may be attributable to its impact on bone metabolism, particularly when our MR analysis suggests that BMD is probably a genetic causal factor for T2DM, with SNPs associated with osteocalcin excluded. Besides osteocalcin, we also showed other bone biomarkers such as CTx, NTx, and P1NP could be a marker or potentially mediate the relationship. Previous studies also demonstrated that other osteokines may also influence glucose homeostasis.^[37,38,] For example, bone morphogenic proteins derived from the bone matrix and osteoprotegerin secreted by osteoblasts have been shown to regulate insulin secretion.^[39,40,] Observational studies examining the effect of denosumab on glucose parameter improvement^[41,] and T2DM risk reduction^[42] also support osteoprotegerin may play a role in mediating the relationship, but there has not been a clinical trial on denosumab with glycemic traits as the intended outcome in those with diabetes. Considering the intricate interplay between bone and glucose metabolism, depicting a complete picture of the crosstalk between these two systems is challenging and requires further research.

Observational studies cannot infer causality. Despite having adjusted for many covariates, we acknowledged several unmeasured or unobserved confounders may exist, including osteocalcin and hyperinsulinemia. Therefore, we conducted MR analyses to investigate the directional causal association of eBMD with 2hGlu and T2DM risk with the help of genetic IVs. A well‐conducted MR study should be free of reverse causation and confounders that are often present in conventional observational studies.^[43,] We found that genetically higher eBMD was significantly and causally associated with higher T2DM. The associations were consistent in all sensitivity analyses, including MR‐Egger, which is well‐documented to have the lowest power in detecting an association.^[44,] Significant associations were also observed for 2hGlu in the IVW and ConMix analyses. The level of 2hGlu is considered the gold standard in diagnosing diabetes, which reflects the key variables affecting glucose metabolism, including endogenous glucose production, glucose absorption and excretion, cellular uptake of glucose, and insulin secretion and sensitivity.^[45,] It should be noted that the ConMix method has a higher power to detect an association in several scenarios when compared to the other methods used in this analysis, even in the presence of some invalid instruments, which is possible when a large number of IVs were used in the analysis.^[44,] In our MR analyses, the ConMix method consistently demonstrated the association of genetically elevated eBMD with increased 2hGlu and T2DM risk, providing robust evidence for the causal inference. One previous MR study was conducted and found that a 1‐SD increase in eBMD was associated with an 8% increased risk of T2DM.^[46,] The present study, utilizing a larger sample size for both exposure and outcome GWAS, showed similar results. This expanded sample size facilitated the inclusion of a greater number of IVs and enhanced statistical power. For the reverse direction, the previous study found no causal effect of T2DM on eBMD.^[46,] Furthermore, two recent MR studies evaluated the genetically causal effects of glucose metabolism on BMD. Genetically determined fasting glucose was not associated with BMD,^[47,] whereas genetically determined fasting insulin was positively associated with BMD.^[48] Together with the current study, we suggest that impaired glucose metabolism is a consequence, instead of a cause, of higher BMD.

Our study has important clinical implications. We showed that a higher BMD observed in diabetic patients may well exist even before the diagnosis of T2DM and predict the future T2DM incidence. Moreover, we showed that low bone turnover is associated with higher fasting glucose. Together with the MR results, we provide robust evidence on the crosstalk between bone and glucose metabolism in humans, and such an impact of bone on glycemic control is probably causal. Although we illustrated a higher BMD could increase the risk of T2DM, it is important to note that the current evidence does not suggest an increased risk of T2DM with the use of anti‐osteoporosis therapies.^[49] Given the complexity of T2DM etiology, further studies are necessary to elucidate the mechanisms by which bone metabolism affects the risk of T2DM.

This study has many strengths. It evaluated the relationship of bone with glycemic traits and risk of diabetes, from cross‐sectional to longitudinal analyses, and the results are consistent. Higher BMD was associated with poor glycemic control and increased risk of T2DM, providing insights into the potential effects of bone homeostasis on glycemic control in humans. In addition, the study has a large sample size and long follow‐up time, providing ample statistical power for detecting the relationship. However, this study also has some limitations. First, volumetric BMD is not available in the current study, although real BMD that is measured by DXA is the more clinically important phenotype, so future studies evaluating the relationship of volumetric BMD with glycemic traits are warranted. Second, incomplete capture of diabetes cases is possible.^[16,] However, the incomplete capture is likely to underestimate the results rather than overestimate them. Third, the interpretation of the cross‐sectional results requires extra caution because they do not indicate whether BMD is the cause or effect, whereas our prospective study and MR analyses elucidate the direction of the relationship. Fourth, HKOS participants are all Chinese only, the generalizability to other populations is unknown. Nevertheless, similar associations of BMD with prediabetes risk were observed among African‐American and European‐American offspring,^[28,] and the MR analysis was conducted using data derived from the European population, both suggesting that such a relationship may also exist in the non‐Chinese population. Fifth, the measurements for glycemic traits and bone turnover markers in HKOS‐B were unavailable and we were unable to examine the prediction of BMD and bone turnover markers on adverse glycemic traits. Therefore, future studies are necessary to address these gaps and provide more insights into the relationship. For the MR, we used genetic instruments of eBMD, instead of DXA‐measured BMD, so the obtained estimates should not be directly compared with those obtained from DXA‐measured BMD. The use of eBMD in the current study is because the number of genetic instruments of DXA‐measured BMD was limited, and the variance explained was small, therefore leading to lower statistical power. Although eBMD is a good proxy of DXA‐measured BMD,^[17,] the analysis should be performed using genetic instruments from the upcoming largest GWAS meta‐analysis of DXA‐measured BMD when it is available. Furthermore, the GWAS of eBMD consisted of UK Biobank participants only and it might be highly overlapped with the GWAS meta‐analysis of T2DM (49.3%). Nevertheless, the bias and Type I error rate incurred were estimated to be minimal. In addition, a recent simulation study suggested that two‐sample MR methods could be safely applied within biobanks of large sample size (>300,000).^[50] Last but not least, although the violation of the MR independence assumption was mitigated largely by excluding IVs associated with confounding traits (Table S1), the possibility of other confounders, such as hyperinsulinemia, which might violate this assumption remains due to the lack of specific GWAS. Nevertheless, none of the IVs utilized in our study were significantly associated with other insulin‐related traits, including insulin sensitivity, insulin secretion, and proinsulin levels.

In conclusion, high BMD and low bone biomarkers, for example, CTx, NTx, and osteocalcin, are cross‐sectionally significantly associated with poor glycemic control. Meanwhile, high BMD was a risk factor for T2DM in the prospective analysis. MR results suggest the relationships of BMD with 2hGlu and T2DM are probably causal. Further studies evaluating the relationship of bone metabolism in glycemic control and the clinical usefulness of bone markers and BMD in the management of T2DM and surveillance of T2DM prognosis are warranted.

Author Contributions

Xiaowen Zhang: Conceptualization; data curation; formal analysis; validation; investigation; visualization; writing – review and editing; writing – original draft; methodology; software. Suhas Krishnamoorthy: Conceptualization; data curation; validation; investigation; methodology; software; writing – review and editing; formal analysis; visualization. Casey Tze‐Lam Tang: Conceptualization; methodology; data curation; formal analysis; investigation; writing – original draft; software; visualization. Warrington Wen‐Qiang Hsu: Investigation. Gloria Hoi‐Yee Li: Writing – review and editing. Chor‐Wing Sing: Writing – review and editing. Kathryn Choon‐Beng Tan: Resources; project administration. Bernard Man‐Yung Cheung: Resources. Ian Chi‐Kei Wong: Supervision. Annie Wai‐Chee Kung: Resources; project administration. Ching‐Lung Cheung: Project administration; conceptualization; methodology; writing – original draft; writing – review and editing; supervision; resources; visualization; funding acquisition.

Acknowledgment

This work was supported by AIR@InnoHK administered by Innovation and Technology Commission.

Disclosures

The authors declare no conflicts of interest.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/jbmr.4924.

Data Availability Statement

The HKOS data are not freely available, as the informed consent forms of the HKOS did not include data sharing. The CDARS is directly under the control of the Hong Kong Hospital Authority. CDARS data can be accessed for research purposes through the application to HA Data Sharing Portal. The related information can be found online (https://www3.ha.org.hk/data). Research data for the Mendelian randomization analysis were obtained from the respective consortiums and can be found through their respective publications.^[17,19,20] The code used to generate the results for the Mendelian Randomization analysis is available at https://github.com/lung1212/BMD_DM.

REFERENCES

Author notes