This study was based on the Dubbo Osteoporosis Epidemiology Study (DOES) with the protocols and study structure described elsewhere.[15] Briefly, DOES was designed as a prospective, population-based investigation that was operational from 1989 to 2020. Approximately 4,000 men and women aged 60 years and older have been recruited and followed up for almost 30 years. The study was approved by the St. Vincent’s Campus Research Ethics Committee and written informed consent was acquired from each participant. As this study was primarily focused on post-menopausal women, we excluded men and any participants without FTO genotype data (Fig. 1).

A nurse co-ordinator used a structured questionnaire to obtain baseline measurements at 1989 and subsequent visits at approximately 2 to 3 yearly intervals. The structured questionnaire included age, weight, height, smoking and alcohol intake, physical activity, calcium intake, medication, number of falls in the last year, and fracture history. Weight (kg) and height (cm), without shoes and in light clothing, was measured on an electronic scale (to the nearest 0.1 kg) and by a wall-mounted stadiometer (to the nearest 0.1 cm). BMI (kg/m²) was calculated as the ratio of weight (kg) and squared height (m²).

BMD (g/cm²) at the femoral neck and lumbar spine was measured at baseline and biennially by a dual energy X-ray absorptiometry (DXA) instrument using a LUNAR DPX densitometer (GE-Lunar, Madison, WI, USA). The coefficient of variations of BMD measurement of the femoral neck and lumbar spine was 1.5% and 1.3%, respectively.[16] All measurements were conducted by the same technicians, following the same standard operating procedure.

All fractures were ascertained from the 2 or 3 radiology centres servicing the Dubbo area, and circumstances surrounding fracture were ascertained by personal interview following the fracture event. Only low-trauma fractures caused by a fall from a standing height or less were included in the study. Fractures caused by high trauma, such as motor vehicle accidents, pathological fractures from bone diseases other than osteoporosis (e.g., hyperparathyroidism, Paget’s disease), or fractures of the skull and digits were excluded. Vertebral fractures were determined either through clinical diagnosis or detected incidentally on X-ray. Deaths were ascertained from funeral lists, obituary reports, and Dubbo media reports, and verified from the New South Wales Bureau of Births, Deaths, and Marriages.

The genotype data for this study was drawn from a previous investigation, which describes the genotyping protocol used.[17] Briefly, blood samples were collected and stored at −80°C. Extraction of DNA was done either by using QIAamp DNA Mini Blood Kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s instructions, or phenol/chloroform. Six SNPs (rs1421085, rs1558902, rs1121980, rs17817449, rs9939609, and rs9930506) in the first intron of the FTO gene were genotyped using a predesign of Taqman SNP Genotyping Assay (Applied Biosystems, Foster City, CA, USA). The ABI 7900 Sequence Detection System (SDS) software was used to compute the allelic discrimination. From the participants of DOES, the DNA of 1,277 women and 758 men were able to be genotyped. The number of fractures was too small to conduct a sensitive analysis for elderly men, therefore, we primarily focused on post-menopausal women with FTO genotype data.

The first analysis included participants with BMD measured at baseline. We performed a multiple-adjusted linear regression to quantify the association between FTO genotypes and baseline BMD at the femoral neck using the AA genotype as the reference group. The model was adjusted for predefined covariates of age and BMI. A predefined sensitivity analysis was also considered for lumbar spine BMD. To control for potential confounding effects of metabolic diseases, an exploratory analysis was conducted to further adjust for the presence of diabetes mellitus.

For the second aim, the rate of change in BMD was calculated for each participant with at least 2 BMD measurements using a linear mixed-effects regression. We also conducted a linear mixed-effects regression, adjusted for the predefined covariates to examine the association between FTO genotypes and bone loss. A linear mixed-effects regression has been shown to be statistically superior to a conventional linear regression in analysis of repeated measurement.[18,19] A linear mixed-effects regression is capable of accounting for not only a regression-toward-the-mean phenomenon, but also missing data which are very common in any analysis with repeated measures.[18] Moreover, a linear mixed-effects regression is also able to account for variation between and within subjects.[19]

For our third aim, the Cox’s proportional hazards regression was used to quantify the association between FTO genotypes and fracture risk, adjusted for age, BMD, and BMI. The follow-up time for participants with a fracture was calculated between the study entry and the date of fracture. For participants without a fracture, their follow-up was computed as a time interval between the study entry and date of death or the study end (March 5, 2018), whichever came first. The assumption of proportional hazards was graphically checked using the scaled Schoenfeld residuals.[20]

We computed the E-value [21] as a predefined sensitivity analysis to assess the robustness of our findings in the potential of residual confounding effects. Technically, the E-value quantifies possibilities that an uncontrolled confounder would make the reported association statistically non-significant. No adjustments were made for multiple comparisons. The analyses were performed using R language [22] with a P-value of less than 0.05 considered statistically significant.

The current study involved 1,277 women with an average age of 69 years (±6.6) whose FTO genotype data were available (Fig. 1). The distribution of FTO genotypes is consistent with the Hardy-Weinberg principle as the proportion of the major (32%-38%), heterozygous (44%-49%) and minor (16%-20%) allele are similar in the FTO SNP rs9930506 (Table 1) and other FTO SNPs-rs1421085, rs1558902, rs1121980, rs17817449, and rs9939609 (Supplementary Table 1). Given the great similarities between the FTO SNPs and their close proximity to each other at intron 1 of the FTO gene, we present the results of rs9930506 in the main paper and leave those of other FTO genotypes in the appendix.

Table 2 illustrates the association between FTO genotypes rs9930506 and baseline BMD. We found FTO genotypes were not associated with BMD after accounting for potential confounding effects of age and BMI (Table 2). Similarly, there was no association between BMD and the other FTO SNPs - rs1421085, rs1558902, rs1121980, rs17817449, and rs9939609 (Supplementary Table 2-6).

In our study population of post-menopausal women, the overall rates of BMD change were −0.004 g/cm² per year at the femoral neck and 0.002 g/cm² per year at the lumbar spine. In the FTO SNP rs9930506, the annual rate of bone change was similar across the FTO genotypes at the femoral neck (−0.004 g/cm²) and lumbar spine (0.002 g/cm²) (Table 3). The analysis did not find a significant difference in the rate of BMD change over time between the FTO genotypes after taking the potential confounding effect of age and BMI into account (Table 3, Supplementary Table 2-6).

During a median follow-up of 14 years (interquartile range, 10-22 years), 605 women sustained at least one incident fracture, yielding a fracture rate of 40/1,000 person-years (95% confidence interval [CI], 37-44). We also found 147 women with a hip fracture rate of 8/1,000 person-years (95% CI, 6-9). We found FTO genotypes were not independently associated with the risk of any fracture (Table 4, Supplementary Table 4, 5). However, women with the minor allele (GG) were associated with a significantly greater risk of hip fracture (hazard ratio, 1.82; 95% CI, 1.10-3.03). The association remained significant after adjusting for potential confounding effects of age, BMI, and BMD. Interestingly, we found the E-value for the FTO-hip fracture association was as high as 3.25 (95% CI, 1.57-5.85), indicating that the finding would have become non-significant only if there was a very strong residual confounding effect associated with almost a 3-fold risk of hip fracture existing in the study population. Similarly, the minor allele of FTO SNPs rs1121980 was also associated with a greater risk of hip fracture (Supplementary Table 4). The association between the minor allele of rs1421085 (P=0.09), rs1558902 (P=0.06), and rs17817449 (P=0.05) with hip fracture only achieved marginally statistically significant results (Supplementary Table 2, 3, and 5). The exploratory analysis further adjusted for the presence of metabolic diseases provided consistent results (Supplementary Table 7), confirming the robustness of our primary findings.