A density of >650 mg cm-3 was used to define cortical bone. Endosteal and periosteal circumference were derived using a circular ring model. 4502 pQCT scans were performed, of which 88 were excluded due to major motion artifacts. Coefficients of variation

for pQCT scans, based on 139 subjects scanned a mean of 31 days apart, were 2.7%, 1.3% and 2.9% for BMCC, BMDC SB202190 and cortical bone area, respectively. Other variables At 15.5 years research clinics, standing height (mm) was measured using the Harpenden Stadiometer (Holtain, Crymych, Wales, UK), and weight using the Tanita Body Fat Analyzer (model TBF 305; Tanita, Arlington Heights, IL, USA). Whole body DXA scans were performed using a Lunar MEK inhibitor Prodigy scanner with paediatric scanning software (GE Lunar Prodigy, Madison, WI, USA), providing measures of total body fat and lean mass. Maternal SEP was recorded at 32 weeks gestation by questionnaire and categorised according to the Office of Population Censuses and Surveys. Maternal learn more education was assessed at the same time by questionnaire. Pubertal stage was assessed using a Tanner stage (pubic hair domain) questionnaire completed

at age 14.7 years [22]. Moderate and vigorous physical activity was assessed by actigraph accelerometre at age 11, and subsequently found to be related to BMD in ALSPAC [23]. Date of birth and sex was obtained from birth notification, and date of the scan was recorded automatically, allowing age at scan to be calculated. Statistical analyses Descriptive statistics show means, standard deviation (SD), medians and lower and upper quartiles. Analyses were performed using seasonally adjusted 25 (OH)D3, which was modelled according to date of blood sampling using linear regression with trigonometric sine and cosine functions. 25(OH)D3 was loge transformed to reduce

heteroscedasticity. The residual was used as the primary 25(OH)D3 exposure variable in subsequent regression analyses. All analyses were performed Non-specific serine/threonine protein kinase on standardised variables, i.e. subtracting the mean and dividing by the SD. To include all participants on whom a 25(OH)D2 was assayed, those with a value below the detectable limit of the assay (0.5 ng ml-1) were assigned a binary variable indicating whether an individual was at or below the lower limit, which was used as a covariable in all regression models. No individuals had 25(OH)D3 below the detectable limit of the assay. Models were checked for linearity by adding higher-order terms into the linear predictor and by comparing the likelihood of nested models. Further analyses were performed using a nonparametric bootstrap procedure in conjunction with OLS linear regression, based on 5,000 replications. Beta (β) estimates and standard errors were calculated from the mean and SD of the bootstrap distribution, respectively. All P values were calculated using bootstrap means and standard errors, compared to a Z-distribution and 95% percentile confidence intervals calculated.