Proteome-wide profiling reveals dysregulated molecular features and accelerated aging in osteoporosis: A 9.8-year prospective study

Abstract

The role of circulatory proteomics in osteoporosis is unclear. Proteome-wide profiling holds the potential to offer mechanistic insights into osteoporosis. Serum proteome with 413 proteins was profiled by liquid chromatography-tandem mass spectrometry (LC-MS/MS) at baseline, and the 2nd, and 3rd follow-ups (7704 person-tests) in the prospective Chinese cohorts with 9.8 follow-up years: discovery cohort (n = 1785) and internal validation cohort (n = 1630). Bone mineral density (BMD) was measured using dual-energy X-ray absorptiometry (DXA) at follow-ups 1 through 3 at lumbar spine (LS) and femoral neck (FN). We used the Light Gradient Boosting Machine (LightGBM) to identify the osteoporosis (OP)-related proteomic features. The relationships between serum proteins and BMD in the two cohorts were estimated by linear mixed-effects model (LMM). Meta-analysis was then performed to explore the combined associations. We identified 53 proteins associated with osteoporosis using LightGBM, and a meta-analysis showed that 22 of these proteins illuminated a significant correlation with BMD (p < 0.05). The most common proteins among them were PHLD, SAMP, PEDF, HPTR, APOA1, SHBG, CO6, A2MG, CBPN, RAIN APOD, and THBG. The identified proteins were used to generate the biological age (BA) of bone. Each 1 SD-year increase in KDM-Proage was associated with higher risk of LS-OP (hazard ratio [HR], 1.25; 95% CI, 1.14-1.36, p = 4.96 × 10^-06 ), and FN-OP (HR, 1.13; 95% CI, 1.02-1.23, p = 9.71 × 10^-03 ). The findings uncovered that the apolipoproteins, zymoproteins, complements, and binding proteins presented new mechanistic insights into osteoporosis. Serum proteomics could be a crucial indicator for evaluating bone aging.

Keywords: biological age; longitudinal study; osteoporosis; proteome-wide study.

FIGURE 1

Overview of the study cohort and design. This study included 3244 participants from the Guangzhou Nutrition and Health Study (GNHS), followed up every 3 years from 2008 to 2019. Serum proteomics was profiled using a liquid chromatography–tandem mass spectrometry (LC–MS/MS) at baseline (F0) and follow‐ups (F) 2 and 3. Bone mineral density (BMD) was measured at F1 through F3 at the lumbar spine (LS) and femoral neck (FN). The LightGBM algorithm was used to identified OP‐proteins from 1746 samples including 314 serum proteins in the discovery cohort (training and testing datasets), and the model was validated in the internal validation cohort with 1220 samples. The associations between baseline serum proteins and BMDs at the lumbar spine (LS) and femoral neck (FN) were estimated, respectively, by linear mixed‐effects model (LMM) in the discovery and internal validation cohorts. The dependent variables were the BMD levels at three follow‐ups. The combined effects of proteins on BMDs from the discovery and internal validation cohorts were assessed using the random‐effects meta‐analysis. We constructed protein risk scores (PRS) based on the protein biomarkers significantly related to the site‐specific bone outcomes to evaluate the joint “effect” on the bone outcomes. We compared protein means by the site‐specific BMD trajectories identified from the latent class trajectory model (LCTM). In addition, Mendelian randomization analyses were performed to validate the potential causal effects of the serum proteins on BMD. We generated the biological ages (BAs) with BMD‐proteins by the Klemera and Doubal method (KDM) to investigate the aging rates of bone. BMD, bone mineral density; DXA, dual‐energy X‐ray absorptiometry; FN, femoral neck; GNHS, Guangzhou Nutrition and Health Study; KDM‐Proage, KDM protein age; LC–MS/MS, liquid chromatography–tandem mass spectrometry; LCTM, latent class trajectory model; LightGBM, Light Gradient Boosting Machine; LMM, linear mixed‐effects model; LS, lumbar spine; MR, Mendelian randomization; PRS, protein risk score.

FIGURE 2

Prospective associations between serum proteins with osteoporosis and BMD. (a) The fold change of serum protein abundance between osteoporosis cases and controls at three time points. Fold change = (mean_‐OP−mean_‐controls)/mean_‐controls. The Benjamini‐Hochberg (BH) false discovery rates (FDR) approach was applied to control alpha error. ***FDR <0.001, **FDR <0.01, *FDR <0.05. (b) The prospective associations between serum proteins and BMDs. The regression coefficients and 95% CIs (in SD/SD) between serum proteins and BMDs at the lumbar spine (LS) and femoral neck (FN) (n = 3244) were estimated by LMM model in the discovery and internal validation cohorts. The longitudinal BMD levels at three follow‐ups as dependent variables. The multivariate regressions were adjusted for baseline age, sex, BMI, waist‐hip ratio, educational level, household income, smoking status, alcohol drinking status, tea consumption, physical activity, total energy intake, total carbohydrate intake, dietary fiber intake, calcium supplement, multivitamins supplement, SBP, DBP, fasting blood glucose, TC, TG, LDL, HDL, and uric acid. The combined effects of proteins on BMDs from the discovery and internal validation cohorts were assessed using the meta‐analysis. The heterogeneity was investigated using the Cochran's Q and I‐square statistics. The Benjamini‐Hochberg (BH) false discovery rates (FDR) approach was applied to control alpha error. (c) The protein–protein interaction of serum proteins. BMD, bone mineral density; BMI, body mass index; CIs, confidence intervals; DBP, diastolic blood pressure; FN, femoral neck; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; LMM, linear mixed‐effects model; LS, lumbar spine; SBP, systolic blood pressure; SD, standard deviation; TC, total cholesterol; TG, triglyceride.

FIGURE 3

Prospective associations of PRS with osteoporosis and BMD. (a) Regression coefficients and 95% CIs (in SD/SD) between baseline protein risk score (PRS) and BMDs. The LMM was used to analyze the associations between baseline PRS and BMD at the LS and FN. (b) The associations between baseline PRS and osteoporosis risk at the LS and FN analyzed using Cox regression. (c,d) Dose–response associations between the baseline PRS and osteoporosis risk at the LS and FN were analyzed by the restricted cubic spline model. The corresponding hazard ratios (95% CIs) for the quartiles (Q1–Q4) were estimated using the Cox regression model. All the analyses were adjusted for baseline age, sex, BMI, waist‐hip ratio, educational level, household income, smoking status, alcohol drinking status, tea consumption, physical activity, total energy intake, total carbohydrate intake, dietary fiber intake, calcium supplement, multivitamins supplement, SBP, DBP, fasting blood glucose, TC, TG, LDL, HDL, and uric acid. BMD, bone mineral density; BMI, body mass index; CIs, confidence intervals; DBP, diastolic blood pressure; FN, femoral neck; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; LMM, linear mixed‐effects model; LS, lumbar spine; SBP, systolic blood pressure; SD, standard deviation; TC, total cholesterol; TG, triglyceride.

FIGURE 4

The cross‐sectional associations of PRS with osteoporosis risk and BMD at three follow‐up visits. The cross‐sectional associations of protein risk score (PRS) with BMD (a) and osteoporosis risk (b) at each visit. The PRS was constructed using the protein concentrations at the baseline, second, and third follow‐ups as well as the beta coefficients from the meta‐analysis. The cross‐sectional associations were replicated using the PRS and BMD/osteoporosis data that determined at the baseline, 2nd, and 3rd follow‐ups and analyzed by generalized linear models (GLM). Covariates adjusted: see Figure 3.

FIGURE 5

The prospective associations between biological age scores and osteoporosis risk. The prospective associations of KDM‐Proage (a) and BioAgeAccel (b) with osteoporosis risk. The associations between baseline biological age scores (in per SD change) and osteoporosis risk were analyzed using Cox regression. Model 1 was adjusted for baseline age and sex. Model 2 was adjusted for Model 1 + BMI, waist‐hip ratio, educational level, household income, smoking status, alcohol drinking status, tea consumption, physical activity, total energy intake, total carbohydrate intake, dietary fiber intake, calcium supplement, and multivitamins supplement. Model 3 was adjusted for Model 2 + SBP, DBP, fasting blood glucose, TC, TG, LDL, HDL, and uric acid. BMD, bone mineral density; BMI, body mass index; CIs, confidence intervals; DBP, diastolic blood pressure; FN, femoral neck; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; LMM, linear mixed‐effects model; LS, lumbar spine; SBP, systolic blood pressure; SD, standard deviation; TC, total cholesterol; TG, triglyceride.