Identification and Functional Characterization of Metabolites for Bone Mass in Peri- and Postmenopausal Chinese Women

Abstract

Context

Although metabolic profiles appear to play an important role in menopausal bone loss, the functional mechanisms by which metabolites influence bone mineral density (BMD) during menopause are largely unknown.

Objective

We aimed to systematically identify metabolites associated with BMD variation and their potential functional mechanisms in peri- and postmenopausal women.

Design and Methods

We performed serum metabolomic profiling and whole-genome sequencing for 517 perimenopausal (16%) and early postmenopausal (84%) women aged 41 to 64 years in this cross-sectional study. Partial least squares regression and general linear regression analysis were applied to identify BMD-associated metabolites, and weighted gene co-expression network analysis was performed to construct co-functional metabolite modules. Furthermore, we performed Mendelian randomization analysis to identify causal relationships between BMD-associated metabolites and BMD variation. Finally, we explored the effects of a novel prominent BMD-associated metabolite on bone metabolism through both in vivo/in vitro experiments.

Results

Twenty metabolites and a co-functional metabolite module (consisting of fatty acids) were significantly associated with BMD variation. We found dodecanoic acid (DA), within the identified module causally decreased total hip BMD. Subsequently, the in vivo experiments might support that dietary supplementation with DA could promote bone loss, as well as increase the osteoblast and osteoclast numbers in normal/ovariectomized mice. Dodecanoic acid treatment differentially promoted osteoblast and osteoclast differentiation, especially for osteoclast differentiation at higher concentrations in vitro (eg,10, 100 μM).

Conclusions

This study sheds light on metabolomic profiles associated with postmenopausal osteoporosis risk, highlighting the potential importance of fatty acids, as exemplified by DA, in regulating BMD.

Osteoporosis is a progressive, age-related metabolic bone disease. It is characterized primarily by low bone mineral density (BMD) and microstructural deterioration of bone tissues, which increases bone fragility and risk of osteoporotic fracture (1). Osteoporosis has become a serious global health problem for middle-aged and older adults, imposing a heavy economic burden on medical and economic systems as the global population continues to age (2). Postmenopausal osteoporosis (PMOP) is the most common form of osteoporosis. Among individuals over 50 years old, the prevalence of osteoporosis is significantly higher in Chinese women (40.1%) than in men (22.4%) (3). Within the 3 years surrounding the final menstrual period (FMP), there is rapid bone loss that is partly attributable to a corresponding reduction in ovarian estrogen production (4). The basic pathophysiology of PMOP involves an imbalance between bone resorption and formation induced by estrogen deficiency (5). Metabolic factors linked to menopause status, such as insulin resistance (6), inflammation (7), and obesity (8), are associated with decreased bone strength and increased fracture risk. Thus, metabolic profiles have an important influence on or associated with bone health.

Metabolomics studies can profile metabolic changes by identifying and quantifying metabolites in biological samples of various health conditions. Since metabolites represent the downstream expression of the genome, transcriptome, and proteome, they are most likely to reveal inherent/intrinsic omics variation related to complex diseases (9). Metabolomic studies have greatly improved our knowledge of specific human diseases such as coronary heart disease (10), type 2 diabetes (11), obesity (12), and breast cancer (13). Liquid chromatography-mass spectrometry (LC-MS)-based untargeted metabolomics, which provides a broad coverage of metabolic products, has proven to be a powerful tool for detecting novel metabolites associated with human disorders (14, 15).

Animal studies have identified a number of metabolites (eg, arachidonic acid (AA) (16), oleic acid (17), and L-hydroxyproline (17)) associated with PMOP. The association between the metabolome and BMD in humans has also been evaluated (18–24). Mendelian randomization (MR) studies have identified specific metabolites causally associated with BMD by analyzing genome-phenotype association data and metabolomics-phenotype association data (25, 26). The existing MR studies mainly focused on European populations. However, knowledge about the causal relationships between metabolites and BMD in other populations, such as Chinese populations, is very limited. Additionally, few, if any, functional studies have explored metabolites and metabolic networks (14) associated with BMD variation. Functional studies are critical for validating the metabolites identified in human studies of BMD but have not been pursued so far.

To better understand the pathogenesis of PMOP, we first investigated metabolites and co-functional metabolite modules associated with total hip BMD (HTOT-BMD), which includes femoral neck, trochanter, and intertrochanteric region, in perimenopausal and early postmenopausal Chinese women. A series of continuous and dynamic changes in BMD as well as various hormonal and metabolic states are observed during the transition between the perimenopausal and early postmenopausal phases (27, 28). Therefore, peri- and postmenopausal women were combined to maximize our study power to identify the metabolites associated with BMD variation during the trans-menopause process. We focused on HTOT-BMD since it is the most commonly used criteria for clinically diagnosing osteoporosis, and low-trauma fractures of the hip are a major cause of morbidity, physical disability, and early mortality (29). Additionally, hip BMD may better reflect the bone mass in the corresponding region compared with spine BMD because abnormal calcium deposition within the field of the dual-energy X-ray absorptiometry (DXA) region of interest can lead to a falsely elevated spine BMD measurement (30). We identified metabolites causally associated with BMD by integrating metabolomics and whole-genome sequencing data of the same study subjects using a 1-sample MR analysis (31). As a demonstration and proof of concept, we subsequently conducted functional studies for a specific novel prominent BMD-associated metabolite in regulating bone metabolism both in vitro and in vivo. We identified a list of BMD-associated metabolites and a co-functional metabolite module that highlighted the importance of fatty acids in bone metabolism and furnished novel potential mechanisms through which specific fatty acids may affect bone metabolism.

Methods

The overview of the workflow in this study is presented in Fig. 1.

Figure 1.

Overview of the workflow in this study. Abbreviations: μCT, micro-computed tomography; MR, Mendelian randomization; WGCNA, weighted gene co-expression network analysis.

Study population

This cross-sectional study included 517 unrelated peri- and postmenopausal Chinese women who were enrolled from local communities from June 2014 to January 2018. All enrollment processes were performed in the outpatient clinic at the Third Affiliated Hospital of Southern Medical University (Guang Zhou, Guang Dong Province, China). The Medical Ethical Committees of Southern Medical University approved the study protocol, and each subject signed an informed consent before participating in this study. This study complies with the principle of the Helsinki Declaration II. The inclusion criteria included participants that are: (1) 40 years of age or older, (2) in the peri or postmenopausal phase, and (3) having lived in Guang Zhou City for at least 3 months. Peri- and postmenopausal phases were defined according to the criteria of the Stages of Reproductive Aging Workshop (32). Women in the perimenopausal phase had irregular menstrual cycles, but less than 12 months had passed since the FMP. Women with no menstrual cycles for more than 12 months were in the postmenopausal phase. Specifically, we included early postmenopausal women with approximately 5 to 8 years since their FMP (32). Subjects taking antibiotics, estrogens, antiosteoporosis medication, anticonvulsant medications or any medications that would affect BMD, and those with diseases or conditions (ie, bilateral ovariectomy and chronic renal diseases) that could have effects on bone mineral density (BMD), were excluded from the present study.

Clinical measurements

An interviewer-assisted comprehensive questionnaire was administered to collect information on demographics, annual family income, lifestyle (ie, smoking, alcohol drinking, calcium supplementation, and regular exercise), years since menopause, disease history, and medication history. Based on the average exercise time per week, subjects were divided into 3 groups (those without exercise, less than 2.5 hours per week, and more than 2.5 hours per week) (33). The annual family income was divided into groups based on a local economic report. Body mass index (BMI) was calculated as weight (kg) divided by the square of the height (m²). We used a dual-energy X-ray absorptiometry machine (DXA version 13.31.016; GE Healthcare, Madison, WI, USA) to measure HTOT-BMD, and the combined BMD of the femoral neck, trochanter, and intertrochanteric region with a standard scan model. The machine was calibrated daily, and software and hardware were updated during the data collection process. The coefficient of variation (CV) was only assessed for spine BMD and was 0.89%, which was used to reflect the accuracy of the HTOT-BMD measurement. Blood samples from each participant were collected after an overnight fast (> 8 hours). The interview, DXA measurement, and blood sample collection were conducted on the same day. Deoxyribonucleic acid (DNA) was extracted from blood samples with a SolPure DNA Kit (Magen, Guangzhou, China), and serum samples were used for serum analyses and extracting metabolites. Serum and DNA were stored in a freezer at -80°C until further analyses.

Metabolomics analysis

An LC-MS system was applied to perform the untargeted metabolomic analysis by LC-Bio Technologies (Hangzhou) CO., LTD. (Hangzhou, China, www.lc-bio.com; this company owns an LC Sciences R&D department in Houston, TX, USA, www.lcsciences.com). Serum metabolites were separated by reverse phase chromatography with an ACQUITY UPLC BEH Amide column (100mm*2.1mm, 1.7 µm, Waters, Wilmslow, UK) on an ultra-performance liquid chromatography (UPLC) system (SCIEX, Framingham, MA, USA) (34). Eluted metabolites were identified with a high-resolution tandem mass spectrometer TripleTOF5600plus (SCIEX). The quadrupole-time of flight was operated in both positive and negative ion modes.

XCMS software (35) was used to perform pretreatments for the acquired mass spectrometry (MS) data, including peak picking, peak grouping, retention time correction, second peak grouping, and annotation of isotopes and adducts. LC-MS raw data files were converted into an mzXML format and then processed by the XCMS, CAMERA, and metaX toolbox using R software. Metabolomics profiling was analyzed in both positive and negative ion modes, and each ion was characterized by retention time and mass-to-charge ratios (m/z) data. Intensities of each peak were recorded, and a three-dimensional matrix was generated, which contained arbitrarily assigned peak indices (retention time-m/z pairs), sample names (observations), and ion intensity information (variables). For metabolite identifications, the Kyoto Encyclopedia of Genes and Genomes and Human Metabolome Database were used to annotate the metabolites by matching the exact molecular mass data of samples with those from the corresponding databases. If a mass difference between observed and the database value was less than 10 ppm, the metabolite would be annotated, and then the molecular formula of metabolites would be identified and validated by isotopic distribution measurements. We validated metabolite identification with an in-house fragment spectrum library (LC-Bio Technologies (Hangzhou) CO., LTD.) of metabolites. MetaX (36) was used to preprocess peak intensity data for further statistical analysis. Metabolomic features that were detected in less than 50% of quality control (QC) samples or 80% of samples were removed, and the k-nearest neighbor algorithm was used to impute the remaining missing values to further improve data quality. Probabilistic quotient normalization (37) was used to minimize technical artifacts. Quality control (robust spline correction) was performed for the postacquisition correction of batch effects. Moreover, the relative standard deviations of the metabolomic features were calculated across all QC samples, and those >30% were then removed. The relative abundance data of metabolites were further log-transformed and autoscaled to have zero mean and unit variance (z-scores) using the R package “specmine.”

Association analyses between BMD and metabolites

To identify metabolites that were significantly associated with variation in HTOT-BMD, we applied partial least squares (PLS) regression and general linear regression analysis to model the metabolomic profile. The PLS regression, which projects latent structures to maximize the covariance between the loading scores of X (metabolites) and Y (HTOT-BMD) is widely used in metabolomics analysis due to the collinearity between functionally related metabolites (38). It provides a variable importance in projection (VIP) score for each metabolite, which represents the relative importance of each metabolite in the model. We also used linear regression analysis to examine the associations between individual metabolites and BMD status. We used the false discovery rate (FDR) method to correct for multiple testing. The PLS method differs from the ordinary linear regression analysis in that all the metabolites are considered simultaneously rather than testing them individually (14). Therefore, these methods serve as complementary approaches for identifying BMD-associated metabolites. The PLS and linear regression models were adjusted for the known/reported factors including age, BMI, regular exercise, calcium supplementation, physically demanding job, annual family income, alcohol drinking, and years since menopause to exclude potential confounding due to these factors. In order to generate a comprehensive list of BMD-associated metabolites in this study for future replication, we selected BMD-associated metabolites with a VIP ≥ 2.0 in PLS regression or FDR < 0.05 in linear regression. The metabolite association analyses were performed for metabolic features, which could potentially be annotated to overlapping metabolites. In cases where overlapping metabolites were identified from both the positive and negative ion modes, we selected the most significant feature in the linear regression and/or the feature with the highest VIP values.

Furthermore, we evaluated the performance of BMD-associated metabolites on the prediction of BMD variation. We calculated 2 types of metabolite scores, based on BMD-associated metabolites identified in the PLS and linear regression analyses, respectively. The prediction performance of the BMD-associated metabolites was assessed by the proportion of variance explained (measured by R²) and prediction error (measured by root mean squared error [RMSE]). The data analyses were performed using R packages, including “mixOmics,” “robustbase,” and “tidyverse.”

Weighted gene co-expression network analysis for clustering metabolites and pathway analysis

For the 381 metabolites identified in this study, clustering of co-abundant metabolites was performed to construct co-functional metabolite modules using the R package weighted gene co-expression network analysis (WGCNA) (37). Default values of the parameters were used (39). We used biweighted midcorrelations, a median-based correlation measure that is more robust to outliers than Pearson correlation (40), to calculate the correlations between any 2 metabolites and between metabolite modules and BMD variation. We used FDR multiple testing correction to identify metabolite module for association with BMD (FDR < 0.2). Bone mineral density status was also adjusted for potential confounding factors similar to previous association analyses. Finally, we performed pathway analysis by MetaboAnalyst 4.0 (41) for BMD-associated modules to explore the potential major functions of the modules.

Mendelian randomization analysis for potential causality between BMD-associated metabolites and HTOT-BMD

Whole-genome sequencing and genome-wide association study (GWAS).

Whole-genome sequencing was performed by BGI Genomics Co Ltd (Shenzhen, China; https://www.genomics.cn/) with BGISEQ-500 platform. We performed genome-wide association studies (GWAS) to select significantly associated single nucleotide polymorphisms (SNPs) for each BMD-associated metabolite and HTOT-BMD by applying PLINK 1.9 software with the default criteria as follows. For subsequent analyses, we included SNPs with missing call rates < 0.1, minor allele frequencies > 0.01, and Hardy-Weinberg equilibrium P-value > 1.0 × 10^–5.

Mendelian randomization analysis.

Based on GWAS results, we used bidirectional 1-sample MR analysis (42) to identify causal relationships between each BMD-associated metabolite and HTOT-BMD. In the direction of metabolite-to-BMD analysis, we first selected independent genetic variants (r² ≤ 0.001) associated with each BMD-associated metabolite (P-values < 1 × 10⁻⁵) as the instrumental variables (43). Second, we obtained the corresponding effect estimates of those SNPs from the GWAS results for HTOT-BMD. Finally, the simple median (31), weighted median method (31), inverse-variance weighted method (44), and MR-Egger method (31) were applied to identify causal relationships between BMD-associated metabolites and HTOT-BMD by weighting the effect estimate of each SNP on HTOT-BMD by its effect on BMD-associated metabolites. We also investigated the reverse causal relationships between HTOT-BMD (as an exposure) and BMD-associated metabolites (as outcomes). We used the intercept that deviates from the MR-Egger method to test the presence of horizontal pleiotropy, which causes bias in the MR estimates. The MR analyses were performed using R packages, including “MRInstruments” and “MendelianRandomization.”

Mouse experiments with dodecanoic acid

Ovariectomy (OVX) and DA administration.

We purchased C57BL/6J mice (wild-type [WT], female, 6-weeks-old, n = 24) from the Guangdong Medical Laboratory Animal Center (Guangzhou, China). Mice were fed freely with food and water for 2 weeks to acclimate to the new environment before experiments. Half of the mice (8-weeks-old) underwent a bilateral OVX operation (n = 12) by the dorsal approach under general anesthesia with 1.2% tribromoethanol (T831042; Macklin, Shanghai, China) and were allowed to recover for 1 week. The normal mice and OVX mice (9-weeks-old), respectively, were then divided into 2 groups (DA-treated groups and nontreated groups, 6 samples/group). The DA-treated groups were supplemented with DA (L556; Sigma Aldrich, St. Louis, Missouri, USA) at 180 mg/kg, as determined based on a previous study (45), in drinking water daily that was changed every 3 days. Dodecanoic acid was suspended in water by using Tween 80 (P4780; Sigma Aldrich, St. Louis, Missouri, USA), and nontreated groups were only given Tween 80 in the drinking water. All the mice, both with and without the DA supplement, were feed freely with the standard laboratory animal mice and rats formula feed (Supplemental Table 1) (46). Transportation, housing, and feeding of mice were all conducted following the recommendations of “the use of nonhuman primates in research.” After 8 weeks of feeding, we sacrificed the mice and collected blood and femurs for clinical measurements. All procedures were approved by the Animal Care and Use Committee of the Third Affiliated Hospital of the Southern Medical University.

Micro-CT analysis.

Quantitative analysis of bone microstructure changes was performed on the left femur at a 12 μm resolution using a micro-computed tomography (μCT) scanner (μCT40; Scanco Medical AG, Bassersdorf, Switzerland). The structures of trabecular bone were analyzed, including bone volume per tissue volume (BV/TV, %), trabecular number (Tb.N, 1/mm), trabecular thickness (Tb.Th mm), trabecular separation (Tb.Sp), and trabecular BMD (Tb.BMD, g/m³). The scan conditions were set at an aluminum filter of 0.5 mm, X-ray voltage of 50 kV, X-ray current of 200 mA, and exposure time of 360 ms.

Histomorphometric analysis of osteoclasts and osteoblasts in vivo.

Right femurs dissected from the mice were fixed in 4% paraformaldehyde at 4°C for 24 hours and then decalcified in 15% EDTA (pH 7.4) at 4°C for 21 days. Decalcified femurs were dehydrated with ethanol (from 70–100%) and embedded in paraffin, and 2 to 4 μm sagittal-oriented sections were prepared for histological analyses. Tartrate-resistant acid phosphatase (TRAP) (G1492; Solarbio, Beijing, China) staining was performed to determine the number of osteoclasts. Alkaline phosphatase (ALP) staining (C3206; Beyotime, Shanghai, China) was performed to determine the number of osteoblasts. Histologic sections were observed and photographed on a BIX53 microscope (Olympus, Tokyo, Japan). In immunohistochemistry assays, cells per bone perimeter (B.Pm) was used to calculate the number of positive cells by detection in the images taken at 200x magnification with Image-Pro Plus 6.0 software (Media Cybernetics, MD, USA).

Measurement of DA concentrations in mouse serum.

We measured the peak intensity of DA in mice serum by targeted LC-MS, which was performed using the Thermo Scientific™ Prelude SPLC™ System (ThermoFisher Scientific, CA, USA) and a mass spectrometer TQH-Q1-0469 (ThermoFisher Scientific, CA, USA). Dodecanoic acid was identified based on the retention time and mass spectra of standard compounds of DA (61609; Sigma Aldrich, Estonia). We calculated DA concentrations in mouse serum based on the peak intensity of DA and the concentrations of standard compounds of DA.

Effects of DA on osteoclastogenesis and osteoblastogenesis

Differentiation of osteoclasts and TRAP staining in vitro.

The murine monocyte/macrophage cell line RAW264.7 (Chinese Academy of Sciences Cell Bank, Shanghai, China) as preosteoclasts were seeded in α-minimum essential medium (α-MEM; Corning, NY, USA) with 10% fetal bovine serum (Atlanta Biologicals, GA, USA) and 50 ng/ml receptor activator of nuclear factor-κB ligand (RANKL; R&D Systems, Minneapolis, MN, USA.) in 96-well plates. Cells were cultured for 5 days with different concentrations (0, 0.1, 1, 10, 100 μM) of DA. Culture medium was changed every other day, cells were fixed after 5 days, and TRAP (G1492; Solarbio, Beijing, China) staining was performed to observe and count TRAP-positive multinucleated (≥ 3 nuclei) cells.

Differentiation of osteoblasts and Alizarin Red staining in vitro.

A mouse monoclonal osteoblastic cell line (MC3T3-E1; American Type Culture Collection, Manassas, VA, USA), as preosteoblasts, were maintained in α-MEM with 10% fetal bovine serum (Atlanta Biologicals, GA, USA) and 1% penicillin/streptomycin (GE Healthcare, TX, USA) at 37°C in a humidified atmosphere with 5% CO_2. For osteogenic differentiation, MC3T3-E1 cells were plated in 12-well plates in growth medium until 100% confluence. Then, the medium was switched to differentiation medium (growth medium + 50 μg/ml L-(+)-ascorbic acid [Alfa Aesar, MA, USA] + 10 mM β-glycerophosphate [Alfa Aesar]) with different concentrations (0, 0.1, 1, 10, 100 μM) of DA. The medium was changed every other day for 21 days. At that time, MC3T3-E1 cells were treated with 10% formaldehyde for 30 minutes, followed by 3 washes with distilled water (5 minutes each), and stained with alizarin red S (Alfa Aesar, MA, USA) to evaluate extracellular matrix mineralization via quantifying areas of mineralized nodules.

Statistical analyses for in vivo and in vitro study

We performed 2-sample 1-sided t-tests to determine whether dietary supplementation with DA significantly increased serum DA concentrations and decreased bone mass reflected by μCT indices in mice in vivo. We performed 2-sided t-tests to determine whether dietary supplementation with DA produced significant changes in the numbers of osteoclasts and osteoblasts in mice in vivo. For multiple comparisons, 1-way analysis of variance (ANOVA) with Tukey’s test was utilized. All experiments were conducted at least 2 times and P-values < 0.05 was considered significant. Graph generation and statistical analyses were performed using Prism 8 software (GraphPad, La Jolla, CA, USA). To evaluate whether DA led to bone loss in vivo systematically when taking all the individual statistical evidence into consideration holistically, we performed 1-sided sign test (47) based on the changes shown by all the μCT indices for their effects on bone mass. Our null hypothesis, H0, was that dietary DA supplementation resulted in no change/increase in bone mass when all μCT bone indices were taken together. The alternative hypothesis, H1, was that dietary DA supplementation resulted in decreased bone mass when all μCT bone indices were taken together.

Validation of the causal association between DA and BMD by 2-sample MR analysis

We performed a 2-sample MR analysis to validate the causal relationship between DA and BMD in the direction of metabolite-to-BMD using published data from the largest GWAS meta-analysis (women only and pooled samples with both women and men, respectively) for BMD measures and the GWAS data of DA (48–50). This analysis may only be considered as partial validation since we were not able to obtain GWAS data for menopausal women only and the BMD measure in the published data was femoral neck BMD (FN-BMD), which is an important component of HTOT-BMD. We followed the same MR procedure as detailed earlier for the 1-sample MR analysis. When target SNPs were not available in the FN-BMD GWAS results, proxy SNPs in high linkage disequilibrium (LD, r² > 0.80) with the target SNPs based on the 1000 genomes project data were used (51).

Results

Basic characteristics of the study population

Table 1 shows basic characteristics of the study subjects, which included 517 Chinese women aged 41–64 years old. Sixteen percent of the subjects were in the perimenopausal phase, and 84% were in the early postmenopausal phase. Years since menopause (YSM) (P-value < 0.001) and follicle-stimulating hormone (FSH) levels (P-value < 0.001), which could reflect menopausal status, were both significantly associated with HTOT-BMD. Body mass index was also associated with HTOT-BMD (P-value < 0.001). In the studied population, alcohol drinking, calcium supplementation, regular exercise, physically demanding jobs, and annual family income were not significantly associated with HTOT-BMD (P-values > 0.05).

Table 1.

Basic characteristics of the study subjects (N = 517)

Phenotypes	Mean (SD) or No. (%)	β	P-value
Age (years)	52.85 (2.92)	-0.001	0.743
YSM (years)	1.96 (0.94)	-0.019	< 0.001
Weight (kg)	57.29 (7.73)	0.006	< 0.001
Height (cm)	157.90 (5.07)	0.003	0.005
BMI (kg/m2)	22.97 (2.87)	0.017	< 0.001
Calcium supplementation (%)	None (58.20%), Sometimes (26.90%), Often (14.90%)	0.001	0.878
Alcohol drinking (%)	28.20%	0.005	0.659
Smoking history (%)	0.00%		-
Regular exercise (%)	None (29.8%), < 2.5h/per week (9.3%), ≧ 2.5h/per week (60.9%)	0.007	0.251
Physically demanding jobs (%)	42.20 %	0.011	0.284
Family annual income (Yuan) (%)	< 36,000 (26.90%), 36,000~120,000 (42.70%), > 120,000 (30.40%)	0.007	0.326
FSH (mIU/ml)	76.24 (32.63)	-0.001	< 0.001
HTOT-BMD (g/cm2)	0.93 (0.12)	–	–
L1-L4 BMD (g/cm 2 )	1.05 (0.16)	–	–

P-value: significance of the association between each continuous/categorical covariates and HTOT-BMD (total hip BMD) in linear regression analysis. β-regression coefficient of the association between HTOT-BMD and each covariate. Calcium supplementation: None—never calcium supplementation; Sometimes—unregular calcium supplementation; Often—daily calcium supplementation.

Abbreviations: BMI, body mass index; FSH, follicle stimulating hormone; L1-L4, lumbar spine; mean (SD), means (standard deviation) for continuous variables; No. (%), percentages for discontinuous variable; YSM, years since menopause.

BMD-associated metabolites, metabolite modules, and causal metabolites

We identified a total of 381 metabolites with known identities that passed quality control. Partial least square regression identified 12 metabolites important to HTOT-BMD variation with VIP values ≥ 2.0 (Table 2). In the linear regression analysis, 8 metabolites were significantly associated with HTOT-BMD variations (FDR q-values < 0.05) (Table 2). These 20 BMD-associated metabolites included 4 fatty acids, 3 glycerophospholipids, 3 sterol lipids, 2 peptides, and 8 other organic compounds. The main reason for the lack of overlap in the results may be due to the relatively strict significance thresholds for each method (VIP values ≥ 2.0 in PLS or FDR q-values < 0.05 in the regression analyses). When we relaxed the thresholds slightly, for example, VIP values ≥ 1.0 and P-values < 0.05, we observed 6 overlapped metabolites between the 2 methods (the metabolites with P-value < 0.05 from linear regression are presented in Supplemental Table 2 (46)). Of the 6 metabolites, 3 metabolites (dodecanoic acid, D-Proline, myristic acid) were included among the 20 BMD-associated metabolites. Compared with models using traditional risk factors only, models that included the BMD-associated metabolites significantly improved predictive performance with respect to R² and RMSE (Table 3). Adding the PLS-derived metabolites into traditional risk factors increased R² by 1.66% and decreased RMSE by 0.22%.

Table 2.

Metabolites significantly associated with HTOT-BMD

Metabolites	Class	m/z	RT	VIP	β	q-value
2-Methyl-3-hydroxybutyric acid	Fatty acids	182.08	305.16	2.80	0.04	0.995
Dodecanoic acid	Fatty acids	199.17	40.28	1.86	-0.16	0.032
Myristic acid	Fatty acids	227.20	38.35	1.31	-0.21	< 0.001
Arachidonic acid	Fatty acids	303.23	193.35	0.71	0.23	< 0.001
PC(18:0/18:1(9Z))	Glycerophospholipids	810.60	155.26	2.40	0.01	0.995
PC(16:0/16:0)	Glycerophospholipids	734.56	161.48	2.32	0.02	0.995
LysoPC(18:0)	Glycerophospholipids	568.34	192.08	2.16	0.00	0.995
D-mannose	Organic compounds	145.05	287.73	2.00	-0.01	0.995
Methoxyacetic acid	Organic compounds	198.10	282.52	2.24	0.06	0.995
Uracil	Organic compounds	113.03	156.80	2.11	-0.02	0.995
L-carnitine	Organic compounds	323.22	325.98	2.01	-0.01	0.995
Alpha-D-glucose	Organic compounds	179.05	283.53	0.57	-0.16	0.030
Glutaraldehyde	Organic compounds	123.04	22.53	2.10	0.01	0.995
Paraxanthine	Organic compounds	217.02	25.36	0.51	-0.21	< 0.001
Glycolate	Organic compounds	135.03	326.78	2.09	-0.02	0.995
D-proline	Peptides	116.07	298.63	1.54	0.17	0.032
Phenylalanyl-cysteine	Peptides	310.13	280.97	0.74	-0.20	0.002
Glycocholic acid	Sterol lipids	464.31	175.03	2.71	-0.07	0.995
Cortisone	Sterol lipids	361.20	36.25	2.44	0.02	0.995
Glycochenodeoxycholate	Sterol lipids	414.30	202.59	0.09	0.20	0.002

Abbreviations: HTOT-BMD, total hip bone mineral density; β-regression coefficient of the association between metabolites and HTOT-BMD; m/z, mass-to-charge ratios; q-value, p-value adjusted by false discovery rate; RT, retention time; VIP, variable importance in projection.

Table 3.

Comparisons among different predictive models

Predictive Models	R-squared			RMSE
	Improvement of R-squared (95% CI)	Difference of R-squared (95% CI)	P-value	Improvement of RMSE (95% CI)	Difference of RMSE (95% CI)	P-value
Model 1: traditional risk factorsa	Reference	Reference	–	Reference	Reference	–
Model 2: Model 1+ PLS-derived scoreb	1.66%	3.45 × 10 –3 (1.65 × 10 –3 -5.24 × 10 -3 )	1.67 × 10 –4	-0.22%	-2.37 × 10 –4 (-4.42 × 10 -4 -3.12 × 10 –5 )	0.02
Model 3: Model 1 + linear regression analysis-derived scorec	0.88%	1.90 × 10 –3 (1.65 × 10–5-3.60 × 10–3)	4.79 × 10 –2	-0.12%	-1.25 × 10–4 (-3.30 × 10–4 -8.05 × 10–5)	0.23

Bolding signifies a p-value < 0.05.

Abbreviations: CI, confidence interval; RMAE, root mean squared error; PLS, partial least squares;

^a Including age, body mass index, regular exercise, alcohol drinking, calcium supplementation, physically demanding job, annual family income, and years since menopause.

^b Generated using 2-Methyl-3-hydroxybutyric acid, PC(18:0/18:1(9Z)), PC(16:0/16:0), LysoPC(18:0), D-mannose, Methoxyacetic acid, Uracil, L-carnitine, Glutaraldehyde, Glycolate, Glycocholic acid, Cortisone.

^c Generated using dodecanoic acid, Myristic acid, Arachidonic acid, Alpha-D-glucose, Paraxanthine, D-proline, Phe-cys, Glycochenodeoxycholate.

We constructed 26 co-abundance metabolite modules (Fig. 2A), 3 of which were significantly associated with HTOT-BMD (FDRs < 0.2). These included the green-yellow module (r = -0.10, P-value = 0.02), the brown module (r = -0.099, P-value = 0.03), and the blue module (r = 0.097, P-value = 0.03). In the green-yellow module (Fig. 2B), DA and myristic acid are also the BMD-associated metabolites identified in the individual metabolite analyses (Table 2). Table 4 showed that metabolites in the green-yellow module were significantly enriched in the pathways of “fatty acid biosynthesis” (P-value = 2.44 × 10^–5), “Biosynthesis of unsaturated fatty acids” (P-value = 3.78 × 10^–4), and “linoleic acid metabolism” (P-value = 0.022). There were no significantly enriched pathways identified in the brown and blue modules.

Figure 2.

Metabolite modules associated with HTOT-BMD constructed by WGCNA. (A) Clustering dendrograms of metabolites, with dissimilarity based on topological overlap together with assigned module colors. (B) The network of metabolites in the green-yellow module. The network reveals all the metabolites in green-yellow module. Dodecanoic acid and myristic acid were highlighted as they were also identified as significant BMD-associated metabolites in the analyses for individual metabolites (Table 2). Abbreviations: HTOT-BMD, total hip bone mineral density; WGCNA, weighted gene co-expression network analysis.

Table 4.

Pathway analysis for the metabolites in the green-yellow module

Pathways	Matched Metabolites	P-value	q-value
Fatty acid biosynthesis	Dodecanoic acid, Myristic acid, Palmitic acid, Capric acid	2.44 × 10–5	0.002
Biosynthesis of unsaturated fatty acids	Palmitic acid, Linoleic acid, Alpha-Linolenic acid	3.78 × 10–4	0.016
Linoleic acid Metabolism	Linoleic acid	0.022	0.63

Note: P-value—calculated by the pathway enrichment analysis which is used to identify the significant relevant pathways in the metabolites of green-yellow module; q-value, P-value adjusted by false discovery rate.

Mendelian randomization analysis revealed that DA (as the exposure) was causally negatively associated with HTOT-BMD (as the outcome) (β’s < -0.015, P-values < 0.05). Mendelian randomization–Egger (intercept) results showed no significant horizontal pleiotropy of the effect of DA on HTOT-BMD (P-value = 0.441). Furthermore, we did not observe a reverse causal relationship between HTOT-BMD (as exposure) and DA (as outcome) (P-values > 0.05) (Table 5). We also found that glycocholic acid (as the outcome) had a reverse causal relationship with HTOT-BMD (as the exposure) (β’s < -1.662, P-values < 0.05) (Table 5); this relationship was not significant in the direct metabolite-to-BMD (as exposure-to-outcome) analysis (P-values > 0.05). The lists of instrumental SNPs used in the 1-sample MR analyses are presented in Supplemental Tables 3–6 (46). The remaining 18 BMD-associated metabolites were not causally associated with BMD variation.

Table 5.

Examination of causal relationship between BMD-associated metabolites and HTOT-BMD using the bi-directional MR approach

Directions	MR Methods	β	Standard Error	P-value
DA to HTOT-BMD	Simple median	-0.015	0.005	0.001
	Weighted median method	-0.015	0.005	0.001
	IVW	-0.020	0.003	< 0.001
	MR-Egger	-0.024	0.007	< 0.001
	MR-Egger (intercept)	-0.003	0.004	0.441
HTOT-BMD to DA	Simple median	-0.828	0.766	0.280
	Weighted median method	-0.848	0.766	0.268
	IVW	-1.002	0.572	0.080
	MR-Egger	-1.334	2.022	0.509
	MR-Egger (intercept)	0.015	0.085	0.864
GA to HTOT-BMD	Simple median	-0.009	0.006	0.122
	Weighted median method	-0.009	0.006	0.121
	IVW	-0.008	0.004	0.065
	MR-Egger	0.006	0.014	0.677
	MR-Egger (intercept)	-0.005	0.005	0.324
HTOT-BMD to GA	Simple median	-1.906	0.733	0.009
	Weighted median method	-1.847	0.734	0.012
	IVW	-1.662	0.545	0.002
	MR-Egger	-4.493	1.835	0.014
	MR-Egger (intercept)	0.124	0.077	0.106

β is the estimated effect size of the causal association between exposure and outcome. DA to HTOT-BMD is DA (as exposure) and HTOT-BMD (as outcome); HTOT-BMD to DA is HTOT-BMD (as exposure) and DA (as outcome); HTOT-BMD to GA is HTOT-BMD (as exposure) and GA (as outcome); GA to HTOT-BMD is GA (as exposure) and HTOT-BMD (as outcome).

Abbreviations: DA, dodecanoic acid; GA, glycocholic acid; HTOT-BMD, total hip bone mineral density; IVW, inverse-variance weighted; MR, mendelian randomization.

In vivo effects of DA on bone metabolism in mice

Dodecanoic acid was found to be causally negatively associated with HTOT-BMD in the MR analysis, but a search of the literature failed to reveal any functional studies that investigated the role of DA in bone metabolism. In our results, DA supplemented normal mice had significantly increased serum DA concentrations compared with controls (121%, P-value = 0.001, Fig. 3A). Micro-computed tomography analysis of the structure of metaphyseal trabecular bone in the femur showed that DA tended to (though not significantly) decrease bone volume per tissue volume (BV/TV) (6.7%, P-value = 0.08, Fig. 3B), trabecular number (Tb.N) (1.5%, P-value = 0.34, Fig. 3C), trabecular thickness (Tb.Th) (2.4%, P-value = 0.21, Fig. 3D), and trabecular BMD (Tb.BMD) (1.12%, P-value = 0.18, Fig. 3E). Dodecanoic acid supplementation also tended to increase trabecular separation (Tb.Sp) (0.9%, P-value = 0.41, Fig. 3F). The 1-sided sign test, based on the changes shown by all the μCT indices for their individual effects on bone mass, showed that DA supplementation significantly decreased bone mass (P-value = 0.03) in normal mice. Furthermore, histomorphometric analysis showed that DA supplementation led to significant increases in both the numbers of osteoclasts (92.7%, P-value < 0.0001, Fig. 3G) and osteoblasts (85.3%, P-value = 0.002, Fig. 3H) in normal mice, which would be anticipated to increases bone turnover rates. Collectively, this evidence might support that dietary supplementation with DA could decrease bone mass in normal mice.

Figure 3.

Effects of DA on bone metabolism in normal WT mice (n = 6). WT mice nontreated (NT) or DA-treated (180 mg/kg/day) in drinking water for 8 weeks. (A) Serum DA concentrations in NT or DA-treated mice by liquid chromatography-mass spectrometry (LC-MS). (B) Bone volume per total volume (BV/TV, %). (C) Trabecular number (Tb.N, 1/mm). (D) Trabecular thickness (Tb.Th, mm). (E) Trabecular BMD (Tb.BMD (g/m³)). (F) Trabecular separation (Tb.Sp). (G) TRAP-stained osteoclast number per analyzed bone parameter (N.TRAP⁺/B.Pm(mm^-1)), and representative microscopic images of TRAP-stained osteoclasts in sections of distal femur (red arrows point to TRAP-stained osteoclasts). (H) ALP-stained osteoblast number per analyzed bone parameter (N.ALP⁺./B.Pm(mm^-1), and representative microscopic images of ALP-stained osteoblasts in sections of distal femur. Red arrows point to ALP-stained osteoblasts. (* P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 1.0 × 10^–4). Abbreviations: ALP, alkaline phosphatase; BMD, bone mineral density; B.Pm, bone parameter; BV/TV, bone volume/tissue volume; DA, dodecanoic acid; LC-MS, liquid chromatography-mass spectrometry; N.ALP, ALP-stained osteoblast number; NT, nontreated; N.TRAP, TRAP-stained osteoclast number; Tb.BMD, trabecular BMD; Tb.N, trabecular number; Tb.Sp, trabecular separation; Tb.Th, trabecular thickness; TRAP, tartrate-resistant acid phosphatase; WT, wild-type.

In addition, we explored the effects of DA on BMD in OVX mice. The BMD of OVX mice was significantly decreased compared with normal mice (Supplemental Figure 1) (46), supporting that the OVX caused bone loss in the studied mice. Dodecanoic acid-supplemented OVX mice had significantly increased serum DA concentrations compared with OVX controls (141%, P-value = 0.002, Fig. 4A). Micro-computed tomography analysis of the structure of metaphyseal trabecular bone in the femur showed that DA supplementation tended to decrease BV/TV (7.6%, P-value = 0.14, Fig. 4B), Tb.N (5.9%, P-value = 0.19, Fig. 4C), Tb.Th (7.5%, P-value = 0.053, Fig. 4D), and Tb.BMD (2.0%, P-value = 0.04, Fig. 4E). Dodecanoic acid supplementation also tended to increase Tb.Sp (9.2%, P-value = 0.14, Fig. 4F). When taking all the indices together, DA supplementation significantly promoted bone loss (P-value = 0.03) via a 1-sided sign test in OVX mice. Histomorphometric analysis also revealed that OVX mice receiving a dietary DA supplementation had increased numbers of osteoclasts (53.8%, P-value = 0.06, Fig. 4G) and osteoblasts (63.7%, P-value = 0.009, Fig. 4H), suggesting that DA treatment might enhance bone turnover rates in OVX mice. The detailed results of mice study are presented in Supplemental Table 7 (46). These collective results might support the conclusion that dietary DA supplementation could increase the numbers of osteoclasts and osteoblasts, enhance bone turnover rates and promote bone loss in both normal and OVX mice.

Figure 4.

Effects of DA on bone metabolism in OVX WT mice (n = 6). OVX mice nontreated (NT) or DA-treated (180 mg/kg/day) in drinking water for 8 weeks. (A) Serum DA concentrations in NT or DA-treated OVX mice by LC-MS. (B) Bone volume per total volume (BV/TV, %). (C) Trabecular number (Tb.N, 1/mm). (D) Trabecular thickness (Tb.Th, mm). (E) Trabecular BMD (Tb.BMD, g/m³). (F) Trabecular separation (Tb.Sp). (G) TRAP-stained osteoclast number per analyzed bone parameter (N.TRAP⁺/B.Pm(mm^-1)), and representative microscopic images of TRAP-stained osteoclasts in sections of distal femur (red arrows point to TRAP-stained osteoclasts). (H) ALP-stained osteoblast number per analyzed bone parameter (N.ALP⁺./B.Pm(mm^-1), and representative microscopic images of ALP-stained osteoblasts in sections of distal femur (red arrows point to ALP-stained osteoblasts). (* P < 0.05, ** P < 0.01). Abbreviations: ALP, alkaline phosphatase; BMD, bone mineral density; B.Pm, bone parameter; BV/TV, bone volume/tissue volume; DA, dodecanoic acid; LC-MS, liquid chromatography-mass spectrometry; N.ALP, ALP-stained osteoblast number; NT, nontreated; N.TRAP, TRAP-stained osteoclast number; OVX, ovariectomized; Tb.BMD, trabecular BMD; Tb.N, trabecular number; Tb.Sp, trabecular separation; Tb.Th, trabecular thickness; TRAP, tartrate-resistant acid phosphatase; WT, wild-type.

In vitro effects of DA on osteoclastogenesis and osteoblastogenesis

Based on the in vivo effects of dietary supplementation with DA observed above, we further examined the effects of DA on osteoclast and osteoblast differentiation in vitro. Treatment with DA (1 and 10 μM) significantly increased the number of mature osteoclast-like cells compared with controls; 0.1 and 100 μM DA did not induce significant changes. A treatment of 10 μM DA had the greatest effects on osteoclast differentiation (125.5% increase, P-value = 0.01, Fig 5A and 5B). Osteoclast fusion was also enhanced at higher concentrations of DA (eg, 10, 100 μM), manifested by increased numbers of osteoclast-like cells with > 10 nuclei (Fig. 5B).

Figure 5.

Effects of DA on osteoclastogenesis and osteoblastogenesis. (A) Quantification of TRAP-positive osteoclast-like cells (multinuclear cells [MNCs]) induced from RAW264.7 cells after 5 days of osteoclastogenesis with/without DA treatment at indicated concentrations. (B) Microscopic images of TRAP staining of osteoclast-like cells with/without DA treatment at indicated concentrations (black arrows indicate osteoclast-like cells with 3–10 nuclei; red arrows indicate osteoclast-like cells with more than 10 nuclei). (C) Quantification of extracellular matrix mineralization (mineralized nodules area (%)) by MC3T3-E1 cells after 14 days of osteoblastogenesis with/without DA treatment at indicated concentrations. (D) Microscopic images of alizarin red S staining for extracellular matrix mineralization with/without DA treatment at indicated concentrations. (* P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 1.0 × 10^–4). Abbreviations: DA, dodecanoic acid; MNCs, multinuclear cells; TRAP, tartrate-resistant acid phosphatase.

Compared with untreated controls, extracellular matrix mineralization was significantly enhanced in MC3T3-E1 cell cultures treated with 1 μM DA (338% increase, P-value < 1.0 × 10^–4, Fig. 5C and 5D), indicating increased differentiation of MC3T3-E1 into mature osteoblasts. Treatments of 0.1, 10, and 100 μM DA did not induce significant changes. Thus, DA treatment differentially enhances differentiation of osteoblasts and osteoclasts in vitro at different concentrations. These results are consistent with the increased numbers of osteoclasts and osteoblasts detected in vivo in normal and OVX mice whose diets were supplemented with DA.

Validation of the association between DA and BMD in 2-sample MR analysis

Two MR methods (simple median and weighted median) revealed that DA (as the exposure) was causally negatively associated with FN-BMD (women only, as the outcome) (β’s < -0.666, P-values < 0.05) (Supplemental Table 8) (46). The results from inverse-variance weighted and MR-Egger methods were not significant. The result of the MR-Egger (intercept) method showed no significant horizontal pleiotropy of the effect of DA on FN-BMD (P-value = 0.876). These findings partially support the conclusion from our population study that DA may causally decrease BMD. We could not observe any significant results (Supplemental Table 8) (46) using the GWAS data from the pooled population (both women and men), which might be attributable to the fact that our study was focused on women. The lists of instrumental SNPs used in the 2-sample MR analyses are presented in Supplemental Tables 9 and10 (46).

Discussion

This study demonstrated that metabolic profiles in perimenopausal and early postmenopausal Chinese women varied substantially among individuals with different BMD levels. We generated a list of 20 BMD-associated metabolites that showed potential for improving the risk prediction, understanding the pathogenesis, and providing new targets for preventative and therapeutic treatment for PMOP (Table 2). Of these 20 metabolites, DA is a strong causality candidate for decreasing HTOT-BMD, as determined by an MR analysis. Subsequent functional studies suggested that dietary supplementation with DA might induce bone loss and increase the number of osteoblasts and osteoclasts in vivo in both normal and OVX mice. Additionally, in vitro experiments suggested that treatment with DA might differentially promote differentiation of preosteoblasts and preosteoclasts into mature forms at different concentration. The collective results highlight the potential importance of fatty acid metabolism in regulating BMD.

There is accumulating evidence that supports the coexistence of disordered lipid metabolism and bone dysfunction (52). Twelve of the BMD-associated metabolites identified in this study are lipids or their derivatives, including fatty acids, glycerophospholipids, and sterol lipids. The significant association between fatty acids and BMD is particularly noted in this study. For example, the associations between arachidonic acid (AA) and BMD has been controversial. Dietary intake of AA was previously reported to be positively associated with BMD in postmenopausal women (53), which is consistent with our results. However, another metabolomics study of 104 pre menopausal and 260 postmenopausal women reported that AA was negatively associated with BMD (18). Our study included peri- and postmenopausal women who would be expected to have lower estrogen levels than premenopausal women. Consequently, these different observations suggest that estrogen levels may impact the effects of AA on BMD. There have also been inconsistent results regarding the effects of AA on bone metabolism in in vitro studies. For example, Coetzee et al demonstrated that AA (at 67 μM) inhibited osteoprotegerin secretion and stimulated RANKL secretion, suggesting that AA increased osteoclastogenesis (54). However, another study reported that AA (at 80 μM) inhibited osteoclast differentiation and activity by regulating the expression of essential osteoclast-specific genes (55). The potentially important role of AA in bone metabolism, in conjunction with conflicting and inconsistent findings from a small number of studies, provides significant motivation for further studies.

Dodecanoic acid and myristic acid, which are both medium-chain saturated fatty acids, were significantly associated with BMD in our study. Dodecanoic acid and myristic acid are also within the co-functional metabolite module (green-yellow module), and the metabolites in this module were significantly enriched in the pathways of “fatty acid biosynthesis,” “biosynthesis of unsaturated fatty acids,” and “linoleic acid metabolism.” These results suggest that multiple fatty acid pathways contribute to the regulation of bone metabolism. Several other metabolites in this module (eg, capric acid (56), palmitoleic acid (57), and myristic acid (58)) are also involved in the pathway of “fatty acid biosynthesis” and were reported to be related to bone metabolism. This metabolite module may be significant to explore further on how fatty acids regulate the biological process of bone metabolism.

Dodecanoic acid was particularly interesting because it appeared to causally decrease HTOT-BMD, as shown by MR analysis. Previous reports have also demonstrated that DA is associated with a variety of pathophysiological conditions. For example, DA prevented testosterone-induced prostatic hyperplasia in rats (45). Dodecanoic acid contributed to energy balance regulation and insulin resistance by activating toll-like receptor 4 signaling (59). Interestingly, DA itself can regulate fatty acid homeostasis via peroxisome proliferator-activated receptors (60), which have been shown to regulate cell development and metabolism of both osteoclasts and osteoblasts (61). We are not aware of any previous functional studies that directly explored the relationships between DA and bone metabolism. In the current study, we observed that dietary supplementation with DA might cause bone mass trends and increase the numbers of osteoclasts and osteoblasts in both normal and OVX mice. These findings might support our MR analysis and provide a potential mechanism. The increased rate of bone turnover associated with estrogen deficiency may be due to the increased production of osteoclasts and osteoblasts (62). The increased rate of bone turnover caused by DA is consistent with that induced by estrogen deficiency after menopause. Estrogen deficiency appears to increase fatty acid biosynthesis rates in the livers of rats (63), and DA is an end product of fatty acid biosynthetic pathway. Therefore, it is reasonable to speculate that estrogen deficiency leads to increased DA biosynthesis, which contributes to bone loss associated with estrogen deficiency.

In OVX mice receiving DA supplementation, osteoblast numbers were increased to a greater extent than osteoclasts. Intuitively, one would expect the greater relative increase of osteoblasts vs osteoclasts in DA supplemented OVX mice to increase bone formation. In contrast to expectations, however, DA supplementation led to bone loss in OVX mice. The amount of bone resorbed or formed by a group of osteoclasts and osteoblasts is dependent on both the total cell number and the functional activity of individual cells (62). The osteoblasts identified by ALP staining in the histomorphometric analysis include both osteoblast progenitors (64) and mature osteoblasts. However, the amount of bone matrix and its mineralization, the major determinant of BMD, mainly depends on mature osteoblasts and their functional activities (65). Furthermore, previous studies suggested that estrogen deficiency increased the number of osteoblast progenitors in murine bone marrow (66) and shortened the working lifespan of osteoblasts (67). On the other hand, a delay of osteoclast apoptosis associated with estrogen deficiency seems to be responsible for the deeper resorption and trabecular perforation (68). This collective evidence may partially explain increased bone loss in DA supplemented OVX mice, despite a relative increase in the number of osteoblasts vs osteoclasts.

The in vitro differentiation of osteoclasts and osteoblasts was both significantly promoted by 1 μM DA treatment, whereas only osteoclast differentiation was significantly promoted at the higher concentrations of DA (eg,10 μM). Moreover, DA (eg, 10, 100 μM) enhanced osteoclast fusion which produced multinucleated cells (69). Multinucleation can cause giant osteoclast formation, which is more likely to cover a relatively large matrix area, potentially leading to increased bone resorption efficiency (70). The results of our in vitro experiments indicate that DA can stimulate both bone formation and bone reabsorption, similar to some other factors such as growth hormone and insulin-like growth factor (71). However, increasing DA concentrations could favor osteoclast differentiation and bone reabsorption, leading to the bone loss observed in our in vivo experiments. These findings suggest the potentially dynamic modulatory roles of DA in bone regulation, which may depend on its concentrations. Figure 6 shows that the potential functional mechanism of DA on bone metabolism furnished in this study; however, it was only a preliminary functional study, which warrants further work in the future.

Figure 6.

The potential functional mechanism of DA on bone metabolism furnished in this study. Estrogen deficiency leads to increased DA biosynthesis by the pathway of “fatty acid biosynthesis,” and DA promotes the differentiations of osteoclasts and osteoblasts, which contributes to bone loss. Abbreviation: DA, dodecanoic acid.

As would be expected for the role of a single molecule in a complex trait, DA did not show a large effect on bone traits. Together with the limited sample size and the relatively large variations within each mice group, these may partially explain why many individual μCT indices showed bone loss trends but were not statistically significant. However, it is very clear that all the indices consistently showed a trend for bone loss caused by DA. We performed functional experiments as a demonstration and proof-of-concept for the results of metabolomics analysis to demonstrate that the novel findings can be identified based on the hypothesis free untargeted metabolomic discovery analyses. We also partially validated that DA was causally negatively associated with BMD by the 2-sample MR analysis using published data, which further supports that DA may lead to decreased BMD. Our findings, in conjunction with results from previous studies in which fatty acids (such as AA, capric acid) impact bone metabolism, provides a strong foundation for the concept that the biosynthesis and metabolism of fatty acids have complex regulatory effects on bone metabolism.

In addition to fatty acids, several other lipids were associated with BMD variation in the current study. For example, MR analysis found that BMD status might causally influence the level of glycocholic acid (secondary bile acids), which demonstrated an interaction between BMD and metabolic profiles in the direction from BMD variation (or associated conditions) to metabolomic profile. Abnormal bile acid metabolism has been linked with osteoporosis by influencing the intestinal absorption of Ca²⁺ and lipid-soluble vitamin D (72). Furthermore, our results first reported that phenylalanyl-cysteine (phe-cys), a dipeptide, was associated with HTOT-BMD. There is very limited knowledge about the roles of phe-cys in bone regulation, and further studies, similar those reported here for DA, are warranted.

The present study has several strengths. First, the fact that several of the metabolites (eg, AA, myristic acid, and bile acids) we identified have been previously linked to bone metabolism enhances our confidence in these findings, and provides motivation to further explore the other metabolites we identified. Second, the identification of some novel BMD-associated metabolites (such as phe-cys and DA) provide new knowledge for PMOP. The identified metabolites improve predictive performance compared to models using conventional risk factors alone. Third, the BMD-associated co-functional metabolite module, consisting of fatty acids, was identified for the first time via WGCNA. Finally, it was found that DA would causally decrease HTOT-BMD through an innovative multiomics analysis (genomics derived from whole genome sequencing and metabolomics). In contrast to the earlier metabolomic studies in humans, we further performed preclinical types of experiments both in vitro in relevant cell models and in vivo in mice, as a proof-of-concept, to demonstrate the potential role of DA in the pathogenesis of osteoporosis. However, our study does have some limitations. First, we did not replicate the results of BMD-associated metabolites because of lack of samples from the population with similar/same genetic and environmental backgrounds. Second, we did not record subjects’ diets (one major determinant of human blood metabolites) and their clinical biomarkers of bone metabolism such as PINP and CTX-I. Third, we did not explore the mechanism of the effects of DA on the differentiation of osteoblasts and osteoclasts. Therefore, further research efforts are still needed to clarify the role of DA in human bone homeostasis and its related mechanisms.

Conclusion

This study used a novel metabolomics approach, followed by functional studies, to assess metabolic changes associated with BMD variation in perimenopausal and early postmenopausal Chinese women, and the potential mechanism of action. Twenty BMD-associated metabolites were identified. Our collective results suggested that fatty acids and their metabolic pathways, appeared to be particularly important in bone health. These findings provided novel insights into new biomarkers (eg, DA), and potential pathophysiological mechanisms (eg, fatty acids biosynthesis) for osteoporosis, especially for PMOP.

Additional Information

Disclosure Summary: The authors have nothing to disclose

Data Availability

The data that support the findings of this study are available from the corresponding author upon request and approval of the team and respective institutions.