應用核磁共振氫譜技術探討台灣女性代謝體特徵與低骨質密度之關聯研究

國立臺灣大學公共衛生學院流行病學與預防醫學硏究所

碩士論文

Institute of Epidemiology and Preventive Medicine College of Public Health

National Taiwan University Master Thesis

應用核磁共振氫譜技術探討台灣女性代謝體特徵 與低骨質密度之關聯研究

Association of Metabolome with Low Bone Mineral Density in Taiwanese Women Revealed

by

¹

H NMR Spectroscopy

尤瀅淑 Ying-Shu You

指導教授：程蘊菁 博士 林靖愉 博士

Advisors: Yen-Ching Chen, Sc.D.

Ching-Yu Lin, Ph.D.

中華民國 101 年 7 月

July, 2012

i

ii

誌謝

本篇碩士論文能順利完成，背後最重要的推手莫過於指導教授程蘊菁博士。

在程老師不辭辛苦與悉心指導之下，才得已讓學生能完成此艱鉅的任務，在此相

當感謝老師辛勤的付出，很榮幸能擔任程老師的指導學生。另外亦感謝共同指導

教授林靖愉博士給予許多研究上的協助和建議，以及林老師的研究助理梁皓然學

長耐心地教導我操作代謝體學的分析方法，使學生獲益良多。同時也感謝台大北

護分院蔡克嵩院長、中研院統計所副所長丘政民教授，在百忙之中仍能撥冗前來

指教學生的論文研究，並給予中肯的建議。

在進行代謝體學實驗的期間裡，多虧了中研院生醫所核磁共振實驗室裡的研

究員和同仁們，總是熱心又親切地幫助我解決許多難題，才能使我順利收集實驗

資料供研究分析。再者，整個碩士班修業期間，也很感謝程老師研究團隊和同屆

的醫師同學們，大家能一起互助共進。

最後，感謝我親愛的家人和朋友們，不斷給予許多支持和鼓勵！感謝流預所

各位老師們諄諄教導！能建立此基石，皆歸功於所有在我身旁無限供給能量予我

的每一個人，謝謝您們。

尤瀅淑 謹誌 2012.8

iii

摘要

背景：近期已有研究指出骨質疏鬆症與體內不斷循環的特定代謝物具有相關；然

而，若單靠過去研究所探討的少數幾項代謝物，很難對此複雜的致病機轉做詳確

的解釋。如今，代謝體學方法可全面性的檢視人體中代謝情形和潛在的臨床意義。

截至目前為止，未有研究加以驗證人體中代謝體特徵和骨質密度之間的相關。

方法：本研究橫斷式研究設計，自 2009 到 2010 年間招募 610 位年齡介於 40 至

55 歲之間在台北美兆健康檢查中心參加健康檢查的台灣女性。使用雙能量放射線

儀測量隻腰椎骨質密度，以骨質密度值的高低排序後研究族群依人數均分為三等

份，定義骨質密度最高的第二和第三分群為「高骨質密度」組別，而最低的第一

群為「低骨質密度」組別。而血漿代謝體特徵資料是透過核磁共振氫譜實驗分析

所收集而成。運用無監督的主成分分析、監督的偏最小二乘法判別分析法及邏輯

斯迴歸模式，來分析代謝體特徵與骨質密度的相關。

結果：研究結果顯示，無監督的主成分分析的分數散佈圖無法完善地分離高低骨

質密度的組別；監督的偏最小二乘法判別分析法亦沒有能力區別高低骨質密度的

組別。根據停經狀態加以分層之後，可透過偏最小二乘法判別分析法於停經婦女

群體觀察到其代謝體特徵與高低骨質密度顯著的差異(R²= 0.12；Q²=0.04；

Ppermutation=0.03)。此外，在停經婦女身上，會隨著 glutamine 的濃度升高，使得低

骨質密度的風險增加；同時，低骨質密度也與 lactate、acetone、lipid 和 very low

density lipoprotein 濃度降低具有相關。

iv

結論： 這是第一篇研究透過核磁共振氫譜實驗所建立的代謝體學與骨質密度之

研究，我們成功地於停經婦女群體辨識出一群代謝物和低骨質密度具有關聯。其

中，特別是 glutamine 的濃度升高會增加低骨質密度發生的風險，可能會因此導

致骨骼流失。

關鍵詞：代謝體特徵、代謝體學、核磁共振氫譜、骨質疏鬆症、停經、女性

v

Abstract

Background. Osteoporosis has been related to the alteration of specific circulating

metabolites previously. However, studies with few metabolites may have difficulty to

explain the pathogenesis of this complex syndrome. Metabolome provide an overview

of the metabolism status in human body and potential clinical implication. Up to date,

no study has investigated the association between metabolome and bone mineral

density (BMD).

Methods. This is a cross-sectional study. A total of 610 healthy Taiwanese women

aged 40 to 55 were recruited from MJ Health Screening Center between 2009 and

2010. High and low bone mineral density (BMD) was defined as the 2^nd plus 3^rd

tertiles and the 1^st tertile of BMD, respectively. The plasma metabolome were

evaluated by ¹H-nuclear magnetic resonance spectroscopy. Principal components

analysis (PCA), partial least-squares discriminant analysis (PLS-DA), and logistic

regression model were used to assess the association between metabolome and BMD.

Results. The unsupervised PCA showed no visual separation between low and high

BMD levels; the supervised PLS-DA model was also unable to distinguish high and

low BMD groups. After stratification by menopausal status, high and low BMD

groups can be differentiated in post-menopausal women using PLS-DA (R²= 0.12;

vi

Q²=0.04; Ppermutation=0.03). In addition, elevated level of glutamine was associated with

the risk of low BMD among postmenopausal women; while low BMD is characterized

by decreased levels of lactate, acetone, lipid, and very low density lipoprotein in the

same sub-population.

Conclusion. This is the first study using ¹H NMR-based metabolomic approach and

successfully identified a group of metabolites representative for postmenopausal

women with low BMD. Especially, elevated level of glutamine may lead to bone loss

via increased risk of low BMD.

Keywords. Metabolome, metabolic profiles, metabolomics, metabonomics, nuclear magnetic resonance, NMR, bone mineral density, osteoporosis, menopause, women.

vii

目錄

審定書 ... i

誌謝 ... ii

摘要 ... iii

Abstract ...v

圖目錄 ... viii

表目錄 ... ix

Chapter 1. Introduction ...1

1.1 Importance of Osteoporosis ... 1

1.2 Bone-Related Proteins and Metabolites ... 2

1.3 NMR-based Metabolomics ... 2

1.4 Aims ... 3

Chapter 2. Materials and Methods ...5

2.1 Study Population ... 5

2.2 Measurement of Bone Mineral Density ... 6

2.2 Sample Collection and Preparation ... 6

2.3 1D ¹H NMR Spectroscopy ... 6

2.4 NMR Spectral Pre-Processing ... 7

2.5 Statistical Analyses ... 8

Chapter 3. Results ... 11

3.1 Characteristics of the Study Population ... 11

3.2 Metabolome of the BMD Levels ... 11

3.3 Candidate Metabolites and the Risk of Low BMD ... 12

Chapter 4. Discussion ...15

4.1 Main findings ... 15

4.2 Evidences of Metabolites and Bone Metabolism ... 15

4.3 Strengths and limitations ... 17

4.4 Conclusions ... 19

References ...20

viii

圖目錄

Figure 1. Flowchart of participant recruitment ... 30 Figure 2. The tertiles of bone mineral density ... 31 Figure 3. PCA score plots from the analysis of CPMG NMR spectra using women

plasma samples ... 32 Figure 4. PLS-DA score plots from the analysis of CPMG NMR spectra using women

plasma samples ... 33 Figure 5. Receiver operating characteristic curves of comparing models for

classification of high and low BMD ... 34 Figure 6. Acetone distribution by fasting glucose level ... 35 Figure 7. Postulated mechanism relates important metabolites with bone mineral

density among postmenopausal women ... 36

ix

表目錄

Table 1. Characteristics of the study population ... 37 Table 2. The PLS-DA parameters and permutation test for differentiating high and low

BMD levels ... 38 Table 3. The change of plasma metabolites in postmenopausal women to distinguish

high and low BMD ... 39 Table 4. Association between plasma metabolites and bone mineral density (T1 vs.

T2+T3) ... 40 Table 5. Association between plasma metabolites and bone mineral density stratified

by menopausal status (T1 vs. T2+T3) ... 41 Table 6. Receiver operating characteristic contrast tests of pairwise comparison

between different models to classify high and low BMD ... 42 Table 7. Model comparisons for the association between plasma metabolites and bone

mineral density (T1 vs. T2+T3) ... 43

1

Chapter 1. Introduction

1.1 Importance of Osteoporosis

Osteoporotic fractures are a major cause of morbidity, disability, and the

subsequent premature death in the elderly.¹According to World Health Organization

criteria, osteoporosis is defined as bone mineral density (BMD) is 2.5 standard

deviations (SD) below the mean. In the United States, spinal osteoporosis was more

prevalent among Mexican American (24.4% for women and 4.6% for men) than that

among either non-Hispanic blacks (5.3% for women) or non-Hispanic whites (10.9%

for women and 2.2% for men) aged 50 years or older in National Health and Nutrition

Examination Survey (NHANES) from 2005 to 2008.² In the 2005 to 2008 Nutrition

and Health Survey in Taiwan (NAHSIT 2005-2008), the prevalence of osteoporosis at

lumbar spine is 12.6 % in women and 4.3% in men aged 50 years or older.³

Furthermore, BMD testing, impaired vision, and neuromuscular deficits before age 65

are recommended for predicting the risk of future fracture⁴, which has known to

associated with high mortality. As the population aging fast worldwide, osteoporosis

has become an important public health issue.

2

1.2 Bone-Related Proteins and Metabolites

Decreased bone mass occurs when bone resorption is excessive or bone

formation is decreased. Bone turnover markers (e.g., alkaline phosphatase, osteocalcin,

type I collagen cross-linked N-telopeptide, hydroxyproline and the pyridinium

crosslinks, etc.) have been used to monitor bone homeostasis and predict fracture

risk.^5,6 However, 20% to 30% of women with high bone turnover rate may be

misclassified as osteoporosis patients in the future.⁷ On the other hand, some studies

found that elevated homocysteine, which interferes with the cross-linking of collagen

in bone, may predict osteoporotic fractures.^8-10 However, the association between

homocysteine and BMD has been inconsistent.^8,9,11,12 Moreover, in vitro and in vivo

studies showed that nuclear factor kappa-light-chain-enhancer of activated B cells

(NF-κB) essential modulator -binding domain (NBD) peptide¹³ and polyunsaturated

fatty acids¹⁴ inhibit osteoclastogenesis for bone loss. In summary, past studies only

explored few metabolites and thus may have difficulty on predicting the complex

syndrome, osteoporosis.

1.3 NMR-based Metabolomics

Metabolomics based on Nuclear Magnetic Resonance spectroscopy (NMR)

3

spectrometry allow the detection of a wide range of structurally diverse metabolites

simultaneously as well as metabolome for the discovery of novel biomarkers.¹⁵ The

NMR spectrum is made up of peaks from each chemically distinct hydrogen atom in

the molecule, and the peak position of a given hydrogen on the frequency axis is

known as a chemical shift [unit: parts per million (ppm), often termed a δ value] from

that of a reference substance.^15-17 In addition, NMR is a robust and reliable technique

for metabolomic applications with high reproducibility.^15,18,19 The NMR-based

spectroscopic method usually uses urine or blood plasma/serum as the primary sample

sources of metabolic fingerprint data.^20,21 Generally, as compared with urine, blood

sample has less pH variations, dietary or drug effects.^15,22 Therefore, NMR-based

metabolomics gradually become an important tool in predicting metabolites for

different research purpose, e.g., drugs, diet, lifestyle, environment, stimuli and genetic

modulations.¹⁵

1.4 Aims

Previous studies evaluated limited number of candidate metabolites, which is

hard to address the overview of the metabolism status in human body. Up to date, the

association between metabolome and BMD has only been explored in ovariectomized

4

rats.^23,24 Therefore, this study adopted the high-throughput ¹H NMR-based

‘metabonomic/metabolomic’ approach^20,21 to assess the association between plasma

metabolome and BMD in healthy middle-aged women. In addition, because estrogen

deficiency-induced bone loss is observed in postmenopausal women, who are more

susceptible to fragility fractures than premenopausal women,²⁵ this study further

accessed the effect modification by menopausal status.

5

Chapter 2. Materials and Methods

2.1 Study Population

This was a cross-sectional study. A total of 773 Taiwanese women aged 40 to 55

were recruited from MJ Health Screening Center, Taipei, Taiwan, between October

2009 and August 2010. Each participant filled out a self-reported questionnaire and

provided a blood sample. The outcome of this study was BMD (g/cm²) at lumbar spine.

Participants with the following conditions or diseases were excluded (n=163): (1)

received hormone replacement therapy (n=34) or other medications (e.g., steroid, n=3)

that may affect BMD, (2) lack of BMD at lumbar spine (n = 51), (3) lack of blood

samples (n=45), (6) or lack of NMR data (n=30). A total of 610 women were included

for data analyses (Figure 1). Informed consent was obtained from each participant.

The study protocol has been approved by the institutional review boards of MJ Health

Screening Center and College of Public Health, National Taiwan University.

A self-report questionnaire was administered to collect information on

demography, life style (e.g., smoking, alcohol consumption, calcium supplement, and

regular exercise), menopausal status, disease history (e.g., hypertension and diabetes),

and medication history (e.g., steroid, anti-inflammatory drugs, Chinese herb, and

gastrointestinal drugs).

6

2.2 Measurement of Bone Mineral Density

BMD (g/cm²) was measured at the lumbar spine by using dual-energy X-ray

absorptiometry (DXA, GE Lunar Health Care, DPX-L, USA), which was calibrated by

a standard automated test program provided by the manufacturer. BMD was tertiled

(T1, T2, and T3) based the whole population. High BMD was defined as T2 plus T3

(reference group) and low BMD was defined as T1 (comparison group, Figure 2).

2.2 Sample Collection and Preparation

Fasting blood samples were collected in tubes containing sodium ethylene

diamine tetraacetic acid (EDTA) from each participant. After centrifugation, plasma

samples were stored in a −80°C freezer. Before NMR procedures, 200 µl of plasma

was diluted with a ratio of 1:2 in physiological saline [0.9% (w/v) NaCl solution] with

10% D₂O. Solution in Eppendorf tubes were then centrifuged at 15,000g for 5 min at

4°C, and 550 µl of supernatant was transferred into 5 mm NMR tubes.

2.3 1D

¹

H NMR Spectroscopy

All NMR spectra were acquired at 500.13 MHz using Avance-500 spectrometers

(Bruker, Fremont, CA) at a ¹H frequency with 300 K internal probe temperature at

7

High Field Nuclear Magnetic Resonance Center in Academia Sinica, Taiwan.

For the plasma samples, Carr-Purcell-Meiboom-Gill (CPMG)-presat pulse

sequence (relaxation delay-90°-(t -180°-t)n -90°-acquired-free induction decay) was

acquired from 1D ¹H NMR analysis. For each sample, 128 scans were collected into

64K computer data points using a spectral width of 10,000 Hz (20 ppm), with a total

echo time of 64 ms (t= 400 µs) during the relaxation delay (RD, 2.0 s) and an

acquisition time of 1.63 s.¹⁵ Additionally, all 1D spectra were applied for analysis

before Fourier transformation with zero-filled to exponential line-broadenings of

0.5 Hz. The acquired NMR spectra were phased, baseline-corrected, and then

calibrated by the methyl lactate doublet (δ 4.12 ppm) using TopSpin 2.0 (Bruker

Biospin Ltd.).

2.4 NMR Spectral Pre-Processing

After acquisition of the NMR data, each spectrum was then segmented into

1,880 chemical shift bins between δ 0.2 and δ 10.0, corresponding to a bin width of

0.005 ppm (2.5 Hz), using custom-written ProMetab software Version 3.3 in

MATLAB (Version 7.0.1, The MathWorks, Natick, MA).²⁶ Following the removal of

the residual water (δ 4.50-6.00 ppm) and EDTA (δ 3.58-3.65 ppm and δ 3.17-3.23

ppm),²⁷ the area within each spectral bin was integrated to yield a 1×1572 vectors

8

containing intensity-based descriptors of the original spectrum. The total spectral area

of the remaining bins was normalized to unity to facilitate comparison between the

spectra. The binned data was subject to the generalized log transformation

(λ = 8 × 10⁻⁹),²⁸ and the columns were mean-centered by Pareto scaling^29,30 before

multivariate analysis.

2.5 Statistical Analyses

The reduced and normalized NMR spectral binned data were uploaded to

Metaboanalyst 2.0

(

http://www.metaboanalyst.ca

)

³¹ for principal components analysis

(PCA), an unsupervised approach, and partial least-squares discriminant analysis

(PLS-DA), a supervised method, to identify which metabolites contributed to specific

clusters. PCA is used to reduce high-throughput metabolomic data into principal

components without referring to the BMD level, and then the low-dimensional PCA

score plots were visually inspected for natural separation of the two BMD levels (low

vs. high, Figure 2). In contrast, PLS-DA uses previous knowledge about the BMD

level during the classification process by multiple linear regression technique to find

the direction of maximum covariance between metabolome and the BMD level. The

PLS-DA score plots were used to assess if the metabolome can be separated based on

BMD levels. Furthermore, PLS-DA was determined by the number of latent variables

9

below to build the model. A default 10-fold internal cross validation was employed,

from which Q² (expressing the cross-validated predictive capability), R² (explained

variance) and variable importance in projection (VIP) score (coefficient reflects the

relative importance of each variable) were used to extract important plasma

metabolites (threshold>1.5) in the PLS-DA model.

Student’s t test (for normally-distributed continuous variables) and Pearson’sχ²

test (for categorical variables) were performed to compared the baseline distribution of

potential confounders by BMD level (T1 vs. T2 plus T3). Logistic regression model

was used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) in women

with low BMD (T1) versus high BMD (T2 plus T3) for selective metabolites

(VIP>1.5). All models were adjusted for age, menopausal status (yes vs. no), body

weight (continuous), height (continuous), waist circumference (continuous), creatinine

(>79 mg/dL, yes vs. no), regular exercise (≥2 times/week for 30 minutes or more, yes

vs. no), and serum alkaline phosphatase (ALP≥ 60 IU, yes vs. no). These variables

were selected by either stepwise model selection (Slentry=0.15, Slstay=0.15) or

because of its biological importance to BMD level. Serum creatinine and ALP were

dichotomized by its median because quite few participants had abnormal level in this

healthy population.

Because menopausal status significantly affect BMD level, effect modification

10

by menopausal status (yes/no) was explored by comparing a model with terms for

main effects and interaction terms to the models for main effect only using the

likelihood ratio test. Stratified analysis was performed to assess the relationship

between metabolome and BMD by menopausal status. Areas under the receiver

operating characteristic (ROC) curve (AUC) were calculated to evaluate the

performance of the candidate metabolites. An AUC value of 1.0 represents a prediction

model with 100% sensitivity and 100% specificity, while an AUC value of 0.5

correspond to a poor predictive model. All analyses were performed by using SAS 9.2

(SAS Institute, Cary, NC). All statistical tests were two-sided and a P value less than

0.05 was considered statistically significant.

11

Chapter 3. Results

3.1 Characteristics of the Study Population

This cross-sectional study included 610 healthy women aged 40 to 55. As

compared to women with high BMD, women with low BMD were older (47.6 vs. 45.4

years old), had lower body weight (53.1 vs. 57.0 kg), body mass index (21.5 vs. 22.7

kg/m²), waist circumference (71.6 vs.73.6 cm), fasting glucose level (97.2 vs. 101.4

mg/dL), and creatinine level (0.78 vs. 0.80 mg/dL), and had higher alkaline

phosphatase level (67.3 vs. 60.2 IU), and more postmenopausal women (36.5% vs.

11.4%, Table 1). The distributions of diastolic blood pressure, triglyceride,

high-density lipoprotein cholesterol, total cholesterol, hypertension status, smoking

history, alcohol consumption, calcium supplement, and regular exercise were similar

between high and low BMD.

3.2 Metabolome of the BMD Levels

Metabolome obtained from unsupervised PCA showed no visual separation

between low and high BMD levels as well as after stratified by menopausal status

(Figure 3).

12

For supervised approach, PLS-DA score plots revealed little evidence of

separation between high and low BMD clusters (Figure 4a). Some parameters of

PLS-DA model were applied to compare high and low BMD and results using one

component again showed little evidence of classification (variation: R²=0.03,

predictive capability: Q² of 10-fold cross validation=-0.002, 1000 random permutation

test: p=0.07, Table 2). After stratification by menopausal status, metabolome showed

better separation by high and low BMD levels in menopausal women (Figure 4b and

4c; R²=0.12, Q²=0.04, permutation test p=0.03) but not in premenopausal women

(R²=0.01, Q²=-0.02, permutation test p=0.76).

3.3 Candidate Metabolites and the Risk of Low BMD

Given the success of the PLS-DA model in classifying high and low BMD

among postmenopausal women (Table 2), a total of 7 significantly altered metabolites

(VIP > 1.5) were identified based on published data^16,17 and Chenomx NMR suite

(Chenomx Inc., Alberta, Canada). As compared with high BMD, low BMD was

related to decreased levels of plasma lactate, acetone, lipid, very low lipoprotein

(VLDL), and glucose, and elevated acetate and glutamine (Table 3).

13

For multivariate logistic regression, women who had an elevated level of

glutamine was significantly associated with 1.6-fold increased risk of low BMD

(AOR=1.55, 95% CI=1.03-2.33, Table 4). In addition, no significant difference

between high and low BMD level was found for lactate, acetone, acetate, lipid, VLDL,

and glucose (Table 4). However, menopausal status significantly modified the

association between lactate (P interaction=0.004), acetone (P interaction=0.01),

acetatate (P interaction=0.02), lipid (P interaction=0.04), VLDL (P interaction=0.02),

glutamine (P interaction=0.04) and the risk of low BMD (Table 5). After stratification,

postmenopausal women with elevated level of lactate (AOR=0.55, 95% CI=

0.33-0.92), acetone (AOR=0.51, 95% CI= 0.31-0.85), lipid (AOR=0.04, 95%

CI=0.001-0.91), and VLDL (AOR=0.49, 95% CI=0.27-0.90) had a significantly

decreased risk of low BMD (Table 5). No significant association was observed for

premenopausal women. In contrast, postmenopausal women with elevated level of

glutamine had an increased risk of low BMD (AOR=6.04, 95% CI=1.57-23.21, Table

5). No significant association was observed for acetate and glucose in either pre- or

postmenopausal women.

Comparing the ROC curves of model 1 (lactate, acetone, lipid, VLDL, and

glutamine), model 2 (variables in model 1 plus acetate and glucose), and glutamine

alone, the AUC was 0.59 for model 1 (95% CI=0.55-0.64, Figure 5a), 0.60 for model 2

14

(95% CI=0.55-0.65), and 0.54 for glutamine alone (95% CI=0.50-0.59) with limited

ability to classify high and low BMD. After stratification by menopausal status, the

AUC became 0.57 for model 1 (95% CI=0.52-0.63, Figure 5b), 0.60 for model 2 (95%

CI=0.55-0.66), and 0.53 (95% CI=0.48-0.58) for glutamine alone for premenopausal

women; AUC was 0.69 for model 1 (95% CI= 0.67-0.72, Figure 5c), 0.70 for model 2

(95% CI= 0.67-0.73), and 0.65 (95% CI= 0.63-0.68) for glutamine alone in

postmenopausal women. For pairwise comparisons between models using ROC

contrast tests, ROC curves were different between model 2 and glutamine alone in all

participants (p=0.03), premenopausal women (p=0.02), and postmenopausal women

(p=0.01, Table 6), respectively. In addition, ROC curve of glutamine alone was

significant different to ROC curve of model 1 in postmenopausal women (p=0.03,

Table 6).

15

Chapter 4. Discussion

4.1 Main findings

This is the first epidemiologic study used ¹H NMR-based metabolomic approach

to investigate the association between plasma metabolome and the risk of low BMD.

Before stratification by menopausal status, metabolome were not associated with the

risk of low BMD; significant association was observed in postmenopausal women

only. Reactive oxygen species has been related to age-related and estrogen-dependent

bone loss.³² Therefore, estrogen plays an important role in maintaining bone strength

by diminish oxidative stress in bone and bone marrow in women. As estrogen level

decreases significantly after menopause, this may explain menopausal status affect the

association of metabolome with BMD level.

4.2 Evidences of Metabolites and Bone Metabolism

Among the top 7 metabolites identified by PLS-DA model (VIP>1.5), elevated

levels of lactate, acetone, lipid, and VLDL were significantly associated with

decreased risk of low BMD; instead, glutamine showed significant increased risk of

low BMD and acetate and glucose showed no association. Below details the postulated

16

mechanism of each metabolite with BMD (Figure 7).

Lactate. Lactate is the end product of glycolysis derived from glucose under low

oxygen conditions.³³ Lactate is also a crucial intermediate for regulating collagen

biosynthesis during osteogenesis.³⁴ This may explain that increased level of lactate

was associated with decreased risk of low BMD via increased bone formation.

Acetone. Acetone is spontaneous produced in the body by the decarboxylation of

acetoacetate, which refers to ketone bodies.³⁵ In human plasma, concentration of

acetone was (generally higher than acetoacetate) correlated with fasting and diabetic

ketosis.³⁶ However, recent studies suggested that BMD was increased in patient with

Type 2 diabetes^37-39 via increased mechanical loading and release of hormonal factors

(oestrogen, leptin and adiponectin).^40,41 This is also observed in our study, that is,

women with the symptom of diabetes mellitus (i.e., fasting glucose > 100 mg/dL) had

higher acetone (Figure 6) and lower risk of low BMD after menopause than those

without elevated level of fasting glucose.

Lipid & VLDL. Some evidence showed that the amount of body lipid across a broad

range of body fat can be reflected by leptin levels,⁴² which related to differentiation of

stromal cells to osteoblasts and thus prevent further loss of bone.^43,44 This may explain

our finding that lipid and VLDL were associated with decreased risk of low BMD,

which was consistent with other studies.^45,46 However, our findings do not support the

17

previous lipid hypothesis of osteoporosis, which suggests that LDL oxidation and

atherogenic lipid profiles promote bone loss,^47-49 probably because the characteristics

of this study population was differ from previous studies and adjusted different

confounding factors in models.

Glutamine. Glutamine may regulate bone metabolism via osteoclast and can

interconvert to glutamate (Glu), a major neuromediator of the central and peripheral

nervous systems.^50,51 In addition, Glu may lead to bone resorption via bone cells

express Glu receptors, especially on osteoclast,⁵² this supports our finding that

elevated glutamine was associated with elevated risk of low BMD. Moreover, some

studies have reported that estrogen facilitates the neuroprotection against oxidative

stress and reduces glutamatergic excitotoxicity.^53,54 This further explains our finding

that elevated glutamine was associated with increased risk of low BMD among

postmenopausal women. However, this finding showed different metabolome as

compared with those from ovariectomized rats.23,24,55,56

It is possible that

epidemiologic and experimental studies differ in many ways, e.g., controlling for

confounding effect, sample size, and biological mechanism, etc.

4.3 Strengths and limitations

This study had several strengths. First, this study had relatively large sample

18

size, especially for metabolomic study on human participants. Second, previous

studies mainly focused on the association of limited metabolites with BMD in

postmenopausal women.^57,58 This study used high-throughput and unbiased approach

to evaluate the metabolome and includes both pre- and postmenopausal women, which

opens a broad view on research in similar topics. In addition, this study additionally

adjusted for numerous important confounders (age, body weight, menopausal status,

creatinine, waist circumference, serum alkaline phosphatase), which provides more

reliable results.

This study had some limitations. First, the cross-sectional design does not allow

us to assess causal inference. Second, this population is relatively healthy and thus it is

not easy to differentiate the metabolome between high and low BMD groups. This is

also why this study did not use clinical cutoff points for BMD level (Figure 2) and

some biochemical values, e.g., serum alkaline phosphatase and creatinine. Third, this

study only assessed metabolome from plasma but not other biomarkers, e.g., urine.

This is because plasma can provide data on lipid-soluble compounds,²² which has

known with extensive association with osteoporosis.^45-49 However, lipid profile is not

available in urine sample as it is hydrophilic. Last, the use of sodium EDTA tubes for

collecting blood samples prevented us from obtaining the information on some

metabolites, e.g., glycerol, choline, and valine, etc.^16,27 However, a recent study

19

showed that even though some endogenous metabolites might be obscured by the

anticoagulant peaks, the information of these anticoagulants binding metabolites can

be identified by other signals [e.g., glycerol (δ 3.87), choline (δ 3.50), valine (δ 2.28)],

which are available in our NMR spectra.²⁷ After removal of anticoagulants, the

inter-sample variation in these lipoprotein signals would allow us to perform the

statistical analyses.

4.4 Conclusions

This study, for the first time, identified metabolome for predicting the risk of

low BMD in postmenopausal women. Metabolomics can quantify a large-scale of

metabolites to characterize response to drugs, diet, lifestyle, environmental stimuli and

genetic modulations.¹⁵ Future studies exploring gene-metabolite interactions using

liquid chromatograph/mass spectrometry for low abundant metabolites and animal

studies are warranted to shed light on the association between metabolome and BMD

in human.

20

References

1. Prentice A. Diet, nutrition and the prevention of osteoporosis. Public Health Nutr

2004;7:227-43.

2. Dawson-Hughes B, Looker AC, Tosteson AN, Johansson H, Kanis JA, Melton

LJ, 3rd. The potential impact of the National Osteoporosis Foundation guidance

on treatment eligibility in the USA: an update in NHANES 2005-2008.

Osteoporosis international : a journal established as result of cooperation

between the European Foundation for Osteoporosis and the National

Osteoporosis Foundation of the USA 2012;23:811-20.

3. Lin YC, Pan WH. Bone mineral density in adults in Taiwan: results of the

Nutrition and Health Survey in Taiwan 2005-2008 (NAHSIT 2005-2008). Asia

Pac J Clin Nutr 2011;20:283-91.

4. Raisz LG. Clinical practice. Screening for osteoporosis. N Engl J Med

2005;353:164-71.

5. Kanis JA. Diagnosis of osteoporosis and assessment of fracture risk. Lancet

2002;359:1929-36.

6. Garnero P, Hausherr E, Chapuy MC, Marcelli C, Grandjean H, Muller C,

Cormier C, Breart G, Meunier PJ, Delmas PD. Markers of bone resorption

predict hip fracture in elderly women: the EPIDOS Prospective Study. J Bone

21

Miner Res 1996;11:1531-8.

7. Garnero P, Mulleman D, Munoz F, Sornay-Rendu E, Delmas PD. Long-term

variability of markers of bone turnover in postmenopausal women and

implications for their clinical use: the OFELY study. J Bone Miner Res

2003;18:1789-94.

8. McLean RR, Jacques PF, Selhub J, Tucker KL, Samelson EJ, Broe KE, Hannan

MT, Cupples LA, Kiel DP. Homocysteine as a predictive factor for hip fracture

in older persons. N Engl J Med 2004;350:2042-9.

9. van Meurs JB, Dhonukshe-Rutten RA, Pluijm SM, van der Klift M, de Jonge R,

Lindemans J, de Groot LC, Hofman A, Witteman JC, van Leeuwen JP, Breteler

MM, Lips P, Pols HA, Uitterlinden AG. Homocysteine levels and the risk of

osteoporotic fracture. N Engl J Med 2004;350:2033-41.

10. Leboff MS, Narweker R, LaCroix A, Wu L, Jackson R, Lee J, Bauer DC, Cauley

J, Kooperberg C, Lewis C, Thomas AM, Cummings S. Homocysteine levels and

risk of hip fracture in postmenopausal women. J Clin Endocrinol Metab

2009;94:1207-13.

11. Enneman AW, van der Velde N, de Jonge R, Heil SG, Stolk L, Hofman A,

Rivadeneira F, Zillikens MC, Uitterlinden AG, van Meurs JB. The association

between plasma homocysteine levels, methylation capacity and incident

22

osteoporotic fractures. Bone 2012.

12. Gjesdal CG, Vollset SE, Ueland PM, Refsum H, Drevon CA, Gjessing HK, Tell

GS. Plasma total homocysteine level and bone mineral density: the Hordaland

Homocysteine Study. Arch Intern Med 2006;166:88-94.

13. Jimi E, Aoki K, Saito H, D'Acquisto F, May MJ, Nakamura I, Sudo T, Kojima T,

Okamoto F, Fukushima H, Okabe K, Ohya K, Ghosh S. Selective inhibition of

NF-kappa B blocks osteoclastogenesis and prevents inflammatory bone

destruction in vivo. Nat Med 2004;10:617-24.

14. Yuan J, Akiyama M, Nakahama K, Sato T, Uematsu H, Morita I. The effects of

polyunsaturated fatty acids and their metabolites on osteoclastogenesis in vitro.

Prostaglandins Other Lipid Mediat 2010;92:85-90.

15. Beckonert O, Keun HC, Ebbels TM, Bundy J, Holmes E, Lindon JC, Nicholson

JK. Metabolic profiling, metabolomic and metabonomic procedures for NMR

spectroscopy of urine, plasma, serum and tissue extracts. Nat Protoc

2007;2:2692-703.

16. Nicholson JK, Foxall PJ, Spraul M, Farrant RD, Lindon JC. 750 MHz ¹H and

1H-¹³C NMR spectroscopy of human blood plasma. Anal Chem

1995;67:793-811.

17. Nicholson G, Rantalainen M, Maher AD, Li JV, Malmodin D, Ahmadi KR,

23

Faber JH, Hallgrimsdottir IB, Barrett A, Toft H, Krestyaninova M, Viksna J,

Neogi SG, Dumas ME, Sarkans U, The Molpage C, Silverman BW, Donnelly P,

Nicholson JK, Allen M, Zondervan KT, Lindon JC, Spector TD, McCarthy MI,

Holmes E, Baunsgaard D, Holmes CC. Human metabolic profiles are stably

controlled by genetic and environmental variation. Mol Syst Biol 2011;7:525.

18. Keun HC, Ebbels TM, Antti H, Bollard ME, Beckonert O, Schlotterbeck G,

Senn H, Niederhauser U, Holmes E, Lindon JC, Nicholson JK. Analytical

reproducibility in ¹H NMR-based metabonomic urinalysis. Chem Res Toxicol

2002;15:1380-6.

19. Dumas ME, Maibaum EC, Teague C, Ueshima H, Zhou B, Lindon JC,

Nicholson JK, Stamler J, Elliott P, Chan Q, Holmes E. Assessment of analytical

reproducibility of ¹H NMR spectroscopy based metabonomics for large-scale

epidemiological research: the INTERMAP Study. Analytical chemistry

2006;78:2199-208.

20. Nicholson JK, Lindon JC, Holmes E. 'Metabonomics': understanding the

metabolic responses of living systems to pathophysiological stimuli via

multivariate statistical analysis of biological NMR spectroscopic data.

Xenobiotica 1999;29:1181-9.

21. Fiehn O. Metabolomics--the link between genotypes and phenotypes. Plant Mol

24

Biol 2002;48:155-71.

22. Gibney MJ, Walsh M, Brennan L, Roche HM, German B, van Ommen B.

Metabolomics in human nutrition: opportunities and challenges. Am J Clin Nutr

2005;82:497-503.

23. Long WF, Li L, Chen HQ, Tang Y, He XL, Jing RZ. ¹H-NMR-based

metabonomics analysis of plasma from osteoporotic rats induced by ovariectomy.

Sichuan Da Xue Xue Bao Yi Xue Ban 2009;40:843-7.

24. Xue L, Wang Y, Liu L, Zhao L, Han T, Zhang Q, Qin L. A ¹HNMR-Based

Metabonomics Study of Postmenopausal Osteoporosis and Intervention Effects

of Er-Xian Decoction in Ovariectomized Rats. Int J Mol Sci 2011;12:7635-51.

25. Lerner UH. Bone remodeling in post-menopausal osteoporosis. J Dent Res

2006;85:584-95.

26. Viant MR. Improved methods for the acquisition and interpretation of NMR

metabolomic data. Biochem Biophys Res Commun 2003;310:943-8.

27. Barton RH, Waterman D, Bonner FW, Holmes E, Clarke R, Nicholson JK,

Lindon JC. The influence of EDTA and citrate anticoagulant addition to human

plasma on information recovery from NMR-based metabolic profiling studies.

Mol Biosyst 2010;6:215-24.

28. Purohit PV, Rocke DM, Viant MR, Woodruff DL. Discrimination models using

25

variance-stabilizing transformation of metabolomic NMR data. OMICS

2004;8:118-30.

29. Eriksson L, Johansson E, Kettaneh-Wold N, Wold S. Introduction to multi-

andmegavariate dta analysis using projection methods (PCA & PLS): Umetrics;

1999.

30. van den Berg RA, Hoefsloot HC, Westerhuis JA, Smilde AK, van der Werf MJ.

Centering, scaling, and transformations: improving the biological information

content of metabolomics data. BMC Genomics 2006;7:142.

31. Xia J, Psychogios N, Young N, Wishart DS. MetaboAnalyst: a web server for

metabolomic data analysis and interpretation. Nucleic Acids Res

2009;37:W652-60.

32. Manolagas SC. From estrogen-centric to aging and oxidative stress: a revised

perspective of the pathogenesis of osteoporosis. Endocr Rev 2010;31:266-300.

33. Cohen SA. Amino acid analysis using precolumn derivatization with

6-aminoquinolyl-N-hydroxysuccinimidyl carbamate. Methods Mol Biol

2000;159:39-47.

34. Sengupta S, Park SH, Patel A, Carn J, Lee K, Kaplan DL. Hypoxia and amino

acid supplementation synergistically promote the osteogenesis of human

mesenchymal stem cells on silk protein scaffolds. Tissue Eng Part A

26

2010;16:3623-34.

35. Frayn KN. Calculation of substrate oxidation rates in vivo from gaseous

exchange. J Appl Physiol 1983;55:628-34.

36. Owen OE, Trapp VE, Skutches CL, Mozzoli MA, Hoeldtke RD, Boden G,

Reichard GA, Jr. Acetone metabolism during diabetic ketoacidosis. Diabetes

1982;31:242-8.

37. Vestergaard P. Discrepancies in bone mineral density and fracture risk in patients

with type 1 and type 2 diabetes--a meta-analysis. Osteoporos Int

2007;18:427-44.

38. Shan PF, Wu XP, Zhang H, Cao XZ, Gu W, Deng XG, Gu C, Liao EY. Bone

mineral density and its relationship with body mass index in postmenopausal

women with type 2 diabetes mellitus in mainland China. J Bone Miner Metab

2009;27:190-7.

39. Ma L, Oei L, Jiang L, Estrada K, Chen H, Wang Z, Yu Q, Zillikens MC, Gao X,

Rivadeneira F. Association between bone mineral density and type 2 diabetes

mellitus: a meta-analysis of observational studies. Eur J Epidemiol

2012;27:319-32.

40. Adami S. Bone health in diabetes: considerations for clinical management. Curr

Med Res Opin 2009;25:1057-72.

27

41. Felson DT, Zhang Y, Hannan MT, Anderson JJ. Effects of weight and body mass

index on bone mineral density in men and women: the Framingham study. J

Bone Miner Res 1993;8:567-73.

42. Frederich RC, Hamann A, Anderson S, Lollmann B, Lowell BB, Flier JS. Leptin

levels reflect body lipid content in mice: evidence for diet-induced resistance to

leptin action. Nat Med 1995;1:1311-4.

43. Thomas T, Gori F, Khosla S, Jensen MD, Burguera B, Riggs BL. Leptin acts on

human marrow stromal cells to enhance differentiation to osteoblasts and to

inhibit differentiation to adipocytes. Endocrinology 1999;140:1630-8.

44. Martin A, de Vittoris R, David V, Moraes R, Begeot M, Lafage-Proust MH,

Alexandre C, Vico L, Thomas T. Leptin modulates both resorption and formation

while preventing disuse-induced bone loss in tail-suspended female rats.

Endocrinology 2005;146:3652-9.

45. Adami S, Braga V, Zamboni M, Gatti D, Rossini M, Bakri J, Battaglia E.

Relationship between lipids and bone mass in 2 cohorts of healthy women and

men. Calcif Tissue Int 2004;74:136-42.

46. Brownbill RA, Ilich JZ. Lipid profile and bone paradox: higher serum lipids are

associated with higher bone mineral density in postmenopausal women. J

Womens Health (Larchmt) 2006;15:261-70.

28

47. Yamaguchi T, Sugimoto T, Yano S, Yamauchi M, Sowa H, Chen Q, Chihara K.

Plasma lipids and osteoporosis in postmenopausal women. Endocr J

2002;49:211-7.

48. Cui LH, Shin MH, Chung EK, Lee YH, Kweon SS, Park KS, Choi JS.

Association between bone mineral densities and serum lipid profiles of pre- and

post-menopausal rural women in South Korea. Osteoporos Int 2005;16:1975-81.

49. Orozco P. Atherogenic lipid profile and elevated lipoprotein (a) are associated

with lower bone mineral density in early postmenopausal overweight women.

Eur J Epidemiol 2004;19:1105-12.

50. Serre CM, Farlay D, Delmas PD, Chenu C. Evidence for a dense and intimate

innervation of the bone tissue, including glutamate-containing fibers. Bone

1999;25:623-9.

51. Skerry TM, Genever PG. Glutamate signalling in non-neuronal tissues. Trends

Pharmacol Sci 2001;22:174-81.

52. Chenu C, Serre CM, Raynal C, Burt-Pichat B, Delmas PD. Glutamate receptors

are expressed by bone cells and are involved in bone resorption. Bone

1998;22:295-9.

53. Behl C, Skutella T, Lezoualc'h F, Post A, Widmann M, Newton CJ, Holsboer F.

Neuroprotection against oxidative stress by estrogens: structure-activity

29

relationship. Mol Pharmacol 1997;51:535-41.

54. Singer CA, Figueroa-Masot XA, Batchelor RH, Dorsa DM. The

mitogen-activated protein kinase pathway mediates estrogen neuroprotection

after glutamate toxicity in primary cortical neurons. J Neurosci

1999;19:2455-63.

55. Zhu X, Liu X, He P, Cao B, Lv Y, Zhang W, Ni X. Metabolomics in serum of

ovariectomised rats and those exposed to 17beta-oestradiol and genistein.

Gynecol Endocrinol 2010;26:760-7.

56. Ma B, Zhang Q, Wang GJ, A JY, Wu D, Liu Y, Cao B, Liu LS, Hu YY, Wang YL,

Zheng YY. GC-TOF/MS-based metabolomic profiling of estrogen

deficiency-induced obesity in ovariectomized rats. Acta Pharmacol Sin

2011;32:270-8.

57. Rapuri PB, Kinyamu HK, Gallagher JC, Haynatzka V. Seasonal changes in

calciotropic hormones, bone markers, and bone mineral density in elderly

women. J Clin Endocrinol Metab 2002;87:2024-32.

58. Frankenfeld CL, McTiernan A, Thomas WK, LaCroix K, McVarish L, Holt VL,

Schwartz SM, Lampe JW. Postmenopausal bone mineral density in relation to

soy isoflavone-metabolizing phenotypes. Maturitas 2006;53:315-24.

30

Figure 1. Flowchart of participant recruitment Participants for statistical

analyses (610)

Lack of blood sample (45) or NMR data (30)

or NMR (30)

Lack of BMD data at spine (51)

Take steroid (3) or have hormone replacement therapy (34)

Women aged 40-55 (773)

31

s o.138

Figure 2. The tertiles of bone mineral density

T1: BMD < 1.138 g/cm²; T2: 1.138 g/cm² ≤ BMD <1.258 g/cm²; T3: BMD ≥ 1.258 g/cm².

1.138 1.258

0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8

0 5 10 15 20 25 30

Percent

Bone mineral density

Bone Mineral Density (g/cm²)

32

Figure 3. PCA score plots from the analysis of CPMG NMR spectra using women plasma samples

(a) High BMD group: n=399; low BMD group: n=211

(b) Premenopausal women. High BMD group: n=349; low BMD group: n=134 (c) Postmenopausal women. High BMD group: n=45; low BMD group: n=77

High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

Abbreviations: PCA, principle components analysis; CPMG, Carr-Purcell-Meiboom-Gill;

BMD, bone mineral density.

33

Figure 4. PLS-DA score plots from the analysis of CPMG NMR spectra using women plasma samples

(a) High BMD group: n=399; low BMD group: n=211

(b) Premenopausal women. High BMD group: n=349; low BMD group: n=134 (c) Postmenopausal women. High BMD group: n=45; low BMD group: n=77

High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

Abbreviations: PCA, principle components analysis; CPMG, Carr-Purcell-Meiboom-Gill;

BMD, bone mineral density.

34

Figure 5. Receiver operating characteristic curves of comparing models for classification of high and low BMD

(a) All participants. Model 1: AUC= 0.59 (95% CI=0.55-0.64), Model 2: AUC=0.60 (95%

CI=0.55-0.65). Glutamine alone: AUC=0.54 (95% CI=0.50-0.59).

(b) Premenopausal women. Model 1: AUC=0.57 (95% CI= 0.52-0.63), Model 2: AUC=0.60 (95% CI=0.55-0.66). Glutamine alone: AUC=0.53 (95% CI: 0.48-0.58).

(c) Postmenopausal women. Model 1: AUC=0.69 (95% CI: 0.67-0.72), Model 2: AUC=0.70 (95% CI: 0.67-0.73). Glutamine alone: AUC=0.65 (95% CI: 0.63-0.68).

Model 1: lactate, acetone, lipid, VLDL, and glutamine.

Model 2: Model 1 plus acetate and glucose.

High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

Abbreviations: AUC, area under the curve; BMD, bone mineral density; CI, confidence interval;

VLDL, very low density lipoprotein.

35

Figure 6. Acetone distribution by fasting glucose level

(High:

glucose> 100 mg/dL; Low: glucose ≤ 100 mg/dL)

≤

≤

36

Figure 7. Postulated mechanism relates important metabolites with bone

mineral density among postmenopausal women

Acetone is referred to ketone bodies which may be generated during diabetic ketosis³⁶, and then promote to release hormonal factors that suppresses estrogen deficiency and the subsequent bone loss.^37-41 Lactate may against bone loss via regulating collagen biosynthesis during osteogenesis.³⁴ Lipid and very low density lipoprotein as well as lipid profiles can be reflected by leptin level which is related to differentiation of stromal cells to osteoblasts and thus prevent further loss of bone mass.^42-44 Bone resorption may occur when glutamine interconverts to excess amount of glutamate and then cause bone loss as the composition of osteoclast expresses glutamate receptors.⁵²

37

Table 1. Characteristics of the study population

Variables BMD

High (n=399)

Low (n=211)

P*

Mean (S.E.)

Age (year) 45.4 (0.19) 47.6 (0.34) <0.0001

Body weight (kg) 57.0 (0.45) 53.1 (0.52) <0.0001

Body mass index (kg/m²) 22.7 (0.16) 21.5 (0.20) <0.0001 Waist circumference (cm) 73.6 (0.35) 71.6 (0.44) 0.0008 Diastolic blood pressure (mmHg) 64.9 (0.51) 63.7 (0.64) 0.13

Fasting glucose (mg/dL) 101.4 (1.16) 97.2 (0.61) 0.01

Triglycerides (mg/dL) 94.1 (2.70) 87.3 (3.10) 0.10

High-density lipoprotein cholesterol (mg/dL)

69.8 (0.77) 72.7 (1.25) 0.05

Total cholesterol (mg/dL) 198.4 (1.66) 202.7 (2.27) 0.13 Alkaline phosphatase (IU) 60.2 (0.84) 67.3 (1.20) <0.0001

Creatinine (mg/dL) 0.80 (0.005) 0.78 (0.006) 0.03

n (%)

Menopause 45 (11.4) 77 (36.5) <0.0001

Hypertension 72 (18.1) 30 (14.2) 0.23

Ever smoker 40 (10.4) 16 (8.0) 0.34

Alcohol consumption 15 (4.0) 14 (7.3) 0.09

Calcium supplement 213 (53.9) 120 (56.9) 0.49

Regular exercise ( ≥2 times/week) 168 (45.7) 91 (49.2) 0.43

Abbreviation: BMD, bone mineral density.

*P-values were obtained from Student’s t tests (normally distributed continuous variables) and Chi-square tests (categorical variables) by comparing participants with high and low BMD.

Number in bold indicates statistically significant finding.

38

Table 2. The PLS-DA parameters and permutation test for differentiating high and low BMD levels

Comparison Groups

PLS-DA parameters Ppermutation testc

No.^b R² Q²†

High BMD－Low BMD^a 2 0.03 -0.002 0.07

Premenopause

High BMD－Low BMD 1 0.01 -0.02 0.76

Postmenopause

High BMD－Low BMD 1 0.12 0.04 0.03

Abbreviations: PLS-DA, partial least squares-discriminant analysis; BMD, bone mineral density.

a High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

b The number of components based on Q² indicates the best classifier of PLS-DA using 10-fold cross-validation method.

c Distribute 1000 permutations.

†Predictive capability.

Number in bold indicates statistically significant finding.

39

Table 3. The change of plasma metabolites in postmenopausal women to distinguish high and low BMD

Metabolites Multiplicity^a Signal assignment^b δH (ppm)

VIP^c High BMD^d Low

BMD^c Pathways*

Lactate d CH3 1.33 1.74-3.31 ↑ ↓ Glycolysis/Gluconeogenesis

Acetone s CH3 2.22 1.84-3.16 ↑ ↓ Synthesis and degradation of

ketone bodies

Acetate s CH3 1.91 1.56-1.82 ↓ ↑ Pyruvate metabolism

Lipid m CH3CH2CH2, (CH2)n

CH3CH2(CH2)n,

1.20-1.26 2.59 ↑ ↓

Fatty acid metabolism

VLDL m CH₂CH₂CO 1.53-1.6 1.63-2.49 ↑ ↓

Glucose ddd Histone H2, H3, H4, H5 3.37-3.57 2.75 ↑ ↓ Carbohydrate metabolism

Glutamine m half γ-CH2 2.42-2.48 1.55-2.09 ↓ ↑ D-Glutamine and D-glutamate

metabolism Abbreviations: BMD, bone mineral density; VLDL, very low density lipoprotein; δH, chemical shift; ppm, parts per million.

a s,singlet; d, doublet; t, triplet; m, complex multiplet; ddd, doublet of doublets of doublets.

b Signal assignment provides precise bioanalytical and dynamic information.

c All metabolites with variable importance in projection (VIP) score >1.5.

d High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

*Pathway information was obtained from KEGG PATHWAY Database (http://www.genome.jp/kegg/).

40

Table 4. Association between plasma metabolites and bone mineral density (T1 vs. T2 + T3)

Abbreviations: AOR, adjusted odds ratio; BMD, bone mineral density; VLDL, very low density lipoprotein; CI, confidence intervals.

a All metabolites with variable importance in projection (VIP) score >1.5.

b High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

c All models were adjusted for age (continuous), body weight (continuous), height (continuous), waist circumference (continuous), menopausal status (yes/no), creatinine (> median, 79 mg/dL, yes/no), regular exercise (≥2 times/week, yes/no), serum alkaline phosphatase (≥ median, 60 IU, yes/no).

Number in bold indicates statistically significant finding.

Metabolites^a

High BMD^b (n=399)

Low BMD^b (n=211)

OR OR (95% CI) P value AOR(95%CI)^c P value

Lactate 1.00 0.92 (0.81-1.04) 0.16 0.90 (0.76-1.05) 0.18

Acetone 1.00 0.90 (0.79-1.02) 0.09 0.87 (0.74-1.03) 0.11

Acetate 1.00 1.40 (0.97-2.03) 0.07 1.32 (0.83-2.10) 0.24

Lipid 1.00 1.08 (0.44-2.66) 0.87 0.72 (0.24-2.18) 0.56

VLDL 1.00 0.88 (0.75-1.04) 0.13 0.85 (0.69-1.04) 0.11

Glutamine 1.00 1.47 (1.07-2.03) 0.02 1.55 (1.03-2.33) 0.04

Glucose 1.00 0.96 (0.91-1.02) 0.18 0.99 (0.93-1.06) 0.83

41

Table 5. Association between plasma metabolites and bone mineral density stratified by menopausal status (T1 vs. T2+T3)

Abbreviations: AOR, adjusted odds ratio; BMD, bone mineral density; VLDL, very low density lipoprotein; CI, confidence intervals.

a All metabolites with variable importance in projection (VIP) score >1.5.

b High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

c All models were adjusted for age (continuous), body weight (continuous), height (continuous), waist circumference (continuous), menopausal status (yes/no), creatinine (> median, 79 mg/dL, yes/no), regular exercise ≥2 times/week (yes/no), serum alkaline phosphatase (≥ median, 60 UI, yes/no).

Number in bold indicates statistically significant finding.

Metabolites^a

Premenopause Postmenopause p interaction

Low BMD/High BMD^b (134/349)

Low BMD/ High BMD (77/45)

OR (95% CI) AOR(95%CI)^b OR (95% CI) AOR(95%CI)^b

Lactate 0.96 (0.83-1.11) 1.01 (0.85-1.21) 0.65 (0.49-0.86) 0.55 (0.33-0.92) 0.004 Acetone 0.90 (0.77-1.06) 0.97 (0.81-1.17) 0.64 (0.48-0.86) 0.51 (0.31-0.85) 0.01 Acetate 1.13 (0.72-1.75) 1.01 (0.60-1.68) 4.02 (1.67-9.67) 3.31 (0.83-13.19) 0.02 Lipid 0.97 (0.33-2.89) 1.32 (0.38-4.56) 0.12 (0.02-0.98) 0.04 (0.001-0.91) 0.04 VLDL 0.89 (0.73-1.08) 0.95 (0.76-1.18) 0.62 (0.43-0.88) 0.49 (0.27-0.90) 0.02 Glutamine 1.32 (0.90-1.95) 1.24 (0.78-1.96) 2.88 (1.45-5.72) 6.04 (1.57-23.21) 0.04 Glucose 0.96 (0.90-1.03) 0.97 (0.90-1.04) 0.95 (0.85-1.06) 1.14 (0.92-1.40) 0.38

42

Table 6. Receiver operating characteristic contrast tests of pairwise comparison between different models to classify high and low BMD

a Model 1 included lactate, acetone, lipid, VLDL, and glutamine; model 2 included variables in model 1 plus acetate and glucose.

*P-values were obtained from Chi-square tests.

Number in bold indicates statistically significant finding.

Contrast models All participants Premenopause Postmenopause P value*

Model 2 vs. Mode 1^a 0.38 0.16 0.74

Glutamine alone vs. Model 1 0.05 0.12 0.03

Glutamine alone vs. Model 2 0.03 0.02 0.01

43

Table 7. Model comparisons for the association between plasma metabolites and bone mineral density (T1 vs. T2+T3)

Lactate Acetone Acetate Lipid VLDL Glutamine Glucose

Low BMD^a (399) / High BMD^a (211) AOR (95% CI)

Model 1^b 0.92 (0.81-1.04) 0.90 (0.79-1.02) 0.42 (0.97-1.03) 1.08 (0.44-2.66) 0.88 (0.75-1.04) 1.47 (1.07-2.03) 0.96 (0.91-1.02) Model 2^c 0.86 (0.76-0.99) 0.82 (0.71-0.95) 1.55 (1.05-2.30) 0.56 (0.21-1.48) 0.80 (0.67-0.95) 1.67 (1.19-2.35) 0.96 (0.91-1.02) Model 3^d 0.99 (0.86-1.15) 0.95 (0.82-1.10) 1.07 (0.70-1.63) 1.04 (0.37-2.90) 0.93 (0.77-1.11) 1.22 (0.84-1.76) 0.99 (0.93-1.06) Model 4^e 0.94 (0.80-1.09) 0.92 (0.78-1.08) 1.17 (0.74-1.83) 0.85 (0.29-2.51) 0.89 (0.74-1.09) 1.40 (0.94-2.09) 1.00 (0.93-1.07) Model 5^f 0.90 (0.76-1.05) 0.87 (0.74-1.03) 1.32 (0.82-2.10) 0.72 (0.24-2.12) 0.85 (0.69-1.04) 1.55 (1.03-2.33) 0.99 (0.93-1.06) Abbreviations: AOR, adjusted odds ratio; BMD, bone mineral density; VLDL, very low density lipoprotein; CI, confidence intervals.

a High BMD indicates the 2^nd and 3^rd tertiles of BMD; low BMD indicates the 1^st tertile of BMD.

b Model 1: Unadjusted model.

c Model 2: Adjusted for age (continuous), menopausal status (yes/no).

d Model 3: Adjusted variables in model 2 plus body weight (continuous) and height (continuous)

e Model 4: Adjusted variables in model 3 plus serum alkaline phosphatase (≥ median, 60 IU, yes/no) and regular exercise (≥2 times/week, yes/no)

f Model 5: Adjusted variables in model 4 plus waist circumference (continuous) and creatinine (> median, 79 mg/dL, yes/no).