Diet-derived fruit and vegetable metabolites show sex-specific inverse relationships to osteoporosis status

Abstract

Background:

The impact of nutrition on the metabolic profile of osteoporosis (OS) is unknown.

Objective:

Identify biochemical factors driving the association of fruit and vegetable (FV) intakes with OS prevalence using an untargeted metabolomics approach.

Design:

Cross-sectional dietary, anthropometric and plasma metabolite data were examined from the Boston Puerto Rican Osteoporosis Study, n=600 (46–79yr).

Methods:

Bone mineral density was assessed by DXA. OS was defined by clinical standards. A culturally adapted FFQ assessed usual dietary intake. Principal components analysis (PCA) of 42 FV items created 6 factors. Metabolomic profiles derived from plasma samples were assessed on a commercial platform. Differences in levels of 525 plasma metabolites between disease groups (OS vs no-OS) were compared using logistic regression; and associations with FV intakes by multivariable linear regression, adjusted for covariates. Metabolites significantly associated with OS status or with total FV intake were analyzed for enrichment in various biological pathways using Mbrole 2.0, MetaboAnalyst, and Reactome, using FDR correction of P-values. Correlation coefficients were calculated as Spearman’s rho rank correlations, followed by hierarchical clustering of the resulting correlation coefficients using PCA FV factors and sex-specific sets of OS-associated metabolites.

Results:

High FV intake was inversely related to OS prevalence (Odds Ratio = 0.73; 95% CI = 0.57, 0.94; P = 0.01). Several biological processes affiliated with the FV-associating metabolites, including caffeine metabolism, carnitines and fatty acids, and glycerophospholipids. Important processes identified with OS-associated metabolites were steroid hormone biosynthesis in women and branched-chain amino acid metabolism in men. Factors derived from PCA were correlated with the OS-associated metabolites, with high intake of dark leafy greens and berries/melons appearing protective in both sexes.

Conclusions:

These data warrant investigation into whether increasing intakes of dark leafy greens, berries and melons causally affect bone turnover and BMD among middle-aged and older adults at risk for osteoporosis via sex-specific metabolic pathways, and how gene-diet interactions alter these sex-specific metabolomic-osteoporosis links.

1. Introduction

Osteoporosis (OS) is an important issue for aging populations and has emerged as a chief public health problem among Puerto Rican adults living on the U.S. mainland [1]. OS is a concern for men and women, as it typically remains undetected until a fracture occurs, which dramatically increases risk of morbidity, institutionalization and mortality [2–4]. However, there are established sex differences in the pathogenesis, prevalence and treatment of OS that manifest from variations in peak bone mass and maturation, rate of annual bone loss, and disease screening methods [5]. Although medications are available to prevent or treat bone loss, and ultimately prevent fracture, adherence is generally poor because of factors such as absence of OS symptoms, and high cost and potential side effects of medication use [6–8]. Thus, identifying and understanding tailored, modifiable health behaviors as targets for prevention of OS are key for improving quality of life among aging adults.

Diet is one such central modifiable risk factor for OS [9, 10]. Higher fruit and vegetable (FV) intake, recommended as part of a healthy diet, has been associated with better bone health in several cross-sectional [11–14] and prospective epidemiologic studies [14–16], but not all [17]. Across many ethnic groups, higher FV intake has been related to lower incidence of fracture in Chinese [11], European [15] and American [18] populations. Results from randomized controlled trials have been inconsistent on the effects of FV intake and bone turnover markers, [19, 20] possibly due to poor compliance or assessment only in men and women with adequate intakes of FV. Thus, while there is evidence that FV intake is vital for bone health, the sex differences, biological processes and pathways underpinning this relationship remain incompletely identified.

Mechanisms proposed for the effects of FV on bone health, though limited, include reduction of oxidative stress and inflammation, reducing bone resorption and increasing bone mass and strength [21, 22]. Diet-derived compounds from fruit and vegetables, their metabolites, and other related molecules that are released in response to FV metabolites, may alter the levels of inflammatory compounds and demonstrate downstream effects on bone health. Therefore, the overall objective of this work was to examine untargeted associations between FV intakes and metabolomic profiles in a cross-sectional cohort of Puerto Rican adults with and without OS. Secondary objectives included: 1) testing associated metabolites for enrichment in various biological pathways; 2) determining which FV food groups were inversely associated with OS-metabolites; and 3) testing for interactions by sex.

2. Material and Methods

2.1. Study population

This study included data for 600 participants aged 46–79 y from the Boston Puerto Rican Osteoporosis Study (BPROS), an ancillary study to the Boston Puerto Rican Health Study (participant flowchart, Figure 1). The parent BPRHS cohort included 1504 Puerto Rican adults aged 45–75 y recruited from the Greater Boston Area through door-to-door enumeration and community-engaged activities [23]. Of 1504 Puerto Rican adults who completed the baseline interview, 1267 participated in a 2-y follow-up visit. All who completed the 2-y visit were invited to join the BPROS. A total of 973 participants were re-consented for the BPROS; 205 declined participation, 13 had moved from the area, 47 had difficulty scheduling the interview, 11 were lost to follow up, two did not participate for other reasons, and 20 had died since the 2-y interview. Four participants did not complete the 2-y interview but were re-consented for BPROS. BPROS participants were invited to the Bone Metabolism Laboratory at the Jean Mayer USDA Human Nutrition Research Center on Aging at Tufts University to complete an in-person interview, body composition and BMD measures with a trained, bilingual interviewer. Of 973 participants consented for BPROS, 21 participants were removed from analyses due to invalid BMD measurements at the lumbar spine and/or femoral neck. 817 BPRHS parent cohort samples were sent to Metabolon, Inc (Morrisville, NC USA) for metabolomic analysis. All outcomes in the current study (including covariates) were measured during the BPROS exam. Metabolites were measured on blood plasma samples from their baseline visit (2 years prior, at the parent BPRHS). As four participants were missing data from the 2y interview and a primary objective of the study was to relate diet to metabolite production, we chose to use dietary data from the parent visit (measured 2y prior to their outcome measurements). Analysis of fruit and vegetable consumption at both baseline and at the outcome visit shows almost identical mean dietary intakes over time (mean intake at baseline: 3.0±1.8; mean intake at 2y follow-up 2.8±1.6). With the final data merge, 635 participants presented with diet, bone and metabolomics data; 35 of these participants were missing one or more FV food groups and/or their diet records were deemed invalid. Thus, the total sample size in current analyses included data from 600 BPROS participants. All participants provided written informed consent. The study was approved by the IRBs at Tufts University and the University of Massachusetts, Lowell, and adhered to the ethical principles of the Helsinki Declaration of 1975 as revised in 1983.

Figure 1.

Participant flowchart from the Boston Puerto Rican Health Study. *Metabolomics data were measured by Metabolon in two batches (2017–2018) and normalization across batches was conducted. A total of 525 metabolites were kept across data sets.

2.2. Data Collection

Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Massachusetts, Lowell [24, 25]. REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies, providing 1) an intuitive interface for validated data capture; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for data integration and interoperability with external sources.

2.3. Measures of Bone Mineral Density and Osteoporosis

BMD was assessed with DXA on a GE-Lunar Model Prodigy scanner (GE Lunar, Madison, WI, USA), with weekly calibration using an external standard (aluminum spine phantom; Lunar Radiation Corp). As reported previously, the root mean square precision was 1.31% and 1.04% for BMD of the femoral neck (FN) and lumbar spine (LS, L2-L4), respectively [26]. The DXA measures were completed using DXA acquisition software version 6.1 and analysis version 12.2. The right hip was scanned per standard procedures, unless the participant reported hip fracture or joint replacement in that hip. The study endocrinologist (BDH) reviewed all scans identified as having a T-score >4.0 for non-anatomical parts and for extraskeletal calcification and excluded 32 participants (25 for LS and 7 for FN). OS at the FN or LS was defined as T-score ≤ 2.5 (2.5 SD or more below peak bone mass) [27]. Individuals were classified as having OS if they had OS at either the hip or spine.

2.4. Dietary Assessment

Analyses used dietary measures from the closest preceding interview to the BMD measurement (at 2-y follow-up visit). A food frequency questionnaire (FFQ) adapted and validated for use in this population of Hispanic adults was used to assess usual dietary intake. The traditional Puerto Rican diet differs considerably from both the general US population and from other Hispanic subgroups (e.g. Mexican Americans). Hence, the food list for the FFQ was developed using the National Cancer Institute/Block food frequency format but modified with data from the Hispanic Health and Nutrition Examination Survey dietary recalls for Puerto Rican adults. For example, foods like plantains, and specific soup and rice-dish recipes, as well as appropriate portion sizes, were added to the FFQ. The BPRHS FFQ is a better estimator of dietary intakes in this Hispanic population than the original Block questionnaire [28]. It has been validated against plasma carotenoids [29], vitamin E [30], vitamin B6 [31] and vitamin B12 [32] in Hispanic adults aged 60 y or older. Information on nutrient intakes were calculated from the FFQs using the Nutrient Data System for Research software.

Servings of FV were obtained by dividing the gram amount of each food by the reference serving amount in the USDA Food Guide Pyramid. A composite variable for total FV intakes (servings/d) was calculated as the sum of all FV intakes, excluding starchy root vegetables (potatoes, cassava, plantains), vegetables considered as fat sources (avocado and olives) and fruit juices. Variety in FV intake was defined as the total number of unique fruits and vegetables consumed at least once per month over the past 12 months. Variety score was regressed on total FV servings/d to generate residuals that were then included in models to predict OS. The use of residuals accounts for variation in types of FV consumed, independent of total FV intake, as reported [33].

To characterize the common groupings of FV consumed by this cohort of Puerto Rican adults, we performed principal components analysis (PCA) on the intakes of 42 FV items, which was accomplished in several steps. First, FV were condensed to 42 predefined groups, based on nutrient-composition similarities [Supplemental table 10]. Only foods included in the total FV intake calculation were included in the food groupings. Each food group was then calculated as a percent contribution to total FV servings/d (food group x1 = x1 servings/d / total FV servings/d * 100). Because PCA is sensitive to outliers, data were evaluated to ensure that no participants with FV contributions from food groups > 0.5 SD beyond the mean intake contribution for that group were included (n = 2). PCA factors were generated using the PROC FACTOR procedure in SAS with varimax rotation (v9.4; SAS Institute Inc., Cary, NC USA). This procedure was run with prespecified numbers of factors (2–8) to determine which solution best explained the variation in the current sample’s FV intake. The 6-factor set was chosen based on scree plot readings, eigenvalues (>1.4), total variance explained (range 1.47–3.52) and meaningful interpretation of the individual factor loadings. Background and discussion on these methods have been described [34].

2.5. Measures of the Metabolites

Metabolic profiling of plasma samples was performed by Metabolon, Inc (Morrisville, NC USA) as described for this population [35]. Briefly, frozen plasma samples were shipped on dry ice, and stored at −80°C until analysis. After methanol extraction of proteins, metabolomic analysis employed ultrahigh-performance liquid chromatography-tandem mass spectroscopy. Individual metabolites were identified by referring to a library of over 4500 purified standards for retention time/index, mass-to-charge ratio, and chromatographic data, and then quantified by estimating the AUC of the peaks. Metabolites were log-transformed for entry into analysis. The median relative standard deviation for internal standards (a measure of instrument variability akin to coefficient of variation) was 5%. After normalization across samples, 525 metabolites passed quality control. See section “Additional Covariates,” below, for details on other measured outcomes. Metabolite abbreviations are defined in Supplemental Table 1.

2.6. Additional Covariates

Data on sex, age, menopausal status, estrogen use, and osteoporosis medication use were obtained through questionnaire during BPROS visits. Hormone status of women was captured through self-reported menopause status (menstruation cessation >1y) and use, vs. non-use of hormone replacement therapy. Height and weight were measured in duplicate per standard methods. BMI was calculated as weight (kg)/height (m)². A fasting blood sample was collected by a bilingual certified phlebotomist to assess plasma 25-hydroxyvitamin D (25OHD). Blood samples were collected in evacuated EDTA tubes and centrifuged to separate plasma. A ¹²⁵I radioimmunoassay (DiaSorin, Inc., Stillwater, MN, USA; manufacturer procedures: 68100E) [36] was used to obtain plasma 25OHD measures and quality control was performed following the College of American Pathology Proficiency Program; the intra-assay coefficient of variation was 10.8% and inter-assay coefficient of variation was 9.4%. All other covariates, including educational attainment (<8^th grade, 9^th–12^th grade and some college or higher), alcohol consumption, smoking status, physical activity, and dietary calcium and energy intakes were obtained by questionnaire during the 2-yr follow-up visit ^[23]. The physical activity index (PAI) is a variation of the Framingham PAI, a weighted 24-h score of typical daily activity, based on hours spent doing heavy, moderate, light, or sedentary activity plus sleeping [37]. Dietary calcium (mg) and energy (kcal) intakes were estimated from the FFQ [28].

2.7. Bioinformatics and Statistical Analysis

A roadmap matching the study hypotheses to supporting analyses and related results can be found in Supplemental Figure 1. The paragraphs below relate to outlined sections in the figure, labeled: [A], [B], [C], and [D].

[A] Associations of FV intakes with OS status. OS status (yes/no) was regressed on fruit and vegetable intakes in logistic models, adjusting for age, smoking, alcohol use, BMI, height, education, total energy intake and physical activity. Models were run with and without fruit juice included in the fruit and vegetable variable.

[B, C] Associations of plasma metabolites with OS status and FV intakes. Differences in plasma metabolite concentrations by OS status were compared using logistic regression, adjusted for age, smoking, alcohol use, BMI, height and physical activity. Metabolites were considered significantly different at P ≤ 0.05. Sensitivity analyses were conducted for BMD at the femoral neck (FN) and lumbar spine (LS) using multivariable regression, with the same covariates. Metabolites related to OS were similar to those related to BMD sites (data not shown). The strength and direction of associations with significant metabolites were similar across outcomes (correlation between slopes of metabolites with OS and metabolites with FN: r=−0.846, P = 3.82E-09; and LS: r = −0.766, P = 8.05E-07). We present results from OS assessment only, as this was our primary research question and due to the similarity between results with the sensitivity analyses. Associations between plasma metabolites and total FV intake were assessed with linear regression while controlling for age, smoking, alcohol use, physical activity, education, and total energy intake (kcal/d). Associations were considered nominally significant at P≤0.05. A positive beta coefficient indicates higher metabolite concentration in those with higher intake of FV as combined servings/d. For OS status, a positive beta coefficient indicates increased metabolite concentration in those with OS, and correspondingly a negative beta coefficient indicates lower concentration in those with OS. Identified metabolites associated (P <= 0.05) with both FV intake and OS were selected as joint FV-OS metabolomic signatures, and were not corrected for multiple testing at each step.

[B, C] Identification of enzyme inhibitors related to metabolites significantly associated with OS status or FV intake. Enzyme inhibitor data were mined from BRENDA to identify metabolites with published data indicating inhibitory action on a given Enzyme Commission (EC) number [38]. EC number-pathway assignments were taken from KEGG. Lower inhibitory action (negative beta) is interpreted as supporting reduced inhibition of that particular enzyme.

[B, C] Pathway analysis. Metabolites significantly associated with OS status or with total FV intake were analyzed for enrichment in various biological pathways, functional modules, bioprocesses and diseases using pathway analysis. Software included Mbrole 2.0 [39], MetaboAnalyst [40], and Reactome [41], with analyses run using default parameters, including FDR correction of P values in significance tests for pathway enrichment, as performed previously by this group (35). These analysis packages support assignment of a metabolite to several, often overlapping, functional pathways and modules. Data were analyzed separately, by sex. Biological interpretation of the resulting enriched functional and pathway sets was conducted for those represented by two or more significantly different metabolites.

[D] Correlation coefficients between OS-associated metabolites and intake of specific FV food groupings were calculated as Spearman’s rho rank correlations, followed by hierarchical clustering of the resulting coefficients, using the top six PCA factors for intake of FV and sex-specific sets of OS-associated metabolites. This work was performed in R (v 3.5.1; R Core Team (2018) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL: https://www.R-project.org/) and RStudio (v 1.1.456) with the tidyverse, ggpubr, Hmisc (type=“spearman”),reshape2 and stats packages. We used the stats::hclust function for hierarchical clustering. All statistical tests were two-sided and considered statistically significant at P < 0.05.

3. Results

A total of 600 men and women had complete dietary, metabolome and OS data (Table 1). The average age was 60.0 ± 7.5 y; 70% were women; 10% had OS (11% of women, 7% of men); and average FV intake, excluding fruit juice, for the total sample was 3.0 ± 1.8 servings/d, range: (0.3–13.0).

Table 1.

Descriptive characteristics of participants (n=600) from the Boston Puerto Rican Osteoporosis Study

Characteristic	Men (n=174)		Women (n=426)
	Mean or n (%)	SD	Mean or %	SD
Age (y) (range: 46–79)	59.8	7.8	60.3	7.3
BMI (kg/m2)	30.1	5.1	33.1	7.0
Serum 25(OH) D (nmol/L)	18.2	7.0	19.8	7.5
Physical activity scorea	33.1	6.5	30.9	3.6
Current smoker (%)	56 (32)		82 (19)
Heavy alcohol consumption (%)b	22 (13)		20 (5)
Education
≤8th grade	73 (42)		214 (50)
>8th grade – high school	79 (46)		143 (34)
diploma	21 (12)		69 (16)
Some college
Total energy intake (kcal/d)	2375	828	2017	851
Total calcium intake (mg/d)	991	535	1017	566
Total fruit & vegetable intake (servings/d)c	2.7	1.8	3.1	1.8
Osteoporosis (yes, %)d	14 (8)		42 (10)
	With Osteoporosis (n=14)		With Osteoporosis (n=42)
Bone mineral density (g/cm2)
Lumbar Spine (L2–L4)	0.930	0.116	0.886	0.123
Femoral Neck	0.821	0.114	0.738	0.077
T-score
Lumbar Spine (L2–L4)	−2.584	0.963	−2.617	1.028
Femoral Neck	−1.918	0.877	−2.157	0.553
	Without Osteoporosis (n=160)		Without Osteoporosis (n=384)
Bone mineral density (g/cm2)
Lumbar Spine (L2–L4)	1.245	0.171	1.160	0.152
Femoral Neck	1.024	0.138	0.935	0.127
T-score
Lumbar Spine (L2–L4)	0.043	1.426	−0.334	1.268
Femoral Neck	−0.354	1.061	−0.738	0.912

^aPhysical activity score: a weighted 24-h score of typical daily activity, based on hours spend doing heavy, moderate, light or sedentary activity.

^bHeavy alcohol consumption defined as >2 drinks/d for men; or >1/d for women.

^cTotal fruit and vegetable intake, excluding starchy vegetables (potatoes, cassava, plantains) and fruit juice.

^dOsteoporosis status = yes if osteoporosis present at the femoral neck OR the lumbar spine, defined as T-scores ≤ 2.5 (2.5 SD or more below peak bone mass).

3.1. [A] FV intake related to OS status

Total FV intake (servings/d) was negatively associated with prevalence of OS (OR = 0.75; 95% CI = 0.58, 0.97; P = 0.02) and remained significant after inclusion of FV variety (OR = 0.74; 95% CI = 0.57, 0.96; P = 0.02); individuals without OS consumed 3.1 ± 1.8 servings FV/d while those with OS, consumed 2.3 ± 1.4 servings/d. Variety of FV comprising overall FV intake was not statistically significantly associated with OS prevalence, when controlling for total FV intake (Odds Ratio (OR) = 1.03; 95% CI = 0.98, 1.08; P = 0.24). Inclusion of fruit juice in the estimate of total FV intake attenuated results but did not change the significance (OR = 0.80; 95% CI = 0.65, 0.98; P = 0.03). FV intakes for women without and with OS were 3.2 ± 1.8 and 2.4 ± 1.4 servings/d, respectively. FV intakes for men without and with OS were 2.8 ± 1.8 and 2.0 ± 1.2, respectively.

3.2. [B] Metabolites significantly related to FV intake differ by sex

Metabolomic analysis of plasma samples from 600 individuals with complete dietary and OS outcomes identified 525 known metabolites that passed specific quality control measures. Among women, 66 metabolites were significantly associated with FV intake (without fruit juice), after adjustment for covariates (P < 0.05), and among men 38 metabolites were significantly associated with FV intake (P < 0.05) (Supplemental Table 2, Supplemental Figure 2). The metabolite patterns of women and men differed substantially, in that 97 of 104 total significant metabolite-FV intake associations involved unique metabolites. A few metabolite classes overlapped between men and women. Seven different constituents of caffeine metabolism were significantly lower with higher FV intake in both women and men. In the carnitine/fatty acid metabolite class, 12 different molecules, including docosahexaenoate (DHA), were higher with higher FV intake in both men and women. Lastly, of 16 glycerophospholipid metabolites significantly associated with FV intake, 13 had positive beta values in either men or women, or both (cf. 1-linoleoylglycerol and 2-linoleoylglycerol (18:2)) (Figure 2).

Figure 2.

Metabolites significantly associating with fruit and vegetable intake matching specific metabolite classes. Each panel depicts a volcano plot, beta coefficient vs −log₁₀(P-value), for all metabolites significantly associated with osteoporosis (gray) and a distinct metabolite class highlighted in color: caffeine metabolites (n=8), carnitines and fatty acids (n=13), and glycerophospholipids (n=16). Data for women are plotted with triangles, for men with circles, and are derived from Supplemental Table 1.

Although several metabolites were associated with FV intake in men and women separately, only seven metabolites were shared as statistically significantly associated with FV intake (Supplemental Table 3): docosahexaenoate (DHA; 22:6n3), iminodiacetate, homoarginine, theobromine, 1-linoleoylglycerol (18:2), 2-linoleoylglycerol (18:2), and S-1-pyrroline-5-carboxylate. Of these seven metabolites, all but homoarginine showed directionally identical beta coefficients in both men and women. Interestingly, for five of these seven metabolites, the beta coefficient was greater in men compared to women.

3.3. [B] FV-associated metabolites and enriched pathways

Pathway enrichment analysis, with FDR correction, was performed on the datasets of plasma metabolites significantly correlated with FV intake in women and men individually, to identify bioprocesses and functional modules relevant to how this food group may support reduced incidence of OS (Supplemental Tables 4 and 5). Notable pathways include amino acid biosynthesis, and caffeine, arginine and proline, and glycerophospholipid metabolic pathways (observed in both sexes); prostaglandin/leukotriene and α-linolenic acid/linoleic acid metabolic pathways along with biosynthesis of unsaturated fatty acids (women); and glutamate metabolism (men). Although satiety is not a defined pathway, BPROS women with high FV intake had higher linoleoyl ethanolamide and oleoyl ethanolamide, metabolites known to correlate with satiety [42].

3.4. [C] Plasma metabolites associated with OS differ by sex

Forty plasma metabolites differed significantly among men by OS status, and 33 in women (Supplemental Table 6, Supplemental Figure 3). Similar to the metabolite sets distinguishing FV intakes, these OS-associated metabolite sets are distinct, as just two were identified as significant in both men and women with OS. N-(2-furoyl)glycine, a product of high-temperature cooking or compromised fatty acid beta-oxidation [43], was higher in both women (P = 0.041, beta = 0.038, beta SE = 0.019, OR = 1.04) and men (P = 0.012, beta = 0.053, beta SE = 0.021, OR = 1.05) with OS. Androstenediol (3β,17β) monosulfate, a metabolite of steroid hormone metabolism, specifically a precursor of testosterone, was lower in women with, vs. without, OS (P = 0.012, beta = −1.72, beta SE = 0.80, OR = 0.18) but higher in men with, vs. without, OS (P = 0.0027, beta = 1.32, beta SE = 0.45, OR = 3.73).

Several sex-specific metabolites merit attention. In women, ten steroid hormone metabolites were significantly lower in those with, vs. without, OS; four in the phytochemical class, and three purine/pyrimidine metabolites. In men, nine different molecules linked to branched-chain amino acid metabolism were higher in those with, vs. without, OS. In addition, six different molecules belonging to the bio-toxin/drug class were associated with OS, including three related to ibuprofen. Both women and men also showed significant associations between OS status and various carnitines/fatty acids, glycerophospholipids, and lysine/nitric oxide metabolites (Figure 3, Supplemental Figure 3).

Figure 3.

Osteoporosis-associated metabolites corresponding to specific metabolite classes. Volcano plots of beta coefficients vs −log₁₀(P-value) depict all significantly associated metabolites (gray) with each specific metabolite class highlighted in color: branched-chain amino acids, carnitines and fatty acids, and steroid hormone metabolites. Data for women are plotted with triangles, for men with circles, and are derived from Supplemental Table 6.

Comparing the metabolites associated with FV intake (n=66) and those that characterized OS status (n=33) there was an indication of five shared molecules in women: glutarate, glycerate, homoarginine, tartarate and tartronate, all of which were higher with higher FV intake, and lower with OS (compared to those without OS). For men, there was no overlap of metabolites associated with FV intake and OS status.

3.5. [C] OS metabolites and enriched pathways

Pathway enrichment analysis was performed separately on the datasets of plasma metabolites significantly correlated with OS for women and men to identify relevant bioprocesses and functional modules (Supplemental Tables 7 and 8). For women, as noted above, ten of the 33 metabolites correlated with OS were sulfate conjugates of steroid hormones. However, many of these compounds have not been mapped to metabolic pathways used by the analysis tools chosen, hindering a thorough enrichment assessment. Nonetheless, a steroid hormone module was significant (FDR P = 0.017). Other pathways and bioprocesses of note include phosphatidylcholine and phosphatidylethanolamine biosynthesis, adenosine/nucleotide metabolism, lysine metabolism, glyoxylate/dicarboxylate metabolism, and degenerative disc disease pathway. Results from the Mbrole tool highlight the degenerative disc disease pathway, driven by significantly lower hypoxanthine, urate and uridine with OS (Supplemental Table 7). These metabolites are also indicators of adenosine degradation, which was significant at FDR P = 0.0296. For men, pathways of note showing enrichment based on metabolite differences between those with and without OS were BCAA metabolism, and glycerophospholipid metabolism, including conversion of phospholipid to linoleate within linoleic acid metabolism and phosphatidylcholine biosynthesis (Supplemental Table 8).

3.6. [D] Link between FV-associated metabolites and the steroid hormone biosynthesis pathway

Ten of 33 OS-associated metabolites are steroid hormone derivatives. Metabolites significantly related to high FV intake that can inhibit one or more enzymatic processes in the KEGG steroid hormone biosynthesis pathway include 1-methylnicotinamide (beta +), arachidonate (+), deoxycholate (−), naproxen (−), piperine (+) (all P < 0.05), and fatty acids linoleate, oleate/vaccinate, palmitate and stearate (all P < 0.097, with beta +) (Supplemental Table 9).

3.7. [D] Correlations between FV PCA factors and OS-associated metabolites

Principal components analysis of total FV intake generated six meaningful factors: 1) traditional/sofrito, 2) American vegetable, 3) tropical fruits, 4) other fruits, 5) berries and melons, and 6) dark leafy greens (Supplemental Table 10). Correlations between FV derived factors and OS-associated metabolites are presented as heat maps (Figure 4). In women (Figure 4a), six OS-associated metabolites correlated with factor 1 (traditional/sofrito, four metabolites significant at P < 0.05; two approached significance, at P < 0.10), all with positive correlation coefficients (metabolite by factor 1) suggesting that they are associated with both OS and higher intakes of these vegetables. In men, this relationship differed (Figure 4b), where ten OS-associated metabolites correlated with FV factor 6 (five metabolites significantly (P < 0.05) and five approached significance (P < 0.10), with correlation coefficients opposite to that for OS. Metabolites significantly associated with PC factors indicating higher intake, while also associated with less likelihood of OS included, in men: 1-(1-enyl-palmitoyl)-GPC (PC: leafy greens), 4-hydroxychlorothalonil (PCs: berries/melons & leafy greens), and hippurate (PC: berries/melons) and; in women: homoarginine (PC: berries/melons), hypoxanthine (PC: tropical fruits & leafy greens), N-acetyl-beta-alanine (PC: leafy greens), uridine (PCs: other fruits & leafy greens) and tartronate (PC: berries/melons).

Figure 4.

Heat maps depicting the Spearman’s rho rank correlations between osteoporosis [OS]-associated metabolites and principal component factors [PC] for fruit and vegetable [FV] intake. Above each heat map are minus / plus symbols indicating the direction of association (beta) of that metabolite with OS, panel a) for women and panel b) for men. The factors for FV intake and the metabolites are both ordered according to hierarchical clustering based on hclust in R. Individual panels are annotated with asterisks (*) for those metabolites whose correlation with the factor reached statistical significance (P < 0.05), or a † when significance of that correlation was < 0.10. Significance of the correlation at P < 0.01 and P < 0.001 are indicated by ** and *** respectively. Panels where the color representing the correlation are opposite to the beta of association with OS can be considered protective to some extent. Abbreviations for metabolites are defined in Supplemental Table 1.

4. Discussion

To our knowledge, this is the first study to characterize plasma metabolite patterns for osteoporosis in conjunction with the beneficial relation of FV intake in men and women with bone. The current study identified metabolite differences associated with greater FV intake and showed sex-specific associations to metabolic pathways in older Puerto Rican adults. Similarly, the pathways represented by the OS-associated metabolites differed by sex, suggesting that diet may exert sex-specific mechanisms to affect bone health. Results in men should be interpreted cautiously, as the sample size was lower than that of women in this cohort. The current results are in agreement with other studies that identified metabolites related to OS status, including results in women from the Twins UK and Hong Kong Osteoporosis Study [44] where 7 identified metabolites overlapped in relation to BMD and OS, and identified metabolites associated with BMD among U.S. non-Hispanic white women[45]. Food group factors explaining significant relations between FV intakes and OS-associated metabolites included green leafy vegetables and berries/melons. These results are exploratory, cross-sectional associations that warrant investigation into whether dietary augmentation with FV (particularly dark leafy greens) may act upon different sex-specific metabolic pathways related to bone.

Cross-sectional studies evaluating BMD (11,14), and prospective fracture studies (18,15,44) demonstrate the protective association of higher FV intake with bone health. In the current study, greater FV intakes were significantly associated with reduced prevalence of OS among Puerto Rican older adults, but not variety. FV provide micronutrients that affect bone positively, including vitamin K, folate, magnesium, potassium, vitamin C and carotenoids [46]. Moreover, greater FV intake may modulate the gut microbiome, thereby altering metabolic pathways related to bone [47, 48]. Identifying such metabolic pathways is required to grasp how diets rich in FV influence bone health and reduce risk of fracture. In the current study, sex-specific plasma metabolites were identified in relation to greater FV intake. In women, higher arachidonate, as well as PUFA DHA, n-3 DPA and EPA were associated with FV intakes. These metabolites may decrease risk of OS by reducing low-grade chronic inflammation and supporting a shift in mesenchymal stem cell lineage commitment [49]. PUFA also have been shown to inhibit RANKL-induced osteoclast formation in a dose-dependent manner, with arachidonic acid- and DHA-mediated inhibition the strongest effectors [50]. Women with high FV intake were shown here to have higher homoarginine and piperine concentrations, and homoarginine deficiency has been associated with increased bone turnover [51], while piperine has been shown to alleviate osteoclast formation [52]. Greater FV intake in women correlated with lower concentrations of several plasma metabolites, including deoxycholate, gamma-carboxyglutamate, kynurenine, S-1-pyrroline-5-carboxylate, S-adenosylhomocysteine and the NSAID naproxen. Lower kynurenine may protect bone, as elevated kynurenine has been linked to accelerated skeletal aging by impaired osteoblastic differentiation and increased osteoclastic resorption [53]. Finally, higher intake of FV was associated with lower deoxycholate, gamma-carboxyglutamate, S-1-pyrroline-5-carboxylate and S-adenosylhomocysteine, metabolites with potential to benefit bone. For example, higher deoxycholic acid and S-adenosylhomocysteine were related to lower BMD [54, 55], and gamma-carboxyglutamate to greater likelihood of OS [56]. Overall, it is clear that several metabolites related to FV intake in this study have the potential to positively influence bone health in women.

In men, higher FV intake was associated with lower nicotinamide and S-1-pyrroline-5-carboxylate, which inhibit osteoclast differentiation [57] and promote osteopenia [58], respectively. Greater FV intake in men was associated with higher dimethylglycine (DMG) and myo-inositol. Low plasma DMG has been associated with low BMD and increased risk of hip fracture [59], while higher BMD has been associated with phytate (myo-inositol hexaphosphate) intake [60]. Men with higher FV intake also had higher DHA levels, which was shown to positively influence bone health (46) potentially through inhibiting late-stage osteoclastogenesis [61], although, in the current study, this association was not as strong as that observed in women. Results shown from men should be interpreted carefully, as the sample size was smaller (n=174) than in women (n=426). Data from the current study can be used in comparison with other larger male cohorts.

Although many metabolites detected as significant differed between men and women, a few pathways were shared: arginine and proline metabolism, protein synthesis, and caffeine metabolism. Caffeine metabolites were lower in both men and women with greater FV intake. However, coffee intake did not differ significantly by FV intake or OS status intake; hence the observed lower caffeine metabolites may indicate altered liver function [62]. Lower arginine and proline metabolism, as observed here for proline, S-adenosylhomocysteine and S-1-pyrroline-5-carboxylate, agree with observations indicating that dysregulation in arginine metabolism may partially explain bone related complication of diabetes [63]. Approximately 43% of the current BPROS cohort had diabetes at the time of measurement. Protein metabolism has been shown to influence bone health positively, where amino acid-sensing mechanisms may be important in calcium metabolism and bone homeostasis [64]. As a whole, pathways enriched by FV intake in this study may influence bone health via modulations in BM fatty acids, and regulation of amino acid metabolism.

Three different compounds positively associated with FV intake in women might partially inhibit ten enzymes and five different reactions in steroid hormone biosynthesis. Another four metabolites with trending associations with FV intake might contribute to inhibiting these same five reactions. Additionally, another three metabolites (one trending) showed negative association with FV intake and potential for reduced inhibition of other enzymes and outputs of this pathway. We interpret these data to suggest that a diet high in FV in women changes the levels of specific metabolites that can alter, perhaps slightly, the flux through the steroid hormone biosynthesis pathway, promoting progesterone, testosterone and estradiol-17β production (Figure 5, Supplemental Table 8). Detailed analysis of OS-associated metabolites supports this interpretation, in that ten different steroid hormone metabolites were negatively associated with OS prevalence in women, and that factor 1 from PCA of these 33 metabolites includes strong loadings for all ten of these OS metabolites. Importantly, age was not a significant factor in predicting OS status, in contrast to the impact of FV intake on OS status via steroid hormone biosynthesis.

Figure 5.

Schematic of steroid hormone biosynthesis, proceeding from cholesterol to various hormones via successive enzymatic reactions. Blue paths indicate synthesis modules predicted not to be repressed in women based on the observed osteoporosis-associated metabolite profiles. Red paths are predicted to be partially inhibited based on enzyme inhibition data mined from BRENDA, with numerals indicating those metabolite-enzyme pairs listed in Supplemental Table 9. THDOC, tetrahydrodeoxycorticosterone (3α,21-dihydroxy-5α-pregnan-20-one).

Steroid 5α-reductase (SRD5A, EC 1.3.1.22), inhibited by five of the FV-associated metabolites, catalyzes dihydrotestosterone (DHT) formation, and its inhibition promotes estrogen production (63). Therapeutic SRD5A inhibitors demonstrate neutral impact on BMD [65–67] with one study of 14,152 men showing reduced fracture rates over 9-yr follow-up [68]. It is postulated that the neutral or slightly positive effects on bone resulted from an indirect increase in circulating estrogen. Thus, habitually high intake of FV may increase estrogen levels via SRD5A inhibition. Upregulation of STS, steroid sulfatase (3.1.6.2), was associated with higher FV intakes in the current study. Post-menopausal women have substantially lower gonadal estrogens in circulation, resulting in altered bone remodeling, decreased bone density and increased OS risk [69, 70]. Yet, postmenopausal women tend to have high circulating sulfated steroids, particularly estrone sulfate and DHEA-S [71]. Conversion of these precursors to active estrogens by STS may provide some of the estrogen needed to maintain bone density in postmenopausal women. Preosteoblastic cells possess STS activity and can upregulate estrogen production in bone [72]. The current study suggests that greater FV intakes may modulate STS activity and promote a healthy bone phenotype. However, current data do not clarify if reduced inhibition of STS via lower deoxycholate in women with high FV intake directly leads to greater STS production and/or activity (Supplemental Table 9). Metabolites associated with a dietary pattern that itself associates with bone health could contribute to inhibiting pathways relevant to bone phenotypes. Further research must determine if specific FV-associated metabolites affect steroid hormone biosynthesis. It is also necessary to ascertain if altered regulation of these biosynthetic enzymes by FV intakes affects bone health via systemic changes in concentration and/or via regulation of sex steroid hormones directly within bone cells.

Principal components analysis produced six major factors representing FV intakes within this adult population, showing that dark leafy greens and berries/melon were significantly associated with reduced likelihood of OS in both men and women. Dark leafy greens are the primary source of vitamin K in the diet, which has been shown to lower risk of fracture [73, 74] and are high in magnesium, which recently was shown to be important for bone health in this cohort [75]. In addition, berries are an excellent source of vitamin C and carotenoids, which have been associated independently with higher BMD [76] and lower risk of fracture [77]. Here, the association between the dark leafy green factor and reduced risk of OS was stronger in men, where dark leafy green intakes were negatively correlated with ten of the 40 OS-associated metabolites, while in women, they correlated with three.

This study had several strengths. To date, this is the largest study of bone health among Puerto Rican adults living on the US mainland, although there was a small sample of men with osteoporosis. A comprehensive analysis of metabolomics was performed on a large sample of community dwelling older adults. In addition, bone measures and assessment of osteoporosis were completed using the benchmark (DXA) and standard definitions for the disease. Tests of association between metabolites, diet and OS controlled for numerous confounders, a unique strength to this study as other studies in this area did not consider covariates. Limitations to the current study include the cross-sectional design and inability to assess causality. The current study cannot fully parse out the concurrent influence of other health constituents on bone. However, results can serve to narrow hypotheses for future diet-controlled clinical trials that can test the direct influence of identified metabolites (related to both FV intake and OS prevalence) on bone tissue. Plasma metabolites were measured in two batches and normalized by Metabolon Inc. The normalization procedure retains metabolites across each batch run, and this cross-validation is a strength. However, due to the normalization procedure, only 525 metabolites were retained in the final data, which could hinder detection of other unknown metabolites that relate to OS. In addition, there was a small sample size of men with the disease. Despite these limitations, the current exploratory analyses point to significant relations between dietary intakes of FV and osteoporosis prevalence. These data warrant investigation to establish clinical causality among a large sample, designed to capture potential sex-differences. The use of the FFQ has inherent limitations because of the lack of detailed information on portion sizes and specific recipes with the potential for systematic errors that could have resulted from under- or over-reporting of food intakes. As was used in the current study, energy adjustment can partially mitigate systematic errors. The FFQ is best suited for ranking typical food group intakes as were used in this study.

5. Conclusions

In summary, women with greater FV intakes had higher concentrations of metabolites known to inhibit specific branches of the steroid biosynthesis pathway, which enhance greater estrogen production. In both men and women, a dark leafy green dietary factor and a berries/melon factor were significantly related to reduced likelihood of OS. These data warrant future investigation into whether increasing FV intake, particularly dark leafy greens and/or berries/melons, may causally affect bone turnover and BMD among adults at risk for osteoporosis. In addition, future research is warranted to investigate tailored nutrition interventions that may differentially alter bone health by sex.

Highlights.

• High fruit and vegetable intakes are inversely related to osteoporosis prevalence among middle aged and older Puerto Rican adults

• Processes identified with osteoporosis-associated metabolites were steroid hormone biosynthesis in women, and branched-chain amino acid metabolism in men

• Sex-specific, fruit and vegetable derived metabolites and related pathways were identified in relation to osteoporosis prevalence

• High intake of green leafy vegetables and berries/melons were protective of osteoporosis via links to osteoporosis-specific metabolites in both men and women

Abbreviations:

AUC

area under the curve

BCAA

branched-chain amino acid

BPRHS

Boston Puerto Rican Health Study

BPROS

Boston Puerto Rican Osteoporosis Study

BMD

bone mineral density

DHA

docosahexaenoic acid

DPA

docosahexaenoic acid

EC

Enzyme Commission

EPA

eicosapentaenoic acid

FDR

false discovery rate

FFQ

food frequency questionnaire

FV

fruit and vegetable

KEGG

Kyoto Encyclopedia of Genes and Genomes

NSAID

nonsteroidal anti-inflammatory drug

OS

osteoporosis

PCA

principal components analysis

PUFA

polyunsaturated fatty acids

USDA

United States Department of Agriculture

Data Availability Statement:

Data described in the manuscript, codebook, and analytic code will be made available upon request pending approval by the corresponding author.