A positive association of dietary protein with lumbar spine bone mineral density is suppressed by a negative association of protein sulfur ^,

Abstract

Dietary protein is theorized to hold both anabolic effects on bone and demineralizing effects mediated by the diet acid load of sulfate derived from methionine and cysteine. The relative importance of these effects is unknown but relevant to osteoporosis prevention. Post-menopausal women (n=161, mean±SD 67.9±6.0 y) were assessed for areal bone mineral density (aBMD) of lumbar spine (LS) and total hip (TH) using dual X-ray absorptiometry, and dietary intakes of protein, sulfur-containing amino acids and minerals using a USDA multiple-pass 24 h recall. The acidifying influence of the diet was estimated using the ratio of protein / potassium intake, the potential renal acid load (PRAL) and intake of sulfate equivalents from protein. aBMD was regressed onto protein intake, then protein controlled for estimated dietary acid load. A step-down procedure assessed potential confounding influences (weight, age, physical activity and calcium and vitamin D intakes). Protein alone did not predict LS aBMD (P=0.81); however, after accounting for a negative effect of sulfate (β=− 0.28, P<0.01), the direct effect of protein intake was positive (β=0.22, P=0.04). At the TH protein intake predicted aBMD (β=0.18, P=0.03), but R² did not improve with adjustment for sulfate (P=0.83). PRAL and the protein / potassium ratio were not significant predictors of aBMD. Results suggest that protein intake is positively associated with aBMD, but benefit at the LS is offset by a negative impact of the protein sulfur acid load. If validated experimentally, these findings harmonize conflicting theories on the role of dietary protein in bone health.

Introduction

Recent literature reflects discordant views on the role of dietary protein in bone health (1). Protein appears to hold an anabolic influence on bone, mediated by bone-active hormones, particularly insulin-like growth factor-1 (1), and may increase calcium absorption (2). Conversely sulfate equivalents derived from methionine and cysteine metabolism are exchanged in the kidney for acid equivalents (3); such a dietary acid load has been demonstrated to cause bone demineralization in animals (4,5) and is associated with reduced bone mineral mass in humans (6,7).

It has been proposed that bone demineralization is promoted by a mild but chronic dietary acid load characteristic of the Western diet (7). This acid load can be characterized by nutrient intake as the estimated net endogenous acid production (NEAP), calculated using a ratio of protein to potassium intake (8,9), or using a function of protein, calcium, potassium, magnesium and phosphorus intake known as the potential renal acid load (PRAL) (3,10). These methods assume the sulfur content of protein is a fixed ratio; however, it is acknowledged that the estimation of actual sulfur intakes improves estimations of dietary acid load (11,12), as actual methionine and cysteine contents vary according to protein source. Currently nutrient databases are available which account for variation in sulfur-containing amino acids (13).

The primary aim of this study was to elucidate the role of dietary protein in bone health status as estimated by dual X-ray absorptiometry (DXA) measures of areal bone mineral density (aBMD). Diet record analysis was used to estimate intakes of total protein and sulfate from amino acids for calculation of the dietary acid load. We anticipated that although dietary protein would be positively related to aBMD, this relationship would be suppressed by a negative association of aBMD with the dietary acid load related to protein intake.

Methods

Participants

Our sample consisted of 161 post-menopausal women (mean±SD 67.9±6.0 y) from Champaign County, IL participating in an ongoing study of the relation between physical activity, gait ability and self-efficacy. Subjects were recruited using local media advertising, churches, senior centers and health facilities. Women with neurological illness or severe orthopedic or cognitive limitations preventing physical testing for the broader study were excluded. Cross-sectional data were used for the present analysis. Study participants provided written, informed consent; all study procedures met ethical standards of and were approved by the Institutional Review Board of the University of Illinois at Urbana-Champaign.

DXA

For bone measures, women changed into medical “scrubs” or wore light-weight clothing and removed all jewelry and other clothing except underwear. Lumbar spine (LS) and total hip (TH) aBMD were measured by DXA using a Hologic QDR 4500A bone densitometer (software version 11.2, Bedford, Massachusetts). LS scans included lumbar vertebrae L1–L4. Short and long-term accuracy of the densitometer were verified by scanning a manufacturer’s hydroxyapatite spine phantom of a known density. All DXA scans were performed by an Illinois state licensed x-ray technologist and analyzed by the same investigator trained in scan analysis by Hologic, Inc. (EM Evans). Precision for DXA BMD measures of interest are 1 – 1.5% in our laboratory with CV% calculated from duplicate scans of both young adults and postmenopausal women.

Dietary Intake and Estimation of Acid Load

Intake was assessed using the USDA multiple-pass 24 h dietary recall method (14,15). Participants completed an interview with researchers to screen for missed foods, portion size clarification, and recall completeness. Diet records were analyzed for total energy, protein, methionine, cystine and micronutrients of interest for calculation estimated NEAP: potassium, calcium, magnesium and phosphorus. Nutrient analysis was performed using Nutritionist Pro version 2.3.1 (First Data Bank, San Bruno, CA).

Protein sulfur load was calculated as mEq/d using intakes of methionine and cystine divided by their molecular weights, as described by Frassetto et al. (8).

$$\mathit{\text{Sulfur}}(\mathit{\text{mEq}})=2·\left(\frac{\mathit{\text{methionine}}(\mathit{\text{mg}})}{149.2}+2·\frac{\mathit{\text{cystine}}(\mathit{\text{mg}})}{240.3}\right)$$ (Eq.1)

The PRAL of the diet was estimated according to the method of Remer, Dimitriou and Manz (3, 10):

$$\mathit{\text{PRAL}}=0.49·\mathit{\text{protein}}(g)+0.037·P(\mathit{\text{mg}})-0.021·K(\mathit{\text{mg}})-0.02·\mathit{\text{Mg}}(\mathit{\text{mg}})-0.013·\mathit{\text{Ca}}(\mathit{\text{mg}})$$ (Eq.2)

The protein / potassium ratio estimation of NEAP was calculated according to the method of Frasetto et al. (8):

$$\mathit{\text{NEAP}}=54.5·\frac{\mathit{\text{protein}}(g/d)}{\mathit{\text{potassium}}(\mathit{\text{mg}}/d)}-10.2$$ (Eq.3)

Statistical Analyses

Distributions were assessed for normality and outliers using the Shapiro-Wilk statistic in conjunction with box plot outlier labeling (16). Correlations for energy intake, protein, protein sulfur, minerals of interest, vitamin D, aBMD and body composition (weight, fat mass and lean mass) were calculated for descriptive purposes.

Statistical tests followed the prescriptions of MacKinnon, Krull and Lockwood (17) for modeling suppression and mediation effects. Mediation effects, more commonly described as intermediate endpoints in epidemiology, describe a situation where all or part of the influence of X on Y is transmitted through a third, intermediate variable. Modeling these effects is similar statistically, but different conceptually from confounding variables, in which a spurious association between two variables is explained by a third, non-intermediate variable which presumably causes both X and Y. A suppressor effect is a special case of an intermediate endpoint in which X has a direct association with Y and an indirect association, transmitted through a third variable, which is opposite in sign to the direct effect (17). For example, our present hypothesis posits that protein intake will be positively associated with aBMD (the direct effect), but that a third, intermediate variable, the acid load associated with protein intake, will be negatively associated with aBMD, thereby suppressing the association of protein intake with aBMD unless the model is statistically controlled for the intermediate variable.

Sulfate, PRAL and the protein / potassium calculation were entered individually as the second block of a regression equation containing dietary protein to predict aBMD of the LS and TH. The change in R² was observed between first and second blocks, indicating the improvement in the model conferred by the inclusion of respective estimates of diet acid load. A step-down regression procedure was then applied to examine the potentially confounding influences of body weight, age, physical activity and calcium as well as vitamin D intakes. Because the interaction of dietary protein and calcium intakes have been reported to impact bone health (1,18), a protein by calcium interaction term was also tested for statistical control. At each step, covariates yielding a P value >0.1 were removed from the model. Covariates were tested before and after the addition of sulfate to the model to ascertain any confounding influences. The variance inflation factor was observed as a test of multicollinearity.

Where change in R² was significant, unique protein effects and sulfur-mediated protein effects were estimated and tested for significance. The “Indirect” mediation macro for SPSS by Preacher and Hayes (19) provided bootstrapped variance estimates for the indirect or sulfur-mediated effect using 10,000 resamples with replacement. To describe the individual contributions of protein and related sulfur intakes, scatter plots and least squares regression lines were produced using sample Z-scores (difference between observation and sample mean divided by SD) of aBMD and protein adjusted for sulfur, then of aBMD and sulfur, adjusted for protein (Figure 1). Additionally mean LS aBMD was compared for subjects split across the median for both total protein and sulfur from protein (Figure 2). SPSS 14.0 was used in all analysis.

Figure 1.

Scatterplot with least squares regression line and mean 95% CI of lumbar spine areal bone mineral density (aBMD, g/cm²) with protein, adjusted for associated sulfur intake (A) and with sulfur from protein, adjusted for total protein intake (B); n=161 postmenopausal women. Values are sample Z-scores, adjusted by ordinary multiple regression. Regressions were also controlled for body weight. For aBMD on protein, β=0.21 (P=0.04). For aBMD on sulfur, β=−0.28 (P<0.01).

Figure 2.

Descriptive representation of median lumbar spine areal bone mineral density (aBMD) in 161 postmenopausal women with intakes above (high) and below (low) median intakes of protein and sulfur from protein.

Results

Among participants, 128 (80%) were taking supplemental calcium (defined as a daily supplement containing at least 100 mg calcium), 34 (20%) were taking a prescribed osteoporosis medication, 50 (31%) were taking hormone replacement therapy and an additional 64 (40%) had taken hormone replacement therapy in the past. One participant completed the LS scan, but not the TH scan, reducing the sample size for TH tests to 160. Parametric and robust descriptive statistics are presented in Table 1. Spearman’s correlations of aBMD with protein, energy and estimates of the diet acid load are presented in Table 2. Protein, sulfur, PRAL, and the protein / potassium index of NEAP were positively correlated with one another (all P<0.01), but of the dietary variables only sulfur exhibited a significant, negative correlation with LS aBMD.

Table 1.

Descriptive statistics, 161 postmenopausal women.

	Mean	SD	Median	IQR1
Age, y	68	6.0	66	63 – 72
Weight, kg	73.1	15.1	70.7	62.7 – 80.7
BMI, kg/m2	28.2	5.3	27.7	24.4 – 31.4
kJ/d	7063	2335	6749	5402 – 8548
Protein, g/d	74.7	25.0	76.9	54.7 – 88.9
Calcium, mg/d	840	34.5	781	513 – 1079
Vitamin D, mg/d	3.36	2.73	2.85	1.08 – 4.90
PRAL2, mEq/d	2.4	16.6	1.9	−9.9 – 13.3
Est. NEAP3	49.2	22.2	45.0	35.2 – 57.6
Protein Sulfur4, mEq/d	28.4	15.6	27.4	17.3 – 38.6
LS aBMD5, g/cm2	0.99	0.15	0.98	0.88 – 1.08
TH aBMD, g/cm2	0.88	0.13	0.87	0.78 – 0.97

¹Inter-quartile range.

²Potential Renal Acid Load (3,10), calculated according to formula 2.

³Estimated Net Endogenous Acid Production (8), calculated using the ratio of protein to potassium intake, according to formula 3.

⁴Sulfur content of reported dietary protein sources, calculated according to formula 1.

⁵Areal Bone Mineral Density.

Table 2.

Correlations of dietary intakes, diet acid load estimations and areal bone mineral density (aBMD) of the lumbar spine (LS) and total hip (TH) in postmenopausal women.

	Protein	Energy	PRAL1	NEAP2	Sulfur	K+	LS BMD
Energy, kJ	0.59
P	<0.01
PRAL1	0.43	0.05
P	<0.01	0.56
Est. NEAP2	0.41	0.02	0.88
P	<0.01	0.82	<0.01
Sulfur3	0.67	0.33	0.33	0.27
P	<0.01	<0.01	<0.01	0.03
K+	0.59	0.56	−0.33	−0.41	0.47
P	<0.01	<0.01	<0.01	<0.01	<0.01
LS aBMD4	−0.01	0.10	−0.03	−0.03	−0.17	−.04
P	0.94	0.19	0.75	0.74	0.03	0.66
TH aBMD	0.08	0.08	0.12	0.08	0.04	−.04	0.59
P	0.30	0.32	0.12	0.30	0.64	0.63	<0.01

¹Potential Renal Acid Load (3,10), calculated according to formula 2.

²Estimated Net Endogenous Acid Production (8), calculated using the ratio of protein to potassium intake according to formula 3.

³Sulfur content of reported dietary protein sources, calculated according to formula 1.

⁴Areal bone mineral density, g/cm². Values are Spearman’s Rho and associated P values, n=161.

At the lumbar spine, the step-down procedure eliminated, in order, vitamin D (P=0.83), calcium×protein interaction (P=0.77), energy/wk expended in physical activity (P=0.55), age (P=0.63) and calcium (P=0.10), leaving only weight as a covariate in the final model (P<0.01). In the first block, protein held no association with LS aBMD (P=0.81). Neither the addition of PRAL (P=0.66) nor estimated NEAP using the protein / potassium ratio (P=0.97) significantly improved the model fit; however, adding sulfate demonstrated a negative association of sulfur from amino acids with LS aBMD, and a positive association of protein with LS aBMD (Table 3).

Table 3.

Regression of lumbar spine areal bone mineral density (aBMD, g/cm²) on protein and sulfur from protein, controlled for body weight¹.

	B	SE	β	P
Entered in separate regression models:
Protein	1.11e−4	4.65e−4	0.02	0.81
Sulfur	−1.35e−3	7.39e−4	−0.14	0.07
Entered simultaneously, i.e., mutually adjusted:
Protein	1.35e−3	6.30e−4	0.22	0.04
Sulfur	−2.82e−3	1.01e−3	−0.28	<0.01

¹Estimated using ordinary multiple regression. Final model controlling for weight was determined using a step-down procedure removing age, physical activity, calcium and vitamin D intakes and an interaction of dietary protein with calcium. n=161 postmenopausal women.

The standardized coefficient for sulfate regressed onto protein intake was 0.69 (P<0.01). The standardized indirect coefficient of protein, or that portion of protein’s total predictive influence that is mediated by its sulfur content, was −0.19 (95% CI, −0.35, −0.04), opposite in sign and similar in magnitude to the estimated β for sulfur-controlled protein, 0.21 (Table 3). Despite correlations between protein and acid-load estimators, the largest Variance Inflation Factor observed was 1.9 (for protein and sulfur), well below the recommended threshold of 10 for detecting problematic multicollinearity (20). The relationships of LS aBMD with protein and LS aBMD with sulfur were linear within the range of reported protein and sulfur intakes (Figure 1).

At the total hip, step-down regression removed energy expended in physical activity weekly (P=0.95) vitamin D (P=0.33) and the calcium · protein interaction (P=0.26) terms, but retained age (P<0.01), weight (P<0.01) and calcium intake (P=0.02) as covariates. Protein was a significant (P=0.03), positive (β=0.18) predictor of TH aBMD; however, no improvement in R² was observed with the addition of PRAL (P=0.87), estimated NEAP using the protein / potassium ratio (P=0.29) or sulfate (P=0.83) to the model. The largest Variance Inflation Factor observed was 2.1.

Discussion

Our findings suggest that, within the range of intake reported in our sample, increasing dietary protein is beneficial to aBMD of the lumbar spine and total hip of postmenopausal women, but that this benefit is suppressed at the lumbar spine by the dietary acid load associated with sulfur containing amino acids. Neither the PRAL nor estimated NEAP using the protein / potassium ratio contributed to the prediction of aBMD at either site. The observed regression coefficients are small but clinically meaningful. A participant consuming mean levels of sulfate, but protein at +1 SD would be predicted to have 3.2% (95% CI: 0.12 to 6.4) additional LS aBMD above the sample mean. Conversely a subject consuming mean protein but +1 SD sulfate would be predicted to have 4.3% (−7.4 to −1.2) lower LS aBMD. Although further research is necessary to validate these cross-sectional data, the observed differences support hypotheses that improving intake of low-sulfate protein sources, or alternatively improving protein intake with a corresponding reduction in the dietary acid load may be beneficial in osteoporosis prevention.

Protein deficiency is detrimental to bone health (21). Observational studies of bone health tend to promote a positive view of protein intakes above 0.8 g.kg⁻¹.d⁻¹ (22–26), though not all studies are favorable (27–29). It has long been established that increasing protein intake elevates calcium losses in urine; in many studies fecal calcium measures indicated no apparent compensation in calcium absorption, promoting the view that urinary calcium must reflect mineral lost from bone (30). More recently, however, Kerstetter et al. (2) have demonstrated not only increased calcium absorption, but a reduction in urinary calcium of bone origin in subjects consuming 2.1 g.kg⁻¹.d⁻¹ protein compared to 1.0 g.kg⁻¹.d⁻¹ in a kinetic study using dual stable calcium isotopes. In the same study, no net difference in bone mineral mass were observed between levels of protein intake, however there was a trend toward reduced bone turnover.

In light of the influence of protein on urinary calcium excretion, Dawson-Hughes (1,31) suggests that increased protein intake is beneficial provided that calcium intake is sufficient to support bone growth and urinary losses. Skov et al. (32) and unpublished data in our lab (in review) demonstrate benefits of protein above current recommended levels to bone mineral mass, in the presence of adequate calcium, during weight loss. In the present study the interaction of calcium and protein intake was not significant.

Proposed mechanisms whereby dietary protein may enhance bone health include providing substrate for collagen deposition and increasing circulating levels of insulin-like growth factor-1, a known growth factor for bone (1). A recent prospective study by Alexy et al. (33) demonstrated improved bone health in youth with increasing protein intake. Conversely, bone status was negatively associated with the diet acid load, estimated by the PRAL, which includes estimation of the acidifying effect of dietary protein. The authors noted that this index of dietary acid load does not directly assess methionine and cysteine content of individual proteins, but assumes a fixed proportion of sulfur to protein. The young, growing population is distinct in several important ways from the older population of the present study, but the results illustrate a similar concept: protein is positively associated with bone mineral mass, but also contributes to an acid load with negative ramifications for bone.

Total protein and protein sulfur are highly and intuitively correlated, but the actual ratio depends on protein source. Methionine is an essential amino acid, and deficiency causes adverse health outcomes. However, our results suggest that the addition of lower sulfate protein to a diet that is already adequate in all essential amino acids may be beneficial in osteoporosis prevention; further research is necessary to test this hypothesis. Soy is implicated as a protein source with a low sulfur to protein ratio (13), estimated by Massey (34) at 39.8 mEq sulfur / 100 g protein, compared to a mid-range 54.8 mEq / 100 g protein for milk and 59.4 mEq / 100 g protein for beef, and a higher 73.0 mEq / 100 g protein for pork and 82.2 mEq / 100 g protein for oatmeal. Experiments have tested the impact of soy protein, with and without isoflavones, on calcium balance with mixed results (35–38). Alternatively increased protein from all sources in connection with enhanced intake of alkalizing nutrients may be beneficial. It has been shown that a more alkaline diet is associated with improved bone density (39–43), and that supplementation of potassium bicarbonate or potassium citrate attenuates bone turnover (44,45). Jajoo et al. (46) demonstrated that replacing some cereal, an acidifying diet component, with more base-producing fruits and vegetables resulted in reduced levels of parathyroid hormone, bone resorption and calcium excretion relative to controls.

It is conceptually important that the positive association in this sample between protein and LS aBMD, controlling for sulfur, was almost perfectly negated by the negative association of sulfate equivalents from methionine and cysteine and aBMD. This suppressor effect indicates that any study evaluating the association between protein intake and bone mineral status without controlling for actual sulfur content of protein may observe no significant correlation, despite real positive and negative protein effects. A neutral combined influence of protein and sulfur on aBMD prediction would mask true effects in both studies hypothesizing detrimental acid load effects to bone, and studies hypothesizing beneficial effects of dietary protein. This suppressor effect might explain the discordance of reports concerning the impact of dietary protein or acid load on bone health. Likewise it is possible that uncontrolled influences of the total dietary acid load might explain conflicting reports. If true, this would suggest that future studies of associations between protein intake and bone density must account for the dietary acid load in order to produce unbiased results.

Our results suggest that direct estimation of sulfur from amino acids may perform better than the protein / potassium ratio or PRAL in studies of acid-base balance and bone mineral mass. These constructs were developed and validated for the estimation of the acidifying influence of dietary intake (3,8–10). However because each estimates the acidifying effect of protein as a fixed ratio of total protein intake, they may not be well suited to the investigation of bone, where the negative acid effect seems to be opposed by an anabolic protein effect. Indeed our results suggest that the negative and positive effects of protein may neatly cancel one another out at the LS, leaving the estimated influence on bone of total protein, as used in the protein / potassium ratio and PRAL, neutral in spite of otherwise reliable estimations of dietary acid load. The PRAL is also a function of calcium and phosphorus intake, which have well established influences on the calcium economy (47). These factors may further confound the influence of the PRAL on bone. In light of these complex associations, it may be advisable to use direct intakes of protein, amino-acid sulfur (the acidifying agent of protein), and other relevant minerals such as potassium in lieu of estimated NEAP when calcium balance or bone mineral mass are investigated.

It is not clear why this suppressive association of sulfur intake would describe aBMD at the lumbar spine and not at the total hip, although differences in the response of these two sites are commonly reported (48–50). The change in R² at the LS indicated a small likelihood of a type I error, or inappropriate rejection of the null hypothesis that sulfur intake does not predict aBMD. Conversely, a post-hoc power analysis of the regression at the hip was performed, and indicated a power of 0.74 to detect an improvement in R² as large as that observed at the spine. Though it is not entirely improbable that our regression failed to detect a true difference at the hip, it is very possible that a real difference between protein associations at the hip and spine explain divergent findings. Differences in observations between these measurement sites may relate to differing levels of trabecular compared to cortical bone content; specifically, the ratio of trabecular to cortical bone is greater at the spine than at the hip. It has been theorized that trabecular bone, being more metabolically active, may be more sensitive to dietary intervention, at least in the short term (51,52). However a post-hoc analysis of the trochanter sub region, with a greater proportion of trabecular bone, did not mimic the findings at the LS in this sample (data not shown). Alternatively it is possible that the weight-bearing load at the hip during ambulation may blunt the effects of dietary factors.

This study is not without limitations; the constraints of cross-sectional data are recognized. This research is performed in the context of a broader body of literature suggesting causal connections between protein intake and bone mineral mass (31,32), sulfate intake and acid-base balance (53,54) and acid base balance and bone demineralization (4,5,6,44,55). The results are theoretically consistent and harmonize apparently conflicting research; however, this observational study offers no basis for new causal inference and is best interpreted as a rationale for additional investigation.

In conclusion, protein intake is positively associated with aBMD in postmenopausal women, but this association is suppressed by a negative association of sulfur from amino acids at the lumbar spine. This observation may reconcile reports of positive impacts of dietary protein on bone health with reports of a negative impact of the acid load from sulfur-containing amino acids. At the total hip dietary protein intake is positively associated with aBMD, and may not exhibit the same negative association with sulfur from protein as observed at the spine. These results highlight the need to evaluate actual sulfur contents of varying dietary protein sources rather than assuming a fixed ratio of sulfur to protein. Future research in this line of inquiry should evaluate the role of dietary protein in preserving bone health in populations at higher risk for fracture such as the elderly.

Abbreviations

aBMD

areal bone mineral density

DXA

dual x-ray absorptiometry

LS

lumbar spine

NEAP

net endogenous acid production

PRAL

potential renal acid load

TH

total hip