Anorexia nervosa (AN) is a primary psychiatric disorder characterized by extreme self-imposed starvation affecting 0.5–1% of college-aged women in the US.1, 2 Among the many significant medical complications and comorbidities associated with the disease, bone loss is the most common. Occurring at both trabecular and cortical sites, an estimated 50% of women with AN have bone mineral density values more than one standard deviation below the mean of a population of young, healthy women, and an additional 35% have bone density values more than two-and-a-half standard deviations below the mean.3
Low bone mass in individuals with AN occurs in the setting of low subcutaneous and visceral fat depots.4 However, despite low peripheral and visceral fat stores, we have shown that bone marrow adipose tissue (MAT) is increased in AN and inversely associated with bone mineral density (BMD).5 Previous studies have shown the clinical significance of elevated levels of MAT and its relationship to bone density and bone strength. Adipose tissue volume is increased in aging patients with osteoporosis6 and has been shown to be inversely associated with BMD in healthy white women7 as well as in obese women.8 Schellinger et al. also showed that individuals with evidence of vertebral bone weakness, as assessed by magnetic resonance findings of Schmorl's nodes, endplate depression, vertebral wedging, and vertebral compression fractures, had increased amounts of vertebral MAT compared with those without such evidence.9 Profound suppression of bone formation is an established mechanism of bone loss in adolescents and adult women with AN.10–12 Because osteoblasts and adipocytes originate from a common progenitor, the human mesenchymal stem cell (hMSC), understanding the factors that potentially regulate the differentiation process of hMSCs into bone and fat may be of great importance in understanding clinical states of low bone mass and how nutritional deprivation affects bone metabolism.
Currently, little is known about the hormonal determinants of MAT. Preadipocyte factor (Pref)-1, a member of the epidermal growth factor–like family of proteins, is expressed in progenitor cell types, including hMSCs and preadipocytes, and is known to suppress adipocyte and osteoblast differentiation. How Pref-1 acts to suppress adipocyte and osteoblast differentiation remains unknown. We have previously shown that Pref-1 levels are elevated in AN and positively associated with MAT and inversely associated with BMD.13 In order to elucidate further the regulation of MAT in humans, we studied women with active AN, women who have recovered from AN (AN-R), and healthy controls. We hypothesized that AN-R would have decreased levels of MAT compared with women with active AN, and that in a nutritionally replete state, Pref-1 levels would be comparable to normal controls.
Materials and Methods
Twenty-nine women were studied: 14 women with active or recovered AN (30.7 ± 2.2 years [mean ± SEM]) and 15 normal-weight controls of comparable age (27.8 ± 1.2 years). Data from 10 of the healthy controls were previously published.5, 13 Women with AN were recruited through referrals from local eating disorder providers and online advertisements, and AN-R subjects and normal-weight controls were recruited through online advertisements. Subjects with AN met DSM-IV weight and psychiatric criteria. None of the subjects received estrogen within 3 months of the study. AN-R subjects were > 85% ideal body weight and had regular menses for at least 3 months. All control subjects had a normal BMI, a history of regular menstrual cycles, and were receiving no medications known to affect bone mass. Control subjects did not have a past or present history of an eating disorder. Subjects with abnormal thyroid function tests, chronic diseases known to affect bone mineral density (other than AN), or diabetes mellitus were excluded from participation.
All subjects were examined and blood was drawn for laboratory studies at a study visit at our Clinical Translational Science Center. Height was measured as the average of three readings on a single stadiometer, and subjects were weighed on an electronic scale while wearing a hospital gown. BMI was calculated using the formula (weight [kg]/height [meter]2), and percent ideal body weight (%IBW) was calculated based on 1983 Metropolitan Life Height and Weight Tables.14
The study was approved by the Partners Institutional Review Board and complied with the Health Insurance Portability and Accountability Act guidelines. Written informed consent was obtained from all subjects.
Pref-1 was measured with the Quantikine Human Pref-1 immunoassay (ELISA) (R&D Systems, Minneapolis, MN, USA) with a mean minimum detectable level of 0.012 ng/mL and intra-assay coefficient of variation (CV) of 3.1% to 4.3%. Leptin was measured using an ELISA (EMD Millipore Corporation, Billerica, MA, USA), with a minimum detectable level of 0.5 ng/mL and an intra-assay CV of 2.6% to 4.6% and an inter-assay CV of 2.6% to 6.2%.
All subjects underwent 1H-magnetic resonance spectroscopy (1H-MRS) of bone marrow of the L4 vertebral body, the proximal femoral epiphysis, metaphysis, and mid-diaphysis to determine lipid content using a 3.0T MR imaging system (Siemens Trio, Siemens Medical Systems, Erlangen, Germany). For lumbar 1H-MRS, a voxel measuring 15 × 15 × 15 mm (3.4 mL) was placed within the L4 vertebral body. Single-voxel 1H-MRS data was acquired using point-resolved spatially localized spectroscopy (PRESS) pulse sequence without water suppression with the following parameters: TE of 30 ms, TR of 3,000 ms, 8 acquisitions, 1024 data points, and receiver bandwidth of 2000 Hz. For femoral 1H-MRS, a voxel measuring 12 × 12 × 12 mm (1.7 mL) was positioned within the proximal femoral epiphysis and single voxel 1H-MRS using the same non–water suppressed PRESS pulse sequence was performed. This process was repeated with voxel placement in the metaphysis at the inter-trochanteric region, and the mid-diaphysis. Automated procedures for optimization of gradient shimming and transmit and receive gain were used. The coefficient of variation for marrow fat quantification was 3%. This was determined by scanning five subjects twice.
The fitting of the 1H-MRS data was performed using LCModel software (version 6.1-4A) (Stephen Provencher, Oakville, ON, Canada). Data were transferred from the scanner to a Linux workstation, and metabolite quantification was performed using an eddy current correction and water scaling. A customized fitting algorithm for bone marrow analysis provided estimates for all lipid signals combined (0.9, 1.3, and 2.3 ppm). LCModel bone marrow lipid estimates were automatically scaled to unsuppressed water peak (4.7 ppm) and expressed as a lipid to water ratio (LWR) (Fig. 1).
A single axial MR imaging slice through the abdomen at the level of L4 and a single slice through the mid-thigh were obtained (Siemens Trio, 3T, Siemens Medical Systems, Erlangen, Germany) to determine abdominal subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), and total adipose tissue, as well as SAT of the thigh. In addition, total cross-sectional area, cortical area, and marrow area of the femur at the level of the mid-diaphysis were determined.
All subjects underwent dual-energy X-ray absorptiometry (DXA) to measure BMD of the PA lumbar spine (L1–L4), lateral spine (L2–L4), total hip, total body, and body composition, including fat mass (kg), lean mass (kg), and percent body fat using a Hologic Discovery A densitometer (Hologic Inc., Waltham, MA. USA). Coefficients of variation of DXA have been reported as <1% for bone,15 1.1% for lean body mass, and 2.7% for fat mass.16
Statistical analysis was performed using JMP 8.0 (SAS Institute, Carry, NC, USA) software. Means and standard error of the mean (SEM) measurements are reported. The means were compared using the Student's t-test with Tukey-Kramer adjustment for multiple comparisons. A p-value of <0.05 was used to indicate significance. Non-normal distributions were compared using the Wilcoxon test and adjusting for multiple comparisons using the Bonferroni correction, in which case a p-value of <0.017 was used to indicate significance. Correlations are for the group as a whole unless otherwise noted.
Clinical characteristics of the study subjects are listed in Table 1. The AN, AN-R, and HC groups were of similar age (p = 0.2). BMI, %IBW, and BMD of the hip, spine, and total body were lower in AN compared with both AN-R and HC. Percent body fat and total fat mass as measured by DXA were significantly lower in AN compared with HC. SAT, total adipose tissue, and SAT of the thigh were also significantly lower in AN compared with HC. There were no significant differences between AN and AN-R with respect to percent body fat, total fat mass, SAT, VAT, total adipose tissue, thigh SAT, total cross-sectional area, cortical area, or marrow area of the diaphysis or leptin levels.
Table 1. Clinical Characteristics of Study Subjects
AN (n = 7)
AN-R (n = 7)
HC (n = 15)
AN, anorexia nervosa; AN-R, women recovered from anorexia nervosa; HC, healthy controls.
Total cross-sectional area of mid-diaphysis of femur (cm2)
5.43 ± 0.32
5.34 ± 0.21
5.35 ± 0.15
Cortical area of mid-diaphysis of femur (cm2)
3.53 ± 0.20
4.02 ± 0.10
3.88 ± 0.13
Marrow area of mid-diaphysis of femur (cm2)
1.90 ± 0.25
1.32 ± 0.19
1.47 ± 0.09
3.7 ± 1.4
5.6 ± 1.6
7.8 ± 1.5
Marrow adipose tissue
MAT of the L4 vertebra was significantly higher in AN (1.01 ± 0.22 LWR) compared with AN-R (0.42 ± 0.05 LWR; p = 0.03) and HC (0.58 ± 0.07 LWR; p = 0.048) (Fig. 2). MAT of the epiphysis, diaphysis, and metaphysis was comparable in all groups (p = 0.99, p = 0.40, and p = 0.57, respectively).
MAT and body composition
MAT of the diaphysis was inversely associated with SAT (Spearman's rho = −0.44; p < 0.04) (Fig. 3A), total adipose tissue (Spearman's rho = −0.44; p < 0.04) (Fig. 3B), and thigh SAT (R = −0.46; p < 0.03). Thigh SAT was also inversely associated with MAT at L4 (Spearman's rho = −0.41; p = 0.047). MAT of the epiphysis was positively associated with percent body fat as measured by DXA (R = 0.50; p < 0.02) and total fat mass as measured by DXA (R = 0.42; p = 0.04).
MAT and bone mineral density
In HC, there was a positive association between MAT of the diaphysis and lateral spine BMD (Spearman's rho = 0.59; p = 0.03), as well as MAT of the diaphysis and total body BMD (Spearman's rho = 0.68; p = 0.007).
Cross-sectional area of the femoral diaphysis
There were no associations between MAT at the diaphysis and total cross-sectional area (p = 0.56), cortical area (p = 0.40), or marrow area (p = 0.38) at the diaphysis in the group as a whole. There was a significant inverse association between cortical area and MAT of the metaphysis (R = −0.53; p < 0.01) and between cortical area and MAT at L4 (Spearman's rho: −0.49; p = 0.01). In the individual groups, there were no significant associations between MAT at the diaphysis and any of the cross-sectional areas.
There were no significant associations between total cross-sectional area, cortical area, or marrow area of the diaphysis and SAT, total adipose tissue, or thigh SAT in the group as a whole. In AN-R, there was a significant inverse association between marrow area and thigh SAT (R = −0.97; p = 0.03).
Pref-1 was significantly higher in AN compared with AN-R (p < 0.02) and HC (p = 0.004) (Fig. 4). Pref-1 levels in AN-R and HC were comparable (Fig. 4).
Pref-1 and MAT
Pref-1 was inversely associated with MAT of the metaphysis in the group as a whole (Spearman's rho = −0.41; p = 0.047). Pref-1 was positively associated with MAT of the L4 vertebra in AN (R = 0.94; p = 0.002) (Fig. 5A). In HC, Pref-1 was inversely associated with MAT of the L4 vertebra (R = −0.71; p = 0.004) (Fig. 5B).
Pref-1 and body composition
There was an inverse association between Pref-1 and percent body fat as measured by DXA (Spearman's rho = −0.46; p = 0.01). There was also a trend toward an inverse association between Pref-1 and total body fat mass as measured by DXA (Spearman's rho = −0.33; p = 0.08).
Pref-1 and BMD
Pref-1 was inversely associated with total body BMD (Spearman's rho = −0.50; p = 0.006) (Fig. 6). There was a trend toward an inverse association between Pref-1 and BMD of the hip (Spearman's rho = −0.35; p < 0.07).
Pref-1 and Leptin
There was a positive association between leptin and total adipose tissue (Spearman's rho = 0.63; p = 0.001), SAT (Spearman's rho = 0.68; p = 0.0003), and thigh SAT (Spearman's rho = 0.73; p < 0.0001). There was a positive association between leptin and femoral neck BMD (Spearman's rho = 0.54; p = 0.02). There was an inverse association between Pref-1 and leptin (Spearman's rho = −0.49; p < 0.01) (Fig. 7).
We have shown that in patients who have recovered from AN, MAT of the L4 vertebra and Pref-1 levels are similar to levels in healthy controls. In all patients, Pref-1 was inversely associated with total body BMD. These data support the hypothesis that elevated Pref-1 levels and MAT in women with active AN is the result of their extreme low weight and further establishes a potential link between nutritional status and bone marrow adipocyte differentiation.
AN is a disease that affects approximately 0.5% to 1% of college-aged women in the US.1, 2 It is characterized by extremely low body weight, low bone mineral density, and decreased amounts of peripheral fat. AN is characterized by significant bone loss, which is associated with a significantly increased fracture risk. A prospective study of 27 women with AN showed a 7-fold increased risk of non-vertebral fracture during a mean 2-year follow-up.17 Similarly, in a retrospective, population-based study, a 3-fold increased risk of fracture was found many years after the initial diagnosis of AN, with the long-term cumulative incidence of fracture being 57%.18 Therefore, gaining a better understanding of the mechanisms of bone loss in AN is especially important.
How fat depots change in response to weight gain in individuals with AN depends on the length of time since the weight gain. Adults with AN who have undergone acute weight-recovery have increased total trunk fat and VAT compared with healthy controls of comparable weight,4 whereas subcutaneous fat in both groups is similar.4 With prolonged weight-recovery in AN, we have shown that the % trunk fat: to % extremity fat ratio is elevated compared with controls.19 Therefore, in adults, peripheral fat depots do not fully normalize, even with prolonged weight recovery. How marrow fat depots change with weight recovery in adults is unknown.
We have previously shown that women with active AN have increased levels of MAT that are inversely associated with bone mineral density and SAT.5 A potential mediator between bone mineral density and fat, Pref-1, has also been shown to be elevated in AN.13 Pref-1 is a member of the EGF-like family of proteins and is a trans-membrane protein highly expressed in preadipocytes, osteoblastic cell lines and hMSCs.20 Pref-1 has been shown to be an important negative regular of adipocyte and osteoblast differentiation. Overexpression of osteoblast-specific Pref-1 in mouse models results in mice with low body weights and significantly reduced BMD.21 Pref-1 has also been shown to be a potentially important metabolic regulator. Overexpression of Pref-1 in a mouse model leads to lower adipose tissue mass but increased insulin resistance compared with wild-type (WT) mice.22 In humans, Pref-1 has recently been shown to be associated with the metabolic profile of obese individuals.23 A cross-sectional study of metabolically healthy obese subjects—defined by the absence of dyslipidemia, impaired fasting glucose, and hypertension—and metabolically unhealthy obese subjects showed that those who were metabolically healthy had lower Pref-1 levels in subcutaneous and omental fat compared with the metabolically unhealthy, and these levels were correlated with inflammatory markers.23 Our data also show that in low-weight and normal weight women, Pref-1 levels are inversely associated with leptin—an important indicator of nutritional stores in low-weight individuals. Therefore, Pref-1 may be an important mediator between an individual's energy balance and the ability to form bone.
Our data support the role of Pref-1 as a negative regulator of osteoblast differentiation because Pref-1 was inversely associated with BMD of the total body. In AN, Pref-1 was also positively associated with vertebral MAT of the L4 vertebra, but in HC, Pref-1 was inversely associated with vertebral MAT. It is possible that Pref-1 functions differently in varying metabolic milieus and therefore, may behave differently in states of nutritional sufficiency compared with nutritional deficiency. Similarly, it is likely that Pref-1 is only one of multiple factors that plays a role in adipocyte differentiation. Further, the role of circulating Pref-1 in humans is currently unknown. Whereas it may seem counterintuitive that Pref-1, a negative regulator of adipocyte differentiation, is positively associated with MAT in AN, it may be that with the differentiation of preadipocytes into adipocytes, Pref-1 is cleaved and released into the circulation, resulting in elevated circulating Pref-1 levels in individuals with elevated levels of MAT. The role of circulating Pref-1 on the varying adipose tissue depots is also currently not known, and future studies will be necessary to delineate the effects of Pref-1 on marrow adipose tissue compared with visceral adipose tissue and subcutaneous adipose tissue.
We have previously shown that MAT of the L4 vertebra and metaphysis are inversely associated with BMD at the spine, hip, and total body in women with AN and healthy controls.5 In this study, we found a positive association between MAT of the diaphysis and BMD of the lateral spine and total body. These disparate data illustrate that it is likely that the relationship between osteoblast and adipocyte differentiation is complex and unlikely to be fully explained as an inverse relationship. For example, there are periods during development, such as puberty, when both MAT and osteoblasts increase, suggesting that this differentiation process does not always occur in an inverse manner. Post et al. also show the presence of independent preosteoblastic and preadipocytic cell populations, providing further evidence that the relationship between osteoblasts and adipocytes is not solely an inverse one.24
We have also previously shown that women with AN have significantly more MAT in the femoral metaphysis and diaphysis compared with HC.5 In this study, we found that women with AN had significantly more MAT in the L4 vertebra compared with both AN-R and HC, but they had similar levels of MAT in the femoral epiphysis, metaphysis, and diaphysis. As the present study was powered to detect a difference in L4 marrow fat, it is possible that we did not have sufficient power to detect this difference in the femur. The fact that we were able to detect a difference in the lumbar spine but not in the femur in a small group of subjects suggests that marrow fat may accumulate preferentially in the lumbar spine, a site composed of predominantly trabecular bone, compared with the femur, a site that contains a higher percentage of cortical bone compared with the spine; this may provide us with more insight into the role of marrow fat in bone mineral metabolism.
Because this is a cross-sectional study, we cannot determine causation based on our data. However, the fact that we saw such a significant decrease in MAT and Pref-1 levels in a small group of women who have recovered from AN suggests that MAT and Pref-1 may be strongly influenced by nutritional factors. MAT is normal in women who have recovered from AN, despite the fact that their SAT is still comparable to that of the AN group, suggesting that MAT may be more sensitive to nutritional changes compared with the other fat depots. Further understanding of factors that influence osteoblast and adipocyte differentiation, including Pref-1, may therefore be of great clinical importance.
All authors state that they have no conflicts of interest.
We would like to that the nurses and bionutritionists of the MGH Clinical Research Center for their expert care. The project described was supported by NIH grant numbers UL1 RR025758, Harvard Clinical and Translational Science Center, the National Center for Research Resources, the KL2 Medical Research Investigator Training (MeRIT) program of Harvard Catalyst and NIH grant R24 DK084970. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Center for Research Resources or the National Institutes of Health.
Authors' roles: PKF worked on the concept and design of the study; acquisition, analysis, and interpretation of data; the writing of the manuscript; and approved the final version of the submitted manuscript. MAB worked on the concept and design of the study; acquisition and analysis of data; revised the submitted manuscript for intellectual content; and approved the final version of the submitted manuscript. LF, BJT, AB, and EM worked on data acquisition for this study, revised the submitted manuscript for intellectual content, and approved the final version of the submitted manuscript. CJR worked on the concept and design of the study, revised the submitted manuscript for intellectual content, and approved the final version of the submitted manuscript. AK worked on the concept and design of the study; acquisition, analysis, and interpretation of data; revised the submitted manuscript for intellectual content; and approved the final version of the submitted manuscript. PKF and AK accept responsibility for the integrity of the data analysis.