Introduction

Osteoporosis is a well-documented complication of anorexia nervosa (AN) and other eating disorders (ED). ED typically present during adolescence when bone mineral accrual is rapid. With starvation and weight loss bone mineral accrual is slowed and turned into loss. Decreased bone mineral content/density can be demonstrated early in the course of disease; it may not be restored despite recovery from the ED [1, 2] and result in a reduced peak bone mass predisposing to osteoporotic fractures later in life [3, 4].

Using biochemical markers of bone metabolism short-term fluctuations of bone formation and resorption can be studied. Serum osteocalcin (OC) concentrations of adolescents are positively correlated with weight and/or weight gain [5–9]. Conversely, markers of bone resorption increase with low weight/negative weight trend [5, 7, 9, 10]. It is conceivable that these relationships can be partly explained by the (lack of) nutritional energy supply and availability of building blocks for skeletal tissue.

In addition to nutritional supply, hormones are essential for the regulation of bone accretion and the maintenance of bone mineral content. Insulin-like growth factor-1 (IGF-1) is central in the regulation of bone formation [11], circulating concentrations are closely related to nutrition and weight (change) [12], and treatment with IGF-1 increases bone mineral density even in underweight AN [13]. Oestrogens prevent osteoporosis in postmenopausal women by inhibiting osteoclastic bone resorption [14, 15]. In long-standing ED, the duration of amenorrhoea is related to the low bone mineral density (BMD) suggesting a similar mechanism [16]. Physiologic oestrogen replacement of adolescents with AN does improve BMD [17] but weight gain appears a more important predictor [18, 19] which suggests that amenorrhoea is a marker of not only oestrogen deficiency but also of underweight/energy deficiency per se [20].

There is a marked heterogeneity in the hormonal response to undernutrition [21] and it may therefore be difficult to discern the effects of nutrition per se from those of hormonal (dys)regulation of bone metabolism. We have therefore presently attempted to investigate the independent roles of nutrition, IGF-1 and oestradiol in bone metabolism in a sizeable sample of adolescent girls with ED.

Methods

Assessment of new patients

Patients referred to Uppsala University Children’s Hospital for medical assessment of weight loss and a suspected or diagnosed eating disorder are subject to a careful medical history and physical examination. Blood sampling is performed in order to exclude somatic disease and evaluate to what degree starvation has compromised metabolism and organ function. During the period 2004–2012, 638 girls born between 1986 and 2001 were assessed. Forty-seven girls with other diseases (type 1 diabetes, coeliac disease, hypothyreosis, rheumatoid arthritis, cystic fibrosis, osteogenesis imperfecta, food allergy, kidney failure, pubertas praecox, neuropsychiatric disorders) and 55 girls on hormonal contraceptives were excluded. Copies of growth charts from the school health services spanning age 6–18 years together with analyses of serum OC, BetaCrosslaps (CTX; C-terminal collagen cross-links), IGF-1 and oestradiol were available for 498 (93 %) of the remaining assessments. Menstrual status at assessment was classified as premenarcheal (n = 68), secondary amenorrhoea (n = 178) when periods had been absent for at least 3 months in postmenarcheal girls or no amenorrhoea (n = 252) when this definition of amenorrhoea was not fulfilled. Twenty-six (5 %) were on medication with selective serotonin reuptake inhibitors (SSRI) at assessment. By DSM-IV [22] criteria 74 girls had AN, 15 bulimia nervosa, and 409 eating disorders not otherwise specified (EDNOS).

Analysis of growth charts

From the growth charts a recorded maximal weight could be obtained. It had been reached less than a year before the initial assessment for 325 (65 %) of the patients. A weight, usually recorded by a family doctor or a nurse in the school health services, 5–88 days (28 ± 17 days; mean ± SD) prior to assessment was available for 467 (94 %) patients. Timing of menarche was by patients/parents recall and corroborated by notes on the growth charts. Weight loss was calculated as the difference between maximal recorded weight and weight at assessment. Duration of weight loss was calculated as the difference in time between the maximal recorded weight and the medical assessment. Rate of weight loss was calculated as weight loss divided by duration. The estimations of weight loss, duration and rate of weight loss thus do not take into account the course of weight changes between the two observations. Final rate of weight loss was calculated for the weight loss from the last recorded weight prior to assessment. Growth rate was calculated as the growth in stature between the point of maximal weight and assessment, divided by duration. Body mass index (BMI) was calculated as weight/length2 (kg/m2) for all observations of weight and stature. Gynaecological age was calculated as the difference between age at assessment and age at menarche. Measures of weight, stature and BMI were recalculated into standard deviation scores (SDS) [23].

Follow-up during treatment

During 2004–2011, 59 patients who could not be managed on an out-patient basis received intensified treatment in an adolescent psychiatric day-care unit specialised in ED. Eighteen were premenarcheal and 41 had secondary amenorrhoea. At the start of day-patient treatment emphasis was on nutritional rehabilitation with an aim of establishing a weight gain of 0.5–1.0 kg/week. Weighing and blood sampling for OC, CTX and hormones were intended to be biweekly but had to follow the patients’ progress through the treatment protocol and attendance to the day-care unit. The average sampling interval was 17.6 ± 6.0 days (range 6–46, median 15). Follow-up was for a median of 22 weeks (range 7–87) and generated a median of 10 (range 4–30) samplings during a net weight gain of 9.8 ± 3.8 kg (range 1.1–18.7). Weight changes between two weighings, which were not exactly 14 days apart, were recalculated to 14-day periods. The average weight gain for the 627 14-day periods analysed was 0.76 ± 0.99 kg (range 3.6 to +4.8). Blood sampling was discontinued if patients (re)gained menstruations (n = 35), if they were transferred to treatment facilities where repeated blood sampling could not be continued (n = 23) or if they declined further blood sampling (n = 1). Thirty-five (59 %) girls were on olanzapine and 28 (47 %) on SSRI during some part of the sampling period. Nineteen (32 %) had been on both drugs but not necessarily simultaneously.

Analysis of serum bone metabolism markers and hormones

Analyses were performed at the Department of Clinical Chemistry at the Uppsala University Hospital as part of the clinical work-up of the patients. The laboratory is certified by a Swedish government authority (Swedac). Non-fasting blood samples were analysed for OC, CTX and oestradiol using an automatic analyser (Cobas E601, Roche Diagnostics, Basel, Switzerland). Total assay variation is continuously monitored and maintained <5 %. In premenopausal women >20 years of age the reference value for OC is 11–43 μg/L and for CTX < 580 ng/L. Age and gender specific reference values for adolescents are not available for a local population. During the study period IGF-1 was measured by two automatic analyser, Nichols Advantage (Nichols Institute, San Juan Capistrano, CA, USA) and Immulite 2000 (Siemens, Los Angeles, CA, USA). The methods correlate well and values from both methods could be used in the study. The reference range for the age group 12–15 years is 115–510 μg/L, for 16–20 years 247–482 μg/L. Detailed age-specific reference values have previously been published [12].

Ethics

The protocol was approved by the Ethics committee of the Faculty of Medicine of Uppsala University.

Statistical analysis

All data were analysed using SPSS version 18.0.2. Values are mean ± SD. Differences between menstrual groups of anthropometric measures, hormones and bone metabolism markers were analysed using a one-sided ANOVA. Predictors of the markers of bone metabolism were analysed using a sequential linear regression analysis. P-values refer to two-tailed tests.

Results

Characteristic of patients

Girls who presented with an ED before menarche were at their maximal premorbid weight close to the population mean as evidenced by weight SDS and BMI SDS around zero (Table 1). At presentation they had lost only 4.2 ± 4.5 kg of weight but emaciation was marked reaching BMI SDS −1.73 ± 1.27. Postmenarcheal girls were at their maximal weight heavier and had higher BMI than the population mean with SDS for weight and BMI above zero. Girls with secondary amenorrhoea had lost more weight compared to those with retained menstruations (p < 0.001). The growth rate up to presentation was highest in the premenarcheal group. Duration of weight loss was on the average around a year for all groups. At assessment a majority of 366 (73 %) girls were on a weight-losing course or only just maintained their low weight. The subgroup of patients who entered the day-patient treatment programme had at their maximal weight characteristics similar to the entire group. They differed in that weight loss was greater (premenarcheal 6.9 ± 3.5 kg; secondary amenorrhoea 12.4 ± 6.2 kg) and that weight (premenarcheal 35.9 ± 7.1 kg; secondary amenorrhoea 43.7 ± 4.4 kg) and BMI SDS (premenarcheal −2.30 ± 0.93 SDS; secondary amenorrhoea −2.51 ± 0.94 SDS) were lower at start of day-patient treatment.

Table 1 Weight, weight changes, serum hormones and serum markers of bone metabolism in 498 teenage girls presenting with eating disorders
Full size table

Serum bone metabolism markers and hormones at assessment

At presentation, premenarcheal girls had higher serum concentrations of both OC and CTX compared to the postmenarcheal groups (p < 0.001) (Table 1). In postmenarcheal girls, those with secondary amenorrhoea had lower serum concentrations of OC (p < 0.001) and higher serum concentrations of CTX (p < 0.001) compared to those with retained menstruations.

Serum concentrations of both IGF-1 and oestradiol were lowest in the premenarcheal group and highest in the postmenarcheal group with retained menstruations (p < 0.001).

For each category of menstrual status, the serum OC concentration at presentation was entered as the dependent variable in a sequential linear regression analysis (Table 2). In a first step, age and growth rate were entered in view of the previously demonstrated effects of developmental stage/age [9]. In a second step, BMI SDS and final rate of weight loss were entered as measures of the degree of emaciation and of short-term energy balance. Finally, serum IGF-1 and/or oestradiol concentrations were entered when significant by forward stepwise selection. When the serum concentrations of hormones were entered in the model, the squared value of their semipartial correlations was calculated as measures of their unique contribution to the variation of the independent variable. In all menstrual groups, low age/high growth rate was associated with higher serum OC concentrations. In premenarcheal girls and those with secondary amenorrhoea low BMI SDS and a high rate of weight loss contributed to the model and were associated with low serum OC concentrations. In all menstrual groups, there was an independent contribution to the model by serum IGF-1 concentrations and in premenarcheal girls and those with secondary amenorrhoea also by serum oestradiol concentrations. In all cases higher serum hormones concentrations were associated with higher serum OC concentrations. In view of that serum oestradiol concentrations vary during the menstrual cycle, being especially high around ovulation, outlying values could influence the model. If such values were omitted or substituted for lower values the contribution of serum oestradiol concentrations to the model was, however, not altered.

Table 2 Relationship between serum osteocalcin concentrations and age, anthropometric measures, serum insulin-like growth factor 1 (IGF-1) and serum oestradiol in adolescent girls presenting with an eating disorder
Full size table

In a similar analysis the serum CTX concentration at presentation was entered as the dependent variable (Table 3). In all menstrual groups, low age/high growth rate was associated with higher serum CTX concentrations. In the postmenarcheal girls, there was also an association between low BMI/high rate of weight loss and high serum CTX concentrations. The serum concentrations of IGF-1 and oestradiol did not enter the model and thus did not contribute to the prediction of serum CTX at presentation.

Table 3 Relationship between serum C-terminal peptide of type 1 collagen (CTX) concentrations and age, anthropometric measures, serum insulin-like growth factor 1 (IGF-1) and serum oestradiol in adolescent girls presenting with an eating disorder
Full size table

Serum bone metabolism markers and hormones during treatment

Throughout treatment there was net weight gain and successive increases of the serum concentrations of IGF-1 and oestradiol (Fig. 1). For each category of menstrual status, the serum OC concentration was entered as the dependent variable in a sequential linear regression analysis (Table 4). In a first step, BMI SDS and the preceding 2-week weight change were entered. In a second step, serum IGF-1 and/or oestradiol concentrations were entered when significant by forward stepwise selection. The squared value of the semipartial correlations was calculated as measures of the unique contribution to the prediction. In both menstrual groups, BMI SDS and weight change were positively correlated with the serum OC concentration (Table 4). There were also independent contributions to the model by IGF-1 and oestradiol, which both were positively associated with OC. Premenarcheal and postmenarcheal girls differed in that the anthropometric measures had a higher predictive value in the premenarcheal girls.

Fig. 1
figure 1

Serum insulin-like growth factor-1 (IGF-1) (a) and serum oestradiol (b) concentrations in premenarcheal (filled circle) girls and in girls with secondary amenorrhoea (open circle) during treatment of eating disorders with weight loss. BMI standard deviation scores (SDS) reached during treatment are categorised in one unit wide groups. Data are given as means ± 95 % confidence intervals and include 59 treatment periods (premenarcheal 18; secondary amenorrhoea 41) and a total of 627 samples (premenarcheal 257; secondary amenorrhoea 370). There is a significant change of serum hormones with BMI by a one-sided ANOVA (p < 0.001) for both hormones in both menstrual groups

Full size image
Table 4 Relationship between serum osteocalcin and serum C-terminal peptide of type 1 collagen (CTX) concentrations and anthropometric measures, serum insulin-like growth factor 1 (IGF-1) and serum oestradiol in adolescent girls with anorexia nervosa and subjected to an intensive treatment programme resulting in net weight gain
Full size table

In a similar analysis, the serum CTX concentration was entered as the dependent variable (Table 4). Considering that serum CTX concentrations decrease during the early phases of treatment and subsequently increase [9], the very low weight range (BMI SDS < −2.00) was analysed separately from the low-normal weight range (BMI SDS ≥ −2.00). In premenarcheal girls in the very low weight range anthropometric measures did not predict serum CTX concentrations but there was a positive correlation with serum oestradiol concentrations. In the higher weight range the serum oestradiol concentration was still the most important predictor but there was also a positive relationship with BMI SDS and weight gain. In postmenarcheal girls there was a positive correlation with BMI SDS, weight gain and serum IGF-1 concentrations throughout the weight ranges. In the very low weight range there was also an independent contribution of the serum oestradiol concentration to the prediction. Overall, predictions of CTX during treatment were stronger in premenarcheal girls.

Discussion

This investigation shows that in adolescent girls with ED weight loss/negative energy balance is per se associated with decreased bone formation and increased bone resorption. The previous observations of a correlation between circulating IGF-1 and OC in adolescents with ED [5–7, 24–26] are extended in that weight and weight changes to a higher degree than IGF-1 predicts serum OC. This could be seen both at presentation, when most patients were on a weight-losing course, and during treatment when there was net weight gain. Nutrition thus independently influences bone formation by providing building blocks and energy supply [27, 28] but hormones, such as IGF-1, have a regulatory role [11]. The positive correlation of IGF-1 and OC at presentation indicates an independent effect of IGF-1 which would support some bone formation even when energy intake is poor. This forms a parallel with the finding that exogenous IGF-1 stimulates bone formation even during negative energy balance [29] and increases bone mineral density in underweight patients with AN [13]. However, since IGF-1 production is sensitive to nutrition and is rapidly reduced during negative energy balance [12], it cannot compensate for the direct effects of starvation. The observations altogether emphasise the mandatory role of nutrition in bone formation and that weight recovery is the most important aspect of treatment for the preservation of bone health.

At presentation, circulating CTX could be predicted by growth and weight changes but not by independent effect of the hormones. There are reports of both a positive [6] and negative [7] correlation between IGF-1 and bone resorption markers in adolescents. By the same token, a negative correlation [24] or none at all [6] with oestrogens has been reported. During adolescence, bone metabolism markers are influenced by pubertal development and longitudinal growth [9, 30], factors not always included in the analyses of these previous investigations. The present data therefore support the notion that negative energy balance per se increases bone resorption [20] independently of the hormones usually associated with bone accretion and maintenance of bone mineral density. However, other hormones may influence the process. Circulating cortisol concentrations are increased in ED with weight loss [31] and related to both starvation and comorbid depression [32]. Cortisol concentrations are negatively correlated with bone resorption [7, 33] and bone mineral density [31]. It is therefore likely that circulating cortisol, which was not measured in the present investigation, contributes to bone resorption. It is therefore notable that not only weight gain but also treatment of comorbid depression may help to preserve bone health [34].

During weight gain, there were unique contributions of the hormones, and especially oestradiol, to the positive prediction of circulating CTX. If it is accepted that oestrogens enhance bone mineral accretion/maintenance by inhibiting osteoclastic resorption of bone [14, 15] this appears the opposite of what would be expected. The findings are, however, better interpreted if it were assumed that they represent a return to normal pubertal development when energy intake is adequate [35]. In evidence of this there is an increase of circulating IGF-1 associated with resumption of growth and of circulating oestradiol when pubertal development/menstrual cycles are resumed. OC and CTX concentrations then parallel the circulating hormones as in healthy adolescents [36] which marks a return to normal active bone metabolism and longitudinal growth [6, 9, 30] with a high bone turnover balanced to promote bone accretion [7, 37].

It is notable that bone resorption during weight loss is directly related to negative energy balance irrespective of menstrual status. Endogenous oestrogens have no effect independent of the weight loss and would therefore not be able to protect against bone loss. Indeed, bone resorption is highly sensitive to nutrition and mineral loss increases within days when energy availability decreases [38, 39]. Oestrogen replacement using oestrogen/progesterone combination pills in amenorrhoeic postmenarcheal adolescents may then be not only inefficient [18, 19] but even detrimental to bone health. If gonadotropin secretion and hence oestradiol production is suppressed by exogenous oestrogens/hormonal contraceptives normal adolescent bone accretion may be negatively influenced. There may also be a suppression of IGF-1 production by oestrogens which would influence bone metabolism [40]. Such mechanisms are indicated by the finding of decreased bone mineral density in young women who have taken oral contraceptives during late adolescence [41, 42]. The use of hormonal contraceptives during young adulthood does not have this effect thus underscoring the vulnerability of bone accretion during adolescence. If oestrogens are administered transdermally to adolescents with AN, doses are lower and mimic puberty. A positive effect on bone mineral accretion during such physiologic oestrogen replacement has been observed [17]. It is, however, notable that this effect is observed during weight gain. The present data suggest that direct effect of negative energy balance would not be counteracted even by such physiologic oestrogen replacement.

A strength of this study is the large number of adolescents investigated and the availability of objective measures of weight (change) for analyses together with hormones and markers of bone metabolism. There is a limitation in that it was not possible to obtain fasting blood samples which, since there are diurnal variations of OC and CTX, would increase variation [43]. The sampling intervals during follow-up were irregular and cyclic changes of oestrogens are likely to increase the distribution width of point samples of the hormone. Another limitation is that an age- and gender appropriate reference population was not available. Neither was the possible effect of SSRI medication on bone metabolism analysed. However, this does not preclude the analyses of bone metabolism markers against weight (change) and circulating hormones within the ED sample. The limitations may make predictions less precise but the associations obtained nevertheless hold true.

This investigation shows that negative energy balance directly affects bone metabolism more than the concomitant changes in circulating hormones. Normalisation of eating and weight gain is a first priority in the successful treatment of adolescent ED [44] which also would prevent further bone mineral loss. It is notable that the short-term weight trend and temporary nutritional insults affect bone resorption [9, 38]. An implication of this observation is the importance of a steady, uninterrupted weight gain during nutritional rehabilitation. It is likely that repeated periods of weight loss which affect bone accretion would lead to a reduced peak bone mineral density. Girls with premenarcheal onset of an ED are at an especially high risk of developing osteoporosis [16, 45, 46]. This is due to not only that they have many years of potential bone mineral accretion ahead of them but also, as presently shown, that their bone metabolism is more sensitive to the effects of starvation. In these young girls, vigorous and uninterrupted nutritional rehabilitation is necessary to preserve bone health.