Menstrual Disruption with Exercise is not Linked to an Energy Availability Threshold

Abstract

Introduction

Chronic reductions in energy availability (EA) suppress reproductive function. A particular calculation of EA quantifies the dietary energy remaining after exercise for all physiological functions. Reductions in LH pulse frequency have been demonstrated when EA using this calculation is < 30 kcal/ kg ffm·d⁻¹.

Purpose

We determined whether menstrual disturbances (MD) are induced when EA is < 30 kcal/ kg ffm·d⁻¹.

Methods

Thirty-five sedentary, ovulatory women 18–24 yr (weight= 59.0 ± 0.8 kg, BMI= 21.8 ± 0.4 kg·m²) completed a diet and exercise intervention over three menstrual cycles. Participants were randomized to groups that varied in the magnitude of negative energy balance created by the combination of exercise and energy restriction. MD were determined using daily urinary estrone-1-glucuronide (E1G) and pregnanediol glucuronide (PdG), mid-cycle LH, and menstrual calendars. In a secondary analysis, we calculated EA from energy balance data and tested the association of EA with menstrual disturbances.

Results

A generalized linear mixed-effects model showed that the likelihood of a MD decreased by 9% for each unit increase in EA (odds ratio 0.91, 0.84 – 0.98, 95% CI, P=0.010). No specific value of EA emerged as a threshold below which MD were induced. When participants were partitioned into EA tertile groups (Low EA =23.4–34.1; n=11, Moderate EA =34.9–40.7; n=12, and High EA =41.2–50.1; n=12 (kcal/ kg ffm·d⁻¹)), E1G (p<0.001), PdG (p<0.001), and luteal phase length (p=0.031) decreased significantly, independent of tertile.

Conclusion

These findings do not support that a threshold of EA exists below which MD are induced but do suggest that MD increase linearly as EA decreases. MD can likely be prevented by monitoring EA using a simplified assessment of metabolic status.

Introduction

Exercise-associated menstrual disturbances are commonly observed in physically active women and female athletes. Prevalence rates vary depending on the type of exercise or sport but have been reported as luteal phase defects (29%), anovulation (20%), oligomenorrhea (7%), and amenorrhea (37%)(1, 2). Clinical sequelae are well documented (3, 4) and include transient infertility(5), low bone mineral density (6, 7), a greater risk of stress fractures(8, 9), and negative alterations in cardiovascular function (10). Research in women and non-human primates on the underlying mechanism of exercise-related menstrual disorders has demonstrated the causal role of low energy availability (EA) in the induction and reversal of exercise- associated menstrual disturbances (11–13).

The suppression of reproductive function by low EA allows for a repartitioning of available energy from the non-essential functions of growth and reproduction toward the essential functions of thermoregulation, locomotion, and cellular maintenance(14). In some studies of the effects of low EA on luteinizing hormone pulsatile secretion, EA has been represented with a particular index, i.e., the difference between the total energy consumed as food and the energy expenditure of exercise normalized for fat free mass (ffm)(15). Short-term (5 days) reductions in EA below a threshold of 30 kcal/ kg ffm·d⁻¹ slowed the pulsatile release of luteinizing hormone (LH), a proxy indicator of hypothalamic gonadotropic-releasing hormone (GnRH) secretion from the anterior pituitary gland (16). A slowing of LH pulse frequency is associated with delays in folliculogenesis and luteal phase shortening and complete menstrual suppression (17–19). Reduced LH pulse frequency in response to low EA occurs whether EA is reduced through diet, exercise, or a combination of both (15). When EA is reduced below 30 kcal/ kg ffm·d⁻¹, metabolic effects that have been observed include reduced serum concentrations of glucose, triiodothyronine, insulin, insulin like growth factor-1, and increased concentrations of growth hormone and cortisol (15, 16). Although these previous studies elegantly identified the importance of EA in the modulation of LH pulse frequency, the magnitude of change in EA that is associated with the induction of menstrual disturbances has not been directly demonstrated through experimentation.

We previously demonstrated that an energy deficit created by the combination of exercise and dietary restriction over three menstrual cycles results in menstrual disturbances in a dose dependent manner such that a greater energy deficit resulted in a greater frequency of menstrual disturbances (20). Luteal phase defects, oligomenorrhea, and anovulation were induced by energy deficits that ranged from −22% to −42% of baseline energy needs and in absolute terms, −470 kcals to −810 kcals below initial energy requirements (20). These results provide practical information about the magnitude of the energy deficit that result in exercise-associated menstrual disturbances. However, the assessment of all the components of energy balance is difficult and expensive and thus not very practical.

Assessing EA as defined by Loucks et al. (15) (16) requires the assessment of 1) 24-hour dietary intake, 2) exercise energy expenditure, and 3) fat free mass, all of which can be calculated by coaches or trainers in a field setting. Moreover, this particular calculation of EA has been recommended for the prevention and management of the Female Athlete Triad (3, 21). Although our previous work documented a causal role of an energy deficit in the induction of menstrual disturbances in exercising women, our results may be more useful if the energy deficit was expressed similarly to the Loucks et al. (15, 16) calculation of EA. Moreover, the induction of menstrual disturbances due to low EA has not been experimentally demonstrated.

The goal of this study was to determine whether low EA , as assessed using the Loucks et al. calculation (15, 16), could induce menstrual disturbances and to explore whether these disturbances only occur below a specific threshold of EA. Another aim was to build on our previous work (20) by reporting the impact of low EA on the urinary concentrations of estrone -1-glucuronide (E1G) and pregnanediol glucuronide (PdG), which are proxy indicators of estradiol and progesterone. Taking advantage of the ability to examine the temporal relation between variables of interest, we also sought to determine whether EA in the previous menstrual cycle as well as in the current menstrual cycle was related to the occurrence of menstrual cycle disturbances. We hypothesized that the incidence of menstrual disturbances will be inversely related to EA, such that a lower EA will be associated with a higher incidence of menstrual disturbances. We also hypothesized that mean E1G and PdG will be linearly related to EA, such that a lower EA will be associated with lower concentrations of these metabolites.

Methods

Experimental Overview

Original Study

This study is a secondary analysis of data collected in our previous study which employed a randomized prospective design to examine the impact of a controlled feeding and supervised exercise intervention on the occurrence of menstrual disturbances among young, untrained, ovulatory women over the course of three menstrual cycles (20). The latter is referred to as “the original study” while this study is referred to as the “current analysis”. The intervention in the original study was initiated after a screening and baseline period. After the baseline cycle, participants were randomly assigned to their experimental groups. Four groups were created representing distinct levels of energy deficiency; one in which participants exercised but energy balance was maintained, and three other groups that represented low, moderate, and high energy deficits created by a combination of exercise and caloric restriction during the intervention. The intervention was subdivided into three menstrual cycles known as intervention cycle 1, 2, and 3 (INT 1, INT 2, and INT 3). Each of the three energy deficit groups (low, moderate, and high) exercised and was prescribed an individual energy intake so that the target negative energy balance ranged from −15% to −60% of baseline energy needs depending on the group. Each of the four study phases i.e., Baseline, INT1, INT2, and INT3 was the length of a given participant’s menstrual cycle. A post-study period of one week where diet and exercise remained controlled allowed for post-intervention measurements.

Current Analysis

For the current secondary analysis, we used the energy balance data from the original study to calculate EA as described by Loucks (15, 16). Daily EA was calculated using that day’s energy intake, net exercise energy expenditure, and the most recent measure of ffm as described below. The calculated energy deficit used in our previous study (20) and our calculated EA values were highly correlated (R=0.88; P<0.001). The average EA for each phase of the study (Baseline, INT1, INT2, and INT3) was calculated as was the EA for entire intervention (INT1 to INT3). These EA data were used in regression analyses. For other analyses, participants were grouped into tertiles based on their EA for the entire intervention, i.e., low, moderate, or high EA. The low EA group consisted of participants whose average EA ranged from 23.4 to 34.1 kcal/ kg ffm·d⁻¹; the moderate EA group consisted of participants whose average EA ranged from 34.9 to 40.7 kcal/ kg ffm·d⁻¹; and the high EA group consisted of participants whose average EA ranged from 41.2 to 50.1 kcal/ kg ffm·d⁻¹.

Experimental Methods in Original Study

Participants

In the original study, inclusion criteria were: 1) no history of serious medical conditions, 2) no current evidence of disordered eating or history of an eating disorder, 3) age 18–30 yr, 4) weight 45–75 kg, 5) body fat 15–35% 6) BMI 18–30 kg·m⁻², 7) non-smoking, 8) no medication use that would alter metabolic hormone concentrations e.g. thyroid medication, glucose lowering drugs, 9) no significant weight loss/gain ± 2.3 kg (± 5 lbs) in the last year, 10) less than one hour of purposeful aerobic exercise/week, 11) not taking hormonal contraceptives for the past 6 months, 12) documentation of at least two ovulatory menstrual cycles during screening. Participants were told the purpose, procedures, and potential risks of participation in the study before they provided their documented informed consent. The original consent was approved by the Penn State University Biomedical Institutional Review Board. Recruitment occurred over three consecutive academic years and drew from the University community during the fall semester of each academic year.

Screening and Baseline

Each participant’s medical history, menstrual history (self-report, prior six months), current and past physical activity (22), eating attitudes and behaviors, anthropometrics, and physical and psychological status were measured during the original study’s screening phase using previously published methods (23). Screening occurred over 2–3 menstrual cycles. All participants completed a menstrual history questionnaire in which they were asked whether they had regular (26–35 days) menstrual cycles during the previous six months, and prior to the previous six months. Participants were excluded if they had a history of menstrual disturbances and or were not eumenorrheic (26–35 days) during the last six months. Over two screening cycles prior to the Baseline period, the participants completed menstrual calendars to monitor menstrual cycle length, menstrual bleeding, and menstrual symptoms. Ovulation was assessed using ovulation detection kits processed in the laboratory on mid cycle urine collections. Blood samples were collected via venipuncture during screening in order to rule out hormonal or metabolic disease. A physical examination was performed by a Clinical Research Center (CRC) clinician, and a clinical interview was performed under the supervision of a clinical psychologist to rule out the history of or any current eating disorders. A CRC dietitian screened participants for inclusion in a controlled feeding study and instructed them on how to complete a three-day food log. Menstrual status was assessed using self-reported menstrual calendars, ovulation detection kits (First Response, Tambrands, Inc.), and three mid-luteal phase serum progesterone measurements (at least one > 5.0 ng/ml) to indicate ovulation obtained via venipuncture during the Baseline cycle (24).

Assessment of Baseline energy needs

Baseline energy needs were assessed during the early follicular phase of the Baseline cycle. Twenty-four-hour energy expenditure was assumed to represent baseline energy needs because all participants were weight stable (± 1.5 kg) during the study’s screening phase. Twenty-four-hour energy expenditure was assessed as the sum of resting metabolic rate (RMR) (kcal/24 hr) and the energy expended that were attributable to that day’s physical activity. Resting metabolic rate was measured during the follicular phase of the Baseline cycle using indirect calorimetry. Expenditure due to physical activity was measured with an accelerometer (RT3, Stayhealthy, Monrovia, CA), where the device was worn 24 hr per day, 7 days per week (except during training bouts, sleeping, swimming, and bathing), during the follicular phase. This sum (RMR kcal/24 hr and RT3 physical activity kcal/24 hr) was operationally defined as the participant’s baseline energy needs (25). The quantity of food intake during baseline was adjusted if the participant’s body weight varied during a one week “calibration” period (see below).

Determination of energy intake during the intervention

Energy intake for each participant assigned to the original study groups was calculated as the percentage change in baseline energy needs. All food consumed over the course of the study was prepared by CRC staff in a metabolic kitchen where food was weighed to the nearest gram to ensure the desired energy level. Energy content of prepared food items was verified with bomb calorimetry as previously described (26). All participants were required to eat at least two of their three daily meals at the CRC dining room while the other daily meal, all weekend meals and a daily snack were packed out. The CRC dietary staff followed an eight-day meal rotation to ensure diet variety. The prescribed diet was comprised of 55% carbohydrates, 30% fat, and 15% protein. To ensure that the energy content of food reflected baseline energy needs, a one week calibration period was initiated where ±100 kcal adjustments were made if the participant’s weight changed more than ± 0.5 kg during the week. After the Baseline cycle, the energy content of prescribed food was adjusted according to the participant’s assigned group. Participants were instructed to eat all and only the foods provided to them. If participants were unable to eat all of the food provided, uneaten food was returned to the laboratory where it was weighed by a CRC dietitian and the alteration in energy intake was taken into account. Participants were discouraged from eating food that had not been provided by the study. If participants did eat food that had not been provided, they were instructed to record what they consumed and this amount was added to the calculation of dietary intake. Total intake for each 24-hour day was analyzed for macronutrient composition and energy content with Nutritionist Pro (First Data Bank, Indianapolis, IN).

Exercise training

Exercise training took place at Noll Laboratory and was supervised by trainers. Aerobic exercise was performed once per day 5 times/week at 70–80% of maximum heart rate as determined from tests of maximal aerobic capacity (VO₂ max). The duration of exercise was dependent on the energy expenditure prescription for a given participant and ranged from 20–75 min/day across the intervention. Energy expended during each exercise session ranged from 350–550 kcal/day and was quantified using the OwnCal feature (27) on the Polar S610 heart rate monitor (Polar Electro Oy, Kempele, Finland). This method was superior to the use of accelerometers in order to capture energy expended during modes of exercise such as stationery cycling, elliptical and treadmill running with a grade. These monitors were continually updated with the most recent values for participant weight, VO₂ max, and age. Several different modes of exercise were utilized, including treadmill running, elliptical machine exercise, stair stepping, and stationary cycling. Accelerometers were removed after participants were equipped with heart rate monitors for a given exercise session and returned and worn when heart rate monitors were removed at the conclusion of an exercise session.

Anthropometrics

Body weight was measured to the nearest 0.01 kg using a digital scale (Seca; Hamburg, Germany) by a CRC dietitian. The CRC dietitian measured each participant’s body weight in the morning (before eating) twice per week while participants wore standard shorts and a tee shirt. Body composition was measured during the first 7 days of baseline, INT 2, and INT 3 phases, and during the post-study period using hydrostatic weighing after correcting for residual lung volume (28).

Resting metabolic rate

Resting metabolic rate was measured in the early follicular phase between 0600 and 1000 h after an overnight fast as previously published (29). Upon arriving at the laboratory, the participant would lie in a supine position on a bed for 20–30 min to acclimate to room temperature and undergo familiarization with the equipment and procedures. A ventilated hood was then placed over the participant’s head for 30 min. Expired air was analyzed each minute for carbon dioxide and oxygen concentration using a carbon dioxide analyzer (URAS 4, Hartmann & Braun, Frankfurt, Germany) and a paramagnetic oxygen analyzer (Magnos 4G, Hartmann & Braun). The values for minutes in which steady state was achieved were averaged and the resting metabolic rate in kilocalories per minute was determined using the Weir equation (30).

Aerobic capacity

A laboratory technician measured VO₂ max during the screening, baseline, INT 1, INT 2, and INT 3 phases, and during the post-study period with an incremental treadmill test (modified Åstrand protocol (31)) to volitional fatigue using indirect calorimetry (Sensormedics metabolic cart model no. 229, Sensormedics Corp., Yorba Linda, CA).

Blood sampling

Screening blood samples were collected during the first week of the menstrual cycle between 0700 hr and 1000 hr in the CRC. Baseline blood samples were collected on three days during the middle of the luteal phase to confirm ovulation in real time during that cycle (the day of ovulation for the purposes of this analysis was measured with the use of urinary LH, as described below). A CRC nurse collected all blood samples after participants lay supine for 15 minutes and had fasted for 12 hours. Samples were allowed to clot at room temperature and were then centrifuged for collection of serum.

Menstrual status and daily urine collection

Menstrual status was monitored with menstrual calendars and daily urine samples during baseline, INT 1, INT 2, and INT 3 cycles using previously published methods (23). Menstrual calendars were kept by the participants who recorded the days of menstrual bleeding, the severity of bleeding, and the occurrence of cramps or other menstrual symptoms i.e., bloating, spotting. Follicular phase length was assessed as the number of days from the first day of menses to the day of ovulation; luteal phase length was assessed as the number of days from the day after ovulation to the day preceding the first occurrence of menses. Confirmation of ovulation, the presence or absence of menstrual disturbances, and the lengths of the follicular and luteal phases were assessed by analyzing daily urine samples for the urinary metabolites E1G, PdG, and LH. Menstrual disturbances that were identified included luteal phase defects, anovulation, and oligomenorrhea. Luteal phase defects were ovulatory menstrual cycles which had short luteal phases (<10 days) and/or inadequate luteal phases utilizing urinary PdG concentrations (peak PdG <5 mcg/ml) (1, 32). Anovulatory cycles were cycles lacking an adequate preovulatory E1G peak, an absence of a midcycle LH surge, and no luteal rise in PdG above 2.49 mcg/ml (1, 32). Oligomenorrheic cycles were cycles that had a length of 36 or more days (21).

Biochemical analyses

A complete blood count, chemistry panel, and an endocrine panel were completed by Quest Diagnostics (Lyndhurst, NJ). The endocrine panel included measurements of LH, follicle stimulating hormone (FSH), prolactin, progesterone, estradiol, thyroxine and thyroid-stimulating hormone. Progesterone was measured in our laboratory using a radioimmunoassay on the serum samples obtained during the mid-luteal phase of the Baseline cycle (Diagnostics Products Corporation, Los Angeles, CA). The sensitivity of the progesterone assay is 0.02 ng/ml. The intra-assay and inter-assay coefficients of variation for the high controls were 2.7% and 3.9%, respectively; the intra-assay and inter-assay coefficients of variation for the low controls were 8.8% and 9.7 %. Competitive enzyme immunoassays were used to measure the major urinary estrogen and progesterone metabolites, E1G and PdG, according to previously published methods (33). Urinary LH was measured using a radioimmunoassay (Diagnostic Products Corporation, Los Angeles, CA). The sensitivity of the LH assay is 0.15 mIU/ml. The intra-assay and inter-assay coefficients of variation for the high controls were 1.0% and 3.4%, respectively; the intra-assay and inter-assay coefficients of variation for the low controls were 1.6% and 7.1%. All biochemical analyses for a given participant were measured in the same assay and all samples were run in duplicate.

Methods for the Current Analysis

Calculation of EA during the study’s intervention phases

EA for a given day was quantified as:

$$\begin{gathered} (\text{Energy}\text{Intake}(\text{kcals})-(\text{Exercise}\text{Energy}\text{Expenditure}(\text{EEE})(\text{kcals})-(\text{RMR} \\ (\text{kcals}/\text{minute})\ast \text{duration}\text{of}\text{exercise}\text{bout}(\text{minutes}))/\text{ffm}(\text{kg}). \end{gathered}$$

Net exercise energy expenditure (EEE) was calculated each exercise day by subtracting the energy expenditure that was attributable to RMR from the total exercise energy expenditure during the exercise period. EA was calculated on non-exercise days and EEE was assumed to be zero. The daily EA values were averaged for each study phase (INT 1, INT 2, INT 3). The above method to calculate EA is very similar to the method of Loucks et al. (15, 16), although there is a slight difference. Loucks et al. (15, 16) calculations represent only exercise days, where as our calculation includes non-exercise days as well as exercise days. Secondly, exercise energy expenditure is defined as that which is over and above an estimate of the typical non-exercise expenditure (NEE) occurring at that time of day, as if participants were not engaging in purposeful exercise, i.e.,

Loucks (15, 16) EA calculation for a given day:

$$(\text{EI}(\text{kcals})-(\text{Exercise}\text{Energy}\text{Expenditure}(\text{EEE})(\text{kcals})-\text{expenditure}\text{of}\text{usual}\text{waking}\text{activities}(\text{kcals}))/\text{ffm}(\text{kg}).$$

The latter approach results in slightly smaller values for EEE than the method we employed, which subtracts only the energy expended due to rest during the exercise bout from the total energy cost of the exercise bout. To test whether these differences are meaningful, we calculated EA using Loucks’ et al. method for the 5 participants with the longest average exercise bout durations. NEE was calculated with the use of accelerometers (RT3, Stayhealthy, Monrovia, CA), where the device was worn 24 hr per day, 7 days per week (except during training bouts, sleeping, swimming, and bathing), every other week. We found the difference between our calculation of EA using only RMR to adjust EEE, and Loucks’ et al. method, which used NEE to adjust EEE, resulted in at most an average of 1.3 kcals/ kcal/ kg ffm·d⁻¹ difference in EA when EA was averaged over a given intervention phase, and at most an average of 1.2 kcal/ kg ffm·d⁻¹ difference in EA when EA was averaged over the whole intervention, with the calculations used in this analysis resulting in higher EA values. Because of this small difference, we believe our calculation of EA is comparable to that of Loucks et al. (15, 16).

Statistical analysis

All continuous variables were assessed for normality and the presence of outliers before any statistical analyses were performed. All PdG data and follicular and luteal phase length data were log transformed as the data were not normally distributed. We analyzed the data provided by participants (n= 35) who completed all phases of the study. We imputed the data (used INT 2 data for INT 2 and INT 3) collected on one participant who had a 59-day INT 2 cycle and had to leave the study due to other commitments. Specifically, this participant’s INT 2 cycle was repeated as her INT 3 cycle. Linear regression was used to compare participants’ mean percent energy deficit during the intervention with their calculated mean intervention EA. Analyses of anthropometric, VO₂ max, EA, menstrual cycle length, E1G, and PdG data were conducted using one-between, one-within repeated measures analysis of variance, with participant EA group as the between-participant variable and time and the group by time interaction were the within-participant variables. Post-hoc testing for group mean differences was performed using Fisher’s least significant difference (LSD). Post-hoc testing for within-group changes in EA over time was completed with paired t-tests adjusted with the Bonferroni correction.

We fit a generalized linear mixed-effects model to test whether previous cycle (lag-1) and (or) current cycle EA were significant predictors of the presence of menstrual disturbances during each intervention phase (INT 1, INT 2, and INT 3) of the study. Previous cycle EA (lag-1) and current cycle EA were treated as fixed effects. The response variable was menstrual disturbance; it was coded as a 0/1 binary variable if it did not or did occur in a given cycle. The model was fit to data on all three cycles from all participants (35 participants, n=105 observations). To account for the dependent data provided by each participant, participant identification number was incorporated into the model as a random effect. Data were analyzed using IBM SPSS Statistics for Windows Version 22.0, Armonk, NY.

Results

Participants

Participants were healthy, sedentary (<1 hour of purposeful exercise/ week) women, aged 18–24 yrs. Participants’ baseline characteristics are presented in Table 1.

Table 1.

Subject characteristics at baseline (n=35)

Variable	Baseline	Range
Demographics
Age (yr)	20.3 ± 0.3	(18.0 – 24.0)
Anthropometrics
Height (cm)	165 ± 1	(156 – 180)
Body Weight (kg)	59.0 ± 0.8	(51.1 – 68.6)
Body Mass Index (kg/m2)	21.8 ± 0.4	(18.2 – 27.7)
Body Fat (%)	28.5 ± 0.8	(18.8 – 38.5)
Fat Mass (kg)	16.9 ± 0.6	(9.9 – 26.4)
Fat Free Mass (kg)	42.1 ± 0.5	(37.3 – 49.6)
Reproductive Characteristics
Age of Menarche (yr)	12.1 ± 0.2	(9 – 15)
Gynecological Age (yr)	8.2 ± 0.4	(5 – 15)
Cycle Length (days)	29.0 ± 0.4	(25 – 35)
Aerobic Capacity
VO2 max (ml/kg · min−1)	37.5 ± 0.8	(27.4 – 49.5)

Data are mean ± SEM.

Changes in EA

Changes in study phase EA by EA group are presented in Figure 1. A one-between, one-within repeated measures ANOVA revealed significant time (P<0.001), group (P<0.001), and group x time effects (P=0.041) for changes in study phase EA. Post-hoc testing using paired t-tests adjusted with the Bonferroni correction revealed no significant change in EA in the high EA group; that EA decreased significantly from INT 1 to INT 2 (42.3 vs. 38.2 kcal/ kg ffm·d⁻¹; P=0.004) and remained lower relative to INT 1 in INT 3 (42.3 vs. 35.7 kcal/ kg ffm·d⁻¹; P=0.003) in the moderate EA group; and that EA decreased significantly from INT 1 to INT 2 (32.5 vs. 25.3 kcal/ kg ffm·d⁻¹; P<0.001) and remained lower relative to INT 1 in INT 3 (32.5 vs. 25.0 kcal/ kg ffm·d⁻¹; P<0.001) in the low EA group.

Figure 1.

Changes in energy availability during the intervention

Repeated measures ANOVA revealed significant time (P<0.001), group (P<0.001), and group x time effects (P=0.041) for changes in study phase EA; Post hoc testing showed no differences over time in High EA; ^aEA decreased from INT 1 to INT 2 (42.3 vs. 38.2 kcal/ kg ffm·d⁻¹; P=0.004) and remained lower relative to INT 1 in INT 3 (42.3 vs. 35.7 kcal/ kg ffm·d⁻¹; P=0.003) in the Moderate EA group; ^bEA decreased from INT 1 to INT 2 (32.5 vs. 25.3 kcal/ kg ffm·d⁻¹; P<0.001) and remained lower relative to INT 1 in INT 3 (32.5 vs. 25.0 kcal/ kg ffm·d⁻¹; P<0.001) in the Low EA group; data are mean ± SEM;

Anthropometrics and aerobic capacity

Changes in anthropometric and aerobic capacity data are presented in Table 2. There were no statistically significant Baseline EA group mean differences with respect to body weight, BMI, fat free mass, percent body fat, fat mass, age of menarche, gynecological age, cycle length, or VO₂ max. A one-between, one-within repeated measures ANOVA indicated that body weight decreased over time (P<0.001); however, this decrease was not significantly different among EA groups. Percent change in weight was −5.7 ± 1.3% in the low EA group, −4.1 ± 1.1% in the moderate EA group, and −3.6 ± 0.9% in the high EA group. Separate one-between, one-within repeated measures ANOVA indicated that percent body fat, fat mass, and BMI all declined significantly over time (P<0.001); however, these decreases were not significantly different among EA groups. In addition, a one-between, one-within repeated measures ANOVA indicated aerobic capacity increased in all groups (P<0.001); however, this increase was not significantly different among EA groups.

Table 2.

Effects of the intervention on anthropometric variables and aerobic capacity by energy availability group

Group	Phase	Body Weight (kg)	Body Mass Index (kg/m2)	Percent Body Fat (%)	Fat Mass (kg)	Fat Free mass (kg)	VO2 max (ml/kg · min−1)
Low Energy Availability	Baseline	59.3 ± 1.3	22.0 ± 0.6	28.6 ± 1.2	16.9 ± 0.8	42.3 ± 1.2	38.1 ± 1.0
	INT 2	56.3 ± 1.3	20.9 ± 0.5	25.5 ± 1.2	14.4 ± 0.8	41.9 ± 1.0	46.1 ± 1.5
	Post	55.9 ± 1.3	20.7 ± 0.6	24.7 ± 1.3	13.8 ± 0.8	42.1 ± 1.1	45.3 ± 1.5
	%Change	−5.7 ± 1.3	−5.7 ± 1.3	−13.6 ± 2.6	−18.3 ± 3.2	−0.5 ± 1.3	19.2 ± 4.2
	N	11	11	11	11	11	10
Moderate Energy Availability	Baseline	58.2 ± 1.4	21.9 ± 0.6	27.8 ± 1.5	16.3 ± 1.1	41.9 ± 0.7	38.4 ± 1.8
	INT 2	56.3 ± 1.2	21.2 ± 0.6	26.2 ± 1.4	14.8 ± 1.0	41.4 ± 0.8	43.3 ± 0.7
	Post	55.8 ± 1.2	21.0 ± 0.6	25.7 ± 1.4	14.4 ± 1.0	41.4 ± 0.8	44.3 ± 1.2
	%Change	−4.1 ± 1.1	−4.0 ± 1.1	−7.1 ± 2.6	−10.5 ± 3.4	−1.3 ± 0.7	17.2 ± 4.8
	N	12	12	12	12	12	11
High Energy Availability	Baseline	59.5 ± 1.5	21.6 ± 0.8	29.1 ± 1.5	17.4 ± 1.3	42.0 ± 0.7	37.4 ± 0.9
	INT 2	58.3 ± 1.5	21.1 ± 0.8	27.0 ± 1.4	15.9 ± 1.2	42.3 ± 0.7	42.3 ± 1.1
	Post	57.3 ± 1.6	20.8 ± 0.9	26.6 ± 1.5	15.4 ± 1.3	41.7 ± 0.8	43.2 ± 1.4
	%Change	−3.6 ± 0.9	−3.6 ± 0.9	−8.5 ± 2.1	−11.7 ± 2.8	−0.7 ± 0.8	15.6 ± 3.4
	N	12	12	12	12	12	11
P-values	Group	0.677	0.972	0.755	0.669	0.890	0.336
	Time	<0.001	<0.001	<0.001	<0.001	0.191	<0.001
	Time X Group	0.179	0.149	0.239	0.226	0.329	0.443

Results of ANOVA with repeated measures comparing Baseline, Intervention Cycle 2 (INT 2), and Post; data are mean ± sem; Low Energy Availability (EA) (mean intervention EA= 27.8 ± 0.9 kcals/kg ffm · d⁻¹, range: 23.4–34.1 kcals/kg ffm · d⁻¹), Moderate EA (mean intervention EA= 38.6 ± 0.6 kcals/kg ffm · d⁻¹, range: 34.9–40.7 kcals/kg ffm · d⁻¹), High EA (mean intervention EA= 44.8 ± 0.9 kcals/kg ffm · d⁻¹, range: 41.2–50.1 kcals/kg ffm · d⁻¹);

Changes in menstrual cyclicity

A generalized linear mixed-effects model fit to menstrual disturbances revealed current cycle EA was a statistically significant predictor of the presence of a disturbance in a given menstrual cycle such that each unit decrease in EA was associated with a 9% increase (odds ratio 0.91, 0.84 – 0.98, 95% CI) in the likelihood of experiencing a menstrual disturbance (P=0.010). Previous cycle EA (lag-1) was not a significant predictor of the presence of a menstrual disturbance in a given menstrual cycle. Overall, the sensitivity of the model was 40.0% (model correctly predicted the presence of a menstrual disturbance) and the specificity of the model was 82.5% (model correctly predicted the absence of a menstrual disturbance).

Figure 2 represents the relation between EA, menstrual disturbances, and the EA threshold of kcal/ kg ffm·d⁻¹ previously demonstrated to disrupt LH pulsatility (16). Of note is the frequent occurrence of menstrual disturbances above the threshold of 30 kcal/ kg ffm·d⁻¹. Notably, when the results of our generalized linear mixed-effects model fit to menstrual disturbances are graphed, an EA of less than 30 kcal/ kg ffm·d⁻¹ is associated with a predicted probability of a menstrual disturbance exceeding 50%.

Figure 2.

Individual energy availability data and the incidence of menstrual disturbances during intervention (INT) cycles, a) INT 1, b) INT 2, c) INT 3, and d) all intervention cycles combined

Y axes notations are 0=No menstrual disturbance i.e., ovulatory cycle; 1=at least one menstrual disturbance i.e., luteal phase disturbance, oligomenorrhea, or anovulation; each dot represents one participant’s average energy availability and whether that intervention cycle had none or at least one menstrual disturbance. Vertical line denotes the theoretical threshold energy availability of 30 kcal/ kg ffm·d⁻¹ demonstrated by Loucks at al. (16)

Luteal phase defects were the most common disturbance during the intervention (57% of total disturbances), followed by anovulation (28% of total disturbances). Oligomenorrheic cycles (15% of total disturbances) were the least common disturbance. A one-between, one-within repeated measures ANOVA indicated there was no change in cycle length or follicular phase length over time (P<0.001); and that there was no difference in these lengths among EA groups. In addition, a one-between, one-within repeated measures ANOVA indicated luteal phase length declined significantly over time (P=0.031); however, this decrease was not significantly different among EA groups.

Urinary hormone concentrations

Table 3 presents changes in mean cycle, follicular, and luteal E1G and PdG by EA group during the four study phases. Figure 3 displays composite graphs representing changes in EA group urinary E1G, PdG, and LH concentrations during Baseline and INT 3.

Table 3.

Effects of the intervention on urinary E1G and PdG by energy availability group

Group	Phase	Cycle Length (days)	Follicular Phase Length* (days)	Luteal Phase Length* (days)	AVG. E1G (ng/ml)	AVG. PdG* (μg/ml)	Follicular E1G (ng/ml)	Luteal E1G (ng/ml)	Follicular PdG*(μg/ml)	Luteal PdG* (μg/ml)
Low Energy Availability	Baseline	29.3 ± 0.5	17.3 ± 1.1	12.3 ± 0.7	35.3 ± 4.3	10.4 ± 2.4	33.8 ± 4.6	45.2 ± 5.5	6.0 ± 1.5	17.2 ± 5.0
	INT 1	28.8 ± 0.6	17.8 ± 1.0	11.9 ± 0.6	36.8 ± 3.6	12.2 ± 2.7	36.2 ± 3.2	47.9 ± 5.9	5.9 ± 1.6	23.9 ± 6.8
	INT 2	29.1 ± 1.1	19.6 ± 1.8	10.7 ± 1.7	32.5 ± 3.4	9.1 ± 2.2	33.9 ± 2.7	43.7 ± 6.0	6.7 ± 2.2	17.0 ± 4.3
	INT 3	28.8 ± 1.6	17.8 ± 1.3	10.3 ± 1.5	29.9 ± 3.6	7.9 ± 1.9	31.2 ± 3.1	39.2 ± 6.8	4.2 ± 1.1	18.2 ± 5.5
	%Change	−1.7 ± 5.2	4.4 ± 7.8	−16.2 ± 13.5	−13.6 ± 5.1	−8.9 ± 20.3	0.1 ± 10.2	−14.8 ± 7.0	−29.3 ± 6.4	48.9 ± 66.9
	N	10	8	7	10	10	8	7	8	7
Moderate Energy Availability	Baseline	28.3 ± 0.5	16.8 ± 0.6	11.0 ± 0.4	30.7 ± 3.2	6.7 ± 1.3	21.9 ± 1.3	35.4 ± 4.3	4.2 ± 0.6	13.5 ± 2.9
	INT 1	27.8 ± 1.1	17.6 ± 0.9	10.6 ± 0.7	29.0 ± 1.7	7.8 ± 1.8	24.2 ± 1.4	33.2 ± 2.0	3.9 ± 0.4	16.3 ± 4.9
	INT 2	26.2 ± 0.9	17.8 ± 1.1	8.7 ± 1.2	28.0 ± 2.5	5.0 ± 0.9	23.3 ± 1.7	37.7 ± 4.6	4.2 ± 0.7	10.4 ± 2.1
	INT 3	27.5 ± 0.5	18.1 ± 1.0	9.8 ± 0.9	26.9 ± 2.1	4.4 ± 0.8	22.5 ± 2.5	30.9 ± 2.8	3.8 ± 0.3	8.3 ± 1.5
	%Change	−2.5 ± 2.8	8.1 ± 5.1	−10.0 ± 9.1	−8.4 ± 5.8	−24.7 ± 12.8	5.3 ± 13.4	−6.9 ± 8.1	3.1 ± 17.0	−22.8 ± 15.3
	N	12	9	9	12	12	7	9	8	9
High Energy Availability	Baseline	29.0 ± 0.6	16.3 ± 0.6	12.6 ± 0.5	28.5 ± 1.8	10.1 ± 1.3	26.8 ± 1.8	30.5 ± 2.2	4.2 ± 0.5	16.4 ± 2.2
	INT 1	30.4 ± 1.4	17.7 ± 1.2	12.6 ± 0.6	28.8 ± 2.3	8.1 ± 1.0	26.7 ± 2.0	28.0 ± 2.9	3.6 ± 0.7	13.1 ± 1.7
	INT 2	28.2 ± 0.9	16.1 ± 0.9	12.1 ± 0.3	25.1 ± 1.4	6.7 ± 0.8	23.4 ± 1.8	26.5 ± 1.8	2.8 ± 0.3	11.6 ± 1.9
	INT 3	27.1 ± 0.9	16.4 ± 1.1	11.3 ± 0.4	23.3 ± 1.8	6.7 ± 1.1	24.2 ± 1.9	27.6 ± 2.6	3.5 ± 0.9	13.0 ± 2.1
	%Change	−6.1 ± 3.9	2.0 ± 8.8	−10.3 ± 3.5	−16.9 ± 5.2	−26.4 ± 17.5	−8.6 ± 6.4	−10.1 ± 4.1	−10.5 ± 22.4	−10.2 ± 21.4
	N	11	9	8	11	11	8	8	9	9
P-values	Group	0.112	0.328	0.124	0.148	0.153	0.007a	0.024b	0.466	0.385
	Time	0.190	0.645	0.031	<0.001	<0.001	0.080	0.071	0.067	0.015
	Time X Group	0.454	0.713	0.519	0.590	0.490	0.650	0.279	0.237	0.336

Results of ANOVA with repeated measures on urinary estrone-1- glucuronide (E1G) and pregnanediol-glucuronide (PdG) comparing energy availability (EA) groups at Baseline, Intervention Cycle 1, Intervention Cycle 2, Intervention Cycle 3; data are mean ± sem; p-values are for main effects (group, time) and interactions (time x group); Low EA (mean intervention EA= 27.8 ± 0.9 kcals/kg ffm · d⁻¹, range: 23.4–34.1 kcals/kg ffm · d⁻¹), Moderate EA (mean intervention EA= 38.6 ± 0.6 kcals/kg ffm · d⁻¹, range: 34.9–40.7 kcals/kg ffm · d⁻¹), High EA (mean intervention EA= 44.8 ± 0.9 kcals/kg ffm · d⁻¹, range: 41.2–50.1 kcals/kg ffm · d⁻¹); %change is from Baseline to INT3.

^*Data are log transformed

^aLow EA > Moderate and High EA; p<0.0167

^bLow EA > High EA; p<0.0167

Figure 3.

Effects of the intervention on urinary estrone-1-glucuronide (E1G), pregnanediol glucuronide (PdG), and luteinizing hormone (LH) concentrations in energy availability groups; data are mean ± sem for each cycle day; cycle day “0” is day of ovulation

Separate one-between, one-within repeated measures ANOVA indicated mean cycle E1G and PdG, and luteal PdG length declined significantly over time (P<0.001), however, this decrease was not significantly different among EA groups. Finally, additional one-between, one-within repeated measures ANOVA analyses indicated there were significant group effects for both follicular (P=0.007) and luteal (P=0.024) phase E1G. Fisher’s LSD post-hoc test indicated follicular E1G concentrations were greater in the Low EA group than in the Moderate EA group (P=0.003), and luteal E1G was greater in the Low EA group than in the High EA group (P=0.007).

Discussion

The major finding of this study was that EA calculated in a similar manner to that in short-term studies of changes in LH pulsatility was also a significant predictor of menstrual disturbances during a controlled, feeding-and-exercise intervention in previously untrained, normally menstruating, young women. This finding is important because consensus statements and position stands frequently incorporate the Loucks et al. index of EA into practical guidelines for prevention and treatment of the Female Athlete Triad even though no experiments have established the significance of this index with respect to ovarian function per se (3, 21). However, in contrast to the studies of LH pulsatility conducted by Loucks et al. (15, 16), we observed no discernible EA threshold below which menstrual disturbances were initiated. As EA dropped below approximately 30 kcal/ kg ffm·d⁻¹, the probability that a participant experienced a menstrual disturbance exceeded 50 percent. Finally, our diet-and-exercise intervention had a suppressive effect on E1G excretion and on PdG excretion during the intervention that was not dependent on EA.

This work expands on earlier work from our laboratory (20) which demonstrated that a diet-and-exercise intervention that produced varying levels of energy deficiency induced menstrual disturbances in a dose dependent manner in previously untrained women. In the current study, we show that a particular calculation of EA that differs from the calculation of energy deficit used by Williams et al. (20) is also a significant predictor of menstrual disturbances. This finding is important because overall energy deficit can often be cumbersome to calculate in field settings due to the difficulties quantifying all three components of energy expenditure (RMR, non-exercise activity thermogenesis (NEAT), and exercise expenditure). EA is easier to calculate because it requires only the purposeful exercise component of 24-hour energy expenditure (34).

Interestingly, our analysis showed that the average percent deficit calculated using all components of energy balance as in Williams et al. (20) was highly correlated with mean EA. This finding indicates that daily non-exercise energy expenditure and RMR remained fairly consistent when menstrual disturbances were newly induced with modest energy deficits created through the combination of energy restriction and exercise. Moreover, a 0 % deficit was equivalent to an EA of approximately 45 kcal/ kg ffm·d⁻¹ which is assumed by Loucks et al. to approximate energy balance (34). The relation between overall energy balance and EA calculated similarly to Loucks et al. has never been reported. This finding is important because it confirms that the EA calculation used by Loucks and replicated in this study provides a meaningful assessment of overall energy balance when used over several months. However, adaptations in components of energy expenditure to conserve energy for more vital bodily processes such as a reduction in RMR can occur over a sustained period of energy deficiency (35). Such compensatory adaptations restore energy balance even though an individual is "energy deficient" relative to a healthy energy status (36). The Loucks et al. calculation of EA for preventing and treating menstrual disturbances in an active population is particularly useful because it is unaffected by compensatory homeostatic mechanisms such a decline in RMR which can occur with severe energy deficits.

We observed that EA in a given study phase was a significant predictor of menstrual disturbances while previous cycle EA was not. This finding is consistent with several studies using both animal and human models in that the effects of changes in EA occur rapidly (16, 37, 38). Following glucose deprivation in sheep, there was an almost immediate suppression of pituitary LH secretion (38). Similarly, a one-day fast of male Rhesus monkeys reduced both LH and testosterone pulse frequency (37). The aforementioned work by Loucks et al.(15, 16) demonstrated reduced LH pulse frequency in young eumenorrheic women following 5-days of varying levels of low EA. Finally, in a previous study, Williams et al. demonstrated that initiating strenuous exercise at the onset of a menstrual cycle could induce luteal phase dysfunction within the same menstrual cycle, and that exercise confined to the luteal phase could also induce luteal dysfunction (39). These findings demonstrate that the hypothalamic-pituitary-ovarian (HPO) axis can be rapidly disrupted by low EA while carry-over effects of EA during the previous cycle appear to be minimal. Our results also support the established importance of readily available oxidative fuels versus energy stores in the induction of menstrual disturbances and help explain why menstrual disturbances can occur across a wide range of body fat levels (14, 40).

Our inability to find an EA threshold that induces menstrual disturbances is consistent with a cross sectional finding reported by Reed at al. (41). The latter reported that the EA threshold of 30 kcal/ kg ffm·d⁻¹ could not distinguish eumenorrheic individuals from individuals with subclinical menstrual disturbances (i.e., LPD, anovulation). Similarly, Guebels et al. showed that when EA was calculated using four different cut-offs of exercise intensity for the EEE that was subtracted from EI, EA was not different between amenorrheic and eumenorrheic women (42). Differences between our own findings and those of the short-term studies by Loucks may be due to several factors. A threshold may exist for the short term effects of metabolic stressors on LH pulsatility but not for more chronic declines in gonadotropin support that would begin to impact ovarian function over a longer time frame. The present study’s longer exposure to low EA may have been accompanied by other factors that potentially mediate or moderate the disruption of ovarian function but do not behave in the way EA and LH pulsatility do in terms of a threshold effect. That is, these factors could have different thresholds and or dose response characteristics in relation to menstrual disruption. Lastly, the sensitivity of an individual to ovarian disruption may vary more than one’s sensitivity to alterations in LH pulse dynamics. A variety of stressors can mediate the central suppression of the reproductive axis, including psychosocial factors such as the perception of life stressors and or psychological disorders (43, 44), or social stressors such as a death in the family or relationship challenges (45). Other factors such as gynecological age may moderate the effects of stress on the reproductive axis (46, 47).

While the underlying mechanism remains unclear, our data suggest that EA values above 30 kcal/ kg ffm·d⁻¹ are associated with a suppression of LH pulsatility and subsequent menstrual dysfunction when experienced chronically. Importantly, these findings do not support the use of messaging by clinicians and sports medicine practitioners that a threshold of EA of 30 kcal/ kg ffm·d⁻¹ exists for preventing exercise-associated menstrual disturbances. Rather, the calculation of EA should be considered to be on a sliding scale for which the risk of experiencing a menstrual disturbance increases as EA decreases. Specifically, our results show as EA drops below 30 kcal/ kg ffm·d⁻¹, the predicted probability of experiencing a menstrual disturbance exceeded 50 percent.

Our study demonstrates the utility of the EA calculation in the prediction of menstrual disturbances. However, its use by practitioners or individuals outside a laboratory as a tool to make decisions about healthy dietary strategies is limited by the accuracy of methods to assess the components (i.e., dietary intake, exercise energy expenditure, and body composition). As well, the measure is best used in conjunction with other measures of energy availability, such as one’s history of and current eating behaviors and attitudes and history of body weight changes (3). Multiple measures of EA in the same individual should be recommended to observe trends, and a target and sustained EA of kcal/ kg ffm·d⁻¹ is a wise recommendation to prevent clinical sequelae associated with the Female Athlete Triad.

Of interest is the finding that exercise seemed to have a progressive inhibitory effect on reproductive hormone excretion. This finding is observed qualitatively in Figure 3 where there was a blunting of E1G, PdG, and LH excretion in the three EA groups as the intervention progressed. This overall decline in mean cycle concentrations of urinary E1G and PdG and mean luteal PdG concentrations was statistically significant for all groups. This finding extends reports from other studies which have shown that exercise has a suppressive effect on sex hormone concentrations in women (48, 49). Specifically, the current study demonstrated that the effects of exercise on the menstrual cycle were independent of effects of EA.

Although our study provides important insights into the impact of EA on menstrual function, there are limitations to be considered. Our original study was not designed to test the effects of EA as calculated by Loucks et al. (16) on menstrual function and thus our results rely on a secondary analysis of our original energy balance data. As well, our secondary analysis did not replicate the specific levels of EA studied by Loucks et al. (16). Additionally, our measures of energy balance are associated with sources of potential error and variability. For example, participants may have consumed food outside the study. Although this could have impacted our EA calculations, we believe that the variability attributable to this is likely to be relatively modest because participants reported a high compliance to their respective energy prescriptions. The use of heart rate monitors to measure EEE is not as accurate when compared to indirect calorimetry (9). Lastly, each EA group included participants with varying combinations of energy restriction and exercise prescriptions and thus subtle differences in effects on ovarian function due to how an energy deficit is created, (i.e., by exercise or energy restriction), cannot be discerned. Our study had many strengths including the randomized, prospective design, the control of both the dietary intake and exercise training, and the detailed characterizations of menstrual disturbances using daily urinary measures. Our secondary analysis included detailed and daily calculations of EA over the entire intervention using updated assessments of fat free mass and daily measures of actual dietary energy intake and exercise energy expenditure. There are currently no other reports of prospective laboratory studies testing the validity of EA calculations in association with exercise related menstrual disturbances.

In conclusion, this study addresses an important gap in the literature by demonstrating the utility of a particular and relatively easy calculation of EA for assessing the risk of menstrual disturbances in active young women. EA was a highly significant predictor of menstrual disturbances during the intervention, but a specific threshold for induction of menstrual disturbances was not demonstrated. Future studies should test whether EA predicts changes in menstrual status over an even longer period of time and should include the reversal of menstrual disturbances in amenorrheic and oligomenorrheic exercising women. Moreover, the utility of field as opposed to laboratory based measurements of EA need to be validated as a means of identifying exercising women at risk for menstrual disturbances.