For the present investigation, we analyzed samples from a randomized controlled trial assessing the impact of an ED with and without exercise on metabolic and behavioral adaptations [24, 25]. To test the independent and combined effects of an ED and exercise, the study was conducted using a four-way cross-over design during which participants underwent two 4-day conditions of ED and two 4-day control conditions in energy balance (CON) (Fig. 1). During one ED and one CON condition, participants conducted aerobic exercise (ED + EX; CON + EX). During the other ED and CON conditions, no exercise was conducted (ED-EX; CON-EX). To generate isocaloric conditions within ED and CON, dietary energy intake was adjusted for the energy expended during exercise. The order of the experimental conditions was randomly assigned. After each condition, participants underwent wash out periods of ad libitum food intake. Wash-out periods were set to at least 4 days following CON and at least 10 days following ED conditions. The study was approved by the ethical review board of the German Sport University Cologne (no approval number was given) and was conducted in accordance with the Declaration of Helsinki.

Overview of the study design. In this four-way cross-over study, participants underwent 4 days of either energy deficit (ED) or energy balance (CON) without (-EX) or with exercise (+ EX). Energy deficit and energy balance were operationally defined as an energy availability of 15 kcal/kg/fat-free mass (FFM) or 40 kcal/kg/FFM, respectively

Participants were recruited from the university community in accordance with the following inclusion criteria: male, 18–30 years of age, ≥ 3 h/week aerobic exercise, normal (body-mass index: 19–25 kg/m2; body fat percentage ≤15%), and stable body weight (± 3 kg during the past 6 months). Exclusion criteria were: smoking, infectious disease within the past 4 weeks, intake of medication that could influence the study outcomes, cardiovascular disease or orthopedic impairment interfering with conducting exercise, diabetes or history of a clinical eating disorder. All participants provided written informed consent prior to participation in the study.

At baseline and the beginning and end of each condition, body weight and body composition were assessed with a bioimpedance scale (Tanita BC 418 MA, Tanita, Amsterdam, The Netherlands). All measurements were carried out in the morning with participants in a fasted (≥ 10 h) and well-hydrated state [24].

Energy deficit was operationally achieved by reducing energy availability, defined as dietary energy intake minus exercise energy expenditure, to 15 kcal/kg FFM, which represents approximately 50% of basal energy requirements [26]. Energy balance was assumed at an energy availability of 40 kcal/kg FFM [27]. Detailed meal plans were developed for each participant, based on individual dietary habits and food preferences. The meal plans served to ensure energy intake was in accordance with the respective conditions. Regarding macronutrient distribution, it was aimed to keep the intakes within the range of the recommendations of the German Nutrition Association (50–55% carbohydrates, 30–35% fat, 10–15% protein) while also assuring that protein intake at least met the recommended daily allowance for adults (0.8 g/kg) [28]. During all conditions, participants weighed all foods consumed as well as leftovers with a food scale and reported their intake daily. Analysis of food consumed occurred on a daily basis (EBIS pro version 7.0, University of Hohenheim, Stuttgart, Germany, 2005) and meal plans were adjusted if reported and prescribed intake differed by ≥ 50 kcal.

During exercise conditions, participants performed supervised exercise on the bicycle ergometer at 60% of their VO2peak until an exercise energy expenditure of 15 kcal/kg FFM was achieved. Outside of the intervention, participants were instructed to abstain from any additional exercise, which was monitored with an activity tracker (SenseWear Pro3 armband, Bodymedia, Pittsburgh, USA).

For analysis of δ15Nurea, participants were asked to collect three urine samples per day. Samples were collected in the morning in a fasted state as well as in the early afternoon and at bedtime. Participants were instructed to record the time of sample collection and the total urine volume. A 50-mL aliquot of each sample was stored in the participants’ refrigerators until delivery to the laboratory, which occurred within 24–48 h. For precipitation of urinary urea, 250 μL of urine were acidified with 375 μL of glacial acetic acid (analytical grade; Merck, Darmstadt, Germany) and 375 μL 10% (w/w) xanthydrol (Sigma-Aldrich, Steinheim, Germany) in methanol (analytical grade; VWR, Darmstadt, Germany) were added and stored at 6 °C for 12 h [21]. Thereafter, the mixture was centrifuged, the supernatant discarded and the precipitate (N,N′-dixanthylurea) washed twice with methanol/water (2:1 v/v). N,N′-dixanthylurea was dried overnight in vacuum over phosphorus pentoxide. For analysis, ~ 500 μg N,N′-dixanthylurea were weighed into tin capsules and δ15N of urea was assessed using elemental analysis–isotope ratio mass spectrometry. Isotope ratios are expressed relative to atmospheric nitrogen (AIR). The elemental analyzer (Eurovektor EA 3000, Hekatech, Wegberg, Germany) was coupled to a Delta C continuous-flow isotope ratio mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). The working standard gas (N2, purity 5.0, Linde, Munich, Germany) and the working standard (creatine-monohydrate, AlzChem, Trostberg, Germany) were scale calibrated using IAEA-N-1 (+ 0.4 ‰) and IAEA-N-2 (+ 20.3 ‰) for δ15N values (IAEA, Vienna, Austria). All measurements were carried out in triplicates and the standard deviation for triplicate measure of the working standard was ± 0.2‰. Instrument stability and analytical performance were checked by regular analysis of the working standard and zero blanks. For the present analysis, data from the three daily samples were aggregated into a daily average of δ15Nurea, and δ15Nurea was further averaged across each of the 4-day study condition. Δ15N was calculated as the difference between δ15Nurea and dietary nitrogen isotope ratio (δ15Ndiet), which was obtained from the prescribed diet as previously described [29].

Statistical analyses were performed with R (version 4.1.1). If not stated otherwise, data are reported as mean ± standard error of the mean (SEM). To account for the small sample size and missing data from one participant, we used linear models to identify changes within conditions, between conditions, and over time. The subject identifier was included in all models to account for the repeated measures. If a trend was observed (p < 0.1), paired one sided t-tests were performed as post-hoc analyses and p-values were adjusted for multiple testing using Holm-correction. Statistical significance was set at < 0.05. Effect sizes were calculated from the difference of means and standard deviation of those differences, with d = 0.2 considered as small, d = 0.5 as medium and d > 0.8 as large effect [30]. To evaluate the impact of potential confounders on Δ15N, multiple linear regression analysis was performed using Δ15N as dependent and protein intake (in g/day), energy availability (in kcal/kg FFM) and the interaction as independent variables. Subject identifiers were included in all models and adjusted R2 was interpreted. The standardized coefficient (β) was obtained from multiple linear regression with Z-transformed variables. Sample size was predetermined through power calculations performed for the primary analysis [24].