The study was conducted as a monocentric, prospective, open-label, matched-group pilot trial at the University of Freiburg, Germany. The primary objective was to assess the additional effects of a GFD on metabolic and performance-related parameters, beyond the effects of a standardized SIT regimen. Moreover, feasibility objectives were evaluated, including recruitment and retention rates, adherence to dietary and exercise protocols, and completeness of outcome data, to inform the design of a future, adequately powered trial. Although the CONSORT 2010 extension is primarily aimed at randomized pilot and feasibility trials, several of its key principles (e.g., clear feasibility objectives, reporting of recruitment and adherence, preliminary effect size estimation) were applied to this non-randomized, matched-group pilot trial (15). Supervised SIT sessions took place at the University of Freiburg between 8 am and 8 pm, with participants assigned to either a GFD or a MD, both provided ad libitum.

A total of 20 healthy male, non-professional endurance athletes, aged 18 to 50 years, were recruited. Participants were classified as trained/developmental athletes, engaging in sport-specific endurance training 2 to 5 times per week, identifying with an endurance sport, and intending to compete at local-level events without national or international representation (16). The groups consisted of 10 athletes each, with one group following the GFD (GFD-G) and the other following the MD (MD-G). Due to the limited sample size and to minimize baseline variability in performance-related parameters, participants were matched based on their peak oxygen uptake (V̇O₂peak) values. They were ranked according to their V̇O₂peak and then alternately assigned to either the GFD-G or MD-G in a pairwise manner, ensuring comparable aerobic fitness levels between groups at baseline. The supervising investigator was responsible for assigning participants to their respective groups.

The sample size for this exploratory pilot trial was pragmatically determined based on a prior study investigating the effects of a short-term GFD on athletic performance in non-celiac athletes (14) and on methodological recommendations for pilot studies suggesting approximately 10 participants per group (17). As this was a pilot study, the sample size was chosen to assess feasibility and to provide preliminary variability estimates across the selected outcome measures. No single primary outcome was pre-specified for sample size calculation. For illustration, based on the variability in the measured outcomes, a future main trial aiming to detect a moderate effect size (Cohen's d = 0.5) with α = 0.05% and 80% power would require approximately 64 participants per group. This calculation is based on a two-tailed independent samples t-test using G*Power 3.1 (University of Düsseldorf, Germany). Exclusion criteria included a prior diagnosis of celiac disease, health problems during or after physical activity, unstable weight or eating behaviors, adherence to special diets (e.g., vegan diet), and contraindications to physical activity (e.g., cardiovascular, metabolic, or renal diseases), as defined by the American College of Sports Medicine guidelines (18).

The study protocol received approval from the Ethics Committee of the University of Freiburg (ETK: 208-18). This study was part of a registered trial (DRKS00025708) comprising two independent sub-studies investigating the effects of different dietary interventions in combination with SIT on metabolic and performance-related parameters in healthy male athletes. For context, one sub-study previously reported on a paleolithic diet (19), the present investigation focuses on a GFD, using the same MD-G as control and identical training protocol to ensure methodological consistency across both sub-studies. Given the exploratory nature, limited sample sizes, and testing feasibility, this approach allowed multiple interventions to be compared without separate control groups (20–22). Reuse of data from the previous sub-study is acknowledged under a CC BY-NC 4.0 license. Although both interventions exclude gluten, they differ conceptually: the paleolithic diet represents a broad evolutionary nutrition approach excluding multiple food groups, whereas the GFD specifically targets gluten avoidance (10, 23, 24). The current dietary intervention specifically assesses whether gluten exclusion alone has measurable effects in non-celiac athletes.

The study spanned approximately nine weeks, including participant screening, baseline (T0) and follow-up (T6) assessments, and the six-week intervention. All participants began the intervention simultaneously. Baseline and follow-up assessments were conducted using multiple morning time slots to accommodate all participants. Each laboratory session lasted around 1.5 h per participant; overlapping slots were possible, as only the treadmill segment required exclusive use. Time trial sessions were scheduled on separate days, with a minimum of two days between sessions to ensure sufficient recovery. The same spacing rules applied between time trial testing and training sessions. Post-testing (T6) followed the same procedure immediately after the intervention. An overview of the study phases and timeline is provided in Figure 1.

After providing written informed consent, participants underwent a screening process that included a detailed medical history questionnaire to confirm eligibility, ensure the inclusion criteria were met, and rule out any risk factors that could be exacerbated by the exercise protocols. During this session, anthropometric measurements (including height, weight, age) were obtained using a calibrated stadiometer and digital BIA scale (OMRON BF-500 Medizintechnik GmbH, Mannheim, Germany), respectively, with participants wearing light clothing and no shoes. Blood pressure was measured on the participant's dominant arm using a validated automated oscillometric device (OMRON M6 Comfort, OMRON Healthcare, Kyoto, Japan). Measurements followed standardized recommendations, with participants seated comfortably, arm supported at heart level, and after a minimum 5 min rest in a quiet environment to ensure accuracy (25).

Body composition assessments were conducted at baseline (T0) and after the intervention (T6), following the procedures described in the efficacy outcomes. Endurance capacity was evaluated using a ramp incremental test and a one-hour time trial at both T0 and T6. Given that substrate oxidation during exercise, as reflected by the respiratory exchange ratio (RER), is influenced by pre-exercise nutrition (26), participants consumed two bananas (125 g each, providing a total of 50 g carbohydrate) 45 min before these tests (27). All study visits, including screening, T0, and T6, were scheduled for the same time in the morning, with participants required to fast for 12 h beforehand. Additionally, they were instructed to empty their bladders before assessments. To ensure proper hydration, participants consumed 1 liter of water the evening prior and 0.5 liters on the morning of each visit (28). Alcohol intake was prohibited for 48 h before any evaluation.

The participants' body composition (fat free mass and fat mass) was assessed by using a bioelectric impedance analysis (BIA) with a high reliability for evaluating body composition in physically active adults (29, 30). Participants were assessed on the BIA scale (OMRON BF-500 Medizintechnik GmbH, Mannheim, Germany). The typical measurement variability (coefficient of variation) for this BIA device series is approximately 1.3% for fat mass indicating that changes exceeding these thresholds can be considered true changes rather than measurement error (31). According to the manufacturer's recommendation, participants were measured in the morning at the same daytime following a fasted period and under the same environmental conditions at each study visit. In addition, the systolic and diastolic blood pressure were recorded at rest as described above.

Spiroergometry (Geratherm Respiratory GmbH, Germany) was conducted on a treadmill (RAM Model 770CE, United Kingdom) in accordance with evidence from recent studies indicating that SIT significantly influences V̇O₂max/peak (32, 33). To assess respiratory parameters (RER, V̇O₂peak, ventilatory threshold [VT], and respiratory compensation point [RCP]), the BRUCE ramp protocol [as illustrated in Supplementary Figure S1 of a previously published investigation by Zdzieblik et al. (19)] was employed. During the initial phase of the protocol (up to 14 min and 20 s), the treadmill grade was progressively increased by 0.2%–0.4% every 20 s. After this, the incline adjustment was set to 0.5% increments every 40 s. The treadmill speed began at 2.7 km/h (1.7 mph) and was raised by 0.2 km/h (0.1 mph) every 20 s from 2 min and 40 s until the 14 min mark. Subsequently speed increased every 40 s (34). The ramp test was preceded by a 3 min rest phase followed by a warm-up period, which included 3 min of light constant-load exercise at 1.7 mph. This preparatory phase aimed to ensure adequate tissue saturation with carbon dioxide, thereby minimizing the risk of misinterpreting the VT (35). VT is determined during incremental exercise as the point where ventilation (VE) begins to increase disproportionately relative to oxygen uptake (V̇O₂), signaling the transition from aerobic to anaerobic metabolism (36). RCP is marked by a pronounced rise in VE relative to carbon dioxide production (V̇CO₂). Both VT and RCP were identified graphically using the V-slope method, which involves plotting V̇CO₂ against V̇O₂ and VE against V̇CO₂ to detect the respective transition points. These thresholds were evaluated independently by three blinded reviewers to ensure accuracy (36). In addition, V̇O₂peak was calculated from the highest 30 s V̇O₂ average during the ramp test (37), and time to exhaustion (TTE) was recorded. Substrate oxidation was also assessed, with maximal fat oxidation (MFO) defined as the highest measured fat oxidation rate, and FatMax as the exercise intensity (%V̇O₂peak) at which MFO occurred (38).

Carbohydrate and fat oxidation rates were calculated using the following Equations (1) and (2) assuming a negligible protein oxidation (39, 40).(1)CarbohydrateOxidation(g⋅min−1)=(4.585×V·CO2)−(3.226×V·O2)(2)FatOxidation(g⋅min−1)=(1.695×V·O2)−(1.701×V·CO2)To assess the RER and substrate oxidation rates, the area under the curve (AUC) was calculated using the trapezoidal method, from the start of the test until the final minute completed by all participants before exhaustion. Substrate oxidation rates were determined only for exercise intensities below the VT, as substrate oxidation estimates become unreliable at higher intensities due to bicarbonate buffering and non-metabolic CO₂ production (26). The inclusion of RER and substrate oxidation rates as outcomes was based on their sensitivity to diet-related changes and their relevance for evaluating substrate utilization during exercise (26). For the prolonged endurance performance assessment, participants completed a fixed-duration running time trial outdoors on a 400 m tartan track under comparable environmental conditions (same location, dry surface). Ambient temperature and humidity were monitored and documented during each session (overall range across all tests:16–26°C, 60%–80% relative humidity). Following a 5 min warm-up on a 400-meter track, the distance covered [km] during the 60 min trial was recorded using a GPS device (Polar M200, Kempele, Finland) (41). Quality of life was measured using the Gastrointestinal Quality of Life Index (GIQLI), a 36-item questionnaire covering five domains: gastrointestinal symptoms (19 items, max 76 points), physical function (7 items, max 28 points), emotional well-being (5 items, max 20 points), social dimension (4 items, max 16 points), and therapeutic component (1 item, max 4 points). The total score ranges from 0 to 144, with higher values reflecting better quality of life. A score below 112 indicates reduced quality of life (42, 43). For gluten sensitivity, the gastrointestinal domain is most relevant, while physical and emotional domains also reflect common symptoms such as fatigue and reduced well-being (44–46).

The statistician responsible for the analysis remained blinded during all procedures. Data unblinding occurred only after data collection was complete and the database was locked, following standard data management protocols.

After attending an information evening to get familiar with nutritional concepts, participants were instructed to follow the dietary pattern according to their respective group (GFD-G or MD-G) over the time course of 6 weeks. The diet of the GFD-G included fresh fruits, vegetables, unprocessed meats, fish, gluten-free grains (e.g., rice, quinoa, corn, buckwheat), legumes, nuts, and dairy products, unless flavored or processed with gluten. Gluten-free flours (e.g., rice, almond, coconut) and certified gluten-free processed foods (e.g., bread, pasta) were permitted.

Excluded from the diet were all wheat-based products (e.g., bread, pasta), barley, rye, oats (unless certified gluten-free), and processed foods containing gluten additives (e.g., malt, wheat starch). Alcoholic beverages such as beer and most processed foods containing gluten-based ingredients (e.g., modified starches) were also restricted. To prevent cross-contamination, careful attention was given to food handling and preparation. To assist participants in identifying gluten-free products, foods labelled with the “Gluten-Free” symbol, often accompanied by a crossed-out wheat ear symbol, were included. This symbol is a licensed trademark of the German Celiac Society (“Deutsche Zöliakie Gesellschaft”). Participants of the MD-G were instructed not to change their nutritional habits. In addition, participants of both groups were asked to complete a three-day nutrition protocol, using digital three-day food diaries provided via the Nutriguide online platform (Nutri-Science GmbH, Pohlheim, Germany), covering two weekdays and one day at the weekend, before, after three and after six weeks of intervention. Participants received a personalized link for each recording period, through which they could either select foods from the integrated German Nutrient Database (Bundeslebensmittelschlüssel, BLS) or enter additional items via a free-text option. They were instructed to record all foods, beverages, and dietary supplements consumed. Portion sizes, meal times, preparation methods, and product brands were noted where possible. Quantification of portion sizes was primarily based on grams or milliliters, but household measures such as tablespoons, teaspoons, and cups were also accepted. All entries were completed by participants themselves and, once submitted, were reviewed by a qualified nutritionist for completeness, plausibility, and consistency with the study protocol. Where entries were ambiguous, incomplete, or implausible, the nutritionist contacted participants directly for clarification or correction. Verified records were then analyzed using the Nutriguide software (Nutri-Science GmbH, Pohlheim, Germany) by the same investigator to maintain consistency in coding and food selection. To ensure quality control, questionable entries were flagged and discussed within the study team before final confirmation. Energy and nutrient intakes were screened for potential under-reporting using plausibility checks (e.g., unusually low total energy intake relative to reported activity levels). While no datasets were excluded on this basis, all flagged cases were clarified with participants whenever possible, and this potential source of error is acknowledged in the study limitations.

Participants were instructed to maintain their usual physical activity levels, excluding the supervised short-term interval training (SIT). The SIT protocol utilized the Wingate anaerobic test format as the exercise component, specifically designed for an active population accustomed to endurance training (47, 48). Each supervised session began with a five-minute warm-up, performed at 70%–80% of maximal heart rate, below the VT (49). The main training consisted of 30-second all-out sprints, followed by a 4 min recovery phase at the same intensity as the warm-up. The number of repetitions was progressively increased: weeks 1–2 involved four repetitions, weeks 3–4 included five, and weeks 5–6 involved six repetitions. A cool-down was performed under the same conditions as the warm-up and recovery phases. All training sessions were conducted at the University of Freiburg between 8 am and 8 pm, with a total of 18 sessions completed during the 6-week intervention.

All data are presented as median and interquartile range (IQR) unless stated otherwise. SPSS statistics (IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.) was used for all statistical analyses. In the present trial, endpoints had been defined with no hierarchy. The statistical evaluation was performed to determine an adequate sample size and the primary outcome of a main RCT study, which will be designed on the basis of the present study protocol. All the tests in the descriptive analysis were performed as two-sided tests and the significance level was set at α = 0.05.

Due to the small sample size and uneven number of subjects per group, the homogeneity of the baseline values between the study groups was checked using the Mann–Whitney U-test. In addition, the mean differences obtained from both groups were compared using the Mann–Whitney U-test. The significance of changes in the respective endpoints during the intervention period within groups were analyzed with the Wilcoxon signed-rank test. For nutrients with an Average Requirement (AR), the proportion of participants below the AR is reported, based on the European Food Safety Authority (EFSA) in their Scientific Opinion on Dietary Reference Values for nutrients. For nutrients with only an Adequate Intake (AI), intakes are described relative to the AI without inferring inadequacy (50). To assess dietary intake during the intervention, data from weeks 3 and 6 were averaged and compared with baseline values.

The magnitude of change in the respective outcomes was expressed as the rank-based effect size r, calculated from the Z-value of Mann–Whitney or Wilcoxon tests, for differences between groups at the end of the investigation and within groups between baseline and post-intervention, using the following Equation (3):(3)ZNwhere N is the total number of observations. Effect sizes were interpreted according to conventional thresholds: small effect: 0.10 ≤ r < 0.30; medium effect: 0.30 ≤ r < 0.50; large effect: r ≥ 0.50 (51).

Recruitment took place over four weeks, during which 35 athletes initially expressed interest; however, many withdrew prior to the formal screening (n = 11) or were excluded based on a preliminary telephone screening (n = 2). Ultimately, 22 athletes were screened. A total of 20 men (91%) met the inclusion criteria and were enrolled in the study (Figure 2). Of these, 15 participants completed the trial and were subsequently included in the final analysis, resulting in a retention rate of 60% in the GFD-G (6/10) and 90% in the MD-G (9/10). Participant dropout was attributed to voluntary withdrawal, as individuals opted not to continue with the dietary intervention or training program. No adverse events or pathological findings related to the intervention were identified through routine anamnesis or the compliance calendar. The analysis of the training protocols revealed no significant differences (p = 0.529) between the GFD-G (17 ± 1) and MD-G (17 ± 1) in terms of the number of completed training sessions. Dietary adherence, based on self-reported food records and compliance calendars, was estimated at 100% in both the GFD-G and MD-G. Completeness of key outcome data was 100%, with no major protocol deviations.

At baseline, no significant differences between groups were observed in the outcome parameters reported in the following sections. Age was comparable between groups (GFD-G: median 24.5 years, IQR 22.5–31.0; MD-G: median 26.0 years, IQR 22.0–28.0; p = 0.955), whereas body height was significantly greater in the GFD-G (1.85 m, IQR 1.82–1.87) compared to the MD-G (1.79 m, IQR 1.77–1.80; p < 0.001).

Between-group comparisons revealed significant differences in weight and BMI changes, favoring the GFD-G over the MD-G (p = 0.018 and p = 0.026, respectively). No significant between-group differences were observed for fat mass, fat-free mass, or blood pressure parameters (Table 1).

Within-group analyses showed that participants in the GFD-G experienced reductions in weight, BMI, and fat mass, along with an increase in fat-free mass percentage. In the MD-G, significant improvements were observed only in fat mass percentage and fat-free mass percentage, while weight and BMI remained unchanged. Systolic and diastolic blood pressure did not change significantly in either group (Table 1).

Table 2 summarizes the nutritional patterns of the intervention groups at baseline and during the intervention period. At baseline, there were no statistically significant differences between the groups in mean energy or nutrient intake, except for the intake of saturated fatty acids.

When comparing intervention assessments to baseline values, the median energy intake in the GFD-G was approximately 100 kcal higher, whereas it was approximately 100 kcal lower in the MD-G. Compared to baseline, the GFD-G showed a non-significant increase in protein intake at the intervention assessments, along with improved micronutrient intake, including a significant increase in vitamin C consumption.

In the MD-G, no significant changes in dietary patterns were observed.

No statistically significant differences in energy or nutrient intake were observed between the groups at the intervention assessments.

In contrast to baseline assessment, the proportion of participants below the AR in the GFD-G was lower at the intervention assessment for several micronutrients, including fiber (50.0 → 16.7%), calcium (66.7 → 33.3%), retinol equivalent (16.7 → 0%), vitamin C (66.7 → 16.7%), riboflavin (83.3 → 33.3%), vitamin B6 (33.3 → 0%), and folate (50.0 → 33.3%). Energy (50.0%), zinc (0.0%), iron (0.0%), thiamin (33.3%), vitamin E (50.0%) and niacin (0.0%) remained unchanged. In the MD-G, proportions below the AR for calcium (55.6 → 44.4%) and vitamin C (44.4 → 33.3%), vitamin E (33.3 → 22.2%), vitamin B6 (33.3 → 22.2%), were lower, whereas proportions for fiber (22.2 → 33.3%), retinol equivalent (0.0 → 11.1%) and riboflavin (22.2 → 33.3%) increased. Energy (44.4%), zinc (11%), iron (11%), thiamin (11.1%), niacin (11.1%) and folate (22.2%) showed no relevant differences between baseline and the intervention assessment. For nutrients with only an AI, intakes are presented relative to the AI without inferring inadequacy.

Detailed results of the GIQLI subscales and total score are provided in Table 3. No significant baseline differences between the study groups were detected in GIQLI-Scores (p = 0.514) Changes in GIQLI -Scores from baseline to post-intervention were small in the GFD-G (p = 0.833; r = 0.019) and medium in the MD-G (p = 0.068; r = 0.251). No significant group differences could be detected for changes in the GIQLI -Score (p = 0.388; r = 0.262).