Endurance exercise intervention reduces vegetable intake which compromises the achievement of glycemic benefit: a randomized controlled, three-period crossover trial Authors: Sihan Wang, Hui Gao, Hao Cai, Ling Liu, Yingmei Li, Mingjing Gao, Yacan Huang, Shuangyan Liu, Wei Wang, Mingjian Fu, Yurong Liu, Yunhua Zi, Xuanwen Mu, Xiaogang Zhang, Kuai Yu, Gaokun Qiu Published: 2026 ABSTRACT AbstractBackgroundThere exists long-lasting concern, but no substantive evidence, that exercise may cause compensatory eating and compromise its glycemic benefits. We performed a randomized controlled, three-period crossover trial of moderate- and vigorous-intensity endurance exercise (MOD-EX and VIG-EX) intervention to fill this knowledge gap, which was also externally validated in the UK Biobank (UKB) cohort.MethodsIn total, 54 healthy college students (aged 22.2 ± 2.5 years) completed the 5-day WHO-recommended doses of MOD-EX, VIG-EX, or control treatments. Outcomes included dietary responses, assessed by daily weighed dietary logs in ad libitum eating condition, and glycemic control status, assessed both at fasting state and during oral glucose tolerance test (OGTT). Cox proportional hazard models were fitted to estimate associations of the interaction between physical activity (PA) and dietary intake with the risk of type 2 diabetes (T2D).ResultsMOD-EX and VIG-EX did not induce significant increases in total energy intake, but 15.67% (−13.5 kcal/day) decreased vegetable intake and 10.92% ( + 31.1 kcal/day) increased intake of added oil in food for VIG-EX (p < 0.05). MOD-EX and VIG-EX both reduced postprandial glucose (HbA1c: −3.82% [-6.40% to −1.24%] and −2.18% [−4.67% to 0.32%]; 2 h total area under the glucose curve (tAUC) during OGTT: −11.92% [−24.48% to 0.64%] and −13.48%, [−25.34% to −1.63%]). The effect had been masked by 7.52% and 10.85% from diet, respectively, primarily attributing to decreased fiber and increased saturated fatty acid (SFA) intakes ( | factors loadings | >0.4), exactly matching to dietary responses to exercise. Interaction analysis supported that higher fiber (p-interaction=0.022) and lower SFA (p-interaction=0.009) intakes augmented exercise-induced glycemic benefit, which was confirmed in the UKB study (N = 50 162) that participants with higher fiber or lower SFA intake showed more pronounced association between PA and reduced risk of incident type 2 diabetes (p-interaction=0.035).ConclusionsWe provide evidence to recommend dietary fiber supplementation and SFA restriction accompanying exercise training to achieve maximum glycemic improvement.Trial registration Chinese Clinical Trial RegistryChiCTR2100055095. 2021/12/31.https://www.chictr.org.cn/showproj.html?proj=139665Subject terms: Nutritional supplements, Pre-diabetes Introduction The epidemic of type 2 diabetes (T2D) has been increasing for three decades, projected to continue to rise up to a worldwide prevalence of 12.2% with a total number of 783 million patients in 2045, posing enormous financial burden to the healthcare system [1]. In the combat with this epidemic, exercise has been proven by compelling evidence to be the one of the most effective non-medical measures not only in treatment but also in prevention of T2D [2], even associated with 51.9% reduced risk of developing T2D risk among lean and healthy women [3]. The natural course of T2D development begins as early as 13 years before diagnosis, with the first manifestation of blunted insulin sensitivity, followed by increased postprandial then fasting glucose, and glucose homeostasis is becoming more and more challenging to regain along the course [4], suggesting that management of T2D risk for primary prevention should begin as early as possible. For this reason, exercise excels itself in that it is a safe lifestyle modification measure with no essential cost attached, and is effective in improving glycemic control even among apparently healthy individuals, making it an ideal prescription for early intervention of T2D before disease onset. Although exercise has been proven effective in T2D treatment and prevention in well-controlled research settings [5], there is a growing concern that exercise alone without dietary considerations might not readily lead to long-term maintenance of the achieved glycemic improvement [4]. Exercise raises energy expenditure. Evidence from studies employing sophisticated energy balance models, including those utilizing the doubly labeled water (DLW) method for energy intake assessment, indicates that increased energy expenditure can induce compensatory increases in energy intake, which may compromise exercise-induced glycemic benefits in long-term [6–8]. However, these findings from various studies concerning the impact of exercise on energy intake have shown a degree of inconsistency. Systematic reviews and meta-analyses highlight that conclusions are inconsistent and significantly influenced by measurement methodology. For instance, a meta-analysis highlighted the discrepancy between self-reported and objectively measured (DLW) energy intake, and further re-evaluated studies that estimated intake via energy balance equations. This analysis reported a median (range) pre- to post-intervention change in daily energy intake of 70 kcal (−381 to 174), indicating a minimal and highly variable pooled average effect [9]. This aligns with broader reviews noting a lack of consistent evidence for a significant effect on energy intake in healthy adults [10]. Furthermore, high-quality evidence from studies using objective measures in free-living settings, particularly among Chinese populations, remains limited. As for total energy intake, the E-MECHANIC study had proved that at doses of 8 and 20 kcal/kg body weight, approximately 1.40 and 3.05 metabolic equivalent of task hours per day (MET-h/day), exercise intervention increased daily energy intake by 71.1 and 90.5 kcal, respectively, by use of DLW method, the gold-standard method for monitoring of dietary energy intake [11]. Dietary pattern involves much more than merely intake of total energy, but also other important aspects such as macro- and micro-nutrient composition, dietary fiber and certain foods which have established associations with glycemic control [12–14], and its assessment entails high-quality dietary monitoring, which is extremely challenging, particularly when it comes to the complex Chinese diet characterized by significant geographic diversity and dynamic temporal transition in food consumption. There were two studies which assessed the effect of exercise intervention on intakes of individual food and nutrients in ad libitum condition, with conflicting results that one reported increase of vegetables intake and decrease of fat intake [15], and the other reported decreased dietary fiber and increased saturated fatty acids (SFAs) intakes [16], both conducted among participants leading a western lifestyle. Studies examining dietary responses to exercise in the scenario of Chinese diet are scarce. More importantly, it remains unknown how dietary responses to exercise would further affect exercise induced glycemic benefits, and what dietary factors can be held responsible, questions critical to the formulation of targeted dietary recommendations for exercise training to maximize the achievement of glycemic improvement. Besides, the difference of effect between exercise modalities, intensity in particular, is also worth pursuing, with the optimal modality yet to be settled [17–19]. In this study, we performed a randomized controlled, three-period, three-treatment crossover trial among 54 apparently healthy Chinese college students to examine ad libitum dietary responses, in terms of intakes of total energy, individual foods and drinks, macro- and micro-nutrient composition, to moderate- vs vigorous-intensity endurance exercise (MOD-EX and VIG-EX) intervention at World Health Organization (WHO)-recommended doses (150 min of MOD-EX or 75 min of VIG-EX per week), and their collective effect on glycemic control, with comparison between the two exercise intensities. We monitored real-time physical activity (PA) and obtained weighed dietary logs from each participant on all food and drinks they consumed on each day of the three study periods, totaling 8945 items of dietary records on 972 person-days (based on each of 54 participants completing a 6-day record in each of three periods). With these data, we systematically evaluated dietary responses to exercise intervention, and more importantly, the effect from such dietary responses on exercise-induced glycemic benefits, with identification of the responsible dietary factors. For the collective effect of exercise and diet, we further sought to externally validate this interaction association in the real-world observational UK Biobank (UKB) cohort, thereby evaluating its generalizability beyond controlled trial conditions. Methods Randomized controlled crossover trial of exercise intervention Participants In the randomized controlled, three-period, three-treatment crossover trial, we recruited healthy, young college students to participate. The study protocol was approved by the Ethics and Human Subject Committee of Tongji Medical College, Huazhong University of Science and Technology ([2021] IEC (S274), and is registered at the Chinese Clinical Trial Registry (ChiCTR2100055095). We restricted recruitment within apparently healthy college students. To be eligible, individuals should be aged 18-35 years, not having been suspended from school or planning to change school, not smoking or having a smoking roommate; not being pregnant or in lactation period; not being inflicted with conditions that disenables participation in exercise intervention. A total of 60 college students were enrolled in this study, and 54 of them (16 male and 38 female individuals) completed the trial according to the study protocol. All participants provided written informed consents. Study design This study was a randomized controlled, three-period, three-treatment crossover trial, and the three treatments were moderate-intensity exercise (MOD-EX) intervention (30 min of cycling at the speed of 9–11 km/h on an ergometer bike per day for five consecutive days), vigorous-intensity exercise (VIG-EX) intervention (15 min of cycling at the speed of 19–21 km/h per day for five consecutive days), and free-living as the control condition. The cycling speed for MOD-EX and VIG-EX interventions were specified according to 2011 Compendium of Physical Activities: a Second Update of Codes and MET Values [20], and the dose of exercise intervention was determined approximately equaling to WHO recommendations of 150 min moderate- or 75 min vigorous-intensity PA per week [21]. To verify the actual exercise intensity at the specified cycling speed of 9–11 and 19–21 km/h, we performed preliminary test among 10 volunteers (also healthy college students), using a three-axis accelerometer (Actigraph WGT3X-BT), a research-grade physical activity monitor with well accepted reliability and validity [22, 23]. The actual intensity during MOD-EX and VIG-EX was 4.67 ± 1.17 METs and 11.00 ± 1.31 METs in the preliminary test, respectively, in accordance with the definitions of moderate- (3-6 METs) and high-intensity ( ≥ 6 METs) exercise in the 2011 Compendium of Physical Activities: a Second Update of Codes and MET Values [20]. During the trial, the exercise intensity during MOD-EX and VIG-EX interventions were 3.87 ± 0.94 METs and 10.35 ± 0.78 METs, respectively. The three study periods were separated by two 14-day washout periods in-between, and a recovery day was emplaced after each study period before blood collection for outcome measurements, in order to minimize confounding from acute effects from last bout of exercise. Sample size Sample size was estimated a priori assuming a small-to-moderate effect size of 10% change from baseline, using the “onesamplemeans” option in Proc Power by t-test in SAS software for three-period crossover of WILLIAMS design [24], and a power of > 80% could be achieved with a sample size of eight participants per sequence (48 participants in total). A total of 60 participants were enrolled for randomization. Randomization All enrolled participants were randomly assigned into one of the six sequence groups of treatment order according standard WILLIAMS design by use of the SAS macro %WILLIAMS [25], and the treatment orders were Control - MOD-EX - VIG-EX, Control - VIG-EX - MOD-EX, MOD-EX - Control - VIG-EX, MOD-EX - VIG-EX - Control, VIG-EX - Control - MOD-EX, and VIG-EX - MOD-EX - Control, 10 participants per sequence, with MOD-EX and VIG-EX representing moderate- and vigorous-intensity exercise interventions, respectively, and Control representing the free-living control condition. During recruitment, we did not place restriction to enroll those having a sedentary lifestyle. Instead, we conducted an interview with the International Physical Activity Questionnaire (IPAQ) [26] to assess their PA level during the past seven days, and carried out randomization according to the assessed basal PA level to ensure its balanced distribution across the six sequence groups, in order to examine whether health benefits derived from exercise intervention is independent of daily-life PA level. To this end, we also monitored non-intervention PA level of each participant on each day of the three study periods using Actigraph WGT3X-BT accelerometers (ActiGraph LLC, Pensacola, FL, USA) wore at the participants’ dominant wrist, and intervention PA level wore at the participants’ right ankle. In the end, 54 participants completed the trial, with 8, 8, 10, 9, 9 and 10 participants within the sequence groups of Control - MOD-EX - VIG-EX, Control - VIG-EX - MOD-EX, MOD-EX - Control - VIG-EX, MOD-EX - VIG-EX - Control, VIG-EX - Control - MOD-EX, and VIG-EX - MOD-EX - Control, respectively. Dietary recording and assessment of intakes of total energy, individual food and drinks, and macro- and micro-nutrient All participants were instructed to keep ad libitum eating habits as usual and record all food and drinks they consumed during each study period, including five-day treatment period and the recovery day, with a smartphone app ‘Boohee’, which has built-in the largest database of Chinese food to date (https://www.boohee.com/food) [27]. Of note, the ‘Boohee’ app was employed solely for ease of diet recording, not as data reference. We obtained a total 8945 food records during the 972 person-days (based on each of 54 participants completing a 6-day record in each of three periods) of dietary monitoring, including 474 kinds of foods and drinks, each with data processing details and nutrient contents provided in supplementary methods. In brief, the amount of food or drinks intake was recorded as weight if the weight was known, such as for snacks with weight per serve labeled, or as portions if the weight was unknown, and the weight of food or drinks was then verified through manual weighting by the research staff. All weighted food and drinks were then decomposed into contents of their major ingredients also through manual weighting, with additional reference to nutrition information on food labels and Chinese Food Composition Tables, supplemented by USDA FoodData Central (https://fdc.nal.usda.gov/) [28, 29]. For each item of food record, we assessed total energy, energy from major macronutrients (fat, protein and carbohydrate), the amount of dietary fiber and micronutrients (18 amino acids, 22 fatty acids, 10 minerals and 15 vitamins), additives including oil, sugar and common sauces. And for drinks, we assessed caffeine intake from coffee, energy intake from sugar-sweetened beverages (including soft drinks [carbonated beverages, sports drinks, fruit-flavored drinks containing added sugar, commercial tea beverages, sweetened water], and boba tea, a hand-made sweetened beverage popular among young people in China) and various dairy products (milk, fermented dairy products, and sweetened-milk beverages), as well as additive sugar in all drinks. Calculation of intakes of major food ingredients and macro- and micro-nutrients was exemplified with a popular Chinese dish, Stir-fried Green Pepper with Eggs. If a participant documented consuming half of a full serving of stir-fried green pepper with eggs from the school cafeteria using the “Boohee” mobile application, we calculated the intakes of eggs, green peppers, added oil and macro- and micro-nutrients according to the following steps: We first purchased a same serving of stir-fried green pepper with eggs from the school cafeteria, and through manual weighing, we documented that this dish contained 80 g of green pepper and 67 g of egg (cooked weight), approximating to 81 g and 62 g in raw weight, respectively, according to the proportion of each food ingredient and change of humidity in the database from USDA FoodData Central. Added oil in this dish was calculated also according to the USDA FoodData Central, assuming that 3 g of oil is added per 100 g (raw weight) of sweet green peppers, and 7 g of oil is added per 100 g (raw weight) of eggs, amounting to a total of 6.8 g added oil in a full serve of this dish. For macro- and micro- nutrients including carbohydrates, protein, fat, 18 amino acids, 22 fatty acids, 10 minerals and 15 vitamins, the calculation was based on their contents in sweet green peppers, eggs and cooking oils according to the USDA FoodData Central, with fatty acid composition in cooking oils estimated assuming intakes of 50% soybean oil, 26% rapeseed oil, 10% palm oil, 10% peanut oil, and 4% sunflowerseed oil, equaling to the consumption ratios of edible oils in Chinese market from 2021 to 2022 according to Table 27 of Oilseeds from the USDA Foreign Agricultural Service’s PSD Online Reports. The consumption by this hypothetical participant would be multiplied by 0.5. Blood collection and assessment of glycemic control status Blood collection for each participant was performed on the day before each study period, and after each study period on the day following the recovery day (7:00 a.m. to 8:00 a.m. for fasting blood, and 8:00 a.m. to 10:30 a.m. for blood samples during the 2-h OGTT). Blood samples were collected in the fasting state and during a standard glucose tolerance test (30, 60 and 120 min after intake of 75 g glucose). We measured levels of glucose, HbA1c (Beckman Coulter AU480, Beckman Coulter, Brea, USA), insulin (ELISA, DRG Diagnostics, Marburg, Germany, Cat#EIA2935), C peptide (ELISA, DRG Diagnostics, Marburg, Germany, Cat#EIA1293) and free fatty acids (Solarbio, Beijing, China, Cat#BC0595) in each blood sample (supplementary methods). The between-batch coefficient of variation for measurements of insulin, C peptide, and free fatty acids were 22.87%, 8.74% and 15.61% for fasting samples, and 16.34%, 17.48%, 11.10% for OGTT samples. Glycemic control was assessed using indicators as follows: overall glycemic control (fasting glucose, fasting HbA1c, and area under the curve of glucose calculated using the trapezoidal rule [30]), insulin sensitivity (McAuley index, calculated as exp [2.63 – 0.28 × ln (fasting insulin mU/L) – 0.31 × ln (TG mmol/L)] [31]; HOMA2-IR, calculated with the website of HOMA2 calculator: https://www.dtu.ox.ac.uk/homacalculator/index.php; Adipo-IR, calculated as fasting NEFAs mmol/L × fasting insulin pmol/L [32]) and β-cell insulin secretion (fasting insulin, Stumvoll-1st phase index, calculated as (1283 + 1.829 × 30-min insulin pmol/L – 138.7 × 30-min glucose mmol/L + 3.772 × fasting insulin pmol/L); and Stumvoll-2nd phase index, calculated as (286 + 0.416 × 30-min insulin pmol/L – 25.94 × 30-min glucose mmol/L + 0.926 × fasting insulin pmol/L) [33, 34]). External validation in the UKB study Data from the UKB study were used for replication purpose. In brief, the UKB study is a prospective cohort of more than 500 000 participants enrolled between March 2006 and July 2010 [35]. At recruitment into the cohort, each participant completed a touchscreen questionnaire covering socio-demographic, lifestyle (covering a 29-item short food frequency questionnaire [FFQ] to evaluate the past-year average diets), and psychosocial characteristics and underwent an anthropometric assessment. Monitoring of physical activity was performed in a substudy of 103 666 participants who responded to wear an Axivity AX3 wrist-worn accelerometer for 7 days, and in the end, acceptable data were obtained from 96 644 participants after excluding those with insufficient wear time, poor quality of calibration, and those with an abnormal average acceleration more than 100 milligravities (mg) [36]. Dietary intakes were assessed using a web-based 24-h recall (Oxford WebQ) acquiring the consumption of about 200 commonly consumed food and drinks within the previous 24 h and administered five times between 2009 and 2012, which has been validated, demonstrating comparable energy and nutrient intakes to those of an interviewer-administered 24-hour recall [37]. The consumed quantity of each food or drink is calculated by multiplying the assigned standard portion size of each food or drink by the consumed amount. Each nutrient intake was calculated by multiplying the nutrient composition by the consumed quantity. Incident T2D was diagnosed by using code E11 (without E10) or E14 (without E10–E13) from the International Classification of Diseases, 10th Revision, after completing a 7-day accelerometer assessment [38]. After excluding those with prevalent diabetes and cancer, or taking insulin, or HbA1c level >48 mmol/mol, or with missing covariate information, a total of 50 162 individuals was available for the final analysis (Fig. S1), and during 6.8 ± 0.9 years of follow-up, a total of 779 (1.6%) incident T2D cases were documented. The present study was conducted under the UKB application project 88159. Statistical analysis A total of 54 participants completed this trial according to the protocol. A 2-hour OGTT was offered as an optional, more time-intensive component of the trial. It was completed by 27 of the 54 participants, as participation was voluntary and subject to individual scheduling constraints. The actual speed of cycling was 10.12 km/h (range: 9.44 km/h–10.98 km/h) during MOD-EX, and 19.94 km/h (19.08 km/h–20.84 km/h) during VIG-EX. We compared baseline characteristics of the participants at each study period, and also between the participants categorized into the six sequence groups, using Analysis of Variance (ANOVA) test for continuous variables, and Chi-Square test for categorical variables. Descriptive data were reported as mean ± SD or median (P25, P75) unless otherwise noted. The primary aims of this trial were two-fold: to assess the effect of exercise intervention on ad libitum dietary intake, and the effect of exercise intervention on glycemic control as well as the role of dietary intake in such effect. For the effect of exercise intervention on dietary intake, we conducted linear mixed models with R package ‘lmer4’ to compare total energy intake, energy intake from fat, protein and carbohydrate, from each of three meals, intakes of individual food and drinks, and micro-nutrient between intervention and control periods. We also performed Principal Component Analysis (PCA) using a total of 96 dietary features derived from the dietary recording data, and calculated euclidean distances from the control period to assess the effect of exercise on overall dietary intake with SAS 9.4. For the effect on glycemic control, we first conducted linear mixed models to compare the differences of change in all glycemic indicators from baseline between intervention and control periods, with adjustment for non-intervention PA and major dietary principal components (PCs) to examine the independence of such effect. In effect estimate from exercise intervention, we fitted the model as where Y denoted the logarithmically transformed level of an outcome variable, X denoted the treatment variable (0=control, 1=exercise intervention), visit denoted the time variable (0=pre-treatment, 1=post-treatment), and the interaction term between the treatment and visit was introduced to assess the effect from the treatment, which was the difference of change from baseline in outcomes comparing the intervention periods to the control period. Due to logarithmical transformation of the outcome variables for their skewed distributions, the different effect was estimated as %change [exp(β2 + β3)-exp(β2)], which is a common practice in studies involving log-transformed dependent variables in linear regression [39]. Effect heterogeneity between sexes and body mass index (BMI) status ( ≥23 kg/m2 or not) was examined with subgroup analysis. The cutoff value of 23 kg/m2 for BMI categorization was based on WHO recommendation for Asian populations [40], also with an aim to maintain close number of participants between subgroups as much as possible. We also constructed RCS to depict the dose-response relationship between total PA level and glycemic control, using HbA1c and glucose tAUC as overall glycemic control assessment. We then performed mediation analyses with the SPSS macro PROCESS (version 4.2) to assess the effect of ad libitum dietary intake on exercise-induced HbA1c reduction, using major dietary PCs representing overall dietary intake, and the responsible dietary factors were then identified according to factor loadings on each PC. For key dietary factors, we then explored their interaction with PA level on HbA1c reduction. In observational analysis with UKB data, we used Cox proportional hazards models to explore the interaction between accelerometer-assessed total PA and dietary factors affected by exercise identified in the previous crossover trial on risk of developing T2D, adjusting for age, sex, BMI, season of wear, race, baseline HbA1c level, parent history of diabetes, educational level achieved, employment status, smoking and alcohol status. A two-tailed p value < 0.05 was considered statistically significant. Ethics approval and consent to participate All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation by the Ethics and Human Subject Committee of Tongji Medical College, Huazhong University of Science and Technology ([2021] IEC (S274), and is registered at the Chinese Clinical Trial Registry (ChiCTR2100055095) and with the Helsinki Declaration. Informed consent was obtained from all participants for being included in the study. This article does not contain any studies with animal subject. All participants in the UKB study have given their informed consent, and the study was approved by the North West National Health Service Research Ethics Committee (Ref 21/NW/0157). Results Characteristics of participants A total of 60 college students were enrolled in this study, and 54 of them (16 male and 38 female individuals) completed the trial according to the study protocol (Fig. 1), with total volume of PA 29.92 ± 3.12 MET-h/day (mean ± SD [Standard Deviation]) during the control period, 31.32 ± 2.98 MET-h/day during MOD-EX intervention period, and 32.21 ± 3.12 MET-h/day during VIG-EX intervention period (p < 0.001; Table 1). PA level from MOD-EX and VIG-EX interventions were 1.98 ± 0.48 and 2.59 ± 0.20 MET-h/day, respectively (Table 1). The average age of the participants were 22.2 ± 2.5 years, with a BMI of 22.36 ± 3.16 kg/m2, and 70.40% of them were female (Table S1). Study flow chart. A total of 60 college students were enrolled in this study, and 54 of them (16 male and 38 female individuals) completed the trial according to the study protocol. Each study period included a five-day treatment period followed by a recovery day, and separated from another study period with a 14-day washout period. Physical examination and blood sample collected was performed for all participants on the day before each treatment period and the day after each recovery day between 7:00 am to 10:30 am (7:00 am to 8:00 am for fasting blood samples and 8:00 am to 10:30 am for blood samples collected during the 2-h OGTT). Data were collected and analyses performed on 54 participants. The total 46-day trial contained three-treatment periods (five consecutive days per period) and two 14-day washout periods. Participants visited physical examination and blood sample collection before and after each treatment period (total six times). The time window between post-treatment blood sample collected and the last bout of exercise intervention was 37.90 ± 4.60 h. We monitored real-time PA level using Actigraph WGT3X-BT accelerometers throughout each treatment period and also on the recovery day, and also obtained weighed dietary logs from each participant of all food and drinks they consumed using the smartphone app ‘Boohee’, total 8945 items of dietary records on 54*18 person*days. Table 1Characteristics of study participants at baseline.ControlMOD-EXVIG-EXpBaseline characteristics Age, years22.24 ± 2.4622.24 ± 2.4622.24 ± 2.46- Female, n (%)38 (70.40)38 (70.40)38 (70.40)- BMI, kg/m221.63 (20.15, 23.89)21.65 (19.98, 23.81)21.79 (19.81, 23.99)0.149 Current drinker, n (%)2 (3.7)2 (3.7)2 (3.7)- Current smoker, n (%)0 (0.0)0 (0.0)0 (0.0)- Moderate PA time, h per week0.42 (0.00, 1.92)0.42 (0.00, 1.92)0.42 (0.00, 1.92)- Vigorous PA time, h per week0.00 (0.00, 0.56)0.00 (0.00, 0.56)0.00 (0.00, 0.56)- Total MVPA time, h per week1.08 (0.00, 2.40)1.08 (0.00, 2.40)1.08 (0.00, 2.40)-Baseline glycemic control In the fasting state Numbers, na535354- Fasting glucose, mmol/L4.96 ± 0.344.92 ± 0.384.92 ± 0.340.520 Fasting HbA1c, %5.10 ± 0.385.16 ± 0.285.14 ± 0.270.995 Fasting insulin, μIU/mL12.48 (10.79, 16.48)13.39 (11.04, 18.58)12.98 (10.57, 17.72)0.316 During the 2-h OGTT Numbers, nb121716 Glucose tAUC, mmol/L·min721.5 (617.6, 768.8)708.0 (646.0, 797.7)747.9 (694.4, 791.3)0.362 Stumvoll 1st phase1634.18 ± 608.691366.19 ± 468.121557.48 ± 466.620.474 Stumvoll 2nd phase414.68 ± 144.12354.19 ± 104.90398.20 ± 111.870.518PA, MET-h/day Total PA during 5-day intervention/control period29.92 ± 3.1231.32 ± 2.98*32.21 ± 3.12*<0.001 Intervention PA0 ± 01.98 ± 0.48*2.59 ± 0.20*<0.001Data are presented as number (%), mean ± SD or median (P25, P75).MOD-EX, moderate-intensity exercise, VIG-EX vigorous-intensity exercise, n number of participants, PA physical activity, MVPA moderate-to-vigorous intensity physical activity, MET metabolic equivalent of task, HbA1c glycated hemoglobin, tAUC total area under the curve, Stumvoll 1st phase index, first-phase Stumvoll index; Stumvoll 2nd phase index, second-phase Stumvoll index.aAll 54 participants completed three period treatments, and one participant missed two blood samples before Control and MOD-EX intervention.bTwenty-seven participants of the whole subjects underwent at least twice OGTT at both pre- and post-treatment.*Significant difference between intervention period and control period (p < 0.05 by Friedman test). Characteristics of study participants at baseline. Data are presented as number (%), mean ± SD or median (P25, P75). MOD-EX, moderate-intensity exercise, VIG-EX vigorous-intensity exercise, n number of participants, PA physical activity, MVPA moderate-to-vigorous intensity physical activity, MET metabolic equivalent of task, HbA1c glycated hemoglobin, tAUC total area under the curve, Stumvoll 1st phase index, first-phase Stumvoll index; Stumvoll 2nd phase index, second-phase Stumvoll index. aAll 54 participants completed three period treatments, and one participant missed two blood samples before Control and MOD-EX intervention. bTwenty-seven participants of the whole subjects underwent at least twice OGTT at both pre- and post-treatment. *Significant difference between intervention period and control period (p < 0.05 by Friedman test). All participants provided fasting blood before and after each study period, with post-intervention blood samples collected after 37.9 ± 4.6 h after last bout of exercise intervention (Fig. 1). A subset of 27 participants, whose baseline characteristics and glycemic homeostasis were comparable with the total 54 participants, also took part in a standard 2 h OGTT before and after at least one of the study periods, with blood samples collected at 0, 30, 60, and 120 min post glucose load (Table S1). We assessed status of glycemic control both in the fasting state and during the 2 h OGTT. The baseline fasting glucose levels of the participants before each study period were 4.96 ± 0.34 mmol/L (Control), 4.92 ± 0.38 mmol/L (MOD-EX), and 4.92 ± 0.34 mmol/L (VIG-EX), and the corresponding glycated hemoglobin (HbA1c) levels were 5.10 ± 0.38%, 5.16 ± 0.28% and 5.14 ± 0.27% (all p > 0.05). Indicators of insulin sensitivity and β-cell insulin secretion were also similar at baseline across the three study periods (p > 0.05; Table 1, Table S2). The participants also had comparable levels of all these glycemic indicators across the six sequence groups of treatment order according to the WILLIAMS design [25] for three-treatment, three-period cross-over trials (all p > 0.05; Table S3). Effect of MOD-EX and VIG-EX on ad libitum dietary intake manifested in reduced vegetable and increased added oil intake Across the three periods (control free-living condition, MOD-EX, and VIG-EX), the mean total energy intakes of male participants were 1973 ± 449, 2011 ± 432, and 2120 ± 450 kcal/d, respectively, whereas those of female participants were 1670 ± 310, 1664 ± 299, and 1675 ± 314 kcal/d (Table S4), close to the recommended energy intakes of 2150 kcal/d for male and 1700 kcal/d for female adults according to DIETARY REFERENCE INTAKES FOR CHINA. We first assessed between-day and between-period variability of intakes of the major categories of food, and macro- and micro-nutrients. We found that, for almost all categories of food, macro- and micro-nutrients, there was good reproducibility in their intakes across the three study periods and 18 study days, with intraclass correlation coefficients (ICCs) > 0.70 (Fig. S2), except intakes of polyunsaturated fatty acids (PUFAs), vitamin B12, and vitamin D with between-day ICCs <0.4, and intakes of sticky rice, red meat, fish and seafood, sauces, PUFAs, Cu, vitamin A, retinol, and vitamin B12 with between-period ICCs <0.4. The good agreement in between-day and between-period dietary intake indicated that although diet monitoring based on voluntary recording of food intakes from the participants might be susceptible to missing report, the possibility remained minimal in our study. Compared with the control free-living condition, no significant differences in total energy intake were observed for either MOD-EX or VIG-EX (7.3 and 46.9 kcal/d, respectively; p = 0.958, 0.402), and the energy intake increase by VIG-EX intervention was mainly attributable to that from fat and carbohydrate intake (23.5 and 33.7 kcal/d, respectively; Fig. 2a). To assess overall dietary structure, we used the first four PCs with eigenvalues greater than 5 derived from PCA of a total of 96 dietary features (intakes of energy [4 features], 22 categories of foods/drinks and 70 macro- and micro-nutrients; Fig. S3) as a representation of overall dietary intake. The first four PCs each explained 37.39%, 7.81%, 5.84% and 5.56% of the variation in overall dietary intake, with eigenvalues of 36.26, 7.58, 5.67 and 5.39 (Table S5). We found that, on the whole, VIG-EX altered overall dietary intake to a similar level compared to the MOD-EX, showed as euclidean distances of 5.46 ([P25-P75]: 3.77-7.50) vs 5.63 (3.74-8.05) (p = 0.698) from the control period determined within the first four PCs, with similar proportions of contributions from the four PCs (49.13%, 22.17%, 15.06%, and 13.65% vs 46.55%, 19.12%, 19.30%, and 15.02%; Fig. 2b). We compared dietary responses to exercise intervention between sexes and those with higher BMI ( ≥23) and those without, and found that in male participants, differences in overall dietary intake between MOD-EX and control periods was greater among those with higher BMI (euclidean distances of 4.15 vs 6.56, p = 0.043; Fig. 2b). Effects of MOD-EX and VIG-EX intervention on ad libitum dietary intake. a Effects of MOD-EX and VIG-EX on energy intake. Data are expressed as mean in figure. b1 Box-and-whisker plot showing euclidean distance from PCA separated by treatments, sex and BMI calculated within the first four dietary PCs with eigenvalues ≥5. b2 Radar chart showing contribution proportions of the first four dietary PCs on the whole Euclidean distance. c Effects of MOD-EX and VIG-EX on food, drinks and nutrients intake. In the analysis, linear mixed models were constructed including variables of treatment, treatment sequence, study period, age, sex, and BMI as fixed effects, and subjects as random effects. Boba tea is a type of hand-made sweetened beverage popular among young people in China, with substrates of green tea, black tea or oolong tea, and addition of milk, sweeteners (e.g. fructose syrup), cream, red bean, nata, tapioca, pudding, or poi. MOD-EX, moderate-intensity exercise; VIG-EX, vigorous-intensity exercise; * p < 0.05; **p < 0.01; † 0.05

|0.4| were shown and considered as main features of each component. MOD-EX, moderate-intensity exercise; VIG-EX, vigorous-intensity exercise; PC, principal component; HbA1c, glycated hemoglobin. As a sensitivity analysis, we also examined mediation proportions of all other dietary PCs with eigenvalues greater than 1 in HbA1c reductions by MOD-EX and VIG-EX intervention, and the mediation proportions of PC5 to PC19 were 0.75% and −0.78% (PC5); 0.37% and 0.75% (PC6); 0.37% and 1.55% (PC7); 0.75% and −5.43% (PC8); 0.75% and 4.65% (PC9); 1.12% and −50.78% (PC10); −13.58% and −4.65% (PC11); 0.75% and −4.65% (PC12); −2.64% and 3.88% (PC13); 0.75% and 1.55% (PC14); 5.62% and -10.77% (PC15); 4.49% and −4.55% (PC16); 2.25% and 0.77% (PC17); 0.37% and −1.55% (PC18); −4.12% and -16.28% (PC19) (Table S5). Notably, PC10, PC11, PC15, and PC19 showed large mediation proportions disproportionate to their explained variation (1.9%, 1.7%, 1.4% and 1.1%, respectively). PC10 represented high intake of sesame paste (loading: +0.43), a noodle sauce popular in the city where this study was performed; PC11 represented low intake of red meat (loading: −0.51); PC15 represented high intake of starchy vegetables (loading: +0.38); while PC19 represented high intake of Boba tea, a hand-made sweetened beverage popular among young people in China, which were consumed among only 30%-40% of our participants (Fig. S3, supplementary methods). In this regard, we suspected that the estimation of their mediation proportions might be biased. Nevertheless, the negative directions of their mediation proportions suggested that our assessment of masking effect from ad libitum dietary intake in exercise-induced HbA1c reduction using the first four dietary PCs might well be underestimated. Higher dietary fiber and lower saturated fatty acid intakes can augment PA-associated glycemic benefits We then performed interaction analysis by categorizing key dietary factors into groups of lower and higher intakes, including intakes of dietary fiber, saturated and other fatty acids, as well as other factors associated with glycemic control, including energy intake, and intakes of protein and branched-chain amino acids. Interaction was assessed with introduction of a multiplicative interaction term of intake level group and total volume of PA, also using HbA1c as assessment of overall glycemic control. In the first place, we found that the association between total PA and reduced HbA1c was significantly enhanced among those with higher intake of dietary fiber and lower intake of total SFAs, with effect sizes of −0.18 (95% CI: −0.30, −0.06; p = 0.004) vs 0.08 (95% CI: −0.03, 0.19; p = 0.158; p-interaction=0.022) and −0.06 (95% CI: −0.18, 0.05; p = 0.283) vs −0.03 (95% CI: −0.14, 0.07; p = 0.545; p-interaction=0.009) compared with their corresponding counterparts, which was in line with the masking effect from decreased dietary fiber and increased SFAs intakes identified in mediation analysis (Fig. 5a). In addition, we found that the association between total PA and reduced HbA1c was also more pronounced among those with lower intake of total fat (p-interaction=0.001) and lower intake of total MUFAs (p-interaction=0.003) (Fig. 5a). For other dietary factors, we also observed stronger association between total PA and reduced HbA1c among those with greater proportion of energy intake from dinner (p-interaction=0.033), with effect sizes of −0.14 (95% CI: −0.26, −0.02; p = 0.028) vs 0.03 (95% CI: −0.07, 0.12; p = 0.593) (Fig. 5a, Table S8), suggesting that exercise may be particularly useful in combating glucose load associated with energy intake at late hours, which is hard to avoid in modern life. Interaction between key dietary factors and total volume of PA in reducing HbA1c and risk of T2D. a Interaction was assessed with linear mixed models with introduction of a multiplicative term of PA level and each dietary factor (high vs low intake) adjusting for treatment sequence, study period, baseline HbA1c level, age, sex, BMI, waist-to-height ratio, hip circumference, and total energy intake as fixed effects, and subjects as random effect. b Interaction between status of dietary fiber and SFA intake and total volume of PA (high vs low level) in reducing risk of incident T2D was assessed with Cox proportional hazards models adjusted age, sex, BMI, season of wear, race, baseline HbA1c level, parent history of diabetes, educational level achieved, employment status, smoking and alcohol status (n = 50 162). Participants were categorized into three groups based on whether their dietary intake exceeded or fell below the median value of the overall study population in UKB. HbA1c, glycated hemoglobin; BCAA, branched-Chain amino acids; AAA, aromatic amino acids; SAA, sulfur-containing amino acids; SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; PA, physical activity. We further examined interaction between accelerometer-assessed PA and dietary intake of fiber and SFA on risk of developing incident T2D in the UKB study with a prospective, external validation analysis. Using Cox models, we found that on the whole, higher total volume of PA (average acceleration >27.49 mg) was associated with lower T2D risk (hazard ratio [HR], 0.67 [95% CI: 0.57–0.79; p < 0.001]), and more pronounced favorable association with T2D risk of total PA was observed among those with higher intake of dietary fiber or lower intake of total SFAs (0.52 [95% CI: 0.40–0.67] vs 0.87 [95% CI: 0.66–1.14] among those without; p-interaction=0.035) (Fig. 5b). However, further reduction in the risk of incident T2D, though still significant, was not observed among those with both higher intake of dietary fiber and lower intake of total SFAs (HR, 0.71 [95% CI: 0.52–0.97]), possibly indicated that T2D risk would be harder to reduce a step further when an individual was already leading a healthy lifestyle. Discussion In this randomized controlled, three-period, three-treatment crossover trial, we performed 5-day moderate- and high-intensity endurance exercise intervention accompanied by weighed logging of consumption of all food and drinks on each day during each study period in ad libitum eating condition, totaling 8945 items of records on 972 person-days (based on each of 54 participants completing a 6-day record in each of three periods), which enabled a systematic evaluation of how exercise intervention affect dietary intake and how dietary responses to exercise intervention further affect exercise-induced glycemic benefits. Here, we report that MOD-EX and VIG-EX induced non-significant increases in total daily energy intake (7.3 and 46.9 kcal/day). Overall dietary structure was changed similarly by MOD-EX and VIG-EX interventions as revealed by PCA analysis, and the most significant manifestations of dietary responses were both decreased vegetable intake, plus increased intake of added oil in food by VIG-EX. We found that MOD-EX and VIG-EX both reduced postprandial glucose, which was assessed 37.9 ± 4.6 h from last bout of exercise intervention, independently of overall diet intake, and the effects had been masked by 7.52% and 10.85%, respectively, from ad libitum eating. The masking effects were primarily attributable to decreased dietary fiber and increased SFA intakes, exactly matching to dietary responses to exercise. Further interaction analysis also revealed augmented glycemic benefit with higher fiber or lower SFA intakes during exercise intervention, confirmed by additional evidence from the UKB study that participants with higher dietary fiber intake or lower SFA intake showed more pronounced favorable association between PA and T2D risk. To enable systematic evaluation of dietary responses to exercise and the effect of such dietary responses on exercise-induced glycemic benefits, high-quality monitoring of dietary intake is a prerequisite, which is challenging in ad libitum eating condition, especially for Chinese diet, one of the most complex dietary systems worldwide. With regard to assessment of total energy intake, the doubly labelled water (DLW) method is the gold standard [41]. Using this method, the E-MECHANIC randomized controlled trial found that DLW-assessed daily energy intake was increased by 71.1 kcal at the exercise dose level similar to that in our study, slightly larger than our observation of 46.9 kcal increase by VIG-EX intervention [11]. To make a comprehensive assessment of overall dietary structure, intake of individual food, and macro- and micro-nutrient composition, there is yet no short-cut method and it still relies on voluntary reports from participants and quantification of food intake in some way. There were two studies to date which assessed the effect of exercise intervention on intakes of both individual food and nutrients in ad libitum condition, one of which reported increase of vegetables intake and decrease of fat intake based on semi-quantification data from food frequency questionnaires among post-menopausal women from the USA [15], while the other, DR’s EXTRA trial, assessed food intake on four consecutive days using a picture booklet of portion sizes with weighing or gauges among Finnish men and women [16], and reported that exercise reduced intakes of vegetables, fruits and fiber, but increased SFAs. In dietary monitoring of our study, we used a smartphone app to facilitate diet recording for our participants, and quantification was made possible by self-report of portion size in the recording, as well as manual weighing of total weight and the weights of main ingredients of all food and drinks within the dietary records for each kind of food and drink by trained staff. Using this data, we found for the first time in the scenario of Chinese diet that both MOD-EX and VIG-EX reduced vegetables intake, and VIG-EX also increased added oil intake, in line with the DR’s EXTRA trial [16]. Of note, it is the elucidation of the ultimate effect of dietary responses to exercise on the achievement of glycemic benefits that helps to make clear what dietary recommendations should accompany exercise intervention in T2D prevention, the very knowledge gap that we intended to fill in the present study. While our primary analysis found no significant overall effect of the exercise interventions on total energy intake, we further explored potential effect modification by sex and BMI. No statistically significant interactions were observed (Table S9), suggesting that, within the statistical power of our study, the effect did not differ markedly across these subgroups. Our study had limited statistical power to detect modest subgroup interactions, given that these analyses require a substantially larger sample size. Most importantly, it underscores the notion that the widely documented heterogeneity in energy compensation [42] is likely governed by a more complex array of factors than sex or BMI alone. Furthermore, a recent review indicates that levels of body adiposity and sex do not modulate changes in appetite perception, energy intake, or macronutrient composition in response to either acute or chronic exercise [43]. Previous reviews highlight that inter-individual variability might be influenced by a confluence of hormonal and neural signals level, sex, body adiposity, and genetic factors [43–45]. Our measures of sex and BMI are broad demographic and anthropometric proxies that may not capture more precise physiological or behavioral phenotypes. For instance, variations in appetite-regulating hormones like ghrelin and peptide YY, or genetic polymorphisms affecting energy homeostasis, might explain differential responses that our study was not designed to measure [43, 44]. Future research aimed at predicting individual energy compensation responses should prioritize high-resolution phenotyping, including dynamic measures of appetite hormones, detailed behavioral assessments, and genomic data, to move beyond broad demographic categories towards truly personalized models of energy balance regulation. In this study, we first confirmed that exercise intervention can effectively reduce postprandial glucose, demonstrated as a 3.82% reduction in HbA1c (with unchanged fasting glucose level) and 11.92% reduction in glucose tAUC during the 2-h OGTT by MOD-EX, and 2.18% reduction in HbA1c and 13.48% reduction in glucose tAUC during the 2-h OGTT by VIG-EX, after an average of 37.9 h from last bout of exercise. It has been a consensus that exercise reduces postprandial glucose, confirmed by studies with 24-h glucose monitoring [17, 46, 47]. Our study extended such consensual evidence in that we can concluded with confidence that exercise-induced glycemic benefits are independent of ad libitum dietary intake and non-intervention PA, enabled by our thorough monitoring data and statistical adjustment. Although it has been established that exercise can improved glycemic control, there remains much controversies on the optimal modality and does of exercise [17–19], and heterogeneity of effect from exercise between subjects of certain characteristics [48]. Our results supported that moderate-intensity might be superior over high-intensity in endurance exercise, as we observed that one-week of MOD-EX intervention had already induced improvement in insulin sensitivity, indicated by McAuley index which well-reflected peripheral insulin sensitivity in general population [31, 49]. With the construction of RCS dose-response curve, we also found that the postprandial glucose-reduction effect of exercise plateaued at the dose around 33 MET-h/day in total PA, equivalent to an intervention does of 3.08 MET-h/day subtracting the total PA level of 29.92 MET-h/day during the control period in our study, which was close to the report of optimal dose of 2.62 MET-h/day from a meta-analysis study including 126 clinical trials [19]. With regard to effect heterogeneity, we found greater improvement in McAuley index and HOMA2-IR (hepatic insulin sensitivity) among male participants, and that exercise reduced adipose tissue insulin resistance among high-BMI participants, indicating that exercise may be of particular use in tackling T2D risk associated with the still-expanding obesity epidemic. Taken together, this study provided another line of evidence on the formidable regulating power of exercise on glycemic control, and also shed some light on the long-lasting controversies around optimal modality and does of exercise, and effect heterogeneities. The most important finding of this study is that, we estimated that ad libitum dietary intake have masked exercise-induced glycemic benefits from MOD-EX and VIG-EX intervention by 7.52% and 10.85%, respectively, and the principal responsible dietary factors were decreased intake of dietary fiber and increased intake of SFAs, exactly matching to the main dietary responses to exercise intervention aforementioned. Further interaction analysis consistently showed that higher dietary fiber and lower SFAs intakes could both augment the achieved glycemic benefit during exercise intervention, which was confirmed by prospective association evidence from the UKB study. In consistency with our findings, low dietary fiber and high SFAs intakes were both associated with increased T2D risk in epidemiological studies [12, 50]. It has been widely accepted that exercise need to be practiced in combination of dietary controls to maintain health benefits in long term [4], considering exercise itself can affect dietary behaviors. In fact, exercise intervention has been explored with popular dietary regimes such as calorie restriction and time-restricted eating. Calorie restriction in combination with exercise has succeeded in the attainment of diabetes remission among overweight or obese T2D patients in Look AHEAD and DiRECT trials [51, 52]. However, calorie restriction is difficult to maintain in long-term, exemplified in the CALERIE 2 trial, the intended 25% reduction in energy intake was achieved by only 12% over a two-year average [53]. Moreover, previous studies have showed that long-term calorie restriction bears high risk for overeating afterwards thorough persistent increase of appetitive signals (e.g., ghrelin) and decrease of satiety signals (e.g., leptin), and worse still, may lead to immune function impairment [54, 55]. As for time-restricted eating, which has attracted much attention as a simple-to-follow weight-loss strategy [56], there has also emerged worrying evidence that time-restricted eating led to aggravated atherosclerosis in an experimental study with animal models and increased risk of cardiovascular mortality in an epidemiological study [57, 58]. Dietary regimes which is safe, easy to adhere in long-term, and effective in improve glycemic control is still in need. From findings of the present study, we can safely recommend supplementation of dietary fiber and restriction on SFA intake to maximize exercise-induced glycemic benefits, while maintaining minimal alteration to day-to-day dietary habit at the same time, which, in this regard, can be expected to foster long-term adherence. We acknowledge several limitations in this study. First, we performed exercise intervention for a short duration of five consecutive days, considering that longer intervention periods presented more challenges in monitoring dietary intake. Nevertheless, we still observed significant reduction in postprandial glucose, even improvement in insulin sensitivity with MOD-EX intervention, highlighting the strong regulating power of exercise in glycemic control from another perspective. Second, the acute effect from the last bout of exercise intervention cannot be completely ruled out in this study with a time-window of 37.9 ± 4.6 h, as acute effects of exercise on glycemic improvement were reported to last up to 24-48 h [59]. However, during this process, such effect will be gradually fading away [59], and after about 38 h, including two nights at resting state, we shall say that, the contribution of acute effects from the last bout of exercise was minimal. Third, one of the primary outcomes in this study, HbA1c, was commonly considered as an indicator of the average blood glucose over an eight to twelve-week period [60], much longer than the intervention period of one week in our study, which might raise doubt in our finding that one-week endurance exercise intervention reduced about 3% reduction in HbA1c. However, in assessment of effect on HbA1c, we have already adjusted for the baseline level before intervention, and significant reduction of HbA1c after short-term exercise intervention was indeed possible, which was revealed by a pooled estimate of 0.11% reduction after a single bout of exercise in a meta-analysis study of 182 individuals [61]. More importantly, this finding was supported by a consistent observation of over 10% reduction in glucose tAUC during 2-h OGTT by intervention within a subsample in our study, indicating that our conclusion that both MOD-EX and VIG-EX interventions reduced postprandial glucose was reliable. A further limitation is that we recorded only whether female participants were menstruating at each period, not the specific phase of their menstrual cycle (e.g., follicular or luteal). The lack of granularity may have obscured more nuanced, phase-dependent associations, and future studies incorporating detailed cycle tracking are warranted. Fourth, exercise intensity was assessed using accelerometer-derived METs. While research-grade accelerometers are standard in epidemiology, their validity for estimating energy expenditure during stationary cycling is suboptimal [62]. We addressed this limitation by implementing a standardized, speedometer-controlled, and supervised exercise protocol to ensure consistent workload across all participants. Direct physiological measures (e.g., energy expenditure, heart rate) were not collected. Future studies incorporating these direct measures would strengthen these findings. Finally, OGTT assessments were performed in a subset of participants (n = 27), which may limit the statistical power of related analyses. As key characteristics were comparable between groups, data from this representative subset offer preliminary insight into potential glucoregulatory mechanisms. Conclusion In conclusion, we found in this study, with 972 person-days of weighed dietary monitoring derived from each of 54 participants completing a 6-day record in each of three periods, that at WHO-recommended does, MOD-EX and VIG-EX both reduced vegetable intake, plus increase of added oil in food intake for VIG-EX, with no significant increase in total energy intake. We also confirmed that both MOD-EX and VIG-EX could improve postprandial glucose, and found that this effect had been masked by 7.52% and 10.85%, respectively, from ad libitum eating, which were primarily attributable to decreased dietary fiber and increased SFA intakes, exactly matching to dietary responses to exercise, with supportive evidence that higher fiber and lower SFA intakes augmented exercise-induced glycemic benefit. Taken together, our findings provided evidence to recommend dietary fiber supplementation and SFA restriction accompanying exercise training to achieve maximum glycemic improvement. Open access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Supplementary information Supplementary Materials Acknowledgements The authors thank the study participants for their participating in this trial. This research has been conducted using the UK Biobank resource under Application Number 88159. The authors thanked all participants of the UKB study and the dedication and commitment of the UKB research team in their work. Author contributions SHW conducted the study, performed the statistical analysis, prepared the first draft of the manuscript, and revised the draft. HG, HC, LL, YML, MJG, YCH, SYL, WW, MJF, YRL, YHZ, XWM, and XGZ conducted the study and organized the trial data. KY conducted and supervised the study. GKQ conceived, designed and supervised the study, acquired data, provided critical revisions to the manuscript and important intellectual content, and acquired funding. All authors edited, reviewed and approved the final draft for publication. GKQ and KY jointly supervised this work. GKQ is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Funding This study was supported by grants from the Supported by Hubei Provincial Natural Science Foundation of China (2025AFB562), National Natural Science Foundation of China (82003422). Data availability The datasets generated and analysed during the current study are not publicly available due to privacy or ethical restrictions, but are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Kuai Yu, Email: yukuai5200@163.com. Gaokun Qiu, Email: qiugaokun@163.com. Supplementary information The online version contains supplementary material available at 10.1038/s41387-026-00442-w. References Associated Data Supplementary Materials Supplementary Materials Data Availability Statement The datasets generated and analysed during the current study are not publicly available due to privacy or ethical restrictions, but are available from the corresponding author on reasonable request.