Low and high carbohydrate isocaloric diets on performance, fat oxidation, glucose and cardiometabolic health in middle age males

Highly trained competitive middle-aged athletes underwent two 31-day isocaloric diet periods (HCLF or LCHF) in a randomized (www.randomizer.org↗), counterbalanced, crossover design with a two-week washout period between dietary interventions without feeding limitations. We assessed both diets while controlling calories and training load. Primary outcomes were performance, substrate oxidation during exercise, continuous glucose and cardiometabolic biomarkers. Each subject visited the laboratory on ten separate occasions, performing testing before (PRE) and at the completion (POST) of each 31-day dietary intervention (Figure 2). Visit one and two consisted of a familiarization of measurement instruments, equipment, perceptual measurements (42, 43), consent, VO₂max test (31), and continuous glucose monitoring (CGM; Freestyle Libre 2, Abbott Diabetes Care Inc; Almeda, CA) sensor application. One-mile [1,609 m] running time trial (TT) and repeated sprint protocol (RSP; 6 × 800 m) were performed twice on each dietary intervention (Pre and Post). One-mile TT was performed on Day –4 and Day 28. Three days later, subjects performed the RSP on Day –1 and Day 31. Gas exchange and perceptual changes were recorded throughout each performance trial Pre and Post each dietary intervention. Body composition and cardiometabolic parameters were measured Pre and Post each dietary intervention. Capillary and interstitial metabolite concentrations were measured throughout each 31-day dietary intervention period. Testing sessions were conducted at the same time of day in an environment with controlled temperature and humidity (19-21 ⁰C, humidity = 35-40%) within the Exercise Science Laboratory of Grove City College. The experimental protocol was approved by the Institutional Review Board of Grove City College prior to implementation (IRB number 110-2021).

Ten middle-aged competitive distance runners (Figure 2) were recruited directly from local running organizations and by advertising within the local community. Inclusion and exclusion criteria were screened utilizing medical history/Physical Activity Readiness Questionnaire (PAR-Q). Inclusion criteria included: (1) 30 to 50-years old males; (2) completion of one-mile (1,609 meter) run in under seven minutes; (3) running ≥ 20 miles (32 kilometers) per week; (4) > 2 years of running experience; (5) currently consuming a carbohydrate-based diet (> 50% kcals). Exclusion Criteria included: (1) cigarette smoker; (2) metabolic or cardiovascular disease, (3) orthopedic, musculoskeletal, neurological, and/or any medical conditions that prohibit exercise; (4) psychiatric disorder; (5) pharmaceutical drugs affecting measurement parameters. Subjects were prohibited from using any ergogenic aids/supplements one month prior to study initiation to study completion. Subjects were instructed to refrain from caffeine and alcohol consumption for 48 h, and racing or training for 24 h before each performance test.

Using direct counseling and prepared educational handouts, a registered dietitian taught and guided each athlete prior to the experimental phase on how to implement the LCHF and HCLF diets at home. Instructional handouts included: (i) summary of key aspects of each diet, (ii) 31-day LCHF and HCLF meal plan but were advised to use this meal plan as a guide rather than a strict protocol; (iii) detailed guide on acceptable low-carbohydrate foods as well as a recommended list of nutritious fat and protein-rich foods. The primary macronutrient targets for LCHF and HCLF were expressed as both a percentage of total daily energy intake and daily gram intake: LCHF: < 50 g/day carbohydrate, 75–80% fat, 15–20% protein; HCLF: 60–65% carbohydrate, 20% fat, 15–20% protein (Table 1). Additionally, the LCHF group was recommended to supplement their diets with added salt to taste at mealtime and supplement 1-2 g/day of sodium from bouillon cubes, or homemade broth (44).

Weekly energy intake and relative macronutrient distribution were monitored and estimated via 3-day weighed food records, capturing two consecutive weekdays and a weekend day via the online smartphone application, MyFitnessPal. Subjects used digital kitchen scales to measure the weights of food portions for total energy intake estimates. Researchers administrated subject’s MyFitnessPal user accounts and therefore had the ability to assess and modify the subject’s macronutrient and micronutrient intake throughout the intervention (45). A two-week recovery macrocycle was incorporated between each dietary intervention without dietary restriction.

Verification of compliance to the LCHF diet was done via capillary blood ketone concentrations on days 3, 7, 14, 21, and 28 before ingesting breakfast and exercising. Capillary blood ketone concentrations (R-β-hydroxybutyrate; Precision Xtra, Abbott Diabetes Care Inc., Almeda, CA) was also measured immediately before and after the LCHF diet by the primary researcher. The results from the dietary intake and ketone measurements enabled the registered dietician to individually fine-tune participants’ diets, if necessary, via phone or email, thus ensuring continuous nutritional ketosis while on the LCHF diet.

Participants were instructed to maintain a training log (mode, duration, and intensity of each workout) during the study intervention without increasing or decreasing the training load (46). Training load was calculated by using the session-RPE method (RPE x session duration [min]) (47).

Participants reported to the laboratory at least ≥ 3 h post-prandial, refraining from exercise for 48 h. The measurements of body mass were performed on a medical scale with a precision of 0.1 kg. Body composition was evaluated using the electrical impedance technique (Tanita^® MC-980U, Tanita Corporation, Inc., Tokyo, Japan).

On the second visit, subjects performed an incremental test to exhaustion on a motorized treadmill (Trackmaster TMX425C treadmill, Newton, KS) utilizing the modified Astrand treadmill protocol. The treadmill was calibrated before each exercise test according to the manufacturer’s instructions. Oxygen consumption (VO₂) and carbon dioxide production (VCO₂) was assessed using an automated metabolic analyzer system (TrueOne 2400, ParvoMedics, Sandy, UT) calibrated prior to each exercise test using standard calibration gases (16% O₂ and 4% CO₂). Subjects wore a Polar heart rate monitor (Polar Electro, Kempele, Finland) during exercise to measure heart rate. After a thorough explanation of the experimental procedures, each subject was instructed to walk on the treadmill for 3 min as a warm-up at a self-selected speed (0% grade). Immediately following the 3-min warm-up, the speed was increased to 8–13 km.hr^–1 for 3 min (0% grade) to achieve the subjects’ comfortable running pace. After 3 min of running at 0% grade, the grade was increased 2.5% every 2 min throughout the test protocol while speed was kept constant. At the end of the test, the highest VO₂ value recorded during the last 30 seconds of exercise was considered the subject’s VO_2max (48). Some or all of the following criteria were used to determine if a physiologically valid VO_2max had been attained: (a) a plateau in VO₂ with increasing exercise intensity (< 150 ml/min or < 2.1 ml/kg/min), (b) a respiratory exchange ratio (RER) of ≥ 1.1, (c) and volitional termination due to exhaustion.

Participants arrived at the laboratory in the morning at least ≥ 3 h post-prandial and performed a one-mile TT on a motorized treadmill (TMX425C treadmill, Trackmaster, Newton, KS, USA). Subjects performed a warmup run consisting of 5 min at 45% VO_2max followed by 5 min at 65% VO_2max (10 min total). After a 5-min passive rest period, subjects then initiated the one-mile TT (1,609 m). Prior to the TT, the treadmill was brought to a standstill (0 km.hr^–1), all timing devices were reset, the distance covered on the treadmill monitor was reset, a 5-s count down was given, and the TT began. Each TT commenced with a 30-second rolling start at a running speed approximating 80% VO_2max, after which subjects could freely adjust their speed. Running speed and time was not visible to subjects during the TT. Subjects were verbally informed of the distance they had covered at 200m intervals. The treadmill gradient was set to 1% throughout the exercise (49). Subjects were instructed to finish the TT as fast as possible and were not informed of the overall performance time until completion of the final testing session. Heart rate (Polar Electro, Kempele, Finland), RPE (RPE-Overall; RPE-Chest; RPE-Legs) and affect (Feeling Scale) were recorded at 200 m intervals during the TT. Lastly, RPE and affect for the entire exercise session (session RPE and session affect) were obtained 5 min following the one-mile TT. Expired respiratory gases were continuously collected during the entire time trial and capillary blood was collected Pre- and Post-TT.

Participants arrived at the laboratory in the morning at least ≥ 3 h post-prandial. Upon arrival to the laboratory subjects completed a 10 min self-paced warm-up on the treadmill. The RSP prolonged high intensity interval protocol was chosen because it requires a high level of aerobic oxidative, as well as anaerobic glycolytic contributions (50). The RSP was performed on a treadmill and consisted of 6 × 800 m (0.5 miles) sprints, with a three-minute passive recovery interval. Subjects were instructed to finish each 800 m sprint as fast as possible. The time it takes to complete each sprint was recorded. No feedback on sprint times was given to the participants during trials. Verbal encouragement was provided during maximum effort sprints in a standardized fashion throughout each visit. Heart rate, blood samples, RPE, affect, and metabolic gases were collected throughout.

Respiratory gas exchange was recorded using an automated metabolic analyzer system (TrueOne 2400, ParvoMedics, Sandy, UT, United States). Prior to each experimental session, the device was calibrated using procedures according to manufacturer instructions. The breath-by-breath measurements were performed for oxygen uptake (VO₂), carbon dioxide production (VCO₂), and respiratory exchange ratio (RER) and was measured continuously throughout trials (one-mile TT and RSP). The average values for VO₂ (L/min) and VCO₂ (L/min) were calculated over the last minute of each 2-min exercise stage in the maximal exercise test, and each minute of the one-mile TT. Whole-body rates (g/min) of CHO and fat oxidation were calculated using intensity dependent equations that assume negligible protein contribution to energy expenditure (51).

Blood samples were measured via fingertip blood samples collected using a lancet following cleaning of the fingertip with an alcohol swab and then dried. The first droplet was wiped away with a cotton swab to remove any alcohol and the subsequent droplets were used for analysis. Samples were immediately processed for measurement of blood lactate (Lactate Plus, Nova Biomedical), ketones (R-β-hydroxybutyrate; Precision Xtra, Abbott Diabetes Care Inc., Almeda, CA) and glucose (Precision Xtra, Abbott Diabetes Care Inc., Almeda, CA) concentrations.

Blood samples were collected using a validated dried blood spot (DBS) card technique (52). DBS cards were allowed to dry for 30 min, and then stored at −30°C prior to shipment to ZRT CLIA-approved Laboratory (Beaverton, OR) for immunoassay analyses as previously described (52, 53). DBS were assayed for HbA_1c, total cholesterol, low-density lipoprotein cholesterol (LDL-C), very low-density lipoprotein cholesterol (VLDL-C), high-density lipoprotein cholesterol (HDL-C), triglycerides, insulin, and high-sensitivity C-reactive protein (hsCRP). Dried blood spot testing has shown a strong correlation with conventional serum tests, making it a reliable and convenient tool for screening cardiometabolic risk factors (52).

Participants’ interstitial glucose concentrations were measured throughout each 31-day dietary intervention via CGM (Freestyle Libre 2, Abbott Diabetes Care Inc; Almeda, CA) utilizing Levels, Inc., mobile application software to capture glucose response and trends in real-time (Levels Health, Levels, Inc., New York, New York). CGM tracks long-term to HbA_1c (54–56), short term CGM readings (10-14d) are good estimates of 3-month CGM averages (57), and (iii) CGM can also capture both fasting and post-prandial differences in glucose (validated diagnostic tools). Participants were instructed to apply a 14-day sensor to the back of the arm following the manufacturer’s written instructions and demonstration video. Participants were instructed to place it in the same location throughout. CGM subcutaneous sensor measured interstitial glucose values every 15 min which was transmitted using near-field communications. Participants obtained their glucose values after scanning for up to 8 h of data. All 24-hour CGM readings were included into daily and 31-day statistics. All CGM sensors were obtained in batch to control for any potential manufacturing discrepancies across different sensor batches.

Statistical analyses were performed using SPSS version 26.0 (SPSS Inc., Chicago, IL), and MATLAB 2022a (MathWorks Inc., Natick, MA). Statistical significance was set a priori at p < 0.05. Descriptive statistics were calculated for all variables. Normality and absence of large outliers were verified by using the Shapiro-Wilks test, observing the normality plots, and residual plots. Repeated measure analyses of variance (ANOVA) were utilized to assess physiologic, metabolic, respiratory, perceptual, and performance time, treatment, and interaction effects. Bonferroni post hoc was utilized to control for multiple comparisons when main effects were observed. A paired samples t-test was used to analyze differences in macronutrient composition and 31-day glycemic variable averages between the two dietary interventions. A one-way repeated measures analysis of variance was used to analyze differences over time for training load. The assumption of sphericity was confirmed using Mauchly’s test. Greenhouse-Geisser epsilon corrections were used when the sphericity assumption was violated. Partial-eta squared (η2p) was used to report the effect sizes for the above metrics, where appropriate. Simple linear regression analyses were run to determine the relationship between glycemic, substrate oxidation, and biochemical parameters. Slope intercept (y = mx + b), r², and p values represent best-fit equation, the goodness of fit, and slope significantly non-zero values, respectively. To avoid (58) assumptions of normality in continuous data, a kruskal Wallis (KW) test with Dunn’s correction for multiple comparisons was utilized to determine differences across the entire 24 h circadian window for the all data in the 31d CGM response window. Time points every 4th h across the entire 24 h period were included in the statistical comparison, corresponding to ∼1 point maximum per ultradian cycle (59, 60). All data are reported as Median ± SD, with exception of circadian glucose patterns presented as Median, 25th, and 75th percentile. Sample size (n = 10) was determined based on prior studies where one-mile TT performance differences were observed using dietary interventions in elite runners which was adjusted for the expected increase in one-mile TT performance time variance across cohort in non-elite athletes (61).

Daily dietary nutrient intakes are summarized in Table 1. All participants (n = 10) completed all the required study duties. The energy intake between LCHF and HCLF treatments remained isocaloric. Significant differences were detected for every diet composition variable, notably for carbohydrate and fat intake. LCHF consumed less absolute and relative carbohydrates compared to HCLF, largely driven by the 10-fold reduction in simple sugars and approximately one-half the dietary fiber. Conversely, LCHF consumed significantly more dietary fat and cholesterol, both as absolute and relative amounts, compared to HCLF. While attempts were made to control protein intake across groups, athletes consumed an extra 31 grams of protein during the LCHF treatment (p < 0.01), likely due to the close association between dietary fat and protein.

Average capillary R-βHB (mean ± SD: 0.76 ± 0.04 mM; range: 0.3 – 2.2 mM) during LCHF increased significantly from baseline and remained within the range of nutritional ketosis (≥ 0.5 mM R-βHB) throughout the intervention (Figure 3).

There were no significant differences for weekly training load, either within- or between dietary phases (Table 2).

Participants started each diet at similar weight and body composition. There were no significant treatment or interaction effects for weight or body composition during either LCHF or HCLF (Table 3). Overall changes in weight and body composition on each diet were similar. Significant time effects were detected for weight and BMI in both LCHF and HCLF treatments. Approximately ∼80% of weight loss composition was derived from fat mass (Δ: 1.8 ± 0.6 kg; p = 0.05) in 4:1 ratio fat:fat-free mass, while the remaining portion of weight loss was derived non-significantly from fat-free mass (Δ: 0.6 ± 0.2 kg).

All 10 participants completed the treadmill one-mile time trial (1,609 m) at a self-selected pace before initiating either diet. No significant baseline (Pre) differences were observed across physiological, metabolic, respiratory, perceptual, or performance parameters. Significant differences were detected within-diet and between-treatments at the post-timepoint (Table 4).

Within-diet, LCHF mean respiratory exchange rate (Δ: −0.08 ± 0.02; −6%), and mean carbohydrate oxidation (Δ:−1.06 ± 0.36 g/min; −20%) decreased significantly from Pre to Post, whereas heart rate (Δ: 4 ± 1 bpm; 3%) and mean fat oxidation (Δ: 0.44 ± 0.16 g/min; 190%) increased significantly from pre to post (all p < 0.05). The significant decrease in respiratory exchange rate from Pre to Post-LCHF revealed a decrease in carbohydrate reliance (91 to 73%) and an increase in fat utilization (9 to 27%). Additionally, these effects were corroborated by the absolute (g/min) carbohydrate and fat oxidation rates, which revealed that for every 1 g/min decrease in carbohydrate oxidation there was an expected 0.42 g/min of increase in fat oxidation during LCHF. There were no significant changes in substrate oxidation and respiratory exchange rate detected pre- to post-HCLF.

Significant interactions revealed that Post-LCHF athletes had a higher heart rate (Δ: 6 ± 2 bpm), mean fat oxidation rate (Δ: 0.62 ± 0.21 g/min), mean respiratory rate (Δ: 1.0 ± 0.3 bpm), and a lower carbohydrate oxidation rate (Δ: 1.75 ± 0.60 g/min) compared to Post-HCLF (all p < 0.05).

All 10 participants completed the treadmill series of 6-sets of 800m sprints at a self-selected pace. Before each dietary phase there were no significant differences in physiological, metabolic, respiratory, perceptual, or performance parameters. Additionally, neither diet influenced repeated sprints running performance post-diet intervention (Table 5 and Figure 4).

During RSP, LCHF induced very high rates of fat oxidation which peaked at 1.58 ± 0.33g/min during 86.40 ± 6.24%VO_2max (range: 0.99 to 2.01 g/min) compared to 0.69 ± 0.24g/min on HCLF during 79.67 ± 3.15%VO_2max (range: 0.32 to 1.13 g/min). Interestingly, 30% of the subjects on LCHF had peak fat oxidation rates > 1.85 g/min (1.95 ± 0.08 g/min; range: 1.86 to 2.01 g/min). To our knowledge, these are the highest rates of fat oxidation ever recorded. Within-diet, LCHF mean respiratory exchange rate (Δ:−0.06 ± 0.02; −7%), and mean carbohydrate oxidation (Δ:−1.69 ± 0.57 g/min; −54%) decreased significantly from pre to post, whereas mean fat oxidation (Δ: 0.69 ± 0.23 g/min; 92%) increased significantly from pre to post (all p < 0.05). The significant decrease in respiratory exchange rate indicated decreased carbohydrate reliance (53 to 35%) toward fat utilization (47 to 65%). Additionally, the absolute carbohydrate and fat oxidation rates (g/min) demonstrated that for every 1 g/min decrease in carbohydrate oxidation there was 0.41 g/min increase in fat oxidation. No significant Pre-Post changes were detected within-HCLF (Figure 5 and Supplementary Table 1).

There was a trend for greater mean relative VO₂ consumption (p = 0.07) and lower rate-of-perceived exertion, both as overall body RPE-O (p = 0.095) and session RPE (p = 0.08), when comparing post-LCHF to post-HCLF. This increased relative oxygen consumption after LCHF is largely explained by weight-loss, and partly explained by the non-significant increase in heart rate (i.e., greater cardiac output) and fat oxidation, altogether increasing tissue demand for oxygen and thereby augmenting relative VO₂.

There were no significant changes over time in any of the variables of interest (Figure 8 and Supplementary Table 4). Between-condition effects reveal higher total cholesterol (Δ: 20.7 ± 3.3 mg/dL; p = 0.001) and LDL-C (Δ: 10.7 ± 4.6 mg/dL; p = 0.03) after LCHF. Interaction effects revealed total cholesterol (Δ: 29.5 ± 6.3 mg/dL; p = 0.007) and HDL-C (Δ: 11.4 ± 3.3 mg/dL; p = 0.045) were significantly greater Post-diet on LCHF treatment.

All glycemic parameters significantly improved on LCHF (Figures 9, 10). Average glucose was significantly lower during LCHF treatment starting day 8, and remained lower on day 13, 15-20, and 22 (Figure 9). 31-day mean and median glucose levels were reduced −15.0% (range: −34.3 to 0.7%; p = 0.0057) and –15.2% (range: −33.2% to 0.4%; p = 0.0058) on LCHF treatment, respectively. 31-day time in range (70-110 mg/dL) increased 35.4% (range: 1.1 to 122.3%), standard deviation reduced −34.9% (range: −57.5 to 5.7%), and coefficient of variance reduced −25.4% (range: −41.0 to 12.4%) on LCHF. 31-day daily glucose minimum and maximums were also reduced on LCHF. 31-day median glucose was significantly lower throughout fasting (00:00 to 07:00) and feeding (07:01 to 23:59) windows (p = 0.004;Figure 10). Interestingly, 30% of subjects had 31-day mean and median glucose (mean glucose range: 111.7-115.2 mg/dL; Figure 9) and fasting glucose > 100 mg/dL (Figure 10) throughout HCLF treatment, which is consistent with prediabetes glycemic values (62). These subjects were also the greatest responders to carbohydrate restriction (range: −21.5 to −34.3% mean glucose). When observing whether there was a significant relationship between the mean glucose while on HCLF versus the percentage change in mean glucose between LCHF and HCLF, we observed a significant (p = 0.0077) interaction with a large effect size (r² = 0.6094) indicating that those individuals with a higher mean glucose, are more responsive to carbohydrate restriction treatment (Figure 9). Importantly, 30% of subjects who had a 31-day average mean, median, and fasting glucose > 100 mg/dL on HCLF (range: 111.68-115.19 mg/dL), were also the largest glycemic responders to carbohydrate restriction and also reported the highest fat oxidation rates during LCHF, with a large inverse relationship (p = 0.0069; r² = 0.6194) across the entire cohort between the percent change in mean glucose when switching to LCHF and the peak fat oxidation rate at 86.4% VO_2max indicating that those individuals with the greatest change in glycemic control also had the greatest shift in global metrics of systemic metabolic adaptation to diet. To explore whether peak fat oxidation rates on LCHF were associated with circulating lipids, we found that higher peak fat oxidation (p = 0.0034; r² = 0.6775) predicted higher total cholesterol, with trends for triglycerides (p = 0.0730; r² = 0.3474; X = Peak Fat Oxidation on LCHF; Y = Triglycerides on LCHF; Y = 29.06*X + 24.62), suggesting a potential relationship between changes in fat oxidation rates and circulating lipid metabolism. These findings in the 30% of subjects with pre-diabetes in our study could not be explained by underlying demographic, body composition or running experience as these subjects with glycemic values consistent with pre-diabetes (62) had near equivalent age (pre-diabetic: 41.67 y/o; cohort: 39.3 y/o), running experience (pre-diabetic: 8.67 y; cohort: 9.70 y), lower weight (pre-diabetic: 84.03 kg; cohort: 86.70 kg), BMI (pre-diabetic: 25.37 kg/m²; cohort: 26.2 kg/m²), body fat [% (pre-diabetic: 14.8%; cohort: 15.7%) & kg (pre-diabetic: 12.7 kg; cohort: 14.1 kg)], and relative VO_2max (pre-diabetic: 60.97 mL/kg/min; cohort: 58.70 mL/kg/min). Additionally, this prediabetic phenotype was present in these subjects despite them losing weight on both nutritional strategies (LCHF: –2.3 ± 1.3 kg; HCLF: –2.1 ± 3.2 kg).

Performance during the 1,609 m time trials was the same when athletes ate HCLF or LCHF diets. This is in keeping with our previous study (31) in which the 5-km time trial performances of athletes, similar in ability to those studied here, were equivalent on either diet. It adds further weight to the conclusions from two recent meta-analyses (63, 64) that that LCHF and HCLF diets produce equivalent performances across a wide range of athletic events.

Further, to search for a performance difference between dietary interventions, we asked the athletes to perform an interval session involving six repetitions of 800m at a pace equivalent to an exercise intensity of ∼85% VO_2max. Our reasoning was that if the pre-exercise muscle glycogen stores are a critical determinant of exercise performance and if the LCHF diet is associated with lower muscle glycogen concentrations in recreational athletes (28) (but perhaps not in highly competitive athletes (65)), and since very high rates of muscle glycogen use are measured during 800m repetitions (66) so that, if significant muscle glycogen depletion can be produced by a high intensity interval session, then any impaired performance of athletes eating the LCHF diet should become apparent in the latter intervals of that session.

For example, Impey et al. reported rates of muscle glycogen use of 12.7 and 7.0 mmol/min in the gastrocnemius and vastus lateralis muscles respectively of recreational male athletes performing 800m repetitions at 100% VO_2max (66). Webster et al. reported that pre-exercise vastus lateralis glycogen concentrations were 85 mmol/kg in well-trained recreational athletes eating the LCHF diet (28). Whilst appreciating that rates of muscle glycogen use are reduced in those eating the LCHF diet (28, 65), according to these data a starting muscle glycogen concentration of 85 mmol/kg in the vastus lateralis muscle would be depleted after just 12 minutes of high intensity exercise. Our athletes exercised for ∼21 minutes during the 6 × 800m repetitions; sufficient time to produce substantial glycogen depletion.

In contrast to our expectation, based on this prediction that significant muscle glycogen depletion would occur in athletes following the LCHF diet and this would impair their performance, in fact exercise performance was identical across all the intervals on either diet (Tables 4, 5; Figure 4 and Supplementary Table 1).

These finding raise the important question of why our two studies have failed to detect diet-induced differences in performance whereas prior meticulously conducted studies (16–18) detected meaningful differences in their studies of Olympic standard race walkers. Five key factors may have contributed to these differences: randomization, dietary controls during exercise, training load, body composition, and dietary habituation timeline. Prior studies with differing results allowed subjects to choose the diet they preferred (16–18). In addition, during the final performance trial involving simulated races of up to 25-km, subjects on the LCHF diet were provided with carbohydrate-free “non-caloric fluid (water or artificially sweetened drinks)” (16) and “LCHF cookies” (17) whereas subjects on the HCLF diet received “sports drink, sports gels and confectionary” providing “∼60g CHO” every hour (16, 17). As a result, blood glucose levels were lower in the LCHF group in the two trials (16, 17) in which it was measured, with a trend toward a progressive hypoglycemia in one trial [figure 5A from (17)]. As the authors of those studies appreciate, even in the absence of hypoglycemia, carbohydrate ingestion alone can have an ergogenic effect even if the carbohydrate is not ingested (67). Thus, these trials did not control for the potential effects of carbohydrate ingestion during exercise. The potential role of hypoglycemia in explaining differences in exercise performance has recently been revisited (29). The intensified training load and across group differences in body composition in these trials (16–18) also illustrates key differences as increased physical activity levels (68) and body weight reductions (69) both illustrate biological stressors requiring adaptation and may independently impact performance. The increased physical activity across groups and more significant reductions in bodyweight in LCHF arm (16–18), on top of introducing a diet which requires systemic metabolic reprogramming (70), illustrate three co-administered biological stressors all requiring adaptions and which may influence performance. Lastly, our analyses allowed for a 4-week adaptation timeline, which is ≥ 33% longer than prior trials describing negative performance impacts (16–18). Thus, it is not surprising that when we controlled randomization (within-subject), dietary controls during performance testing, calories, training load, and body compositional changes across groups to allow for the isolation of diet-induced changes across these key parameters, we observed different results from prior observations (16–18). Of note, when Burke et al. (16) attempted to repeat their findings (18), they did not detect differences between LCHF and HCLF in real world race performance via IAAF points [figure 5C from; (16)], a point acknowledged by the authors in the abstract.

The described method for measuring maximal rates of fat oxidation during exercise is to have subjects exercise for short periods of approximately 3 minutes at exercise intensities that gradually increase (22–24). Using this method, the highest rates of fat oxidation are generally achieved at submaximal exercise intensities of between 55-72% VO_2max. Maximal rates of fat oxidation measured with this method are usually in the range of 0.5-0.6 g/min. Importantly, at higher exercise intensities rates of fat oxidation fall precipitously, reported reaching 0% at exercise intensities > 85% VO_2max.

Higher rates of fat oxidation have been measured in athletes adapting to the LCHF diet. Volek et al. measured rates of 1.2 g/min in elite ultra-marathon runners performing prolonged exercise (180 minutes) at 64% VO_2max with peak fat oxidation of 1.54 g/min at 70.3% VO_2max (65). Webster et al. measured identical values in well-trained recreational athletes during 120 minutes’ exercise at 55% of peak power output (28); whereas Burke et al. measured rates in excess of 1.4-1.5 g/min during the final km of 25-km time trials in Olympic class race walkers (16, 17). In a case study of an elite Ironman triathlete, Webster et al. reported a peak fat oxidation rate of 1.6 g/min in a single cyclist for the full duration of a 100-km cycle trial performed at an average power output of 224 W (28, 71). Shaw et al. demonstrate fat oxidation ranging from 0.88 to 1.51 g/min following 31-day of LCHF habituation (72). Our data is in line with these prior studies showing elevated fat oxidation rates (LCHF: 1.58 ± 0.33 g/min).

Thus, the values for fat oxidation rates measured in these recreational athletes are unusually high, particularly when they were measured under experimental conditions expected to produce values close to 0g/min. However, what is particularly unique in our findings is that we observed the peak fat oxidation rates (LCHF: 1.58 ± 0.33 g/min) at 86.40 ± 6.24% VO_2max. Additionally, 30% of subjects reported record high fat oxidation rates > 1.85 g/min (1.95 ± 0.08 g/min), ranging from 1.86 to 2.01 g/min, which to our knowledge is the highest rates ever recorded. These unique findings may be a result of subject age and athletic status and/or due to the unique controls incorporated into this study to isolate diet-induced effects (i.e., randomization, calories, training load, and body composition changes controlled across groups).

Three out of eight cardiometabolic markers were significantly modulated by diet, most notably post-diet LCHF vs. HCLF total cholesterol (238 vs. 208 mg/dL, 14%; p = 0.007) and HDL-C (68 vs. 57 mg/dL, 20%; p = 0.045) concentrations. Although the overall LDL-C condition effect was significant (p = 0.03), the LCHF diet did not increase LDL-C concentrations beyond HCLF after four-weeks (155 vs. 137 mg/dL; 18% p = 0.30), implying that post-diet hypercholesterolemia effects were predominantly determined by changes in HDL-C fraction. We anticipated based on our prior work (73) and others (65, 74, 75) that competitive runners would experience significant blood lipid changes within weeks after starting a LCHF designed to induce nutritional ketosis (≥ 0.5 mM R-βHB). The significant main effects in this study were directly attributable to the between-diet differences in dietary fat and fat composition, however, the fact that more than half of participants had borderline elevated total cholesterol, LDL-C, and HDL-C at baseline was somewhat unexpected. While carryover effects were ruled out by identical concentrations at baseline (i.e., both conditions started the same), this effect may have indirectly revealed that a peak or plateau had not been reached over our four-week intervention, making blood lipid trends harder to speculate. Moreover, it is unclear if similar dietary interventions in mildly hypercholesterolemic athletes will exert any significant impact on cardiometabolic indices that are beyond the effects induced by their habitual diet. Individual cardiometabolic responses are available for review in the supplement (Supplementary Figure 1).

We detected a small, but significant change in weight over time, primarily derived from fat mass. Despite energy intake and training load being similar between LCHF and HCLF diets, weight-loss and diet did not modulate triglycerides, insulin/HbA_1c, and inflammation (hsCRP). Based on prior evidence (73, 76) the LCHF diet was projected to lower cardiometabolic markers beyond a HCLF diet, even in the absence of weight-loss (77); however, we did not observe these results with our between-diet isocaloric feeding design. Additionally, it is important to acknowledge that HbA_1c is a 2–3-month biomarker that we quantified to predict directional trends rather than significant changes over four weeks. Based on our findings, we expect that these markers will continue decreasing after four-weeks of LCHF, similar to isocaloric HCLF feeding when duration, energy intake, and weight are controlled between conditions.

Continuous glucose monitors are unique tools which allow researchers to extract changes in glycemic control every ≤ 15 min over extended periods of time without relying on the limitations at a single timepoint which may only provide limited biological feedback (78, 79). Importantly, (i) CGM tracks long-term with HbA_1c (54–56), (ii) shorter term CGM readings (10-14d) are good estimates of 3-month CGM averages (57), and (iii) can also capture both fasting and post-prandial differences in glucose (validated diagnostic tool) demonstrating a powerful monitoring tool for glycemic control, particularly in interventions not long enough to confirm alterations in HbA_1c (< 2-3 months in length). The value of these tools has been shown in previous observations utilizing lower carbohydrate (80) or ketogenic interventions (81) where continuous glucose patterns found key shifts in glycemic control before and/or in the absence of changes in traditional biochemical cardiometabolic biomarkers. When measuring continuous glucose levels every 15 min over a 31-day period, we observed improvements across all glycemic parameters in virtually all subjects on the LCHF diet, with initial significant differences in mean glucose observed on day 8 of dietary habituation. Carbohydrate restriction is a known therapeutic strategy to help facilitate improvements in glycemic control and other key metabolic parameters in other clinical conditions such as obesity (82), type-1 diabetes (36), and type-2 diabetes (33, 35). The improved glycemic mean, median, and variability on the LCHF diet in middle-aged athletes was observed in a rigorous, randomized cross-over study design without the confounding influence of caloric, training load/physical activity, and body composition differences across diets. These illustrate critical controls allowing us to extract diet-induced impact on glycemic parameters as prior observations have shown that caloric intake and changes in body weight both influence glucose levels regardless of diet (83, 84). Additionally, prior evaluations have found that intensified training programs can disrupt not only glycaemia and mitochondrial function, but also performance (68).

While there have been small short-term investigations exploring glycemic control during exercise in athletes (79, 85), very few studies have investigated the relationship between the long-term (i.e., ≥ 1 month) 24-h glycemic control while also observing performance. Nolan et al. reported the impact of a ketogenic diet on an individual type-1 diabetic cyclist during a 20-day, 4011-km race (86). This case report demonstrated remarkable glycemic control for a Type-1 Diabetic compared to historical glycemic norms for Type-1 Diabetics during this 20d race window, but nature of the report did not allow for the comparison of performance on- and off-diet.

While all subjects were “healthy,” with normal bodyweight and BMI, and competitive middle-aged athletes without any medical diagnoses, we observed that when continuously monitoring glucose parameters over a 31-day period, 30% of subjects on a HCLF diet had mean, median, and fasting BG > 100 mg/dL, consistent with pre-diabetes interstitial glucose values using analogous technology (62). This is consistent with a prior analysis which found that 30% of sub-elite endurance athletes exercising > 6-hour per week had undetected pre-diabetes when measured via continuous glucose monitoring devices (87). These subjects fitting the pre-diabetes glycemic phenotype in our study could not be explained by underlying demographics, body composition or physical activity differences as these pre-diabetic subjects had near equivalent age (pre-diabetic: 41.67 y/o; cohort: 39.3 y/o), running experience (pre-diabetic: 8.67 y; cohort: 9.70 y), lower weight (pre-diabetic: 84.03 kg; cohort: 86.70 kg), BMI (pre-diabetic: 25.37 kg/m2; cohort: 26.2 kg/m²) and body fat [% (pre-diabetic: 14.8%; cohort: 15.7%) & kg (pre-diabetic: 12.7 kg; cohort: 14.1 kg)], and higher VO_2max (pre-diabetic: 60.97 mL/kg/min; cohort: 58.70 mL/kg/min) when compared to the entire cohort. This is in line with the understanding that multiple factors contribute to diabetes onset (88, 89), some of which may go undetected until overt diagnosis. Potential explanations for early pathogenic progression of diabetic dysglycemia include genetic predisposition, adiposity-induced insulin resistance, fasting insulin, and beta-cell dysfunction (88). However, markers of elevated adiposity were not higher in the prediabetic group. In fact, this sub-cohort lost weight on both dietary protocols. Additionally, circulating lipids tended to be lower on the HCLF diet suggesting lipids could not explain dysglycemia on HCLF. Fasting insulin was not different across diet groups, nor were baseline or post-HCLF diet fasting insulin levels associated with elevated mean glucose on HCLF (p = 0.4048). However, this does not exclude the possibility that dynamic changes in post-prandial insulin or beta-cell function couldn’t be contributing to this effect which we did not measure herein. While intense exercise overtraining has also been demonstrated to acutely disrupt mitochondrial and glycemic function (68), this dysfunction was reversed following reduction in activity and cannot explain our results as our subjects did not increase or decrease physical activity levels. While genetic predisposition may explain why 30% of healthy, active and normal weight individuals had pre-diabetes on HCLF, it cannot explain across treatment effects as the crossover design controlled for this variable. Our results demonstrate that all these subjects were able to reduce their mean and median glucose < 100 mg/dL on LCHF diet. While prior observations have demonstrated the ability to improve glycemic control in individuals with established pre-diabetes and obesity in the absence of exercise consumed a low-carbohydrate diet (< 100 g carbohydrates/day) (62), to our knowledge, this is the first observation to demonstrate the ability to detect and resolve pre-diabetes without changes in activity or changes across groups in body composition using carbohydrate restriction in competitive middle aged athletes. Importantly, Al-Ozairi et al. found that a 6-day LCHF diet in Type-2 Diabetic subjects who kept calories and bodyweight controlled were unable to find differences in mean and post-prandial glycaemia utilizing CGM devices (90). This could be due to the short treatment duration as we observed significant differences on day 8 of the isocaloric HCLF and LCHF diets. Alternatively, it may be explained by the influence of engaging in physical exercise regularly as Moholdt et al. found a 5% reduction in mean glucose levels using CGM without changes in bodyweight following a 5-day lower carbohydrate diet (i.e., 15% kcal carbohydrates) and exercise in obese subjects who were of similar age to our cohort (80). Our larger reduction in mean (15%) and median glucose (15.2%) compared to Moholdt’s (5%) is likely explained by Moholdt’s higher percentage of carbohydrate (15%), shorter duration of diet (5-days), and different metabolic phenotype in their cohort (i.e., sedentary; BMI > 30 kg/m²). Our observation of consistent improvements in glycaemia in middle-aged athletes (30% pre-diabetic glycemic phenotype), without impairing performance, illustrates a promising therapeutic strategy for improving glycemic control, without requiring body composition and physical activity change.

Our study found that 30% of subjects who had a 31-day average mean and median glucose > 100 mg/dL on HCLF (range: 111.68 – 115.19 mg/dL; pre-diabetic phenotype), were also the largest glycemic responders to carbohydrate restriction. Importantly, when looking to observe if the entire cohort also observed a relationship between 31-day average mean glucose on HCLF diet and percentage change in mean glucose between LCHF and HCLF diet, we observed a large significant inverse relationship, indicating that those individuals with a higher mean glucose, are more responsive to carbohydrate restriction treatment, not just those with pre-diabetic glycemic phenotypes. As our study and prior literature suggests this change is in response to diet and not other factors (i.e., calories, body composition, altered activity levels, cardiometabolic factors), thus indicating that as individuals develop increasing levels of mean BG, they may become more responsive to therapeutic carbohydrate restriction. Interestingly, the 30% of subjects who had a 31-day average mean and median glucose > 100 mg/dL on HCLF (range: 111.68 – 115.19 mg/dL; pre-diabetic phenotype), also reported the highest fat oxidation rates during LCHF, with a large inverse relationship (r² = 0.6194; p = 0.0069) across the entire cohort between the percent change in mean glucose when switching to LCHF and the peak fat oxidation rate at 86.4% VO₂max indicating that those individuals with the greatest change in glycemic control also had the greatest shift in global metrics of systemic metabolic adaptation to diet. While multiple studies have shown reductions in glucose (35, 80, 83) and elevations in fat oxidation (16–18, 28, 31, 91) on a LCHF diet, we are unaware of any data which has demonstrated that the magnitude of glycemic changes across diet predicted the magnitude of peak fat oxidation rates. Interestingly, we also observed that higher peak fat oxidation levels on LCHF predicted higher total cholesterol on LCHF suggesting a potential interaction between higher rates of fat turnover and higher levels of circulating lipids while on a diet that restricts carbohydrates and increases fat intake. While elevated fat oxidation rates have been observed on LCHF diet in the absence in changes of insulin or calories, explained by elevated fat intake, (91), they did not see a change in glucose levels nor did they explore whether the magnitude of fat oxidation rate was associated with glucose or lipid parameters. In line with our data, there has been a report demonstrating that individuals with healthy bodyweight undergoing a LCHF diet can have elevated circulating lipid (i.e., LDL-C) (92). While this prior observation did not look at either total cholesterol or fat oxidation rates, in light of our data, there remains a possibility that these individuals (92), have elevated levels of systemic fat oxidation which requires further analyses. However, in disagreement with these findings we did not observe a relationship between fat oxidation and LDL, a relationship between baseline Trig/HDL predict LDL-C, nor any LDL-C levels “hyper-responders” (LDL-C: > 200 mg/dL) which indicates our findings may not translate to this unique cohort (92). The ability for (i) 31-d mean glucose on HCLF to predict changes in mean glucose following carbohydrate restriction, (ii) changes in mean glucose with carbohydrate restriction to predict peak oxidation rates, and (iii) peak fat oxidation to predict total cholesterol suggests a unique predictable physiologic relationship between glycemia, substrate oxidation, and circulating lipids biomarkers which requires further validation.