Carbohydrate ingestion during prolonged exercise and net skeletal muscle glycogen utilization: a meta-analysis
Rothschild et al. pooled every crossover trial that measured net muscle glycogen use (biopsy or MRS) in fixed-work cycling or running with carbohydrate versus placebo: 31 studies, 48 effect sizes, n = 279 participants (273 male), exercise 117.4 ± 51.6 min on average.
Carbohydrate spared muscle glycogen, but only a little: SMD −0.16 (95% CI −0.30 to −0.02), P = 0.021, with no heterogeneity (I2 = 0%) and no small-study effect. Translated to a typical trial (start 550 mmol/kg dry weight, 110 min), placebo used 270 and carbohydrate 247 mmol/kg, a sparing of ~24 mmol/kg dry weight (95% CI 4–45).
No moderator mattered: not the ingestion rate, carbohydrate type, feeding frequency, duration, exercise mode, intermittent vs continuous exercise, or starting glycogen. Intakes of ≥100 g/h were barely represented.
The authors' reading: the effect is small but consistent; single studies (mean n = 8.8) were far too small to see it, which is why 46 of 48 effect sizes were individually non-significant.
Included studies
Inclusion: peer-reviewed English papers; healthy, non-diabetic, non-obese humans; crossover with and without carbohydrate; exercise >20 min, cycling or running, external work and duration fixed between trials; net glycogen by muscle biopsy or magnetic resonance spectroscopy from the start of exercise (stable-isotope estimates excluded); vastus lateralis/quadriceps or gastrocnemius; end biopsy within 5 min.
Searched MEDLINE/EBSCO, PubMed and Scopus; dual screening; Cochrane RoB 2 for crossover trials.
31 studies, 48 effect sizes (different arms or time points), 279 participants (273 male, 6 female; age 26.8 ± 6.0 yr). Effect sizes: 40 cycling, 8 running. Exercise duration 117.4 ± 51.6 min. Glycogen units: dry weight (n = 22), wet weight (n = 23), protein (n = 1), mmol/L (n = 2). Every sample was from the quadriceps (vastus lateralis in all but one study, which reported quadriceps femoris); none from the gastrocnemius.
Risk of bias: low 10 studies, some concerns 19, high 2 (mainly randomisation, blinding and unclear drink composition).
Analysis: multilevel random-effects model (random intercept per study), SMD from paired data with a within-subject pre-post correlation of 0.5 where not reported (the median of 16 values from raw data); sensitivity across r = 0.2–0.9.
Funding: none stated in the published version; no conflicts of interest declared (Disclosures). Data and code on OSF.
Key Results
| Outcome | Result (full text) |
|---|---|
| Pooled effect, net muscle glycogen use | SMD −0.16 (95% CI −0.30 to −0.02), P = 0.021 — small sparing with carbohydrate |
| Heterogeneity | Q(47) = 31.93, P = 0.95, I2 = 0%; REML τ2 = 0; Egger-type test P = 0.989 |
| Single effect sizes | varied widely, e.g. −1.09 to +0.93 |
| Running (8 effect sizes) | SMD −0.38 (−0.74, −0.02), P = 0.036 |
| Cycling (40 effect sizes) | SMD −0.12 (−0.28, 0.03), P = 0.107 |
| Running vs cycling | not different (moderation P = 0.195) |
| Dry-weight / wet-weight units | −0.20 (−0.41, −0.009), P = 0.060 / −0.10 (−0.30, 0.11), P = 0.347 |
| Moderators (ingestion rate, frequency, type, duration, mode, starting glycogen, pre-exercise carbohydrate) | none significant (all P > 0.05) |
| Sensitivity to imputed correlation (0.2–0.9) | significant at every value |
| Absolute translation (start 550 mmol/kg dw, 110 min) | placebo −270 → 280; carbohydrate −247 → 303 mmol/kg dw; sparing ~24 mmol/kg dw (95% CI 4–45) |
| Power of single studies | mean n = 8.8 per effect size; n = 9 gives 80% power only for an effect of 1.07; 309 participants needed for 0.16 |
Mechanisms
Measured (pooled): net muscle glycogen use was slightly lower with carbohydrate, consistently in direction; ingestion rate was not a significant moderator across the rates represented.
Authors' interpretation: ingested carbohydrate raises plasma glucose, GLUT4 translocation and muscle glucose uptake, and some of the extra plasma glucose oxidation replaces glycogen. The sparing is small because carbohydrate also suppresses fat oxidation, so the drop in glycogen use must be smaller than the rise in glucose oxidation. Some glycogen synthesis during exercise may also contribute.
Authors' interpretation, not tested: if carbohydrate preferentially spares intramyofibrillar glycogen (linked to calcium release), the performance effect could exceed what whole-muscle glycogen suggests.
Authors' interpretation: any benefit of very high intakes is unlikely to come through further muscle glycogen sparing; if ≥100 g/h helps, other mechanisms are probably involved. Inference, not tested: this fits the base's reading that higher intakes act mainly through liver glycogen and whole-body carbohydrate supply (blood glucose did not differ between 60 and 120 g/h in REF-1122) rather than muscle stores.
Practical Application
Does carbohydrate during a ride save my muscle glycogen? A little: across 31 crossovers, ~24 mmol/kg dry weight over ~110 min (95% CI 4–45), against ~270 used. Ingestion rate did not significantly change the sparing, though ≥100 g/h was barely tested. Inference: the muscle saving is small and was not detectably larger at higher doses, so most of what extra grams per hour add probably comes through liver glycogen and whole-body carbohydrate supply, not muscle stores.
The sparing is real and consistent in direction, so "carbohydrate makes no difference to muscle glycogen" is wrong. Single studies called it null only because they were small (mean n = 8.8). Equally, the claim that carbohydrate speeds up muscle glycogen use (REF-476) is not supported by the pooled crossover data.
Do not use muscle glycogen to justify a higher row on the ladder. Ingestion rate did not significantly moderate the effect, so the case for 90-120 g/h rests on oxidation, gut tolerance and the performance trials in SYNTH-02, not on muscle sparing. Late-ride power at 120 vs 60 g/h is a separate question (REF-1122).
Why a small saving can still matter is the authors' argument, not a result: they note suggestions that high-intensity performance may suffer below ~250–300 mmol/kg dry weight and that 24 mmol/kg could matter in long or repeated efforts; the analysis itself says it cannot determine the practical significance.
Limitations
Mostly young men (273 of 279); the authors state that effects in women cannot be determined.
Intakes of ≥100 g/h were rarely represented, so "no dose effect" covers the range studied, not 100-120 g/h.
Every sample was quadriceps; the gastrocnemius, more used in running, was never sampled, and the running subgroup has only 8 effect sizes.
Pre-post correlations were imputed as 0.5 wherever raw data were unavailable (observed values came from 16 datasets); results held across 0.2–0.9, but observed correlations varied widely.
Many included studies are 30-40 years old; 19 of 31 carry some concerns for bias and 2 are high risk; drink composition was often unclear.
Internal inconsistencies in the paper: the abstract gives the absolute translation for "~100 min" of exercise while the discussion uses 110 min and the descriptive mean is 117.4 min; the dry-weight subgroup is called significant with a CI excluding zero but P = 0.060.
The outcome is net whole-muscle glycogen, not performance, liver glycogen or fibre/subcellular pools.
How this entry was checked and changed
Why this level
31 small crossovers (48 effect sizes, 279 participants), preregistered, no heterogeneity or small-study signal; held below A- because the CI upper bound sits at −0.02 (fragile), most trials carry "some concerns" for bias, and the outcome is a mechanism (net glycogen use), not performance; read from the full text (author repository copy of the published version, CC BY-NC-ND)
Full text (checked 2026-10-13)
31 studies, 48 unique effect sizes, 279 participants (273 male, 6 female; age 26.8 ± 6.0 yr); 40 cycling and 8 running effect sizes; exercise duration 117.4 ± 51.6 min (full text).
Pooled SMD -0.16 [95% CI: -0.30 to -0.02], P = 0.021; Q(47) = 31.93, P = 0.95, I2 = 0%; Egger test P = 0.989 (full text).
Running SMD = -0.38 [-0.74, -0.02], P = 0.036; cycling SMD = -0.12 [-0.28, 0.03], P = 0.107; mode difference P = 0.195 (full text).
Risk of bias: low risk 10 studies, some concerns 19 studies, high risk 2 (full text).
Simulation: 550 mmol·kg-1 start, 110 min; placebo decline 270 to 280 mmol·kg-1, carbohydrate decline 247 to 303 mmol·kg-1; sparing ~24 mmol·kg-1 (95% CI: 4–45 mmol·kg-1) (full text).
46 out of 48 effect sizes non-significant; average n = 8.8; 309 participants needed to detect 0.16 (full text).