Do light loads grow slow-twitch fibres more than heavy ones?
Not shown: in 8 small biopsy trials type I growth was flat across loads; type II grew a little more with heavy loads, all CIs crossing zero. Mostly untrained.
Varovic, Larsen and Grgic asked whether the load lifted (as %1RM) decides which muscle fibres grow. Earlier meta-analyses had split loads into "low" and "high" and found no significant fibre-type difference; this one modelled load as a continuous variable with multilevel meta-regressions.
Eight randomised trials of at least 4 weeks met the criteria (195 young adults, 6-12 weeks, all with vastus lateralis biopsies; six trials in untrained or physically active people, two in resistance-trained lifters; every trial but one trained to failure).
Pooling both fibre types, load added nothing to a model holding only the study design: fibre area grew by 0.489 SMC (95% CI 0.297 to 0.682), and the authors call the data insufficient to determine an effect of load.
Modelling the fibre types separately, the type II slope was positive (b = 0.054 per 10% 1RM, 95% CI −0.038 to 0.145) and the type I slope flat (b = −0.005, −0.082 to 0.072). The type II minus type I contrast was −0.147 SMC at 20% 1RM and −0.088 at 30%, near zero at 40-50%, and rose from 0.089 at 60% to 0.265 at 90%; every one of these 95% CIs included zero.
The prediction intervals ran from about −0.7 to +0.8 SMC per 10% 1RM, so a new trial could land almost anywhere. The authors conclude the pattern "suggested a potential fiber-type specific pattern" that is preliminary. Our reading: the data are compatible with a small tilt toward type II growth at heavy loads and with none.
On the primary model's point estimates the type I slope is near zero, so most of the contrast's swing from 20% to 90% 1RM (about 0.38 of 0.41 SMC) comes from type II fibres growing less at light loads rather than type I fibres growing more; but the type I slope's 95% CI (−0.082 to 0.072 per 10% 1RM) allows type I growth to differ by about half an SMC across that range, and in the aggregated type II model the two slopes were nearly equal and opposite (−0.017 and 0.019), so which fibre type drives the contrast is not settled (our inference from the slopes).
"Light loads grow slow fibres more" is not shown by the model. Fibre area is not strength or performance, and no endurance athlete was studied.
Protocol
Design: systematic review and exploratory meta-regression reported to PRISMA 2020; not prospectively registered (the authors say the analytical approach was agreed before analysis); data, code and supplementary material on OSF.
Search: PubMed/MEDLINE, Scopus and SPORTDiscus from January 2020 to 27 February 2026 (an earlier meta-analysis by one of the authors, Grgic 2020, had searched to January 2020, and all five studies of the latest meta-analyses were carried over), plus a theses database, forward and backward citation tracking in Google Scholar and the reference lists of three reviews; completed 2 March 2026.An updated search without the date restriction in August 2026 found no further study. Screening by two authors, disagreements to a third.
Inclusion: adults under 60 with no known medical condition; randomised within- or between-group resistance-training trial of at least 4 weeks; at least two groups or conditions differing in training load; fibre cross-sectional area by biopsy before and after. Type IIb fibres were coded as type IIx.
Studies (Table 1): Campos 2002 (27 physically active men; 3-5, 9-11 and 20-28 RM), Cumming 2025 (14 resistance-trained men and women; 3-5 vs 20-25 repetitions), Lees 2025 (20 untrained men; 8-12 vs 20-25 repetitions), Lim 2019 (21 untrained men; 80% vs 30% 1RM to failure; REF-1178), Mackey 2011 (12 untrained men; 70% vs 15.5% 1RM, the one trial not to failure), Mitchell 2012 (18 physically active men; 80% vs 30% 1RM), Morton 2016 (49 resistance-trained men; 8-12 vs 20-25 repetitions, full body), Schuenke 2012 (34 untrained women; 6-10 vs 20-30 RM).Six trials men only, one mixed, one women only; ages 20.8-25.0; 6-12 weeks (mean 9.6 ± 2.0); four between-participant and four within-participant (one leg each) designs; four used mATPase histochemistry and four MyHC antibodies.
Load coding: an exact %1RM was used as given; a %1RM range by its midpoint; a repetition range only (Campos 2002 alone) converted to %1RM with a published repetitions-to-%1RM meta-regression (Nuzzo 2024). Observed loads: 12 discrete values from 15.5% to 92.5% 1RM, none between 53% and 70%.
Effect size: standardised mean change (post minus pre over the baseline SD, small-sample corrected). No trial reported pre-post correlations; correlations were taken from individual data digitised from one trial (Cumming 2025) and imputed, r = 0.9 for pooled type II fibres. Lees 2025 reported change scores, so its pre mean was set to zero with the pooled baseline SD.
Models: three-level random intercepts (study, group, observation) with REML and cluster-robust variance in metafor. Linear, linear-log, two-knot restricted cubic spline and quadratic forms compared by BIC-approximated Bayes factors (Kass and Raftery scale), each with and without study design as a covariate; the simpler model kept when close.Estimated marginal means and slopes (per 10% 1RM) with 95% CIs and prediction intervals at anchors from 20% to 90% 1RM; type II minus type I contrasts at each anchor.
Secondary analyses: an aggregated type II model (type II subtypes IIa, IIa/x and IIx dropped), leave-one-out (which also tests the non-failure trial, Mackey 2011), and moderators (training status, direct sets per week, volume equating, design, rest), each fitted separately. Publication bias not assessed (k = 8).
Quality: TESTEX (0-15), two assessors.
Funding and interests: no financial support; the authors declare no commercial or financial conflict. ChatGPT (GPT-5.6) was used for proofreading and language, per the authors not for scientific content.
Key Results
Type II minus type I fibre growth, by training load (negative = more type I growth, positive = more type II growth; estimated marginal means after adjusting for study design):
| Load | Fibre-type-specific model, SMC (95% CI) | Aggregated type II model, SMC (95% CI) |
|---|---|---|
| 20% 1RM | −0.147 (−0.345 to 0.052) | −0.108 (−0.319 to 0.103) |
| 30% 1RM | −0.088 (−0.252 to 0.076) | −0.073 (−0.254 to 0.109) |
| 40-50% 1RM | negligible (no numbers given) | negligible (no numbers given) |
| 60% 1RM | 0.089 (−0.076 to 0.253) | 0.035 (−0.093 to 0.162) |
| 70% 1RM | 0.147 (−0.052 to 0.347) | 0.070 (−0.059 to 0.200) |
| 80% 1RM | 0.206 (−0.037 to 0.449) | 0.106 (−0.037 to 0.249) |
| 90% 1RM | 0.265 (−0.026 to 0.556) | 0.142 (−0.023 to 0.306) |
No data were observed between 53% and 70% 1RM, and the 20%, 60% and 90% anchors had no observed load near them (the authors' limitation).
Overall model (both fibre types): study design was the only useful predictor (2×ln(BF) = 6.62 over intercept-only); adding load gave no further evidence. Fibre growth 0.489 SMC (95% CI 0.297 to 0.682; 95% PI −0.364 to 1.343), a small effect on the authors' Cohen scale; R² 36.1% (fixed) and 70.7% (fixed plus random); I² 69.1%, mostly at group level (τ² 0.06).
Fibre-type model: linear-log (2×ln(BF) = 7.16) and linear (7.12) forms with study design fitted alike; linear kept. Slopes per 10% 1RM: type II 0.054 (95% CI −0.038 to 0.145; 95% PI −0.729 to 0.836), type I −0.005 (95% CI −0.082 to 0.072; 95% PI −0.786 to 0.776). R² 41.5% and 73.4%; I² 69.4% (group-level τ² 0.06). The load × fibre type interaction was positive; its estimate is not given in the text.
Aggregated type II model: slopes 0.019 (95% CI −0.016 to 0.055; PI −0.680 to 0.718) for type II and −0.017 (95% CI −0.052 to 0.019; PI −0.716 to 0.682) for type I, the same direction; the type II slope about a third of the primary model's, the type I slope about three times larger (−0.017 vs −0.005), so the gap between them shrank from 0.059 to 0.036 per 10% 1RM; R² 27.7% and 66.4%; I² 67.9%, now at study level (τ² 0.05).
Leave-one-out: the interaction stayed positive in all eight runs (0.043 to 0.069), with the 95% CI excluding zero in two. Dropping Campos 2002 gave the smallest estimate, 0.043 (95% CI 0.005 to 0.081); dropping Mackey 2011, the only trial not to failure, 0.059 (95% CI −0.017 to 0.134).
Volume equating (moderator): type II slope 0.060 (95% CI −0.028 to 0.149) in volume-equated trials (k = 3) and 0.048 (95% CI −0.092 to 0.188) in the others (k = 5); type I slopes 0.001 (−0.045 to 0.048) and −0.011 (−0.151 to 0.13). Other moderators had wide, overlapping intervals (hypothesis-generating in the authors' words).
Quality: TESTEX mode 7/15 (range 6-11); study quality mode 2/5 (1-3), reporting mode 5/10 (5-8); no trial reported its randomisation method or allocation concealment.
The arithmetic that does add up (our check): the gap between the two fibre-type slopes (0.054 − (−0.005) = 0.059 per 10% 1RM) equals the step between successive contrasts (for example 0.265 − 0.206), and 0.019 − (−0.017) = 0.036 matches the aggregated model's steps (0.035, 0.070, 0.106, 0.142); the 195 participants equal the sum of Table 1's samples; 9.6 ± 2.0 weeks matches Table 1's durations.
Numbers that do not add up: (1) Table 1 marks Mitchell 2012 and Schuenke 2012 as volume-equated ("Yes/Yes"), while the text lists only Campos 2002, Mackey 2011 and Lees 2025 as volume-equated and the moderator analysis uses that split (k = 3 vs 5); one of the two is wrong.(2) Table 1's TESTEX scores (7, 11, 9, 8, 8, 6, 10, 7) give two modes, 7 and 8, not a single mode of 7 (our count). (3) The median sample is given as 20; Table 1's samples (12, 14, 18, 20, 21, 27, 34, 49) give 20.5 (our arithmetic; trivial).(4) The Discussion calls both aggregated-model slopes "attenuated" against the primary model, but only type II shrank (0.054 → 0.019); type I grew in magnitude (−0.005 → −0.017).
Abstract vs body: the abstract reports only the fibre-type contrasts; that load added nothing when both fibre types were pooled is stated in the body and conclusion but not in the abstract. The numbers the abstract gives match the body.
Mechanisms
Measured: fibre cross-sectional area from vastus lateralis biopsies; the type II slope was positive and the type I slope flat across 15.5-92.5% 1RM, both with 95% CIs including zero.
Authors' interpretation: near failure, light and heavy loads both recruit high-threshold motor units (the size principle), which is why whole-muscle growth is similar across loads; at fibre level the opposing directions of type I and type II slopes may cancel when the fibre types are pooled, hiding a fibre-type-specific response.Rare type IIx and IIa/x fibres, few or absent in some biopsies, may have added group-level variance; dropping them kept both slopes' direction, but the type II slope fell to about a third (0.054 → 0.019) while the type I slope grew more negative (−0.005 → −0.017); the authors call both "attenuated", which holds for type II only (our reading).
Inference: on the primary model's point estimates type I fibres grow about the same at 20% and at 90% 1RM, so the negative low-load contrasts come mostly from type II fibres growing less; but the type I slope's 95% CI (−0.082 to 0.072 per 10% 1RM) is wide, and in the aggregated type II model the type I slope (−0.017) was about as large as the type II slope (0.019), so a real type I advantage at light loads is not excluded.This fits REF-1178 (one of the eight trials), where 80% and 30% 1RM to failure both increased type I fibre area.
Inference: fibre area is fibre size, not oxidative capacity. Nothing here measured mitochondria, capillaries or fatigue resistance, so it gives no support to the claim that light loads build the slow-fibre engine (SYNTH-89).
Inference: in an endurance athlete, concurrent endurance training and a different fibre-type mix could change any of these responses; no trial in the review tested that.
Practical Application
Do light loads grow slow-twitch fibres more than heavy ones? Not shown: in 8 small biopsy trials type I growth was flat across loads; type II grew a little more with heavy loads, all CIs crossing zero. Mostly untrained.
Do not choose a light load to "target slow fibres" (A-, imprecise, untrained people): type I fibre growth did not depend on load in this model, and in one of the included trials heavy loads to failure grew type I fibres too (REF-1178, B+). Choose the load by the goal: heavy for strength, economy and late-race power (SYNTH-43, SYNTH-44).
For muscle size, loads of roughly 30-90% 1RM taken near failure grew fibres alike as far as this review can tell: pooled fibre growth was about 0.49 SMC and adding load to the model gave no further evidence (the 15.5% arm, Mackey 2011, was the one trial not to failure), which agrees with whole-muscle data (REF-677, A); a possible extra for type II fibres at heavy loads is a hypothesis, not a reason to change a programme.
Fibre growth is not strength or performance: the review measured fibre area only; for strength the heavier load still wins (REF-677, A), and no endurance outcome was measured.
What it cannot show: any effect in trained endurance athletes, between 53% and 70% 1RM, in muscles other than the vastus lateralis, in women (two trials), or of tempo, continuous tension or proximity to failure (seven of eight trials trained to failure).
Limitations
Few, small, mostly untrained trials. Eight trials of 12 to 49 participants, six of them in untrained or physically active people; the authors say this makes it hard to generalise to trained people, and the two trials in resistance-trained lifters reported negligible fibre growth per the authors.
Imprecision. Every pairwise contrast and every slope has a 95% CI including zero, and the prediction intervals span roughly −0.7 to +0.8 SMC per 10% 1RM; the authors call the findings exploratory and hypothesis-generating.
Gaps in the load range. No observed loads between 53% and 70% 1RM; the 20%, 60% and 90% anchors are model estimates with no observed data near them; loads were estimated from repetition or %1RM ranges, and for one trial converted from repetitions.
Model choices. Pre-post correlations were imputed (r = 0.9 for type II fibres, derived from one trial); type II subtypes were pooled with aggregated type II in the primary model; the review was not pre-registered; moderator subgroups rest on k = 2 to 6 (2 trained vs 6 untrained trials, 3 vs 5 volume-equated, 4 vs 4 by design); publication bias was not assessed.
Trial quality. TESTEX mode 7/15; no trial reported randomisation method or allocation concealment; several did not report drop-outs or adherence.
One muscle, few women. All biopsies from the vastus lateralis; women in two trials only, so no sex analysis.
Reporting. Table 1 and the text disagree on which trials equated volume (Mitchell 2012 and Schuenke 2012); Table 1 lists only the two failure arms of Lim 2019, not its volume-matched 30% arm short of failure (REF-1178), and only the TS and TE arms of Schuenke 2012, whose title names a slow-speed regimen; the sample sizes of 21 and 34, and so the total of 195, appear to count participants in arms not listed (our reading); the primary interaction estimate is not given numerically.
Funding and interests. No financial support; no conflicts declared. Grgic authored the earlier meta-analysis this one updates. Generative AI was used for proofreading (disclosed).
How this entry was checked and changed
Why this level
meta-analysis of mostly small trials (12 to 49 participants), read from the full text (open access, CC BY). The rubric places a systematic review of mostly small trials at A-; the trials scored a mode of 7/15 on TESTEX per the text (Table 1's scores have two modes, 7 and 8) and none reported its randomisation method or allocation concealment, the review was not pre-registered, and every load contrast is imprecise (all 95% CIs include zero). The level grades how the synthesis was done; the fibre-type pattern itself is, in the authors' words, preliminary and hypothesis-generating
Full text (checked 2026-10-26)
Eight studies involving 195 participants were included (full text)
Pairwise contrast estimates were negative at low loads (20–30% 1RM: −0.147 to −0.088 SMC) and positive at higher loads (60–90% 1RM: 0.089–0.265 SMC); the 95% confidence intervals included zero for all pairwise contrasts (full text)
Overall model: estimated marginal mean 0.489 SMC (95% CI: 0.297, 0.682; 95% PI: −0.364, 1.343); adding load as a predictor did not provide additional evidence beyond study design (full text)
Marginal slope of training load for type II fibers b = 0.054 (95% CI: −0.038, 0.145; 95% PI: −0.729, 0.836); for type I fibers b = −0.005 (95% CI: −0.082, 0.072; 95% PI: −0.786, 0.776) (full text)
The observed data spanned only 12 discrete load values between 15.5% and 92.5% 1RM, with a complete absence of observed data between 53% and 70% 1RM (full text)
Leave-one-out: interaction estimates ranging from 0.043 to 0.069; the 95% CIs excluded zero in two of the eight iterations (full text)
Volume-equated studies: b = 0.060 (95% CI: −0.028, 0.149); non-volume-equated: b = 0.048 (95% CI: −0.092, 0.188) for type II fibers (full text)
The mode score on the TESTEX scale was 7/15 (range: 6–11/15); none of the included studies reported methods of randomization or allocation concealment (full text)
Financial support was not received for this work; ChatGPT (GPT-5.6) was used for proofreading and language correction (full text)