Does a poor responder to one kind of training do better on another?
34 untrained young twin pairs: strength followed 12 wk of lifting, VO2max 12 wk of endurance; "responder" was any change >0 in one test, so the rescue claim is unproven.
Marsh, Thomas and colleagues (University of Western Australia, with Twins Research Australia) recruited 42 same-sex twin pairs (30 monozygotic, MZ; 12 dizygotic, DZ; zygosity confirmed by DNA test) who were untrained, and trained 34 of the pairs (24 MZ, 10 DZ).
Each pair was randomised together to start with 12 weeks of resistance training (RES) or endurance training (END), three 1-h sessions a week, then had a 12-week washout and crossed over to the other mode; twins in a pair trained together at matched relative intensities.
On average strength rose only after RES (leg press +47.0 kg vs +3.0 kg after END) and VO2max only after END (+0.25 vs +0.04 L/min after RES).
The share of people with any positive change was higher for the mode-specific training (leg press: 100% after RES, 51% after END; VO2max: 86% after END, 57% after RES), and a person's change after one mode was not correlated with their change after the other (VO2max r = -0.10, leg press r = 0.10).
MZ co-twins' training responses were correlated (leg press after RES rMZ = 0.62) while DZ co-twins' were not significantly, and the variance models attributed the response mostly to shared and unshared environment.
The authors conclude that responsiveness is mode-specific, that low responders to one mode may be "rescued" by switching, and that genes may matter less for trainability than cross-sectional twin data suggest.
Our reading: the mode specificity is solid; the rescue and the genes conclusions are not, because "responder" was any change above zero in one test (VO2max test error was 0.15 L/min), the spread of changes was about as wide after the mode that did not move the mean as after the one that did, twins trained together (a shared environment during the trial itself), and there were only 10-12 DZ pairs.
Protocol
Participants: 42 same-sex twin pairs (n = 84; 30 MZ pairs, 17 female and 13 male; 12 DZ pairs, 8 female and 4 male), 24.9 ± 5.4 y, recruited in Perth through Twins Research Australia and by advertising (newspapers, social media, university email lists, word of mouth). Inclusion: healthy, relatively unfit (below 150 min a week of activity), non-smokers, no medication. Zygosity by DNA test; 10 pairs who self-reported as DZ were MZ.
Who trained: 34 pairs (n = 68; 24 MZ, 10 DZ). Table 2 gives n = 64 for RES and n = 68 for END; Figure 3 says 64 completed both modes. Baseline MZ participants were heavier than DZ (74.1 vs 62.6 kg, P = 0.01).
Design: pairs randomised together to RES or END first; 3 months of training, a 3-month washout with usual activity and diet, then 3 months of the other mode. Tests at weeks 0, 12, 24 and 36. No group stayed untrained, so there is no non-exercise control.
Training (details in a separate protocol paper): 3 × 1-h sessions a week for 12 weeks, progressively overloaded. END: 2 running sessions and 1 cycling session a week, progressing from 60% to 90% of VO2max. RES: alternating upper- and lower-body sessions, progressing from 60% to 90% of 1RM. Twins trained together at matched relative intensities in the same period. Compliance 94% (RES) and 95% (END).
Outcomes: VO2max on a treadmill graded test (3-min stages; the four highest consecutive 15-s values), treadmill time to exhaustion, resting and maximal heart rate, peak blood lactate; 1RM leg press and bench press.
Responder definition: a "positive response" was any change above zero (post minus pre) in a single test; no threshold for test error or typical error was applied. The paper gives the metabolic cart's technical error for VO2max as 0.15 L/min.
Analysis: linear mixed models adjusted for twin correlation, sex and age; z-tests for response rates between modes; intraclass correlations for MZ and DZ pairs and ACE/CE variance-component models (A additive genetic, C shared environment, E unshared environment) in STATA. Carry-over (week 0 vs week 24 baselines) and order effects tested.
Funding: an NHMRC Principal Research Fellowship to Green (APP1080914); Twins Research Australia and Scurrah supported by an NHMRC Centre of Research Excellence grant (ID 1079102); an Exercise and Sport Science Australia Clinical Exercise Physiology Research Grant. Competing interests: none declared.
Key Results
| Outcome (change, mean (SD)) | After RES | After END | RES vs END |
|---|---|---|---|
| Leg press 1RM (kg) | 47.0 (29.4) | 3.0 (26.4) | P < 0.0001 |
| Bench press 1RM (kg) | 5.1 (5.0) | -0.4 (3.4) | P < 0.0001 |
| VO2max (L/min) | 0.04 (0.25) | 0.25 (0.26) | P < 0.0001 |
| VO2max (ml/kg/min) | 0.03 (3.57) | 3.61 (3.77) | P < 0.0001 |
| Time to exhaustion (s) | -3 (131) | 170 (137) | P < 0.0001 |
| Resting HR (bpm) | -0.1 (0.9) | -3.4 (6.5) | P = 0.009 |
| n (Table 2) | 64 | 68 | — |
| Share with any positive change (> 0) | After RES | After END | P |
|---|---|---|---|
| Leg press | 100% | 51% | < 0.0001 |
| Bench press | 79% | 28% | < 0.0001 |
| VO2max (L/min) | 57% | 86% | < 0.0001 |
| Time to exhaustion | 35% | 86% | < 0.0001 |
| Resting HR (fall) | 40% | 63% | 0.02 |
Concordance, VO2max (L/min): 47% improved with both modes, 4% with neither, 10% only with RES, 39% only with END; concordant 51% vs discordant 49% (P = 0.84). Leg press: 50% improved with both, 50% only with RES; no one had a negative change after both modes (no -/- quadrant).
Within-person correlation of the two responses: leg press r = 0.10 (P = 0.45), bench press r = -0.10 (P = 0.42), VO2max r = -0.10 (P = 0.22), time to exhaustion r = -0.10 (P = 0.43), resting HR r = 0.01 (P = 0.93).
Twin correlations of the training response: leg press after RES rMZ = 0.62 (P = 0.002), rDZ 0.49 (not significant); leg press after END rMZ = 0.36 (P = 0.04); VO2max after END rMZ = 0.45 (P = 0.02), rDZ 0.18; VO2max after RES rMZ 0.26, rDZ 0.39, both not significant.
Variance components: at baseline (cross-sectional), VO2max (L/min) A 0.3, C 0.63, E 0.07; leg press C 0.79, E 0.21. For the training response, leg press after RES C 0.64 (95% CI 0.39-0.90), E 0.36; VO2max after END C 0.43 (95% CI 0.14-0.72), E 0.57 (95% CI 0.28-0.86).No additive genetic share was estimated for any training response because rMZ and rDZ did not differ significantly or rDZ was not significant (the table's footnote says C from the MZ-only model "has a different meaning" from the usual C).
Carry-over and order: no difference between the week-0 and week-24 baselines; those who did RES first gained 0.12 L/min less VO2max from RES (P = 0.04), within the 0.15 L/min technical error.
Spread of changes (our calculation from Table 2): the SD of VO2max change was 0.25 L/min after RES, the mode that did not move the mean, and 0.26 L/min after END; for leg press 26.4 kg after END and 29.4 kg after RES.Taking the non-specific mode as a rough comparator, the true between-person SD of response would be about 0.07 L/min for VO2max and about 13 kg for leg press (REF-430 method). RES is not a true no-training control, so this is an approximation, not the paper's analysis.
Mechanisms
Measured: the training effect followed the mode: strength rose after RES and not after END, VO2max and time to exhaustion rose after END and not after RES; the between-mode differences were all P < 0.0001.
Measured: a person's response to RES did not correlate with their response to END for any outcome (all |r| ≤ 0.10).
Authors' interpretation: low responders to one mode can be "rescued" by switching mode, since "non-responders to one modality almost always respond to the other".Inference: with responder defined as any change above zero, a person with no true change has roughly even odds of looking like a responder in each block; the 57% of people whose VO2max "responded" to lifting, whose mean change was 0.04 L/min against a test error of 0.15 L/min, is close to what noise alone would produce: a mean change of 0.04 with an SD of 0.25 L/min puts about 56% above zero (our calculation).The rescue claim needs repeated tests or repeated blocks (REF-441, REF-498).
Authors' interpretation: genes "may not play as large a role" in training adaptation as cross-sectional twin data suggest, and shared environment shapes the response.Inference: the twins trained together, at the same time, at matched relative intensity, so the trial itself was a shared environment that inflates C; with 10 DZ pairs in training, a genetic share could not be separated from shared environment even if present.The HERITAGE family study gave a maximal heritability estimate of 47% for the VO2max response (REF-429); this trial does not refute that, it lacked the power to estimate A.
Inference: the SD of change after the non-specific mode was nearly as large as after the specific one (VO2max 0.25 vs 0.26 L/min), which by the comparator logic of REF-430 leaves little true individual variation in VO2max response to explain. Much of the apparent responder/non-responder split here may be measurement error and day-to-day variation.
Practical Application
Does a poor responder to one kind of training do better on another? 34 untrained young twin pairs: strength followed 12 wk of lifting, VO2max 12 wk of endurance; "responder" was any change >0 in one test, so the rescue claim is unproven.
Specificity is the solid result (B+): lifting raised leg-press 1RM by 47 kg and endurance training raised VO2max by 0.25 L/min in untrained adults, and neither did much for the other quality on average. Train the quality you want to change.
Do not label an athlete a non-responder from one block (the base's practice): this trial's own numbers (VO2max test error 0.15 L/min; 57% "responded" in VO2max to lifting with a mean change of 0.04 L/min) show how easily a single before-and-after test mislabels people. Judge a lever over 2-3 blocks with repeated tests (REF-441, REF-498); if one block is flat, more dose or more intensity in the same mode has evidence behind it (REF-996, REF-995).
Switching mode is not a cure for a flat VO2max: the lifting block did not raise VO2max on average (0.04 L/min; half the pairs lifted first), and the 10% whose VO2max rose only after lifting are within what one test's noise can produce (error 0.15 L/min). For a flat endurance block, the tested options are more dose or more intensity, not a change of mode.
Genetics talk with athletes: this trial cannot say how much of trainability is inherited (too few DZ pairs, twins trained side by side). Do not quote it as "genes don't matter" (SYNTH-13).
Limitations
"Responder" was any change above zero in one before-and-after test; no typical-error threshold, no repeated baseline, no repeat of the same block. The response rates and the "rescue" conclusion are therefore sensitive to measurement noise.
No non-exercise control group; the other training mode served as the only comparator.
10-12 DZ pairs: too few to estimate an additive genetic share of the training response; the authors name this limitation. Twins trained together, which makes the trial itself a shared environment.
Untrained young adults (24.9 ± 5.4 y), 12-week blocks: large early gains in strength and VO2max; nothing in trained, older or diseased people.
The numbers of participants do not quite reconcile across sections (68 trained, 64 in RES and in Figure 3, 68 in END); the CONSORT flow is in a figure not available in the text version.
Read from the accepted author manuscript; the typeset version may differ in detail. The pdf-to-text conversion lost ± signs in the abstract; numbers here are from the tables.
Funding: NHMRC fellowship and Centre of Research Excellence grant, and an ESSA research grant; no competing interests declared.
How this entry was checked and changed
Why this level
randomised crossover in which n = 68 trained and 64 completed both modes, registered before recruitment, read from the full text (accepted author manuscript in the University of Melbourne repository, not the typeset version). The group result (mode specificity) carries B+; the responder counts and the "rescue" claim rest on a change above zero in a single before-and-after test, with no repeated testing and no non-exercise control, and the twin heritability estimates rest on 10-12 DZ pairs, so those parts are weaker than the level suggests
Full text (checked 2026-10-22)
We recruited 42 healthy same-sex twin pairs (30 MZ and 12 DZ; aged 24.9 ± 5.4 yr); 34 of these pairs (n=68; 24 MZ and 10 DZ) participated in the training interventions (full text).
Inclusion: relatively unfit individuals who did not meet Australian guidelines for physical activity recommendations (<150 min/week); 10 pairs who self-reported as DZ were actually found to be MZ (full text).
3x 1 h sessions per week for 12 weeks; END 2x running and 1x cycling session per week progressing from 60-90% VO2max; RES progressing from 60-90% one-repetition maximum; compliance 94% for RES and 95% for END (full text).
Table 2: leg press 47.0 (29.4) after RES vs 3.0 (26.4) after END; VO2max 0.04 (0.25) vs 0.25 (0.26) L.min-1; n = 64 RES, n = 68 END (full text).
Positive response rate (>0) to RES and END: 100% and 51% for leg press, 79% and 28% for bench press, 57% and 86% for VO2max, 35% and 86% for TTE (full text).
VO2max concordance: 47% (+/+), 4% (-/-), 10% (+/-), 39% (-/+); P=0.84 for discordant vs concordant (full text).
No correlations between responses within subjects: leg press r=0.10, P=0.45; VO2max r=-0.10, P=0.22 (full text).
rMZ=0.62, P=0.002 for strength change to RES; rMZ=0.36, P=0.04 to END; VO2max change to END rMZ=0.45, P=0.02 (full text).
C 0.64; 95% CI 0.39-0.90 for leg press to RES; C 0.43 (95% CI 0.14-0.72) and E 0.57 (95% CI 0.28-0.86) for VO2max to END (full text).
Those completing RES first gained 0.12 L.min-1 less VO2max from RES (P=0.04), within the technical error of 0.15 L.min-1 (full text).