This is the adaptation side of the resistance-training lever, and it is the evidence Challenge #10 flagged as missing when it separated training-for-strength-gain from training-for-mortality (Muscle-Strengthening Activity and Mortality). Two things distinguish it: it is RCT-grade (unlike the observational mortality data), but it is on surrogates (muscle mass and strength), not patient-important outcomes. The headline: resistance training is the driver; protein is a modest adjunct, and only up to about 1.6 g/kg/day.
(inferred from Morton et al., 2017)
Protein augments resistance-training gains — modestly
From «Data from 49 studies with 1863 participants» (RCTs, «RET ≥6 weeks»), protein supplementation added to resistance training, versus placebo/no-supplement:
| Outcome | Added effect of protein (MD) |
|---|---|
| 1RM strength | «2.49 kg (0.64, 4.33)» |
| Fat-free (lean) mass | «0.30 kg (0.09, 0.52)» |
| Muscle fibre CSA | «310 µm2 (51, 570)» |
| Mid-femur muscle CSA | «7.2 mm2 (0.20, 14.30)» |
| Fat mass | «−0.41 kg (−0.70,– 0.13)» |
| Maximal voluntary contraction | «SMD: 0.04 (-0.09, 0.16)» — no effect |
| Total body mass | «0.11 kg (−0.23, 0.46)» — no effect |
Read the magnitudes before the significance. +2.5 kg on a 1RM and +0.3 kg of lean mass over weeks of training are small increments on top of what the training itself delivers — protein does not build muscle, it lets the training build slightly more. (And it did not move total body mass or maximal voluntary contraction at all.)
How much protein — a plateau near 1.6 g/kg/day, with real uncertainty
The decision-relevant number is a dose-response knee: «Protein supplementation beyond total protein intakes of 1.62 g/kg/day resulted in no further RET-induced gains in FFM». This is a mechanism-supported but statistically non-significant plateau (the break-point fit was non-significant — detailed below) — «muscle protein synthesis (MPS)… shows a saturable dose-response relationship», so above the point where synthesis saturates, extra protein has nothing to add. The acute-MPS saturation is a real mechanism; it is the chronic-FFM break-point operationalizing it as “1.62” that is not statistically established.
But hold the number loosely (the lesson from Challenge #10): the break point is «1.62 (1.03, 2.20)» g/kg/day — the confidence interval runs from ~1.0 to ~2.2, so “~1.6” is a central estimate with wide uncertainty, not a precise threshold. And the baseline matters: control groups were already eating «pre: 1.4±0.3» g/kg/day, so the useful move is reaching roughly 1.6, not exceeding it — going higher buys no further lean mass.
Two different objects, not a contradiction: reach ~1.6, don’t exceed describes the plateau of the mean FFM effect within the observed range; the bias up to 1.8-2.2 for hypertrophy move (Decision relevance, below) is a decision under the wide CI — hedging against undershoot where the knee’s true location is uncertain — not a claim that more protein raises the mean effect.
The acute mechanism is real — a saturable MPS dose-response is the kind of curve where a knee is expected, and a saturating supply (protein → synthesis) plateaus where a risk-reduction curve need not. But the existence of the chronic-FFM knee, not merely its location, is statistically uncertain, which downgrades this from a located knee to one asserted on mechanism. On the daily-intake axis Morton’s biphasic fit «explained more variation than a linear regression» but «is presented as a segmental regression despite not being statistically significant (p=0.079» (Morton et al., 2017), with only R2=0.19; the Figure 5 caption labels the break point «break point=1.62 g protein/kg/day, p=0.079» (Morton et al., 2017). On the baseline-intake axis the data went the other way: «linear regressions explained significantly more variance than biphasic regressions» in both young and old participants (Morton et al., 2017) — a straight (monotone) line fit significantly better than a knee. By the wiki’s own dose-response rule — the burden is on whoever asserts a knee to locate one, and a shape equally consistent with monotone has no diagnostic value — this plateau is therefore held as weak: mechanism-supported, but not a statistically-established knee, and not clearly distinct from the monotone nutrition-reduction curves. Do not read “1.62” as a demonstrated threshold.
The denominator is TOTAL body mass — and the number is undefined until the kg is named. Morton’s break-point regressed the change in fat-free mass against «baseline protein intake (g/kg/day)» (Morton et al., 2017): the intake is grams per kg of scale-weighed body mass, while «total body mass (TBM; measured by any scale)» and FFM/FM are the outcomes, not the denominator (Morton et al., 2017). So “1.62 g/kg” means 1.62 g per kg of total body weight, not per kg of lean mass — the page’s ~1.6 target is g/kg-total. This matters only where the two diverge: in a lean person total and lean mass differ ~1.2x, so the denominator barely moves the gram target; in an obese person they diverge ~2x, so “1.62 g/kg” is not even well-defined until the kg is named.
Whom the number was measured on — the transportability envelope. The 49 RCTs (17 countries, 1962-2016) required adults who were «healthy and not energy-restricted» and «performing RET at least twice per week»; among them «10 studies in resistance-trained participants and 14 study groups in exclusively female participants», and «a total of 1863 participants (mean±SD; 35±20 years)» (Morton et al., 2017). No baseline BMI, body-fat %, or body weight appears in the held text for the pooled sample [searched: BMI / body-mass-index / body-fat / adiposity / obese / waist / kg-m2 across the Participant-characteristics + Methods/Inclusion sections of chunk 01 — zero hits; the two BMI hits in the source sit in chunk 03 competing-interests/references, not the participant description]. This is an absence from the MAIN TEXT, not a demonstrated absence from the paper: Morton reports «Participant details and outcomes are presented elsewhere (see online supplementary table 1» (Morton et al., 2017), and that supplementary table is not held — participant body composition may be characterized there. And the intakes actually sampled were narrow but ran higher than the CI: the protein group’s relative intake went from «pre: 1.4±0.4, post: 1.8±0.7» g/kg/day (Morton et al., 2017), so with a post-mean of 1.8 many study arms exceeded ~2.2. The break-point CI (1.03-2.20, above) is the interval on the knee’s location, not the x-axis span of the data — the knee and its CI sit inside the sampled intakes, and the flat segment above 1.62 was fit from the higher arms. (Superseded 2026-08-07: an earlier version here read 1.03-2.20 as the sampled range; that conflated the location-CI with the data span.)
The observed-range caveat governs the whole curve. The whole analysis lives in a fairly narrow band of total-protein intakes (post-mean 1.8±0.7 g/kg-total) in a non-energy-restricted, non-obese sample, so read the shape as described within that band, not as a universal dose-response. The lower tail (below ~1.0) is barely sampled; the upper arms extend above ~2.2 but thin out, so the plateau above 1.62 rests on relatively few high-intake arms.
The obese are OFF-SUPPORT — an insufficient-evidence gap, NOT a dose on lean mass finding. Two hard facts: (a) Morton reports no obese stratum and the held text leaves adiposity uncharacterized — «healthy and not energy-restricted» was required, and the energy-restricted trials where the obese typically concentrate were excluded (Morton et al., 2017); (b) as above, the g/kg denominator is total body mass, and total-vs-lean diverges ~2x in the obese, so “1.62 g/kg” is undefined for them until the kg is named. It is tempting to close this by dose on lean mass instead, but that move is mechanism-only, discounted, and does not settle its own sign: the obese often carry more absolute lean mass (which pushes the gram target up, not down), while obesity brings anabolic resistance (plausibly raising the per-lean-kg requirement). Net-effect-not-intended: do not conclude obese -> lower target. The honest object for the obese is a gap — (i) do not transport 1.62 g/kg-total to them; (ii) specify which denominator any target uses; (iii) the right obese target is unobserved here.
Consolidated GAP — a defensible target by stratum, on a named basis (a soft knee is not guidance)
[2026-08-28]. The number that would change a decision — g/kg, on which denominator, for whom — is
unresolved wherever the person departs from Morton’s lean/normal-BMI trained sample: non-athletes, the
obese, older adults, and anyone in a caloric deficit. Two reasons it is a gap and not a finding: 1.62 is
the CI-midpoint of a non-significant knee (p=0.079), so it is not robust guidance even in-sample; and the
strata that matter are off-support. What is held is stratum-scattered and mostly answers does protein help,
not what target — Bauer PROT-AGE, Coelho-Junior, Moore (older adults); Refalo, Kim, Wycherley (deficit);
Morton/Tagawa (trained adults). The obese target on a named basis is the thinnest cell, and a gold SR+MA
aimed squarely at it is the acquisition that would move this — reported to give a protein amount for preserving
muscle during weight loss in overweight/obesity; held as a candidate, not a finding, until ingested.
confidence: low
The independent second opinion — Tagawa’s 1.3 g/kg knee (a different quantity, not a contradiction)
Tagawa 2020 is a non-Phillips-lineage dose-response MA — a Japanese group (Miyachi lab), zero author overlap with Morton — pooling «138» RT + non-RT trials / «5866» individuals (~2-3x Morton’s 49). It lands a diminishing-returns knee at 1.3 g/kg BW/d, and the first instinct — 1.3 contradicts 1.62 — is wrong. The parameter table shows why they are not the same quantity:
Independence is partial, not clean — disclose it before leaning on it. Zero author overlap is necessary, not sufficient: Tagawa cites Morton (ref 14) as the finding it is «consistent with», and its 138-trial pool re-analyzes some of Morton’s own RT constituent trials. So for the RT knee Tagawa is not independent evidence (shared trials defeat independence exactly as shared authors do); the genuine independence is the without-RT leg, built on non-RT trials Morton’s RT-only pool excludes. Hence F-refinement, not a clean type-E backing. (inferred from Tagawa et al., 2020)
| Parameter | Morton 2018 | Tagawa 2020 | Same quantity? |
|---|---|---|---|
| Knee location | «break point=1.62 g protein/kg/day, p=0.079» (chunk 02) | «1.3 g/ kg BW/d» slope-drop (chunk 01) | see rows below |
| Outcome | RET-induced FFM | LBM change | ~yes (LBM≈FFM) |
| Population | RT adults, «performing RET at least twice per week» (chunk 01) | «diverse population», WITH or without RT (chunk 01) | NO — Morton RT-only |
| Estimator | biphasic/segmented break-point (non-sig, p=0.079) | multivariate-adjusted spline slope-drop | NO — different method |
| Shape above knee | plateau: «no further RET-induced gains in FFM» (chunk 01) | overall spline «over a wide range of doses (from 0.5 to 3.5 g/kg BW/d) was positively correlated with an increase in LBM» (no plateau); in model 2, above 1.3 with-RT «continued to rise», without-RT «declined» (chunk 01) | NO — Tagawa’s overall curve has no plateau, and its RT arm keeps rising |
Fourth column is NO -> a refinement/distinction, not a tension. The two measure diminishing returns in different populations by different estimators, so their numbers are not rival estimates of one knee. What Tagawa adds is decisive for the hold-it-loosely reading: (Tagawa et al., 2020)
- The knee is population-dependent, and Tagawa’s RT arm does NOT plateau at 1.3. «In model 2 … after a total protein in- take of 1.3 g/kg BW/d was exceeded, the effect on LBM change continued to rise with resistance training and declined without resistance training.» So the post-1.3 decline is the non-RT subgroup; the RT subgroup keeps gaining above 1.3 — directionally consistent with a higher RT knee like Morton’s 1.62, not against it. The one place a genuine clash could live — the RT population above 1.62, where Morton says plateau and Tagawa’s RT arm says still-rising — is defused only because Morton’s plateau is itself non-significant (p=0.079). The inflection is thus population-dependent (general/non-RT ~1.3, RT higher/absent) -> the target is a wide, population-dependent region (~1.3-1.6), not a point -> The Underivable Optimum.
- The curve is monotone-positive to 3.5, never flat. «total protein intake over a wide range of doses (from 0.5 to 3.5 g/kg BW/d) was positively correlated with an increase in LBM» — the slope drops at 1.3 («0.39 kg (95%CI, 0.36- 0.41) and 0.12 kg (95%CI, 0.11-0.14) per 0.1 g/kg BW/d increment … below and above 1.3 g/ kg BW/d») but stays positive. This matches Morton’s significantly-better linear fit on the baseline axis and Refalo’s monotone-wins result: no lineage locates a true plateau.
- Protein raises LBM even WITHOUT resistance training — new to the fabric. «this meta-analysis demon- strates for the first time that protein supplementation is significantly effective without resistance training». RT is additive, not synergistic («no syner- gistic effects, but it may have a simple additive effect»; Table 2 with-RT 0.48 vs without-RT 0.53 kg). This refines the page’s headline: RT is the driver of hypertrophy, but protein alone still adds ~0.5 kg LBM — relevant where RT is not on the table (frail/elderly/dysphagia). Even «less than 0.3 g/kg BW/d (0.17 g/kg BW/d, on average) was suffi- cient to significantly increase LBM».
Weighting caveat (symmetric standards). Tagawa is gold-tier (large SR-MA, spline dose-response) but 5 of 7 authors incl. the first author «are employees of Meiji Co, Ltd» (a protein/dairy company), «No external funds supported this work.», and blinding-related bias was high (whole-food protein cannot be double-blinded — same design ceiling Morton hits). With sponsor-employed authors making the analysis choices on an unblindable exposure, the commercial interest discounts the framing and cautions the magnitude (the direction is held by Morton/Refalo independently; the effect size is the exposed parameter). (Tagawa et al., 2020)
Protein timing — the peri-workout “anabolic window” is small-to-null once daily total is adequate
The dose sections above answer how much; this answers when — and the answer is that when barely matters once the daily total is met. Schoenfeld’s 2013 meta-regression is the first MA to test the peri-workout “anabolic window” on chronic strength and hypertrophy (not the acute muscle-protein-synthesis spike). It pooled 23 studies (525 subjects, 132 effect sizes for hypertrophy; 478 subjects, 20 studies for strength), mean PEDro 8.7, defining a timing “treatment” as >=6 g essential amino acids taken <=1 h pre- and/or post-exercise vs a control taking no protein within 2 h. (Schoenfeld et al., 2013)
The apparent timing benefit is total protein in disguise. The unadjusted model showed a small-moderate, significant hypertrophy edge for peri-workout timing («difference = 0.24 ± 0.10; CI: 0.04, 0.44; P = 0.02»; strength was non-significant). But it vanished once total protein was controlled: «any positive effects associated with protein timing on muscle protein accretion disappeared after controlling for covariates», and «discrepancies in total protein intake ex- plained the majority of hypertrophic differences noted in timing studies.» The mechanism is mundane — the timing groups simply ate more: «The average protein intake for controls in the un- matched studies was 1.33 g/kg/day while average intake for treatment was 1.66 g/kg/day.» Total protein was the strongest predictor of effect-size magnitude («estimate = 0.41 ± 0.14; CI: 0.14, 0.69; P = 0.004»), at «a ~0.2 increase in ES noted for every 0.5 g/kg in- crease in protein ingestion». (Schoenfeld et al., 2013) (Schoenfeld et al., 2013)
Author verdict + decision. «These results refute the commonly held belief that the timing of protein intake in and around a training session is critical to muscular adaptations and indicate that consuming adequate protein in com- bination with resistance exercise is the key factor for maximizing muscle protein accretion.» Practically: «current evidence does not appear to sup- port the claim that immediate (<=1 hour) consumption of protein pre- and/or post-workout significantly en- hances strength- or hypertrophic-related adaptations to resistance exercise.» So the peri-workout window is a small lever, dominated by the same total-daily-protein rock the dose sections above are built on — one fewer thing to optimize for someone already hitting ~1.6 g/kg/day. (Schoenfeld et al., 2013) (Schoenfeld et al., 2013)
Hold the null honestly — four author-stated limits keep this “wide window”, not “no window”.
- The window is wide, not absent. «if a peri-workout anabolic window of oppor- tunity does in fact exist, the window for protein consumption would appear to be greater than one-hour before and after a resistance training session.» The null is against a narrow (<=1 h) window, not against distributing protein sensibly across the day. (Schoenfeld et al., 2013)
- Causality is not established. The timing groups’ higher intake was observational within the pool: «Since causality cannot be directly drawn from our analysis … we must acknowledge the possibility that protein timing was in fact responsible for producing a positive effect and that the associated in- crease in protein intake is merely coincidental.» (Schoenfeld et al., 2013)
- Mostly untrained subjects. «statistical power was low because only 4 studies using trained subjects met inclusion» — the timing question matters most for the trained, and that is exactly the thin cell here. (Schoenfeld et al., 2013)
- Only 3 protein-matched studies. The cleanest test — hold total protein equal, vary only timing — had just 3 qualifying studies (2 of 3 showed no timing benefit): «The sum results of the matched-protein studies suggest that timing is superfluous provided adequate protein is ingested, although the small number of studies limits the ability to draw firm conclusions on the matter.» (Schoenfeld et al., 2013)
Independence + funding note. (inferred from Schoenfeld et al., 2013) Schoenfeld/Aragon/Krieger have zero author overlap with Morton (Phillips/McMaster) or Tagawa (Miyachi/Meiji), so the timing-null is a genuinely independent, new contribution. The total-protein- dominates convergence is directionally concordant with the dose sections above but only partially independent — this timing MA’s 2000-2013 RT pool shares some constituent trials with Morton’s supplementation pool, so it corroborates the direction that total protein is the operative variable, not an independent magnitude. The Dymatize Nutrition grant runs against the finding (the sponsor markets around-workout products), so the null despite that commercial pressure strengthens it — symmetric-standards caution on framing, as with Tagawa’s Meiji funding.
Wirth 2020 cashes that second SR+MA — a modern, larger, GRADE-rated timing replication (type-F refinement, NOT clean E). Wirth is exactly the second timing SR+MA the handle awaited: «Data from 65 studies with 2907 participants (1514 men and 1380 women, 13 unknown sex) were included in the review» (Wirth et al., 2020), searched to March 2019 and GRADE-rated — ~3x Schoenfeld’s 23-study pool, extending explicitly to older adults. It reaches the same verdict — «the timing of intake did not influence the results» (Wirth et al., 2020) — and does so on a hard-pooled LBM endpoint, not only a meta-regression effect size: protein raised LBM in every timing subgroup but not differently between them: «Protein supplementation improved LBM in all subgroups (after exercise MD: 0.51 kg; 95% CI: 0.13-0.89 kg …; before and after exercise MD: 0.70 kg; 95% CI: 0.28-1.13 kg …; and other timing not around exercise MD: 0.52 kg; 95% CI: 0.22-0.82 kg …), with no significant difference between the 3 timings (P = 0.76)» (Wirth et al., 2020). The distinct within-day distribution sub-question is null too — «In meta-analysis comparing even with skewed protein supplementation, no pattern presented a superior effect to the other (MD: −0.29 kg; 95% CI: −1.20 to 0.62 kg, P = 0.62 …)» (Wirth et al., 2020) — as are the two strength endpoints (handgrip «No difference (P = 0.90) was observed between protein supplementation at breakfast … and at breakfast plus other timing»; leg press «None of the timings of protein intake had an impact on leg press strength») (Wirth et al., 2020). Author decision rule (the source sentence is split by a mis-placed Figure 4 caption in the two-column PDF; quoted as its continuous prose): “the important aspect is to increase the overall protein intake as opposed to altering the distribution throughout the day” (Wirth et al., 2020).
Independence is author-clean but evidence-shared -> F, not E. Author-list diff first (the cheapest test): Wirth/Hillesheim/Brennan (UCD Dublin) have zero author overlap with Schoenfeld/Aragon/Krieger — the necessary condition for type-E. But it is not sufficient here, because Wirth cites Schoenfeld and frames its own result as corroboration: «Our results corroborate those from Schoenfeld et al. (81) who performed a meta-regression and concluded that the immediate protein supplementation, pre- or postexercise, did not improve hypertrophy and muscle strength in adults and older adults» (Wirth et al., 2020). A self-stated convergence over a citation-as-antecedent is F/attribution by the strict rule, never clean E; and both are SRs of the same peri-workout timing RCT literature over overlapping windows (Schoenfeld to 2013, within Wirth’s window to 2019), so shared constituent trials are expected, not independent replication. So Wirth composes as F (claim-refinement: a larger, newer, GRADE-rated, older-adult-inclusive base that bounds and extends the earlier null onto a hard LBM endpoint), not a second independent backing.
| Parameter | Schoenfeld 2013 | Wirth 2020 | Same quantity? |
|---|---|---|---|
| Timing question | peri-workout window (<=1 h pre/post) | after-ex / before+after / not-around-ex subgroups | ~yes — both ask does when matter |
| Timing verdict | benefit vanishes after controlling total protein | «no significant difference between the 3 timings (P = 0.76)» | YES — both null |
| Estimator | meta-regression, covariate control | GRADE indirect-comparison subgroup pooling | NO — different method |
| Pool | 23 studies, mostly untrained, to 2013 | 65 studies (26 in LBM MA), adults+older, to 2019 | NO — Wirth larger, newer |
| Author overlap | Schoenfeld/Aragon/Krieger | Wirth/Hillesheim/Brennan | zero |
| Citation link | — | cites Schoenfeld (81), «corroborate» | self-stated -> F, not E |
Funding closes the symmetric-standards loop the Schoenfeld note opened. Schoenfeld’s null came despite a Dymatize grant (a sponsor marketing around-workout products); Wirth reports «The authors report no conflicts of interest», on academic funding (Marie Curie / ERC / CAPES) that «had no role in the design, analysis, or writing of this article» (Wirth et al., 2020). So the timing-null now holds both against a commercial pressure (Schoenfeld) and free of one (Wirth) — the composite is more robust than either alone (the F payoff). The decision is unchanged and reinforced: once daily total protein is adequate, when it is eaten is not a lever worth optimizing.
Who it helps more, and who less
- Reduced with age: «reduced with increasing age (−0.01 kg (−0.02,–0.00), p=0.002)» — older adults gain less lean mass from the supplement (their training still works; the added protein does less).
- Greater in the already-trained: «more effective in resistance-trained individuals (0.75 kg (0.09, 1.40), p=0.03)» — the supplement’s edge grows once someone is past the untrained-beginner phase.
In HEALTHY older adults, nutrition adds nothing over training alone — except creatine
Morton’s age covariate (the supplement effect «reduced with increasing age», above) reaches its end-point in a meta-analysis built only on older adults. Choi 2021 (high-tier MA, 22 RCTs) pooled trials that compared resistance training + a nutritional intervention against resistance training alone in «healthy community-dwelling older adults», and found the added nutrition bought nothing on any outcome family: «The results of the meta-analysis showed no significant differences between groups in muscle mass, muscle strength, or physical functional performance.» (Choi et al., 2021)
- The one exception is creatine, on lean body mass. «In the subgroup analysis regarding the types of nutritional interventions, creatine showed significant effects on lean body mass (n = 4, MD 2.61, 95% CI 0.51 to 4.72).» Every other subgroup was null — hand grip «(χ2 = 0.12, p = .73)», appendicular skeletal muscle mass «(χ2 = 0.62, p = .43)», knee-extension «(χ2 = 4.89, p = .09)», chair-stand, timed-up-and-go. (Choi et al., 2021) Read the magnitude with its pool: n=4 trials, and the between-subgroup test was NOT significant, so creatine is a within-arm signal on a small set, not a demonstrated superiority over protein. It aligns with the separate creatine evidence base -> Creatine Supplementation, but does not by itself settle it.
- This EXTENDS Morton, it does not contradict it. Morton’s protein effect (+0.30 kg FFM) was measured in a mostly-young sample and decays with age by his own covariate; Choi is the older-adult end-state where the added-nutrition effect has decayed to null. Same added effect over training, different stratum along one continuous age gradient — a refinement, not a clash.
- Choi’s own reading is a ceiling in the nutrient-replete — and it PREDICTS the sarcopenic exception. He reads the null as headroom: «despite the nonsignificant results, nutritional interventions may still be beneficial for older adults who do not lack nutrients.» And he flags the stratum where it should bite differently — «protein supplements for sarcopenic older adults along with exercise showed a larger effect size than exercise alone… Individuals with existing nutritional deficiencies or poor muscle function might have been shown to respond better to accompanying nutritional supplements than to exercise alone.» (Choi et al., 2021) So the decision rule is baseline-status-dependent: for a healthy, replete older adult, training is the whole lever and added protein/co-supplements do little; the deficient or diagnosed-sarcopenic are the stratum where nutrition might still add (the sarcopenic case is taken up below).
The sarcopenic exception — where nutrition may still add (grip, not mass)
Song 2023 (moderate-tier MA [downgrade: Frontiers venue], 12 trials, «713 older adults diagnosed with sarcopenia») runs the same RT+nutrition-vs-RT-alone contrast Choi does, but in the stratum Choi flagged — the diagnosed-sarcopenic. It splits the outcome: «resistance training combined with additional nutritional supplementation, especially compound nutritional supplements that included protein and vitamin D, might further enhance grip strength rather than muscle mass in older adults with sarcopenia.» (Song et al., 2023)
Mass stays null even here; grip is a borderline, heavily-hedged positive:
| Quantity | Choi 2021 (high; HEALTHY older adults) | Song 2023 (moderate; SARCOPENIC older adults) | Same quantity? |
|---|---|---|---|
| Comparator | RT + nutrition vs RT alone | RT + nutrition vs RT alone | YES — identical contrast |
| Lean/muscle mass | «no significant differences between groups in muscle mass» (creatine exception MD 2.61) | «no significant difference in lean body mass … [SMD = 0.10, 95% CI (−0.14, 0.34), P = 0.422]» | YES — both NULL on mass |
| Grip strength | null «(χ2 = 0.12, p = .73)» | «[WMD = 1.87, 95% CI (0.01, 3.74), P = 0.049]» borderline positive | same quantity; result DIVERGES |
| Population | healthy, community-dwelling, mostly nutrient-replete | diagnosed sarcopenic, «in poorer health» | NO — the stratum differs |
(Choi et al., 2021; Song et al., 2023)
This is a DISTINCTION (baseline-status effect-modification), not a tension — and the sources agree it is. The mass-null is concordant across both strata, so there is no clash there. The grip result diverges, but only where the population differs, and both authors close the gap the same way: Choi predicted it (the deficient «respond better»), and Song attributes the divergence to baseline health — «In contrast, the present study centered on older individuals with sarcopenia and found different results. The reason for the difference may be that older patients with sarcopenia were in poorer health and could gain more benefits from nutritional supplements.» (Song et al., 2023) The not-joined check fires on scope (ii): the two are consistent once the stratum is matched. So the decision rule is one continuous story — replete healthy older adult: training is the whole lever; diagnosed-sarcopenic / deficient: a compound protein+vitamin-D supplement may add a little grip strength on top of training.
Hold the sarcopenic grip signal weakly — it does NOT lift the null to a benefit. Four discounts, all on
Song’s own numbers: the grip effect is borderline (P=0.049, lower CI 0.01, a whisker from null);
high-heterogeneity (I²=68.8%); it sits entirely in the compound protein+vitamin-D arm (protein/vitD-free
subgroup «[WMD = 0.35, 95% CI (−1.52, 2.22), P = 0.713]» — null), so the design cannot isolate which nutrient
does the work; and it is diagnostic-criterion-dependent — «the grip strength of the EWGSOP subgroup was
significantly improved [WMD = 5.41, 95% CI (3.74, 7.09), P = 0.000], and there was no difference in the AWGS
subgroup [WMD = 0.54, 95% CI (−1.34, 2.41), P = 0.574]», i.e. the pooled 1.87 averages subgroups that
disagree. (Song et al., 2023)
This is a between-MA contrast (two separate pools on two strata), not a within-study interaction test — a
route-(a)/(b) stratum candidate held directionally, confidence: low, and it does not overturn Choi’s
higher-tier null. The clinical read is unchanged in weight: even in sarcopenia, the supplement is a small
adjunct to training, and grip — not mass — is the surrogate it nudges.
The surrogate boundary — this is the mechanism, not the outcome
Muscle mass and strength are surrogates (Surrogate Outcomes), not patient-important endpoints. Morton (Morton et al., 2017) is the mechanism half of the resistance-training story — RT (plus adequate protein) builds muscle, RCT-grade — while Muscle-Strengthening Activity and Mortality is the outcome half — strength activity associates with lower mortality, observational-grade. Neither shows that protein supplementation reduces mortality; the composite is “RT builds muscle (proven) and strength associates with living longer (associational)”, with protein a small lever on the first half only. Muscle mass/strength do matter directly for function and sarcopenia, which are on the outcome menu — so the surrogate is not worthless, it is just not the mortality endpoint.
-
The mortality endpoint IS held now, but for a different decision — protein source, not the muscle amount. Dietary Protein and Mortality (Naghshi 2020, observational) finds plant-protein intake associates with lower all-cause/CVD mortality while animal protein is null and total amount does little. That is a substitution decision (shift sources plant-ward), orthogonal to this page’s ~1.6 g/kg quantity decision — you satisfy both by reaching the amount and biasing the sources. The two protein questions are distinct, not two facets of one.
-
Among the two surrogates, strength beats mass — so read this page’s strength effect (+2.49 kg 1RM) as the more outcome-relevant one. The sarcopenia case definition demotes muscle mass precisely because «strength is better than mass in predicting adverse outcomes» (Cruz-Jentoft et al., 2018), and EWGSOP2 names inadequate protein/energy intake as a secondary-sarcopenia cause — direct support for this page’s lever from the geriatric side -> Sarcopenia Definition and Diagnosis. Protein’s +0.30 kg lean-mass effect moves the confirmatory parameter; its +2.49 kg strength effect moves the primary one.
-
But the mass surrogate now has its own hard-outcome link — partly cashing the boundary. Low muscle mass independently predicts all-cause mortality (de Santana SR-MA: ASMI SMD −0.18, «cannot be completely explained by differences in muscle strength») (de Santana et al., 2021) -> Low Muscle Mass and Mortality. So the quantity this page’s protein lever moves is not only a proxy for strength — it tracks mortality on its own. The caveat is the same predictor-vs-target line: that low mass predicts death (observational) is not that raising mass via protein/RT reduces death (no RCT). The composite tightens to “RT+protein raises muscle mass (proven surrogate); low mass predicts mortality (observational); raising mass -> lower mortality (unproven)”.
The hormonal alternative lever — and why it does not displace this one
Testosterone therapy is the other anabolic lever aimed at the same surrogates (lean mass, strength), and the comparison sharpens why this page’s lever is preferred: TRT raises lean mass +1.6 to 3.6 kg but its strength/function gains are modest and, in the TRAVERSE RCT, it increased fractures (HR 1.43) and added AFib/AKI/PE while showing no mortality benefit -> Testosterone Adiposity and Muscle. Resistance training is the proven driver of strength here and carries none of those harms; in an obese man, losing fat also raises testosterone naturally (secondary hypogonadism is reversible). So the hormone is a narrow option for confirmed hypogonadism, not a substitute for the training. (No head-to-head RT-vs-TRT trial is held — this is an evidence-weighting judgment, not a trial result.)
Decision relevance
-
If you do resistance training, aim for roughly 1.6 g/kg/day of total protein — around there the lean-mass benefit appears to flatten (a mechanism-supported but statistically non-significant plateau, so read it as a region — see below), and more is unlikely to add lean mass. Most people who already eat ~1.4 g/kg are close, so the move is topping up, not loading.
- Treat 1.6 as a REGION, not a point, and for a hypertrophy objective bias UP (~1.8-2.2 g/kg/day). The break-point is 1.62 g/kg/day with a wide 95% CI of 1.03-2.20, so “1.6” is a central estimate, not a threshold — reading it as a precise target is false precision. The direction of the bias follows bias away from the costly tail -> The Estimate-to-Action Gap: for someone whose objective is muscle, the plateau’s location is uncertain and undershoot forfeits the objective (the costly tail is on the LOW side), while overshoot is low-harm for healthy kidneys (higher protein does not change GFR in adults without CKD, now source-backed -> Protein Intake and Kidney Function) — so aim the upper end of the CI plus a margin. The up-bias has a ceiling (~2.0-2.2 is well-supported; beyond it the evidence thins and the practical costs of displacement, satiety and expense rise), and it shifts higher in an energy deficit or older age (to protect lean mass). The one stratum flip is pre-existing renal disease (eGFR <30 / CKD), where high protein is a contraindication, not a cheap overshoot -> Protein Intake and Kidney Function. — the CI is Morton’s (extracted above); the region-not-point reading and the directional bias are the wiki’s, via The Estimate-to-Action Gap.
- The target is protein quantity, and it silently assumes quality. Morton’s trials were mostly high-quality supplemental protein (whey). Hitting ~1.6 g/kg from low-DIAAS plant sources (peas 64, wheat 40 vs milk 122) delivers fewer digestible indispensable amino acids per gram, so it needs more (Food and Agriculture Organization of the United Nations, 2013) grams — hence more food mass — or deliberate complementation to be equivalent -> Protein Quality and the DIAAS Score. This is a source caveat on the number, not a change to it.
-
The training is the lever; protein is the adjunct. Do not expect protein alone (without the training) to build muscle — every effect here is during resistance training. How to program that training — load for strength, volume for hypertrophy, the minimal effective dose, and the same surrogate-only caveat — is Resistance Training Prescription - Load Sets and Frequency (Currier NMA). Note the two are not independent evidence: Currier and Morton share Stuart Phillips’ McMaster lab, so treat them as one lineage’s two levers on the same adaptation, not as corroborating each other. (inferred from Morton et al., 2017)
-
Creatine is the second ergogenic adjunct, through a different pathway — additive, not redundant. Where protein feeds muscle-protein synthesis, creatine buffers ATP resynthesis (the phosphocreatine energy system), so it augments the same resistance-training surrogates (strength, lean mass) by a distinct mechanism and stacks with protein rather than substituting for it — including in older adults, where creatine + RT beat RT alone on muscle mass, strength and functional capacity -> Creatine Supplementation. Same boundary as this page: RT is the driver, the supplement is a small adjunct on surrogates, and creatine monohydrate (3-5 g/day) is the cheap, best-evidenced form.
-
Age and training status modulate it, not whether to do it: a 46-year-old sits near the age boundary (Morton splits at 45) where the supplement’s edge is modestly reduced but the training’s is not.
-
The over-65 stratum has its own target — a different decision, not a contradiction. This page’s ~1.6 g/kg/d is the hypertrophy-during-RT number for healthy adults; the older-adult maintenance target is 1.0-1.2 g/kg/d (1.2-1.5 in illness), set to overcome Anabolic Resistance with a raised per-meal threshold (~25-30 g protein / 2.5-2.8 g leucine per meal) -> Protein Intake for Older Adults. Same unit (g/kg/d), different objective and population — a ladder by goal, so an older adult training for hypertrophy aims high while one merely maintaining aims 1.0-1.2. The per-meal figures are a different denominator again (the commensurability table lives on that page). — the distinction and the ladder reading are the wiki’s; the 1.6 is Morton (above), the 1.0-1.2 is Bauer (extracted on the linked page).
- The young 1.62 does NOT transport UP to older adults — it is a weak, young-population anchor. The break-point is statistically non-significant even in the largely-young pooled sample (p=0.079, above), and the supplement’s own effect is «reduced with increasing age (−0.01 kg (−0.02,–0.00), p=0.002)» (Morton et al., 2017) — the number was measured where the effect is strongest and decays with age. So for an older adult ~1.6 is a floor, not a ceiling: Anabolic Resistance means old muscle needs at least as much, plausibly more per anabolic stimulus, so a non-significant young knee is no reason to aim low in the old. The older-stratum target proper comes from a different evidence base (PROT-AGE 1.0-1.2 g/kg maintenance, 1.2-1.5 in illness, plus a raised ~25-30 g per-meal threshold to clear anabolic resistance) -> Protein Intake for Older Adults. The honest object is a direction + a floor, not a point optimum -> The Underivable Optimum.
-
From what age is resistance training essential rather than nice-to-have? is the wrong axis — it is a lifelong lever with no on/off age.** Its mortality/CVD benefit runs across adulthood at a small dose (Muscle-Strengthening Activity and Mortality); what rises with age is its priority, because muscle and strength decline from midlife and RT is the primary lever preserving function, balance and independence — outcomes that themselves become critical with age (Lifetime Benefit - The Frame for Younger Adults). So “essential” tracks falling muscle/function status, not a birthday: RT’s rank rises continuously with sarcopenia risk rather than switching on at an age.
-
During weight loss, adequate protein + resistance training is how you keep lean mass while losing fat (the fat-mass fell and lean mass rose here) — the practical reason it belongs in a weight-loss program, not just a strength one.
- The deficit RAISES the target — a different-stratum decision, now held on its own page. In an energy deficit the requirement rises from this page’s ~1.6 g/kgBM (energy balance) to ~1.9 g/kgBM / ~2.5 g/kgFFM (the ES-zero crossing) for FFM retention, up to 3.2/4.2 for gain -> Protein Intake During Energy Restriction (Refalo 2025 meta-regression, nonobese RT adults; authors’ own word EXPLORATORY, small effect vs the RT stimulus). Same unit, different energy state and construct (a plateau of the build effect here vs a loss-to-hold crossing there) — a stratum ladder, not a contradiction. — the numbers are Refalo’s (extracted on the linked page); the ladder reading is the wiki’s.
- The curve SHAPE corroborates independently — the knee stays unestablished across two labs and two
populations. Refalo (Deakin/AUT; energy deficit) found a linear model beat quadratic/cubic (97%
prob monotone-positive), matching Morton’s own non-significant 1.62 knee and the significantly
better linear fit on Morton’s baseline-intake axis. Reached by different data, different labs, no
shared author -> a genuine independent-shape convergence (
[E-independent]): the burden-of-proof dose-response rule holds here too, no located knee -> The Underivable Optimum. But the per-FFM-denominator point is NOT independent — Refalo’s per-FFM finding tests Helms 2014’s own speculation, and Helms co-authors Refalo, so that leg is one lineage, not corroboration. (inferred from Morton et al., 2017; Refalo et al., 2025) - A worked failure of exactly this: in the TREAT time-restricted-eating trial, a self-selected 16:8 window (no protein guidance) lost weight that was «approximately 65% … lean mass» vs a normal 20-30% — the authors suggest (protein intake was not measured) it may reflect a short ad-libitum window cutting protein intake (protein is eaten mostly at meals), a «caution for patient populations at risk for sarcopenia» -> Time-Restricted Eating. The mitigation is this page’s number: keep meals/protein up inside the window. (Lowe et al., 2020)
Limits
- Surrogates, not outcomes — no mortality/function endpoint; RCT-grade for muscle/strength only.
- Supplementation, not total dietary protein per se — the trials add protein on top of a diet; whole- food vs supplement was mostly supplement (largely whey), so this is not a food-quality (DIAAS) claim -> Protein Quality and the DIAAS Score holds the source-quality dimension.
- The plateau CI is wide (1.03-2.20); the fibre-CSA effect is fragile; healthy adults only.
- The number is g/kg TOTAL body mass, over a narrow sampled intake band (~1.0-2.2 g/kg-total), in a sample with no reported baseline body composition. The obese are off-support (no obese data), so neither the target nor its denominator transports to them — an insufficient-evidence gap, not a lower-target finding (see How much protein).
- One meta-analysis; the moderator model overall explained little variance, so age/training effects are subgroup signals, not a full explanation.
(inferred from Morton et al., 2017)