Working research document, copied from the author’s notes on 2026-09-06. Rough, long, and unedited apart from removing personal details. The finding page summarizes it.

Nutrient Thresholds & Dose-Response — Machine-Encodable Evidence Base

Purpose: score ~800 real meals with per-meal nutrition data. Calibration target: a healthy general-population adult. [personal profile and health goals removed] Date: 2026-09-05

Evidence-handling rules used throughout

Every quantitative claim carries design + n + effect size. Designs are labelled RCT / prospective cohort / meta-analysis / mechanistic.
Claims retrieved and verified against a primary record during research are unmarked. Anything not verifiable is tagged UNVERIFIED RECALL or LOW CONFIDENCE.
Numbers I computed myself from published inputs are tagged (derived).
Scoring bands are engineering choices calibrated to the evidence, not retrieved findings. Where a band edge is anchored to a specific paper or regulation, it says so. Where it is a judgment call, it says that too.
“Statistically significant” is never treated as “worth optimizing.” Effect sizes are stated in absolute terms wherever available.

0. BOTTOM LINE — LEVERS RANKED BY EXPECTED MAGNITUDE FOR THIS PERSON

Ranked by expected effect on the adult’s stated goals (visceral fat, VO2max, joints), not by general population importance.

#	Lever	Expected magnitude	Why	Evidence tier
1	Food-only energy density (kcal/g)	Largest. 25–30% ED reduction → 20–30% ad-libitum intake reduction at constant food weight	People eat a roughly constant weight of food. This is the mechanism behind both UPF trials, and it produces the energy deficit that is the dominant VAT lever	RCT (Rolls-lineage ED manipulations; Hall n=20; Dicken n=50)
2	Total energy deficit itself	Largest, and ED is how you get it. VAT is preferentially lost at modest weight loss; ~−22.5% VAT over 18 mo	Chaston & Dixon: the method of weight loss did not matter — only the amount. This is the single most-buried finding in the VAT literature	SR of 61 imaging studies, 98 time points
3	Eliminate liquid sugar / caloric beverages	Large and depot-specific. At matched weight gain, fructose beverages loaded VAT; glucose beverages loaded SAT. Men +18.1% intra-abdominal fat vs women −0.6%	The only clean depot-specific dietary finding in the literature, and the sex interaction favours exactly the adult’s demographic. Plus: liquid energy compensation is −17% vs +118% for solid	RCT (Stanhope n=32, CT; Maersk n=47, MRS; DiMeglio n=15)
4	Protein ≥35–45 g per main meal	Moderate-large for body composition quality. 2.4 vs 1.2 g/kg/d in a 40% deficit: LBM +1.2 kg vs +0.1 kg, FM −4.8 vs −3.5 kg, 4 weeks	Doesn’t burn fat faster; makes the deficit lean-sparing, which is the entire difference between “lost 8 kg” and “lost 8 kg of the right thing”	RCT (Longland n=40); MA (Wycherley 24 trials n=1,063)
5	Adequate carbohydrate around VO2max sessions	*Moderate — and it is a protective* lever, not an additive one. LCHF raised VO2peak identically (+2.55–5.20%) but negated** the 5–7% performance gain	The failure mode is dieting into under-fuelled interval sessions. “Train low” is a null meta-analysis in trained athletes (SMD 0.17, p=0.29)	RCT (Burke n=29 elite); MA (Gejl 9 studies)
6	Fiber, as a whole-food marker	Moderate — but mostly as a proxy. Cohorts say RR 0.85 all-cause; high-GRADE RCTs of the same exposure say −0.37 kg and −1.27 mmHg	Chase the foods, not the grams. A chicory-fortified bar hits the number and buys nothing	Cohort (185 studies) vs RCT (58 trials) — the gap is the finding
7	Added sugar (non-liquid), fructose <50 g/d	Moderate. Below ~50 g/d fructose the evidence is neutral-to-beneficial; at 80 g/d hepatic DNL roughly doubles isocalorically	The adult’s realistic added-sugar ceiling (~36 g/d) puts the adult at ~18 g fructose — comfortably under every threshold	RCT (Geidl-Flueck n=94; Schwarz n=8)
8	Potassium up (not sodium down)	Small-moderate, and the ratio beats either alone. ~75% of US adult men are below the 3,400 mg AI	Hitting Na:K ≤0.6 mass is far more sensitive to raising K than cutting Na. In model selection with gold-standard 24h urines, the ratio had the lowest BIC — sodium alone was p=0.38, ratio p=0.04	Prospective (Cook n=2,974; Yang n=12,267; Ma n=10,709)
9	Sodium down	Small for the adult. −0.4 to −1.4 mmHg SBP per 1,000 mg/d in normotensives. No RCT has ever shown a mortality benefit from sodium reduction alone	Real but tiny at the individual level. Its public-health value comes from applying to a whole distribution, not to one normotensive adult	MA (Filippini 85 RCTs; Graudal Cochrane)
10	SFA → unsaturated substitution	Near-zero for the adult’s stated goals; genuinely worth doing anyway. 5% energy SFA→PUFA = HR 0.75 for CHD	This is a 50-year cumulative-LDL-exposure play, and in early adulthood the duration term is the entire reason to care. It does nothing for visceral fat, VO2max, or joints this year	Cohort substitution (n=127,536); RCT MA (8 trials, n=13,614)
11	Omega-3 EPA+DHA ~1–2 g/d	Small. TG −32.6 mg/dL pooled; CRP −0.34 mg/L (trivial at the adult’s baseline); CV RR ~0.92–0.95 at 1 g/d	2 servings oily fish/week. Do not megadose: 4 g/d has an AF signal (HR 1.49) with no indication in a lean normotensive	MA (171 RCTs for lipids; 13 RCTs n=127,477 for events)
12	Micronutrient density (Mg, K, Ca, Fe, Zn, C, A, folate)	Small, and mostly a byproduct of #1. The ED fix and the micronutrient fix are the same intervention	Vitamin D and ferritin are blood tests, not meal scores. Choline and iodine are effectively unmeasurable from meal data	NHANES usual-intake + repletion RCTs
13	Meal timing / protein distribution	Second-order at best; possibly zero. The flagship distribution RCT was P=0.06 with n=26	TRE is null beyond calorie control in both large trials (n=139 12-mo; n=116 12-wk). Distribution “cannot be disentangled from protein quantity”	RCT (Liu n=139; Lowe n=116; Yasuda n=26)

The three special questions, answered up front

Visceral fat. The dominant lever is total energy deficit, and how you create it barely matters (Chaston & Dixon, 61 imaging studies: “the method of weight loss was not an influence”). Ranked after that: aerobic exercise independent of weight loss (−6.1% VAT at zero weight change, n=4,815) — note resistance training is null for VAT (ES 0.09, 95% CI −0.17 to 0.36, 35 studies), which matters given the adult’s joint limits; then eliminating liquid fructose, the only clean depot-specific dietary finding; then modest carb restriction (one 8-week eucaloric trial, n=69, −11% vs −1% IAAT — real but contradicted by the weight of evidence); then soluble fiber (observational only). Protein does not target visceral fat specifically in an adequately-fed young man — the one clean VAT RCT was 0.8 vs 1.3 g/kg in men ≥65, i.e. deficient-vs-adequate, a part of the curve the adult isn’t on.

Joint/back pain. Be skeptical, and the honest answer is uncomfortable: diet composition is not an analgesic. The two interventions with real effect sizes are exercise therapy (MD −15.2 points on 0–100, 249 trials) and fat loss (1 kg lost ≈ 4 kg less knee compressive load per step; IDEA n=454). Below those: omega-3 has genuine but modest evidence in rheumatoid arthritis only (SMD −0.26 to −0.43, and physician-assessed outcomes were null while patient-reported ones moved — the signature of incomplete blinding), and is flatly null in osteoarthritis (Hill n=202: a 10× dose was, if anything, worse; Laslett JAMA 2024 n=262: null in MRI-confirmed inflamed knees). Zero RCTs of omega-3 for non-specific low back pain were found. Glucosamine/chondroitin is definitively negative (GAIT n=1,583; Wandel n=3,803 — no estimate crossed the MCID). Vitamin D in replete people is null (CLBP meta SMD −0.130, CI touching zero). The Dietary Inflammatory Index is circular by construction and failed to associate with knee OA symptoms in a trial where the DII’s own developer was an author. For this person: load management, strength work, and total adiposity — not meal composition.

VO2max. Only three levers have real evidence, and one of them is a blood test. (1) Adequate carbohydrate — protective rather than additive; under-fuelling costs adaptation. (2) Iron, if deficient — g = 0.610 (0.399–0.821) on VO2max, one of the largest effects in this entire document — but prior probability of deficiency in a young male is ~1%, so this is one ferritin test, then stop thinking about it. Do not supplement blind; iron overload is a real harm in males. (3) Caffeine 3–6 mg/kg — g=0.392 for TTE; it raises training quality, not the ceiling. Beetroot/nitrate does not raise VO2max — the GXT meta-analytic effect is non-significant (ES 0.25, −0.06 to 0.56) and it is cleanly null in trained athletes (n=8 elite 1500m runners at 19.5 mmol: nothing; n=9 trained cyclists: nothing). Actively avoid high-dose supplemental vitamin C (1,000 mg) / E (235 mg–400 IU) during training blocks — VO2max was unaffected in the trial (+8% both arms, n=54) but mitochondrial signalling was blunted (COX4, PGC-1α) and training-induced insulin-sensitivity gains were abolished (n=39). No upside to offset it.

1. SODIUM

1.1 Dose-response shape: approximately LINEAR, no threshold, and it gets STEEPER at low intake

Source	Design	n	SBP effect	Per 1,000 mg Na (derived)
Filippini 2021, Circulation 143:1542	Dose-response MA, 1-stage cubic spline	85 RCTs, >10,000	−5.56 mmHg (−6.59 to −4.52) per 100 mmol/d	−2.42
— hypertensive stratum			−6.50 (−7.79 to −5.22)	−2.83
— normotensive stratum			−2.30 (−3.27 to −1.33); DBP −0.80 (−1.89 to +0.29, NS)	−1.00
The adult/Li/MacGregor 2013, BMJ 346:f1325	MA, RCTs ≥4 wk	34 trials, 3,230	meta-regression −5.8 per 100 mmol	−2.52
— normotensive			−2.42 (−3.56 to −1.29)	−1.40
Graudal 2020, Cochrane CD004022.pub5	MA incl. short trials	5,982 (95 trials), white normotensive	−1.14 (−1.65 to −0.63); DBP +0.01	−0.36
— white hypertensive		3,998 (88 trials)	−5.71 (−6.67 to −4.74)	−1.80
INTERSALT (Dyer 1994, regression-dilution corrected)	Cross-sectional, 52 centres	10,079	−3.1 per 100 mmol	−1.35

The honest normotensive number: −0.4 to −1.4 mmHg SBP per 1,000 mg/day. For the adult going from typical US male intake (~4,000 mg) to 2,500 mg: −0.5 to −2.1 mmHg SBP. Real, but small at an individual level.

DASH-Sodium (Sacks 2001, NEJM 344:3) — the definitive shape experiment. n=412; parallel control vs DASH diet, with a crossover through high/intermediate/low sodium, 30 days each, controlled feeding. Achieved 24h Na: 143 / 107 / 66 mmol/d = 3,290 / 2,460 / 1,520 mg (derived).

Control diet: high→intermediate −2.1 mmHg; intermediate→low −4.6 mmHg.
DASH diet: high→intermediate −1.3; intermediate→low −1.7.
Combined DASH+low vs control+high: −7.1 mmHg (non-hypertensives), −11.5 mmHg (hypertensives).

Two structurally important readings (slopes derived):

The curve gets STEEPER as intake falls — −2.54 mmHg/1,000 mg in the upper range vs −4.88 mmHg/1,000 mg in the 2,460→1,520 mg range. This directly contradicts “no benefit below 2,300.”
Potassium blunts sodium. On the high-potassium DASH background the low-range slope collapses to −1.80 mmHg/1,000 mg — ~2.7× flatter. RCT-grade evidence that the ratio is more mechanistically apt than sodium alone.

Salt sensitivity is not a rare subtype. Gupta 2023, JAMA 330:2258, prospectively allocated crossover, n=213, 24h ambulatory BP: median within-individual MAP change 4 mmHg (IQR 0–8); 73.4% declined on low sodium; 46% met the conventional ≥5 mmHg “salt sensitive” threshold; effect did not differ by hypertension status. (Caveat: ages 50–75, 64% Black — not a adult.)

1.2 The J-curve controversy — resolved, and it’s a measurement artifact

Pro-J evidence (all spot-urine + Kawasaki formula):

PURE individual-level (O’Donnell 2014, NEJM 371:612): n=101,945, 17 countries, 3.7 y follow-up, 3,317 events. vs 4.00–5.99 g/d ref: ≥7 g/d OR 1.15 (1.02–1.30); <3 g/d OR 1.27 (1.12–1.44).
PURE community-level (Mente 2018, Lancet 392:496): 255 communities / 82,544 for events, median 8.1 y. Nonlinearity p=0.043. Stroke association strong in China (mean 5.58 g/d) and inverse elsewhere, heterogeneity p<0.0001.
ONTARGET/TRANSCEND (O’Donnell 2011, JAMA 306:2229): n=28,880 with established CVD/diabetes. CV death HR 1.66 (1.31–2.10) at >8 g/d and HR 1.37 (1.09–1.73) at <2 g/d.
UCC-SMART (Groenland 2022, PLoS One 17:e0265429): n=7,561, J-shaped, nadir 4.59 g/d. Also found higher potassium excretion associated with higher mortality (HR 1.25/g) — implausible against all other literature, and a red flag impeaching the Kawasaki method in sick cohorts.

The decisive rebuttal — The adult, Campbell, Ma, MacGregor, Cogswell, Cook 2018, Int J Epidemiol 47:1784. This is the most important paper in the debate because it runs both methods on the same people: TOHP participants, n=2,974, median 24 y follow-up, 272 deaths, sodium assessed 4 ways.

Gold-standard mean intake 3,769 ± 1,282 mg/d. Kawasaki over-estimated by 1,297 mg/d (95% CI 1,267–1,326) with systematic bias: over-estimating at low intakes, under-estimating at high intakes — exactly the differential misclassification that manufactures a J.
Result: measured 24h sodium → LINEAR with mortality. Kawasaki-estimated sodium → J-shaped. Same cohort, same outcomes, method determines shape.

Regression dilution compounds it. INTERSALT reliability coefficients from 805 repeat collections (Dyer 1994): 24h Na 0.37–0.40; K 0.47–0.52; Na:K 0.32–0.36. Even a single gold-standard 24h urine is a noisy estimate of habitual intake; INTERSALT’s corrected slope was 44–50% larger than uncorrected.

But the pro-reduction camp is also over-read. Graudal’s counter-signals are real and unrebutted: sodium reduction raises renin +1.56 ng/mL/h, aldosterone +104 pg/mL, noradrenaline +62.3 pg/mL, cholesterol +5.19 mg/dL, triglycerides +7.10 mg/dL — and the adult notes these were more consistent than the BP effect in normotensives. Whether they offset the BP benefit is genuinely unknown. And randomized sodium reduction has never significantly reduced mortality: Cook 2016 (JACC 68:1609) randomized 24-year mortality HR 0.85 (0.66–1.09), p=0.19; Adler 2014 Cochrane (8 RCTs, n=7,284) all-cause RR 1.00 (0.86–1.15), and its positive CV-event pooled estimate was “driven by one trial among retirement home residents.”

SSaSS is the one adequately-powered hard-outcome RCT (Neal 2021, NEJM 385:1067): cluster-randomized, 600 villages, n=20,995, 88.4% hypertensive, mean age 65.4, 4.74 y. Salt substitute 75% NaCl / 25% KCl. Stroke RR 0.86 (0.77–0.96); major CV events RR 0.87 (0.80–0.94); all-cause death RR 0.88 (0.82–0.95); serious hyperkalemia RR 1.04 (0.80–1.37) — no signal. Greenwood 2024 Ann Intern Med: 16 RCTs, all-cause RR 0.88 (0.82–0.93), low certainty, “the evidence base is dominated by a single, large RCT,” 7 of 8 from China/Taiwan.

Verdict. (a) The J-curve is most likely a measurement artifact; He 2018 is close to dispositive. (b) The BP relationship is linear with no threshold down to at least ~1,500 mg. (c) But hard-outcome benefit is proven only via salt substitution in high-risk elderly hypertensives, and SSaSS confounds Na-lowering with K-raising — you cannot separate the arms. Note that what SSaSS actually tested is shifting the Na:K ratio, which is the strongest single argument for scoring the ratio. (d) For a normotensive adult, the evidence of benefit below ~2,300 mg is extrapolation from BP slopes, not outcome data. Expected SBP delta: ~1–2 mmHg.

1.3 Where 2,300 and 1,500 actually come from

2,300 mg: IOM 2005 set it as the Tolerable Upper Intake Level — a safety ceiling, not a target. NASEM 2019 removed the UL and re-issued 2,300 as a CDRR (Chronic Disease Risk Reduction intake, the first ever assigned to any nutrient). Numerically identical, conceptually different, and phrased as a reduction directive — “reduce sodium intakes if above 2,300 mg per day” — not a target to hit.
1,500 mg — this is the number most people get wrong. It is the Adequate Intake, and per the IOM 2005 process it was NOT a chronic-disease target. It was set on adequacy grounds, on two criteria: (i) the sodium level at which a diet can still meet all other nutrient recommendations, and (ii) coverage of sweat losses in unacclimatized individuals newly exposed to heat or physical activity. This directly answers the active adult question: the 1,500 mg AI already has moderate sweat losses baked in.
NASEM 2019 explicitly declined to endorse 1,500 mg as a chronic-disease target, citing “limited evidence on sodium intakes below 1,500 mg per day for adults.” It also withdrew the UL for sodium and potassium (“insufficient evidence of risk of excess… within healthy populations”).
AHA’s <1,500 mg rests on a 2011 presidential advisory restated in Whelton 2012, Circulation 126:2880, which argues the observational J-curve studies have flaws that “limit their usefulness in setting, much less reversing, dietary recommendations.”

Is 1,500 defensible? Split verdict. For: dose-response is linear with no observed threshold; DASH-Sodium’s steepest slope was in the 2,460→1,520 range; DASH-Sodium proves it’s achievable under feeding conditions. Against: NASEM, reviewing the same evidence, explicitly declined; essentially no free-living population achieves it (US adult men averaged 3,996 mg/d, NHANES 2015–2016, CDC MMWR 2021;70:1478); no hard-outcome RCT has tested it; Graudal’s counter-signals are largest under aggressive restriction. Working position: 2,300 mg is the defensible anchor; 1,500 is an AI floor, not a goal.

1.4 Active adult sweat losses — real, but they do not justify an exemption

Baker 2016, J Sports Sci 34:358 — normative dataset, n=506 athletes: forearm sweat [Na⁺] 43.6 ± 18.2 mmol/L; predicted whole-body [Na⁺] 35.9 ± 10.4 mmol/L = 826 ± 239 mg/L (derived); whole-body sweating rate 1.21 ± 0.68 L/h. Sweat [K⁺] is 2–8 mmol/L and does not rise with intensity.

Closest real data to indoor recreational sport — Baker 2022, IJSNEM (PMID 35477899): n=53 elite male basketball players, coach-led practice 98±30 min at 21.0±1.2 °C — whole-body sweating rate 0.97 ± 0.41 L/h. Contrast Zetou 2008 (n=47 beach recreational sport at 33.6 °C): ~2.0 L/h (derived). Environment dominates. No indoor-recreational sport sweat sodium data exists — any such number is extrapolated from basketball.

Estimated 2-h indoor recreational sport losses (all derived): low ~500 mg; typical ~1,650 mg (1.0 L/h × 36 mmol/L); high “salty sweater” in a warm gym ~4,900 mg. Potassium: ~390 mg — trivial against a 3,400 mg AI.

Does this justify higher intake? Evidence quality: LOW.

The AI already accounts for it (see §1.3). The CDRR sits 800 mg above the AI.
Acclimation is the body’s answer — sweat [Na⁺] falls with heat acclimation via aldosterone; sodium restriction raises aldosterone by 104 pg/mL, which is exactly the conservation machinery.
Two null supplementation RCTs. Cosgrove & Black 2013, JISSN 10:30: n=9 trained cyclists, 72 km TT ~3 h, 700 mg/h salt vs placebo — no performance effect (171 vs 172 min, p=0.46). Earhart 2015, J Sports Sci Med 14:172: n=11, 1,800 mg sodium vs placebo — no difference in TTE (6.88 vs 6.96 min, p=0.919), sweat rate, cardiovascular drift, or RPE.
Exercise-associated hyponatremia is an overdrinking problem, not a sodium-intake problem (Hoffman & Myers 2015: symptomatic EAH at serum Na 122 mEq/L in a runner who drank 9.2–10.6 L and took >6,500 mg sodium).
ACSM’s 1996 stand explicitly noted “there is little physiological basis for the presence of sodium in an oral rehydration solution… as long as sodium is sufficiently available from the previous meal.”

Practical: on days with ≥90 min hard indoor play, permit +500 to +800 mg to the daily budget — roughly half the estimated typical sweat loss, mirroring the ACSM convention of deliberately incomplete replacement. Flag this as physiologically reasonable but NOT outcome-validated. No trial supports it.

1.5 NORMALIZER: absolute mg is primary; density is a secondary override

“1 mg sodium per kcal” is arithmetic folklore. It comes from 2,300 ÷ 2,300. No primary IOM/NASEM/AHA/FDA document states it as a recommendation. Treat as a mnemonic with no institutional standing.

What IS institutionally grounded:

USDA/FSRG uses <1,150 mg per 1,000 kcal as “consistent with the DGA recommendation of <2,300 mg/day” (i.e. 2,300 on a 2,000 kcal reference diet).
Actual US sodium density: 1,631 mg/1,000 kcal (WWEIA/NHANES 2007-2008, n=8,529). USDA notes the male/female sodium difference disappears once adjusted for calories — which is the strongest argument for energy-adjustment as a food-choice metric.
FDA regulatory anchors (21 CFR 101.61; 21 CFR 101.65(d), Dec 2024 “healthy” final rule), Na DV = 2,300 mg: “low sodium” ≤140 mg per RACC; “healthy” main dish ≤20% DV = 460 mg; “healthy” meal product ≤30% DV = 690 mg. The meal-product number is the only regulatory figure that is explicitly per meal.

Why absolute wins as primary:

Physiology is absolute. BP responds to sodium load; ECF expansion is a function of total mmol. Every effect size in §1.1 is per mmol/day, never per kcal. 24h urinary excretion ≈ intake is an absolute quantity.
All guidance is absolute — CDRR, AI, DV, WHO.
Density penalizes low-calorie healthy foods and rewards calorie-dense ones. 25 kcal of steamed vegetables with 90 mg Na scores 3.6 mg/kcal (catastrophic by density) while contributing 90 mg. A 900-kcal fried entrée with 1,300 mg scores 1.44 mg/kcal — “only slightly above average” — while eating 57% of the CDRR.
Density rewards eating more. At [calorie target removed], the DGA-consistent 1,150 mg/1,000 kcal yields 3,278 mg/day (derived) — 43% above the CDRR. Density silently converts “I train hard so I eat more” into “I get more sodium,” which is exactly the inference the evidence does not support.

Why density still earns a slot: it strips out portion size, making it the right metric for comparing two foods, and it catches the failure mode absolute mg misses — a large, calorie-appropriate meal that is nonetheless salt-drenched.

Verdict: absolute mg is the score; density is an override flag. Do NOT scale the daily sodium budget with the adult’s [calorie target removed] — the budget stays 2,300 mg; the density band tightens instead. (Derived personalized figure: 2,300 / [calorie target removed] = 807 mg per 1,000 kcal — meaningfully stricter than the generic 1,150.)

1.6 Sodium vs the adult’s actual goals — skeptical assessment

Visceral fat: NO evidence sodium reduction reduces it. Cross-sectional associations exist and are large (Santalahti 2026: OR 4.30–6.05 highest vs lowest urine-sodium quartile) but are implausibly large for a nutrient effect and almost certainly reflect residual confounding by total energy and UPF intake. Tellingly, in Lee 2023’s review of reviews (39 observational studies, 35 cross-sectional), the BMI effect shrinks monotonically as measurement gets less biased by total intake (2.27 kg/m² with 24h urine → 0.85 with dietary assessment) — a classic confounding signature. If the adult cuts sodium by cutting processed food, visceral fat may fall — because of the food, not the sodium.
Joint pain: mechanistically interesting, clinically unevidenced for the adult. Kleinewietfeld 2013, Nature 496:518 shows high NaCl boosts pathogenic TH17 induction via SGK1 — in mice and in vitro. Downstream rodent work extends it to collagen-induced arthritis. This pathway is about autoimmune disease, not mechanical joint/back pain. No human evidence connects sodium intake to non-autoimmune joint pain. Do not score sodium on a joint-pain rationale.
VO2max: no benefit from more, no demonstrated harm from moderate restriction. Two null supplementation RCTs (n=9, n=11). No RCT with VO2max as an endpoint for sodium restriction was located. Seals 2001 (n=35, 3 months) found restriction to <100 mmol/d produced no change in plasma volume. At 2,300–2,800 mg/day expect zero performance effect in either direction. The performance-relevant variable is acute fluid/electrolyte replacement around sessions — a hydration protocol question, not a habitual-intake question.

2. POTASSIUM AND THE Na:K RATIO

2.1 Potassium → BP: real, but concentrated in hypertensives

Aburto 2013, BMJ 346:f1378 (WHO-commissioned):

RCTs: 22 trials, n=1,606. SBP −3.49 mmHg (1.82–5.15), DBP −1.96 (0.86–3.06).
Critical subgroup finding: “an effect seen in people with hypertension but not in those without hypertension.”
At 90–120 mmol/d: SBP −7.16 (1.91–12.41) — but “without any dose response.” This is precisely what NASEM later cited when declining to set a potassium CDRR.
Cohorts: 11 studies, n=127,038. Stroke RR 0.76 (0.66–0.89). CVD 0.88 (0.70–1.11) NS. CHD 0.96 (0.78–1.19) NS. The hard-outcome signal is stroke-specific.

Filippini 2020, JAHA 9:e015719 — dose-response, 32 RCTs ≥4 wk, cubic spline: U-shaped, not linear. BP-lowering weakens above a 30 mmol/d differential and BP rises above ~80 mmol/d. Effects stronger in hypertensives and at higher sodium intakes — potassium’s benefit is conditional on the sodium context, the pharmacological analogue of the DASH-Sodium interaction.

Recommendations & reality. WHO 2012: ≥3,510 mg/d, conditional recommendation. NASEM 2019 AI: 3,400 mg/d men (down from the 2005 AI of 4,700), no CDRR, no UL. US mean intake (all ages ≥2 y) is 2,496 mg/d (WWEIA 2017-2018); among males 20+, only ~25% exceed 3,400 mg.

For the adult: normotensive, so the RCT-demonstrated BP benefit is weakest exactly in the adult’s subgroup. Being honest: potassium’s per se BP effect for the adult is likely near zero. The case rests on the ratio (§2.2) and on potassium being a whole-food marker (§2.3).

2.2 Is the RATIO better than sodium alone? Yes — and this is the strongest single finding in the sodium/potassium literature

Study	Design	n	Sodium alone	Potassium alone	Na:K ratio
Cook 2009, Arch Intern Med 169:32 (TOHP follow-up)	Prospective; mean of 3–7 measured 24h urines	2,974 enrolled / 193 CVD events, 10–15 y	Trend NS (p=0.38); RR 1.42 (0.99–2.04) per 100 mmol	Trend NS (p=0.08)	Trend p=0.04; RR 1.24 (1.05–1.46) per unit, p=0.01. The ratio model had the LOWEST BIC — formally the best fit.
Yang 2011, Arch Intern Med 171:1183 (NHANES III LMF)	Prospective cohort	12,267; 2,270 deaths, mean 14.8 y	All-cause HR 1.20 (1.03–1.41) per 1,000 mg	All-cause HR 0.80 (0.67–0.94) per 1,000 mg	Q4 vs Q1: all-cause HR 1.46 (1.27–1.67); CVD 1.46 (1.11–1.92); IHD 2.15 (1.48–3.12). No effect modification by sex, race, BMI, HTN status, education, or physical activity
Ma 2022, NEJM 386:252	Pooled IPD, 6 cohorts, ≥two 24h urines each	10,709; 571 CV events, median 8.8 y	Q4 vs Q1 HR 1.60 (1.19–2.14)	Q4 vs Q1 HR 0.69 (0.51–0.91)	Q4 vs Q1 HR 1.62 (1.25–2.10)
Cook 2016, JACC 68:1609	as above, median 24 y	2,974	HR 1.12 per 1,000 mg (1.00–1.26)	—	HR 1.13 per unit (1.01–1.27), p=0.04

Note the pattern in Cook 2009: sodium alone p=0.38, potassium alone p=0.08, ratio p=0.04 with the best BIC. With gold-standard repeated 24h urines and a formal model comparison, the ratio wins. That is a model-selection result, not a post-hoc preference. Ma 2022’s median 24h urinary sodium was 3,270 mg — this dose-response was demonstrated across a real-world Western range.

Corroborating RCT mechanism: SSaSS is literally a ratio intervention (swap 25% of NaCl mass for KCl → lowers Na and raises K simultaneously) and is the only trial that has moved hard outcomes. DASH-Sodium shows the sodium slope flattening 2.7× on a high-potassium background.

2.3 MASS vs MOLAR — state it explicitly or you will get this wrong

Urinary ratios in epidemiology are molar (mmol/mmol, because urine is assayed in mmol/L). Food-label-derived ratios are mass (mg/mg).

Conversion: molar = mass × 1.70 (derived: 39.10 / 22.99); mass = molar × 0.588.

Reference point	Na	K	Mass ratio	Molar ratio
WHO 2012 targets	2,000 mg	3,510 mg	0.57	0.97 (WHO states “approximately 1:1 molar”)
NASEM 2019 men: CDRR / AI	2,300	3,400	0.68	1.15
NASEM AI / AI	1,500	3,400	0.44	0.75
US adult men, actual	~3,996	~2,900–3,100 (approximate)	~1.3	~2.2

Supported target: molar Na:K ≤ 1.0 = mass Na:K ≤ 0.60. This is the WHO-implied operating point, derived from two independently-set guideline values rather than one cohort’s spline nadir. A pragmatic intermediate step from the US baseline is molar ≤1.5 (mass ≤0.88). Do not anchor on Groenland’s J-curve nadir (molar 2.6–2.7) — Kawasaki spot urines in vascular-disease patients, and the same study produced the implausible “more potassium = more death” result.

Internal consistency check (derived): 2,300 mg Na with 3,800 mg K gives mass 0.61 / molar 1.03 — right at target. Achieving the ratio target is far more sensitive to raising potassium than to cutting sodium further, and ~75% of US adult men are below the potassium AI. For the adult, the highest-leverage single electrolyte change is potassium, not sodium.

2.4 Is potassium’s benefit just “fruits and vegetables”? Substantially yes — and that’s fine for scoring

Arguing it’s a confounded marker:

Top US potassium contributors: fruits/vegetables/100% juice 23%, grain-based mixed dishes 15%, meats/poultry 10%. Anyone eating high potassium eats more produce, fiber, magnesium, nitrate, polyphenols, and less UPF.
The original DASH trial cannot separate them. Appel 1997, NEJM 336:1117, n=459, sodium and body weight held constant: the fruits-and-vegetables arm lowered SBP 2.8 mmHg; the combination diet (adding low-fat dairy and reduced SFA) lowered SBP 5.5. The high-potassium-produce component delivered roughly half the total effect — and even that half bundles fiber, magnesium, and nitrate with potassium.
Non-replication: Kwon 2022, Front Nutr, n=143,050 Korean, 10.1 y: potassium inversely associated with mortality (Q5 HR 0.79), but sodium was unassociated and the Na:K ratio was NOT significantly associated in adjusted models.
NASEM 2019 declined to set a potassium CDRR citing “unexplained inconsistencies… lack of intake-response relationship, and limited evidence,” and lowered the AI from 4,700 to 3,400. That is a formal downgrade of confidence by the body that reviewed everything. NASEM’s stated rationale for the AI is telling: “Given the lack of evidence of potassium deficiency in the population, median intakes observed in an apparently healthy group of people are appropriate for establishing the potassium AI.” “Below the AI” is therefore close to a tautology, not a deficiency signal.

Arguing it’s at least partly real: Aburto’s 22 RCTs used KCl pills, not vegetables — no produce confound possible, and the result was positive (in hypertensives). SSaSS is a pure KCl intervention in 20,995 people that moved stroke, MACE and mortality. Mechanism is well-characterized (natriuresis, vascular smooth-muscle hyperpolarization, WNK-SPAK signalling).

Honest verdict: potassium has a genuine but modest independent BP effect, concentrated in hypertensives; in normotensive free-living cohorts a large fraction of the apparent benefit is dietary-pattern confounding. For scoring purposes this doesn’t matter and may be a feature: a meal-level potassium term rewards vegetables, legumes, potatoes, dairy and fish and penalizes refined/processed items. Whether the causal agent is K⁺ or its usual company, the food-selection behaviour it induces is the one you want. Just don’t claim the mechanism is proven potassium, and don’t call it a “deficiency.”

3. PROTEIN

Assumed anthropometry: 85 kg (state and revise if wrong), ~13–16% BF → FFM ≈ 72 kg.

3.1 g/kg/day

(a) Muscle retention in a deficit — the strongest case for going high

Source	Design	n	Result
Longland 2016, AJCN 103:738	Single-blind parallel RCT, 4 wk, ~40% energy deficit, 6 d/wk RT + HIIT	40 (20/group)	2.4 g/kg/d: LBM +1.2 ± 1.0 kg vs 1.2 g/kg/d: +0.1 ± 1.0 kg (P<0.05). FM −4.8 ± 1.6 vs −3.5 ± 1.4 kg (P<0.05)
Helms 2014, IJSNEM 24:127	SR, energy-restricted lean resistance-trained athletes	6 studies, 13 groups	FFM decreased in 9 of 13 groups. Recommends 2.3–3.1 g/kg FFM, scaled with deficit severity and leanness
Wycherley 2012, AJCN 96:1281	MA, isocaloric high- vs standard-protein energy restriction, mean 12.1 wk	24 trials, 1,063	Weight −0.79 kg (−1.50, −0.08); FM −0.87 kg (−1.26, −0.48); FFM preserved +0.43 kg (0.09, 0.78); REE preserved +595.5 kJ/d

Note the size gap. In the general (mostly non-active adult, mild-deficit) literature the protein effect on FFM is ~0.4 kg — real but small. Longland’s 1.1 kg between-group LBM difference in 4 weeks is far larger because it stacked a severe deficit, hard training, and a wide protein contrast. That combination is close to this subject’s situation, which is why it’s the most relevant single trial. Helms applied here: 2.3–3.1 g/kg FFM × 72 = 166–223 g/d (= 1.95–2.63 g/kg BW).

(b) Hypertrophy at energy balance — the ceiling is lower, and 1.6 is routinely over-read

Morton 2018, Br J Sports Med 52:376 — 49 studies, 1,863 participants. Supplementation effects: 1RM +2.49 kg (0.64, 4.33); FFM +0.30 kg (0.09, 0.52); fibre CSA +310 µm² (51, 570). Efficacy declines with age (−0.01 kg/yr) and is greater in resistance-trained (+0.75 kg).

Breakpoint 1.62 g/kg/day — but the 95% CI is 1.03 to 2.20 and the breakpoint fit was p=0.079, i.e. the plateau itself did not reach conventional significance. The authors themselves write it “may be prudent to recommend ~2.2 g protein/kg/d for those seeking to maximise” FFM gains.
“1.6 is the ceiling” is not what this paper shows. “Somewhere between 1.0 and 2.2, and 2.2 is the defensible upper anchor” is what it shows.
Also note the headline supplementation effect is 0.30 kg FFM — supplements are a rounding error next to training. Total intake is what matters.

(c) Satiety — real but narrower than claimed. Leidy 2015, AJCN 101:1320S: benefits at 1.2–1.6 g/kg/d with ≥25–30 g/meal, but the honest limit is that acute trials confirm “a modest satiety effect… but does not support an effect on energy intake at the next eating occasion.” Long-term studies are “limited and conflicting”; adherence is the primary driver. So protein makes a deficit feel tolerable and protects lean mass; the claim that it makes you spontaneously eat less at the next meal is not supported.

(d) Antonio’s high-protein work and its limits

Study	Design	n	Result
Antonio 2014, JISSN 11:19	8 wk, no training change	30 trained	HP 4.4 ± 0.8 g/kg/d vs 1.8. No significant change in BW, FM, FFM or %BF in either group, despite hypercaloric HP
Antonio 2015, JISSN 12:39	8 wk + periodized heavy RT	48	HP 3.4 vs NP 2.3 g/kg/d. FM −1.7 vs −0.3 kg. FFM change IDENTICAL: +1.5 ± 1.8 vs +1.5 ± 2.2 kg (NS)
Antonio 2016, J Nutr Metab	1-year randomized crossover	14 trained men	3.32 vs 2.51 g/kg/d. No harmful effects on lipids, liver or kidney panels; no fat gain despite higher energy intake

Limits that matter: intake is self-reported; n=14 over a year cannot exclude uncommon harm (absence of evidence ≠ evidence of absence); kidney function assessed by creatinine-based panels, which are themselves elevated by high meat intake and muscle mass; not blinded; and the “control” arms already ate 2.3–2.5 g/kg, so these test 3.4 vs 2.3, not high vs normal. Net read: very high protein doesn’t make you fat and doesn’t visibly break healthy kidneys in the short-to-medium term. It does NOT demonstrate that 3.4 g/kg builds more muscle than 2.3 — in Antonio’s own data, FFM gain was identical.

Recommendation for this subject

Goal state	g/kg BW	g/day @ 85 kg
Maintenance / hypertrophy at balance	1.6–2.2	136–187
Current: deficit + muscle retention + already lean	1.9–2.4	165–205
Aggressive deficit (>30–40%) or very lean	2.6–3.0	220–255

At [calorie target removed], 180 g/d = 720 kcal = 25% of energy — comfortably inside the AMDR, leaving 2,130 kcal for carbohydrate and fat. That matters: for recreational sport + VO2max work, carbohydrate availability is the actual performance-limiting nutrient (§10). Going to 250 g protein costs 280 kcal of carb/fat budget for no demonstrated FFM benefit.

3.2 Per-meal dose and the leucine threshold — the ceiling is collapsing

Study	Design	n	Result
Moore 2009, AJCN 89:161	Crossover, leg RT, 0/5/10/20/40 g egg protein, tracer, 4 h	6 young men	MPS maximally stimulated at 20 g; 40 g gave no further MPS. Leucine oxidation rose at 20 and 40 g
Macnaughton 2016, Physiol Rep 4:e12893	Crossover, whole-body RT, 20 vs 40 g whey	30 trained men (15 low-LBM, 15 high-LBM)	40 g > 20 g: 0.059 ± 0.020 vs 0.049 ± 0.020 %·h⁻¹, P=0.005 (~20% higher). LBM did not modify the response — “bigger men need more” was tested and NOT supported
Areta 2013, J Physiol 591:2319	80 g whey over 12 h, three patterns	24 trained men	PULSE 8×10 g, INT 4×20 g, BOLUS 2×40 g. INT beat both by 31–48% (P<0.02). But signalling ranked BOLUS>INT>PULSE — phosphorylation and MPS disagreed within one study
Trommelen 2023, Cell Rep Med 4:101324	Quadruple tracer, intrinsically labelled milk protein, post-RT, 12 h	36 (12/arm: 0 / 25 / 100 g)	100 g > 25 g in magnitude AND duration (>12 h). Myofibrillar MPS ~20% higher at 0–4 h, ~40% higher at 4–12 h. Amino acid oxidation was negligible (<15%); ~13 g of the 100 g appeared in muscle protein. Explicitly: the response “is not restricted and has previously been underestimated”

Read as a sequence this is a twenty-year retreat from the ceiling. Moore (n=6, isolated leg, fast protein) → “20 g is the max.” Macnaughton (n=30, whole-body) broke it. Trommelen (n=36, whole-food-like protein, 12 h tracing) demolished it and directly falsified the “excess is just oxidized” premise.

The leucine threshold — the number, and how weak it is. ISSN Position Stand (Jäger 2017, JISSN 14:20), verbatim: “Acute protein doses should strive to contain 700–3000 mg of leucine.” That is a 4.3-fold range presented as a recommendation — the range width is the honest answer about evidence strength. The commonly cited “~2.5–3 g leucine per meal” is the top of that band, not a measured breakpoint.

Direct causal evidence that leucine is the trigger — Churchward-Venne 2014, AJCN 99:276. Parallel, double-blind, unilateral knee-extensor RT, tracer, n=40 men. MPS was greatest after 25 g whey (~267%) and after 6.25 g whey + leucine to 5.0 g total (~220%) (P=0.002). A 6.25 g dose matched a 25 g dose if leucine hit 5.0 g. Two caveats: the 3.0 g leucine top-up did not reproduce it — only 5.0 g did, which sits above the ISSN’s stated ceiling; and the low-protein-plus-leucine condition supplies inadequate total EAA for sustained net balance.

Practical leucine content (Gorissen 2018, Amino Acids 50:1685, UPLC-MS/MS): leucine as % of protein — milk 9.0%, egg 7.0%, human muscle 7.6%; plants 5.1% (hemp) to 13.5% (corn). EAA: whey 43%, milk 39%, casein 34%, egg 32% vs oat/lupin 21%, wheat 22%. Methionine and lysine: plant 1.0 ± 0.3% and 3.6 ± 0.6% vs animal 2.5 ± 0.1% and 7.0 ± 0.6%. → At 8–9% leucine, 2.5 g leucine needs ~28–31 g of dairy/meat/egg protein. At 5–6%, a plant-only meal needs ~42–50 g. That difference is the entire practical content of “protein quality.” (Note corn’s 13.5% leucine but severe lysine limitation — quality = EAA total × digestibility × absence of a limiting AA, not leucine alone.)

Is “0.4 g/kg/meal × 4 meals” well-supported? Schoenfeld & Aragon 2018, JISSN 15:10 recommend exactly that (and up to 0.55 g/kg/meal to reach 2.2 g/kg/d). Honest assessment: the recommendation is arithmetically derived, not experimentally tested. Read the structure — it is 1.6 g/kg/d ÷ 4 meals, with the meal count chosen so the division lands above the acute-MPS saturating dose. No trial randomized people to 0.4 g/kg × 4 vs an alternative and measured hypertrophy. That doesn’t make it wrong; it is a sensible, safe, non-falsified heuristic erring on the generous side. It should just not be cited as an experimentally established optimum. For 85 kg: 0.4 = 34 g/meal; 0.55 = 47 g/meal.

3.3 Does distribution change OUTCOMES, or only acute MPS?

This is the weakest link in the entire per-meal-protein edifice, and it should be stated bluntly: the acute-MPS → chronic-hypertrophy inference is assumed, not validated.

Acute case for even distribution — Mamerow 2014, J Nutr 144:876. 7-day crossover, isoenergetic and isonitrogenous, tracer, 24-h MPS with biopsies. n=8. EVEN (31.5/29.9/32.7 g) vs SKEW (10.7/16.0/63.4 g): 24-h FSR 0.075 vs 0.056 %/h, +25%, P=0.003, holding after 7 days. Clean and well-controlled — but n=8, total protein only ~1.0 g/kg (well below this subject), and the outcome is a synthesis rate, not tissue.
The chronic case — Yasuda 2020, J Nutr 150:1845. 12-week parallel RCT, supervised RT 3×/wk, DXA lean tissue as primary outcome, n=26, matched total protein 1.30 g/kg/d. HBR (0.33/0.46/0.48 g/kg) vs LBR (0.12/0.45/0.83): LTM +2.5 ± 0.3 vs +1.8 ± 0.3 kg, P = 0.06, d = 0.795. Read that P value. The title says distribution “Augments” hypertrophy; the primary outcome was not statistically significant. A large effect size at n=26 is exactly what an underpowered true-null or an inflated true-effect looks like. The most-cited chronic distribution RCT is a non-significant trend.
The systematic assessment — Hudson, Bergia & Campbell 2020, Nutrients 12:1441, verbatim: “The current evidence… is limited and inconsistent. The effect of protein distribution cannot be sufficiently disentangled from the effect of protein quantity.” And: “for adults already consuming 0.8–1.3 g·kg⁻¹·d⁻¹, the preponderance of evidence supports that consuming at least one meal that contains sufficient protein quantity to maximally stimulate MPS, independent of daily distribution, is helpful.” That last clause directly contradicts the strong version of the distribution doctrine.
Why the acute→chronic link is suspect mechanistically: Areta 2013 found MPS and signalling ranked differently within a single study; Trommelen 2023 showed the anabolic window from one feeding extends beyond 12 h, meaning 12-h acute windows systematically mis-measure what a 24-h day of eating does.
Meal frequency per se does nothing. Schoenfeld, Aragon & Krieger 2015, Nutr Rev 73:69, 15 studies: frequency initially appeared positively associated with fat loss and FFM, but “sensitivity analysis… showed that the positive findings were the product of a single study.”

Verdict: distribution is second-order. Total daily protein is first-order by a wide margin. The realistic magnitude of a distribution effect, if it exists, is the Yasuda gap (~0.7 kg lean over 12 weeks) with a CI including zero. Score per-meal protein anyway — it is easy, non-falsified, and errs generous — but do not weight it as if it were established.

3.4 Is there a ceiling?

Stops helping — yes, fairly early. Morton: plateau at 1.62 (CI 1.03–2.20). Antonio 2015: 3.4 vs 2.3 g/kg → identical FFM gain. Antonio 2014: 4.4 g/kg hypercaloric for 8 weeks → no change in anything.

Starts hurting — no evidence of harm within the tested range. What is real is opportunity cost, not toxicity: at a fixed [calorie target removed], every gram above ~200 g displaces carbohydrate, a direct performance cost for a court + VO2max active adult. (A nitrogen/urea-cycle ceiling around ~3.5–4.5 g/kg/d or ~35% of energy is often cited — UNVERIFIED RECALL, no primary source retrieved. Antonio’s cohorts reached 4.4 g/kg without reported issue, so if a hard ceiling exists it is above anything relevant here.)

Practical ceiling: ~2.4 g/kg (205 g/d). Above that the adult is trading carbohydrate for nothing measurable.

3.5 Kidney safety — settled for healthy people, NOT for reduced kidney function

Devries 2018, J Nutr 148:1760. SR + MA of RCTs >4 d, high protein (≥1.5 g/kg or ≥20% E or ≥100 g/d) vs lower, adults without kidney disease, GFR outcome. 28 studies, 1,358 participants.
- Post-intervention GFR: SMD 0.19 (0.07, 0.31), P=0.002 — a trivially higher GFR (the authors’ own word).
- Change in GFR: SMD 0.11 (−0.05, 0.27), P=0.16 — no difference.
- Dose-response present for post GFR (r=0.332, P=0.03) but not for change in GFR (r=0.184, P=0.33).
- Conclusion: “HP intakes do not adversely influence kidney function on GFR in healthy adults.”
- The post-vs-change distinction is the whole argument: GFR sits slightly higher on high protein but does not progressively decline. That is the signature of adaptation (recruited functional renal reserve, as in pregnancy), not injury.
Van Elswyk 2018, Adv Nutr 9:404. 26 studies. Of 13 RCTs comparing GFR, 8 reported significantly higher GFR with more protein, 5 did not — and all rates were within the normal range. No adverse effect on BP; kidney-stone evidence “limited and inconsistent.” Declared limitations: all studies at moderate-to-high risk of bias; all but 2 cohorts <6 months. COI note: authors affiliated with the National Cattlemen’s Beef Association — though the conclusions align with the independent Devries MA.
The genuine residual uncertainty: no RCT has run decades in healthy people. This is uncertainty, not evidence of harm.
Where “protein is safe” must stop — reduced kidney function. ISRNM/KDOQI 2020 commentary (J Ren Nutr 31:116): CKD without diabetes 0.55–0.60 g/kg/d; diabetic kidney disease 0.6–0.8 g/kg/d. That is roughly one quarter of this subject’s target. Same logic applies to reduced nephron mass (single kidney, prior nephrectomy, prior AKI).
Action item: know the baseline. One metabolic panel with eGFR — ideally cystatin C-based, since creatinine-based eGFR is systematically biased downward in muscular people eating a lot of meat — converts an unknown risk into a known one. Note this bias contaminates most of the “no harm” literature, including Antonio’s.

Bone / acid-ash hypothesis — dead, and the direction is if anything protective.

Fenton 2009, Nutr J 8:41. MA of the acid-ash hypothesis, 12 studies, 30 arms, 269 subjects: all analyses showed significant decreases in urinary calcium with phosphate supplementation; no analysis showed lower calcium balance. Verbatim: “All of the findings… were contrary to the acid ash hypothesis.”
Shams-White 2017, AJCN 105:1528, National Osteoporosis Foundation-commissioned, 16 RCTs + 20 cohorts: moderate evidence higher protein protects lumbar spine BMD, +0.52% (0.06, 0.97), I²=0%. No effect at hip/femoral neck/total body. Verbatim: “Current evidence shows no adverse effects of higher protein intakes.”
Mechanism, resolved: higher protein increases intestinal calcium absorption (more in, more out, balance unchanged or improved) and raises IGF-1, which is anabolic to bone. The observation that drove the hypothesis — more urinary calcium after a protein load — was misinterpreted.

3.6 Protein and visceral fat specifically — weak and indirect

One clean RCT with VAT as an outcome, in the wrong population. Huang 2021, J Gerontol A 76:1084 — 2×2 RCT, 6 months, 0.8 vs 1.3 g/kg/d × testosterone/placebo; 92 randomized, VAT by DXA in 56. 1.3 vs 0.8 g/kg: VAT difference −17.3 cm² (−29.7, −4.8), P=0.008, regardless of testosterone; no difference in glucose, insulin, HOMA-IR, leptin, adiponectin, IL-6 or hs-CRP. Why it barely transfers: participants were ≥65, functionally limited, habitually eating ≤0.8 g/kg. The comparison is deficient vs adequate, not adequate vs high. A adult already at 1.8+ g/kg is nowhere on this curve. DXA-estimated VAT is also weaker than CT/MRI.

The general route is better supported: Wycherley (24 trials, n=1,063) — more fat lost (−0.87 kg) and more FFM retained (+0.43 kg) per unit deficit. Since VAT is preferentially mobilized during energy deficit, protein’s visceral benefit is downstream of a better-quality deficit, not a direct visceral-specific effect. That is a sufficient reason on its own.

3.7 Protein quality — a dilution problem, not a biology problem

DIAAS vs PDCAAS (Marinangeli & House 2017, Nutr Rev 75:658): DIAAS uses true ileal amino-acid digestibility rather than faecal crude-protein digestibility (which is confounded by colonic microbial metabolism), and is untruncated — PDCAAS caps at 1.00, erasing the difference between an adequate protein and a surplus one and hiding a high-quality protein’s ability to compensate for a poor one in a mixed meal. (Specific per-food DIAAS decimals — whey ~1.1, soy ~0.9, pea ~0.6–0.8, wheat ~0.4–0.5 — are UNVERIFIED RECALL. The ordering is well established; the decimals are not confirmed.)

Does quality matter at high total intake? Largely no. Lim 2021, Nutrients 13:661 — SR+MA of RCTs, animal vs plant protein, 18 in review / 16 in MA, with total protein “generally above the RDA at baseline and end of intervention”: protein source did NOT affect absolute lean mass or muscle strength. In adults <50 y specifically, animal protein gained WMD 0.41 kg (0.08, 0.74) absolute lean mass and 0.50% (0.00, 1.01) percent lean mass — statistically real, practically trivial, with the percent-lean CI touching zero.

Synthesis. At 1.0 g/kg from wheat and oats, leucine and lysine shortfalls bite. At 2.0+ g/kg from a mixed diet, every EAA requirement is overshot regardless of source. For this subject at 180–200 g/d, source is close to irrelevant, with two practical exceptions: (1) a fully plant-based meal needs roughly 1.3–1.6× the grams to match leucine; (2) blend plant sources (legume + grain) for lysine/methionine complementarity.

3.8 NORMALIZER for protein — the gate/pacing architecture, and when per-calorie is WRONG

Normalizer	What it governs	Behavior
Absolute g/meal	Whether the meal crosses the anabolic threshold	Threshold. Below ~25–30 g the meal contributes little to MPS; above it, returns diminish
g/kg BW/meal	Same, cross-person comparable	Threshold, body-size adjusted
g per 100 kcal	Whether the adult can hit the daily target inside the calorie budget	Linear / pacing. Only meaningful when calories are constrained
% of meal kcal	Identical information	%kcal = 4 × (g per 100 kcal). Redundant — pick one

Architecture: absolute grams (or g/kg) is the GATE; per-calorie density is the PACING score. Never score protein on density alone.

Per-calorie is WRONG when:

The meal is small. A 90-kcal nonfat Greek yogurt with 10 g protein scores 11.1 g/100 kcal — near-perfect density — but 10 g crosses no threshold and delivers <1 g leucine. Any density score needs a minimum-calorie floor (~200 kcal) below which you report grams only.
The meal is large. A 950-kcal post-training meal with 55 g protein scores 5.8 g/100 kcal — “below pace” — yet 55 g is more than enough. The low density flags a calorie problem, not a protein problem.
Peri-workout. What drives the response is the absolute EAA/leucine bolus; the ratio to accompanying carbohydrate is irrelevant, and you want carbohydrate there.
The adult is not in a deficit. Density exists to ration a scarce calorie budget.
The denominator is near zero. Protein powder in water = 25 g/100 kcal; the metric explodes and stops discriminating. Cap it.

BW vs FFM scaling: Helms scales to FFM, which is more physiologically correct. For this subject they nearly coincide — 2.0 g/kg BW (85 kg) = 170 g; 2.4 g/kg FFM (72 kg) = 173 g. The distinction only matters at higher body fat. Use bodyweight here.

Daily anchor: 180 g / [calorie target removed] = 2.1 g/kg = 6.3 g per 100 kcal = 25.3% of calories. That 6.3 g/100 kcal is the break-even pace.

4. FIBER

4.1 The anchor: Reynolds et al., Lancet 2019 — and the gap that matters

Reynolds A, Mann J, Cummings J, et al. Lancet 2019;393(10170):434–445. PMID 30638909. WHO-commissioned. Scope: “Just under 135 million person-years of data from 185 prospective studies and 58 clinical trials with 4,635 adult participants.”

Higher vs lower total fibre — cohorts:

Outcome	Studies	Cases / PY	Effect	GRADE
All-cause mortality	10	80,139 deaths / 12.3M PY	RR 0.85 (0.79–0.91)	Moderate
CHD mortality	10	7,243 / 6.9M PY	RR 0.69 (0.60–0.81)	Moderate (Egger p=0.0040 — publication bias)
CHD incidence	9	7,155 / 2.7M PY	RR 0.76 (0.69–0.83)	Moderate
Stroke incidence	9	13,134 / 4.6M PY	RR 0.78 (0.69–0.88)	Low
Cancer mortality	5	29,593 / 11.2M PY	RR 0.87 (0.79–0.95)	Moderate
Type 2 diabetes	17	48,468 / 6.9M PY	RR 0.84 (0.78–0.90)	Moderate
Colorectal cancer	22	22,920 / 16.9M PY	RR 0.84 (0.78–0.89)	Moderate

Absolute: 13 fewer deaths (8–18) and 6 fewer CHD cases (4–7) per 1,000 over study duration.

The RCT arm — where the story weakens:

Outcome	Trials	n int / ctrl	Effect	GRADE
Body weight	27	1,294 / 1,201	MD −0.37 kg (−0.63 to −0.11)	High
HbA1c	6	191 / 189	SMD −0.35 (−0.73 to +0.03) ns	Low
Total cholesterol	36	1,832 / 1,671	MD −0.15 mmol/L (−0.22 to −0.07)	Moderate
Systolic BP	15	1,064 / 988	MD −1.27 mmHg (−2.50 to −0.04)	Moderate

The central tension: cohorts say 15–31% risk reduction; the HIGH-quality RCT evidence says 0.37 kg, 5.8 mg/dL total cholesterol, 1.27 mmHg. Those risk-factor deltas are far too small to generate a 15% mortality reduction. Reynolds argues concordance in direction implies causality. The magnitude gap is the argument against.

4.2 Dose-response: linear, and 25–29 g is NOT an optimum

Per 8 g/day increment: all-cause mortality RR 0.93 (0.90–0.95) (68,183 deaths); CHD incidence 0.81 (0.73–0.90); T2D 0.85 (0.82–0.89); colorectal cancer 0.92 (0.89–0.95). Whole grains, per 15 g/day: all-cause 0.94 (0.92–0.95); CHD 0.93; T2D 0.88; CRC 0.97 — all weaker per unit than fibre, GRADE Low. Shape, verbatim: “many of which are linear with no sign of a plateau within the available data.”

Where “25–29 g” actually comes from, and it is widely misreported. It is not an inflection point, threshold, or curve maximum. The authors compared lowest consumers against pre-specified bins (15–19, 20–24, 25–29, 30–34, 35–39 g/d) and counted how many of 7 critical outcomes showed a significant improvement in each bin: 15–19 g → 3 of 7; 20–24 g → 4 of 7; 25–29 g → 6 of 7. That is a vote-count of statistical significance across sparse bins, heavily driven by how many studies populate each bin — higher bins have fewer cohorts and wider CIs, so they lose the count regardless of true effect. The authors are explicit it is a floor: “intakes… should be no less than 25 to 29 grams per day with additional benefits likely to accrue with higher intakes.”

Bottom line: the curve keeps falling. 25–29 g is a defensible minimum, not a target. 35–40 g/day is the evidence-consistent aim at [calorie target removed], with no evidence of harm or plateau in these data.

4.3 Which fractions do what

Viscous / gel-forming — the only fraction with replicable, dose-predictable metabolic effects:

Psyllium — Jovanovski 2018, AJCN 108:922. 28 RCTs, n=1,924, median 10.2 g/d → LDL-C −0.33 mmol/L (−0.38 to −0.27) ≈ −12.8 mg/dL; apoB −0.05 g/L. GRADE moderate (LDL), high (apoB).
Oat β-glucan — Ho 2016, Br J Nutr 116:1369. 58 RCTs, n=3,974, median 3.5 g/d → LDL-C −0.19 mmol/L ≈ −7.3 mg/dL. I²=79%.
Barley β-glucan — Ho 2016, EJCN 70:1239. 14 RCTs, n=615, median 6.5 g/d → LDL-C −0.25 mmol/L ≈ −9.7 mg/dL; apoB not significantly changed.
Roughly 1–1.3 mg/dL LDL per g/day of viscous fiber, flattening at higher doses.

Viscosity also drives satiety — Wanders 2011, Obes Rev 12:724: more viscous fibers reduced appetite in 59% of comparisons vs 14% for less viscous (n=58 comparisons); acute energy intake 69% vs 30% (n=26). Verbatim caveat: “effects on energy intake and body weight were relatively small, and distinct dose-response relationships were not observed.”

Soluble vs insoluble for mortality — Front Nutr 2023 dose-response MA (PMID 37854351), per 10 g/day: soluble all-cause 0.83 (0.74–0.92), CVD 0.62 (0.47–0.84); insoluble 0.86 / 0.81; cereal 0.82 / 0.84; fruit 0.99 (0.92–1.07) ns; vegetable 0.88 ns. Note this MA found a non-linear relation for total fibre and all-cause mortality, directly contradicting Reynolds — see Contested.

Cereal fiber dominates CHD — Pereira 2004, Arch Intern Med 164:370, pooled analysis of 10 cohorts, 91,058 men + 245,186 women, 5,249 coronary events, 2,011 coronary deaths. Per 10 g/d energy-adjusted, error-corrected total fibre: events RR 0.86 (0.78–0.96); coronary death RR 0.73 (0.61–0.87). By source (events / deaths): cereal 0.90 / 0.75; fruit 0.84 / 0.70; vegetable 1.00 (0.88–1.13) / 1.00 (0.82–1.23) — flatly null. Threapleton 2013, BMJ 347:f6879, 22 cohorts, per 7 g/d: CVD 0.91 (0.88–0.94).

Fermentable / SCFA — mechanistically rich, clinically thin. The best human trial is Chambers 2015, Gut 64:1744: novel inulin-propionate ester delivering propionate to the colon, n=60, 24 weeks, 10 g/d vs inulin control → reduced weight gain, intra-abdominal adipose tissue, and intrahepatocellular lipid. But the comparator was inulin itself — so it demonstrates propionate does something, not that fermentable fiber does. Roager 2019, Gut 68:83 is the relevant cold water: n=60 randomized / 50 completed, 8-wk crossover, whole grain 179 ± 50 g/d vs refined 13 ± 10 g/d — a very large real-food contrast — reduced body weight and low-grade inflammation but did NOT alter insulin sensitivity and did NOT meaningfully change the gut microbiome. That is a direct problem for the fiber → microbiome → health chain.

4.4 Intrinsic vs isolated/fortification fiber — NOT equivalent

FDA’s definitional history. May 2016 Nutrition Facts rule accepted 7 isolated/synthetic non-digestible carbohydrates (beta-glucan soluble fiber, psyllium husk, cellulose, guar gum, pectin, locust bean gum, HPMC). June 2018 guidance announced intent to add 8 more under enforcement discretion — mixed plant cell wall fibers, arabinoxylan, alginate, inulin and inulin-type fructans, high-amylose starch (RS2), galactooligosaccharide, polydextrose, resistant maltodextrin/dextrin. Later: RS4 (2019), glucomannan (2020), acacia gum (2021).

The criterion is the weak point. FDA requires a “physiological effect that is beneficial to human health,” and the accepted effects are: lowering postprandial glucose/insulin; lowering fasting LDL or fasting glucose; lowering BP; increased frequency of bowel movements; increased mineral absorption; reduced energy intake. Read that carefully: a substance qualifies as “dietary fiber” on a US label if it makes you poop more often, or slightly nudges a postprandial glucose curve. None of the accepted endpoints is a hard clinical outcome. The FDA list is a laxation-and-biomarker registry, not a claim of equivalence to the fiber measured in the Reynolds cohorts.

What the outcome evidence says:

Viscous isolates DO work for their specific biomarker — psyllium and β-glucan genuinely lower LDL (above). These are the honest exceptions.
Non-viscous fermentable isolates (inulin, oligofructose, polydextrose, soluble corn fiber, resistant maltodextrin) do not show comparable effects. Wanders 2011’s classification is the cleanest evidence: 14% effect rate on appetite vs 59% for viscous; 30% vs 69% for acute energy intake.
No dedicated meta-analysis of polydextrose, soluble corn fiber, or resistant maltodextrin on body weight or glycemic control could be retrieved. That absence is itself informative, given these are among the most heavily used fortification fibers in the US food supply.

Practical rule: if the ingredient list names the fiber, discount it. If the fiber has no name because it came with the food, count it fully.

4.5 Is fiber causal, or a marker for whole plant food? Mostly a marker

The RCT record is bad for fiber-as-agent:

Trial	Design	n	Intervention	Result
DART — Burr 1989, Lancet	Factorial RCT, post-MI men	2,033	Cereal fibre advice	*“Subjects given fibre advice had a slightly higher* mortality” (ns). Same trial: fatty fish advice → 29% reduction in 2-y all-cause mortality**, significant
Polyp Prevention Trial — Schatzkin 2000, NEJM	RCT, 4 y	2,079	18 g fibre/1000 kcal + 3.5 servings F&V/1000 kcal	Adenoma recurrence 39.7% vs 39.5%; RR 1.00 (0.90–1.12)
Wheat Bran Fiber — Alberts 2000, NEJM	RCT, ~3 y	1,429	13.5 vs 2 g/d wheat bran	Adenoma OR 0.88 (0.70–1.11), p=0.28
Cochrane 2017 (CD003430.pub2)	SR of RCTs	4,798 / 7 studies	Wheat bran, ispaghula, or high-fibre diet	≥1 adenoma RR 1.04 (0.95–1.13)

And the Cochrane result that almost never gets quoted: “The results on the number of participants diagnosed with colorectal cancer favoured the control group over the dietary fibre group (2 RCTs, n=2,794, RR 2.70, 95% CI 1.07 to 6.85, low-quality).” Also: ispaghula husk, ≥1 recurrent adenoma RR 1.45 (1.01–2.08). Cochrane cautions these are low-quality with >16% attrition and not robust to sensitivity analysis. This is not a claim that fiber causes colorectal cancer. It is a claim that the RCT literature provides ZERO support for fiber-as-isolated-agent preventing colorectal neoplasia, and contains a nominally significant signal in the wrong direction.

The pattern to notice: cohorts RR 0.85 all-cause / 0.76 CHD → high-GRADE RCTs −0.37 kg, −1.27 mmHg → hard-outcome RCTs null. High-fiber eaters in cohorts smoke less, weigh less, exercise more, eat less processed meat, drink less, and are more educated; FFQ measurement error further biases adjustment.

Most defensible reading: dietary fiber is largely a marker for a whole-plant-food dietary pattern, with a genuine but modest causal core confined to (a) viscous fiber → LDL/apoB, (b) bulk/laxation, and (c) energy-density displacement. That is not nothing — it is just much smaller than “RR 0.85.” Actionable consequence: chase the foods, not the grams.

4.6 Fiber and visceral fat

Hairston 2012, Obesity 20:421. IRAS Family Study, OBSERVATIONAL, n=1,114 (339 African American, 775 Hispanic American), abdominal CT at baseline and 5 years. “For each 10 g increase in soluble fiber, rate of VAT accumulation decreased by 3.7% (P=0.01)” and “Soluble fiber was not associated with change in SAT (0.2%, P=0.82)” — independent of BMI change. Moderately active participants: −7.4% VAT accumulation (P=0.003). Caveat: FFQ administered only at follow-up, so exposure is measured after most of the outcome accrued. The VAT-specific/SAT-null pattern is biologically interesting; the design cannot establish causality.
RCTs measuring VAT directly are sparse and mostly null. J Diet Suppl 2013;10:129, double-blind RCT, n=30 Japanese men with VAT >100 cm², 7.5 g/d fiber for 12 wk: fasting glucose improved, weight and BMI fell, but “visceral and subcutaneous fat areas did not change significantly.”
Katcher 2008, AJCN 87:79. n=50 obese adults with metabolic syndrome, 12 wk hypocaloric, whole grain vs avoid-whole-grain: both lost weight; significantly greater decreases in CRP and in percentage body fat in the abdominal region with whole grains (P<0.05).

Verdict: for a lean active 29M, fiber is a small lever on visceral fat, not a large one. The verified large levers are energy deficit, physical activity, and removing liquid fructose.

4.7 The 14 g/1,000 kcal recommendation — and whether per-calorie is the right normalizer

Origin: IOM 2002 DRI, Adequate Intake set at 14 g per 1,000 kcal → 38 g/day men, 25 g/day women, based on observational CHD-protection data. The key primary source is Pietinen 1996, Circulation 94:2720 (ATBC, 21,930 Finnish male smokers, 6.1 y, 1,399 major coronary events, 635 CHD deaths): highest quintile of total fibre (median 34.8 g/d) → coronary death RR 0.69 (0.54–0.88).

What US adults actually eat against that target: NHANES 2009-2010 mean daily fiber intake is 16.2 g/day (Quagliani D, Felt-Gunderson P. “Closing America’s Fiber Intake Gap: Communication Strategies From a Food and Fiber Summit.” Am J Lifestyle Med 2017;11(1):80-85, PMID 30202317, https://pmc.ncbi.nlm.nih.gov/articles/PMC6124841/), against the IOM AI of 25 g/day for women and 38 g/day for men. The same paper reports only ~5% of the population meets the recommendation. [V]

Two real problems. (1) It is an AI, not an EAR/RDA — IOM set an AI precisely because the evidence was judged insufficient to define a requirement. (2) The entire dose-response literature is expressed in absolute g/day, not g/1,000 kcal. Reynolds’ curves, Pereira’s per-10 g, Threapleton’s per-7 g — all absolute. There is no dose-response evidence in energy-normalized units. The per-calorie normalizer was an extrapolation device, not a finding.

Arguments against per-calorie: fiber’s mechanisms are absolute-mass phenomena — luminal viscosity depends on grams of gel-former per volume of chyme; fermentation on grams of substrate reaching the colon; fecal bulk on grams of water-holding matrix. The adult’s colon is not larger because the adult eats [calorie target removed] instead of 2,000. Scaling to energy over-prescribes (14 × [calorie target removed] = 40 g/day), pushing into GI-tolerance territory.

Argument for per-calorie: fiber density is an excellent proxy for degree of food processing. A diet at 14 g/1,000 kcal is almost necessarily built from intact plants; one at 6 g/1,000 kcal is almost necessarily refined/UPF. Given §4.5’s conclusion that fiber is mostly a marker, the per-calorie metric may measure the thing that matters (pattern quality) better than the absolute count measures the thing that doesn’t (fiber as agent).

Resolution — use both for different jobs:

Absolute g/day = the physiological dose. Target 35–40 g/day.
g/1,000 kcal = the food-quality / processing score. Target ≥12–14 g/1,000 kcal. This is the better per-MEAL scoring metric — scale-invariant, cannot be gamed by eating more.

4.8 Downsides: tolerance, FODMAPs, athletes

Dose-finding — Bonnema 2010, J Am Diet Assoc 110:865. Randomized double-blind crossover, n=26 healthy adults, GI tolerance scored across 7 domains. Verbatim: “Doses up to 10 g/day of native inulin and up to 5 g/day of oligofructose were well-tolerated.” The 10 g oligofructose dose substantially increased GI symptoms. Most frequent: flatulence, then bloating. Chain length matters more than gram count — short-chain FOS ferments fast and proximally; long-chain native inulin roughly doubles the tolerated dose.
Mechanism by imaging — Gunn/Whelan 2022, Gut 71:919. Randomized 4-period crossover, n=19 IBS, MRI-quantified colonic gas. A single 20 g inulin drink: gas AUC median 3,145 mL·min; co-administering 20 g psyllium cut it to 618 mL·min (p=0.02), placebo-equivalent, with breath H₂ AUC falling 7,230 → 1,035 ppm·h. Useful: psyllium blunts inulin-driven gas without inhibiting fermentation.
Athletes pre-competition — weak evidence, be explicit about that. Gaskell/Costa 2025, IJSNEM: a 48-h high-carbohydrate low-FODMAP diet before the target event reduced exercise-associated GI symptoms — but this is a case series, n=9, the lowest interventional tier. Lis 2019 (review): no direct benefit of gluten avoidance in healthy athletes; apparent benefits are indirectly attributable to concurrent FODMAP reduction.
Practical, explicitly labelled as hypothesis: for a [calorie target removed] active adult habitually at 35–40 g/d, dropping to ~15–20 g/d for 24 h pre-event with fiber concentrated away from the 3–4 h pre-session window is a reasonable operationalization. No RCT establishes a specific gram threshold.
Minerals/phytate: UNVERIFIED RECALL — the classic concern is phytate-bound rather than fiber-bound mineral loss and is generally considered clinically minor below ~50 g/day in mixed Western diets with adequate mineral intake. Not verified.

5. ADDED SUGAR vs TOTAL SUGAR

5.1 The added/total distinction is WEAKER than the guidelines imply

Nutrition 2023;112:112032 (PMID 37182401) — SR + dose-response MA of prospective cohorts, highest vs lowest:

Exposure	All-cause mortality	CVD mortality	Cancer mortality
Total sugars	1.09 (1.02–1.15), I²=71.9%	1.10 (1.02–1.18)	1.00 (0.94–1.05)
Fructose	1.09 (1.03–1.16)	1.11 (1.03–1.20)	1.00 (0.95–1.06)
Added sugars	null	null	null
Sucrose	null	null	null

Verbatim: “No association was found for the added sugars and sucrose with all-cause, CVD, and cancer mortality.” Splines showed non-linear associations for total sugars and fructose with risk rising above ~10% of energy.

This is the opposite of what the added-sugar framing predicts. Plausible explanations: added-sugar exposure is measured far worse in FFQs (it requires recipe-level composition databases that changed repeatedly over the cohorts’ lifetimes); fewer cohorts report it, so power is lower; and total sugar in these populations is dominated by added sugar anyway. But you cannot claim the added/total distinction is settled epidemiologically. It is not. Against it stands Yang 2014 (§5.4), which found strong added-sugar gradients. Likely explanation is differential measurement error — but flag this as genuinely contested.

Is intrinsic sugar benign? For whole fruit, yes — and the fruit-vs-juice contrast is the cleanest natural experiment in the field.

Muraki 2013, BMJ 347:f5001. NHS + NHS II + HPFS = 187,382 participants; 3,464,641 person-years; 12,198 incident T2D cases. Pooled HR per 3 servings/week:

Food	HR (95% CI)
Total whole fruit	0.98 (0.97–0.99)
Blueberries	0.74 (0.66–0.83)
Grapes/raisins	0.88 (0.83–0.93)
Apples/pears	0.93 (0.90–0.96)
Bananas	0.95 (0.91–0.98)
Oranges	0.99 (0.95–1.03) ns
Strawberries	1.03 (0.96–1.10) ns
Cantaloupe	1.10 (1.02–1.18) ← increased risk
Fruit juice	1.08 (1.05–1.11)

Heterogeneity among individual fruits P<0.001. Replacing 3 servings/week of juice with whole fruit → ~5% lower T2D risk (95% CI 3–7%).

Read carefully: the whole-fruit protective effect is 0.98 per 3 servings/week — statistically robust across 12,198 cases but tiny. The striking numbers are the berry-specific effect (0.74) and the juice penalty (1.08) — and cantaloupe was significantly harmful, which should temper any “all whole fruit is equivalently good” claim. The active ingredient is much more plausibly polyphenols and fiber matrix than the sugar being intrinsic.

(Dairy lactose: UNVERIFIED RECALL — lactose is excluded from WHO “free sugars” and FDA “added sugars” on matrix/glycemic grounds, but no cohort isolating lactose intake and cardiometabolic outcomes was retrieved. Treat “dairy lactose is benign” as plausible but unverified.)

Bottom line: the distinction that actually carries the signal is MATRIX and PHYSICAL FORM, not the added/intrinsic bookkeeping category.

5.2 Fructose: where the dose actually matters

The “it’s just calories” camp (Sievenpiper/Toronto):

Sievenpiper 2012, Ann Intern Med 156:291. 31 isocaloric trials (n=637) → body weight MD −0.14 kg (−0.37 to +0.10), null. 10 hypercaloric trials (n=119) at +104 to 250 g/d fructose → +0.53 kg (0.26–0.79). Authors’ own caveat: studies were “small (<15 participants), short (<12 weeks), and of low quality.”
Chiu 2014, EJCN 68:416. 13 trials, n=260. 7 isocaloric trials → no effect on intrahepatocellular lipid or ALT. 6 hypercaloric (+21–35% energy) → IHCL SMD 0.45 (0.18–0.72); ALT +4.94 U/L. Conclusion: “may be more attributable to excess energy than fructose.”
Sievenpiper 2012, Diabetes Care 35:1611. 18 trials, n=209: isocaloric exchange of fructose for other carbohydrate improved glycated proteins, SMD −0.25 ≈ −0.53% HbA1c.
Nutrients 2018;10:1805. Small doses (≤50 g/d or ≤10% E): HbA1c −0.38% (−0.64 to −0.13), 14 comparisons n=337, GRADE low.

The “fructose is specifically lipogenic at ordinary doses” camp:

Schwarz 2015, JCEM 100:2434. n=8 healthy men, inpatient, consecutive 9-day periods, weight-maintaining and ISOCALORIC, 25% of energy as fructose vs complex carbohydrate. DNL 18.6 ± 1.4% vs 11.0 ± 1.4% (P=0.001); liver fat median +137% (P=0.016); blunted insulin suppression of EGP (P=0.013). An isocaloric, weight-stable design showing exactly what Chiu’s isocaloric pooling said doesn’t happen. n=8 is the limitation; tracer methodology and inpatient control are the strengths.
Geidl-Flueck 2021, Journal of Hepatology 75(1):46–54, PMID 33684506. (Note: J Hepatol, not Lancet.) n=94 healthy men, double-blind randomized, 7 weeks, four arms: fructose / sucrose / glucose SSB at 80 g/day in addition to usual diet, or abstinence. “Total energy intake was similar across groups.” Basal hepatic fractional secretion rate of newly synthesized fatty acid, median %/day: control 9.1; glucose 11.0 (ns); fructose 19.7 (p=0.013); sucrose 20.8 (p=0.0015).
- ~2-fold increase in hepatic fatty acid synthesis at 80 g/day of fructose OR sucrose, without excess calories, in lean healthy men. Sucrose ≥ pure fructose — ordinary table sugar, not just isolated fructose, drives this.
The reverse experiment — Schwarz 2017, Gastroenterology 153:743. n=41 obese children with habitual fructose >50 g/d; all meals provided 9 days, isocaloric and macronutrient-matched, starch substituted for sugar → final fructose 4% of kcal. Liver fat 7.2% → 3.8% (P<0.001); VAT 123 → 110 cm³ (P<0.001); DNL AUC 68% → 26% (P<0.001). Weight controlled by design.

Synthesis — where the dose matters:

Fructose dose	Evidence
≤50 g/day (≤10% E)	Neutral to beneficial for glycemia (HbA1c −0.38%, n=337). No liver-fat signal
~80 g/day, isocaloric, 7 wk	~2× hepatic DNL (n=94, lean healthy men). Sucrose equally potent
~25% of energy (~170 g/d here), isocaloric, 9 d	DNL 11% → 18.6%; liver fat +137% (n=8)
25% of energy, hypercaloric, 10 wk	VAT increase, DNL increase, dyslipidemia, insulin resistance (Stanhope, §5.5)
+104–250 g/day hypercaloric	+0.53 kg weight; IHCL SMD 0.45; ALT +4.94 U/L

Practical threshold: ~50 g/day fructose is where the evidence flips from neutral/beneficial to lipogenic. Since sucrose is 50% fructose, 50 g fructose ≈ 100 g sucrose ≈ 400 kcal ≈ 14% of [calorie target removed]. An added-sugar ceiling of ~36 g/day puts the adult at ~18 g fructose/day from added sources — comfortably below every threshold, even adding fruit.

5.3 Liquid vs solid — genuinely worse, and here is the size of it

The mechanism trial, and the single most useful number in this section: DiMeglio & Mattes 2000, Int J Obes 24:794. Crossover, n=15, 1,880 kJ/day (~450 kcal) of carbohydrate as soda vs jelly beans, two 4-week periods, 4-week washout.

Solid (jelly beans): dietary energy compensation = 118% — subjects spontaneously ate slightly more than the full load less at other meals. Free-feeding intake fell significantly.
Liquid (soda): compensation = −17% — no decrease in free-feeding intake; total daily energy rose by the full amount of the load.
“Body weight and BMI increased significantly only during the liquid period.” Physical activity and hunger ratings unchanged.

A ~135-percentage-point swing in energy compensation for identical calories. Caveat: n=15, 4 weeks.

Cohorts and trials:

Malik 2010, Diabetes Care 33:2477. T2D: 8 cohorts, 310,819 participants, 15,043 cases, highest (1–2 servings/day) vs lowest → RR 1.26 (1.12–1.41). Metabolic syndrome: 3 studies, 19,431 participants → RR 1.20 (1.02–1.42).
Malik 2013, AJCN 98:1084. Adults, cohorts: 7 studies, n=174,252, 1 serving/day increment = +0.22 kg/y (0.09–0.34). Adults, RCTs: 5 trials, n=292 (small!) → +0.85 kg (0.50–1.20). Children RCTs: 5 trials, n=2,772 → −0.17 (−0.39 to +0.05), ns.
Xi 2014, Atherosclerosis 234:11. SSBs and CHD: 4 prospective studies, 7,396 cases among 173,753. Highest vs lowest RR 1.17 (1.07–1.28); per 1 serving/day RR 1.16 (1.10–1.23).
Counterweight — Khan/Sievenpiper 2023, PLoS One 18:e0264802. 93 reports, 147 trial comparisons, N=5,213, four prespecified energy-control designs. Total fructose-containing sugars had no effect on BP in substitution, subtraction, or ad libitum trials. Food source mattered: fruit and 100% juice at ≤10% E produced small BP decreases; mixed sources including SSBs at high doses increased BP; removing SSBs decreased BP in a linear dose-response gradient.

Honest quantification: the satiety-compensation differential is large and well-characterized (118% vs −17%). The cohort differential is real but modest (RR 1.16–1.26 per serving/day). The adult RCT weight evidence rests on n=292 total. So: liquid sugar is genuinely worse, primarily via failure of energy compensation — mostly an energy-balance effect rather than a unique toxicity — with Geidl-Flueck’s isocaloric DNL doubling as the one verified exception.

5.4 Thresholds — what actually underlies each number

WHO 2015 “Guideline: Sugars intake.” Free sugars <10% E: STRONG. <5% E: CONDITIONAL.

Two commissioned reviews:

Body weight (Te Morenga): RCTs reducing free sugars −0.80 kg (−1.21 to −0.39); RCTs increasing +0.75 kg (0.30 to 1.19); isoenergetic exchange of sugars for other carbohydrate: +0.04 kg (−0.04 to 0.13) — NULL. GRADE moderate. Read the third line. WHO’s own evidence base says the body-weight effect of sugar is entirely an energy effect. The <10% recommendation is a calorie-displacement recommendation dressed as a sugar recommendation.
Dental caries (Moynihan): 8 cohorts; 5 permitted the <10% vs >10% comparison and all 5 reported higher caries above 10%. The <5% threshold rests on three national/ecological population studies conducted in Japan, showing lower caries at <10 kg free sugars/person/year with a log-linear gradient extending below 5% E. So the <5% conditional recommendation is essentially a DENTAL CARIES recommendation from ecological, largely pre-fluoridation-era data. It is the weakest-evidenced number in mainstream sugar guidance and should not be treated as a cardiometabolic target.

AHA 2009 — Johnson RK et al., Circulation 120:1011. Verbatim derivation: “A prudent upper limit of intake is half of the discretionary calorie allowance, which for most American women is no more than 100 calories per day and for most American men is no more than 150 calories per day from added sugars.” This is almost universally misunderstood: the AHA limit is not derived from any dose-response relationship with a hard outcome. It comes from the 2005 DGA discretionary calorie allowance — a nutrient-adequacy budgeting construct — then halved as a “prudent” judgment. The statement itself concedes “although trial data are limited.” At 4 kcal/g: 100 kcal = 25 g (women), 150 kcal = 37.5 g (men); AHA consumer materials round to 6 tsp / 25 g and 9 tsp / 36 g. (The 150 kcal derivation is verified from Circulation; the exact “36 g” consumer figure is the standard arithmetic rounding, not separately verified.)

Yang 2014, JAMA Intern Med 174:516 (2014, not 2015). NHANES III Linked Mortality, n=11,733; 831 CVD deaths over 163,039 person-years, median 14.6 y. Fully adjusted HRs across quintiles of % calories from added sugar: Q2 1.07 (1.02–1.12); Q3 1.18 (1.06–1.31); Q4 1.38 (1.11–1.70); Q5 2.03 (1.26–3.27), P=0.004. By threshold vs <10% of calories: 10.0–24.9% → HR 1.30 (1.09–1.55); ≥25% → HR 2.75 (1.40–5.42). Consistent across strata except non-Hispanic blacks. Note the CIs widen dramatically at the top, reflecting few events.

Threshold convergence at [calorie target removed]:

Source	Rule	At [calorie target removed]
WHO strong (<10% E)	free sugars	71 g/day
Yang 2014 inflection	<10% E added sugar	71 g/day
AHA 2009 (men)	150 kcal	37.5 g/day = 5.3% E
WHO conditional (<5% E)	free sugars	36 g/day

AHA’s men’s limit and WHO’s <5% coincide at ~36–37 g/day at the adult’s energy intake. That is the defensible working ceiling — not because either derivation is strong, but because two independently-derived numbers converge there and it sits comfortably below every fructose threshold in §5.2.

5.5 Sugar and visceral vs subcutaneous fat — the single most goal-relevant finding in this document

Stanhope KL et al., J Clin Invest 2009;119:1322–1334. PMID 19381015. n=32 completers (fructose 17, glucose 15), age 40–72, overweight/obese. 2-wk inpatient baseline → 8-wk outpatient beverages at 25% of energy requirements (fructose or glucose) → 2-wk inpatient. CT imaging.

Verbatim: “Total and visceral adipose tissue (VAT) volumes were not significantly changed in subjects consuming glucose; however, subcutaneous adipose tissue (SAT) volume was significantly increased. In contrast, both total abdominal fat and VAT volume were significantly increased in subjects consuming fructose.”

Both groups gained similar weight. Same calories, same weight gain, different fat depot. That is the cleanest human demonstration that sugar type determines where fat is deposited.

DNL: glucose group unchanged (fasting 8.8→9.5%, P=0.47; postprandial 13.4→14.2%, P=0.31). Fructose group postprandial 11.4 ± 1.3% → 16.9 ± 1.4% (P=0.021).

The sex interaction — the highest-leverage finding here for this specific person: “The total and percentage increases of fat mass (men +4.4% ± 0.8%; women +1.5% ± 0.7%; P=0.020) and intraabdominal fat volume (men +18.1% ± 5.1%; women −0.6% ± 4.4%; P=0.049) were greater in men than in women. Men consuming fructose also had larger increases of intraabdominal fat compared with women consuming fructose (P=0.033).”

Men gained 18.1% intra-abdominal fat; women gained essentially none. Caveats: n=32 total (fructose arm n=17), participants were 40–72 and overweight/obese (not lean young adults), and the dose was 25% of energy ≈ 178 g/day at [calorie target removed], far above realistic intake. The direction and depot-specificity transfer; the magnitude does not. (The commonly quoted “+14% vs +3%” VAT figures are in Figure 1B; only direction, significance and n were verified.)

Corroboration: Maersk 2012, AJCN 95:283 — n=47 overweight, randomized to 1 L/day for 6 months of regular cola / isocaloric semi-skim milk / aspartame diet cola / water, ¹H-MRS + MRI: liver fat relative increase 132–143% higher in the regular-cola group than the other three; visceral fat also increased in the cola arm. And Schwarz 2017 in reverse: 9 days of isocaloric fructose restriction, VAT 123 → 110 cm³, liver fat 7.2% → 3.8%, weight controlled.

6. SATURATED FAT

6.1 The honest current state

Hooper et al., Cochrane 2020 (CD011737.pub3) — 15 RCTs (16 comparisons), 56,675 participants, minimum 24 months:

Outcome	RR (95% CI)	Trials / n
Combined cardiovascular events	0.83 (0.70–0.98)	12 / 53,758 (8% event rate)
All-cause mortality	0.96 (0.90–1.03)	11 / 55,858
Cardiovascular mortality	0.95 (0.80–1.12)	10 / 53,421
Non-fatal MI	0.97 (0.87–1.07)	—
CHD mortality	0.97 (0.82–1.16)	—

NNTB = 56 (primary prevention, ~4 y), 53 (secondary prevention).

The key caveat, verbatim: “Subgrouping did not suggest significant differences between replacement of saturated fat calories with polyunsaturated fat or carbohydrate, and data on replacement with monounsaturated fat and protein was very limited.” Cochrane could not resolve the replacement question — which is the whole question.

Honest reading: reducing SFA for ≥2 years buys a 17% relative reduction in CV events with no detectable mortality signal. Real but modest, and the trials are old, heterogeneous, and mostly pre-statin.

The Ramsden re-analyses.

Minnesota Coronary Experiment (BMJ 2016;353:i1246) — n=9,423 institutionalized adults, double-blind, corn oil/corn-oil margarine vs control, 1968–73, recovered from magnetic tapes. Serum cholesterol fell −13.8% vs −1.0%. Yet each 30 mg/dL reduction in serum cholesterol was associated with a 22% HIGHER risk of death (HR 1.22, 1.14–1.32; P<0.001). No mortality benefit in any prespecified subgroup. Accompanying MA of 5 RCTs: CHD mortality 1.13 (0.83–1.54).
Sydney Diet Heart Study (BMJ 2013) — n=458 men aged 30–59 with recent coronary events, safflower oil (nearly pure linoleic acid, essentially zero omega-3) replacing animal fat. All-cause mortality 17.6% vs 11.8%, HR 1.62 (1.00–2.64); CVD 1.70 (1.03–2.80); CHD 1.74 (1.04–2.92).

Legitimate criticisms of Ramsden: MCE had massive attrition from institutional turnover, so most randomized subjects had well under a year of exposure (the frequently-quoted “>83% loss to follow-up” is LOW CONFIDENCE — search snippet only); the MCE margarine used era-appropriate hydrogenation and plausibly delivered substantial trans fat, confounding “linoleic acid” with a known harm; Sydney Diet Heart used a linoleic-acid isolate with zero n-3, not representative of “replace SFA with unsaturated fat” as anyone practices it; both are post-hoc resurrections of decades-old data with non-prespecified analyses. Net: Ramsden legitimately demolishes the claim that lowering cholesterol via linoleic acid alone has proven mortality benefit. It does not demolish LDL causality, and it does not license high SFA intake.

6.2 PURE — what it found and how much to weight it

Dehghan et al., Lancet 2017;390:2050. n=135,335, 18 countries, median follow-up 7.4 y, 5,796 deaths, 4,784 major CVD events. Q5 vs Q1 total mortality: carbohydrate HR 1.28 (1.12–1.46); total fat 0.77 (0.67–0.87); saturated fat 0.86 (0.76–0.99); MUFA 0.81; PUFA 0.80. SFA and stroke HR 0.79 (0.64–0.98) — inverse. Critically: “Total fat and saturated and unsaturated fats were not significantly associated with risk of myocardial infarction or cardiovascular disease mortality.”

Valid criticisms:

The contrast is not “Western high-fat vs Western low-fat.” The highest-carbohydrate quintile averaged ~77% of calories from carbohydrate — a subsistence white-rice/maize pattern. The comparison being made is malnutrition vs adequacy, not steak vs oatmeal. (The 77% figure is LOW CONFIDENCE, from secondary coverage.)
Confounding by poverty. High carbohydrate % proxies for poverty — least medical access, highest infectious/trauma death. Authors adjusted for five SES markers and the association persisted, which blunts but does not eliminate the objection: residual confounding is essentially guaranteed when the exposure gradient is the development gradient.
FFQ validity across 18 countries spanning Sweden to Bangladesh cannot be assumed comparable. Differential misclassification across the exposure range is the biggest technical weakness.
Carbohydrate quality is uncollapsed — “carbohydrate” pools kale and cake.
Total mortality, not CVD, drove the headline — the result most consistent with a nutritional-adequacy story rather than an atherogenesis story.

How to weight it: PURE is strong evidence that very high carbohydrate, low total fat diets in low-income settings predict higher mortality. It is weak evidence about whether a well-fed adult American should eat 30 g or 15 g of SFA per day. It does not overturn the RCT/genetic evidence on LDL because it found no association with MI or CVD death in either direction — a null on the mechanism-relevant endpoint is not a refutation of the mechanism.

6.3 Replacement framing — the only framing that means anything

Cohort substitution — Li et al., JACC 2015. NHS (84,628 women) + HPFS (42,908 men), 7,667 incident CHD cases. Replacing 5% of energy from SFA with:

Replacement	HR (95% CI)
PUFA	0.75 (0.67–0.84)
MUFA	0.85 (0.74–0.97)
Whole-grain carbohydrate	0.91 (0.85–0.98)
Refined starch / added sugar	not significantly associated

RCT meta-analysis — Mozaffarian, Micha & Wallace, PLoS Med 2010. 8 RCTs, 13,614 participants, 1,042 CHD events. Increasing PUFA in place of SFA: RR 0.81 (0.70–0.95), p=0.008. Meta-regression: 10% lower CHD risk per 5% energy increase in PUFA — RR 0.90 (0.83–0.97). Study duration was an independent determinant (p=0.017): longer = larger.

AHA Presidential Advisory (Sacks 2017, Circulation): core trials replacing SFA with polyunsaturated vegetable oil reduced CVD by ≈30%, similar in magnitude to statin treatment. Core-trial criteria required ≥2 years, controlled intake in both arms, biomarker-verified adherence, and exclusion of trials where trans fat was a major component — precisely the filter that separates Sacks’s conclusion from Ramsden’s.

Consolidated, per 5% energy substitution: SFA→PUFA ≈ 20–25% lower CHD; SFA→MUFA ≈ 15%; SFA→whole grains ≈ 9%; SFA→refined carbs/sugar ≈ 0%. At [calorie target removed], 5% energy = ~16 g of fat shifted.

This is the load-bearing conclusion of the whole SFA section. The question is never “is saturated fat bad.” It is “what goes in its place.” Swapping butter for white bread buys nothing. Swapping butter for olive oil, walnuts, or fatty fish buys most of the available effect.

6.4 LDL / ApoB mechanism, and reconciling it with “SFA isn’t bad”

Mensink 2016 WHO meta-regression, 84 RCTs: replacing 1% of total daily calories from SFA with PUFA, MUFA, or carbohydrate lowers LDL-C by 2.1, 1.6, and 1.3 mg/dL respectively. Individual SFAs differ: lauric (12:0), myristic (14:0) and palmitic (16:0) raise TC/LDL/HDL vs carbohydrate; stearic acid (18:0) does not — relevant because stearic is ~⅓ of cocoa butter and a large share of beef fat. (Coefficients retrieved via Endotext NBK326737 quoting Mensink 2016, not the WHO primary; the older Mensink & Katan 60-trial coefficient is LOW CONFIDENCE.)

Independent confirmation — RISSCI-1, n=109 healthy males (mean age 48, BMI 25.1), sequential dietary intervention reducing SFA from 19.1% → 8.9% of energy (−10.2 %E): LDL-C −0.50 mmol/L (−0.58 to −0.42) ≈ −19.3 mg/dL, ~−15.7% — i.e. ~1.9 mg/dL per 1% energy, closely matching Mensink’s PUFA coefficient of 2.1. Two independent methods converging is the strongest thing in this section.

Is LDL→ASCVD causality settled? Yes. Ference et al., EAS Consensus Panel, Eur Heart J 2017 synthesized >200 prospective cohorts, Mendelian randomization studies, and RCTs — >2 million participants, >20 million person-years, >150,000 cardiovascular events — finding “a remarkably consistent dose-dependent log-linear association between the absolute magnitude of exposure of the vasculature to LDL-C and the risk of ASCVD; and this effect appears to increase with increasing duration of exposure.”

Reconciling the two literatures — this is the crux and most public argument collapses it:

MR studies estimate the effect of lifelong LDL exposure. Diet RCTs measure 2–5 years of modest LDL change in middle-aged/elderly people who already have decades of plaque. A 12 mg/dL LDL difference over 4 years is a tiny cumulative-exposure delta; failing to detect a mortality effect in that design is exactly what causality predicts. Absence of a trial signal is not evidence of absence of a mechanism.
SFA is a weak lever on a strong causal pathway. Both statements are true simultaneously.
For a adult, the cumulative-exposure framing is the one that matters. Ference’s “duration of exposure” term is the entire reason this is worth any attention at the adult’s age — the adult has ~50 years of area-under-the-LDL-curve ahead, which is where diet compounds.

6.5 Dairy fat vs meat fat — real at the food level, much weaker at the molecule level

MESA (de Oliveira Otto 2012, AJCN) — prospective cohort, n=5,209, ages 45–84, 316 CVD events:

Exposure	HR (95% CI)
Dairy SFA, +5 g/d	0.79 (0.68–0.92)
Dairy SFA, +5% energy	0.62 (0.47–0.82)
Meat SFA, +5 g/d	1.26 (1.02–1.54)
Substituting 2% energy meat SFA → dairy SFA	0.75 (0.63–0.91)

Biomarker evidence — Imamura 2018, PLoS Med. Pooled analysis of 16 prospective cohorts, 12 countries, 63,682 participants, 15,180 incident T2D cases, HRs (10th→90th percentile): 15:0 0.80 (0.73–0.87); 17:0 0.65 (0.59–0.72); t16:1n-7 0.82 (0.70–0.96); sum 0.71 (0.63–0.79).

Food-matrix RCTs: cheese vs butter at matched fat content lowers LDL-C — pooled −0.21 mmol/L (4 RCTs) and −6.5% / −0.22 mmol/L (5 RCTs). Mechanisms proposed: calcium-fatty-acid soap formation reducing absorption, the casein matrix, fermentation. MODERATE CONFIDENCE — from secondary coverage of the meta-analyses.

Is the divergence real or confounded? Partly confounded — and there is a specific, damaging problem with the biomarker story.

Odd-chain fatty acids are not clean dairy biomarkers. 15:0 and 17:0 are also synthesized endogenously from gut-derived propionate. Weitkunat 2017, AJCN: inulin and propionate raised 15:0 by ~17% and ~13%, and 17:0 by ~11% and ~13%, in nearly all participants. This means the “dairy fat → lower T2D” biomarker literature is partially a FIBER literature wearing a dairy costume. The HR 0.65 for 17:0 is very likely picking up fermentable-fiber intake — one of the most robustly protective dietary exposures there is.

Additional confounding: dairy-SFA-heavy patterns in US cohorts are yogurt/milk/cheese patterns correlated with lower processed-meat intake, higher SES, and non-smoking; meat SFA travels with processed meat, sodium, nitrates, and heme iron. MESA’s substitution model cannot separate “dairy fat is benign” from “processed meat is harmful.”

Verdict: the FOOD-level divergence (fermented dairy and cheese behave better than processed meat) is real and reproducible. The FATTY-ACID-level claim (“dairy SFA molecules are metabolically different”) is substantially weaker than usually presented.

6.6 Is SFA relevant to THE ADULT’S goals?

Goal	SFA relevance	Evidence
Long-run ASCVD	Moderate — the only place it genuinely matters	Ference 2017 (cumulative exposure); Li 2015; Mensink. In early adulthood the duration term dominates
Visceral / liver fat	Small but real, better-evidenced than expected	LIPOGAIN (Rosqvist 2014, Diabetes) — double-blind RCT, n=39 young normal-weight adults, 7 weeks of muffin overfeeding, palm oil (SFA) vs sunflower oil (n-6 PUFA). *Equal weight gain, but SFA “markedly increased liver fat” and caused a twofold larger increase in visceral adipose tissue; PUFA caused a nearly threefold larger increase in lean tissue. Caveat: this is an OVERFEEDING model at n=39. In energy balance the effect is much smaller*
VO2max	Essentially zero for SFA specifically	The relevant fat finding is about total fat/carb ratio, not fat type — see §10
Back/joint pain	No credible direct evidence	No RCT shows SFA composition changes joint or back pain in non-arthritic adults (§9)

Ranking for the adult’s goals: VO2max (carb availability) > visceral fat (total energy balance ≫ fat type) > joints (SFA: no evidence) — with long-run ASCVD as the one genuine reason to care, and it’s a 30-year play, not a this-quarter play.

6.7 Normalizer and source differentiation

Anchors: AHA target <6% energy; DGA/WHO <10% energy. At [calorie target removed]:

Threshold	% energy	g SFA / 1,000 kcal	g/day	g per ~950 kcal meal
Aggressive (AHA)	6%	6.7	19 g	6.3 g
Moderate	7%	7.8	22 g	7.4 g
Standard ceiling	10%	11.1	32 g	10.6 g

Use g/1,000 kcal — portion-size invariant. Raw grams per item will mis-rank a 300 kcal snack against a 1,100 kcal dinner.

Source differentiation: yes, but modestly — no more than one band’s worth. In descending order of evidential support:

Adjust by whole food, not fatty-acid profile. The reproducible signal is at the food level: fermented dairy/cheese/yogurt ≥ unprocessed meat > processed meat (MESA substitution HR 0.75).
Half-credit for cheese, yogurt and fermented dairy SFA (matrix RCTs). Do NOT extend this to butter or cream — the matrix effect is specifically about cheese structure.
No credit for “it’s dairy fat, therefore the molecules are fine.” The odd-chain biomarker basis is contaminated by fiber-derived endogenous synthesis.
Full penalty for processed-meat SFA.
(Weak, optional) discount stearic acid (18:0), LDL-neutral vs carbohydrate. Practically this means dark chocolate and to a lesser degree beef fat score slightly better than their total-SFA number implies. Low priority — the measurement burden isn’t worth it in a meal scorer.
Always score the replacement. A meal red for SFA because of a large fatty-fish or nut portion is not the same as one red from processed meat. If the scorer can carry only one extra dimension, carry the (PUFA+MUFA):SFA ratio rather than SFA alone — that’s the axis the substitution evidence actually measures.

7. OMEGA-3 (EPA/DHA)

7.1 Dose-response

Triglycerides — AbuMweis et al. MA (J Hum Nutr Diet), 171 placebo-controlled RCTs of acceptable quality (Jadad ≥3):

Outcome	Pooled effect (95% CI)
Triglycerides	−0.368 mmol/L (−0.427 to −0.309) ≈ −32.6 mg/dL
LDL-C	+0.150 mmol/L (0.058 to 0.243) ≈ +5.8 mg/dL
HDL-C	+0.039 mmol/L (0.024 to 0.054)
Systolic BP	−2.195 mmHg (−3.172 to −1.217)
Diastolic BP	−1.08 mmHg (−1.716 to −0.444)
CRP	−0.343 mg/L

The TG effect is explicitly dose-dependent and reaches ≥30% at 4 g/d in hypertriglyceridemic patients (Skulas-Ray, AHA Science Advisory, Circulation 2019). No clean published mg/dL-per-gram coefficient could be retrieved — do not quote one.

The LDL trade-off is the underrated finding here. EPA+DHA supplements raise LDL-C (+5.8 mg/dL pooled). Skulas-Ray 2019 confirms 4 g/d EPA+DHA agents produce concurrent LDL increases, whereas EPA-only did not raise LDL-C. A real reason to prefer EPA-predominant formulations if dosing high, and a real reason not to megadose “for general health.”

Inflammation markers — heterogeneous, and I will not pretend otherwise:

Umbrella review of 8 MAs (Abbasi 2025): TNF-α SMD −0.34 (−0.56 to −0.11); IL-6 −0.30 (−0.49 to −0.12); CRP MD −5.46 (−10.06 to −0.87).
Huang 2026, Front Nutr (9 RCTs, 504 participants): no significant effects on IL-6, CRP, or TNF-α (p>0.05).
AbuMweis pooled CRP: −0.343 mg/L — statistically present, clinically trivial in someone whose CRP is already ~1 mg/L.

Honest summary: effects on inflammatory markers are small, inconsistent across pooling strategies, and largest where baseline inflammation is elevated. In a lean adult with normal CRP, the expected change is close to nothing.

7.2 REDUCE-IT vs STRENGTH — the mineral oil problem

REDUCE-IT (Bhatt 2019, NEJM) — n=8,179, statin-treated, elevated TG, icosapent ethyl 4 g/d vs MINERAL OIL placebo: primary composite 17.2% vs 22.0%, HR 0.75 (0.68–0.83) — ARR 4.8 pp, NNT ≈ 21; CV death HR 0.80 (0.66–0.98); AF/flutter hospitalization 3.1% vs 2.1% (P=0.004); serious bleeding 2.7% vs 2.1% (P=0.06).

STRENGTH (Nissen 2020, JAMA) — n=13,078, omega-3 carboxylic acid (EPA+DHA) 4 g/d vs corn oil: HR 0.99 (0.90–1.09). Stopped early for futility. AF increased ~69% (2.2% vs 1.3%).

Four candidate explanations, ranked by weight:

(a) The comparator — the strongest single explanation. Ridker 2022, Circulation, biomarkers in the REDUCE-IT arms, median percent increases from baseline at 12 months in the MINERAL OIL group: interleukin-1β +28.9%; hsCRP +21.9%; Lp-PLA2 +18.5%; IL-6 +16.2%; oxidized LDL +10.9%; Lp(a) +2.2%; homocysteine +1.5%. “In the icosapent ethyl group, there were minimal changes.” Authors’ own conclusion: “The effect of these findings on interpretation of the overall risk reductions… is uncertain.” A placebo arm in which every atherogenic and inflammatory marker moved the wrong way over a year. It does not prove the placebo caused harm, but it makes “part of the 25% relative reduction is placebo-arm deterioration rather than drug benefit” the default hypothesis rather than a fringe one. (A REDUCE-IT placebo-arm LDL-C/apoB rise is UNVERIFIED RECALL — Ridker’s retrieved abstract did not report those.)

(b) EPA-only vs EPA+DHA — plausible on paper, but the best test of it FAILED. STRENGTH secondary analysis (Nissen 2021, JAMA Cardiol), n=10,382 with baseline and 12-month levels: adjusted HRs for the highest achieved tertile vs corn oil: EPA 0.98 (0.83–1.16), P=.81; DHA 1.02 (0.86–1.20), P=.85. Conclusion: “the highest achieved tertiles of EPA and DHA were associated with neither benefit nor harm.” If EPA level were the mechanism, this analysis should have detected it. It didn’t. That substantially weakens (b) and correspondingly strengthens (a).

(c) JELIS as supporting context — Yokoyama 2007, Lancet: n=18,645 hypercholesterolemic Japanese, 1,800 mg EPA + statin vs statin alone, mean 4.6 y — major coronary events 2.8% vs 3.5%, 19% relative reduction (p=0.011). Supports EPA at a much lower dose, but open-label with no placebo at all, in a very-high-fish-intake population.

(d) Population/formulation differences — real but small.

Best current read: REDUCE-IT is a genuinely positive trial whose effect size is probably inflated by the mineral oil comparator, and the magnitude of that inflation remains unresolved. Given STRENGTH’s flat result and its null achieved-level analysis, “4 g/d omega-3 reduces CV events” should NOT be treated as established for the general population.

The 1 g/d trials are null-ish:

VITAL (Manson 2019, NEJM) — n=25,871, 1 g/d marine n-3, median 5.3 y: major CV events HR 0.92 (0.80–1.06), P=0.24 — null; invasive cancer 1.03 (0.93–1.13) — null; total MI HR 0.72 (0.59–0.90) — a real secondary signal; stroke 1.04; CV death 0.96. (The widely-cited low-fish-intake and Black-participant subgroups are UNVERIFIED RECALL and are subgroups of a null primary regardless.)
ASCEND (NEJM 2018) — n=15,480 diabetics without ASCVD, 1 g/d vs olive oil, mean 7.4 y: serious vascular events 8.9% vs 9.2%, RR 0.97 (0.87–1.08) — flatly null.
Meta-analytic middle ground — Hu 2019, JAHA, 13 RCTs, 127,477 participants (excluding REDUCE-IT): MI 0.92 (0.86–0.99); CHD death 0.92 (0.86–0.98); total CHD 0.95; CVD death 0.93; total CVD 0.97. Statistically significant linear dose-response. So: small, real, dose-related — roughly a 5–8% relative reduction on hard coronary endpoints, not the 25% of REDUCE-IT.

7.3 Dose thresholds with real evidence

Purpose	Evidence-backed dose	Quality
Triglyceride lowering	2–4 g/d EPA+DHA; ≥30% TG reduction at 4 g/d	Strong. Note LDL rises with EPA+DHA formulations but not EPA-only
General CV	~1 g/d for the small MI/CHD-death effect; 4 g/d only in high-risk, high-TG, statin-treated — and contested	Weak-to-moderate
Inflammation	No defensible threshold	Weak. Pooled CRP change ~−0.34 mg/L; meta-analyses disagree on whether markers move at all

For this person: ~1–2 g/d EPA+DHA, preferentially from 2 servings of oily fish per week. There is no evidence base for a lean, low-CRP adult taking 4 g/d, and there is an AF signal against it.

7.4 The atrial fibrillation signal — the real risk of megadosing

Gencer 2021, Circulation — MA of CV-outcome RCTs, 7 trials, 81,210 patients, 2,905 AF events:

Overall HR 1.25 (1.07–1.46), P=0.013
>1 g/d (n=22,271): HR 1.49 (1.04–2.15), P=0.042
≤1 g/d (n=58,939): HR 1.12 (1.03–1.22), P=0.024

Abuknesha 2026, Circ Arrhythm Electrophysiol — 35 trials, 114,592 individuals: in high-CV-risk patients on >1,500 mg/d EPA/DHA, pooled OR 1.43 (1.14–1.79); low-dose subgroups non-significant. Consistent with trial-level data (REDUCE-IT 3.1% vs 2.1%; STRENGTH 2.2% vs 1.3%).

The uncomfortable detail: even the ≤1 g/d stratum is statistically significant (HR 1.12, 1.03–1.22). Absolute risk in a healthy adult is very low, so this is not a reason to avoid fish or 1 g/d supplements. It IS a reason not to take 4 g/d without a specific indication. The risk-benefit at 4 g/d for someone with no CVD, no hypertriglyceridemia, and a joint complaint is unfavorable.

8. ENERGY DENSITY (kcal/g)

8.1 Hall 2019 — the trial everyone cites, and what it actually shows

Hall KD et al., Cell Metab 2019;30(1):67-77.e3. PMID 31105044. n=20 weight-stable adults, NIH inpatient metabolic ward, 28 days, randomized crossover, two 14-day arms (UPF vs unprocessed), ad libitum, matched for presented calories, macronutrients, sugar, sodium and fibre.

Energy intake +508 ± 106 kcal/d on UPF (p=0.0001). Weight +0.9 ± 0.3 kg (UPF) vs −0.9 ± 0.3 kg (unprocessed).

The energy-density confound is in the paper itself, not a critic’s invention:

ED of foods and beverages consumed: 1.36 ± 0.02 vs 1.09 ± 0.02 kcal/g (p<0.0001).
ED of presented non-beverage foods: 1.957 vs 1.057 kcal/g — ~85% higher on the UPF arm.
The gap between those two numbers is the whole story: fiber-supplemented beverages on the UPF arm diluted the whole-diet ED metric and hid an 85% difference in the solid food. Hall’s own text notes “beverages have limited ability to affect satiety.”

Eating rate (Forde, Mars & de Graaf, Curr Dev Nutr 2020): UPF 37 g/min vs 30 g/min; energy intake rate 48 vs 31 kcal/min (>50% higher).

Fazzino, Courville, Guo & Hall, Nature Food 2023;4:144. 2,733 meals from 35 inpatient participants across two feeding studies and four dietary patterns. Energy density, eating rate, and hyper-palatable-food share were each positively and consistently related to meal energy intake across all four diets. Protein was positively related in two of four patterns — the opposite sign to protein leverage at the meal level. (Exact regression coefficients are paywalled — direction and n verified only.)

8.2 Dicken 2025 (UPDATE trial) — the follow-up

Dicken SJ et al., Nat Med 2025. 2×2 crossover feeding RCT, n=55 randomized, 50 ITT, 43 per-protocol. Adults in England, BMI 25–<40, habitual UPF ≥50% kcal/d. Two 8-week ad libitum diets, both built to the UK Eatwell Guide, 4-week washout.

Weight: MPF −2.06% (−2.99, −1.13); UPF −1.05% (−1.98, −0.13). Difference −1.01% (−1.87, −0.14), P=0.024, Cohen’s d = −0.48. Absolute −1.84 vs −0.88 kg.
Fat mass −1.59 vs −0.61 kg (difference −0.98 kg, P=0.004). Waist −1.70 vs −0.18 cm, NS (P=0.148).
Self-reported energy intake −327.3 kcal/d on MPF.
Energy density: MPF 1.25 vs UPF 1.60 kcal/g. Again not matched.

Published criticism (both Nat Med 2026): Ludwig, Willett & Putt (PMID 41495404) and Robinson & Forde (PMID 41495406). The Forde/Robinson thrust is exactly the ED point: a 0.35 kcal/g ED gap plus texture differences is a sufficient explanation without invoking “processing” as a causal category.

8.3 The Rolls-lineage manipulation trials — the actual causal evidence

Manipulation	Result
Soup as a first course	~20% reduction in total lunch energy intake
Same water served in the food (soup) vs alongside a casserole	26% less energy at lunch
ED reduced ~30% over 4 days, macros held constant	~30% less cumulative energy intake, same weight of food eaten
ED reduced 25% over 2 days	−24% energy intake (≈575 kcal/d) `[PARTIAL — secondary source]`
Ello-Martin 2007, 12 mo, n=97 obese women	F&V (low-ED) arm lost 17.4 lb vs portion-restriction 14.1 lb
PREMIER secondary analysis, n=658, 6 mo	Large ED reduction 13 lb / modest 9 lb / slight-or-none 5 lb

The recurring signature: people eat a roughly constant WEIGHT of food, so kcal/g maps almost linearly onto kcal. That is the mechanistically important finding and it is replicated many times.

8.4 What the best evidence isolates

Energy density — strongest. The only mechanism with direct, repeated, isocaloric-macro manipulation trials showing dose-dependent intake changes (20–30% intake swings for 25–30% ED swings). It also fully accounts for the direction of both Hall 2019 and Dicken 2025, neither of which matched ED.
Eating rate — strong, and largely collinear with ED. 48 vs 31 kcal/min in Hall; independently predictive across 2,733 meals in Fazzino. But in real foods ED and eating rate co-vary heavily, so “which one” is partly unanswerable from observational meal data.
Hyper-palatability — real but smallest and worst-defined. The HPF definition is itself a nutrient-combination rule (fat+sodium / fat+sugar / carb+sodium), which correlates with ED by construction.
Protein dilution — real but only at the low end, and asymmetric.

Verdict: energy density and eating rate are the same lever viewed from two sides, and together they are the dominant one. “Ultra-processing” is not an independent mechanism in any trial run to date, because no trial has matched ED between arms.

8.5 Protein leverage

Gosby 2011, PLoS ONE 6:e25929. n=22 lean healthy adults, three separate 4-day ad libitum in-house periods at 10%, 15%, 25% protein, composition disguised.

Total energy over 4 days: 41.45 ± 2.43 MJ (10%), 37.11 ± 2.08 (15%), 37.07 ± 2.16 (25%).
10% vs 15%: +4.34 MJ over 4 days = +12% energy intake (P<0.0001). 15% vs 25%: no difference (P=1.0).
Below the 15% target, every 1 kJ decrease in protein → +4.5 kJ non-protein intake; above it, +1 kJ protein → −1 kJ non-protein (4.5× asymmetry).

How much evidence is this really? Modest. One 22-person, 4-day-per-arm crossover is the flagship human experiment, and Fazzino 2023 found protein positively associated with meal energy intake in two of four patterns.

Honest reading: protein leverage is a FLOOR effect, not a dial. Below ~15% of energy from protein it reliably drives overconsumption; 15% → 25% buys nothing on intake. For this person it is almost certainly already satisfied — 15% of [calorie target removed] = 107 g/d, and the adult will be well above that. Protein leverage is not an available lever for the adult; the adult is on the flat part of the curve.

8.6 The beverage problem, and concrete kcal/g bands

Ledikwe et al., J Nutr 2005;135:273. PMID 15671225. Eight ED calculation methods applied to CSFII 1994-96: mean daily dietary ED ranged from 0.94 to 1.85 kcal/g depending purely on which beverages you include. A ~2× swing from a methodological choice. [PARTIAL — abstract via search index]

Recommendation: compute ED on FOODS ONLY (exclude ALL beverages, including water, milk, juice, soda) and score caloric beverages as a separate penalty term. Justification, in order of strength:

Hall 2019 is the proof case: including beverages compressed a 1.96 vs 1.06 kcal/g food difference down to 1.36 vs 1.09. The metric that mattered was the food-only one.
Including water/diet drinks makes ED trivially gameable — drink a liter of water with a meal, “improve” the score, change nothing physiologically.
Liquid calories are poorly compensated for, so the satiety mechanism is different; pooling them into a satiety-proxy metric is a category error.

Banding — Rolls’s 4-category system (Rolls BJ, Nutrition Bulletin 2017;42:246, DOI 10.1111/nbu.12280; also used in CDC materials). [Cutoffs verified via multiple secondary sources citing Rolls 2017; primary PDF not fetched.]

Band	kcal/g	Typical foods
Very low	< 0.6	Non-starchy vegetables, broth soups, most fruit, skim milk
Low	0.6 – 1.5	Starchy vegetables, cooked grains, legumes, lean meat/fish, low-fat mixed dishes
Medium	1.5 – 4.0	Most meat, bread, cheese, dense mixed dishes, dried fruit
High	4.0 – 9.0	Nuts, oils, butter, chips, crackers, cookies, most snack foods

If a 5-band scheme is wanted, split “high” at ~5.5 kcal/g — but that split has no published basis; Rolls’s system is 4 categories. Do not present a 5th band as sourced.

Calibration for this person: at [calorie target removed]/d, a food-only ED of 1.0 kcal/g means eating [calorie target removed] g of food/day — practically very hard. At 1.5 kcal/g it’s 1,900 g; at 2.0 it’s 1,425 g. Prepared/packaged entrées typically land 1.5–2.0 kcal/g. So the realistic target is not “very low ED” — it is pulling the meal average from ~1.8 down toward ~1.2–1.4 by volume-adding vegetables, which is exactly the manipulation the 20–30% trials used. Note the scoring-band choice below is therefore centred on the 1.0–2.5 range where the adult’s meals actually live, not on Rolls’s whole-food-supply categories.

9. MICRONUTRIENTS

9.1 NHANES prevalence of inadequacy — the actual numbers

Reider et al., Nutrients 2020;12:1735. NHANES 2005–2016, n=26,282 adults ≥19 y, two 24-h recalls, usual-intake modelling:

Nutrient	% below EAR, food only	% below EAR, food + supplements
Vitamin D	95%	65%
Vitamin E	84%	60%
Vitamin C	46%	33%
Vitamin A	45%	35%
Zinc	15%	11%
Folate	12%	—
Vitamin B6	11%	—
Copper	6%	—
Iron	5%	—
Selenium	<1%	—

(This paper does not report sex-stratified numbers and does not cover magnesium, calcium, or B12.)

Nutrients with an AI rather than an EAR — where “% below” means much less:

Nutrient	% below AI	Source
Potassium	~100% vs the old 4,700 mg AI; ~75% of men <3,400 mg	FSRG Dietary Data Brief (NHANES 2017-18)
Choline	89.2% overall; 84.4% of males fail the AI; only 6.6 ± 0.5% of adults ≥19 y meet it	Wallace & Fulgoni 2016, J Am Coll Nutr 35:108, NHANES 2009-2012, n=16,809
Vitamin K	~67%	`[PARTIAL — secondary source]`
Magnesium	~48–55% of US adults below EAR; ~57% of men	`[PARTIAL — secondary summaries citing NHANES; NIH ODS returned 403]`

9.2 The five deep-dives: real vs over-hyped

MAGNESIUM — partly real, but the EAR itself is soft.

Is the EAR trustworthy? This is the key skeptical point, and it cuts both ways. Rosanoff, Adv Nutr 2021;12:298: the 1997 Mg DRIs were derived from balance studies using standard reference body weights below current US mean weights, and assumed a 10% CV where balance data suggest 20–30%. Body-weight correction alone raises the men’s EAR 17% (330-350 → 386-409 mg/d); RDA recalculations range 463 to 654 mg/d. So “50% inadequate” rests on a requirement estimate contested in both directions. And serum magnesium is a poor status biomarker (only ~1% of body Mg is extracellular), so there is no good population estimate of true functional deficiency.
Blood pressure — real but small. Zhang 2016, Hypertension 68:324: 34 double-blind placebo-controlled trials, n=2,028, median dose 368 mg/d, median 3 months. SBP −2.00 mmHg (0.43–3.58); DBP −1.78 (0.73–2.82). Clinically trivial in a normotensive active adult.
Muscle cramps — NULL, at high certainty. Garrison 2020, Cochrane CD009402.pub3: 11 trials, 735 participants. Idiopathic cramps: % change in cramps/week at 4 wk MD −9.59% (−23.14 to +3.97); cramps/week MD −0.18 (−0.84 to 0.49); responders (≥25% reduction) RR 1.04 (0.84–1.29), HIGH certainty. Conclusion: “unlikely that magnesium supplementation provides clinically meaningful cramp prophylaxis.” Critically: ZERO RCTs exist for exercise-associated muscle cramps. Anyone claiming Mg fixes active adult cramps is extrapolating from nothing.
Sleep — weak. Arab 2023, Biol Trace Elem Res 201:121, 9 studies, 7,582 subjects: observational studies show associations, “the RCTs reported contradictory findings.”

Verdict: mildly real for BP (trivial for the adult), OVER-HYPED for cramps (high-certainty null) and sleep. Get it from food, where it comes bundled with the ED fix. Don’t supplement.

POTASSIUM — mostly an artifact of the old AI. NASEM 2019 found insufficient evidence to set an EAR or RDA for potassium at all and lowered the men’s AI from 4,700 → 3,400 mg/d, stating: “Given the lack of evidence of potassium deficiency in the population, median intakes observed in an apparently healthy group of people are appropriate for establishing the potassium AI values.” That sentence dismantles the “100% of Americans are potassium deficient” talking point — the AI is now DEFINED as the median intake of healthy people, so “below AI” is close to a tautology. There is no biomarker-defined dietary potassium deficiency in healthy US adults with normal renal function.

Verdict: OVER-HYPED as a deficiency. Still worth tracking as a DIETARY-PATTERN PROXY (high potassium tracks vegetables, fruit, legumes, dairy) and for the Na:K ratio (§2) — but not as a “deficiency to correct.”

VITAMIN D — real deficiency is common; supplementation benefit is largely null.

Prevalence: 95% below EAR from food alone. Biomarker: 41.6% of US adults have 25(OH)D ≤20 ng/mL (Forrest & Stuhldreher 2011, NHANES 2005-06, N=4,495; blacks 82.1%, Hispanics 69.2%).
The three mega-trials are null for hard outcomes:
- VITAL (Manson 2019, NEJM 380:33), n=25,871, 2,000 IU/d, median 5.3 y: invasive cancer HR 0.96 (0.88–1.06); major CV event HR 0.97 (0.85–1.12).
- D-Health (Neale 2022, Lancet Diabetes Endocrinol 10:120), n=21,315, 60,000 IU monthly, median 5.7 y, achieved 25(OH)D 115 vs 77 nmol/L: all-cause mortality HR 1.04 (0.93–1.18); CVD mortality 0.96; cancer mortality 1.15 (0.96–1.39), and an exploratory analysis excluding the first 2 y gave cancer mortality HR 1.24 (1.01–1.54), P=0.05 — a signal in the WRONG direction.
- DO-HEALTH (Bischoff-Ferrari 2020, JAMA 324:1855), n=2,157, age ≥70, 2×2×2, 3 y: no significant benefit on any of 6 primary endpoints. SBP with vitamin D: −0.8 mmHg (99% CI −2.1 to 0.5).
- Athletes — also null for performance. Farrokhyar 2017, Sports Med 47:2323: 13 RCTs, 532 athletes. Repletion works biochemically (3,000 IU/d → +15.2 ng/mL (10.7–19.7); 5,000 IU/d → +27.8) but “of 13 included trials, only seven measured different physical performances and NONE demonstrated a significant effect.”
  
  Verdict: deficiency is genuinely prevalent and worth MEASURING; repletion is cheap and safe; but the performance and hard-outcome case is OVER-HYPED and empirically null, including in athletes. “Test, don’t blanket-supplement.”

CHOLINE — under-evidenced in both directions; the clearest case of a paper-only inadequacy.

Only 10.8 ± 0.6% of Americans ≥2 y meet the AI (15.6% of males); 6.6% of adults ≥19 y.
No RDA/EAR exists — only an AI (550 mg/d men), derived from small early-1990s human depletion studies showing liver dysfunction on choline-free diets. (Zeisel’s specific 1991 study n and design: UNVERIFIED RECALL.)
What is verified: “Deficiencies of choline have only been reported in experimental situations or total parenteral nutrition” (Wortmann & Mayr, J Inherit Metab Dis 2019;42:237) — no free-living dietary deficiency syndrome has been described. The NAFLD link is at the level of “studies reporting a relation of low choline levels to subclinical organ dysfunction” — associations, not RCT causation. The same review flags the opposite risk: dietary choline → TMAO → increased atherosclerosis in mice.
Repletion RCTs showing functional benefit in healthy adults: none identified.

Verdict: an AI so high that 89% of the country “fails” it, derived from artificial depletion protocols, with no free-living deficiency syndrome and no benefit RCTs. OVER-HYPED — but genuinely under-studied, so “we don’t know” is the correct posture, not “it’s fine.”

IODINE — no re-emerging US population deficiency; real risk in subgroups.

The trend everyone cites is real but stopped 30 years ago. Caldwell 2005, Thyroid 15:692: median urinary iodine NHANES I (1971-74) 320 ± 6 µg/L → NHANES III (1988-94) 145 ± 3 → NHANES 2001-02 167.8 (159.3–177.6). The drop stabilized; the authors confirm “the current stability of the U.S. iodine intake and continued adequate iodine nutrition for the country.” WHO’s adequacy threshold is median UIC ≥100 µg/L.
Subgroups are where it gets thin. Herrick 2018, Nutrients 10:874, NHANES 2011-2014, n=4,613: women of reproductive age median UIC 110 µg/L — barely above threshold; non-Hispanic Asian WRA 81 µg/L (below). Dairy is the dominant US iodine source.
(The frequently repeated claim that US commercially processed food uses largely non-iodized salt is UNVERIFIED RECALL — could not confirm from a primary US source. It is consistent with the Herrick finding that dairy, not salt, drives US intake.)

Verdict: population-level deficiency is OVER-HYPED. Real risk concentrates in low-dairy + low-iodized-salt eaters and in pregnancy. For a non-pregnant male eating packaged food this is low-probability — and it is genuinely unmeasurable from nutrition data.

9.3 Which micronutrients are worth encoding at all

Nutrient	In USDA FoodData Central?	On US labels?	Encode?
Magnesium	Yes, well-covered	No (voluntary)	Yes — good coverage, real inadequacy, cheap
Potassium	Yes, well-covered	Yes (mandatory since 2016)	Yes, as a dietary-pattern proxy and for Na:K — not as a deficiency metric
Vitamin D	Yes, but sparse (few foods contain it)	Yes (mandatory)	Marginal. Intake is nearly irrelevant vs sun exposure. Handle by blood test
Fiber (total)	Yes	Yes	Yes. Soluble fiber is poorly covered — do not require it
Calcium, iron, zinc, vit C, vit A, folate	Yes	Ca/Fe mandatory	Yes — all cheap and reliable
Vitamin E	Yes	No	Yes, but note the 84%-below-EAR figure is partly an artifact of poor tracking of cooking oils in recalls
Choline	Frequently missing	No	No. Missing data will dominate the signal
Iodine	Largely absent (separate limited DB, not integrated)	No	No. Effectively unmeasurable from meal data
Vitamin K	Yes (phylloquinone)	No	Optional, low priority

Practical rule: encode magnesium, potassium, fiber, calcium, iron, zinc, vitamin C/A/E, folate. Do NOT encode choline or iodine. And do not encode “% below EAR” for potassium or choline as if it were a deficiency — both are AI-based and near-universally “failed” by construction.

10. MEAL TIMING AND DISTRIBUTION

10.1 Protein distribution — see §3.3

Short version: second-order at best. Acute MPS favours even distribution (Mamerow n=8: +25% 24-h FSR, P=0.003), but the flagship chronic RCT (Yasuda n=26) was P=0.06 on its primary outcome, the systematic review says the effect “cannot be sufficiently disentangled from the effect of protein quantity,” and Trommelen 2023 undermines the saturable-per-meal-dose premise the whole doctrine rests on. Meal frequency per se does nothing — Schoenfeld 2015’s positive pooled result was driven entirely by a single study.

10.2 Time-restricted eating — null beyond calorie control

Trial	Design	n	Result
TREAT — Lowe 2020, JAMA Intern Med 180:1491	12-wk RCT, app-based, 16:8 (12:00–20:00) vs 3 structured meals, no calorie prescription	116	TRE −0.94 kg (−1.68, −0.20); CMT −0.68 (−1.41, 0.05); between groups −0.26 kg (−1.30, 0.78), P=.63 — NULL. No difference in estimated energy intake. “TRE, in the absence of other interventions, is not more effective in weight loss than eating throughout the day.”
Liu 2022, NEJM 386:1495	12-month RCT, TRE 08:00–16:00 + CR vs CR alone, both 1500–1800 kcal (men)	139 randomized, 118 completed	TRE −8.0 kg (−9.6, −6.4); CR alone −6.3 kg (−7.8, −4.7); net difference −1.8 kg (−4.0, 0.4), P=0.11 — NULL. Waist, BMI, body fat, lean mass, BP and metabolic risk factors all consistent with the null
Lin 2023, Ann Intern Med 176:885	12-month RCT, 8-h TRE vs 25% CR vs control	90	Energy intake fell −425 ± 531 kcal/d (TRE) vs −405 ± 712 (CR) — TRE reduced calories just as much as counting them. Weight vs control: TRE −4.61 kg; CR −5.42 kg
Jamshed 2022, JAMA Intern Med 182:953	14-wk RCT, early TRE (07:00–15:00) + ER vs ≥12-h window + ER	90	eTRE more effective for weight loss; fat loss between groups P=.09 (NS); DBP −4 mmHg (−8, 0; P=.04). Body-fat benefit appeared only in a secondary analysis of 59 completers
Jin 2024 MA, Crit Rev Food Sci Nutr	SR+MA, RCTs + non-randomized trials	23 studies, 1,867	Claims the weight-loss effect “remained significant” even when energy restriction was applied in both arms. But it pooled non-randomized trials

Honest synthesis: the two best-designed, longest, energy-matched trials (Liu n=139, 12 mo; TREAT n=116) are both NULL for any effect of TRE beyond calorie control. TRE works in the real world because it CUTS CALORIES — Lin 2023 shows an 8-hour window achieves the same −400 kcal/d as explicit 25% calorie counting, without counting. That’s a genuine adherence tool, not a metabolic effect.

The TREAT lean-mass signal, and its criticism. TREAT found appendicular lean mass index differed: −0.16 kg/m² (−0.27, −0.05; P=.005), favouring the NON-TRE group — i.e. TRE lost more lean mass. Four published comment letters plus two errata followed. Fair criticisms: one of many secondary outcomes in a 50-person in-person subcohort (multiplicity, low power); DXA lean-mass changes over 12 weeks with ~1 kg weight change are near the measurement noise floor. Fair defense: the direction is mechanistically plausible (TRE arms typically eat less protein and fewer protein feedings), and Liu 2022 — larger and longer — reported lean mass “consistent with” the null, so it did NOT replicate. Verdict: not established, but not dismissible, and it is the single most relevant risk for someone whose priority is muscle retention.

10.3 Breakfast skipping — the folk wisdom is unsupported

Sievert 2019, BMJ 364:l42. SR+MA of RCTs in high-income countries, 13 trials (7 for weight, 10 for energy intake):

Weight: MD 0.44 kg (0.07, 0.82) favouring those who SKIPPED breakfast (I²=43%).
Energy intake: breakfast eaters consumed 259.79 kcal/day MORE (78.87, 440.71) (I²=80%).
All trials at high or unclear risk of bias in ≥1 domain; mean follow-up 7 weeks (weight), 2 weeks (energy).
Conclusion: “Caution is needed when recommending breakfast for weight loss in adults, as it could have the opposite effect.”

The effect is small, the evidence low-quality, and follow-up short — so read it as “breakfast is not required for weight control,” not “skip breakfast to lose weight.” The observational literature linking breakfast skipping to obesity is confounded by healthy-user effects.

Important tension for this subject: Sievert says breakfast is optional for weight, but Mamerow and Yasuda both suggest a protein-containing breakfast is the lever for lean mass — because breakfast is where most people’s protein distribution is worst (Mamerow’s SKEW arm: 10.7 g). Not contradictory: skip the carbohydrate-heavy breakfast, not the protein.

10.4 Late-night eating — mechanistic signal, no body-composition data

Vujović 2022, Cell Metab 34:1486. Randomized controlled crossover, rigorously controlling nutrient intake, activity, sleep and light. n=16 completed, late condition shifted 250 minutes later, ~4 days/condition:

Hunger: late eating roughly doubled the odds of being hungry (P<0.0001); raised waketime and 24-h ghrelin:leptin ratio.
Energy expenditure: −59.4 ± 13.9 kcal per waking day (−5.03%), P=0.002.
24-h core body temperature −0.19 °C.
Adipose gene expression (n=7 subset) shifted toward decreased lipolysis / increased adipogenesis.
Body composition and fat mass were NOT measured.

This is the best-controlled late-eating study and it is mechanistic only. A ~59 kcal/day expenditure difference is ~2% of the adult’s intake. Anyone citing this as “eating late makes you fat” is extrapolating past the data.

10.5 Athletes specifically

Moro 2016, J Transl Med 14:290. 8 weeks, 34 resistance-trained males, 16:8 TRF (13:00/16:00/20:00) vs normal diet (08:00/13:00/20:00), matched for kcal and macros (TRF 2,826 ± 412 kcal, 22.1% protein; ND 3,007 ± 445, 21.4%), standardized RT program, DXA.

Fat mass decreased in TRF vs ND (P=0.0448).
Fat-free mass, arm and thigh muscle area, and maximal strength were maintained in both groups.
Total and free testosterone and IGF-1 DECREASED significantly in TRF (P=0.0476; P=0.0397), no change in ND. T3 ↓, RER ↓, REE unchanged.

Note 2,826 kcal/day in the TRF arm is almost exactly this subject’s [calorie target removed] — the closest population match in this entire review. Reassuring on lean mass and strength over 8 weeks; the testosterone and IGF-1 declines are a genuine yellow flag for someone prioritizing muscle retention. The ~180 kcal difference between arms was not perfectly matched, which weakens the fat-mass finding.

Pre-sleep protein: ISSN Position Stand (Jäger 2017), verbatim: “Pre-sleep casein protein intake (30–40 g) provides increases in overnight MPS and metabolic rate without influencing lipolysis.” (Whether this retains any benefit once total daily protein is matched at a high level is unresolved by the acute data. The Snijders longitudinal trial was not retrieved — UNVERIFIED RECALL.)

10.6 Practical prescription on timing

~180 g protein/day (2.1 g/kg), floor 165 g, ceiling ~205 g.
4 meals × 40–50 g. Do it because it’s easy and non-falsified, not because distribution is proven to matter.
Protein at breakfast is the highest-leverage single distribution change — it is where nearly everyone’s distribution is worst, and it is the one thing Yasuda tested.
Don’t sweat exact peri-workout timing. ISSN: “the anabolic effect of exercise is long-lasting (at least 24 h).”
Skip TRE, or keep any window ≥10 h. The two strongest trials show no benefit beyond calorie control; TREAT’s lean-mass signal and Moro’s testosterone/IGF-1 declines are unnecessary risks when the goal is muscle retention and calories are already being tracked.

11. VISCERAL FAT — WHAT ACTUALLY MOVES IT

11.1 Total energy deficit is the dominant lever, and HOW you get there barely matters

Chaston TB & Dixon JB, Int J Obes 2008;32:619. PMID 18180786. Systematic review of imaging-based (CT/MRI) VAT and SAT before/after weight loss. 61 studies, 98 cohort time points.

Percentage weight loss was the ONLY variable that influenced the ratio of %ΔVAT to %ΔSAT (r = −0.29, P = 0.005).
VAT is lost preferentially with modest weight loss, and the preference is attenuated as weight loss gets larger.
“The method of weight loss was not an influence with one exception” — very-low-calorie diets produced exceptional short-term preferential VAT loss.

This is the single most important finding in the whole VAT literature and it is routinely buried: at matched weight loss, how you got there barely matters. Only how much.

Magnitude anchor: pooled MRI data from the CENTRAL and DIRECT PLUS 18-month trials, n=572 (Yaskolka Meir 2025): baseline mean VAT 144.8 cm²; over 18 months of lifestyle intervention −28 cm² VAT (−22.5%).

11.2 Macronutrient composition, head-to-head, with imaging

CENTRAL — Gepner Y et al., Circulation 2018;137:1143. PMID 29142011. n=278, 18 months, abdominal obesity or dyslipidemia, 89% men, mean age 48, BMI 30.8, isolated workplace with monitored provided lunch, 86% completion. Isocaloric low-fat vs Mediterranean/low-carb + 28 g walnuts/day, each with or without added moderate physical activity. MRI of VAT, deep/superficial SAT, liver, pericardial, muscle, pancreas, renal sinus.

Final weight loss did not differ between diets.
Exercise, with either diet, produced significantly greater VAT loss: mean difference −6.67 cm² (95% CI −14.8 to −0.45).
MED/LC beat low-fat for intrahepatic, intrapericardial and pancreatic fat — but the VAT advantage came from the EXERCISE arm, not the diet arm.
Authors’ conclusion: “exercise has an independent contribution to VAT loss.”

Gower BA & Goss AM, J Nutr 2015;145:177S. PMID 25527677. Study 1: n=69 overweight/obese, 8 wk eucaloric then 8 wk hypocaloric, lower-carb (43% CHO / 18% P / 39% F) vs lower-fat (55/18/27), CT.

After the EUCALORIC phase: intra-abdominal adipose tissue −11 ± 3% (lower-carb) vs −1 ± 3% (lower-fat), P<0.05 — at NO weight loss.
This is the strongest single result FOR a carb effect on VAT — but note it is a modest carb reduction (55→43%), n=69, 8 weeks.

Hall KD et al., Cell Metab 2015;22:427. n=19 adults with obesity, metabolic ward, exercising daily, 5-day baseline then 6 days of selective ISOCALORIC carbohydrate vs fat restriction, randomized order.

Carb restriction: 53 ± 6 g/day body fat loss. Fat restriction: 89 ± 6 g/day (P=0.002). Fat restriction won.
The authors’ own caveat: model simulations predict the body minimizes fat-loss differences over prolonged isocaloric diets. This is a 6-day mechanistic study.

11.3 The one clean depot-specific dietary finding: liquid fructose

Full detail in §5.5. Summary:

Stanhope 2009 (n=32, CT, 25% of energy from beverages, 10 weeks): both groups gained similar weight; fructose significantly increased total abdominal and visceral fat while glucose significantly increased subcutaneous fat. Men +18.1% intra-abdominal fat vs women −0.6% (P=0.049).
Maersk 2012 (n=47, 1 L/day for 6 months, MRS/MRI): liver fat relative increase 132–143% higher in the regular-cola group than isocaloric milk, diet cola, or water; visceral fat also increased in the cola arm.
Schwarz 2017 (n=41, 9 days isocaloric fructose restriction, weight controlled): VAT 123 → 110 cm³ (P<0.001); liver fat 7.2% → 3.8% (P<0.001); DNL AUC 68% → 26% (P<0.001).

11.4 Exercise independent of weight loss — and the resistance-training null

Verheggen RJ et al., Obes Rev 2016;17:664. 117 studies, n=4,815, radiographic imaging. Both diet and exercise reduce VAT (P<0.0001). Diet produced larger weight loss (P=0.04); a trend toward larger VAT reduction with exercise (P=0.08). Weight-change-to-VAT-change correlation R² = 0.737 after diet vs R² = 0.451 after exercise — exercise decouples the two. Headline: in the absence of weight loss, exercise is associated with a 6.1% decrease in VAT; diet with 1.1% (virtually none).
Vissers D et al., PLoS One 2013;8:e56415. 15 studies, 852 subjects, exercise without caloric restriction: Hedges’ g = −0.497 (−0.655 to −0.340), P<0.001. Aerobic training at moderate or high intensity had the greatest effect.
Ismail I et al., Obes Rev 2012;13:68. 35 studies, CT/MRI required. Aerobic vs control ES −0.33 (−0.52 to −0.14, P<0.01). RESISTANCE training vs control ES 0.09 (−0.17 to 0.36, P=0.49) — NULL.

This matters for the adult specifically: given back/joint limits, the temptation is to substitute lifting for cardio. For the visceral-fat goal, resistance training is null. Whatever else it is good for — and it is good for a lot, including the joint-load story in §12 — it does not move VAT.

11.5 Alcohol

Honest answer: no high-quality causal evidence was found. No Mendelian randomization study with VAT as an outcome; no RCT of alcohol reduction with imaged VAT. The existing literature is observational, heavily confounded by total energy and the J-shaped drinker/abstainer problem, and typically uses waist circumference rather than imaged VAT.

The defensible statement: alcohol contributes 7.1 kcal/g of poorly-compensated liquid energy, which is a strong mechanistic route to positive energy balance and therefore to VAT via the dominant lever. A specific, depot-selective visceral effect of alcohol independent of energy is not established. [Absence of found evidence — not the same as evidence of absence.]

11.6 Does any macronutrient preferentially mobilize VAT at matched weight loss?

No — with one qualified exception and one strong caveat.

Chaston & Dixon (61 studies, 98 time points): “The method of weight loss was not an influence.” Only the amount predicted VAT-vs-SAT preference. This is the best available answer and it is a null.
CENTRAL (n=278, 18 mo, MRI): weight loss did not differ between diets, and the VAT advantage came from exercise. MED/LC did win on liver, pericardial and pancreatic fat — so macronutrients appear to move ectopic depots differentially even when they don’t move VAT differentially.
The qualified exception: Gower & Goss (n=69) found −11% vs −1% IAAT at eucaloric. Real, imaged, isocaloric — but a single 8-week study, and Hall’s metabolic-ward work found fat restriction produced more total fat loss over 6 days.
The one thing that IS depot-specific: liquid fructose.

11.7 Ranked VAT levers

Rank	Lever	Best effect size	Evidence grade
1	Total energy deficit	VAT lost preferentially at modest loss; ~−22.5% VAT over 18 mo (n=572, MRI)	SR of 61 imaging studies
2	Aerobic exercise (independent of weight loss)	−6.1% VAT at zero weight change; Hedges’ g −0.497; −6.67 cm² on top of diet in an 18-mo MRI RCT	3 concordant MAs (852 / 4,815 subjects) + RCT
3	Eliminate liquid fructose / SSBs	VAT ↑ significantly at 25% E fructose at matched weight gain (n=32); liver fat +132–143% at 1 L cola/d (n=47)	2 RCTs with CT/MRS
4	Alcohol (as liquid energy, not depot-specific)	No imaged causal estimate found	Weak / mechanistic only
5	Modest carbohydrate restriction	−11% vs −1% IAAT eucalorically (n=69, 8 wk, CT)	One trial; contradicted at total-fat-loss level (n=19)
6	Soluble fiber	−3.7% VAT accumulation rate per 10 g/d over 5 y	Observational only (n=1,114)
7	Protein above adequacy	−17.3 cm² at 1.3 vs 0.8 g/kg — but in men ≥65 with habitual intake ≤0.8 g/kg (n=56 imaged)	RCT, wrong population; the adult is not on this part of the curve
—	Resistance training	ES 0.09 (−0.17 to 0.36) — NULL for VAT	35-study MA
—	Sodium reduction	No evidence. Cross-sectional associations shrink monotonically as measurement improves — a confounding signature	No RCT with a VAT endpoint located

12. JOINT AND BACK PAIN — BE SKEPTICAL, THIS AREA IS FULL OF WEAK CLAIMS

12.1 Tier 1: body weight / fat loss — the only dietary lever with large, replicated, mechanistically-grounded effects

IDEA trial — Messier SP et al., JAMA 2013. Single-blind RCT, n=454 (399 completed, 88%), overweight/obese adults with knee OA, 18 months:

Arm	Weight loss	WOMAC pain (95% CI)	IL-6
Diet + Exercise	10.6 kg (11.4%)	3.6 (3.2–4.1)	2.7 pg/mL
Diet alone	8.9 kg (9.5%)	4.8 (4.3–5.2)	2.7 pg/mL
Exercise alone	1.8 kg (2.0%)	4.7 (4.2–5.1)	3.1 pg/mL

Diet+Exercise vs Exercise: pain difference 1.02, P=0.004; IL-6 difference 0.39 pg/mL, P=0.007. Knee compressive force: diet group 2,487 N vs exercise 2,687 N — difference 200 N (55–345), P=0.007.

Mechanism quantified — Messier SP et al., Arthritis Rheum 2005, 18-month RCT with 3D gait analysis, n=142: “A weight reduction of 9.8 N (1 kg) was associated with reductions of 40.6 N and 38.7 N in compressive and resultant forces… each weight-loss unit was associated with an approximately 4-unit reduction in knee-joint forces.” Also 1 kg → 1.4% reduction in knee abduction moment. This is the origin of the “1 lb lost = 4 lb off the knee per step” figure, and it is real and correctly cited.

Low back pain — Shiri R et al., Am J Epidemiol 2010. 95 studies reviewed, 33 in meta-analysis. Obesity vs normal weight: LBP in past 12 months OR 1.33 (1.14–1.54); seeking care OR 1.56 (1.46–1.67); chronic LBP OR 1.43 (1.28–1.60); incident LBP in cohort studies OR 1.53 (1.22–1.92). Overweight showed intermediate risk — a dose gradient, which strengthens causal inference. The cohort-based incident finding is the important one because it partially answers reverse causation (pain → inactivity → weight gain). (No Mendelian randomization study of BMI → back pain was found — do not claim one exists.)

The systemic-vs-mechanical test (non-weight-bearing joints): genuinely mixed, weaker than usually claimed. A 389,807-person 13-year cohort found high BMI elevated risk of both knee and hand OA “irrespective of metabolic status”; a cross-sectional study (n=858) found BMI OR 1.04 (1.00–1.09) and waist-to-height OR 1.03 (1.01–1.06) for symptomatic hand OA — very small; and HALLOA (n=231) found “no significant relationships between hand OA and obesity or serum leptin levels.” Honest verdict: hand OA associations with adiposity exist but are small and inconsistent — suggestive of a systemic component, not proof of one. Do not oversell it.

12.2 Tier 2: nothing. There is a genuine gap between weight/adiposity and everything below.

12.3 Tier 3: small, unreliable, or null

Omega-3 — real in rheumatoid arthritis, null in osteoarthritis, absent for back pain.

RA — Goldberg & Katz, Pain 2007, MA, 17 RCTs, 3–4 months: patient-reported joint pain intensity SMD −0.26 (−0.49 to −0.03), p=0.03; morning stiffness −0.43 (−0.72 to −0.15); painful/tender joints −0.29 (−0.48 to −0.10); NSAID consumption −0.40 (−0.72 to −0.08). Physician-assessed pain: not significant. Ritchie articular index: not significant. Note the pattern: patient-reported outcomes move; physician-assessed outcomes don’t. In a partially-blinded supplement literature with fishy-tasting capsules, that is exactly the signature of incomplete blinding. Effect sizes are small-to-moderate; the NSAID-sparing finding is the most clinically meaningful.
OA — the two best-designed trials are NEGATIVE.
- Hill CL et al., Ann Rheum Dis 2016. Randomized, double-blind, multicentre, n=202, symptomatic knee OA, 24 months. High dose 4.5 g/d omega-3 vs low dose 0.45 g/d. “The low-dose fish oil group had greater improvement in WOMAC pain and function scores at 2 years compared with the high-dose group”; no difference in cartilage volume loss. A 10× dose increase produced, if anything, WORSE outcomes — a strong argument against a dose-dependent anti-inflammatory analgesic mechanism in OA.
- Laslett LL et al., JAMA 2024;331:1997. Randomized, double-blind, placebo-controlled, n=262, knee OA plus MRI-confirmed effusion-synovitis (i.e. enriched for the inflamed population where marine oils should work best), 2 g/d krill oil, 24 weeks. VAS pain change −19.9 vs −20.2; between-group difference −0.3 (−6.9 to 6.4). “These findings do not support krill oil for treating knee pain.”
Non-arthritic musculoskeletal pain / DOMS — two meta-analyses of overlapping tiny trials reach OPPOSITE verdicts. Lv 2020 (12 RCTs, 145/156): DOMS at 2 days MD −0.93 (−1.44 to −0.42), but the authors’ own conclusion is “Low-quality evidence that n-3 PUFA supplementation does NOT result in a clinically important reduction of muscle soreness.” Yaghoobi 2026 (9 RCTs): DOMS Hedges’ g = −0.75 (−1.14 to −0.36), but “due to the methodological limitations… it was not possible to assess effective dosing strategies.” Median trial size is roughly 10–25 people. I would not tell anyone that omega-3 treats DOMS.
Low back pain specifically: NO RCT evidence that omega-3 treats non-specific low back pain was found. Not “weak evidence” — no trials surfaced. Any claim otherwise should be treated as fabricated until a citation is produced.

Dietary Inflammatory Index — largely circular, almost entirely cross-sectional. The DII was constructed by scoring foods/nutrients from the literature on inflammatory markers, then correlated with outcomes in the same self-report datasets.

Wu 2025, Front Nutr, hospital cross-sectional, n=598 sciatica patients: per 1-unit DII, VAS +0.48 (0.42–0.53), ODI +4.75 (4.16–5.34). But the authors note DII was NOT significantly associated with CRP in their own sample — the “inflammatory” index failed to predict inflammation while predicting pain, which points to reporting/behavioral confounding rather than an inflammatory pathway.
Marques 2025, BMC Oral Health: no significant DII differences in painful TMD vs controls.
The most informative datapoint: Law et al., J Am Nutr Assoc 2025 — FEAST trial baseline, n=144 knee OA, with James R. Hébert (the DII’s developer) as an author — found “no associations between DII/E-DII and KOOS subscales” except ADL, despite a relatively pro-inflammatory baseline diet.

Verdict: DII is not evidence for a causal diet→pain pathway. Cross-sectional, self-reported, plausibly confounded by depression/SES/activity, subject to reverse causation, and null in the developer’s own trial baseline. Treat DII-based claims as hypothesis, not evidence.

Vitamin D — essentially null in replete people.

Chronic low back pain: Lee 2024, In Vivo, MA of 10 RCTs — “vitamin D supplementation did not significantly reduce pain scores (SMD −0.130, 95% CI −0.260 to 0.000)”; long-term SMD −0.097. “Vitamin D supplementation does not substantially alleviate CLBP.”
Knee OA: McAlindon 2013, JAMA — 2-year randomized, placebo-controlled, double-blind, n=146, achieved 25(OH)D +16.1 vs +2.1 ng/mL: WOMAC pain −2.31 vs −1.46, no significant differences at any time; cartilage volume loss −4.30% vs −4.25%, P=0.96.
Contrary signals exist but are subgroup/post-hoc (a VIDEO post-hoc, n=413, found benefit only at baseline 25(OH)D ≤43 nmol/L).

Verdict: correcting genuine deficiency (<20 ng/mL) is worth doing on general grounds. Supplementing a replete person for joint or back pain is NOT supported — the best-powered CLBP meta-analysis has a CI touching zero at its most favorable edge.

Glucosamine / chondroitin — negative in the trials designed to be definitive.

GAIT — Clegg 2006, NEJM. n=1,583, 24 weeks, 5 arms. Response rates vs placebo 60.1%: glucosamine 64.0% (P=0.30); chondroitin 65.4% (P=0.17); combination 66.6% (P=0.09); celecoxib 70.1% (P=0.008). Note the 60% placebo response rate — that alone explains most supplement testimonials. The moderate-to-severe subgroup (n=354) showed 79.2% vs 54.3% (P=0.002) — unreliable: exploratory, unadjusted for multiplicity, and it failed to replicate in GAIT’s 2-year extension.
Wandel 2010, BMJ — network MA, 10 trials, 3,803 patients, prespecified MCID −0.9 cm on a 10 cm VAS: glucosamine −0.4 cm (−0.7 to −0.1); chondroitin −0.3 (−0.7 to 0.0); combination −0.5 (−0.9 to 0.0). “For none of the estimates did the 95% credible intervals cross the boundary of the minimal clinically important difference.”

Collagen peptides — positive pooled estimates, but the effect shrinks as trial quality rises. Classic small-study-bias signature. Best and largest: Liang 2024, Osteoarthritis Cartilage — 35 RCTs / 3,165 participants: pain SMD −0.35 (−0.48 to −0.22), moderate certainty; function SMD −0.31, high certainty. Others report much larger effects with I²=88% and “all these trials were considered to have a high risk of bias.” Several recent positive RCTs are manufacturer-supported. A small effect, roughly the size of the omega-3-in-RA analgesic effect.

Curcumin — positive pooled estimates, catastrophic evidence quality. Zhao 2024 (Bayesian NMA, 23 studies / 2,175): VAS MD −1.63 (−2.91 to −0.45); WOMAC −18.85. Hsueh 2025 (21 studies / 1,705): CRP SMD −0.906, TNF-α −0.921, no effect on ESR, IL-1β, IL-6, PGE-2. But Chen 2025, Front Pharmacol — critical overview of 7 systematic reviews — rated methodological quality “extremely low,” with only 5 of 48 outcomes rated medium-quality evidence. The pooled numbers look impressive; the critical appraisal of those same pools says the underlying trials are near-worthless — small, from a narrow set of groups, testing proprietary bioavailability-enhanced formulations, with an outcome (self-reported pain) maximally sensitive to unblinding.

12.4 Final honest ranking for musculoskeletal pain

Rank	Intervention	Effect size	Evidence quality
1	Exercise therapy (NOT dietary)	Pain MD −15.2 (−18.3 to −12.2) on 0–100 vs no treatment	249 trials, Hayden Cochrane 2021, moderate certainty, exceeds MCID
2	Fat loss / weight reduction	1 kg → ~4 kg less knee load per step; 10.6 kg loss → WOMAC pain 3.6 vs 4.7	RCT (n=454) + mechanism (n=142) + cohort (33 studies)
3	Omega-3 — RA only	SMD −0.26 to −0.43 on patient-reported outcomes	17 RCTs, but physician-assessed outcomes null
4	Collagen peptides	SMD −0.35	35 RCTs, heterogeneity + industry funding; effect shrinks with quality
5	Curcumin	VAS −1.63 nominally	Underlying quality “extremely low” (5/48 outcomes above that)
6	Vitamin D (if genuinely deficient)	Post-hoc only	Best CLBP MA SMD −0.130, CI touches 0
7	Glucosamine / chondroitin	−0.3 to −0.5 cm vs MCID −0.9 cm	Definitively negative (n=1,583 + n=3,803)
8	Omega-3 for OA	Null	Hill n=202 (10× dose = no benefit); Laslett n=262 (null in inflamed knees)
9	DII / “anti-inflammatory diet” scores	Cross-sectional only	Circular construction; null in the developer’s own trial baseline
10	Omega-3 for DOMS / back pain	Contradictory / absent	Two MAs of tiny trials reach opposite conclusions; zero LBP trials found

12.5 Do specific meals cause acute inflammation relevant to the adult’s back/joint issues?

No — and this should be unambiguous.

No evidence was found that a single meal’s macronutrient or fatty-acid composition measurably changes joint or back pain in a non-arthritic person. What exists:

Postprandial lipemia transiently raises inflammatory markers after high-fat meals. This is a BIOMARKER finding. No study links postprandial inflammatory-marker excursions to same-day or next-day joint/back pain in healthy adults. Inferring from marker to symptom is exactly the reasoning error that produced the DII literature.
Elimination diets (gluten, nightshades) for non-celiac musculoskeletal pain: no RCT evidence found. Claims in this space should be treated as unsupported.

The honest answer for this specific person: it is load management, strength training, and total adiposity — not diet composition. Diet earns its place in the adult’s plan for visceral fat (energy balance), training quality (carbohydrate availability), and 50-year ASCVD risk (SFA→unsaturated substitution) — NOT as an analgesic.

Scoring implication: do NOT build a “joint/inflammation” sub-score into the meal scorer. There is nothing to encode. The joint benefit of the scorer is entirely indirect, via the energy-deficit path.

13. VO2MAX — DIETARY LEVERS WITH REAL EVIDENCE

13.1 Dietary nitrate / beetroot — works on economy, not on VO2max, and probably not on the adult

Mechanism study — Bailey SJ et al., J Appl Physiol 2009;107:1144. n=8 men, double-blind placebo-controlled crossover, 500 mL/day beetroot juice ≈ 5.6 mmol nitrate for 6 days. Plasma nitrite 273 ± 44 vs 140 ± 50 nM; SBP −8 mmHg (P<0.01). Moderate exercise: O2 uptake gain reduced 19% (8.6 ± 0.7 vs 10.8 ± 1.6 mL·min⁻¹·W⁻¹). Severe exercise: VO2 slow component reduced; time-to-exhaustion +16%. VO2max was not the outcome and was not raised. The effect is a reduced oxygen COST at a given power — efficiency, not ceiling.

Meta-analysis — McMahon, Leveritt & Pavey, Sports Med 2017;47:735. 47 studies / 76 trials.

Time trial (28 trials): ES −0.10 (−0.27 to 0.06) — NON-significant.
Time to exhaustion (22 trials): ES 0.33 (0.15–0.50) — small-to-moderate, significant.
Graded exercise test (8 trials): ES 0.25 (−0.06 to 0.56) — NON-significant. GXT is the protocol class closest to a VO2max test.

The trained-active adult caveat, with two clean nulls:

Christensen PM et al., Scand J Med Sci Sports 2017;27:1616. Endurance-trained cyclists n=9 (VO2max 64 ± 3) vs recreationally active n=8 (46 ± 3), acute ~9 mmol nitrate. Plasma nitrate +1200%. No change in submaximal VO2 (leg or arm) or resting MAP in either group. VO2max was not affected. Only incremental peak power rose in the trained group. Authors note high baseline plasma nitrite may blunt the effect — trained athletes may already be at ceiling.
Boorsma RK, Whitfield J & Spriet LL, Med Sci Sports Exerc 2014;46:2326. n=8 elite male 1500-m runners, VO2peak 80 ± 5, PB 3:56 ± 9. 19.5 mmol nitrate (a very large dose), acute and 8-day chronic. Plasma nitrate 37 → 615 → 870 µM. No VO2 differences at 50%, 65% or 80% VO2peak. 1500-m TT completely unaffected. Two individual responders out of eight.

Mouthwash caveat (Bryan, Burleigh & Easton, Nitric Oxide 2022): “There is unequivocal evidence that dysbiosis… or disruption (e.g. by use of antiseptic mouthwash or antibiotics) of the oral microbiota will suppress nitric oxide production… and negatively impact blood pressure.” But the same review states it is “yet to be established whether purposefully altering the oral microbiome can have a meaningful impact on exercise performance.”

Verdict for this person: dose is 6–8 mmol NO3⁻, ~2–3 h pre-exercise. It does NOT raise VO2max — no meta-analysis shows a significant GXT effect. It reduces submaximal O2 cost and extends TTE in untrained-to-moderately-trained people; in genuinely trained athletes it attenuates toward null. For recreational sport — a repeated-short-effort sport — the TTE benefit is the least transferable of the three outcomes.

13.2 Iron — big effect if deficient, but the adult almost certainly isn’t

Burden RJ et al., Br J Sports Med 2015;49:1389. MA, 17 studies, iron-deficient non-anaemic endurance athletes. Hedges’ g: serum ferritin 1.088 (0.914–1.263); serum iron 1.004; transferrin saturation 0.741; haemoglobin 0.695; VO2max 0.610 (0.399–0.821) — all P<0.001.

Is it relevant to a adult male? Almost certainly not. Looker 1997, JAMA 277:973 (NHANES III, n=24,894): “Iron deficiency occurred in… no more than 1% of teenage boys and young men.” The 2024 update (JAMA Netw Open 7:e2433126, NHANES 2017–2020, n=8,021) finds 14% absolute and 15% functional deficiency across all US adults, but that burden is concentrated in premenopausal women. (The “<35 ng/mL ferritin” active adult cutoff commonly quoted is UNVERIFIED RECALL; the IDNA literature operationally uses roughly <30–35 ng/mL with normal haemoglobin.)

Verdict: g = 0.610 on VO2max is one of the largest effects in this entire document — but it only exists if you are deficient, and the prior probability for a adult male is ≈1%. Correct action: measure ferritin ONCE, then stop thinking about it. Do NOT supplement iron blind — iron overload is a real harm in males.

13.3 Carbohydrate availability — the real lever, and it is protective not additive

“Train low” does not work in trained athletes. Gejl & Nybo, JISSN 2021;18:37. MA restricted to genuinely trained athletes (VO2max ≥55 women / ≥60 men), interventions ≥1 week, periodized CHO restriction ≥3×/week. 9 studies. SMD 0.17 (−0.15 to 0.49), P=0.29. No effect. Authors: “periodized CHO restriction does not per se enhance performance in endurance-trained athletes,” explicitly flagging that “compromised training quality and particularly lower intensities in peak intervals seem to be a major drawback.”

Low CHO availability impairs economy and performance DESPITE raising VO2peak — the single best illustration that VO2max ≠ performance. Burke LM et al., J Physiol 2017;595:2785. n=29 elite race walkers, 3 weeks of intensified training, three isoenergetic diets: HCHO (n=9, 8.6 g/kg/d CHO), PCHO (n=10, same macros periodized), LCHF (n=10, <50 g/d CHO, 78% energy as fat).

VO2peak increased in ALL THREE groups (P<0.001, 90% CI 2.55–5.20%).
LCHF peak fat oxidation reached 1.57 ± 0.32 g/min — a large, real metabolic adaptation.
But economy worsened: O2 uptake at race pace (as % of the new VO2peak) fell in HCHO and PCHO, while it was maintained at pre-intervention levels in LCHF — i.e. LCHF got no economy improvement.
10-km race walk time: HCHO +6.6% (90% CI 4.1–9.1), PCHO +5.3% (3.4–7.2), LCHF −1.6% (−8.5 to +5.3).

Verdict: adequate CHO is a genuine VO2max-TRAINING lever, but it works by protecting interval quality, not by any direct effect. For a recreational sport player doing VO2max work, under-fuelling carbohydrate is a way to LOSE adaptation, not gain it. This is the main reason not to push protein above ~200 g/d at a fixed [calorie target removed].

13.4 The rest, ranked

Supplement	Best evidence	Effect size	Raises VO2max?
Iron (if deficient)	Burden 2015, 17 studies	g = 0.610 (0.399–0.821)	Yes, directly — but only if deficient
Caffeine	Grgic 2020 BJSM umbrella (11 reviews / 21 MAs); Wang 2022 Nutrients (21 RCTs, 3–9 mg/kg)	TTE g = 0.392 (0.214–0.571); trained runners g = 0.344 (0.122–0.566); time trial g = −0.101; Yo-Yo +7.5%	No — improves performance/tolerance, not the ceiling. Larger effect for aerobic than anaerobic
Sodium bicarbonate	Peart 2012 JSCR; Grgic 2020 JSAMS	“Moderate” ES overall, significantly lower in specifically-trained vs recreationally-trained; Yo-Yo SMD 0.36	No. Buffering; helps 1–10 min max efforts
Beta-alanine	Saunders 2017 BJSM. 40 studies, 65 protocols, 1,461 participants	Overall ES 0.18 (0.08–0.28). In the 0.5–10 min window: capacity ES 0.4998 (0.246–0.753); performance ES 0.108 (ns). No moderation by training status	No. Buffering. Recreational sport rallies are far shorter than the 0.5–10 min window — low relevance
Dietary nitrate	McMahon 2017 (47 studies)	TTE ES 0.33; GXT ES 0.25 (−0.06 to 0.56), NS; TT ES −0.10, NS. Null in elites	No
Creatine	Gras 2023, Crit Rev Food Sci Nutr. 19 RCTs, 424 individuals	VO2max increased LESS with creatine: ES −0.32 (−0.51 to −0.12), P=0.002	Negative — almost certainly a mass artifact (VO2max is mL/kg/min and creatine adds water weight). Real value is repeated-sprint/jump power, which does matter for recreational sport

13.5 Antioxidant blunting — widely mis-stated; the effect is on signalling, not VO2max

Paulsen G et al., J Physiol 2014;592:1887. Double-blind RCT, n=54, 1,000 mg vitamin C + 235 mg vitamin E vs placebo daily, 11 weeks, HIIT + steady-state.

VO2max: +8 ± 5% in the vitamin group vs +8 ± 5% in placebo — IDENTICAL. 20-m shuttle +10 ± 11% vs +14 ± 17%.
But mitochondrial markers were blunted: COX4 +59 ± 97% and PGC-1α +19 ± 51% in placebo vs COX4 −13 ± 54% and PGC-1α −13 ± 29% in the vitamin group (P ≤ 0.03).
Authors: “no clear interactions were detected for improvements in VO2max and running performance… although this did not translate to the performance tests applied in this study, we advocate caution.”

Ristow M et al., PNAS 2009;106:8665. Vitamin C 1,000 mg/d + vitamin E 400 IU/d, 4 weeks of exercise, n=19 untrained + n=20 pretrained = 39 analyzed (commonly cited as n=40; the correct figure is 19 + 20). Outcome was INSULIN SENSITIVITY, not VO2max. GIR and adiponectin increased ONLY in the absence of antioxidants, in both subgroups (P<0.001 each). PPARγ, PGC1α, PGC1β, SOD1, SOD2 and glutathione peroxidase induction were all blocked.

Companion strength trial (Paulsen, J Physiol 2014;592:5391, n=32, 10 weeks): altered protein signalling but no effect on muscle growth.

Verdict: the mitohormesis story is real at the level of cell signalling and insulin sensitivity, and has NEVER been shown to blunt VO2max in a controlled trial. The practical implication is narrow: avoid high-dose SUPPLEMENTAL vitamin C (1,000 mg) and E (235 mg / 400 IU) around a training block. Food-level vitamin C and E — a bell pepper has ~150 mg C — are nowhere near these doses. DO NOT encode dietary vitamin C/E as a “blunting” penalty; that would be a dose-category error.

13.6 Encodability from meal nutrition data

Lever	Encodable?	Notes
CHO grams / g·kg⁻¹·d⁻¹	Yes, cleanly	The highest-value VO2max signal that is actually computable. Flag days where CHO is low and a high-intensity session is scheduled
Iron (mg)	Yes	But intake is a poor proxy for status — this should trigger “get ferritin measured,” not “eat more iron”
Caffeine (mg)	Partially	In USDA FDC for coffee/tea/soda; usually absent from packaged-meal data, never on labels
Dietary nitrate (mmol)	No	Nitrate is not in standard nutrition databases. Best proxy is a curated list of nitrate-dense vegetables (beetroot, rocket, spinach, lettuce, celery, rhubarb) scored by serving weight — a hand-built lookup, not a database field. Given the null GXT effect, low priority
Vitamin C / E (mg)	Yes, but do not use for blunting	Food-level doses are an order of magnitude below trial doses
Beta-alanine, bicarbonate, creatine	No	Supplements, not food constituents