data/vo2max/README.md

VO2 Max — aerobic fitness and training response (version 2.0.0)

This module looks at spots in your DNA linked to cardiorespiratory fitness — how much oxygen your body can use during hard exercise (VO2max) — and, in the original rows, to how much VO2max improves with training. Every effect here is small. Fitness depends far more on training, age, sex and body composition than on any DNA spot, and nothing in this module is a diagnosis.

What is in it

Group (category)	Rows	What it means
VO2max GWAS lead (direct fitness test)	14 variants × 3 genotypes	Genome-wide significant (p < 5×10⁻⁸) in the UK Biobank fitness GWAS (N = 69,416 with a submaximal cycle-ergometer test) — Cai et al. 2023, Nat Commun, PMID 37400433, Supplementary Table 3 / Supplementary Data 1
Heart-rate-informed fitness locus	123 variants × 3 genotypes	Genome-wide significant for resting heart rate in the same paper (UK Biobank, >450,000), kept in the 160-variant fitness instrument because their effect on measured fitness was directionally consistent (nominal p < 0.05 for fitness). Their fitness effects are weak individually and partly reflect resting heart rate
Candidate gene (tested, small or no effect)	2 variants × 3 genotypes	ACTN3 R577X (rs1815739, weight 0 — no endurance effect in 2025 meta-analysis) and PPARGC1A Gly482Ser (rs8192678)
(blank — original rows)	13 variants, 39 rows	Olga Borysova's expert-curated rows on VO2max training response (HERITAGE and related studies), kept unchanged

Totals: 152 variants, 456 genotype rows, 161 study rows citing 12 papers.

How the rows were chosen (selection rule, 2026 rebuild)

• Tier A (GWAS). Anchor: Cai L, et al. Causal associations between cardiorespiratory fitness and type 2 diabetes. Nat Commun 2023;14:3904 (PMID 37400433) — the largest genome-wide study of measured cardiorespiratory fitness available 2021–2026. Every variant of its 160-variant fitness instrument (Supplementary Data 1: 14 direct fitness-GWAS leads + 146 resting-heart-rate variants retained after radial filtering) was taken, with the effect allele, effect-allele frequency, β (ml O₂·min⁻¹·kg⁻¹ fat-free mass per allele), s.e. and p transcribed from that table. 22 were dropped because they cannot be represented honestly (see below). rs7412 (APOE) is in the instrument and already in the legacy rows, so only a studies.csv row was added for it.
• Tier B (candidate/functional). Only candidate variants with a 2019–2026 meta-analysis were added: ACTN3 rs1815739 (Chelly 2025, PMID 40845214; El Ouali 2024, PMID 38609671) and PPARGC1A rs8192678 (Su 2025, PMID 41574190; Chen 2019, PMID 31938000). Candidate-gene evidence for athletic performance is weak and inconsistent: a 2026 systematic review of 204 studies (PMID 42931598) concluded that individual variants have modest, context-dependent effects and that the evidence does not support talent identification or training prescription. ACE I/D, EPAS1 and EPOR were considered and not added (below).
• Tier C (legacy). Olga Borysova's 39 rows are unchanged.

Considered and not used as row sources: HUNT direct VO2peak GWAS (Klevjer 2022, PMID 35482759; N = 4,525, loci largely unreplicated), HUNT and FinnGen polygenic-score papers (PMIDs 41402372, 40672480 preprint, 42087449 — genome-wide scores, no per-variant lead table), Predict-HIIT training-response GWAS (PMID 33985508; no genome-wide-significant variant), small athlete GWAS (PMID 41345885) and a 228-person HIIT GWAS (PMID 37170954). No cardiorespiratory-fitness score is in the PGS Catalog as of this rebuild.

Dropped Tier A candidates (22 of 160)

• 8 indels published only as GRCh37 markers with no rsID (e.g. 6:122514171_CTTTT_C) — nothing here lifts coordinates between builds.
• 12 multi-allelic repeat or homopolymer indels whose biallelic I/D coding in the study cannot be mapped to a single VCF allele (including rs763773724, the HTT CAG repeat): rs753131216, rs781155278, rs373999074, rs72394060, rs781388561, rs564190295, rs763773724, rs138564056, rs375963061, rs35987999, rs796269914, rs5879832.
• 2 strand-ambiguous (C/G) SNPs with effect-allele frequency between 0.42 and 0.58: rs7136054 (0.484), rs2054266 (0.501). The 16 other A/T or C/G SNPs were kept because the study's EAF is far from 0.5 and agrees with UK reference frequencies (TwinsUK/ALSPAC in dbSNP).

Weights (two scales — read before summing)

• New GWAS rows: weight = +β/SD × number of copies of the fitness-raising allele (0, 1 or 2), with SD = 6.9 ml O₂·min⁻¹·kg⁻¹ FFM (implied by the paper's "3% lower diabetes risk per unit = 19% per SD"; sex-specific SDs 6.4 and 7.1). Zero copies = 0.0 (neutral). Range 0.010–0.180.
• Candidate rows: ln(OR) of the published genotype odds ratio — PPARGC1A Gly/Gly (C/C) = +0.174 (OR 1.19 for elite endurance-athlete status, Caucasians); ACTN3 = 0.0 for all genotypes.
• Legacy rows (blank curator/method): Olga Borysova's ordinal scale {−1, −0.5, 0, 0.5, 1} for training response.
• Positive = favourable (higher fitness). The scales are not comparable: a single legacy row outweighs the sum of all new GWAS rows at typical genotypes. Do not read a module "net weight" as a fitness score. The whole 160-variant score explained about 1% of fitness variance in the Fenland validation cohort.

Limits

• All Tier A effects come from people of European ancestry in the UK; transfer to other ancestries is unknown.
• VO2max in UK Biobank was estimated from heart rate during a submaximal cycle test, not measured maximally; heart-rate genetics therefore enter both the trait and the instrument.
• The 146 heart-rate variants were selected for consistency with fitness, so their fitness betas are optimistic (selection on the same data).
• Fitness level (new rows) and training response (legacy rows) are different traits.
• Effect alleles were taken from the paper. The GWAS Catalog record GCST90310239 lists rs67622553-G as fitness-raising, which contradicts the paper's own Supplementary Table 3 and Supplementary Data 1 (T raises fitness); the paper was followed. For SLC35F1 the instrument uses rs11153730 where Supplementary Table 3 names rs1572226.
• No provenance_quote was authored for the new rows; their numbers come from a supplementary workbook, which the literature checker does not open.

Provenance

Original module: Olga Borysova (expert curation, v1.x). 2026 rebuild (v2.0.0): AI-assisted, recorded in authorship: as ai, agent; no human has reviewed the new rows yet. Rows added in 2026 carry curator = ai-agent (2026 rebuild, just-module-creator).

Evidence tiers

Every variant carries an evidence tier in flags (tier:S … tier:D) and in priority (S = high, A = medium, B/C/D = low). The tier never changes weight. Most variants in this module are weak signals, so the top tier is S, kept for the few variants that matter on their own. New (2026) rows also start their conclusion with the tier's label.

Tier	Meaning	Variants (of which legacy)
S	Strong evidence: an effect large enough to matter on its own (OR/HR ≥ 1.5 or ≤ 0.67, or ≥ 0.5 SD), robustly supported	0 (0)
A	Well established, small effect (research grade): genome-wide significant in the anchor GWAS and in ≥ 1 further publication, same direction	3 (0)
B	Single-study signal, small effect (research grade): genome-wide significant in the anchor GWAS only	11 (0)
C	Early evidence (research only): not genome-wide significant for this trait	137 (13)
D	Not supported: contradicted, or null in large GWAS / meta-analysis	1 (0)

No variant reaches S. Most rows are C, because the heart-rate-informed loci are only nominally associated with measured fitness and the training-response rows come from candidate-gene studies. Replication was read from the GWAS Catalog (REST v2, retrieved 2026-11-03; cohorts of different publications may overlap). also_in:<module> marks the 6 variants that also appear in another just-dna module. Summing weights across modules counts those twice. The rules are adapted from the Longevity Map v2 requirements (docs/longevitymap_v2/).

Corrections to the original expert-curated rows

The 2026 audit of the Generation I content (Supplementary Table S14 of the Just-DNA-Lite paper) found no errors in this module's original rows that were still present in v1.1.0; the one Generation I error (a mistyped PMID) was already fixed when the module was converted to the current format.