Genomic predictors of the maximal O2 uptake response to standardized exercise training programs

473 sedentary white adults from 99 families (HERITAGE); 324,611 SNPs

HERITAGE Family Study, the same cohort as Bouchard 1999. Adults, 20-week standardized cycle training. Partial look-ups in STRRIDE, DREW and the HERITAGE black cohort.

J Appl Physiol 110(5):1160-1170 (published online December 2010; some indexes list it as 2010). THE HEADLINE NUMBER IS AN OVERFIT AND SHOULD NOT BE QUOTED AS A VARIANCE-EXPLAINED ESTIMATE. The 21-SNP panel was produced by stepwise multiple regression over 39 SNPs that had themselves been pre-selected from 324,611 on the same 473 people. Selecting 21 predictors out of a 324,611-wide search space in n=473 and then reporting in-sample R2 is textbook optimism; NONE of the SNPs reached genome-wide significance, and the panel was never validated as a panel in an independent cohort. Only partial single-SNP look-ups were attempted elsewhere.

The strongest-looking molecular case for trainability, and it does not hold. Single-SNP analysis found 39 SNPs at P < 1.5e-4; stepwise regression selected 21 of them into a panel reported as accounting for 49% of the variance in VO2max trainability, with carriers of <=9 favourable alleles gaining 221 ml/min versus 604 ml/min for carriers of >=19. The strongest single SNP accounted for ~6% on its own. Read against the rest of this topic: the reported 49% is an in-sample stepwise fit, not a replicated predictor, and it happens to land suspiciously close to the 47% "maximal heritability" from the same cohort. Later work found only partial single-SNP replication and a systematic review of trainability genes (Williams et al. 2017, BMC Genomics) reported that only 13 variants across ~97 candidate genes had been reproduced by more than two independent author groups.

This IS the genetically-informative design, so the usual selection worry does not apply. The failure mode is different and just as fatal: multiple comparisons and model selection in a small family sample. Consistent with Rankinen 2016 (GAMES), where 45 candidate markers for elite endurance status replicated in zero of seven independent cohorts.

Not replicated as a panel. Partial single-SNP replication only (e.g. the SNP nearest ZIC4 in STRRIDE). Systematic review evidence indicates poor reproducibility across the trainability candidate-gene literature.