This is a correction to: Xavier Farré, Ruben Molina, Fabio Barteri, Paul R H J Timmers, Peter K Joshi, Baldomero Oliva, Sandra Acosta, Borja Esteve-Altava, Arcadi Navarro, Gerard Muntané, Comparative Analysis of Mammal Genomes Unveils Key Genomic Variability for Human Life Span, Molecular Biology and Evolution, Volume 38, Issue 11, November 2021, Pages 4948–4961, https://doi.org/10.1093/molbev/msab219
Subsequent to the original publication1 of this article, we noted two mistakes in the calculations presented in Supplementary Tables 9 and 10.
First, enrichment in heritability of human parental longevity in the three sets of detected genes (genes harboring discovered CAAS, genes harboring validated CAAS, and genes presenting significant gene-phenotype coevolution with PGLS) was evaluated using the LDSC2 software. To do so, we computed the heritability explained by the SNPs harbored by genes in each of our three detected sets and measured enrichment relative to the SNPs in all the genes screened in our comparative genomic analysis. In our calculations, we attempted to trim GWAS data to evaluate heritability in the screened geneset only. However, the software still calculated the size of the partitions relative to the whole genome set of SNPs and so the estimation we reported was incorrect.
With invaluable help from the authors of LDSC we have now performed correct calculations, following Hujoel et al. 20193, of relative enrichment of the hit genesets (gene lists containing discovered CAAS, validated CAAS and PGLS-significant genes) relative to the background geneset (the set of genes evaluated in our paper). Our new estimations show that there is no significant enrichment in heritability for parental longevity in any of the genesets relative to the evaluated genes partition (ALLGENES) (p-value=0.98, 0.28, and 0.71 for the discovered CAAS, validated CAAS and PGLS gene lists, respectively). As the authors of LDSC reported4, one of the reasons for obtaining non-significant results is lack of statistical power due to the relative amount of SNPs we were testing. Basically, when applied to small sets of SNPs, the block jackknife-based significance testing used in S-LDSC does not always control for Type 1 error. In our case, we were evaluating 5.6%, 2.9%, and 1.6% of total SNPs, respectively, so the power we have might not be enough to test our hypothesis.
Second, we used the Pascal5 software as an additional method to evaluate heritability enrichment. The method calculates an enrichment score for pathways or sets of genes, again relative to all the genes tested, from SNP-phenotype association summary statistics. It generates a p-value of the significance of the scores in a pathway or a set of genes based either on the sum or the maximum of chi-squared statistics, which measure the strongest and the average association signals per gene, taking into account the LD of genes in the geneset. The authors tested the software extensively with many known biological pathways and so the method is well-calibrated for these. However, we noted that the p-values calculated by the software might not be well-calibrated for large gene sets (>50-300 genes). Therefore, again with help from Pascual´s authors, we have run 1,000 permutations of genesets the size of each of the gene lists (CAAS discovery, 2004 genes; CAAS validated, 996 genes; PGLS, 705 genes), with genes sampled at random from the ALLGENES set, and found empirical p-values of 0,15, 0.52 and 0.33 for the CAAS discovery, CAAS validated and PGLS lists.
Unfortunately, in one case due to a software failure to calculate the intended relative enrichment and in the other to insufficient calibration, both analysis mistakenly pointed in the same direction (heritability enrichment in the genes detected by our comparative genomics analyses) and so we were confused by the very exercise we had designed to double-check our results.
Because of the nature of these two errors, our results must be considered inconclusive and the statement that our lists of CAAS are enriched with GWAS heritability on parental longevity cannot be made anymore. While this does not affect any of the principal results and conclusions of our paper -in particular in what refers to the lists of candidate CAAS linked to longevity in mammals- the issue about whether macro and micro evolutionary changes in lifespan are mediated by the same genes remains undecided.
References
1. Farré X, Molina R, Barteri F, et al. Comparative Analysis of Mammal Genomes Unveils Key Genomic Variability for Human Life Span. Nowick K, ed. Mol Biol Evol. 2021;38(11):4948-4961. doi:10.1093/molbev/msab219
2. Finucane HK, Bulik-Sullivan B, Gusev A, et al. Partitioning heritability by functional annotation using genome-wide association summary statistics. Nat Genet. 2015;47(11):1228-1235. doi:10.1038/ng.3404
3. Hujoel MLA, Gazal S, Hormozdiari F, van de Geijn B, Price AL. Disease Heritability Enrichment of Regulatory Elements Is Concentrated in Elements with Ancient Sequence Age and Conserved Function across Species. Am J Hum Genet. 2019;104(4):611-624. doi:10.1016/j.ajhg.2019.02.008
4. Tashman KC, Cui R, O’Connor LJ, Neale BM, Finucane HK. Significance Testing for Small Annotations in Stratified LD-Score Regression. Genetic and Genomic Medicine; 2021. doi:10.1101/2021.03.13.21249938
5. Lamparter D, Marbach D, Rueedi R, Kutalik Z, Bergmann S. Fast and Rigorous Computation of Gene and Pathway Scores from SNP-Based Summary Statistics. PLOS Comput Biol. 2016;12(1):e1004714. doi:10.1371/journal.pcbi.1004714
Substantial changes have been made to the text of the article to accommodate the corrections identified by the authors.