Evolutionary constraint and innovation across hundreds of placental mammals

Key results

1. ShownSampling 240 species provides enough branch length to call constraint at single human bases; at least 10.7% of the human genome appears under purifying selection, versus 5.5% estimated earlier from 29 mammals.

   3.6 million perfectly conserved positions vs 191 expected by chance; ≥10.7% (332 Mb) constrained; single-base calls at FDR<0.05 cover 3.53% (101 Mb); non-degenerate vs four-fold degenerate coding sites constrained 74.1% vs 18.5%; TOPMed variants at constrained bases have lower minor allele frequency (0.0026 vs 0.004 at unconstrained bases; Wilcoxon p = 9.5×10^−13). · Evolutionary constraint in humans; Single-base resolution of constraint

2. ShownStringent constraint thresholds expand the ultraconserved element catalog (zooUCEs) more than seven-fold, and 424,179 unannotated intergenic constrained regions (UNICORNs) with depleted human variation suggest function in contexts ENCODE3 did not assay.

   4,552 zooUCEs (≥20 bp, identical in ≥235 species; 3,799 new) vs 481 original UCEs; 48.5% of constrained bases lack ENCODE3 annotation; UNICORNs have lower minor allele frequencies and fewer TOPMed variants than random unannotated sequence (pWilcoxon<2.2 x 10^−16); 17% of UNICORNs overlap brain open chromatin; 53 of 28,218 100kb bins significantly constrained (q<0.05). · Abstract; Regions of constraint; Unannotated constraint; Figure legends

3. ShownCross-species phenotype analyses tie olfactory receptor repertoire size to nasal turbinal counts, and find coding regions where deep hibernators stay closer to the reconstructed ancestor, including mitophagy-related MFN2 and PINK1, plus genes whose evolutionary rates shift in hibernators.

   Olfactory receptor count vs turbinal count: Spearman rho=0.71, phylolm coefficient 0.014, phylogenetic permutation p = 0.0013 (N=63). Hibernation: 14 regions in 11 genes more conserved in 22 deep hibernators than 154 homeotherms (pFDR<0.01); RERconverge: 764 rate-shifted genes (511 slower, 253 faster) in 21 hibernators vs 146 non-hibernators (pFDR<0.05). · Olfaction; Hibernation

4. ShownProjecting human, mouse and goat annotations through the Cactus alignment yields gene models for 238 placental species (129 lacking NCBI or Ensembl annotation); TOGA (Tool to infer Orthologs from Genome Alignments) improves completeness over NCBI annotations for most benchmarked assemblies.

   Per genome, 3,871 to 20,168 (mean 12,905 ± 2,925) of 22,799 input genes annotated, declining with lower scaffold N50 (Pearson r=0.37, p=3×10^−9) and greater phylogenetic distance from human (r=-0.41, p=7×10^−11); TOGA: mean 19,144 orthologs for ape vs 17,779 for non-ape assemblies; completeness improved for 108 of 118 NCBI-annotated assemblies. · Annotating genes through alignment; New 428-way protein-coding alignment in mammals

5. ShownHeterozygosity, long-term effective population size (Ne), recent decline, missense load and conserved-site mutation counts differ on average between International Union for Conservation of Nature (IUCN) threatened and other species, though no single metric is diagnostic.

   Threatened species have lower heterozygosity (phylogenetic regression p=0.016), smaller long-term Ne (p<3.30×10^−5), larger recent declines (p=0.012), higher missense proportions (p=3.63×10^−4), and less mutation at the most conserved positions (phyloP kurtosis, p=0.004); modeled threat probability 12-14% for Upper Galilee Mountains blind mole rat vs 53-69% for killer whale. · Biodiversity