Rare variants drive high variance in human ancestral fitness at mutation-selection-drift balance

It is an open question whether variation in the genetic load of unconditionally deleterious mutations contributes substantially to the variability in human disease. Here, we solve for mutation-selection-drift balance and predict variation in genetic load given a realistic human genome-wide deleterious mutation rate, U , and a distribution of fitness effects (DFE). Empirical estimates of U come from sequence constraint, which fails to count slightly deleterious mutations that nevertheless fix. We use the inferred DFE to correct for this and conclude that total human U >3.8. Two humans typically differ in ancestral fitness by 17-33% given uncertainty in U , or by 6-49% when we consider a broad range of alternative DFEs. Results are similar for other species with larger mean selection coefficients, such as other mammals. Most variation in load comes from rare variants with frequencies below 1%, with a substantial fraction coming from ultra-rare variants below 0.01%. This could help explain why some of the heritability observed in pedigree studies is missing from genome-wide association studies. Accounting for rare and ultra-rare variants, e.g., via variant-effect prediction of unique mutations from whole-genome sequencing rather than via polygenic risk scores, could help identify individuals at high risk of disease.