Comparison of the Distribution of Fitness Effects Across Primates

Key results

1. ShownLarger effective population size (Ne) moves population-scaled DFE mass from effectively neutral toward strongly deleterious classes, a pattern robust to parametrization, SFS folding, common-gene-set restriction and GC-conservative spectra.

   Grafen PGLS on 30 species: fraction S ∈ (−∞, −10] r = 0.82, p = 3.6 × 10^−8 (discrete) and r = 0.83, p = 1.6 × 10^−8 (gamma); fraction S ∈ (−1, 0] r = −0.72, p = 8.1 × 10^−6 and r = −0.70, p = 1.4 × 10^−5; moderate class (−10, −1] not significant (discrete r = −0.28, p = 0.14). · 3.1 The DFE across catarrhines; 2.9 Regression analysis; 3.6 GC-biased gene conversion; 3.7 Gene-set differences

2. ShownA tail-robust quantile diagnostic finds the unscaled deleterious DFE consistent with simple Ne rescaling, giving no evidence of epistasis across the ∼20-fold catarrhine Ne range.

   PGLS slope of log10 |Sq| on log10 Ne did not differ from the no-epistasis expectation of 1 at f = 0.4, 0.5 and 0.6 under all four model variants (all p > 0.05), e.g. slope 1.20, p = 0.20 for deleterious-only gamma at f = 0.6; a simulated fixed unscaled DFE rescaled by 4Ne gives slope exactly 1. · 3.3 Can variation in Ne explain the observed…; 2.9 Regression analysis; Tables

3. SuggestedUnder deleterious-only fits, predicted total dN/dS (ω) falls with Ne; polymorphism alone cannot resolve the adaptive rate ωa, but it rises with Ne once lineage-specific divergence is fit jointly, a trend that persists on GC-conservative sites under the discrete parametrization.

   Deleterious-only ω vs Ne: discrete r = −0.75, p = 1.9 × 10^−6. Polymorphism-only ωa: r = 0.16, p = 0.41 (gamma), r = −0.12, p = 0.52 (discrete). With divergence: ωa gamma r = 0.57, discrete r = 0.64 (both p < 2 × 10^−3); GC-conservative discrete ωa r = 0.65, p = 1.1 × 10^−4; 14 independent reference lineages. · 3.4 Substitution rates and adaptive evolution; 3.6 GC-biased gene conversion; 2.7 Inference from SFS and divergence counts

4. ShownReanalyzing the nine great apes of Castellano et al. (2019) reproduces their Ne association for strongly deleterious mutations but not a significant unscaled sd-Ne trend or outlier bonobos and western chimpanzees, which undercuts their positive-epistasis interpretation.

   Fraction S ∈ (−∞, −10] vs Ne: r = 0.89, p = 1.68 × 10^−2 (new data) vs r = 0.89, p = 1.82 × 10^−2 (original); unscaled sd vs Ne: r = 0.75, p = 8.27 × 10^−2 (old data), r = 0.44, p = 3.87 × 10^−1 (new data), turning weakly negative under independent DiscreteParametrization fits. · 3.5 Comparison with previous estimates; 4 Discussion

5. ShownModeling partial recessivity that intensifies with selection strength shifts inferred DFEs only modestly and uniformly, leaving Ne associations intact; simulations show h is weakly identifiable because it trades off against Sd.

   Sensitivity model h(S) = 0.4 exp(−0.02|S|), with h(−100) ≈ 0.05, makes DFEs slightly more deleterious across species. In SLiM runs, simulating h = 0.2 but fitting h = 0.5 biases toward weaker selection; in fastDFE-simulated spectra (Sd = −300, b = 0.3, sample size 20), joint (h, Sd) bootstraps show strong negative covariance. · 3.8 Dominance; Demographic scenarios and model violations; 4 Discussion