Positive Selection

Positive selection (also termed directional selection or Darwinian adaptive selection) is the evolutionary regime in which newly arisen or standing genetic variants that confer a survival or reproductive advantage (s > 0) systematically increase in allele frequency within a population until they reach fixation (p = 1.0). Grounded mathematically in the population-genetic syntheses of Ronald Fisher, J. B. S. Haldane, and Sewall Wright (1920s–1930s) and extended to DNA and protein alignments through molecular evolution and codon substitution models, positive selection is the primary mechanistic driver of adaptive phenotypic innovation, host–pathogen evolutionary arms races, and functional divergence following gene duplication.

Within evolutionary genomics and statistical phylogenetics, positive selection leaves distinct quantitative footprints at two complementary timescales. Over deep macroevolutionary branches (millions of years), recurrent positive selection accelerates the rate of non-synonymous amino acid replacements (d_N) above the neutral baseline rate of synonymous substitutions (d_S), yielding an elevated ratio ω = d_N / d_S > 1.0. Over recent microevolutionary timescales (tens to thousands of generations within a single species), the rapid transit of a beneficial allele to fixation drags linked neutral variants along with it (genetic hitchhiking), carving a selective sweep characterized by depleted nucleotide diversity, a skewed site-frequency spectrum, and extended haplotype homozygosity.

Haldane's Fixation Probability (u ≈ 2*s*), Logistic Trajectories, and Soft vs. Hard Sweeps

Consider a diploid population of effective size N_e in which a new mutation arises as a single copy (initial frequency p₀ = 1 / 2*N*_e) with heterozygote fitness 1 + s and homozygote fitness 1 + 2*s* (additive selection, h = 0.5). Using branching-process and diffusion theory, J. B. S. Haldane (1927) and Motoo Kimura (1962) showed that the probability u(s) that this single advantageous copy escapes early stochastic loss by random genetic drift and reaches fixation is:

Thus, a beneficial allele with a 1% selective advantage (s = 0.01) has an ~2% chance of fixing from a single copy—orders of magnitude higher than a neutral allele's fixation probability of 1 / (2*N*_e). Once a beneficial allele escapes drift and enters the deterministic phase, the ratio of beneficial to ancestral alleles grows according to a logistic growth phase p(t) / (1 − p(t)) = [ p₀ / (1 − p₀) ] exp(st), completing its transit from frequency 1/(2*N*_e) to 1 − 1/(2*N*_e) in a mean sojourn time of only t_fix ≈ (2 / s) ln(4*N*_e s) generations—far faster than the 4*N*_e generations required for a neutral allele to drift to fixation.

As conceptualized by John Maynard Smith and John Haigh (1974), this rapid fixation drives genetic hitchhiking: neutral polymorphic variants physically linked on the same chromosome within a recombination distance r < s do not have time to recombine off the adaptive background and are swept to fixation alongside the beneficial mutation. A hard selective sweep (originating from a single de novo mutation) purges surrounding genetic variation, leaving a valley of reduced Watterson θ_W and nucleotide diversity π flanked by an excess of rare derived singletons (negative Tajima's D < −2.0, positive Fay and Wu's H) and a single long-range haplotype block measured by Extended Haplotype Homozygosity (EHH) and the Integrated Haplotype Score (iHS). Conversely, when positive selection acts on preexisting standing genetic variation or recurrent beneficial mutations, multiple haplotypes rise simultaneously in a soft selective sweep.

Goldman–Yang Codon Likelihoods, Site Models (M1a/M2a, M7/M8), and Branch-Site Tests

In protein-coding genes, synonymous nucleotide substitutions (d_S, which do not change the encoded amino acid) accumulate largely neutrally (d_S ≈ μ₀), providing an internal benchmark against which non-synonymous amino acid replacements (d_N) are compared via the ratio ω = d_N / d_S:

• Purifying (Negative) Selection (ω < 1.0): Deleterious amino acid changes are purged; typical housekeeping proteins exhibit whole-gene averages of ω ≈ 0.02 to 0.25.
• Neutral Evolution (ω = 1.0): Non-synonymous changes fix at the exact same rate as synonymous changes, characteristic of nonfunctional pseudogenes.
• Positive Darwinian Selection (ω > 1.0): Advantageous amino acid replacements fix faster than neutral drift (d_N > d_S).

Because structural and catalytic core residues of a functional protein remain under strong purifying constraint (ω << 1) even when surface or active-site loops experience intense adaptive evolution, averaging d_N/d_S across an entire gene length almost never exceeds 1.0. To detect localized adaptation, Ziheng Yang, Rasmus Nielsen, and Nick Goldman formulated maximum-likelihood codon substitution models (implemented in PAML codeml and HyPhy MEME/BUSTED using the Felsenstein pruning algorithm) that allow ω to vary across individual codon sites (site models) or across both sites and specific phylogenetic branches (branch-site models).

A Likelihood Ratio Test (LRT) compares twice the log-likelihood difference 2Δℓ = 2(ℓ₁ − ℓ₀) between a null model forbidding positive selection—such as M1a (Nearly Neutral) with two site classes (ω₀ < 1, ω₁ = 1) or M7 (beta distribution) on [0, 1]—and an alternative model adding an extra site class with ω₂ > 1 (M2a Positive Selection or M8 beta + ω > 1). When 2Δℓ exceeds the χ² critical value, Bayes Empirical Bayes (BEB) posterior probabilities identify the exact codon residues under positive selection.

Uniting Polymorphism and Divergence (D_N/D_S vs. P_N/P_S)

To bridge population genetics and macroevolutionary divergence within a single test, John McDonald and Martin Kreitman (1991) compared the ratio of non-synonymous to synonymous fixed differences between species (D_N / D_S) against the ratio of non-synonymous to synonymous polymorphisms segregating within species (P_N / P_S) in a 2 × 2 contingency table. Under strict neutrality, both ratios equal the fraction of neutral sites (D_N/D_S = P_N/P_S). However, because advantageous mutations fix rapidly via selective sweeps, they contribute minimally to transient within-species polymorphism (P_N) while accumulating substantially in inter-species divergence (D_N). Thus, D_N / D_S > P_N / P_S demonstrates recurrent positive selection and quantifies the proportion of amino acid substitutions driven by positive selection:

Across empirical genomic datasets, the strongest signatures of positive selection occur in coevolutionary host–pathogen arms races (Red Queen dynamics)—including vertebrate Major Histocompatibility Complex (MHC) peptide-binding grooves, antiviral restriction factors (TRIM5α, PKR, APOBEC3G), and viral surface glycoproteins in virology—alongside reproductive gamete-recognition proteins (abalone lysin andVERL), venom toxins (cone snail conotoxins and snake phospholipases), and recent human physiological adaptations such as lactase persistence (LCT), high-altitude hypoxia tolerance (EPAS1 in Tibetans), and malaria resistance (HBB HbS and DARC Duffy-null alleles).

Why is a whole-gene average d_N/d_S > 1 rare even in genes undergoing adaptive evolution?

In almost every functional protein, the majority of amino acid residues form the hydrophobic folding core or catalytic scaffold and are subject to strong purifying selection (ω < 0.05) to prevent misfolding. Positive selection typically acts on only a small subset of 3 to 15 surface or ligand-contacting codons (where ω may reach 3 to 15). Averaging ω across all 300 to 1,000 codons dilutes the adaptive signal below 1.0, which is why site-specific and branch-site codon models (such as PAML M8 and HyPhy MEME) are required.

How does the McDonald–Kreitman test distinguish positive selection from relaxed purifying selection?

An elevated inter-species d_N/d_S ratio on a phylogenetic branch can result either from genuine positive selection or from relaxed purifying selection (for example, during a population bottleneck where slightly deleterious mutations drift). The McDonald–Kreitman test separates the two by examining within-species polymorphism: relaxed purifying selection increases both within-species non-synonymous polymorphism (P_N/P_S) and between-species divergence (D_N/D_S) equally, whereas positive selection elevates fixed divergence (D_N/D_S) well above standing polymorphism (P_N/P_S).

What is the difference between a hard selective sweep and a soft selective sweep?

A hard selective sweep occurs when a single newly arisen beneficial mutation sweeps to fixation, replacing all ancestral chromosomes in the surrounding genomic region with a single homogeneous haplotype and eliminating linked nucleotide diversity. A soft selective sweep occurs when positive selection acts on an allele already present on multiple haplotype backgrounds in standing genetic variation (or via recurrent mutation at a high population mutation rate 2*N*_eμ > 0.01), carrying several distinct haplotypes to high frequency simultaneously and preserving moderate linked diversity.

1. Maynard Smith, J., & Haigh, J. (1974). "The hitch-hiking effect of a favourable gene." Genetical Research, 23(1), 23–35. https://doi.org/10.1017/S0016672300014634
2. McDonald, J. H., & Kreitman, M. (1991). "Adaptive protein evolution at the Adh locus in Drosophila." Nature, 351(6328), 652–654. https://doi.org/10.1038/351652a0
3. Nielsen, R., & Yang, Z. (1998). "Likelihood models for detecting positively selected amino acid sites and applications to the HIV-1 envelope gene." Genetics, 148(3), 929–936. https://doi.org/10.1093/genetics/148.3.929
4. Sabeti, P. C., Reich, D. E., Higgins, J. M., Levine, H. Z. P., Richter, D. J., Schaffner, S. F., Gabriel, S. B., Platko, J. V., Patterson, N. J., McDonald, G. J., Ackerman, H. C., Campbell, S. J., Altshuler, D., Cooper, R., Kwiatkowski, D., Ward, R., & Lander, E. S. (2002). "Detecting recent positive selection in the human genome from haplotype structure." Nature, 419(6909), 832–837. https://doi.org/10.1038/nature01140