Selective Sweep

A selective sweep is a process in population genetics and molecular evolution in which strong positive natural selection (s > 0) drives a beneficial allele rapidly to high frequency or fixation (p → 1) within a population, carrying along the linked neutral (or nearly neutral) DNA sequence variants sitting on the same ancestral chromosome via genetic hitchhiking and thereby purging standing nucleotide diversity (π) across a surrounding chromosomal window.

First modeled mathematically in 1974 by John Maynard Smith and John Haigh in their classic paper "The hitch-hiking effect of a favourable gene" and extended to coalescent theory and genome-wide scan statistics by Richard Hudson, Wolfgang Stephan, Charles Langley, and Rasmus Nielsen, selective sweeps leave diagnostic footprints in present-day genomes—including valleys of reduced heterozygosity, an excess of rare derived single-nucleotide polymorphisms (negative Tajima's D), and extended haplotype homozygosity (EHH)—that allow geneticists to pinpoint the specific loci underlying recent human adaptation (such as LCT lactase persistence, SLC24A5 skin pigmentation, and EPAS1 Tibetan high-altitude hypoxia tolerance), agricultural domestication, and antimicrobial or insecticide resistance.

🧬 Interactive Population Genetics Simulation: Track logistic fixation trajectories and selective sweeps under positive directional selection (s > 0) in Allele Drift.

Deterministic Sojourn Time of a Beneficial Mutation

Consider a new beneficial allele B with additive heterozygous selection coefficient s > 0 (where relative genotypes bb, Bb, and BB have fitnesses 1, 1 + s, and 1 + 2*s*, with 4*N*_es ≫ 1) arising as a single copy (p₀ = 1 / [2*N*_e]) in a diploid population of effective population size N_e. Once allele B survives initial stochastic drift loss (probability Π ≈ 2*s* under the Wright–Fisher model), its frequency p(t) follows the continuous logistic differential equation:

d*p* / d*t* = s · p (1 − p) ⟹ p(t) / [1 − p(t)] = [ p₀ / (1 − p₀) ] · exp(s t)

Integrating from initial establishment (p₀ ≈ 1 / [2*N*_e]) to near-fixation (p_final = 1 − 1 / [2*N*_e]) yields the total sweep duration (fixation time T_sweep):

T_sweep ≈ ( 2 / s ) · ln( 4 N_e s )

Compare this selective sweep duration against the 4*N*_e generations required for a neutral allele to drift to fixation: for a human population with N_e = 10,000 and a strong beneficial allele with s = 0.05 (5% advantage per generation), neutral drift fixation takes 4*N*_e = 40,000 generations (~1,000,000 years), whereas a selective sweep completes in T_sweep ≈ (2 / 0.05) ln(2,000) ≈ 304 generations (~7,600 years)—more than 130 times faster!

Genetic Hitchhiking, Recombination Escape (r), and the Footprint Width (d ∝ s/r)

Because the single ancestral chromosome on which mutation B first arose carries a specific combination of pre-existing neutral single-nucleotide polymorphisms (SNPs) at neighboring linked sites (a single ancestral haplotype), if zero genetic recombination (crossing over) occurred during the T_sweep generations, the entire chromosome carrying B would replace all other chromosomes in the population, reducing nucleotide diversity (π) to zero along the entire chromosome.

In a recombining sexual organism, however, a neutral locus located at a genetic map distance r (recombination fraction per generation, in Morgans or cM/100) away from the selected site B can escape the sweep if a meiotic crossover occurs in a Bb heterozygote during the sweep—recombining allele B onto a different background chromosome from the ancestral population and preserving standing diversity at that neutral distance. Integrating the probability of no recombination across the logistic trajectory of Bb heterozygotes, John Maynard Smith and John Haigh (1974) and Wolfgang Stephan, Thomas Wiehe, and Marcus Lenz (1992) showed that the expected post-sweep heterozygosity H(r) at recombination distance r relative to the pre-sweep equilibrium diversity H₀ = θ = 4*N*_eμ is:

H(r) / H₀ ≈ 1 − ( 2 N_e s )^{−2*r* / s} ≈ 1 − exp[ − (2 r / s) ln(2 N_e s) ] = 1 − exp( − r · T_sweep )

Consequently, the characteristic chromosomal width of the heterozygosity valley (H(r)/H₀ = ½) scales directly with the selection coefficient (s) and inversely with the local recombination rate per base pair (ρ_rec, typically ~10⁻⁸ per bp per generation in humans):

Sweep Window Width (L_bp) ≈ ( 0.1 to 0.2 ) · ( s / ρ_rec )

• A strong selective sweep (s = 0.05) in a human region with average recombination (ρ_rec = 10⁻⁸ bp⁻¹ gen⁻¹) carves out a wide ~500 kb to 1.0 Mb valley of depleted diversity (such as the ~1 Mb haplotype surrounding the European lactase persistence allele LCT).
• A moderate sweep (s = 0.005) or a sweep occurring in a recombination hotspot leaves a narrow ~20–50 kb footprint.

Modern evolutionary genomics distinguishes three distinct regimes by which natural selection shapes linked neutral variation:

Why a Hard Selective Sweep Creates a "Star-Like" Coalescent Tree

Viewed backward in time through Kingman's coalescent, a neutral locus in a constant-size population has a deep, hierarchical bifurcating genealogy whose total time to the Most Recent Common Ancestor averages T_MRCA ≈ 4*N*_e generations, with half of that total time spent waiting for the final two ancestral lineages (k = 2 → 1) to coalesce—producing an equilibrium Site Frequency Spectrum (SFS) where the number of segregating sites at derived allele count i ∈ {1, …, n − 1} scales as ξ_i = θ / i.

By contrast, directly at the core of a recently completed hard selective sweep, every chromosome in the present-day sample descends from the single individual in which the beneficial mutation B arose just T_sweep ≪ 4*N*_e generations ago. Tracing the genealogy of n sampled chromosomes backward in time, all n lineages are forced to coalesce nearly simultaneously inside the narrow temporal bottleneck at the base of the sweep (a "star-like" genealogy with short terminal twigs of length ≤ T_sweep and zero deep internal branches):

1. Tajima's D (< 0 at Core and Flanks): Formulated by Fumio Tajima in 1989, Tajima's D compares pairwise nucleotide diversity (π, which weights intermediate-frequency alleles heavily) against Watterson's θ_W = S / ∑(1/i) (which weights all segregating sites equally, including rare singletons):

D = ( π − θ_W ) / √Var(π − θ_W) After a hard sweep wipes out all pre-existing variation at the core, every new neutral mutation that arises on the post-sweep star genealogy sits on a single terminal twig (a rare singleton, i = 1), inflating θ_W relative to π and driving Tajima's D strongly negative (D < −2.0).

1. Fay and Wu's H (< 0 in Recombining Flanks): In the flanking regions just outside the zero-diversity core where one or two crossovers occurred during the sweep, ancestral linked neutral variants are dragged up to very high derived allele frequencies (80%–95%), producing a strongly negative Fay and Wu's H statistic.
2. Extended Haplotype Homozygosity (EHH, iHS, and XP-EHH): Introduced by Pardis Sabeti and Eric Lander in 2002 (and standardized into the Integrated Haplotype Score, iHS, by Benjamin Voight and Jonathan Pritchard in 2006), haplotype tests measure how far linkage disequilibrium (haplotype identity) extends outward from a focal allele before being broken up by historical recombination. For an incomplete (ongoing) sweep that has reached 60%–85% frequency in just a few thousand years, the young beneficial allele sits on an exceptionally long, unbroken haplotype block (hundreds of kilobases) compared to the short, recombined haplotype blocks surrounding old neutral alleles of the exact same frequency!

1. Lactase Persistence (LCT / MCM6, Chromosome 2q21): In most mammals, intestinal transcription of the lactase-phlorizin hydrolase gene (LCT, which cleaves milk lactose into glucose and galactose) is downregulated after weaning. With the rise of dairy pastoralism in the Neolithic (~5,000–9,000 years ago), a regulatory enhancer mutation −13910\*T inside intron 13 of the upstream MCM6 gene underwent one of the strongest hard selective sweeps in the human genome across northern Europeans (s ≈ 0.04 to 0.08, sweeping from <1% to >75%–90% in ~200 generations and leaving a >1 Mb homozygous haplotype block), while pastoralist populations in East Africa (Masai, Tutsi, Beja) evolved lactase persistence via convergent soft/hard sweeps on three completely independent nearby enhancer mutations (−14010\*C, −13907\*G, and −13915\*G).
2. Malaria Resistance at the Duffy Antigen Receptor (ACKR1 / DARC FY\*O) and Sickle Cell (HBB): The malaria parasite Plasmodium vivax requires the erythrocyte Duffy chemokine receptor to invade human red blood cells. A GATA-1 promoter mutation (c.−67T>C, the Duffy-null *FY\*O* allele) that abolishes Duffy expression specifically on red blood cells (while preserving expression on endothelial cells) underwent a complete hard selective sweep to ~98%–100% fixation across sub-Saharan Africa, eradicating P. vivax across West and Central Africa.
3. Tibetan High-Altitude Adaptation at EPAS1 (Adaptive Introgression Sweep): Living at >4,000 m elevation on the Tibetan Plateau (where atmospheric O₂ partial pressure is 40% lower than at sea level), Tibetans avoid chronic mountain sickness (polycythemia and pulmonary hypertension) via a 32.7-kb haplotype in the hypoxia-inducible transcription factor gene EPAS1 (HIF-2α) that rose from <3% in lowland Han Chinese to >85% in Tibetans over ~10,000 years—shown by Emilia Huerta-Sánchez and Rasmus Nielsen (2014) to have entered the ancestral Tibetan gene pool via adaptive introgression from archaic Denisovan hominins.

How can researchers distinguish a true selective sweep from a severe population bottleneck?

A genome-wide population bottleneck (such as the Out-of-Africa human migration bottleneck ~60,000 years ago, or a domestic crop domestication bottleneck) and a hard selective sweep both reduce nucleotide diversity (π), increase linkage disequilibrium, and skew Tajima's D during post-bottleneck population expansion. The decisive difference is genomic scope: a demographic bottleneck or population expansion affects the coalescent genealogy of every locus across all 22 autosomes simultaneously, whereas positive natural selection acts locus-specifically on the single adaptive gene and its immediate linked physical neighborhood (L ≈ 0.1 s/ρ_rec). Therefore, modern genome scans first fit a genome-wide demographic null model from thousands of putatively neutral intergenic loci, and then identify selective sweeps as extreme locus-specific outliers (>99.9th percentile of |iHS|, XP-EHH, or composite likelihood ratio CLR) that deviate sharply from the rest of the genome.

Why do rapidly evolving pathogens like HIV-1 and Plasmodium falciparum frequently evolve drug resistance via "soft" sweeps rather than "hard" sweeps?

Whether an adaptive trait sweeps as a single-origin hard sweep or a multi-origin soft sweep is governed by the population-scaled adaptive mutation rate Θ_b = 2 N_e μ_b (where μ_b is the per-generation mutation rate producing the resistant phenotype):

• When Θ_b ≪ 0.01 (small population size N_e or an adaptation requiring one ultra-specific base substitution with μ_b ≈ 10⁻⁹), the population must wait many generations for the first lucky adaptive mutation to arise and establish; that first mutant sweeps to fixation before a second independent mutation ever establishes (a hard sweep).
• When Θ_b ≥ 0.1 to 1.0—as occurs in an untreated HIV-1 patient (where billions of virions are produced daily at a reverse-transcriptase point mutation rate of μ ≈ 3 × 10⁻⁵ per base) or across continental populations of Anopheles mosquitoes exposed to pyrethroid insecticides—multiple independent copies of the exact same drug-resistance mutation (such as the K103N mutation in HIV-1 reverse transcriptase upon efavirenz therapy) arise simultaneously on different haplotype backgrounds within days, sweeping together as a multi-origin soft sweep.

What is "Hill–Robertson interference" (clonal interference), and why did sexual recombination evolve to overcome it?

In an asexual population (or a chromosomal region with zero genetic recombination, such as the mammalian Y chromosome), suppose two different beneficial mutations (A at locus 1 and B at locus 2) arise simultaneously in two different individuals (genotype Ab and genotype aB). Without meiotic crossing over to combine them onto the same chromosome (AB), the two beneficial Selective Sweeps compete directly against each other (clonal interference, or the Hill–Robertson effect, formulated by William G. Hill and Alan Robertson in 1966): for Ab to reach fixation, it must drive the other beneficial mutation aB to extinction! Furthermore, if a moderately beneficial mutation arises on a chromosome that also happens to carry linked weakly deleterious mutations (ruby-in-the-rubbish), selection efficiency is crippled across the genome. Sexual recombination breaks down negative linkage disequilibrium between beneficial alleles on different chromosomes, allowing two simultaneous selective sweeps (A and B) to recombine into the double-mutant AB haplotype and fix in parallel.

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. Sabeti, P. C., Reich, D. E., Higgins, J. M., Levine, H. Z. P., Richter, D. J., Schaffner, S. F., ... & 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
3. Hermisson, J., & Pennings, P. S. (2005). "Soft sweeps: molecular population genetics of adaptation from standing genetic variation." Genetics, 169(4), 2335–2352. https://doi.org/10.1534/genetics.104.036947
4. Messer, P. W., & Petrov, D. A. (2013). "Population genomics of rapid adaptation by soft selective sweeps." Trends in Ecology & Evolution, 28(11), 659–669. https://doi.org/10.1016/j.tree.2013.08.003