Selective Sweep

How a selective sweep works

Imagine a population of chromosomes, each a slightly different string of DNA. Most of the differences between them are neutral — single-nucleotide variants that do nothing. Now a beneficial mutation appears on exactly one of those chromosomes. That chromosome already carries a particular set of neutral variants, a specific haplotype. Natural selection cannot pick out the one beneficial base and copy it alone; it can only favor the whole chromosome. So as the beneficial allele rises from one copy to high frequency, every neutral variant riding on the same haplotype rises with it. The variants that happened to sit on rival chromosomes are driven out. By the time the favored allele reaches fixation, the population is nearly uniform around the selected site — that is the selective sweep.

John Maynard Smith and John Haigh formalized this in 1974 in a paper titled "The hitch-hiking effect of a favourable gene." Their key insight was that selection at one locus has collateral consequences for linked loci, and that the reach of those consequences is set by a tug-of-war between selection and recombination. Selection drives the favored haplotype up; recombination tries to peel neutral variants off it and onto other backgrounds. The closer a neutral site sits to the selected locus, the harder it is for recombination to rescue its variation before the sweep completes. The farther away, the more chances recombination has. The outcome is a V-shaped valley of diversity centered on the selected site, deep in the middle and recovering toward the edges.

Three things have to be true for a strong, clean sweep. The selection must be strong (a large selection coefficient s) so the rise is fast. The mutation must be new and on a single background (a hard sweep) so there is only one winning haplotype. And recombination near the locus must be modest, so it cannot break the linkage in time. Loosen any of these and the sweep softens, broadens, or fades.

The genomic signature step by step

A completed hard sweep stamps a recognizable, three-part fingerprint on the genome. Read in order, these are the clues geneticists use to find sweeps without ever seeing them happen.

• 1. A valley of reduced diversity. Nucleotide diversity, written π (pi) — the average number of pairwise differences per site — collapses near the selected locus. In a deep sweep it can fall close to zero at the center and rise back to the genome-wide background over hundreds of kilobases. Heterozygosity (the fraction of individuals carrying two different alleles) drops the same way.
• 2. A skewed site-frequency spectrum. Right after the sweep, the swept region is a single haplotype, so almost no variation exists. New mutations then start accumulating, and because they are all young, they are rare — an excess of low-frequency variants. This drives Tajima's D strongly negative. Meanwhile, the few neutral variants that hitchhiked all the way up are now at high frequency in the derived (mutant) state — an excess of high-frequency derived alleles that drives Fay and Wu's H negative. The pair together is a much stronger signal than either alone.
• 3. Long, shared haplotypes. Because the winning haplotype spread faster than recombination could chop it up, many individuals carry the same long unbroken stretch of identical DNA around the locus. This long-range linkage disequilibrium is measured by extended haplotype homozygosity (EHH) and summarized as iHS (within one population) and XP-EHH or Rsb (comparing two populations). Long haplotypes are the best signal for recent or ongoing sweeps, because the diversity valley needs time to fully form.

The first two signatures fade as time and mutation refill the valley. The third fades fastest of all because recombination steadily erodes long haplotypes — which is why long-haplotype tests in humans detect mostly sweeps from the last ~30,000 years, while the diversity valley can reveal older events.

When sweeps happen — players and conditions

Selective sweeps are the genomic footprint of strong, recent, positive selection. They show up whenever an environment changes fast enough that one allele confers a large advantage:

• New diets and lifestyles. Lactase persistence is the textbook case: the spread of dairying gave milk-digesters a major caloric and nutritional edge, and the -13910*T regulatory allele swept across Europe in roughly 350 generations, leaving a ~1 Mb haplotype.
• Pathogen pressure. Malaria has driven some of the strongest known human sweeps. The Duffy-null FY*O allele, which switches off the Duffy antigen receptor that Plasmodium vivax uses to enter red blood cells, reached near-fixation across sub-Saharan Africa. Sickle-cell and G6PD-deficiency alleles show related, more complex selection signatures.
• New altitudes and climates. The EPAS1 allele that blunts the overproduction of red blood cells at altitude reached ~87 percent in Tibetans — and was introgressed from Denisovans, an archaic human group. Light-skin pigmentation swept in Europeans at SLC24A5 (the Ala111Thr / A111T change, rs1426654), one of the largest-effect pigmentation loci known.
• Human-imposed selection. Pesticide and insecticide resistance produces rapid sweeps in pests — the kdr sodium-channel mutations and Ace-1 acetylcholinesterase variants in Anopheles and Drosophila. Domestication left sweep signatures at coat-color, body-size and behavior loci in dogs, cattle and maize (e.g. tb1 for plant architecture).
• Standing variation and recurrence. When the beneficial allele is already common before selection starts, or arises many times, you get a soft sweep — several haplotypes rising together. Soft sweeps dominate in large, mutation-rich populations and fast-adapting microbes and viruses.

Hard sweep vs soft sweep

Property	Hard sweep	Soft sweep
Origin of beneficial allele	Single new mutation, one copy	Standing variation or recurrent mutation
Number of winning haplotypes	One	Two or more
Diversity reduction	Near-total at the center	Partial — several backgrounds survive
Tajima's D signal	Strongly negative	Weak or near zero
Long-haplotype signal (iHS/EHH)	Strong, single dominant haplotype	Diffuse across multiple haplotypes
Detectability	Easy with classical scans	Hard — often missed by standard tests
Favored by	Small populations, low mutation supply, novel environments	Large populations, high mutation supply, common targets
Typical example	Lactase persistence, Duffy-null	Insecticide resistance in mosquitoes, viral drug resistance

Selective sweep vs other ways diversity is lost

A sweep is not the only process that erases variation. Two others look superficially similar in some statistics, and telling them apart is the central challenge of detecting selection.

Property	Selective sweep	Genetic drift / bottleneck	Background selection
Cause	Positive selection on one linked locus	Random sampling; population crash	Purifying selection removing deleterious mutations
Genomic scale	Localized — one valley at one locus	Genome-wide	Localized to low-recombination, gene-dense regions
Diversity effect	Sharp local valley of π	Uniform reduction everywhere	Gradual reduction, no sharp center
Site-frequency spectrum	Excess rare + excess high-freq derived	Excess rare only (across whole genome)	Excess rare, milder
Long shared haplotypes	Yes — one dominant haplotype	No specific haplotype enrichment	No
Population specificity	Often one population (local adaptation)	Affects all loci in that population	Same across populations
Key discriminator	Local outlier vs genome background	Genome-wide flatness	Correlates with recombination + gene density, not with a single peak

The numbers — how wide, how fast, how strong

The reach of a sweep is governed by the ratio of the selection coefficient s to the recombination rate r between the selected site and a neutral site. As a rule of thumb, diversity is substantially depressed out to a genetic distance on the order of s. Concretely:

• Selection coefficients. "Strong" human sweeps run s ≈ 0.01–0.10 — a 1 to 10 percent fitness advantage per generation. Lactase persistence is estimated around s ≈ 0.01–0.10 depending on the study; some malaria-resistance loci are even higher.
• Time to fixation. A new additive allele with selection coefficient s in a population of effective size N_e typically fixes in roughly (2/s)·ln(2N_e) generations. For s = 0.05 and N_e ≈ 10,000, that is on the order of ~400 generations — a few thousand years in humans (~25–30 years per generation). Fast enough that recombination cannot keep up nearby.
• Window width. Human recombination averages ~1 centimorgan per megabase (cM/Mb), i.e. ~1 crossover per 100 Mb per generation. A sweep with s = 0.05 flattens diversity out to roughly the distance where r ≈ s, which is on the order of hundreds of kilobases to ~1 Mb — exactly the span of the observed lactase-persistence haplotype.
• Recombination matters as much as selection. The same s in a recombination cold spot clears a far wider region; in a hotspot, a much narrower one. This is why sweep windows vary by an order of magnitude across the genome.
• Detection horizon. Long-haplotype statistics (iHS, XP-EHH) in humans are sensitive to sweeps younger than ~30,000 years (~1,000–1,200 generations); the diversity valley and SFS-based scans (SweepFinder) can reach somewhat older events before mutation refills the signal.

Worked example — sizing the lactase sweep

Take the European lactase-persistence allele -13910*T. Suppose it confers a per-generation advantage of s = 0.05 in a dairying population of effective size N_e ≈ 10,000. The expected time to fixation is about (2/s)·ln(2N_e) = (2/0.05)·ln(20,000) ≈ 40 × 9.9 ≈ ~400 generations, or roughly 10,000–12,000 years at ~25–30 years per generation — squarely consistent with the spread of dairying since the Neolithic.

How wide a region should hitchhike? Diversity is cleared out to about the map distance where the recombination rate equals the selection coefficient. At ~1 cM/Mb, a genetic distance of r = s = 0.05 cM corresponds to ~0.05 Mb on one side and similar on the other — but because the sweep is fast and the relationship is approximate, the empirically observed strong-LD haplotype extends to roughly 1 Mb, among the longest in the human genome. The allele itself reached ~70 percent or more in northern Europeans, and the same selective pressure produced independent persistence alleles (such as -14010*C) that swept separately in African pastoralists — convergent adaptation at one gene, with three or more distinct sweep signatures around the world.

Where sweeps show up in real organisms

• Humans — recent local adaptation. Lactase persistence (LCT/MCM6), Duffy-null malaria resistance (FY/DARC), Tibetan altitude tolerance (EPAS1, introgressed from Denisovans; EGLN1), light pigmentation (SLC24A5, SLC45A2, KITLG), and ethanol/arsenic-metabolism loci all carry sweep signatures of varying age and strength.
• Drosophila. A founding system for sweep theory. Sweeps at loci like the insecticide-resistance gene Ace and the Cyp6g1 DDT-resistance allele are textbook examples; Drosophila's large effective population size makes soft sweeps common, which reshaped the field's expectations.
• Pathogens and pests. Drug-resistance alleles sweep through malaria parasites (pfcrt, dhfr for chloroquine and pyrimethamine resistance), bacteria (antibiotic resistance), and viruses. Insecticide resistance (kdr, Ace-1) sweeps repeatedly through mosquito vectors — often softly.
• Domestication. Maize architecture (tb1, teosinte branched1), dog coat and body-size loci, chicken thyroid-stimulating-hormone receptor (TSHR), and many crop "domestication syndrome" genes sit under sweep signatures left by thousands of years of artificial selection.
• Conservation and disease genomics. Sweep scans help map adaptation in threatened species and identify candidate loci for human disease where selection on a pathogen-defense or metabolic allele also nudged disease risk.

Common misconceptions and pitfalls

• "A sweep removes the beneficial mutation's variation too." The opposite — the favored allele itself goes to high frequency. What is removed is the neutral variation linked to it. The selected site becomes uniform because everyone now carries the beneficial allele, not because it was deleted.
• "Negative Tajima's D proves positive selection." No. A population that recently expanded after a bottleneck shows negative Tajima's D genome-wide, mimicking a sweep at every locus. The discriminator is that a true sweep is a local outlier — a sharp valley against a flatter genomic background — plus the long-haplotype and high-frequency-derived-allele signals that demography does not produce locally.
• "All sweeps are hard sweeps." Early theory assumed single new mutations, but standing variation and recurrent mutation produce soft sweeps that retain multiple haplotypes and largely escape classical scans. In large populations soft sweeps may be the rule, not the exception.
• "Background selection and sweeps are the same." Background selection removes deleterious variants and also lowers diversity in low-recombination regions, but it produces a gradual reduction correlated with gene density and recombination — not a single sharp valley with high-frequency derived alleles. Separating the two is an active methodological problem.
• "A wider valley means stronger selection." Width depends on s/r, not s alone. A modest sweep in a recombination cold spot can clear a wider region than a strong sweep in a hotspot. You must condition on the local recombination rate before reading off selection strength.
• "Sweeps last forever in the genome." The signatures decay. Long haplotypes erode by recombination within tens of thousands of years; the diversity valley refills by mutation over hundreds of thousands. Ancient sweeps may leave only faint traces, which is why most detected human sweeps are recent.