Linkage Disequilibrium

Linkage disequilibrium (LD)—also known as gametic phase disequilibrium—is the non-random statistical association of alleles at two or more polymorphic loci in a population, such that the observed frequency of a multi-locus gametic haplotype differs from the product of its constituent single-locus allele frequencies. First studied quantitatively by Hilda Geiringer (1944) and named by Richard C. Lewontin and Ken-ichi Kojima in 1960, linkage disequilibrium is a population-level property of haplotype distributions rather than a direct measure of physical genetic linkage on a chromosome: unlinked loci on separate chromosomes can exhibit LD due to population admixture or epistatic selection, whereas tightly linked loci on the same chromosome can exist in complete linkage equilibrium.

Within population genetics and medical genomics, the spatial patterns and temporal decay of linkage disequilibrium record a population's evolutionary history. Because homologous genetic recombination during meiosis breaks down haplotype associations at a rate proportional to the recombination fraction c, the extent of LD across chromosomes enables Genome-Wide Association Studies (GWAS) to map complex disease variants, reveals recombination hotspots mediated by PRDM9, and provides molecular signatures of recent selective sweeps, demographic bottlenecks, and ancestral admixture.

🧬 Interactive Population Genetics Simulation: Examine how genetic drift, selection, and population size alter allele and haplotype frequencies in Allele Drift.

The Coefficient D and Gametic Haplotype Determinants

Consider two diallelic loci in a population: locus A segregating for alleles A (frequency p_A) and a (frequency p_a = 1 − p_A), and locus B segregating for alleles B (frequency p_B) and b (frequency p_b = 1 − p_B). In a pool of haploid gametes, four two-locus haplotypes exist (AB, Ab, aB, and ab) with population frequencies P_AB, P_Ab, P_aB, and P_ab summing to 1. If alleles at the two loci associate independently at random (linkage equilibrium), the expected frequency of each haplotype equals the product of its marginal allele frequencies (for instance, P_AB = p_A p_B).

Lewontin and Kojima (1960) defined the fundamental coefficient of linkage disequilibrium (D) as the deviation of the observed gametic haplotype frequency from random expectation, or equivalently as the determinant of the 2 × 2 gametic frequency matrix:

D = P_AB − p_A p_B = P_AB P_ab − P_Ab P_aB

Here, P_AB P_ab represents the product of coupling (cis) gametes and P_Ab P_aB represents the product of repulsion (trans) gametes. All four gametic frequencies can be written in terms of marginal allele frequencies and D: P_AB = p_Ap_B + D, P_Ab = p_Ap_b − D, P_aB = p_ap_B − D, and P_ab = p_ap_b + D.

Normalized Statistics: Lewontin's D′ and Hill-Robertson r²

Because gametic frequencies cannot be negative (P_ij ≥ 0), the numerical range of raw D is constrained by the marginal allele frequencies (p_A, p_B), making raw D unsuitable for comparing LD across SNP pairs with different minor allele frequencies (MAFs). Population geneticists therefore standardize D using two distinct metrics:

1. Lewontin's Normalized D′ (1964): Divides D by its theoretical maximum possible magnitude D_max given the observed allele frequencies:

D′ = D / D_max, where D_max = min(p_Ap_b, p_ap_B) if D > 0, and min(p_Ap_B, p_ap_b) if D < 0 The value of |*D*′| ranges from 0 to 1. |*D*′| = 1.0 (complete LD) occurs whenever at least one of the four possible two-locus haplotypes is absent from the population (P_ij = 0), indicating that since the younger mutation arose on a single ancestral chromosome, no historical recombination or recurrent mutation has yet created the fourth gamete (the basis of Hudson and Kaplan's Four-Gamete Test).

1. Squared Correlation Coefficient (r² or Δ²): Formulated by William G. Hill and Alan Robertson in 1968, r² scales D² by the product of the single-locus binomial variances:

r² = D² / ( p_A p_a p_B p_b ) Also ranging from 0 to 1, r² equals the Pearson squared correlation between indicator variables for alleles at the two loci. Unlike D′, which can equal 1.0 even when one allele is rare and the other is common, r² = 1.0 (perfect LD) requires BOTH complete absence of recombinant gametes AND identical allele frequencies at the two loci (p_A = p_B).

Exponential Decay via Meiotic Recombination

In an infinitely large, randomly mating diploid population free of selection and new mutation, marginal allele frequencies p_A and p_B reach Hardy-Weinberg equilibrium in a single generation. By contrast, linkage disequilibrium decays gradually across generations as meiotic crossing over in double heterozygotes (AB/ab versus Ab/aB) exchanges alleles between homologous chromosomes. If c ∈ (0, 0.5] denotes the recombination fraction per generation between loci A and B, the disequilibrium after one generation is D₁ = (1 − c) D₀. Iterating across t generations yields the exponential decay law:

D_t = D₀ (1 − c)^t ≈ D₀ exp(−*c t*)

Even for unlinked loci located on different chromosomes (c = 0.5), linkage disequilibrium is not erased in a single generation; instead, it halves every generation (D_t = D₀ (0.5)^t), requiring 6 to 10 generations of panmixia to approach zero. For tightly linked SNPs separated by 10 kilobases on a human autosome (c ≈ 10⁻⁴ per generation), reducing initial D₀ by half requires t_1/2 = ln(2) / c ≈ 6,931 generations (~173,000 years).

Generation of LD by Drift, Admixture, and Epistatic Selection

Because recombination continually erodes D, persistent linkage disequilibrium in natural populations must be generated or maintained by countervailing evolutionary forces:

1. De Novo Mutation: Every new point mutation arises on one specific individual chromosome, initially exhibiting complete LD (|*D*′| = 1) with all pre-existing polymorphic alleles along that chromosome until recombination breaks down the association.
2. Genetic Drift and Recombination-Drift Balance: In a finite population of effective population size N_e, random gamete sampling generates stochastic fluctuations in haplotype frequencies. In 1971, Tomoko Ohta and Motoo Kimura (following Sved, 1971) proved that the balance between random genetic drift (generating LD at rate ~1/[2*N*_e]) and recombination (eroding LD at rate 2*c*) establishes a steady-state expectation for r²:

E[r²] ≈ 1 / (1 + 4 N_e c) (or ≈ 1 / [2 + 4*N*_ec] with mutation) Consequently, small bottlenecked populations (low N_e) exhibit extensive long-range LD, whereas large ancestral populations (high N_e) exhibit rapid short-range decay of LD.

1. Population Admixture and Substructure: When two previously isolated populations with different allele frequencies (p_*A*1 ≠ p_*A*2 and p_*B*1 ≠ p_*B*2) interbreed in proportions m and 1 − m, the admixed F₁ cohort instantly develops admixture linkage disequilibrium across the entire genome—even between unlinked chromosomes—of magnitude D_admix = m(1 − m)(p_*A*1 − p_*A*2)(p_*B*1 − p_*B*2). Measuring the exponential decay of admixture LD as a function of genetic distance c (via methods such as ROLLOFF and ALDER, developed by Nick Patterson and David Reich) allows geneticists to date ancient human admixture events to within a few generations.
2. Natural Selection and Epistasis: During a positive selective sweep, genetic hitchhiking drags a single ancestral haplotype to high frequency, creating an extended block of elevated LD (measured by Extended Haplotype Homozygosity, EHH). Furthermore, synergistic epistatic selection—where two alleles A and B confer high fitness together (w_AB > w_Aw_B) while recombinant combinations (Ab, aB) are deleterious—actively maintains positive D against recombination, favoring the evolution of chromosomal inversions ("supergenes") that suppress crossing over between co-adapted gene complexes.

In 2001–2005, the International HapMap Consortium and researchers including Mark Daly, Gabriel et al., and Jeffreys et al. discovered that human chromosomes are organized into discrete haplotype blocks (typically 10–100 kb in length) of high internal LD (|*D*′| > 0.8) separated by narrow 1–2 kb recombination hotspots where the histone methyltransferase PRDM9 binds specific DNA motifs to initiate meiotic double-strand breaks via SPO11.

This block architecture underpins Genome-Wide Association Studies (GWAS): if an ungenotyped causal disease mutation at locus A has correlation r² with a genotyped marker SNP ("tag SNP") at locus B, genotyping N individuals at the tag SNP captures the equivalent statistical power of genotyping N · r² individuals directly at the causal mutation. However, because LD decays much faster in African populations (N_e ≈ 20,000, average haplotype block ~10–15 kb) than in European or East Asian populations (N_e ≈ 10,000, post-bottleneck block ~30–50 kb), polygenic risk scores and GWAS tag SNPs discovered in European cohorts lose predictive accuracy when transferred across ancestries, whereas multi-ancestry cohorts leverage the narrower LD of African genomes for high-resolution fine-mapping of causal variants.

Why can two SNPs have |*D*′| = 1.0 while having a very low r² (such as r² = 0.02)?

|*D*′| = 1.0 only requires that no historical recombination has occurred between the two mutations, so at most three of the four possible two-locus haplotypes exist in the population (P_ij = 0 for one gamete). When a newly arisen rare mutation at locus A (p_A = 0.01) occurs on a chromosome carrying a common ancestral allele at locus B (p_B = 0.50), the rare allele A is always found with B (so |*D*′| = 1.0), yet knowing an individual carries common allele B barely predicts whether they carry rare allele A, yielding r² = (p_Ap_b) / (p_ap_B) ≈ 0.01 / 0.99 ≈ 0.01.

Can linkage disequilibrium exist between genes on different chromosomes?

Yes. Despite its name, linkage disequilibrium measures statistical non-independence between alleles (P_AB ≠ p_Ap_B), not physical chromosomal linkage. Recent population admixture, assortative mating, cryptic population stratification, or strong epistatic natural selection can generate significant LD between loci on different autosomes (c = 0.5), which then decays by 50% per generation once random mating resumes.

How is linkage disequilibrium used to estimate past effective population size (N_e)?

Because the equilibrium expectation E[r²] ≈ 1 / (1 + 4*N*_ec) depends on the recombination fraction c, pairs of SNPs separated by large genetic distances (c ≈ 0.05–0.20 Morgans) recombine rapidly and reflect N_e over the most recent 2 to 10 generations, whereas tightly linked SNPs (c ≈ 0.0001 Morgans) reflect N_e thousands of generations ago. In general, LD at genetic distance c reflects the harmonic mean N_e roughly t ≈ 1 / (2*c*) generations in the past.

1. Lewontin, R. C., & Kojima, K. (1960). "The evolutionary dynamics of complex polymorphisms." Evolution, 14(4), 458–472. doi:10.1111/j.1558-5646.1960.tb03113.x
2. Hill, W. G., & Robertson, A. (1968). "Linkage disequilibrium in finite populations." Theoretical and Applied Genetics, 38(6), 226–231. doi:10.1007/BF01245622
3. Ohta, T., & Kimura, M. (1971). "Linkage disequilibrium between two segregating nucleotide sites under the steady flux of mutations in a finite population." Genetics, 68(4), 571–580. doi:10.1093/genetics/68.4.571
4. Slatkin, M. (2008). "Linkage disequilibrium — understanding the evolutionary past and mapping the medical future." Nature Reviews Genetics, 9(6), 477–485. doi:10.1038/nrg2361