Polygenic Inheritance

Polygenic inheritance (also termed quantitative inheritance or multifactorial inheritance) is the mode of genetic transmission in which a measurable phenotype—such as adult human stature, systolic blood pressure, agricultural grain yield, or Drosophila abdominal bristle number—is determined by the cumulative small effects of many distinct genetic loci (polygenes or quantitative trait loci, QTLs) acting alongside environmental variation. Unlike monogenic Mendelian traits (such as cystic fibrosis or sickle cell disease), which segregate into discrete qualitative categories corresponding to genotypes at a single gene, polygenic traits exhibit continuous, bell-shaped Gaussian distributions across populations.

For two decades following the rediscovery of Gregor Mendel's laws in 1900, biology was divided between Mendelians (led by William Bateson, who argued that evolution proceeds via discrete macromutations) and Biometricians (led by Karl Pearson and W. F. R. Weldon, who measured continuous Gaussian variation and linear parent-offspring correlations following Francis Galton). This conflict was resolved between 1909 and 1918 when Swedish geneticist Herman Nilsson-Ehle demonstrated additive multi-locus kernel color ratios in wheat and British statistician Ronald Fisher published his landmark 1918 paper "The Correlation between Relatives on the Supposition of Mendelian Inheritance", proving via the Central Limit Theorem that particulate Mendelian inheritance across a large number of independent loci generates continuous quantitative variation and launching the Modern Synthesis.

🧬 Interactive Population Genetics Simulation: Model allele frequency changes, quantitative trait shifts, and selection differentials in Allele Drift.

Nilsson-Ehle's Wheat Crosses and the Binomial Expansion

To understand how discrete alleles generate a continuous phenotypic spectrum, consider k independent diallelic loci (A₁/a₁, A₂/a₂, …, A_k/a_k) that contribute additively and equally to a quantitative trait z, where each uppercase "+" allele adds a fixed phenotypic increment +*α* and each lowercase "−" allele adds 0. In a cross between two doubly or multiply heterozygous F₁ parents (each carrying k uppercase and k lowercase alleles), the number of "+" alleles inherited by an F₂ individual can range from 0 to 2*k*, following a binomial distribution with coefficients given by the expansion of (½ + ½)^2*k*:

• 1 Locus (k = 1, 2*k* = 2): Yields 3 genotypic classes (0, 1, or 2 "+" alleles) in a 1 : 2 : 1 ratio (4¹ = 4 gametic combinations).
• 2 Loci (k = 2, 2*k* = 4): Yields 5 phenotypic classes in a 1 : 4 : 6 : 4 : 1 ratio (4² = 16 combinations).
• 3 Loci (k = 3, 2*k* = 6): Demonstrated empirically by Herman Nilsson-Ehle (1909) in hexaploid wheat (Triticum aestivum) kernel coat redness (R₁, R₂, R₃), yielding 7 shading classes from white (0 R alleles, 1/64) to deep red (6 R alleles, 1/64) in a 1 : 6 : 15 : 20 : 15 : 6 : 1 ratio (4³ = 64 combinations).

Fisher's (1918) Infinitesimal Model and the Central Limit Theorem

In 1918, Ronald Fisher generalized this multi-locus binomial progression into the infinitesimal model. Suppose a trait is influenced by k → ∞ unlinked Mendelian loci at allele frequencies p_i and q_i = 1 − p_i, where each individual locus contributes an infinitesimal effect scaling as α_i ∝ 1 / √*k* so that the total additive genetic variance V_A = ∑_*i*=1^k 2*p*_iq_iα_i² remains finite and constant. By the Central Limit Theorem of probability theory, the sum of a large number of independent binomial genetic contributions converges to a smooth Gaussian (normal) probability density function:

f(z) = [ 1 / √(2π V_P) ] · exp[ − (z − μ)² / (2 V_P) ]

When non-heritable environmental variance (V_E, arising from nutrition, temperature, microhabitat, and developmental noise) is added to the genetic value, any remaining step-like discontinuities between the 2*k* + 1 genotypic classes are smoothed out completely. A key mathematical consequence of Fisher's infinitesimal model is that short-term directional or stabilizing selection shifts the population mean μ via infinitesimal allele-frequency adjustments across thousands of loci without substantially depleting the standing additive genetic variance (V_A) over tens of generations.

Decomposition of Phenotypic Variance (V_P)

Quantitative genetics models an individual's measured phenotype P as the sum of a genetic value G and an environmental deviation E (P = G + E). Assuming genotypes and environments are uncorrelated (Cov(G, E) = 0) and without genotype-by-environment interaction (V_G×E = 0), Fisher (1918) partitioned the total phenotypic variance (V_P) into orthogonal components:

V_P = V_G + V_E = ( V_A + V_D + V_I ) + V_E

1. Additive Genetic Variance (V_A): The variance of breeding values—the sum of the average linear effects of the individual alleles an organism carries: V_A = ∑_i 2*p*_iq_i [ α_i + d_i(q_i − p_i) ]². Because diploid parents transmit only a single haploid allele per locus to each gamete rather than intact diploid genotypes, V_A is the primary cause of resemblance between parents and offspring.
2. Dominance Genetic Variance (V_D): Variance arising from non-additive interactions between alleles at the same locus (V_D = ∑_i [2*p*_iq_id_i]²). Full siblings share both parental alleles (identical by descent at both chromosomes) with 25% probability, so V_D contributes ¼*V*_D to the covariance of full sibs, but zero to parent-offspring covariance.
3. Epistatic Interaction Variance (V_I): Variance resulting from non-additive epistasis across distinct loci (V_AA + V_AD + V_DD + …).

Broad-Sense (H²) vs. Narrow-Sense (h²) Heritability and the Breeder's Equation

From this decomposition, quantitative geneticists define two distinct measures of heritability:

• Broad-Sense Heritability (H² = V_G / V_P): The proportion of total phenotypic variance attributable to all genetic differences (additive, dominance, and epistatic combined), estimated from comparisons of monozygotic (identical) versus dizygotic (fraternal) twins or isogenic clonal lines.
• Narrow-Sense Heritability (h² = V_A / V_P): The proportion of phenotypic variance attributable solely to additive genetic variance (V_A), estimated from the linear regression slope of mid-parent phenotype against offspring phenotype (b_OP = h²).

As formulated by Jay Laurence Lush (1937) in the Breeder's Equation, narrow-sense heritability determines the single-generation evolutionary response (R) to directional natural selection or artificial breeding:

R = h² · S = ( V_A / V_P ) · S

where S = μ_selected − μ₀ is the selection differential and R = μ_offspring − μ₀ is the shift in population mean phenotype in the next generation. Extended to multivariate correlated trait vectors z by Russell Lande (1979), Δz̄ = G β, where G is the additive genetic variance-covariance matrix (reflecting pleiotropy and linkage disequilibrium) and β = P⁻¹S is the directional selection gradient vector.

Falconer's Liability-Threshold Model for Dichotomous Diseases

Many common human clinical disorders—including type 2 diabetes, schizophrenia, coronary artery disease, cleft palate, and autoimmune disorders—are binary (present or absent) rather than continuous, yet fail to follow simple Mendelian ratios in pedigrees. In 1965, Douglas S. Falconer resolved this paradox with the liability-threshold model: thousands of polygenic risk alleles and environmental exposures sum additively to form an underlying, unobserved Gaussian variable called liability (L ~ N(0, 1)). Clinical disease manifests whenever an individual's total liability exceeds a developmental or physiological threshold (T), explaining why familial recurrence risk drops much faster than ½ as genetic relatedness decreases from monozygotic twins (r = 1.0) to first-degree (r = 0.5) and second-degree (r = 0.25) relatives.

GWAS, "Missing Heritability," and the Omnigenic Model

With the advent of Genome-Wide Association Studies (GWAS) genotyping millions of single-nucleotide polymorphisms (SNPs) across biobank cohorts of >10⁶ individuals, geneticists directly mapped the architecture of human polygenic traits. For adult human height (pedigree h² ≈ 0.80), early GWAS found that top genome-wide significant SNPs (p < 5 × 10⁻⁸) each altered stature by only 1 to 3 millimeters and collectively explained <10% of variance—a discrepancy termed the "missing heritability" problem (Manolio et al., 2009).

In 2010, Jian Yang, Peter Visscher, and colleagues developed GREML (GCTA) to estimate SNP heritability (h²_SNP) from all common SNPs simultaneously without requiring individual variants to pass genome-wide significance, proving that ~45%–50% of height variance is tagged by common additive SNPs of very small effect, with the remaining 30% residing in rare variants (low minor allele frequency) verified by whole-genome sequencing (Wainschtein et al., 2022) and a 5.4-million-person GWAS by Yengo et al. (2022) that saturated common-SNP heritability across ~12,000 independent loci. To explain why so many thousands of genomic regions affect each complex trait, Evan Boyle, Yang Li, and Jonathan Pritchard (2017) proposed the omnigenic model: because cellular regulatory networks are densely interconnected, a small set of core genes with direct biochemical roles in the phenotype are modulated in trans by thousands of peripheral genes expressed in the relevant tissue, which collectively contribute the bulk (>70%) of total heritable variance. Summing an individual's allele dosages weighted by GWAS effect sizes (β̂_i) yields a Polygenic Risk Score (PRS = ∑ β̂_i G_i) used for clinical risk stratification in pharmacogenomics and personalized medicine.

If a trait has a heritability of h² = 0.80, does that mean 80% of an individual's trait is caused by genes and 20% by the environment?

No. Heritability (h² = V_A / V_P) is strictly a population-level variance ratio, not a partition of an individual organism's phenotype. It states that 80% of the differences (variation) among individuals within that specific population at that specific time are attributable to additive genetic differences. Furthermore, because V_P = V_G + V_E, if environmental inequality (V_E) decreases—for example, through universal access to childhood nutrition and healthcare—heritability h² increases toward 1.0 even though the underlying genetic architecture has not changed at all.

Can a highly heritable polygenic trait still change dramatically when the environment changes?

Yes. High within-population heritability (h² = 0.80) does not restrict the magnitude of a plastic response to a large secular shift in the environment. For example, although adult human height is ~80% heritable within any birth cohort, average adult height in South Korea, Japan, and the Netherlands increased by 8 to 15 cm over the twentieth century due to improvements in nutrition and infectious disease control—a shift of more than 1.5 phenotypic standard deviations driven by environmental change rather than allele frequency evolution.

Why do Polygenic Risk Scores (PRS) lose predictive accuracy when applied across different human ancestry groups?

Most large-scale GWAS discovery cohorts have predominantly sampled individuals of European ancestry. Because GWAS measures association at marker tag SNPs that stand in linkage disequilibrium (r²) with ungenotyped causal variants, and because both LD block structure and minor allele frequencies differ across continental populations due to the Out-of-Africa bottleneck and independent genetic drift, a PRS trained in European cohorts exhibits roughly a 50%–75% reduction in predictive R² when applied to African-ancestry cohorts unless multi-ancestry fine-mapping is performed.

1. Fisher, R. A. (1918). "The correlation between relatives on the supposition of Mendelian inheritance." Transactions of the Royal Society of Edinburgh, 52(2), 399–433. doi:10.1017/S0080456800012163
2. Falconer, D. S. (1965). "The inheritance of liability to certain diseases, estimated from the incidence among relatives." Annals of Human Genetics, 29(1), 51–76. doi:10.1111/j.1469-1809.1965.tb00500.x
3. Yang, J., Benyamin, B., McEvoy, B. P., Gordon, S., Henders, A. K., Nyholt, D. R., ... & Visscher, P. M. (2010). "Common SNPs explain a large proportion of the heritability for human height." Nature Genetics, 42(7), 565–569. doi:10.1038/ng.608
4. Boyle, E. A., Li, Y. I., & Pritchard, J. K. (2017). "An expanded view of complex traits: from polygenic to omnigenic." Cell, 169(7), 1177–1186. doi:10.1016/j.cell.2017.05.038