Continuous Variation and Polygenic Traits

A discrete (qualitative) trait — pea seed shape, human ABO blood group — is controlled by one or a few genes with alleles of large, distinguishable effect, producing a small number of clearly separable phenotypic classes. A continuous (quantitative) trait — height, skin pigmentation, grain yield, blood pressure — is controlled by many genes (polygenes), each contributing a small, roughly additive amount to the phenotype, so that the possible genotypes produce many overlapping phenotypic classes. Add in environmental variation — nutrition, temperature, disease exposure, developmental noise — and the discrete classes blur entirely into a smooth curve. This is the polygenic or multifactorial model of inheritance, first formalized by Herman Nilsson-Ehle and George Udny Yule in the early 1900s and placed on a rigorous statistical footing by Ronald Fisher in 1918. Three properties distinguish quantitative traits from Mendelian ones. First, they are measured on a continuous numerical scale (centimeters, milligrams, milliliters) rather than sorted into named categories. Second, the number of genes involved (often called the trait's polygenic complexity) may range from just two or three loci (as in Nilsson-Ehle's wheat) to hundreds or thousands of loci each of tiny effect (as modern genome-wide association studies have shown for human height). Third, because environmental variation is superimposed on genetic variation, two individuals with identical genotypes at every relevant locus can still show different phenotypes — a fact with no analog in the clean genotype-determines-phenotype logic of Mendel's pea traits. The mathematical signature of a polygenic trait is the normal (Gaussian) distribution: a symmetric bell curve described by a mean (μ) and a variance (σ²). This shape is not a special biological law but a direct mathematical consequence of summing many independent, small, roughly additive contributions — a manifestation of the Central Limit Theorem. Each segregating gene pair contributes a term to the total; as the number of independently assorting gene pairs increases, the distribution of possible genotypic values approaches a smooth bell curve even before environmental variance is added.

Step 1 — Additive gene action at a single locus. Consider one gene with two alleles that act additively (no dominance): allele A contributes +1 unit to the phenotype, allele a contributes 0 units, and the effects of the two alleles at a locus simply add. A single locus therefore gives three genotypic classes: aa = 0, Aa = 1, AA = 2 units above baseline. This is still ordinary Mendelian segregation — nothing new has been invented — but now the phenotype is measured in units rather than sorted into "round vs. wrinkled."

Step 2 — Adding loci multiplies the number of classes. If a second, independently assorting locus (B/b) acts the same way and contributes to the same trait, an AaBb individual crossed with another AaBb individual produces offspring whose total "dose" of capital-letter alleles ranges from 0 (aabb) to 4 (AABB). The number of phenotypic classes for n additive loci, each with two alleles, is 2n + 1. With one locus you get 3 classes; two loci, 5 classes; three loci, 7 classes; and so on. The frequency of each class follows the binomial expansion of (¼ high : ½ mid : ¼ low)ⁿ combined across loci — most offspring end up in the middle classes (many ways to combine some high and some low alleles) and only rare individuals inherit all-high or all-low combinations at every locus.

Step 3 — Environmental variance smooths the classes into a continuum. Even a genetically identical individual does not develop into an identical phenotype every time: nutrition, temperature, uterine environment, and random developmental noise all perturb the trait value. This adds a roughly normally distributed "smear" (environmental variance, V_E) around each genotypic class. With only two or three loci, the discrete classes are still visible as bumps in the histogram (as Nilsson-Ehle actually observed). With many loci (five, ten, fifty or more) the classes are so numerous and so close together, and the environmental smearing so substantial relative to the gap between adjacent classes, that the histogram becomes a smooth, single-peaked bell curve — indistinguishable, without genetic analysis, from pure environmental variation.

Step 4 — Total phenotypic variance is partitioned. The total variance in a population for a quantitative trait, V_P, can be decomposed as: V_P = V_G + V_E + V_(G×E) where V_G is genetic variance (further splittable into additive, dominance, and epistatic components — the subject of the Additive Genetic Model), V_E is environmental variance, and V_(G×E) is variance from genotype-environment interaction. The proportion of phenotypic variance attributable to genetic differences, V_G/V_P, is the trait's heritability (broad-sense H², or narrow-sense h² when restricted to additive variance) — a population-specific statistic, not a fixed property of a trait, and never a measure of how "genetic" an individual's phenotype is (see Common Misconceptions below).

Step 5 — Predicting proportions with the binomial rule. For n additive loci each segregating independently, the fraction of F2 offspring that reach either extreme phenotype (all capital or all lowercase alleles) is (¼)ⁿ. For one locus that fraction is ¼; for two loci, 1/16; for three loci, 1/64; for four loci, 1/256. This is precisely why polygenic extremes are rare and why the population piles up near the mean — most gamete combinations mix high- and low-contributing alleles rather than sorting all of them to one extreme.

The proof that continuous variation is built from ordinary Mendelian genes, rather than some separate "blending" hereditary substance, is one of the pivotal experiments of early genetics: Herman Nilsson-Ehle's wheat and oat kernel-color crosses, published in 1909 at the Swedish Seed Association in Svalöv. Before Nilsson-Ehle, there was a real controversy. Francis Galton and the biometrician school (Karl Pearson, W.F.R. Weldon) had shown that human height and other traits follow smooth, continuous, bell-shaped distributions and regress toward the population mean across generations — patterns that seemed incompatible with Mendel's discrete 3:1 ratios. Mendelians and biometricians were, for roughly a decade after Mendel's rediscovery in 1900, arguing past each other about whether particulate (Mendelian) inheritance could possibly explain continuous variation at all. Nilsson-Ehle resolved this by crossing wheat varieties with red kernels and white kernels. When he crossed a red-kerneled variety with a white-kerneled variety, the F1 was intermediate (pink/light red) — exactly the "blending" pattern that worried biometricians. But when he self-pollinated the F1 and grew out the F2 generation, he did not get a simple 3:1 ratio. Instead, depending on which parental varieties he used, he obtained:

• 3 red : 1 white (when only one gene pair differed between the parents)
• 15 red (of varying intensity) : 1 white (when two gene pairs differed)
• 63 red : 1 white (when three gene pairs differed)

Crucially, within the "red" category the F2 plants were not uniform — they fell into a graded series of color intensities (from very dark red through medium to very light red/pink), in proportions matching the binomial expansion for two or three independently segregating, additively acting gene pairs. Only the rare 1/16 or 1/64 class — the individuals that had inherited all lowercase (non-pigmenting) alleles at every locus — showed the fully recessive white phenotype. Nilsson-Ehle correctly interpreted this as multiple gene pairs, each obeying standard Mendelian segregation and assortment, each adding a small increment of pigment, acting cumulatively on a single trait. He confirmed the model by test-crossing F2 plants of intermediate color back to the white parent and recovering further segregating ratios consistent with the number of loci he had proposed. This experiment was decisive because it showed, using a trait that any observer would naively call "continuous" and "blending" (kernel color grading smoothly from white to dark red), that the underlying inheritance was discretely Mendelian at every individual locus. The "blending" appearance was not a different kind of inheritance — it was the visual, macroscopic consequence of summing several Mendelian units plus a modest amount of environmental variation. Nilsson-Ehle's work, together with the independent theoretical treatment of skin-color inheritance in humans by geneticists in the 1910s–1920s and, definitively, R.A. Fisher's 1918 paper "The Correlation Between Relatives on the Supposition of Mendelian Inheritance," unified the Mendelian and biometrician camps. Fisher showed mathematically that if a continuous trait is controlled by many independently segregating Mendelian loci of small, roughly additive effect, the statistical patterns Galton and Pearson had documented — correlations between relatives, regression to the mean, normally distributed phenotypes — are exactly what population genetics predicts. No separate "blending" mechanism was needed; Mendelian particulate inheritance, multiplied across many loci, was sufficient. A complementary, later line of evidence came from Edward East's studies of corolla (flower tube) length in tobacco (Nicotiana longiflora) in 1910–1916. East crossed a long-corolla variety with a short-corolla variety; the F1 was intermediate, and the F2 showed a continuous, essentially normal distribution of lengths spanning between and slightly beyond the parental extremes, with variance far greater than in either true-breeding parental line or the uniform F1. This increase in F2 variance relative to the F1 — despite F1 individuals being genetically uniform (all heterozygous) — is the hallmark signature of genetic segregation contributing to variance, and it let East estimate the approximate number of gene pairs involved using variance-based formulas that are direct ancestors of the modern heritability calculations covered in Partitioning Phenotypic Variance.

Example 1 — A two-locus wheat-kernel-color cross, worked to ratios and pigment units. Suppose kernel color in a wheat variety is controlled by two independently assorting gene pairs, A/a and B/b, each acting additively: each capital allele contributes 1 unit of red pigment, each lowercase allele contributes 0 units, and units simply sum (this mirrors Nilsson-Ehle's actual model). A true-breeding dark red parent (AABB, 4 units) is crossed with a true-breeding white parent (aabb, 0 units).

• F1: all AaBb, 2 units — a uniform intermediate (medium red) phenotype, as observed.
• F1 × F1 → F2: each parent produces gametes AB, Ab, aB, ab in equal (¼) proportion. The 16 equally likely zygote combinations sort into pigment-unit classes as follows:

This reproduces Nilsson-Ehle's observed 15 red : 1 white ratio (15/16 of F2 show some pigment; only the 1/16 aabb class is white), with the "red" category itself graded into a 1:4:6:4 series of intensities — a visibly discrete approximation of a bell curve, exactly matching his published counts. If a third independently assorting locus (C/c) were also involved, the ratio becomes 63 red : 1 white, and the extreme (all-recessive) class shrinks to (¼)³ = 1/64 — illustrating the general rule that adding loci makes the extremes progressively rarer and the middle classes progressively more crowded, pushing the distribution toward a smooth continuum.

Example 2 — Estimating the (minimum) number of loci and reasoning about a breeding outcome. A plant breeder crosses two true-breeding tomato lines that differ sharply in fruit weight: a small-fruited line averaging 20 g per fruit and a large-fruited line averaging 60 g per fruit. The F1 is uniform at 40 g, exactly intermediate — consistent with additive gene action and no dominance. The F1 is self-pollinated, and among 1,000 F2 plants, the breeder finds 4 plants with fruit as small as the small parental line (≈20 g) and 4 plants as large as the large parental line (≈60 g); the rest form a continuous spread between and slightly beyond these extremes. Using the extreme-class formula for n additive loci, (¼)ⁿ estimates the fraction of F2 reaching either extreme phenotype. Observed extreme fraction ≈ 4/1000 = 0.004. Solve (¼)ⁿ = 0.004 for n: n = log(0.004) / log(0.25) ≈ (−2.40) / (−0.602) ≈ 3.98 ≈ 4 loci. So the breeder estimates that at least four gene pairs of roughly equal, additive effect underlie fruit weight in this cross (this is a minimum estimate — it assumes complete additivity, equal effect sizes, no environmental variance, and no linkage, all of which real data violate to some degree, which is precisely why more sophisticated variance-partitioning methods, covered under Partitioning Phenotypic Variance, are needed for real quantitative traits). Each locus contributes roughly (60 − 20) / (2 × 4) = 5 g per capital allele. A plant with, say, 5 capital alleles out of 8 total loci would be predicted to weigh approximately 20 g + (5 × 5 g) = 45 g — a prediction the breeder can test by genotyping and phenotyping additional F2 individuals, an approach that anticipates modern QTL (quantitative trait locus) mapping.

Herman Nilsson-Ehle trained as a botanist in Lund, Sweden, at a moment when Mendel's 1866 paper had just been rediscovered (1900) and was being tested against decades of biometrical data on continuously varying traits in humans, plants, and animals. Working as a wheat and oat breeder at the Swedish Seed Association in Svalöv — a practical plant-breeding institution, not a pure research laboratory — Nilsson-Ehle had access to large, carefully maintained collections of true-breeding cereal varieties differing in kernel color, a trait of real agronomic interest because pigment relates to grain quality and market value. Rather than treating the smooth, graded color variation he observed as evidence against Mendelism (as some biometricians were inclined to do), Nilsson-Ehle systematically crossed varieties differing by one, two, and three gene pairs and counted F2 progeny by color class with a breeder's meticulous eye, obtaining the 3:1, 15:1, and 63:1 ratios described above. His 1909 paper effectively founded the field that would become quantitative genetics, by demonstrating that Mendelian factors, simply added together across multiple loci, generate exactly the kind of continuous variation biometricians had been measuring in humans and livestock. He went on to a distinguished career in cereal breeding and served as a professor at Lund University, where his multiple-factor hypothesis became a standard tool for interpreting polygenic traits in both plants and animals.

Polygenic thinking underlies modern plant and animal breeding: selection indices and genomic selection in dairy cattle, pigs, and crops (maize, wheat, rice) rely on estimating the additive genetic value of individuals across many marker loci simultaneously, a direct descendant of Nilsson-Ehle's multiple-factor model combined with Fisher's variance framework. In human medicine, the recognition that most common diseases (type 2 diabetes, coronary artery disease, many psychiatric conditions) and continuous traits (height, BMI, blood pressure, cholesterol) are polygenic underlies the construction of polygenic risk scores (PRS), which sum the small effects of thousands of single-nucleotide variants identified by genome-wide association studies (GWAS) to estimate an individual's genetic liability. In conservation genetics, understanding quantitative trait variation helps managers judge whether a shrinking population retains enough additive genetic variance in fitness-related traits (disease resistance, thermal tolerance) to adapt to environmental change. In agriculture and food security, marker-assisted and genomic selection for polygenic traits like drought tolerance and yield has accelerated crop improvement far beyond what phenotypic selection alone could achieve, precisely because breeders can now track the many individual loci that Nilsson-Ehle inferred indirectly from ratios alone.

Ask the tutor: How do modern GWAS and polygenic risk scores extend Nilsson-Ehle's multiple-factor hypothesis to traits controlled by thousands of loci?