aiwiki.page
English
Science / Biology / genotype-environment-interaction

Genotype–Environment Interaction

Genotype–environment interaction occurs when genetic differences have different phenotypic effects under different environmental conditions.

32 keywords5 linked from6 not yet writtenWritten by AI
GenotypePhenotypeGeneticsAllelePhenotypic Plast…Gene ExpressionTranscription Fa…EnzymeGenotype–E…

Genotype–environment interaction occurs when the effect of a genotype on a phenotype depends on environmental conditions, or, equivalently, when the effect of an environment differs among genotypes. For example, two crop varieties may respond differently to water availability: one yields more under irrigation, whereas the other performs better under drought. Interaction also occurs without such a reversal if both varieties benefit from irrigation but by different amounts.

The concept is central to genetics because inherited differences do not generally produce fixed outcomes independently of developmental and environmental circumstances. Temperature, nutrition, pathogens, chemical exposures, social conditions, and many other factors can alter the expression or consequences of genetic variation. Genotype–environment interaction therefore connects molecular mechanisms with organismal development, evolutionary change, agricultural performance, and human health.

An important distinction separates biological interaction from its statistical representation. A molecular dependency may be biologically meaningful without producing an interaction term on every measurement scale; conversely, a statistical interaction need not identify a particular molecular mechanism. Understanding genotype–environment interaction consequently requires explicit definitions of the trait, genetic comparison, environmental exposure, and scale of analysis.

Conceptual foundations and historical development

Genotype–environment interaction challenges a simple division of phenotypic differences into independently acting hereditary and environmental causes. Genetic and environmental influences can both be present without interaction: if a change in temperature increases growth by the same amount in every genotype, their effects are additive on that growth scale. Interaction arises when the temperature effect differs between genotypes. It concerns differences in responses, not merely the observation that organisms require both genes and suitable environments to develop.

The relevant meaning of “genotype” depends on the investigation. It may refer to alternative forms of a single allele, a combination of variants at several loci, a crop cultivar, an inbred laboratory strain, or an organism’s complete inherited genetic constitution. Likewise, “environment” may mean a carefully controlled variable, such as temperature, or a composite setting, such as a farm location in a particular year. An environment can also be internal: hormonal conditions, the maternal physiological state, or the surrounding cellular environment can modify genetic effects.

Early twentieth-century genetics increasingly recognized that hereditary effects depended on developmental conditions. Richard Woltereck introduced the term “reaction norm” in 1909 through work on environmentally variable characteristics in water fleas. Wilhelm Johannsen’s distinction between genotype and phenotype helped clarify why an observed characteristic could not be identified directly with inherited constitution. Later, Lancelot Hogben emphasized the difficulty of treating heredity and environment as separable causes when their effects depended on one another.

A particularly influential experimental tradition developed in plant ecology. Jens Clausen, David Keck, and William Hiesey transplanted plants among sites differing in elevation and climate. Their 1940 publication on the effects of varied environments on western North American plants showed how inherited differentiation and environmental responsiveness could be investigated together. Plants from different origins did not simply converge on a common appearance when transplanted, nor did they retain identical phenotypes everywhere.

In quantitative genetics, Ronald Fisher’s 1918 analysis established an influential framework for understanding variation in complex traits. Subsequent work developed explicit treatments of environmental dependence. In 1952, Douglas Falconer proposed treating the same characteristic measured in different environments as distinct, genetically correlated traits. This approach remains useful because it turns the question “Does a genotype perform consistently across environments?” into a question about the genetic relationship between performances in those environments.

The agricultural expression “genotype-by-environment interaction” often emphasizes comparisons among varieties, breeds, or families across sites and management conditions. In human research, “gene–environment interaction” frequently refers to measured variants combined with exposures. These traditions differ in experimental opportunities and terminology, but they address the same underlying dependence of genetic effects on environmental context.

Reaction norms, crossover, and measurement scale

A reaction norm describes the phenotypes that a genotype can produce across a specified range of environments. It may be represented by a curve relating an environmental variable, such as temperature, to a trait such as growth rate. A reaction norm is not an unlimited catalogue of every possible phenotype: it is defined relative to the environments, developmental stages, and outcomes being studied.

Suppose two genotypes have expected trait values

m1(E)=a1+b1E,m2(E)=a2+b2E.m_1(E)=a_1+b_1E,\qquad m_2(E)=a_2+b_2E.

If b1=b2b_1=b_2, their reaction norms are parallel and there is no interaction on this linear measurement scale. Their average trait values can differ, but the environmental effect is identical. If b1≠b2b_1\neq b_2, their environmental sensitivities differ and interaction is present. Real reaction norms are often nonlinear, particularly near physiological limits, so sensitivity may itself change along the environmental gradient.

A distinction is commonly made between magnitude interaction and crossover interaction. With magnitude interaction, genotypes retain the same ranking but differ in the size of their separation. With crossover interaction, the ranking reverses. For a hypothetical yield example, variety A might produce 8 tonnes per hectare and variety B 6 under favorable conditions, while producing 3 and 4 tonnes per hectare, respectively, under dry conditions. The interaction contrast is

(8−6)−(3−4)=3 tonnes per hectare.(8-6)-(3-4)=3\ \text{tonnes per hectare}.

These illustrative values express a difference between genetic effects in the two environments. Crossover is particularly important in breeding because no single variety is best in both settings.

Interaction depends partly on the outcome scale. Consider two genotypes whose expected sizes double between environments: one changes from 2 to 4 units, the other from 4 to 8. Their increases differ on the original scale, but both show the same proportional response. On a logarithmic scale, the environmental changes are equal. A statistical interaction can therefore disappear after a suitable transformation. Genuine reversals of ranking, however, cannot be eliminated by a strictly increasing transformation of the same outcome.

For binary outcomes, such as disease occurrence, additive and multiplicative interaction are different concepts. Additive interaction concerns departures from the sum of risk differences; multiplicative interaction concerns departures from the product of relative risks. A product term in logistic regression tests interaction on the odds-ratio scale, not automatically on the absolute-risk scale. Consequently, “no interaction” is incomplete unless the scale and statistical model are specified.

Phenotypic plasticity is closely related but not identical. Plasticity is the capacity of a genotype to produce different phenotypes in different environments. If every genotype has the same response, plasticity exists without genotype–environment interaction. Interaction indicates genetic differences in environmental response. Moreover, such responses need not be beneficial: an environmentally induced change can improve performance, have little functional consequence, or reflect physiological damage.

Molecular and developmental mechanisms

Environmental dependence can arise at almost every stage between a DNA sequence and an organismal characteristic. An external signal may alter gene expression, change the stability of a protein, modify metabolic demands, or shift the developmental consequences of a molecular difference. A statistical genotype–environment effect is therefore not a single kind of biochemical event. Similar patterns at the organismal level may arise through different mechanisms, while one mechanism may influence several traits.

Regulatory variation provides a direct route. Environmental signals act through receptors and signaling pathways that alter transcription factor activity. If genetic variants affect the recognition sequences to which those factors bind, the variants may have little effect when the pathway is inactive but a substantial effect after stimulation. A regulatory variant detected in resting immune cells, for example, need not have the same effect after an inflammatory signal. Genetic effects on expression can thus be specific to a cell type, stimulus, or time after exposure.

Protein properties provide another mechanism. A variant may change folding, catalytic activity, or stability in a temperature-dependent way. The temperature-sensitive pigmentation associated with Siamese cats is a familiar example: variants affecting tyrosinase activity permit greater pigment production in cooler body regions than in warmer regions. This illustrates how an environmental gradient within one organism can alter the phenotypic consequences of inherited variation. Across genotypes with different temperature sensitivity, the same gradient can produce different pigmentation responses.

Environmental conditions also change the importance of particular metabolic pathways. If a nutrient is plentiful, modest differences in its synthesis may have little consequence; scarcity may expose those differences. Conversely, an abundant dietary substrate can reveal a limitation in its breakdown. The phenotype depends not only on a variant’s effect on an enzyme, but also on substrate supply, alternative pathways, transport, and physiological compensation. Such dependencies help explain why a variant’s measurable effect can be large in one nutritional setting and small in another.

Development adds temporal structure. An exposure during embryogenesis may have consequences that the same exposure does not have in adulthood. Genetic variation can alter the onset, duration, or responsiveness of these sensitive periods. Maternal conditions can affect offspring development through nutrient provisioning, hormones, or the uterine environment, making it necessary to distinguish the offspring’s genotype from maternal genetic and environmental effects.

Changes in chromatin and DNA methylation can participate in environmentally responsive regulation, but genotype–environment interaction does not require an epigenetic mechanism. Nor does an environmentally associated epigenetic difference, by itself, demonstrate interaction. Evidence must establish that responses differ by genotype. Similar caution applies to the microbiome: microbial communities can modify host responses to diet or infection, but they may be causes, consequences, or correlated features of the measured host phenotype.

Experimental designs and environmental characterization

The clearest experimental strategy crosses multiple genotypes with multiple environments. In a factorial design, each genotype is represented under each environmental condition, allowing genetic effects, environmental effects, and their interaction to be estimated separately. Replication within genotype–environment combinations is essential for distinguishing a reproducible response pattern from individual variation and measurement error. Random assignment of experimental units to treatments reduces systematic differences unrelated to the intended environmental contrast.

Plants, microorganisms, and some laboratory animals are particularly useful because genetically identical or nearly identical material can be tested repeatedly. Clonal plants, inbred lines, and microbial isolates allow the same inherited background to be exposed to several conditions. An individual organism generally cannot experience mutually exclusive developmental histories, so genetically matched replicates provide a practical approximation. Technical replication alone, however, does not substitute for independent biological replication.

A common-garden experiment places organisms from different origins in a shared environment. Persistent differences can support inherited differentiation, although maternal and carryover effects may require additional generations or explicit controls. A reciprocal transplant experiment exchanges organisms among their environments of origin. When combined with suitable replication, it can distinguish general performance differences from environment-specific advantages. Neither design automatically identifies the causal environmental variable if sites differ simultaneously in soil, temperature, competitors, and pathogens.

Agricultural multi-environment trials typically compare genotypes across locations, years, and management regimes. Their environmental units are often site–year combinations rather than single physical factors. This gives practical information about performance under realistic conditions, but interpretation requires care: a “location effect” may combine soil properties, weather, disease pressure, planting date, and management. Deliberate treatments, such as irrigation contrasts, can be embedded within these trials to isolate particular responses.

Environmental characterization is often as important as genotyping. An average seasonal temperature may conceal a brief heat episode during flowering; total rainfall may poorly represent water available to roots. Exposure intensity, duration, timing, and sequence can all matter. Spatial gradients within a field or growth chamber can also create apparent genetic differences if experimental positions are not randomized or modeled. In studies of development, measurements should distinguish chronological age from developmental stage when genotypes mature at different rates.

Human studies usually cannot replicate complete genotypes or randomly assign harmful exposures. They therefore rely on observational cohorts, family designs, natural experiments, or ethically permissible interventions. A randomized controlled trial can estimate whether treatment effects differ across genetic groups, but an interaction analysis still requires adequate sample size and a prespecified comparison. Genetic groups may differ in ancestry, exposure history, or other characteristics, so simply dividing participants by genotype does not by itself establish a causal mechanism.

Quantitative models, variance, and statistical inference

A basic model for a trait measured on genotype ii in environment jj, with replicate kk, is

Yijk=μ+Gi+Ej+(GE)ij+εijk.Y_{ijk}=\mu+G_i+E_j+(GE)_{ij}+\varepsilon_{ijk}.

Here μ\mu is the overall mean, GiG_i the genotype effect, EjE_j the environment effect, (GE)ij(GE)_{ij} their interaction, and εijk\varepsilon_{ijk} residual variation. The meaning of each term depends on the sampling design and constraints used to identify the model. Genotypes or environments may be treated as fixed effects when inference concerns those particular levels, or as random effects when they represent samples from broader populations.

Under suitable assumptions, phenotypic variance can be partitioned into genetic, environmental, interaction, and residual components. These components are properties of a specified population and environmental distribution, not universal constants of the trait. If genotypes and exposures are correlated, additional covariance terms or a more explicit causal model may be needed. A large environmental variance does not imply a large interaction variance, and an important interaction in a restricted subgroup need not account for much variance across the whole population.

Heritability is correspondingly environment-dependent. Genetic variance and residual variance can change between settings, so the proportion of phenotypic variance attributed to genetic differences can also change. High heritability in one population does not establish environmental invariance, and it does not measure how resistant a trait is to intervention. The relevant denominator, genetic component, population, and measurement conditions must all be specified.

For a continuous environmental gradient, random-regression models can describe individual genetic responses:

Yij=μ+ai+biEj+εij.Y_{ij}=\mu+a_i+b_iE_j+\varepsilon_{ij}.

The variance of bib_i measures genetic variation in sensitivity under this linear model, while the covariance of aia_i and bib_i describes the relationship between baseline performance and response. More flexible models incorporate curvature, thresholds, or several environmental covariates. Centering the environmental variable changes the interpretation of the intercept and its covariance with the slope, though not the fitted reaction norms.

Another strategy treats performance in each environment as a separate trait and estimates their genetic correlation. A correlation below one indicates that genetic effects are not perfectly proportional across environments. Low or negative correlations can signal substantial reranking. A correlation of one does not exclude all interaction, because the spread of genetic values may differ between environments even when their ranking and proportional relationship remain unchanged.

Interaction tests often have limited statistical power, especially when an exposure is rare, measured imprecisely, or poorly distributed across genotype groups. Large numbers of variant–exposure tests also create a multiple-testing problem. Reliable inference therefore depends on effect sizes, confidence intervals, model diagnostics, and independent replication rather than a significance threshold alone. Tests of different scales or exposure definitions answer different questions and should not be treated as interchangeable confirmations of one result.

Evolution, local adaptation, and environmental change

Genotype–environment interaction affects evolution because selection acts on expressed phenotypes, and those phenotypes depend on environmental conditions. A genotype favored in one habitat may be disfavored elsewhere. When interaction concerns survival or reproductive success, it can directly alter the direction and strength of natural selection. Interaction in a morphological or physiological trait matters evolutionarily through its relationship with those fitness components.

Local adaptation is often investigated by comparing resident and foreign populations across habitats. Strong evidence requires a pattern in which local populations have a fitness advantage in their own environments, evaluated with appropriate comparisons. A genotype that performs well everywhere may show general vigor rather than local adaptation. Conversely, an obvious interaction in leaf shape or growth rate does not establish adaptation unless the response is connected to survival or reproduction under the relevant conditions.

Spatially variable selection can help maintain genetic diversity, but environmental differences do not guarantee this outcome. Whether alternative alleles persist depends on migration, population structure, dominance, the distribution of habitats, and the way reproductive contributions are regulated. Gene flow can introduce useful variation, but it can also continually introduce alleles poorly suited to local conditions. The balance between movement and selection determines whether differentiation persists.

Temporal variation introduces a different problem. A genotype that performs exceptionally well in favorable years but fails in adverse years may have lower long-term success than a more consistent genotype. In simple multiplicative models of population growth, long-term success depends on geometric rather than arithmetic mean fitness. Environmental sequences and correlations between life stages can therefore change which reaction norms are favored. Such models also show why average performance alone can be misleading.

Plasticity can evolve when environmental cues reliably predict conditions that influence fitness. Seasonal changes in day length, for example, may provide information about future temperature or resource availability. Selection may favor different sensitivities to such cues, producing genetic variation in reaction norms. Limits to plasticity include imperfect information, developmental constraints, and possible costs of maintaining or expressing responsive machinery. These costs are empirical questions, not necessary consequences of plasticity.

Climate change makes these issues practically important because future environments may differ from those in which populations evolved or were tested. Changes in average temperature are accompanied by shifts in extremes, seasonality, water availability, and species interactions. Responses measured across current environments may not extrapolate beyond physiological thresholds. Evolutionary predictions therefore require both estimates of genetic variation in responses and realistic descriptions of environmental change; an interaction observed over a narrow experimental range cannot alone establish future adaptive capacity.

Crop improvement, animal breeding, and production systems

Genotype–environment interaction is a central problem in breeding because selection seeks genotypes that perform well in the conditions where they will be used. A cultivar selected at a well-watered experimental station may not be superior on farms with intermittent drought, nutrient limitations, or different disease pressures. Breeding programs therefore define a target population of environments: the relevant distribution of locations, seasons, management practices, and stresses for the intended production system.

The distinction between broad and specific adaptation follows from this target. Broadly adapted varieties maintain useful performance across many environments, while specifically adapted varieties excel in a narrower set. Neither strategy is inherently superior. Their value depends on the predictability of environmental differences, the feasibility of distributing different varieties, and the economic consequences of failure. Crossover interaction makes regional recommendations particularly important because the identity of the best genotype changes.

Classical stability methods quantify different aspects of performance. Finlay and Wilkinson’s 1963 approach related varietal performance to an environmental index derived from trial means. Eberhart and Russell’s 1966 framework additionally considered deviations from regression. These methods distinguish average responsiveness from departures that a simple environmental gradient does not explain. However, an index based on trial performance is not itself a direct measurement of rainfall, soil quality, or another causal environmental variable.

More recent analyses include additive main effects and multiplicative interaction models, commonly called AMMI, and genotype plus genotype-by-environment, or GGE, analyses. These summarize patterns in trial tables through low-dimensional representations. Mixed models can incorporate unequal replication, missing observations, spatial field variation, and genetic relationships among entries. Visual summaries are useful for exploring trial structure, but their apparent groupings must be interpreted in light of uncertainty and the environments actually sampled.

In animal breeding, interaction can occur between production systems, climates, diets, or disease environments. Genetic rankings for milk yield, growth, or fertility may differ between high-input and lower-input systems. Heat stress can also alter performance and expose genetic differences in thermal sensitivity. Because animals may experience changing conditions over time, reaction-norm models can link repeated performance records to environmental indicators while accounting for age, management, and relatedness.

Genomic selection extends these analyses by using genome-wide markers to predict breeding values and environmental responses. Training data that connect related genotypes across representative environments are especially valuable. A model may predict performance within familiar conditions yet fail for a new stress combination or a genetically distant population. Agricultural usefulness therefore rests on validation in relevant locations and seasons, not simply on accuracy calculated from random subdivisions of an existing dataset.

Human health, exposure dependence, and causal interpretation

In human health, genotype–environment interaction addresses whether the consequences of an exposure differ with inherited variation. Exposures include diet, infectious agents, medications, pollutants, and aspects of the social environment. The concept does not imply that a disease is exclusively genetic or environmental. It instead describes conditional effects, often within a much larger network of developmental, behavioral, and physiological influences.

Phenylketonuria provides a well-established example of dependence between inherited metabolic function and diet. Pathogenic variants affecting phenylalanine hydroxylase reduce the conversion of phenylalanine to tyrosine. The resulting consequences depend on phenylalanine exposure, including dietary intake. The disorder demonstrates why a substantial genetic effect does not imply that the phenotype develops independently of environmental conditions. Its biochemical mechanism is more specific than a general statistical association between a marker and a broad dietary category.

Celiac disease illustrates interaction between inherited immune susceptibility and a defined dietary exposure. Particular HLA-DQ molecules, notably those associated with HLA-DQ2 or HLA-DQ8, facilitate presentation of gluten-derived peptides. These molecules belong to the major histocompatibility complex. The susceptibility-associated HLA types are common relative to the disease and are not sufficient to determine its occurrence. The example therefore combines a mechanistically important exposure with a genetic contribution that is substantial but not individually deterministic.

Xeroderma pigmentosum offers another clear connection. In many forms of the disorder, pathogenic variants impair nucleotide excision repair, a pathway that removes certain types of DNA damage. Ultraviolet radiation produces lesions whose consequences are especially severe when the relevant repair machinery is deficient. Here the environmental agent, molecular damage, and inherited repair defect can be connected directly. This differs evidentially from an observational interaction in which neither the causal variant nor the relevant exposure mechanism has been established.

A major complication in population studies is gene–environment correlation. Genetically influenced characteristics can affect exposure, and parental genotypes can influence both the genes and environments experienced by children. This correlation is conceptually distinct from interaction: correlation concerns which environments people experience, whereas interaction concerns whether environmental effects differ by genotype. The two can coexist, and failing to distinguish them can lead to incorrect causal interpretation.

Human interaction research also faces population structure, exposure misclassification, selection bias, and large testing burdens. Some early candidate-gene findings, particularly those involving complex behavioral outcomes and broadly defined adversity, did not show consistent replication. Robust interpretation requires evidence appropriate to the claim, including adequate sample sizes, reproducible exposure definitions, and replication. Even a well-supported population interaction does not, by itself, yield an accurate individual prognosis or establish that genetically stratified interventions improve outcomes.

Genome-wide analysis and predictive limits

Modern genomic methods allow environmental dependence to be studied across the genome rather than only at selected candidate loci. A genome-wide association study can include variant-by-exposure terms or compare genetic effects across environmental strata. Such analyses may reveal variants whose effects are weak when averaged across the entire sample but substantial under particular conditions. They can also test whether apparently stable average associations conceal heterogeneity across populations or settings.

At the molecular level, expression quantitative trait locus studies connect genetic variants with differences in RNA abundance. Repeating these studies across tissues, developmental stages, or experimental treatments identifies context-dependent regulatory effects. A variant may influence expression only after immune stimulation, for example, or affect different genes in different cell types. RNA sequencing and related assays allow the response to be measured broadly, but observed expression changes still need to be distinguished from changes in the proportions of different cell populations.

Perturbation experiments help move from association toward mechanism. Genome editing can introduce or alter a candidate variant while environmental conditions are experimentally controlled. Factorial combinations of edited genotypes and exposures can then test whether the variant changes the response. Such experiments are powerful, but their scope is limited by the model system: a response in cultured cells may not reproduce organism-level physiology, developmental history, or interactions among tissues.

For highly polygenic traits, individual interaction effects may be too small to estimate reliably. Models can instead examine whether aggregate genetic variance or the performance of polygenic scores changes across environments. These analyses require particular care because scores reflect the populations, exposure distributions, and statistical procedures used to construct them. Differences in predictive accuracy across settings can arise from many sources besides biological interaction, including allele-frequency differences, measurement differences, and changes in correlations among genetic variants.

Machine learning methods can represent nonlinear responses and high-dimensional combinations of genomic and environmental predictors. Their flexibility does not automatically establish causation or provide reliable extrapolation. Randomly assigning observations to training and test sets may overstate performance when closely related individuals, repeated measurements, or nearly identical environments occur in both. Holding out entire locations, years, families, or exposure ranges can provide a more demanding assessment of the intended prediction task.

A continuing difficulty is the mismatch between richly measured genomes and incompletely measured environments. Lifetime exposures vary in timing, intensity, and combination, while many studies capture only a few snapshots. Another is distinguishing true changes in genetic effects from changes in outcome variability or measurement scale. Progress depends on integrating carefully characterized exposures, explicit biological hypotheses, appropriately structured statistical models, and independent experiments capable of testing the predicted response under new conditions.