The breeder’s equation is a classical equation in evolutionary theory, and basically states that the response of a trait to selection is equal to the strength of selection on this trait multiplied by the heritability of the trait. There can be several reasons why reality does not conform to a narrow interpretation of this equation, but in spite of that, let us actually indulge in a broad interpretation of the equation. Typically, the heritability of a trait is interpreted to refer to the genetic basis of a trait (the additive genetic variance). However, a broad interpretation of the breeder’s equation would be that if selection acts on a trait and that trait is somehow heritable, we should expect to observe an evolutionary change in that trait. In other words, the source and mechanism of inheritance does not matter for the breeder’s equation to have applicability.

In this paper, Pujol et al. (2025) follow this rationale for epigenetic variation, and more specifically for differentially methylated regions in the model plant Arabidopsis thaliana. They used 120 lines which vary across 126 differentially methylated regions of the genome, but are genetically identical, thereby eliminating additive genetic variance from the heritability component of the breeder’s equation. Previous studies have already shown that these differentially methylated regions are associated with different phenotypes, suggesting they somehow functionally affect the development of the plants. Previous studies have also shown that these differentially methylated regions are stably transmitted across generation, i.e., that they are heritable. This predicts that if one selects for different phenotypes, there should be an evolutionary response in the phenotypes, and an associated shift in the differentially methylated regions that are responsible for the phenotypic variation.

By and large, this is what Pujol et al. (2025) found. They selected divergently on four different plants traits (biomass, rosette size, flowering time, and height at first fruit), in two different populations, and with two different selection strengths. For most traits they found a phenotypic shift, and this largely corresponded with associated shifts in the differentially methylated regions. The patterns were very congruent between the different selection strengths, and very similar for the two different populations. We can therefore confidently conclude that epigenetic variation has effects on fitness-relevant phenotypic traits, that therefore natural selection can act on this epigenetic variation, and if so that this would cause an evolutionary change in the phenotypes under selection and in the differentially methylated regions causing this phenotypic variation. I.e., the breeder’s equation can indeed by applied to more than just genetically heritable variation, it can also be applied to non-genetically heritable variation. 

While this seems like a very obvious result, there are still very few studies that show how selection on non-genetic heritable variation results in a between-generation shift in this non-genetic heritable variation. In that sense, this study can function as a case in favour of a broader interpretation of evolution; to refer to a heritable between-generation change in the composition of a population, independent of whether the heredity is genetic or non-genetic. Critics might say that this study has only limited validity, since the epigenetic variation has been artificially increased in the used plant lines, and may not reflect naturally occurring variation. Indeed, more studies on natural selection on epigenetic variation are called for, but these will have the added complication that they need to correct for genetic variation, which is often correlated to epigenetic variation. Here, the authors could exclude genetic variation by working with epigenetically different lines that were genetically identical.

The study stands out in yet another respect – it uses a highly unusual and apparently novel selection design (although the authors do not know for sure if their idea is actually novel). Typically, in a selection study only a subset of parents is allowed to reproduce out of the total set of potential parents. If one wants to select for different traits separately, or with different selection strengths, separate selection studies need to be undertaken. Here the authors took a different approach. They let all their parents reproduce (3 offspring per plant), and phenotyped all offspring. They then, virtually, selected for different traits or for different selection strengths in the parents by excluding from their dataset the offspring of parents that are selected not to reproduce – after they had actually already reproduced! It is almost as if selection acts with a time delay, destroying not only the selected parents but also their offspring. This has the benefit that a single dataset is generated (all phenotyped offspring of all parents), and then selection can be exerted as often as one wants, for example for distinct types of selection, or on different (combinations) of traits. While the initial effort to phenotype offspring of all potential parents is large, and exponentially so if multiple generations are selected, the subsequent evaluation of distinct types of selection on any of the phenotyped traits is virtually effortless. For example, the authors applied divergent selection, but could reuse their dataset to apply stabilising selection. This novel selection design may therefore be a very attractive approach for many other researchers.

References

Benoit Pujol, Mathieu Latutrie, Pierick Mouginot, Nelia Luviano- Aparicio, Jésaëlle Piquet, Sara Marin, and Stéphane Maury (2025) Experimental evidence for short term directional selection of epigenetic trait variation. Zenodo, ver.2 peer-reviewed and recommended by PCI Evolutionary Biology https://doi.org/10.5281/zenodo.15227609