Suter, Boffelli and Martin (SBM)  have usefully engaged with our recent article  and in doing so have helped to expose our core concerns. Fundamentally, they are in agreement that there is no requirement for a paradigm shift away from the modern synthesis (MS).
We also agree with SBM on a core definitional issue. We used licking and grooming work in rodent models as an example of environmentally induced, transgenerational epigenetic inheritance. SBM categorize this as a serial parental effect, and not a case of epigenetic inheritance, noting that parental effects need not rely upon genetic or epigenetic change in the germ line. In this way, they are keeping to a version of Weismann's principle [3,4] and we think that this strategy is sound. But we note that those scholars we were discussing do not adopt this principle. For example, Danchin et al.  make a distinction between germline and experience-dependent epigenetic inheritance, the latter capturing our example. Our argument accepted their taxonomy and discussed its relation to the MS.
SBM believe that we make the type-identity claim that epigenetic states are genetically encoded and that the inheritance of an epigenetic state reflects the inheritance of a genetic state. At no point did we suggest that genetic state A is equivalent to epigenetic state A′. Our argument was that epigenetic mechanisms come under genetic control, and more generally that there is probably selection on epigenetic mechanisms as a number of their effects look like candidate adaptations. Given that we clearly understand epigenetic processes to operate above the gene and alter gene expression without recourse to altering DNA sequence, we can have no reason to assert an identity. Instead, we think that epigenetic actions are interestingly limited in terms of the effects they can produce, and those limitations are undoubtedly imposed by the nature of the DNA code they affect. This might not be the only cause in play. We used the term ‘control’ and we have since realized that this can be interpreted as a statement of Laplacian determinism. Really, this is an issue of constraint, and it is a view about how epigenetic effects come to have an informational role in the construction of the phenotype .
SBM shift focus from epigenetic mechanisms and states to epigenetic variation. The preceding discussion is an operational one about the functionality of epigenetic mechanisms. An interest in epigenetic variation is something else. SBM express the relationship between operational concerns and variation as follows:
Epigenetics is the study of phenomena in which highly complex molecular accretions to the genome determine stable states of gene expression. An epigenetic state is a functional state (active or inactive) of a transcriptional regulatory element such as a promoter or enhancer. Such alternative states are part of normal processes of gene regulation, as in cell differentiation, but they can also occur as aberrations, which may be termed epigenetic variants or ‘epimutations’.
(, p. 1)
SBM later refer to Richards's  taxonomy of obligate, facilitated and pure epigenetic variation. This is the difference between direct genetic control of epigenetic variation, genetically potentiated epigenetic variation and epigenetic variation generated by ‘stochastic events that are largely independent of genetic variation’ (, p. 398). They continue:
there is clear and abundant evidence for multigenerational inheritance of epigenetic states that are independent of genotype. This makes it possible for purely epigenetic states to participate in evolution if the traits they specify are subject to natural selection.
(, p. 2)
Our recommendation was to treat epigenetic systems to an adaptationist analysis and see them as part of the phenotype. SBM are making a different argument. Pure epigenetic variation could produce variants that affect gene expression in a novel way, and in turn affect the developing phenotype. In this way, ‘pure epigenetic inheritance is fully capable of serving as an ultimate mechanism of evolution in the sense that Dickins and Rahman have used that term: if an epigenetic variant is sufficiently stable over multiple generations, its associated phenotype may be subject to natural selection’ (, p. 1).
It is correct that there is much experimental evidence for the inheritance of epigenetic variation, as SBM defined it, and some evidence for the occurrence of this naturally . SBM's idea is that stochastic events can introduce random variation into natural epigenetic systems. If a randomly produced variant affects gene expression such that a novel phenotypic variant is produced that has an effect upon average lifetime inclusive fitness, then are we dealing with the domain of the special theory, the MS? It would be an interesting case because no change in DNA code had occurred, but rather gene regulation would have been affected.
In our view, any epimutation would have its effect in part as a consequence of the underlying DNA code it operates over. While epimutations might be produced at random, they only have bite when married to the gene. This is a key biological constraint that is ultimately in place because of natural selection. Perhaps this is an issue of levels of explanation, or rather levels of concern. The concern with epimutations is a purely proximate one.
Purely random epigenetic variation might be understood as a process used in order to deliver encapsulated phenotypic plasticity . In essence, this would permit a constrained random walk through phenotype space. The encapsulation will be delivered by the DNA code upon which the epimutations operate. Where a novel epimutation enhances fitness, over time, we would expect the interaction between such processes to affect selection pressures and lead to changes in gene frequencies. So, genetic mutations that alter the opportunity for randomly produced epimutations to have an effect could be favoured. This is the business of the MS.
Finally, if there are epimutational systems in play, as envisaged above, we ought to ecologically ground our understanding of them. Surely such constrained random walks within phenotypic space will only have adaptive advantage in very particular circumstances and the transition to near independence described by Richards may well be very rare? This is an empirical issue.
We should like to thank Ben Dickins for useful discussions around this topic.
The accompanying comment can be viewed at http://dx.doi.org/10.1098/rspb.2013.0903.
- Received July 13, 2013.
- Accepted September 11, 2013.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.