Which Generation Was Never Exposed? The Test Most Epigenetic Inheritance Studies Skip — Epoche C2
The claim is now common outside the specialist literature: a famine, a stress or a toxicant experienced by one generation leaves epigenetic marks — chemical modifications of DNA and of the histone proteins around which it is wound, which alter transcription without altering sequence — that descendants inherit. The problem is that most published support does not meet the one requirement distinguishing that claim from something far more ordinary. The requirement is a counting requirement, and it is met by breeding far enough that some generation contains no cell which was itself present during the exposure. This essay sets out that requirement, identifies three ways in which the mammalian evidence base fails it, and then asks what a study would have to look like to succeed — the requirement being demanding, not impossible. The problem: exposure is not confined to the generation that is dosed If a phenotype traces to an ancestor's exposure, two explanations compete: something heritable passed through the germ line, or the descendant was itself exposed. Distinguishing them is arithmetic, not biology. Treating a pregnant female doses three generations at once: her own body, the fetus inside her, and the primordial germ cells inside that fetus, which found the generation after it. Generation Directly exposed? By what route F0 — the treated pregnant female yes her own tissues F1 — her fetus yes in utero F2 — the fetus's germ cells yes as primordial germ cells inside F1 F3 — the first clean generation no — The first generation carrying no directly exposed cell is therefore F3. For a treated male the count is shorter: his germ cells are exposed, so the F1 conceived from them derives from an exposed cell, but F1's own germ line forms afresh after fertilisation and was never present during the dosing. F2 is the first clean generation. The convention is not the author's; it was set out in the toxicology literature by Michael Skinner in 2008 — eighteen years ago — and it is worth stressing that Skinner is among the most prominent advocates of mammalian transgenerational inheritance rather than a critic of it. The standard is the field's own, proposed by someone whose results it constrains. Two riders follow from the same counting. First, the lineages must be outcrossed: an F3 produced by mating F2 siblings is not a clean generation from an independent line but one branch of a single pedigree, and any founder effect propagates undetected through it. Second, many reported effects stop at F2 after a gestational exposure, and those results are entirely consistent with a direct developmental effect on cells present at the time of treatment. They do not test inheritance at all. The clearest illustration is also the best-designed human study in the area, and it is instructive precisely because its design is exemplary and its conclusion still says nothing about inheritance. Bastiaan Heijmans and colleagues (2008) compared individuals conceived during the Dutch Hunger Winter of 1944–45 with their own unexposed same-sex siblings, and measured DNA methylation at the differentially methylated region of the imprinted IGF2 gene in blood some six decades later. Methylation was lower in the periconceptionally exposed siblings by a few percentage points, and — the detail that makes the study convincing — no difference appeared in those exposed only late in gestation, which is what one expects if the window of vulnerability is the period of germ-line and early-embryonic reprogramming rather than a general effect of malnutrition. The sibling design removes most of the confounding that wrecks observational epigenetics. And the subjects were exposed. They are F1 by construction; the study is a study of developmental programming, not of inheritance. Two further routes transmit a trait vertically without the germ line, and both mimic inheritance closely enough that a design which ignores them cannot claim to have excluded it. The first is behaviour: a dam whose own development was perturbed may rear her pups differently, so a phenotype propagates down the maternal line through nursing and care. Cross-fostering at birth — assigning pups to dams of the other treatment group — is the standard control, and its absence is a design defect rather than an oversight. The second is the transfer of material from mother to offspring outside the gamete: milk constituents, and the founding microbial community of the gut. Neither is an epigenetic mark, and neither is excluded by measuring one. Three reasons the shortfall persists First, the mammalian germ line erases marks twice. DNA methylation, the most frequently invoked carrier, is removed in two genome-wide waves. After fertilisation the paternal genome is actively stripped and the maternal genome loses methylation passively across the cleavage divisions; imprinted regions, whose parent-of-origin methylation must persist for normal development, are protected in this first wave. Then, in the primordial germ cells that found the next generation, a second and deeper wave removes what the first left. Stefanie Seisenberger and colleagues (2012) sequenced the methylomes of mouse primordial germ cells through this window and found the fraction of methylated CpG dinucleotides falling from roughly 70 per cent in the epiblast to about 14 per cent at embryonic day 11.5 and below 10 per cent by day 13.5. An inherited mark must survive both waves; imprinted loci escape only the first, and have their marks erased and re-set according to the sex of the germ line they are passing through, which is exactly why they cannot carry a message about the grandmother's diet. Some sequences do escape both. The best-characterised mammalian case is the agouti viable yellow allele of the mouse — an epiallele, meaning a variant distinguished by its epigenetic state rather than by its sequence — described by Hugh Morgan and colleagues in 1999: an intracisternal A particle retrotransposon inserted upstream of the