Your liver cells and your neurons share nothing: not their shape, not their lifespan, not the work they do. Yet pull the DNA out of each and measure it, and it is the same DNA.
The same book, and two departments have built completely different things from it. The difference has to lie somewhere else: which pages are sealed shut and which ones open.
That layer of seals is real, and it is copied along when a cell divides. The trouble starts with "copied along" — it gets pushed all the way to "your ancestors' experiences are written into your genes." Several checkpoints sit in between, and one of them the body itself passes through twice every generation.
There are two kinds of seal. One is stuck directly onto the DNA: a tiny chemical part (one carbon carrying three hydrogens, called a methyl group) is attached to certain letters, and once it is there that stretch usually cannot be read. The other is attached to the packaging: DNA does not lie around loose as a thread, it is wound turn after turn onto a string of protein spools, and the spools carry marks of their own that set how tightly a stretch is wound — where it is wound dead tight, the machinery that reads DNA simply cannot reach.
The second fact is the one that matters: these seals are copied when a cell divides. A freshly copied DNA strand starts out blank, and there is an enzyme whose whole job is to walk along it, spot a methyl group on the old strand where the new strand has none, and add one at the matching position. So the two cells a liver cell divides into are still liver cells. "A cell remembers what it is" is not a metaphor; it is this row of seals being copied down, cell generation after cell generation.
The cleanest evidence you can see with your eyes is a calico cat. One of the genes for coat colour sits on the X chromosome. Every cell of a female mammal has two X chromosomes, and very early in the embryo each cell shuts one of them down entirely — the mechanism Mary Lyon proposed in 1961. Which one gets shut down is random, but once it is settled, every cell descended from that one shuts down the same one. So the orange patches and black patches on a calico are each the spread-out descendants of a single early cell, and the boundaries between patches are the lines left by that random division of labour. It also explains in passing why calicos are almost always female: you need two X chromosomes before there is anything to shut down.
None of this touches a single letter of the DNA. The sequence sits there intact; some passages have simply been tagged "do not read." That is what the "epi-" in epigenetics means.
If seals are copied along at cell division, passing them to the next generation sounds like it should follow. The body will not have it.
Mammals go through two rounds of wholesale erasure per generation. The first comes after fertilisation: sperm and egg are each highly specialised cells covered in "I am a sperm" and "I am an egg" seals, while the fertilised egg has to be a blank sheet that can still turn into anything, so methylation across the early embryo is largely wiped and reapplied. The second is more thorough: very early on, the embryo sets aside a small group of cells that will later make sperm and eggs, and methylation in that group is stripped down to nearly zero.
So for a seal the environment wrote on you to reach your grandchildren, it has to dodge two erasures back to back. That is where transmission across generations gets hard in mammals — a physical checkpoint, not a philosophical disagreement.
Some things do slip through. A small set of genes carry seals that survive the erasures, tagged with whether the copy came from the father or from the mother; only one parent's copy does any work, while the other stays sealed. This is called genomic imprinting.
There is an exceptionally convincing human case. Prader-Willi syndrome and Angelman syndrome look nothing alike, yet both trace to the same small stretch of the same chromosome: the same deletion gives the first when the father's copy is the one missing, and the second when the mother's is. Same DNA, same damage, two different conditions purely because of which parent it came from — the parent-of-origin tag really does pass through the germ cells intact. As for why such tags exist, the mainstream explanation is a conflict of interest between the father's copy and the mother's copy over how much a fetus should draw from the mother, with the father's copy pushing for more and the mother's pushing back; that is still argued over.
Plants are a different situation entirely, and the evidence there is far harder. Plants do not set germ cells aside early — an ordinary cell on a branch can grow on to become a flower, and gametes. In 1744 Linnaeus described a toadflax whose flowers, normally bilaterally symmetrical, had turned radially symmetrical. In 1999 it was worked out that the two forms have identical sequence at the gene governing symmetry, and that the only difference is that the gene has been sealed shut by methylation — a seal that has held steady for more than two centuries. So the question "can epigenetic marks be inherited across generations" has to be asked with a species attached: in plants and nematodes, yes, stably. In mammals, no.
From the winter of 1944 into the spring of 1945 the western Netherlands was blockaded and urban rations fell very low — the Hunger Winter. It has clear start and end dates, and Dutch birth records are complete, which made it a natural experiment nobody could have designed.
In 2008 Heijmans and colleagues measured a group of people conceived during the famine: six decades later, methylation at one imprinted gene in their blood was lower than in their own same-sex siblings — same parents, simply not conceived in those famine months. And only those whose conception coincided with the famine showed the difference; those who met it late in gestation did not.
What does that establish? That a single environmental shock can leave a decades-long mark on the person it hit directly. That half is very solid. It does not establish the second half: that the mark was passed to someone who never experienced it.
In between sits a confound that is very easy to skip over. When a pregnant woman goes hungry, three generations are exposed directly at once: the woman herself; the fetus she is carrying; and the germ cells already formed inside that fetus — the future grandchildren. So a mark found in the grandchildren may still be nothing more than direct exposure. To claim it was passed down, you need at least a fourth generation: the first one never touched at any point. In humans that bar is nearly out of reach.
On the mouse side there is a wildly travelled experiment. In 2014 Dias and Ressler trained male mice to fear a cherry-blossom-like odour (odour paired with a mild shock) and reported that their offspring and grand-offspring were more sensitive to the same odour, with altered methylation at the matching odorant-receptor gene in sperm. The result was later picked apart on statistical grounds — ten tests in one paper, all of them significant, which given the effect sizes the paper itself reported is an improbable run — and to this day it has not been robustly replicated by anyone else. The honest description: an unsettled result, not a foundation you can build on.
So where does "Lamarck is back" go wrong? Lamarck's claim was never "the environment can leave a mark" — everyone agrees on that. His claim was that changes produced by use and disuse, pointed in a useful direction, get passed on. And the part for which evidence exists is exactly the part that excludes "pointed in a useful direction." The row of seals a famine leaves is not a clever energy-saving scheme; it is the trace of a shock, and nobody picked where it landed or whether it helps.
Turn it around: if some seal really could pass stably for many generations and also affect how many offspring you leave, it would be a target of selection, no different in kind from a change in the sequence. The sticking point is stability — sequence lasts millions of years, while in mammals a seal blurs out within a few generations and cannot accumulate anything long-term.
What epigenetics actually rewrites is not genetics but the nature-versus-nurture split. The environment's contribution is not a separate item added alongside the genes; it lands on the switches of the genes themselves. The piece on evolutionary developmental biology made the point that changing a switch is cheaper than changing the gene; this is the same thing happening on the scale of a single lifetime.