你的肝细胞和你的神经细胞，形状、寿命、干的活，没有一样是像的。可把两者的 DNA 抽出来测，是同一份。
同一本书，两个部门照着它做出了完全不同的东西。差别只能在别处：哪几页被封住了、哪几页翻得开。
这层封条是真的，而且会随着细胞分裂被抄下去。麻烦出在这个「能被抄下去」上——它一路被推成了「你祖辈的经历会写进你的基因」。中间跨过了好几道关，其中一道，身体每传一代就要过两遍。
封条有两种。一种直接挂在 DNA 上：在某些字母上贴一个很小的化学零件（一个碳拖着三个氢，叫甲基），贴上之后那一段通常就读不出来了。另一种挂在包装上：DNA 不是一根线散在细胞里，它一圈圈缠在一串蛋白质线轴上，线轴上也带标记，决定这一段缠得紧还是松——缠得死紧的那段，负责读取的机器根本够不着。
要紧的是第二件事：这些封条在细胞分裂时会被抄一份。DNA 刚复制出来的新链上本是空白的，有一种酶专门沿着链走，看见老链这个位置贴着甲基、新链上没有，就在新链对应位置补一个。于是一个肝细胞分裂出来的两个细胞还是肝细胞。所谓「细胞记得自己是谁」不是比喻，就是这排封条被一代代抄下去。
最干净的肉眼证据是三花猫。管毛色的一个基因坐在 X 染色体上。雌性哺乳动物每个细胞有两条 X，胚胎很早的时候，每个细胞会把其中一条整条封住——1961 年 Mary Lyon 提出这个机制。封哪一条是随机的，可一旦定了，这个细胞往后分裂出来的所有后代都封同一条。所以三花猫身上那些橙色块和黑色块，每一块就是当初某一个早期细胞的后代摊开的样子，块与块的边界，是那次随机分工留下的界线。顺带也解释了为什么三花猫几乎都是母猫：得先有两条 X，才有得可封。
这一层从头到尾没动 DNA 上的任何一个字母。序列完好地待在那儿，只是有些段落被贴了「别读」。这就是「表观」两个字的意思。
EN There are two kinds of seal. One is attached directly to the DNA: a very small chemical part (one carbon carrying three hydrogens, called a methyl group) is stuck onto certain letters, and once it is there that stretch usually cannot be read. The other is attached to the packaging: DNA does not lie about loose as a single thread, it is wound around 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 tight, the reading machinery simply cannot reach it. The second point is what matters: these seals are copied when a cell divides. A freshly copied DNA strand starts out blank, and there is an enzyme whose job is to walk along it, see a methyl on the old strand where the new strand has none, and add one at the matching spot. So the two cells a liver cell divides into are still liver cells. "A cell remembering what it is" is not a metaphor; it is this row of seals being copied down the generations of cells. The cleanest visible evidence is a calico cat. One of the genes controlling coat colour sits on the X chromosome. Each cell of a female mammal has two X chromosomes, and very early in the embryo every 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 descendant of that cell shuts down the same one. So the orange patches and black patches on a calico are each the spread-out descendants of one 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 "epi-" means here — the layer sitting on top of inheritance.
既然封条能随分裂抄下去，传给下一代好像顺理成章。身体偏不让。
哺乳动物每传一代要过两次大清洗。第一次在受精之后：精子和卵子各自是高度特化的细胞，一身「我是精子」「我是卵子」的封条，而受精卵必须是一张什么都能变的白纸，所以早期胚胎里的甲基化被大面积抹掉再重贴。第二次更彻底：胚胎很早就划出一小群细胞，将来专门做精子和卵子，这群细胞里的甲基化被抹到接近零。
所以，环境在你身上写下的封条要传到孙辈，得连着躲过两次清洗。哺乳动物这边「跨代传递」难，难在这儿，是道物理上的关卡，不是哲学分歧。
但确实有漏网的。有一小批基因，封条能穿过清洗，而且带着「我来自父亲」还是「我来自母亲」的记号，这批基因只有来自一方的那份在干活，另一份被封着。这叫基因组印记。
人身上有个说服力极强的例子。Prader-Willi 综合征和 Angelman 综合征表现完全不同，病因却落在同一条染色体的同一小段上：同样一段缺失，缺的是父方那份，长成前一种；缺的是母方那份，长成后一种。同一段 DNA、同样的损伤，只因为来自哪一方不同，就是两种病——「来源标记」确实完整地穿过了生殖细胞。至于为什么会有这种标记，主流解释是父方和母方的那份基因在「胎儿该从母亲身上拿走多少」上利益不一致，父方那份推着多拿、母方那份压回去；这个还在争。
植物那边情况完全不同，证据也硬得多。植物不早早划出生殖细胞——枝条上一个普通细胞，长着长着就能变成花、变成配子。林奈 1744 年描述过一种柳穿鱼：正常的花左右对称，这一种变成了辐射对称。1999 年查清楚，两种花在那个管对称性的基因上序列一模一样，差别只是这个基因被甲基化封死了，而且这个封条稳稳传了两百多年。所以「表观能不能跨代传」这个问题，问的时候必须带上物种：在植物和线虫里是稳的，在哺乳动物里不是。
EN If seals are copied 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 become 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 almost to zero. So for a seal the environment wrote on you to reach your grandchildren, it must dodge two erasures back to back. That is where "transgenerational transmission" 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 works and 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. Why such tags exist at all is usually explained by a conflict of interest between the father's copy and the mother's copy over how much a fetus should draw from the mother: the father's copy pushes for more, the mother's pushes back. That explanation 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 go on to become a flower, and gametes. In 1744 Linnaeus described a toadflax whose flowers, normally bilaterally symmetrical, had become radially symmetrical. In 1999 it was worked out that the two forms have identical sequence at the gene controlling symmetry, and that the only difference is that the gene has been sealed shut by methylation — a seal that has been passed down stably for more than two centuries. Nematodes offer cases that persist for many generations too. 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.
1944 年冬到 1945 年春，荷兰西部被封锁，城市配给一度掉得很低，史称饥饿冬天。它有明确的起止日期，荷兰又有完整的出生记录，成了一个谁也设计不出来的天然实验。
2008 年 Heijmans 一组人测了一批在饥荒期间受孕的人：六十年后，他们血里一个印记基因的甲基化，比他们自己的同性别兄弟姐妹（同一对父母，只是没赶在饥荒那几个月受孕）低。而且只有受孕前后正好撞上饥荒的那批有这个差别，怀到后期才撞上的没有。
这证明了什么？证明「一次环境冲击能在被它直接打到的那个人身上留下几十年的痕迹」。这一半非常硬。它没有证明第二半：这个痕迹被传给了没经历过的人。
中间隔着一个特别容易跳过去的混淆。一个怀孕的女人挨饿时，同时有三代人被直接暴露：她本人；她肚子里的胎儿；以及那胎儿体内当时已经形成的生殖细胞——将来的孙辈。所以在孙辈身上看到痕迹，仍然可能只是直接暴露的结果。要说「传下去了」，至少得看到第四代：第一个从头到尾没被碰过的世代。在人身上，这道门槛几乎够不着。
老鼠那边有个传播极广的实验。2014 年 Dias 和 Ressler 把公鼠训练成害怕一种樱花似的气味（气味配轻微电击），报告说它的子代和孙代对这个气味也更敏感，精子里对应那个气味受体基因的甲基化也变了。这个结果后来被专门挑过统计上的毛病——同一篇里十项检验全部显著，按它自己报出的效应量算，连中十次本身就不太可能；到今天也没被稳固地独立重复出来。诚实的说法是：它是个悬着的结果，不是一块能往上盖东西的地基。
那「拉马克复活了」错在哪？拉马克的主张从来不是「环境能留下痕迹」——这个谁都同意。他的主张是：用进废退产生的、朝着有用方向的改变会被传下去。而现在有证据的那部分，恰恰不含「朝着有用方向」这一条。饥荒留下的那排封条不是一套「省着用能量」的聪明方案，它是一次冲击的痕迹，落在哪儿、是好是坏，没有谁替它挑过。
反过来说，假如某个封条真能稳定传很多代、又影响后代的数量，那它就是选择的对象，跟序列上的变异没有本质差别。卡点在稳定性：序列能传几百万年，封条在哺乳动物里传几代就糊了，攒不住长期的东西。
表观遗传真正改写的不是遗传学，是「先天对后天」这个二分。环境的作用不是加在基因之外的另外一项，它就落在基因的开关上。「演化发育生物学」那期讲过改开关比改基因本身便宜，这里是同一件事在个体一生的尺度上发生。
EN 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 the Dutch birth records are complete, so it became a natural experiment nobody could have designed. In 2008 Heijmans and colleagues measured 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 seen in the grandchildren may still be nothing more than direct exposure. To claim transmission you need at least a fourth generation: the first one never touched at any point. In humans that bar is nearly out of reach. In mice there is a famous 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 criticised on statistical grounds — ten tests in one paper, all significant, which given the effect sizes the paper itself reported is an unlikely run — and it has still not been robustly replicated independently. The honest description is that it is an unsettled result, not a foundation to build on. The Swedish archival study in the far north (harvest quality in a grandparent's childhood against grandchildren's mortality) likewise has small samples and heavy dispute. So where exactly 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, are passed on. And the part for which evidence exists is precisely the part that excludes "pointed in a useful direction." The seals a famine leaves are not a clever energy-saving scheme; they are the traces of a shock, and nobody selected where they landed or whether they help. There is a deeper reason too. 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 — variation plus inheritance plus differential reproduction runs just the same. 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.