玉米粒的颜色有时候是花的。同一穗上，有的粒匀匀的深紫，有的粒上溅着斑点，像被谁拿笔尖戳过几下。
上世纪四十年代，有位研究者盯着这些斑点看了很多年，结论当时几乎没人接受：玉米染色体上有一段东西会自己挪窝——挪进管色素的基因里，那块就不上色；哪个细胞里它又挪出去，这个细胞的后代就恢复上色。
真正麻烦的问题在后面。这段会挪窝的东西，对玉米没有任何用处。那它凭什么还在，而且越来越多？
先把词拆开。一个生物的全部 DNA 合起来叫基因组，里头只有一小块是造蛋白质的说明书。人的基因组里大约有一半，是一类叫转座子的序列留下来的——这类序列彼此差别很大，共同点只有一个：它能让自己的拷贝数变多。有的先转出一份 RNA 副本、再倒写回 DNA 插到别处，等于复印；有的干脆从原地剪下来贴到别处。
数量比想象的夸张。人身上有一个叫 Alu 的家族，拷贝数超过一百万份，占全部 DNA 的一成以上。玉米更极端，基因组里八成五都是这类序列——那位盯斑点的研究者后来拿了诺贝尔奖，而她当年看的那株玉米，绝大部分基因组本来就是会挪窝的东西。
这里有一句特别容易顺口说出来的错话：不能讲它们「想要扩散」。没有谁想要什么。真正在跑的逻辑很枯燥：某段序列如果碰巧带着让自己被复制的性质，下一代里带着它的拷贝就比不带的多。它不需要对携带它的个体有任何好处，只需要复制自己的速度快过被清除的速度。这是「基因视角」那期讲的复制子逻辑最赤裸的形态——选择从不结算「对谁好」，只结算谁留下的拷贝多。
而它对宿主常常有害：插进一个正在工作的基因中间，那个基因就废了，人类有好几种遗传病就是这么来的。好在大部分拷贝早就报废、跳不动——那一百万份 Alu 里还能产生新插入的只有极少数几个源头，另一类叫 LINE-1 的序列，全基因组还有活性的大约只剩八十到一百份。
EN Take the word apart first. All of an organism's DNA together is its genome. Only a small slice of it is instructions for making proteins; most of the rest is not. Roughly half of the human genome is what a class of sequences called transposable elements has left behind. These sequences differ wildly from one another and share exactly one property: they can increase their own copy number. Some make an RNA copy and write it back into DNA somewhere else, like running off a photocopy; others simply cut themselves out and paste themselves in elsewhere. The numbers are more extreme than you would guess. One family in humans, called Alu, runs to over a million copies and makes up more than a tenth of all our DNA. Maize is more extreme still: about eighty-five percent of its genome is made of such sequences — and the researcher who spent years staring at those speckled kernels, who later received a Nobel Prize, was looking at a plant whose genome is overwhelmingly composed of things that move. One sentence is dangerously easy to say here, and it is wrong: they do not "want" to spread. Nothing wants anything. The logic actually running is dull: if a stretch of sequence happens to carry a property that gets itself copied, then in the next generation there are more copies of it than of things without that property. It needs no benefit at all for the individual carrying it; it only needs to copy itself faster than it gets cleared out. This is the replicator logic from the gene's-eye view in its barest form — selection never settles accounts on who benefits, only on who leaves more copies. And the host often pays: land in the middle of a working gene and that gene is wrecked, which is the origin of several human genetic diseases. Fortunately most copies are long dead and can no longer move — of that million Alu copies only a handful of source elements can still generate new insertions, and of the family called LINE-1 only some eighty to a hundred copies in the whole genome remain active.
你从父母各拿到一整套染色体。做精子或卵子时两套要分开，每个配子只带走一套——所以对任何一个位置，父方那份和母方那份各有一半机会被传下去。这条「一半」是遗传学里最基本的账，基本到我们从来不问它凭什么成立。
有些东西能改这个比例。家鼠身上有一段染色体片段（研究者管它叫 t 单倍型），带着它的公鼠，后代里拿到这段的比例不是一半，是九成以上。
办法并不体面。这只公鼠做出来的精子里，带它的和不带它的都有；带它的那套东西会让整批精子都中一种毒，而解药只在带它的那一半精子里。结果不是它跑得更快，是别的精子被搞坏了。果蝇身上有一段染色体在用差不多的手法。
代价明摆着。这类片段常常拖着致命缺陷：两份都拿到的胚胎活不下来，或者公鼠彻底不育。从「这群老鼠整体划不划算」看，这是纯亏损。但选择不结算群体的总账——它在每个携带者身上把自己的传递概率从五成抬到九成，抬得比它造成的死亡损失更快，它就留得下来，在野外鼠群里一直都在。
这类现象统称减数分裂驱动，母方也有版本：做卵子时分裂出四个产物，只有一个能变成卵、其余三个作废——「谁占到那个名额」天生就是一条能被偏置的通道，玉米里就有一种染色体结构更容易占到它。
知道这些之后，问题整个反过来了：不该问为什么有人作弊，该问为什么绝大多数位置上不作弊。
EN You get one full set of chromosomes from each parent. When sperm or eggs are made the two sets have to separate, and each gamete carries away only one — so at any given position, the father's copy and the mother's copy each have a fifty percent chance of being passed on. That "half" is the most basic accounting in genetics, so basic that we never ask what makes it hold. Some things can change the ratio. House mice carry a stretch of chromosome (researchers call it the t-haplotype) that gets passed to more than ninety percent of a carrier male's offspring instead of half. The method is not dignified. The male makes both kinds of sperm, with and without it; the version that carries it poisons the whole batch, and the antidote is present only in the half that carries it. It does not swim faster — the others were sabotaged. A stretch of chromosome in fruit flies uses much the same trick. The cost is out in the open: such stretches usually drag a lethal defect along, so embryos that inherit two copies do not survive, or carrier males are entirely sterile. Judged by whether the mouse population as a whole comes out ahead, this is a pure loss. But selection does not settle the group's books — inside each carrier the stretch lifts its own transmission from fifty percent to ninety, faster than the deaths it causes drag it down, so it persists, and it is indeed found in wild mouse populations. The phenomenon is called meiotic drive, and there is a maternal version too: making an egg, meiosis produces four products and only one becomes the egg while the other three are discarded — "who gets that slot" is a channel that can be biased by its nature, and maize has a chromosome structure that makes itself more likely to take it. Once you know this, the question inverts: not why anything cheats, but why almost nothing does.
人的 15 号染色体上有一小段，缺了会出问题。奇怪的是出哪种问题，取决于缺的是父亲给的那条还是母亲给的那条：缺在父源那条上，孩子会得一种病（婴儿期吸吮无力，之后食欲极难控制）；缺在母源那条上，是一种症状完全不同的病。同样大小的缺口，两种结果。
这说明细胞认得出每份拷贝来自谁。精子和卵子形成时会给某些区段打上化学标记——和「表观遗传与遗传的边界」那期是同一套标记系统——被标记的那份就关着不表达。这叫基因组印记。人和小鼠身上这么办的基因各只有一两百个，是很小的一批。
而这一小批的功能高度扎堆：大多和「胎儿从母体身上拿多少资源」有关。小鼠有个促进生长的基因，正常只有父源那份在工作；把父源那份弄坏，幼崽出生时只有正常体重的六成，把母源那份弄坏则毫无影响——它本来就关着。而压住生长的那一侧，往往是母源在表达。
Haig 给的解释是这样：在一个母亲可能和不同雄性生下多胎的物种里，某个胎儿体内那份来自父亲的拷贝，未必出现在母亲将来的其他孩子体内；而来自母亲的那份，在她每个孩子体内都有一半机会在场。于是「多从母体索取一点」这件事，两份拷贝的账不一样：索取会消耗母亲、压低她将来的后代数，这笔代价母源拷贝要承担，父源拷贝承担得少。结果是，让胎儿多要一点的变异，从父亲那边传下来时更容易被留下；压住索取的，从母亲那边更容易被留下。同一个基因的两份拷贝就这样被推成了两种设定。
这个解释还不是定论。它对管生长的那批印记基因解释得漂亮，但印记基因里还有一批长在脑子里、功能说不清，另有几种假说在竞争。这块目前还有争论。
把三节放一起看：基因组里有自我复制的段落、有偏置传递的作弊者、有互相拉扯的父母双份，它居然还能当成「一个个体」运转——这不是天然的。生殖细胞里有一整套小 RNA 专门按住转座子；每代重组把染色体打散，让作弊者拖不动帮凶。「主要演化转变」那期说过，个体性是被压制出来的，不是给定的——基因组内部就是这句话最小的例子。
EN There is a small stretch of human chromosome 15 that causes problems when it is missing — and which problem depends on whether the missing copy came from the father or the mother. Lost from the paternal chromosome, the child has one condition (weak sucking in infancy, then appetite that is extremely hard to control); lost from the maternal one, a condition with entirely different symptoms. Same DNA, same size of gap, two outcomes. So cells can tell which parent each copy came from. As sperm and eggs form, certain regions get chemical marks — the same marking system covered in the piece on epigenetics and the limits of inheritance — and the marked copy stays switched off. This is genomic imprinting, and only one or two hundred genes in humans and in mice do it. The functions of that small set cluster tightly: most of them bear on how much a fetus draws from its mother. Mice have a growth-promoting gene where normally only the paternal copy works; break the paternal copy and the pups are born at about sixty percent of normal weight, while breaking the maternal one changes nothing, since it was switched off anyway. On the restraining side, it is often the maternal copy that is expressed. Haig's explanation runs like this: in a species where a mother may bear litters by different males, the paternal copy inside one fetus is not necessarily the same copy present in the mother's future offspring, whereas her own copy has a fifty percent chance of being present in every child she has. So "draw a bit more from the mother" scores differently on the two ledgers: drawing more depletes her and lowers her future offspring count, a cost the maternal copy bears and the paternal copy largely does not. The result is that variants letting a fetus take more are more readily retained when inherited from the father, and variants restraining the taking are more readily retained when inherited from the mother — pushing two copies of one gene into two different settings. This is not settled: it explains the growth-related imprinted genes beautifully, but others sit in the brain with unclear functions, and rival hypotheses are in play. Put the three sections together and something more surprising appears — a genome holds self-copying stretches, transmission cheats, and two parental copies pulling against each other, and still runs as "one individual". That is not free. Germ cells carry a dedicated small-RNA system holding transposable elements down, and recombination each generation breaks chromosomes up so a cheat cannot drag a whole set of accomplices along. As the piece on the major evolutionary transitions put it, individuality is something suppression produces rather than something given — and the inside of a genome is the smallest example of that sentence.