你上一次感冒，大概觉得那几天身体里在打仗。
可是病毒本身连打仗的家伙都没有。它不吃东西、不产生能量、不会自己动，离开细胞就是一粒漂在空气或水里的化学颗粒——一份说明书，裹在一层蛋白壳里，如此而已。
按直觉，这种什么都不带的东西应该是生物界边角上一小撮寄生虫。
实际上，你的 DNA 里大约有十二分之一是它们几千万年前留下的残骸，而胎盘之所以能长成，靠的正是其中一段。
细胞要多复制一份自己，需要一整套家什：把基因上的序列拼成蛋白质的装配车间（叫核糖体，细胞里成千上万台），供能的一套化学反应，把原料运进来的通道。病毒一样都没有。
一颗病毒颗粒挤进细胞之后，干的事只有一件：让这个细胞的车间照它的说明书开工。
别把这讲成精明。跑的账很枯燥：在碰巧出现的各种变异里，凡是丢掉了「自己造车间」那部分、又还进得去细胞的，说明书更短、复制更快、一次感染放出去的拷贝更多，于是这些留了下来。丢掉工具不是它付出的代价，是这条路上的默认结局——「内共生与真核革命」那期讲的线粒体，走的是同一种减法。
那这东西最初从哪来？桌上主要有三种想法。一是逃逸：细胞里本来就有能自我复制的 DNA 或 RNA 片段，其中某些碰巧配上了一层壳，从此能出门。二是退化：本来是能独立过日子的细胞，寄生久了一路丢基因，丢到只剩说明书。三是更早：细胞出现之前那锅汤里就有会自我复制的分子，病毒是没被并进细胞的那一支的后代。
要紧的是，这不是三选一的考题。病毒很可能根本不是一个有共同祖先的类群，不同的病毒走的是不同的路。2003 年，研究者在冷却塔水里的变形虫身上认出一类「巨病毒」，基因组一百多万个碱基对——碱基对就是 DNA 上的一个字母配对——比不少寄生细菌还大，甚至带着一批造蛋白用的零件，唯独没有核糖体。它究竟是缩水下来的细胞，还是一路捡基因胀起来的病毒，还在争论。另一条线索指向多次起源：许多亲缘上八竿子打不着的病毒，外壳蛋白折出来的形状却是同一种。
EN To copy itself, a cell needs a full kit: assembly plants that turn gene sequences into proteins (ribosomes, thousands per cell), a set of reactions for energy, channels for raw materials. A virus has none of it. Once a viral particle squeezes into a cell it does exactly one thing: make that cell's plants run off its instructions. Do not read this as cunning. The accounting is dull: among the variants that happen to arise, the ones that dropped the "build your own plant" part and could still get into cells had shorter instructions, copied faster, and released more copies per infection — so those are what persisted. Losing the toolkit is not a price paid; it is the default outcome on this road, the same subtraction mitochondria made in the piece on endosymbiosis and the eukaryotic revolution. So where did the thing come from? Three ideas are on the table: escape (self-copying stretches of DNA or RNA inside cells that happened to acquire a coat), reduction (once free-living cells that parasitised their way down to instructions only), and priority (self-copying molecules older than cells, of which viruses are the branch never absorbed). This is not a multiple-choice question. Viruses very likely are not a group with one common ancestor, and different viruses took different roads. One spoiler: in 2003 researchers identified "giant viruses" in amoebae from cooling-tower water, with genomes above a million base pairs — larger than many parasitic bacteria, and relatives found later reach beyond two million. They even carry some protein-making parts, but no ribosome. Whether they are shrunken cells or viruses bloated by picking up genes is still contested. Another clue points to multiple origins: viruses with no detectable relationship often fold their coat proteins into the same shape, as if "build a shell" was independently hit upon a handful of times.
有一类病毒叫逆转录病毒，进了细胞会把自己的说明书写进宿主的 DNA 里，插在某个位置上，从此跟着宿主的 DNA 一起被复制。
插进普通细胞，它随这个人一起消失。但万一插进的是将来要变成卵或精子的那批细胞，它就跟着传给下一代，永远留在那条世系里。
人的 DNA 里约 8% 是这种残骸。绝大多数早被突变打坏，又被细胞压着不让出声。它们和「基因组内的冲突」那期讲的转座子不一样：那些是自家在基因组里跳来跳去的段落，这些原本能裹上壳、出细胞、去感染别的个体，是从外面进来的。
少数被拿去用了。哺乳动物的胎盘上有一层特别的组织：许多细胞融成一大片，共用一层膜，母体和胎儿之间那道界面才成立。让它们融起来的那个蛋白，来自逆转录病毒的外壳基因——病毒本来用它把自己的膜和细胞膜融到一起好进门，现在同一个蛋白用来把胎盘细胞彼此融起来。小鼠身上把对应的基因敲掉，胚胎在孕期中段就死了，该融的那层没融（2009 年）。
这不是病毒帮了谁的忙。同类的捕获在不同哺乳动物支系里各自发生过：灵长类、小鼠、兔、食肉类用的是来源不同的病毒基因。插入一直在发生，绝大多数没用、被压住；偶尔一处碰巧插在能被读到的地方，又碰巧带着一个会融膜的蛋白，就可能被留下——不是一次幸运，是一个不断被撞上的机会。
这件事顺带给了演化生物学一件硬工具。插入位置基本是随机的，可选的落点上亿个。同一段残骸出现在人和黑猩猩基因组的同一个位置上，最省事的解释只剩一个：插进去的时候，我们还是同一群祖先。两次独立感染刚好插到同一个字母上的概率，小到可以不算。反过来，年轻的那批插入只在一支里找得到——这也正是分家之后各自又被感染该有的样子。
EN One class of virus, the retroviruses, writes its instructions into the host's DNA on entry and is copied along with it ever after. In an ordinary cell that copy dies with the person. But if it lands in cells destined to become eggs or sperm, it travels to the next generation and stays in that lineage forever. About 8% of human DNA is this kind of wreckage. Almost all of it was long ago broken by mutation and is kept silenced. It differs from the transposons in the piece on conflict inside the genome: those are stretches that jump around within a genome, while these once wrapped themselves in a coat, left the cell and infected other individuals — they came from outside. A few got put to use. The mammalian placenta has a tissue where many cells fuse into one sheet sharing a single membrane; without it the interface between mother and fetus does not exist. The protein that fuses them comes from a retroviral coat gene — the virus used it to fuse its own membrane with a cell's to get in; now the same protein fuses placental cells to each other. Knock the corresponding gene out in mice and the embryos die mid-gestation with that layer unfused (2009). This is not a virus doing anyone a favour. Captures of this kind happened separately in different mammal lineages: primates, mice, rabbits and carnivores use virus-derived genes of different origins. Insertions happen all the time, nearly all of them useless and suppressed; occasionally one lands where it can be read and happens to carry a membrane-fusing protein, and it may be kept — not one piece of luck, but a chance that keeps being run into. The whole business also handed evolutionary biology a hard tool. Insertion sites are essentially random, with hundreds of millions of possible landing spots. The same piece of wreckage found at the very same position in the human and chimpanzee genomes leaves one economical explanation: when it went in, we were still one ancestral population. The odds of two independent infections hitting the identical letter are negligible. Conversely, the young insertions are found in one lineage only — exactly what being infected separately after the split should look like.
医院里常见这么一件事：一种细菌先对某种抗生素不再敏感，几个月后，另一种完全不同的细菌也不敏感了。第二种不是自己慢慢突变出来的——那段耐药基因是被递过去的。
递法有好几种：细菌之间能搭一根管子，把一小圈独立的 DNA 送过去；噬菌体（专门感染细菌的病毒）在打包时装错，会把上一个宿主的一段 DNA 带进下一个；细菌也能直接从水里捡起死掉同类散出的 DNA。
机制上只要一句话：一段 DNA 只要能进到新细胞里、并在那儿被复制，就扩散了，完全不必经过繁殖。后果不小。在细菌那一层，「生命之树」不太像树，更像一张网；问某个基因「属于哪个物种」经常没有意义。「适应与约束」那期说演化受历史包袱限制，这里是那条规则的一个缺口：一个新本事可以不靠自己一步步磨出来，而是整套搬进来。
动物和植物里这事罕见，但真有。所有检查过的栽培甘薯——两百九十多个品种——体内都带着一段农杆菌插进去的 DNA，而它的野生近亲没有；也就是说，人类开始种甘薯之前，它已经是一种天然的转基因作物了。蚜虫能自己造类胡萝卜素（让身体发红的那类色素），这套基因来自真菌，动物本来没有。
但这块最该学的是判定门槛。横向转移是最容易被污染伪造出来的一类结论：样品里混进一点别的生物，看上去就像一段外来基因。2001 年人类基因组第一篇论文写道，有两百多个基因是从细菌直接转进脊椎动物祖先的；同年就被人重算，剩下的少数用「别的支系各自把这些基因丢了」解释得一样好。今天的看法是：人基因组里直接来自细菌的横向转移，即使有也极少。这恰好反衬出上一节那类证据为什么硬——插入位点对上的不是「像」，是坐标，污染伪造不出来。
EN A familiar hospital story: one bacterial species stops responding to an antibiotic, and months later a completely different species stops responding too. The second one did not mutate its way there — the resistance gene was handed to it. There are several ways to hand it over. Bacteria can build a tube between them and pass across a small independent ring of DNA; a bacteriophage (a virus that infects bacteria) packaging itself wrongly carries a stretch of the last host's DNA into the next; bacteria can also pick up DNA shed by dead neighbours straight from the water. The mechanism takes one sentence: a piece of DNA that can get into a new cell and be copied there has spread, with no reproduction involved. The consequences are large. At the bacterial level the "tree of life" is less a tree than a net, and asking which species a gene belongs to is often meaningless. The piece on adaptation and constraint described evolution as hemmed in by historical baggage; this is a hole in that rule — a new capability need not be ground out step by step, it can be moved in whole. In animals and plants this is rare but real. Every cultivated sweet potato examined — over two hundred and ninety varieties — carries a stretch of DNA inserted by Agrobacterium, while its wild relatives do not: it was a naturally transgenic crop before anyone farmed it. Aphids make their own carotenoids (the pigments behind red bodies) using genes that came from fungi, which animals otherwise lack. The real lesson here, though, is the standard of proof. Horizontal transfer is the conclusion most easily faked by contamination: a trace of another organism in the sample looks exactly like a foreign gene. In 2001 the first human genome paper reported over two hundred genes transferred directly from bacteria into a vertebrate ancestor; within the same year others recounted, and the few that survived were explained just as well by "other lineages each lost these genes". Today's view is that direct bacteria-to-human transfer, if it happens at all, is very rare. Which is exactly why the evidence in the previous section is hard: matching insertion sites are not a resemblance but a coordinate, and contamination cannot forge a coordinate.