玫瑰枝上那种绿色的小虫子，蚜虫，一辈子只吸植物的汁液。
按理说这东西养不活它。植物汁液糖多得发腻，却缺一批动物自己造不出来、只能吃进去的氨基酸——氨基酸是拼成蛋白质的零件，缺几种，身体就搭不起来。蚜虫应该慢慢饿死才对。
把蚜虫切开，会看到肚子里有几十个特别大的细胞，里面塞满了细菌。缺的那几种氨基酸，是这些细菌造的。
问题就在这儿：这批细菌住在别人身体里、吃着人家的东西，它凭什么不趁机多拿一点、少给一点？
这类细菌有个名字，叫布赫纳氏菌。它住在蚜虫专门腾出来的那几十个大细胞里，一辈子不出来。它传到下一代只有一条路：钻进母虫的卵。没有第二条。
代价看得见。它的全部 DNA 只剩六十四万个碱基对——碱基对就是 DNA 上的一个字母配对，可以当成一个字——大约五百八十个基因，只有近亲大肠杆菌的七分之一。丢掉的正好是「在外面单过」需要的那些：应付环境变化的、造那些蚜虫本来就能造的氨基酸的。留下的恰好是造蚜虫缺的那几种的。而蚜虫这边，自己造这几种氨基酸的本事也早没了。
两边现在谁离了谁都活不成。这不是最近才谈拢的：蚜虫的家谱和这种菌的家谱形状几乎重合——每一次蚜虫分成两支，它身上的菌也跟着分成两支。不是各家各自捡来的，是一起分的叉，至少一亿六千万年。
要紧的是别把这讲成一个道德故事。它不是「学会了忠诚」。跑的账很枯燥：它的拷贝只能坐在卵里出去，所以宿主多留一个后代，它就多一批拷贝；宿主死了，它这条线整条断。于是「多拿一点」的变异就算当场占到便宜，也同时把自己唯一的出口弄小了——这种变异留不下来。不是它选择了不作弊，是作弊的那些品系没传下来。传递通道把两边的繁殖成功绑成了同一个数字。
「内共生与真核革命」那期讲的线粒体，是这条路走到最尽头的样子：基因丢到只剩十几个，谁也不再把它当成一个独立生物了。
EN These bacteria have a name: Buchnera. They live inside a few dozen oversized cells the aphid sets aside for them and never leave. They have exactly one route to the next generation — getting into the mother's eggs. There is no second route. The cost is visible. Their entire DNA is down to about 640,000 base pairs (a base pair is one paired letter of DNA) and roughly 580 genes, about a seventh of what their relative E. coli carries. What they lost is precisely what "living on your own outside" requires: coping with a changing environment, making the amino acids the aphid can already make. What they kept is precisely the machinery for the ones the aphid lacks. And on the aphid's side, the ability to make those few amino acids is long gone too. Neither can now survive without the other, and this is not a recent arrangement: the aphid family tree and the bacterial family tree have almost the same shape — each time aphids split into two lineages, their bacteria split too. They did not each pick up a partner separately; they branched together, for at least 160 million years. The important thing is not to tell this as a morality tale. Nothing "learned loyalty". The accounting is dull: copies of the bacterium can only leave inside an egg, so every extra offspring the host raises means another batch of copies, and if the host dies that whole line ends. A "take a bit more" variant may win locally, but it shrinks its own only exit at the same time — so such variants do not persist. Nothing chose not to cheat; the cheating strains simply left no descendants. The transmission channel welds both sides' reproductive success into one number. Mitochondria, from the piece on endosymbiosis and the eukaryotic revolution, are what this road looks like at the very end: genes reduced to barely a dozen, and nobody counts them as a separate organism any more.
另一种菌叫沃尔巴克氏体。它感染节肢动物（昆虫、蜘蛛、甲壳类这一大群）的比例，不同调查给出的数从四成到七成都有——方法差别太大，但总之极多。
它也只随卵传，精子里挤不进去。于是雄性对它来说是死胡同。它身上出现了各种把宿主性别比往雌性掰的手段：杀掉受感染母亲产下的雄性胚胎；把遗传上本该是雄性的个体发育成雌性；让雌性不用交配就能产卵。
最常见的一手更绕。带菌的雄虫和不带菌的雌虫交配，胚胎发育不起来；带菌的雌虫跟谁配都没事。结果不是带菌的更能生，是不带菌的雌虫被相对惩罚了——菌因此在群体里一路铺开，对虫子本身没有任何好处。这和「基因组内的冲突」那期的减数分裂驱动跑的是同一条逻辑，只不过这回作弊的不是自家的一段染色体，是另一个物种。
可是同属的沃尔巴克氏体，在丝虫身上完全是另一回事。丝虫是引起象皮病和河盲症的那类寄生线虫，它体内这种菌是必需的：用抗生素把菌打掉，虫子就不育，寿命从十来年掉到两三年。这已经成了一种在用的临床手段——打菌，间接治那个寄生虫病。而近亲罗阿丝虫身上干脆没有这种菌。
轴的右半边还有一个反例。夏威夷短尾乌贼靠肚里的费氏弧菌发光，在月夜抹掉自己的影子。这批菌不随卵传，每代小乌贼都要从海水里重新招。没有卵可搭，怎么保证招进来的不揩油？靠筛：发光器的入口只放这一种菌进去，之后每天清晨排掉九成到九成五，剩下的重新长满。绑定和筛选是两条路，通向同一件事——让占便宜的那一支没有出口。
EN A second bacterium: Wolbachia. Estimates of how many arthropod species (insects, spiders, crustaceans) carry it range from forty to seventy percent depending on the survey — the methods differ too much to pin down, but either way it may be the most widespread bacterium on Earth. It too travels only in eggs; it cannot get into sperm. So males are a dead end for it, and it has acquired a whole repertoire for tilting host sex ratios female: killing the male embryos of infected mothers, developing genetic males into females, letting females lay without mating. The commonest trick is more roundabout. An infected male crossed with an uninfected female produces embryos that fail to develop, while an infected female is fine with any partner. The infected are not more fertile — the uninfected females are relatively penalised. So the bacterium spreads through the population while doing nothing good for any individual insect. This is the same logic as meiotic drive in the piece on intragenomic conflict, except the cheat is not a stretch of the host's own chromosome but another species. Yet the same genus of Wolbachia is something else entirely inside filarial nematodes, the parasitic worms behind elephantiasis and river blindness. There it is indispensable: clear it with antibiotics and the worms become sterile and their lifespan drops from around ten years to two or three. That is now a real clinical approach — hit the bacterium to treat the worm infection. And the close relative Loa loa simply has no such bacterium at all. There is a telling counter-case on the other side of the axis too. The Hawaiian bobtail squid uses Vibrio fischeri in its light organ to erase its own shadow on moonlit nights. These bacteria are not passed in eggs; every young squid has to recruit them from seawater afresh. With no egg to ride, what stops the recruits from skimming? Filtering: the light organ's entrance admits only this one species, and each dawn the squid vents ninety to ninety-five percent of them, leaving the rest to regrow. Binding and filtering are two different roads to the same thing — leaving no exit for the lineage that takes advantage.
先承认真实的那部分。在完全无菌的环境里养大的小鼠确实不正常：肠道附近的淋巴组织发育不全，胸腺偏小，而且吃同样多的东西反而更瘦，要维持同样体重得吃得更多。菌不是可有可无的房客。
于是有人提出一个很顺理成章的想法：把宿主和它身上全部的微生物当成一个东西看（有人管它叫「全生物」），两边的基因合起来算一份基因组，选择就作用在这个整体上。
反对意见抓的是一个很硬的点：一个东西要能被当成单位来结算，前提是这一整套可靠地、成套地一起传到下一代。蚜虫和它的菌满足——菌就在卵里，整套一起走。人不满足：你肠道里的大多数菌是从环境、食物、身边人那里陆续捡来的，一次抗生素、一趟旅行、换种主食就变掉一大块。组成不成套传递，「整体」就没有一条连续的世系可供选择去筛；被稳定筛的还是宿主的基因和每个菌各自的基因。有位研究者说得到位：把宿主和菌当成一个演化单位，该是需要被检验的假设，不是默认前提。
这块最该注意的是别被数字带着跑。流传最广的那句「体内细菌是人体细胞的十倍」是错的，2016 年重算下来大约一比一（细菌约三十八万亿，人体细胞约三十万亿）。一个放大十倍的数字流传几十年没人核，本身就说明这个领域多容易过热。
因果证据到哪一步了？确凿的是艰难梭菌反复感染——把健康人的肠道菌群整套移植过去，治愈率八成以上，远高于继续用抗生素。而「改菌群能减肥」这类说法，小鼠身上做得漂亮，人身上一直没复现出同样强度的因果，近几年综述的口径明显收紧了，这个还有争论。两者的差别正好落回本期的杠杆：移植管用，是因为一个被清空的位置被整套填上；而长期改造一套本来就没有传递通道的组合，改完自然漂回去。
EN Start by granting the real part. Mice raised in a completely germ-free environment genuinely are abnormal: the lymphoid tissue near the gut is underdeveloped, the thymus is small, and they are leaner on the same food, needing to eat more to hold the same weight. Microbes are not optional lodgers. So a very natural idea appeared: treat a host and all its microbes as one thing (some call it a holobiont), add the two sets of genes into a single genome, and let selection act on the whole. The objection lands on something hard: for anything to be settled as a unit, the whole package has to be passed to the next generation reliably and as a set. Aphids and their bacteria qualify — the bacteria are inside the egg and the set travels together. People do not: most of the bacteria in your gut are picked up piecemeal from the environment, from food, from the people around you, and one course of antibiotics, one trip, one change of staple food can swap out a large chunk. If the composition is not transmitted as a set, the "whole" has no continuous lineage for selection to sift; what is stably sifted is still the host's genes and each microbe's own genes. As one researcher put it, treating host plus microbes as an evolutionary unit should be the hypothesis under test, not the default assumption. The thing to watch here is not being carried along by numbers. The most-repeated line — that bacteria outnumber human cells ten to one — is wrong; a 2016 recalculation put it at roughly one to one (about thirty-eight trillion bacteria against thirty trillion human cells). That a tenfold-inflated figure circulated for decades unchecked tells you how easily this field overheats. How far does the causal evidence go? Solid for recurrent Clostridioides difficile infection: transplanting gut microbes from a healthy donor cures well over eighty percent of cases, far better than continuing antibiotics. Claims that changing the microbiome drives weight loss look beautiful in mice but have not reproduced at the same strength in humans, and recent reviews have visibly narrowed their language. That one is still contested. Interestingly, the difference between the two lands back on this piece's lever: a transplant works where a niche emptied by antibiotics gets refilled as a set, whereas permanently reshaping a combination that has no stable transmission channel to begin with will drift back.