你读这行字用掉的能量，来自细胞里一堆米粒形状的小颗粒。奇怪的地方在于：它们里面有一份自己的 DNA，是环状的，跟你染色体上那种长条的不一样，反倒跟细菌的那份长得像。
更奇怪的是，有几类专门对付细菌的抗生素，会连带干扰这些小颗粒。那类药盯的是细菌造蛋白质的机器，而这些小颗粒里那台机器，也是细菌那一版——所以药分不清。
如果它只是你身体自己长出来的一个零件，这些痕迹一条都不该有。
一百多年前就有人在显微镜下注意到，植物细胞里那些绿色颗粒会像细菌一样自己一分为二，而不是被细胞从零件堆里现造出来。当时这只是个古怪的观察。真正让它变成一条能查的线索的，是后来陆续凑齐的几个痕迹。
第一是膜。你细胞里那些供能的小颗粒（线粒体）外面裹着两层膜，不是一层。一只细菌被另一个细胞整个包进去、又没被消化掉，剩下的正好是这个结构：里面那层是它自己的皮，外面那层是包它的那只手。第二是那份环状 DNA，前面说过了。第三是造蛋白质的机器——每个细胞里都有一批叫核糖体的小装置负责把氨基酸串成蛋白质，而这些小颗粒里那批核糖体不跟细胞其余部分同款，跟细菌的更近。第四，它不能从零件现造，只能由已有的一个分裂成两个。
最后一条是决定性的：等到能读序列了，人们把这份环状 DNA 拿去跟各种细菌一条条比对，线粒体稳稳落在一类叫 α-变形菌的细菌里面，植物的绿色颗粒（叶绿体）落在蓝细菌里面。它们不是「像」细菌，它们就在细菌的族谱上，各自有确定的位置。
这里要小心一句话：不是谁「为了拿到能量」去吞了谁。碰巧吞进来、碰巧没消化掉、碰巧两边都还活着——这样的组合出现过无数次，绝大多数不了了之；只有极少数留下了更多后代，我们今天看到的就是那一支。「自然选择的逻辑」那期讲的三个条件，一条也没有被绕开。还有一点：你的线粒体几乎全部来自母亲，精子那边带的基本进不去或被清掉。这是一种把内部冲突摁下去的办法——两套来历不同的小颗粒混在一起，就会有专挑自己复制得快的变体冒出来。
EN Over a century ago microscopists noticed that the green granules in plant cells divide in two rather than being built from parts. Four traces turned that oddity into a case. Mitochondria are wrapped in two membranes, not one — exactly what is left when one cell engulfs a bacterium whole: the inner skin is the bacterium's, the outer one belongs to the hand that grabbed it. They carry their own circular DNA. Their ribosomes — the devices that string amino acids into protein — resemble bacterial ones rather than the cell's own, which is why some antibacterial drugs disturb them. And they can only arise by division of an existing one. The decisive evidence came with sequencing: mitochondrial DNA sits inside the alphaproteobacteria and chloroplast DNA inside the cyanobacteria. They are not bacteria-like; they have addresses on the bacterial tree. Nothing here was done "in order to" gain energy: engulfment without digestion happened countless times, almost always to no consequence, and the lineages that left more descendants are the ones still here. Mitochondria come almost entirely from the mother, which suppresses a conflict — mixing two lineages of organelles invites variants that simply copy themselves faster.
1967 年，一位当时署名 Lynn Sagan、后来以 Lynn Margulis 之名为人所知的研究者，发表了一篇把这套说法系统摆出来的长文。据后来的记述，这份稿子在见刊之前被十几家期刊退过，有的连读都没读完。
今天回看，退稿的人未必都在耍脾气。那个年代能拿出来的证据主要是形态：形状像细菌、分裂像细菌、膜有两层。而「像」永远可以有第二种解释——两层膜也可以是细胞自己的膜折进来包出来的。当时的手段判不了，双方吵的其实是同一堆图片。
局面是被一种新测量翻过来的。能读 DNA 序列之后，问题第一次变成可判定的：把线粒体的基因和一大堆细菌的基因摆在一起，看它落在族谱的哪根枝上——这个答案不依赖你觉得它「像」什么。一个悬了几十年的争论，不是靠哪一边把话说得更漂亮收尾的，是靠出现了一种能把两个说法分开的观察。
还有一件同样值得记住的事：她那篇文章里提了三条，线粒体和叶绿体这两条进了教科书，第三条——细胞表面那些会摆动的鞭毛也来自另一种细菌——至今没有得到支持，一般被认为不成立。同一个人、同一篇文章，一部分站得住一部分站不住，这是常态，不是例外。
那被吞的一方查清了，吞它的那一方是谁？现在的主流答案是古菌——一大类外表像细菌、内部机制却和我们更近的单细胞生物。2015 年，有人从深海沉积物里拼出了一类古菌的基因组，发现它们身上带着一批原本以为只有复杂细胞才有的基因。2020 年日本一组研究者真的把其中一株养了出来：从取样到拿到稳定培养花了十几年，它长得极慢，数量翻一倍要两周上下，还得靠把代谢废料交给搭伙的伙伴才活得下去。于是今天的图景是：复杂细胞不是和细菌、古菌并列的第三支，而是从古菌里面长出来的一支，外加一只细菌。这条也还在被继续检验。
EN In 1967 a researcher then publishing as Lynn Sagan — later known as Lynn Margulis — laid the case out in full. By later accounts the manuscript had been turned down by more than a dozen journals, some without a full reading. The rejections were not simply stubbornness: the available evidence was morphological, and looking like a bacterium always admits a second reading, since a cell's own infolded membranes could produce the same picture. The deadlock broke on a new measurement, not a better argument. Once DNA could be sequenced, the question became decidable — where the organelle's genes sit on the bacterial tree does not depend on what anything looks like. Worth remembering too: of the three claims in that paper, mitochondria and chloroplasts entered the textbooks, while her third — that the whip-like flagella also came from a bacterium — has found no support. As for the host, the current answer is the archaea, single-celled organisms that look bacterial but whose machinery is closer to ours. In 2015 genomes assembled from deep-sea sediment revealed archaea carrying genes thought exclusive to complex cells; in 2020 a Japanese group finally grew one, after more than a decade of work, on a doubling time of a couple of weeks and only in partnership with organisms that take its waste. Complex cells look less like a third branch than like a branch growing out of the archaea, plus one bacterium.
你线粒体里那份环状说明书，只剩 37 个基因。而它祖先那类细菌，基因数量是好几千。中间那几千个去哪了？一部分是真的丢了——住在别人细胞里，很多本事用不上了，坏掉也没人罚。但更值得注意的是另一部分：它们搬进了细胞核。今天在线粒体里干活的蛋白质，有一千多种是细胞核里的基因先造好，再运送进去的。
这个搬家不是推理出来的，有人直接量过。2003 年一组研究者在烟草上做了这件事：他们把一个只有落到细胞核里才起作用的抗性基因，塞进叶绿体的说明书里，然后大批种后代，看有多少株表现出「这个基因已经在核里了」。答案是每约一万六千粒花粉出现一次。这个频率高得出乎当时的预料——搬家不是远古发生完就停了的事，它现在还在进行。
留下的那几十个，管的多半是同一件事：盯着膜上那条产能的电子流，随时按状态调整。Allen 提出的解释是，这类调整必须就地立刻做完——把信号送到细胞核、等指令回来，黄花菜都凉了；所以基因得待在离现场最近的地方。这个解释有支持也有反对，还没定论。
更大的一场争论在能量上。Lane 和 Martin 2010 年提出：细菌产能靠的是自己那层外皮，可皮的面积跟不上体积的增长，于是它撑不大；复杂细胞把产能的膜挪进体内，配上成百上千个各自带着一小段说明书的小控制点，于是平均每个基因能分到的能量大了几个数量级（他们估的是二十万倍这个量级），大基因组这才养得起。2015 年 Lynch 和 Marinov 用另一套核算反驳说，这个差距根本不存在，真正卡住细菌的是群体和随机漂变，不是能量。两边至今没吵完。这条现在只能记成「未决」，不能当结论用。
EN The circular manual inside a human mitochondrion holds just 37 genes; its bacterial ancestors had several thousand. Some were simply lost — skills you no longer use inside someone else's cell. The more interesting fate is relocation into the cell nucleus: over a thousand proteins working inside mitochondria today are built from nuclear genes and shipped in. This was measured directly. In 2003 a group inserted into the tobacco chloroplast a resistance gene that only works once it lands in the nucleus, then screened the offspring: roughly one transfer per sixteen thousand pollen grains — far more often than expected, and still ongoing. Why do a few dozen genes stay put? Most of them govern the flow of electrons across the energy-producing membrane, and Allen's proposal is that such adjustments must be made on the spot: send the signal to the nucleus and wait, and it is too late. Supported by some evidence, disputed by other. A larger dispute is about energy. Lane and Martin argued in 2010 that a bacterium's surface cannot keep up with its volume, whereas a complex cell moves the energy membranes inside and adds hundreds of small control points each carrying a scrap of manual, so the energy available per gene rises by orders of magnitude — they put it around two hundred thousand fold — which is what makes a large genome affordable. In 2015 Lynch and Marinov ran a different accounting and concluded the gap does not exist, and that population size and drift, not energy, are what limit bacteria. No measurement both sides accept has yet separated them, so this one stays open rather than settled.