Yang Liu

← Primer

HSC · Healing Source Code

脾集落实验：干细胞第一次被数出来——以及 CFU-S 为什么不是今天的 LT-HSC The spleen colony assay: the first time a stem cell could be counted — and why CFU-S is not today's LT-HSC

2026-09-27 · CFU-S, 脾集落实验, Till-McCulloch, 自我更新, LT-HSC, 极限稀释

一条本来不值得注意的疙瘩

二十世纪五十年代末，多伦多 Ontario Cancer Institute 里有两个人在做一件当时看来相当乏味的事：测量小鼠骨髓细胞到底有多怕辐射。做法直接得近乎粗暴——把小鼠全身照射到骨髓几乎清空，再回输已知数量的正常骨髓细胞，看动物能不能活下来。这类救援实验在当时已经不新鲜。新鲜的是他们决定把死去动物的脾脏取出来仔细看。

脾脏表面长着一些肉眼可见的小疙瘩，像撒在表面的白色颗粒。切开看，里面是正在分裂的血细胞，而且不止一种：红系的、粒系的，有时还有巨核细胞。

1961 年发表在《Radiation Research》上的那篇论文，标题里根本没有「干细胞」三个字——它叫《正常小鼠骨髓细胞辐射敏感性的直接测量》。J. E. Till 与 E. A. McCulloch 本来想画的是一条剂量—存活曲线。让这篇论文成为现代血液学起点的，不是他们看见了什么，而是他们数了什么。

那条直线是论证，不是插图

核心观察是一个线性关系：脾脏上集落的数目，与注入的活骨髓细胞数成正比。按后来官方给出的量级，大约每注入一万个骨髓细胞，才能在脾上换来一个集落——这个东西在骨髓里稀有得离谱。

一条过原点的直线在这里不是漂亮的数据，而是一整个论证。如果一个集落必须由若干细胞协同才能长起来，集落数就该随注入量呈高次方上升；只有当「一个集落对应一个起始细胞」时，才会得到严格的正比。于是那个稀有成分被赋予了一个操作性的名字：脾集落形成单位（colony-forming unit–spleen，CFU-S）。命名方式本身值得留意——它不叫「干细胞」，它叫「能形成集落的那个单位」，只承诺已经被看到的功能。这个连载的第一篇谈的是造血干细胞的定义为什么至今仍由移植功能说了算；这里讲的是更早一步、也更素朴的一步：在争论如何定义之前，得先让这个东西变成一个可以被数出来的量。

同一条染色体裂痕

正比关系只是推论。真正把「一个集落来自一个细胞」钉死的，是 1963 年《Nature》上 A. J. Becker、McCulloch 与 Till 的那篇短文。他们用的标记古典得可爱：辐射本身会在染色体上留下随机的结构畸变，每一处畸变在形态上都近乎独一无二。如果一个集落里几乎所有正在分裂的细胞都带着同一处特征性畸变，这些细胞就只能有一个共同祖先。

结果正是如此。克隆性由此成立——而且是在流式细胞术、PCR、条形码测序统统不存在的年代，靠显微镜下一张张核型图完成的。

自我更新第一次被实验拿住

克隆性说明集落有单一起点，却没说这个起点能不能复制自己。同一年，L. Siminovitch 与 McCulloch、Till 在《Journal of Cellular and Comparative Physiology》上报告了一个更狠的实验：把长到第十天前后的单个脾集落挑出来打散，注入第二批致死照射的小鼠。第二批小鼠的脾上，又长出了集落。

集落里含有能再生集落的细胞。自我更新——干细胞定义里最难捉住的那一半——第一次以可计数的形式落到了实验台上。

同一篇论文还带出一个当时让人不适、后来却成为整个领域方法论底色的结果：每个集落里新生的集落形成细胞数目极不均匀。少数集落含有很多，多数含有很少。Till、McCulloch 与 Siminovitch 1964 年在《PNAS》上给出的解释是一个随机模型——每次分裂时是否自我更新，本身带有概率性，像放射性衰变那样：群体行为可以精确预测，单个细胞的命运不能。这是干细胞生物学的第一个定量理论。紧随其后的另一篇同样值得记住：骨髓细胞被连续系列移植下去，集落形成能力会逐代衰减。自我更新是真的，但不是无限的。

数出来的东西，未必是你想数的东西

此后近二十年，CFU-S 就是造血干细胞的操作定义。它的失宠不是因为实验做错了，而是因为分辨率提高之后，人们发现「脾上一个疙瘩」这个读数里混装着不同的东西。

第一道裂缝来自动力学。1982 年 M. C. Magli、N. N. Iscove 与 N. Odartchenko 在《Nature》上指出，早期出现的脾集落是短命的，会自行消退。同一只脾上第 8 天数到的集落，与第 12 天数到的集落，并不是同一批细胞的作品。稍早一些，G. S. Hodgson 与 T. R. Bradley 用 5-氟尿嘧啶处理骨髓后发现，存活下来的重建能力似乎位于 CFU-S 的上游，他们谨慎地称之为「pre-CFU-S」——一个问号式的命名，本身就说明当时的不安。

第二道裂缝是决定性的。1990 年 R. J. Jones、S. J. Sharkis 等人在《Nature》上用逆流离心淘洗（counterflow centrifugal elutriation）按大小和密度把骨髓分开：富含 CFU-S 的中间组分给出的是短暂植入；而持续的长期重建来自一个既不含 CFU-S、也不含粒-单集落形成细胞的、形态上像淋巴细胞的组分。他们的结论写得很直白——CFU-S 不是多能造血干细胞。早期植入与长期重建由两群不同的细胞承担，而且长期重建的那一群必须与一批能早期植入的细胞一起输进去，宿主才熬得过最初的骨髓空窗。

第三道裂缝把账彻底算清了。2002 年 T. Na Nakorn、D. Traver、I. L. Weissman 与 K. Akashi 在《Journal of Clinical Investigation》上，把纯化的下游祖细胞单独移植进致死照射小鼠：巨核/红系限制性祖细胞（MEP）里装着第 8 天 CFU-S 活性的绝大部分；第 12 天的 CFU-S 大约有一半仍落在 CMP/MEP 这些已经定向的群体里，另一半才在 HSC/MPP 区间。这些定向祖细胞足以让小鼠度过辐射致死期，但供者来源的细胞在四周左右就消失了。

换句话说，那条奠基性的直线数的主要不是干细胞，而是已经定向的祖细胞，尤其是红系方向的祖细胞。脾集落是造血重建的短期前哨，不是它的源头。

计数的标准迁移到了哪里

今天定义长期造血干细胞（long-term HSC，LT-HSC）的读数，已经不在脾脏表面，而在受体的外周血里，而且要等上几个月。核心是两件事：长期的多谱系重建，以及极限稀释下的定量。1990 年 S. J. Szilvassy 与 C. J. Eaves 等人在《PNAS》上把竞争性重建与极限稀释结合起来，定义了竞争性重建单位（competitive repopulating unit，CRU）——把细胞按梯度稀释注入，用有多少受体出现供者来源的多谱系造血，反推出这种细胞的频率。1991 年 L. G. Smith、I. L. Weissman 与 S. Heimfeld 在同一刊物上做了克隆层面的分析，把「一个细胞能供养几个谱系、供养多久」逼到单克隆分辨率。再加上二次移植这一关：只有能在第二个宿主体内重新撑起完整造血的细胞，才算真正通过。

值得说清的是，CFU-S 并没有被废除。作为一个几天就能读出结果、反映短期祖细胞数量的指标，它在今天仍然有用；只是没有人再把它当作干细胞的计数器。

能被数出来，才能被改造

回到那些疙瘩。它们的科学价值并不在于「证明了干细胞存在」——严格说，1961 年那篇论文什么也没有证明。它做的是一次转换：把一个此前只活在形态学猜想里的实体，变成一个有单位、有频率、可以画剂量曲线的量。之后的一切都建立在这次转换上：能数，才能比较；能比较，才能富集；能富集，才能问那个基因治疗真正关心的问题——我改了它之后，还剩多少个。

这条线一路延伸到今天的病床。一个接受造血干细胞基因治疗的患者，回输的剂量写成每公斤体重多少个 CD34+ 细胞；一份细胞产品的放行标准里写着平均载体拷贝数、写着编辑效率、写着集落形成能力。每一个数字都默认了一个前提：造血干细胞是可以被计数的。这个前提在 1961 年之前并不存在。

也正因如此，今天真正悬着的问题依然是计数的问题。在照射小鼠体内被数出来的干细胞，和一个没有被照射、没有被移植的人体里默默维持造血的那些细胞，是不是同一群、是不是同一个数量级？克隆条形码与体细胞突变谱系追踪正在尝试给出不依赖移植的答案，而这些答案与移植读数并不总是吻合。那意味着，六十多年之后，「一个干细胞」到底该怎么数，仍然是一个开放的问题——只是问法比脾脏上的疙瘩精致了许多。

参考文献

  1. Till JE, McCulloch EA. A direct measurement of the radiation sensitivity of normal mouse bone marrow cells. Radiat Res. 1961;14(2):213–222. DOI
  2. Becker AJ, McCulloch EA, Till JE. Cytological demonstration of the clonal nature of spleen colonies derived from transplanted mouse marrow cells. Nature. 1963;197:452–454. DOI
  3. Siminovitch L, McCulloch EA, Till JE. The distribution of colony-forming cells among spleen colonies. J Cell Comp Physiol. 1963;62(3):327–336. DOI
  4. Till JE, McCulloch EA, Siminovitch L. A stochastic model of stem cell proliferation, based on the growth of spleen colony-forming cells. Proc Natl Acad Sci USA. 1964;51:29–36. DOI
  5. Siminovitch L, Till JE, McCulloch EA. Decline in colony-forming ability of marrow cells subjected to serial transplantation into irradiated mice. J Cell Comp Physiol. 1964;64:23–31. DOI
  6. Wu AM, Till JE, Siminovitch L, McCulloch EA. Cytological evidence for a relationship between normal hematopoietic colony-forming cells and cells of the lymphoid system. J Exp Med. 1968;127(3):455–464. DOI
  7. Hodgson GS, Bradley TR. Properties of haematopoietic stem cells surviving 5-fluorouracil treatment: evidence for a pre-CFU-S cell? Nature. 1979;281:381–382. DOI
  8. Magli MC, Iscove NN, Odartchenko N. Transient nature of early haematopoietic spleen colonies. Nature. 1982;295:527–529. DOI
  9. Jones RJ, Wagner JE, Celano P, Zicha MS, Sharkis SJ. Separation of pluripotent haematopoietic stem cells from spleen colony-forming cells. Nature. 1990;347:188–189. DOI
  10. Szilvassy SJ, Humphries RK, Lansdorp PM, Eaves AC, Eaves CJ. Quantitative assay for totipotent reconstituting hematopoietic stem cells by a competitive repopulation strategy. Proc Natl Acad Sci USA. 1990;87(22):8736–8740. DOI
  11. Smith LG, Weissman IL, Heimfeld S. Clonal analysis of hematopoietic stem-cell differentiation in vivo. Proc Natl Acad Sci USA. 1991;88(7):2788–2792. DOI
  12. Na Nakorn T, Traver D, Weissman IL, Akashi K. Myeloerythroid-restricted progenitors are sufficient to confer radioprotection and provide the majority of day 8 CFU-S. J Clin Invest. 2002;109(12):1579–1585. DOI

A nodule nobody had bothered to look at

In the late 1950s, two researchers at the Ontario Cancer Institute in Toronto were doing something that looked, at the time, rather dull: measuring how sensitive mouse bone marrow cells are to radiation. The design was almost brutally simple. Irradiate a mouse until its marrow is effectively emptied, infuse a known number of normal marrow cells, and see whether the animal survives. Rescue experiments like this were already routine. What was not routine was the decision to take the spleens out and look at them closely.

The spleens carried nodules visible to the naked eye, like pale grains scattered across the surface. Cut one open and it was full of dividing blood cells — and not of one kind only. There were erythroid cells, granulocytic cells, sometimes megakaryocytes.

The 1961 paper in Radiation Research does not contain the phrase “stem cell” in its title. It is called A direct measurement of the radiation sensitivity of normal mouse bone marrow cells. J. E. Till and E. A. McCulloch had set out to draw a dose–survival curve. What turned the paper into a founding document of modern hematology was not what they saw. It was what they counted.

The straight line was the argument

The central observation was a linear relationship: the number of colonies on the spleen rose in direct proportion to the number of viable marrow cells injected. By the figure given in the official account of their later Lasker award, roughly ten thousand marrow cells had to go in for each nodule that came out — the responsible cell was absurdly rare.

A line through the origin, here, is not a pretty graph. It is an entire argument. If a colony required several cells acting together, colony counts would rise as a higher power of the inoculum; strict proportionality is what you get only when one colony corresponds to one founding cell. So the rare component was given an operational name: the colony-forming unit–spleen, or CFU-S. The naming deserves attention. It does not say “stem cell.” It says “the unit that forms a colony” — a promise limited to the function actually observed. The first piece in this series argued that the definition of a hematopoietic stem cell is still settled by transplantation rather than by surface markers. What happened in 1961 was one step earlier and plainer than that: before you can argue about how to define something, you have to make it into a quantity that can be counted.

One chromosome scar, shared

Proportionality was still an inference. What nailed down the claim that a colony descends from a single cell was a short 1963 paper in Nature by A. J. Becker, McCulloch and Till. Their marker was charmingly old-fashioned: irradiation itself leaves random structural aberrations on chromosomes, and each aberration is close to morphologically unique. If nearly every dividing cell in a colony carries the same distinctive aberration, those cells can only share a single ancestor.

That is what they found. Clonality was established — in an era with no flow cytometry, no PCR and no barcode sequencing, one karyotype at a time under a microscope.

Self-renewal, caught in an experiment

Clonality says a colony has a single origin. It says nothing about whether that origin can copy itself. In the same year, L. Siminovitch with McCulloch and Till reported a harder experiment in the Journal of Cellular and Comparative Physiology: pick a single spleen colony at around day ten, break it up, and inject the cells into a second cohort of lethally irradiated mice. Colonies grew in the second cohort too.

A colony contained cells capable of regenerating colonies. Self-renewal — the harder half of any stem cell definition — had for the first time been reduced to something countable.

The same paper carried a result that was uncomfortable at the time and later became the methodological temperament of the whole field: the number of new colony-forming cells per colony was wildly uneven. A few colonies held many; most held far fewer. The explanation Till, McCulloch and Siminovitch offered in PNAS in 1964 was a stochastic model — whether a cell self-renews at a given division is itself probabilistic, much as with radioactive decay: the behaviour of the population is precisely predictable, the fate of any single unit is not. This was the first quantitative theory in stem cell biology. A companion paper is worth remembering alongside it: put marrow through serial transplantation and colony-forming ability declines from round to round. Self-renewal is real, but it is not unlimited.

What you count may not be what you meant to count

For nearly two decades CFU-S simply was the operational definition of the hematopoietic stem cell. Its fall from that position was not caused by a flawed experiment. It was caused by rising resolution, which revealed that “a nodule on a spleen” was a readout bundling different things together.

The first crack was kinetic. In 1982, M. C. Magli, N. N. Iscove and N. Odartchenko reported in Nature that the earliest spleen colonies are transient — they regress. Colonies scored at day 8 and colonies scored at day 12 on the same spleen are not the work of the same cells. Slightly earlier, G. S. Hodgson and T. R. Bradley, treating marrow with 5-fluorouracil, found that the regenerative capacity that survived appeared to sit upstream of CFU-S; they called it, cautiously, a “pre-CFU-S” cell. A name with a question mark in it is itself a record of unease.

The second crack was decisive. In 1990, R. J. Jones, S. J. Sharkis and colleagues used counterflow centrifugal elutriation in Nature to separate marrow by size and density. The intermediate fraction, enriched for CFU-S, gave only transient engraftment; sustained long-term reconstitution came from a lymphocyte-like fraction devoid of both CFU-S and granulocyte–macrophage colony-forming cells. Their conclusion was blunt: CFU-S do not represent pluripotent hematopoietic stem cells. Early engraftment and durable reconstitution are carried by two different populations — and the durable one has to be transplanted together with a source of early-engrafting cells if the host is to survive the initial aplasia at all.

The third crack settled the accounts. In 2002, T. Na Nakorn, D. Traver, I. L. Weissman and K. Akashi reported in the Journal of Clinical Investigation what happens when purified downstream progenitors are transplanted alone into lethally irradiated mice. Megakaryocyte/erythrocyte-restricted progenitors carried the vast majority of day-8 CFU-S activity; roughly half of day-12 CFU-S activity still sat within the already-committed CMP/MEP compartment, with the other half in the HSC/MPP compartment. Those committed progenitors were enough to carry mice through the radiation-lethal window — yet donor-derived cells had disappeared by about four weeks.

Put plainly: the founding straight line was counting mostly committed progenitors, weighted toward the erythroid side, rather than stem cells. The spleen colony is a short-term outpost of hematopoietic recovery, not its source.

Where the counting moved

The readout that defines a long-term hematopoietic stem cell today is not on the surface of a spleen. It is in the recipient’s peripheral blood, and it takes months. Two things carry it: long-term multilineage reconstitution, and quantification under limiting dilution. In 1990, S. J. Szilvassy, C. J. Eaves and colleagues combined competitive repopulation with limiting dilution in PNAS to define the competitive repopulating unit — inject graded dilutions of cells, and infer the frequency of the responsible cell from the proportion of recipients showing donor-derived multilineage blood. In 1991, L. G. Smith, I. L. Weissman and S. Heimfeld published a clonal analysis in the same journal, pushing the question of how many lineages one cell can supply, and for how long, to single-clone resolution. Then there is the secondary transplant: only a cell that can rebuild full hematopoiesis in a second host is taken to have passed.

Worth saying clearly: the CFU-S assay was never abolished. As a readout that returns a number in a few days and reflects short-term progenitor content, it remains useful. It is simply no longer treated as a stem cell counter.

Nothing gets engineered until it can be counted

Back to those nodules. Their scientific value does not lie in having “proved that stem cells exist” — strictly speaking, the 1961 paper proved nothing of the kind. What it performed was a conversion. It turned an entity that had lived only in morphological conjecture into a quantity with a unit, a frequency and a dose–response curve. Everything afterward rests on that conversion: once you can count, you can compare; once you can compare, you can enrich; once you can enrich, you can ask the question gene therapy actually cares about — after I modify these cells, how many are left?

That line runs all the way to the bedside. A patient receiving hematopoietic stem cell gene therapy is dosed in CD34-positive cells per kilogram of body weight. The release criteria of a cell product specify a mean vector copy number, an editing efficiency, a colony-forming capacity. Every one of those numbers presupposes that hematopoietic stem cells are the sort of thing you can count. Before 1961, that presupposition did not exist.

Which is also why the question still hanging open is a question about counting. The stem cells enumerated inside an irradiated, transplanted mouse, and the cells quietly sustaining blood production in a person who has been neither irradiated nor transplanted — are they the same population, at the same order of magnitude? Clonal barcoding and somatic-mutation lineage tracing are trying to answer that without relying on transplantation, and the answers do not always agree with the transplant readout. Sixty-odd years on, how to count a single stem cell remains unsettled. Only the asking has grown more refined than nodules on a spleen.

References

  1. Till JE, McCulloch EA. A direct measurement of the radiation sensitivity of normal mouse bone marrow cells. Radiat Res. 1961;14(2):213–222. DOI
  2. Becker AJ, McCulloch EA, Till JE. Cytological demonstration of the clonal nature of spleen colonies derived from transplanted mouse marrow cells. Nature. 1963;197:452–454. DOI
  3. Siminovitch L, McCulloch EA, Till JE. The distribution of colony-forming cells among spleen colonies. J Cell Comp Physiol. 1963;62(3):327–336. DOI
  4. Till JE, McCulloch EA, Siminovitch L. A stochastic model of stem cell proliferation, based on the growth of spleen colony-forming cells. Proc Natl Acad Sci USA. 1964;51:29–36. DOI
  5. Siminovitch L, Till JE, McCulloch EA. Decline in colony-forming ability of marrow cells subjected to serial transplantation into irradiated mice. J Cell Comp Physiol. 1964;64:23–31. DOI
  6. Wu AM, Till JE, Siminovitch L, McCulloch EA. Cytological evidence for a relationship between normal hematopoietic colony-forming cells and cells of the lymphoid system. J Exp Med. 1968;127(3):455–464. DOI
  7. Hodgson GS, Bradley TR. Properties of haematopoietic stem cells surviving 5-fluorouracil treatment: evidence for a pre-CFU-S cell? Nature. 1979;281:381–382. DOI
  8. Magli MC, Iscove NN, Odartchenko N. Transient nature of early haematopoietic spleen colonies. Nature. 1982;295:527–529. DOI
  9. Jones RJ, Wagner JE, Celano P, Zicha MS, Sharkis SJ. Separation of pluripotent haematopoietic stem cells from spleen colony-forming cells. Nature. 1990;347:188–189. DOI
  10. Szilvassy SJ, Humphries RK, Lansdorp PM, Eaves AC, Eaves CJ. Quantitative assay for totipotent reconstituting hematopoietic stem cells by a competitive repopulation strategy. Proc Natl Acad Sci USA. 1990;87(22):8736–8740. DOI
  11. Smith LG, Weissman IL, Heimfeld S. Clonal analysis of hematopoietic stem-cell differentiation in vivo. Proc Natl Acad Sci USA. 1991;88(7):2788–2792. DOI
  12. Na Nakorn T, Traver D, Weissman IL, Akashi K. Myeloerythroid-restricted progenitors are sufficient to confer radioprotection and provide the majority of day 8 CFU-S. J Clin Invest. 2002;109(12):1579–1585. DOI