Yang Liu

← Primer

HSC · Healing Source Code

定义造血干细胞——为什么说了算的至今仍是移植,而不是标记 Defining the hematopoietic stem cell—why transplantation, not markers, still has the final word

2026-07-16 · 造血干细胞, HSC, CD34, 移植, 干细胞定义

把一滴骨髓抹在玻片上,你看到的是一群大同小异的细胞。其中藏着那种能重建一个人全部血液的细胞,但没有任何一种染色、任何一台显微镜,能把它当场指认出来。要确认它,唯一的办法是把它放进一只已经失去造血能力的动物体内,等上几个月,看血液是否重新长了回来。也就是说,只有当它做完了那件事,你才知道它就是那个东西——而做完的那一刻,这个细胞早已在体内分裂、分化、散入血流,不可能再被拿回手里端详。这个尴尬的循环,是造血干细胞这门学问的起点,也是它至今没能彻底摆脱的东西。

争论从一场屠杀开始

二战结束后的那几年,辐射生物学突然有了紧迫的现实动机。人们知道足量的全身照射会杀死动物,而死因很大程度上是造血系统的崩溃:血细胞是一群寿命以天计的细胞,一旦补充停止,机体就在几周内失去携氧、止血与抗感染的能力。

真正的谜题出现在 1949 年。Jacobson 与 Marks 等人在 Proceedings of the Society for Experimental Biology and Medicine 上报告,如果在照射时用铅把小鼠的脾遮住,让这一小块组织免于辐射,动物就能活下来。两年后,Lorenz 与同事在 Journal of the National Cancer Institute 上发现,不遮挡也行——照射之后给动物注射骨髓细胞,同样能把它们从死亡里拉回来,小鼠和豚鼠都是如此。

现象足够清楚,解释却分成两派。一派认为,被保护的脾或者被输入的骨髓释放出某种可扩散的物质——一种”体液因子”,它顺着血流去唤醒宿主残存的造血组织,让机体自己修好自己。另一派认为不必这么绕:救命的就是细胞本身,它们迁移、定居、繁殖,把死掉的造血系统整个换掉。

这不是措辞之争。如果是因子,那么治疗的方向应当是分离并合成这种物质;如果是细胞,那么就存在某种能重建整个造血系统的细胞实体——而它可以被寻找、被计数、有朝一日被改造。

分辨这两种可能,需要的不是更巧的生理学,而是一个标记。1956 年,Ford、Hamerton、Barnes 与 Loutit 在 Nature 上用当时刚刚趁手的细胞遗传学给出了答案:他们照射受体、输入染色体核型可辨的供体细胞,然后去看恢复后的骨髓里,分裂中的细胞究竟带着谁的染色体。答案是供体的。恢复的造血不是宿主被劝醒的,而是外来细胞长出来的。他们把这种动物称为”辐射嵌合体”——一个身体,两套基因组。体液因子的解释就此让位。

把它数出来

细胞说立住之后,一个更锋利的问题接着浮现:注射进去的骨髓里,到底是每一个细胞都能救命,还是只有极少数能?这是个计数问题,而当时没人知道该数什么。

1961 年,多伦多的 James Till 与 Ernest McCulloch 在 Radiation Research 上报告了一个意外发现。他们本意是量化骨髓细胞的辐射敏感性,方法是给致死剂量照射过的小鼠回输不同数量的骨髓细胞。做着做着,他们注意到受体的脾表面在十天左右会鼓起一粒粒肉眼可见的结节。更关键的是那个定量关系:结节的数目与注入的细胞数成正比。

这条线性关系的含义相当重。它意味着每一个结节都源自注入群体里的某一个细胞,于是结节数就成了那类稀有细胞的读数——按他们当时的直接读数,每注入约一万个骨髓细胞,才在脾上长出一个结节。这种细胞后来被命名为 CFU-S(colony-forming unit–spleen)。血液学第一次有了一把可以计数的尺子,而尺子的刻度,是”能不能在体内长出一团东西”。

不过”成正比”只是间接推理。一个集落真的出自一个细胞吗?1963 年,Becker 与 McCulloch、Till 在 Nature 上用一个漂亮的设计把这个洞补上:他们先用较低剂量的辐射在供体细胞里随机制造染色体畸变,给每个细胞打上一枚各不相同的、可以在显微镜下认出的”随机指纹”,再把这些细胞移进受体。结果是,同一个脾集落里的细胞携带同一枚特殊的畸变标记,不同集落之间则不同。集落是克隆——一个细胞的全部后代。

同年稍晚,Siminovitch 与 McCulloch、Till 在 Journal of Cellular and Comparative Physiology 上把最后一块拼上。他们把长出来的脾集落一个个切下来,分别打散,再移进新的照射小鼠。新的集落又长了出来。这意味着最初那个细胞在制造大量分化后代的同时,还复制出了与自己同类的细胞——self-renewal 第一次被实验证明,而不再只是一个名词。同一批数据里还藏着一个当时令人不安的观察:各个集落所含的新集落形成细胞数量差别极大,有的极多,有的寥寥。同样的起点,给出天差地别的结果;干细胞的自我更新看上去不像执行指令,更像是掷骰子。

剩下的是多能性。1967 年到 1968 年,Wu 与 Till、Siminovitch、McCulloch 用同样的染色体标记思路追问:一个克隆里长出来的,是不是不止一种血细胞?先是在 Journal of Cell Physiology 上确认单个克隆能同时给出多种髓系细胞,随后在 Journal of Experimental Medicine 上,他们在胸腺和淋巴结里也找到了与造血集落形成细胞属于同一克隆的细胞。不过 Wu 等人自己把话只说到这里:要么淋巴组织里那些细胞的前体就是造血集落形成细胞本身,要么两者同出于一个尚未找到的更早祖先——他们无法在两者之间做出区分,因此只写道:“这些结果与造血系统的细胞和免疫系统的细胞可能来自同一种干细胞的看法相容。”

于是,不到十年间,造血干细胞被三条实验性质完全一样的证据定义了下来:它能克隆性地增殖、能自我更新、能给出多种谱系。请注意这个定义的形式——它通篇由动词构成。它没有说干细胞长什么样、表达什么、在哪里,它只说这个细胞能做成什么事。这是一个纯粹的功能定义,而它唯一的检验场,是一只被清空了造血系统的动物。

尺子本身也需要被校准

功能定义的第一次自我修正,来自它自己。CFU-S 曾经就是”干细胞”的同义词,可十天后长在脾上的那团东西,真的出自那种能维持一生造血的细胞吗?1990 年,Jones 与 Sharkis 等人在 Nature 上把这两件事分开了:通过物理分选,他们把形成脾集落的能力和长期重建造血的能力落到了不同的细胞群上。数得最欢的那种细胞,并不是最终救命的那一种。

这是个值得停一下的教训。CFU-S 是一个真实、可重复、可定量的读数,它推动了整个领域;但它测的是”十天里长出一团后代”的能力,而人们真正想知道的是”几十年里持续供血”的能力。两者高度相关,却不相等。一个检验方法从来只回答它实际问出的那个问题,而不是我们心里想问的那个。功能定义因此收紧了:今天所谓的金标准,不是长出集落,而是 long-term multilineage engraftment——移植后长期、稳定、多谱系地重建,并且最好能在二次移植的受体里再来一遍。

想在事前就认出它

功能定义有一个无法回避的代价:它是回溯性的,而且是毁灭性的。要证明一个细胞是干细胞,你必须用掉它。你无法拿着一管确认无疑的干细胞去做别的事——一旦确认,它已经不在了。对于任何想要富集、研究、乃至改造这种细胞的人来说,这是个死结。

出路只有一条:找到一些在细胞还活着的时候就能读出的特征,让它们替功能”提前发言”。1984 年,Civin 与同事在 Journal of Immunology 上报告了一株识别人造血祖细胞表面抗原的单克隆抗体,那个抗原当时叫 My-10,后来定名为 CD34。这是造血干细胞第一次有了可供操作的表面把手:它能被抗体抓住,于是细胞可以被分选、被富集、被拿去做事。1992 年,Baum、Weissman 等人在 PNAS 上用 CD34 与 Thy-1 的组合从人胎儿骨髓里圈出了一个候选干细胞群体。小鼠这边,1988 年 Spangrude、Heimfeld 与 Weissman 在 Science 上给出了同类的方案:去掉表达成熟谱系标记的细胞,再按 Thy-1 与 Sca-1 挑选,得到的那一小群细胞,重建能力比未分选的骨髓高出若干个数量级。

必须说清楚的是,这些标记从来不是靠”看起来像干细胞”选出来的。它们的合法性,恰恰来自功能检验:分选出来的细胞被移进动物体内,重建了造血,标记才因此获得意义。表型从诞生的第一天起,就是功能的代理人。

代理人用久了,容易被当成本人。CD34 最典型:它高效、临床可用,几乎所有自体造血干细胞采集与基因治疗流程,今天都以 CD34⁺ 作为起始细胞群。但 CD34⁺ 从来不是一群干细胞——它是一个混杂的群体,里面绝大多数是各类祖细胞,真正能长期重建的细胞只占极小的一部分。

更麻烦的是反方向的例子。1996 年,Osawa 与 Nakauchi 等人在 Science 上报告,在小鼠体内,单个 CD34 低表达或不表达的细胞,就足以长期重建淋巴与髓系造血。1998 年,Bhatia 与 Dick 等人在 Nature Medicine 上在人的细胞里找到了对应的现象:一类 CD34 阴性的细胞,同样具备在免疫缺陷小鼠体内重建人造血的能力。那个被当作身份证用了十几年的分子,原来既会把不是干细胞的放进来,也会把是干细胞的挡在外面。

标记随后越做越细。2005 年,Kiel 与 Morrison 等人在 Cell 上引入 SLAM 家族受体(CD150、CD48 等),用一组简洁的组合大幅提纯了小鼠的干细胞群,并因为这些标记同时可用于组织切片染色,第一次让人能就地看见干细胞待在什么样的微环境里。2011 年,Notta 与 Dick 等人在 Science 上把 CD49f 加进人的方案,做到了单个细胞移植后的长期多谱系重建——这是人类细胞层面最接近”纯”的一次。

然而纯度提高,并没有取消那道裂缝。2010 年,Morita、Ema 与 Nakauchi 在 Journal of Experimental Medicine 上做了一件很不客气的事:他们在当时最严格的表型定义之内,逐个移植单细胞,发现这群”同一种细胞”内部依然分层——有的能长期多谱系重建,有的偏向某些谱系,有的很快就熄火了。表型把纯度一步步推高,却始终没能推到”每一个都是”——即便是当时最好的标记组合,提高的也只是命中的概率。它给的是概率,不是身份。

尺子在被检验的东西上留了指纹

到这里,故事本可以收尾:表型是好用的近似,功能是最终的裁判。但过去十年,裁判本身被送上了被告席。

移植是一个极端场景。受体被照射或化疗清空,骨髓龛位空出来,幸存的细胞在剧烈的应激里被迫全速增殖——这套条件跟一个人平静地活着、骨髓按部就班地供血,几乎没有共同之处。那么,“能在这种废墟里重建造血”所测量的,究竟是这个细胞在体内的日常职责,还是它在灾难中被征召时的应激能力?

回答这个问题需要一种不移植也能追踪细胞谱系的方法,而这在近十几年才成为可能。2014 年,Sun 与 Camargo 等人在 Nature 上用可诱导的转座子在小鼠体内给细胞随机打上遗传条形码,然后在完全不受扰动的动物里读它们的去向。2015 年,Busch 与 Rodewald 等人在 Nature 上用另一套原位标记方案做了平行的追问。两项工作指向同一个方向:在稳态造血中,承担日常供血的是大量克隆各自贡献一点点,而干细胞对成熟血细胞的补充,比移植实验描绘的要缓慢和迂回得多——很大一部分工作是由干细胞下游那些寿命颇长的祖细胞完成的。2018 年,Rodriguez-Fraticelli 与 Camargo 等人在 Nature 上进一步把克隆命运拆开,给出了一张与教科书很不一样的原位路线图:巨核系基本独立于其他血液命运而出现;多能祖细胞群体内部存在一个由单谱系与寡谱系克隆构成的功能层级;而按传统定义的长期造血干细胞,竟是巨核系限制性祖细胞的一个重要来源——也就是说,巨核系很可能才是长期造血干细胞在原位的主要命运。

这些结果并没有推翻移植实验的任何一条结论——移植是真的,重建是真的,自我更新也是真的。它们改变的是这些结论的适用范围:金标准测出来的是一种在应激下被激发的潜能(potential),而不是这个细胞在完好的身体里实际做的事(fate)。潜能与命运是两回事,而所有基于移植的定义,测的都是前者。Rodriguez-Fraticelli 等人把这层意思直接写进了摘要:“鉴于绝大多数针对谱系产出的研究都是在造血移植的语境下做的,现有的谱系分支模型更可能代表的是谱系潜能的路线图,而不是天然状态下的命运。“

于是今天怎么办

这场百年争论并没有分出胜负,它的结局是一种分工。

表型是所有事情得以发生的前提。没有 CD34,就没有采集、没有体外编辑、没有回输;没有 SLAM 和 CD49f,就没有可以拿去做单细胞测序、拿去筛药、拿去成像的干细胞群。前瞻性分选把一群不可见的细胞变成了可操作的材料,这是整个领域能够工程化的基础。

而功能仍是最终的裁定。任何一个新标记、新培养体系、新编辑方案,最后都要回到同一个问题:这些细胞移进去以后,长期、多谱系地重建了吗?再移一次呢?这套检验之所以不能被取代,不是因为它优雅——它昂贵、缓慢、粗暴、还带着自己的偏倚——而是因为迄今没有任何分子读数能替它作答。

与此同时,那张画了几十年的层级图正在被重画。2017 年,Velten 与 Haas 等人在 Nature Cell Biology 上以单细胞数据描绘人的造血,看到的不是一级级离散的台阶,而是一片连续的分化景观,细胞的谱系倾向在其中逐渐显形而非骤然跳变。2018 年,Laurenti 与 Göttgens 在 Nature 上把这条线索总结成领域的新共识:从层级到景观。这场重画同样是定义之争的余波——如果连”分几层”都由测量方式决定,那么”哪一层是干细胞”就更不可能是一个纯粹的分子问题。

这件事对基因治疗尤其不抽象。一份 CD34⁺ 产品报出的编辑效率,是整个混杂群体的平均数;而决定一个患者十年后血里还有没有治疗性细胞的,是那极小一撮真正长期重建的细胞被改成了什么样。这两个数字可以相差甚远,而只有前者能在放行前测出来。产品的效力(potency)该怎么定义、平均编辑率为什么会骗人,是这个连载后面要专门拆的题目;但那些难题的根都在这里——我们至今只能对着代理人下手,却要为本人的行为负责。

一个仍然是动词的名词

回到玻片上那滴骨髓。七十年过去,我们已经能把里面的稀有细胞富集到相当高的纯度,能读出它的转录组,能在它的基因组上做单碱基的改写,能把它做成价值百万美元的药。唯独有一件事没有变:要确认手里这个细胞是不是造血干细胞,仍然得把它放进一个身体里,然后等。

这不是技术的失败,而是这个概念本身的性质。造血干细胞不是一种物质,而是一种能力——一个细胞在被需要的时候,能不能把整个系统重新长出来。能力只在被行使的时候存在,所以它注定只能被事后证实。标记、图谱、算法能做的,是把这个赌注的胜算一点点推高;而扣动扳机的那一刻,答案永远在细胞那边,不在我们手上。

参考文献

  1. Jacobson LO, Marks EK, et al. The role of the spleen in radiation injury. Proc Soc Exp Biol Med. 1949;70(4):740-2. DOI
  2. Lorenz E, Uphoff D, Reid TR, Shelton E. Modification of irradiation injury in mice and guinea pigs by bone marrow injections. J Natl Cancer Inst. 1951;12(1):197-201. (PMID: 14874130)
  3. Ford CE, Hamerton JL, Barnes DW, Loutit JF. Cytological identification of radiation-chimaeras. Nature. 1956;177(4506):452-4. DOI
  4. Till JE, McCulloch EA. A direct measurement of the radiation sensitivity of normal mouse bone marrow cells. Radiat Res. 1961;14:213-22. (PMID: 13776896;1961 年原文无 DOI,2011 年经典重印见 DOI)
  5. Becker AJ, McCulloch EA, Till JE. Cytological demonstration of the clonal nature of spleen colonies derived from transplanted mouse marrow cells. Nature. 1963;197:452-4. DOI
  6. Siminovitch L, McCulloch EA, Till JE. The distribution of colony-forming cells among spleen colonies. J Cell Comp Physiol. 1963;62:327-36. DOI
  7. Wu AM, Till JE, Siminovitch L, McCulloch EA. A cytological study of the capacity for differentiation of normal hemopoietic colony-forming cells. J Cell Physiol. 1967;69(2):177-84. DOI
  8. Wu AM, Till JE, Siminovitch L, McCulloch EA. Cytological evidence for a relationship between normal hemotopoietic colony-forming cells and cells of the lymphoid system. J Exp Med. 1968;127(3):455-64. DOI
  9. Civin CI, Strauss LC, Brovall C, Fackler MJ, et al. Antigenic analysis of hematopoiesis. III. A hematopoietic progenitor cell surface antigen defined by a monoclonal antibody raised against KG-1a cells. J Immunol. 1984;133(1):157-65. (PMID: 6586833)
  10. Spangrude GJ, Heimfeld S, Weissman IL. Purification and characterization of mouse hematopoietic stem cells. Science. 1988;241(4861):58-62. DOI
  11. Jones RJ, Wagner JE, Celano P, Zicha MS, et al. Separation of pluripotent haematopoietic stem cells from spleen colony-forming cells. Nature. 1990;347(6289):188-9. DOI
  12. Baum CM, Weissman IL, Tsukamoto AS, Buckle AM, et al. Isolation of a candidate human hematopoietic stem-cell population. Proc Natl Acad Sci USA. 1992;89(7):2804-8. DOI
  13. Osawa M, Hanada K, Hamada H, Nakauchi H. Long-term lymphohematopoietic reconstitution by a single CD34-low/negative hematopoietic stem cell. Science. 1996;273(5272):242-5. DOI
  14. Bhatia M, Bonnet D, Murdoch B, Gan OI, Dick JE. A newly discovered class of human hematopoietic cells with SCID-repopulating activity. Nat Med. 1998;4(9):1038-45. DOI
  15. Kiel MJ, Yilmaz OH, Iwashita T, et al. SLAM family receptors distinguish hematopoietic stem and progenitor cells and reveal endothelial niches for stem cells. Cell. 2005;121(7):1109-21. DOI
  16. Morita Y, Ema H, Nakauchi H. Heterogeneity and hierarchy within the most primitive hematopoietic stem cell compartment. J Exp Med. 2010;207(6):1173-82. DOI
  17. Notta F, Doulatov S, Laurenti E, et al. Isolation of single human hematopoietic stem cells capable of long-term multilineage engraftment. Science. 2011;333(6039):218-21. DOI
  18. Sun J, Ramos A, Chapman B, et al. Clonal dynamics of native haematopoiesis. Nature. 2014;514(7522):322-7. DOI
  19. Busch K, Klapproth K, Barile M, et al. Fundamental properties of unperturbed haematopoiesis from stem cells in vivo. Nature. 2015;518(7540):542-6. DOI
  20. Velten L, Haas SF, Raffel S, et al. Human haematopoietic stem cell lineage commitment is a continuous process. Nat Cell Biol. 2017;19(4):271-281. DOI
  21. Rodriguez-Fraticelli AE, Wolock SL, Weinreb CS, et al. Clonal analysis of lineage fate in native haematopoiesis. Nature. 2018;553(7687):212-216. DOI
  22. Laurenti E, Göttgens B. From haematopoietic stem cells to complex differentiation landscapes. Nature. 2018;553(7689):418-426. DOI

Smear a drop of bone marrow across a slide and what you see is a crowd of cells that look much alike. Hidden among them is the kind that can rebuild a person’s entire blood system, yet no stain and no microscope can point it out on the spot. To confirm it, the only way is to put it into an animal that has lost its capacity to make blood, wait a few months, and see whether the blood grows back. Which is to say: only once it has done that thing do you know it was that thing—and at the moment it finishes, the cell has long since divided, differentiated, and scattered into the bloodstream inside that body, impossible to retrieve and examine. This awkward loop is where the science of the hematopoietic stem cell begins, and it is also what the field has yet to fully escape.

The argument begins with a massacre

In the years right after the Second World War, radiation biology suddenly acquired an urgent practical motive. It was known that a sufficient dose of total-body irradiation would kill an animal, and that the cause of death was in large part the collapse of the blood-forming system: blood cells are a population whose lifespans are measured in days, and once resupply stops, the body loses within weeks its ability to carry oxygen, to stop bleeding, and to fight infection.

The real puzzle appeared in 1949. Jacobson, Marks, and colleagues reported in Proceedings of the Society for Experimental Biology and Medicine that if the mouse spleen was shielded with lead during irradiation, sparing that small piece of tissue from the radiation, the animal would survive. Two years later, Lorenz and coworkers found in the Journal of the National Cancer Institute that shielding was not even necessary—injecting bone marrow cells into the animals after irradiation pulled them back from death just the same, in both mice and guinea pigs.

The phenomenon was clear enough; the explanation split into two camps. One held that the protected spleen, or the infused marrow, released some diffusible substance—a “humoral factor”—that traveled through the bloodstream to rouse the host’s surviving hematopoietic tissue, letting the body repair itself. The other held that no such detour was needed: what saved the animal was the cells themselves, which migrated, settled, and multiplied, replacing the dead blood-forming system outright.

This was not a quarrel over wording. If it was a factor, then the direction of therapy ought to be to isolate and synthesize that substance; if it was cells, then there existed some cellular entity capable of rebuilding an entire hematopoietic system—and it could be sought, counted, and one day engineered.

Telling these two possibilities apart required not cleverer physiology but a marker. In 1956, Ford, Hamerton, Barnes, and Loutit gave the answer in Nature, using the cytogenetics that had just become tractable: they irradiated recipients, infused donor cells with a distinguishable chromosomal karyotype, and then looked at whose chromosomes the dividing cells in the recovered marrow actually carried. The answer was the donor’s. The recovered hematopoiesis had not been coaxed out of the host; it had grown out of the foreign cells. They called such an animal a “radiation chimaera”—one body, two genomes. The humoral-factor explanation gave way.

Counting it

Once the cellular account stood, a sharper question followed: among the marrow cells injected, is it every cell that can save the animal, or only a very few? This is a counting problem, and at the time no one knew what to count.

In 1961, James Till and Ernest McCulloch in Toronto reported an accidental finding in Radiation Research. Their intent had been to quantify the radiosensitivity of bone marrow cells by infusing different numbers of them into mice given a lethal dose of irradiation. As they worked, they noticed that the surface of the recipient’s spleen would, at around ten days, swell into nodules visible to the naked eye. More crucial was the quantitative relationship: the number of nodules was proportional to the number of cells injected.

The implication of that linear relationship is weighty. It means each nodule arose from one particular cell in the injected population, and so the nodule count became a readout for that class of rare cell—by their direct readout at the time, roughly ten thousand marrow cells had to be injected to grow a single nodule on the spleen. This cell type was later named CFU-S (colony-forming unit–spleen). For the first time hematology had a ruler that could count, and the ruler’s graduations were “can it grow a mass of something inside a body.”

But “proportional” was only indirect inference. Does one colony really come from one cell? In 1963, Becker, McCulloch, and Till filled that hole in Nature with an elegant design: they first used a lower dose of radiation to create random chromosomal aberrations in donor cells, stamping each cell with a distinct “random fingerprint” recognizable under the microscope, and then transplanted these cells into recipients. The result was that cells within the same spleen colony carried the same particular aberration marker, while different colonies differed. A colony is a clone—the entire progeny of one cell.

Later that same year, Siminovitch, McCulloch, and Till fitted the last piece into place in the Journal of Cellular and Comparative Physiology. They excised the spleen colonies one by one, dissociated each, and transplanted them into fresh irradiated mice. New colonies grew. This meant that the original cell, while producing a large number of differentiated descendants, had also copied out cells of its own kind—self-renewal was demonstrated experimentally for the first time, rather than remaining merely a word. Buried in the same set of data was an observation that was unsettling at the time: the number of new colony-forming cells contained in each colony varied enormously, some very many, some very few. The same starting point gave wildly different outcomes; the self-renewal of stem cells looked less like following instructions than like rolling dice.

What remained was multipotency. From 1967 to 1968, Wu, Till, Siminovitch, and McCulloch pressed the question using the same chromosome-marking approach: does more than one kind of blood cell come out of a single clone? First, in the Journal of Cell Physiology, they confirmed that a single clone could give rise to multiple myeloid cell types at once; then, in the Journal of Experimental Medicine, they also found cells in the thymus and lymph nodes belonging to the same clone as the hematopoietic colony-forming cells. Wu and colleagues, however, took their own claim only this far: either the precursors of those cells in lymphoid tissue are the hematopoietic colony-forming cells themselves, or both descend from a still-undiscovered earlier ancestor—they could not distinguish between the two, and so wrote only that “[t]he results are compatible with the view that cells of the hematopoietic system and the immune system may be derived from the same stem cell.”

And so, in under a decade, the hematopoietic stem cell was defined by three lines of evidence, all of exactly the same experimental character: it can proliferate clonally, it can self-renew, and it can give rise to multiple lineages. Note the form of this definition—it is composed throughout of verbs. It does not say what a stem cell looks like, what it expresses, or where it is; it says only what the cell can get done. This is a purely functional definition, and its sole proving ground is an animal whose blood-forming system has been emptied out.

The ruler itself needs calibrating

The functional definition’s first self-correction came from within. CFU-S was once synonymous with “stem cell,” yet was that mass growing on the spleen after ten days really the product of the kind of cell that sustains hematopoiesis for a lifetime? In 1990, Jones, Sharkis, and colleagues pulled the two apart in Nature: by physical sorting, they landed the ability to form spleen colonies and the ability to reconstitute hematopoiesis long-term on different cell populations. The cell that had been counted most eagerly was not the one that ultimately saved the animal.

This is a lesson worth pausing on. CFU-S is a real, reproducible, quantifiable readout, and it drove the entire field; but what it measures is the capacity to “grow a mass of progeny in ten days,” while what people actually wanted to know was the capacity to “supply blood continuously for decades.” The two are highly correlated, but not equal. A test only ever answers the question it actually asked, not the one we had in mind. The functional definition therefore tightened: today’s so-called gold standard is not growing a colony but long-term multilineage engraftment—long-term, stable, multilineage reconstitution after transplantation, and preferably a repeat performance in a secondary recipient.

Trying to recognize it in advance

The functional definition carries an unavoidable cost: it is retrospective, and it is destructive. To prove a cell is a stem cell, you must use it up. You cannot hold a tube of unambiguously confirmed stem cells and go do something else with them—once confirmed, they are already gone. For anyone who wants to enrich, study, or even engineer such cells, this is a dead knot.

There is only one way out: find features that can be read while the cell is still alive, and let them speak for function in advance. In 1984, Civin and colleagues reported in the Journal of Immunology a monoclonal antibody recognizing a surface antigen on human hematopoietic progenitor cells; that antigen was then called My-10 and was later named CD34. For the first time the hematopoietic stem cell had a workable surface handle: it could be grabbed by an antibody, and so the cells could be sorted, enriched, and put to work. In 1992, Baum, Weissman, and colleagues used the combination of CD34 and Thy-1 in PNAS to circumscribe a candidate stem cell population from human fetal bone marrow. On the mouse side, in 1988 Spangrude, Heimfeld, and Weissman offered a counterpart in Science: remove the cells expressing mature lineage markers, then select on Thy-1 and Sca-1, and the small population you obtain has a reconstituting capacity several orders of magnitude above unsorted marrow.

It must be said plainly that these markers were never chosen because they “looked like stem cells.” Their legitimacy came precisely from the functional test: the sorted cells were transplanted into animals and reconstituted hematopoiesis, and only then did the markers acquire meaning. From its first day, the phenotype has been function’s proxy.

Use a proxy long enough and it starts getting mistaken for the principal. CD34 is the classic case: it is efficient and clinically usable, and nearly every autologous hematopoietic stem cell collection and gene therapy workflow today takes CD34⁺ as its starting population. But CD34⁺ has never been a population of stem cells—it is a mixed population, the vast majority of it progenitors of various kinds, with the cells genuinely capable of long-term reconstitution making up only a very small fraction.

More troublesome are the examples running the other way. In 1996, Osawa, Nakauchi, and colleagues reported in Science that in mice, a single cell expressing CD34 at a low level or not at all was sufficient to reconstitute lymphoid and myeloid hematopoiesis long-term. In 1998, Bhatia, Dick, and colleagues found the corresponding phenomenon in human cells in Nature Medicine: a class of CD34-negative cells with the same ability to reconstitute human hematopoiesis in immunodeficient mice. The molecule that had served as an ID card for more than a decade, it turned out, both let in cells that were not stem cells and shut out cells that were.

The markers grew finer from there. In 2005, Kiel, Morrison, and colleagues introduced the SLAM family receptors (CD150, CD48, and others) in Cell, using a concise combination to greatly purify the mouse stem cell population; and because these markers could also be used to stain tissue sections, it became possible for the first time to see in situ what kind of microenvironment stem cells reside in. In 2011, Notta, Dick, and colleagues added CD49f to the human scheme in Science, achieving long-term multilineage reconstitution after transplantation of a single cell—the closest anyone has come to “pure” at the level of human cells.

Yet rising purity did not close the rift. In 2010, Morita, Ema, and Nakauchi did something quite unaccommodating in the Journal of Experimental Medicine: within the strictest phenotypic definition then available, they transplanted single cells one at a time and found that this group of “one and the same cell type” was still stratified inside—some could reconstitute long-term and multilineage, some were biased toward certain lineages, and some burned out quickly. Phenotype pushed purity up step by step but never managed to push it to “every one of them”—even the best marker combination of the day raised only the probability of a hit. What it gives you is a probability, not an identity.

The ruler left fingerprints on the thing it measured

The story could have ended here: phenotype is a serviceable approximation, function is the final judge. But over the past decade, the judge itself has been put in the dock.

Transplantation is an extreme scenario. The recipient is emptied out by irradiation or chemotherapy, the marrow niches are vacated, and the surviving cells are forced to proliferate at full speed under violent stress—these conditions have almost nothing in common with a person living quietly while the marrow supplies blood in its ordinary routine. So what does “able to rebuild hematopoiesis out of such rubble” actually measure: the cell’s day job inside the body, or its stress capacity when conscripted into a catastrophe?

Answering this required a way to trace cell lineage without transplantation, and that has only become possible in the last decade or so. In 2014, Sun, Camargo, and colleagues used an inducible transposon in Nature to randomly stamp genetic barcodes onto cells in mice, then read where those cells went in entirely undisturbed animals. In 2015, Busch, Rodewald, and colleagues pressed the same question in parallel in Nature with a different in-situ labeling scheme. Both works pointed the same direction: in steady-state hematopoiesis, everyday blood supply is borne by a large number of clones each contributing a little, and the stem cells’ resupply of mature blood cells is far slower and more roundabout than the transplantation experiments had portrayed—a large part of the work is done by the fairly long-lived progenitors downstream of the stem cells. In 2018, Rodriguez-Fraticelli, Camargo, and colleagues went further in Nature, taking clonal fate apart and producing an in-situ roadmap quite unlike the textbook one: the megakaryocytic lineage emerges largely independently of other blood fates; within the multipotent progenitor population there exists a functional hierarchy made up of single-lineage and oligo-lineage clones; and long-term hematopoietic stem cells as traditionally defined turn out to be a major source of megakaryocyte-restricted progenitors—which is to say, the megakaryocytic lineage may well be the principal in-situ fate of long-term hematopoietic stem cells.

These results did not overturn any conclusion of the transplantation experiments—the transplant is real, the reconstitution is real, and the self-renewal is real too. What they changed is the scope over which those conclusions apply: what the gold standard measures is a potential provoked under stress, not what the cell actually does in an intact body (its fate). Potential and fate are two different things, and every transplantation-based definition measures the former. Rodriguez-Fraticelli and colleagues wrote this point straight into their abstract: “Given that the bulk of studies addressing lineage outcomes have been performed in the context of haematopoietic transplantation, current models of lineage branching are more likely to represent roadmaps of lineage potential than native fate.”

So what do we do today

This century-long argument has not produced a winner; its outcome is a division of labor.

Phenotype is the precondition for everything happening at all. Without CD34 there is no collection, no ex vivo editing, no reinfusion; without SLAM and CD49f there is no stem cell population to take to single-cell sequencing, to drug screening, to imaging. Prospective sorting turned an invisible population of cells into workable material, and that is the basis on which the whole field can be engineered.

And function remains the final adjudication. Every new marker, every new culture system, every new editing scheme comes back in the end to the same question: after these cells were transplanted, did they reconstitute long-term and multilineage? And in a second transplant? The reason this test cannot be replaced is not that it is elegant—it is expensive, slow, brutal, and carries a bias of its own—but that to date no molecular readout can answer in its place.

Meanwhile, the hierarchy diagram drawn over so many decades is being redrawn. In 2017, Velten, Haas, and colleagues used single-cell data in Nature Cell Biology to depict human hematopoiesis, and what they saw was not a set of discrete steps but a continuous differentiation landscape, within which a cell’s lineage bias takes shape gradually rather than jumping abruptly. In 2018, Laurenti and Göttgens summarized this thread in Nature as the field’s new consensus: from hierarchy to landscape. This redrawing is likewise an aftershock of the argument over the definition—if even “how many tiers there are” is determined by the mode of measurement, then “which tier is the stem cell” is all the less likely to be a purely molecular question.

For gene therapy, none of this is abstract. The editing efficiency reported for a CD34⁺ product is an average over the whole mixed population; but what determines whether a patient still has therapeutic cells in their blood ten years later is what was done to that tiny handful of cells that genuinely reconstitute long-term. These two numbers can differ greatly, and only the former can be measured before release. How a product’s potency should be defined, and why average editing rates can deceive, are topics this series will take apart later; but the root of those difficulties is right here—we can still only act upon the proxy, while being answerable for the conduct of the principal.

A noun that is still a verb

Return to that drop of marrow on the slide. Seventy years on, we can enrich the rare cells inside it to a fairly high purity, read out their transcriptome, rewrite single bases in their genome, and turn them into a million-dollar drug. Only one thing has not changed: to confirm whether the cell in your hand is a hematopoietic stem cell, you still have to put it into a body, and then wait.

This is not a failure of technology but the nature of the concept itself. A hematopoietic stem cell is not a substance but a capacity—whether a cell, when it is needed, can grow the whole system back. A capacity exists only while it is being exercised, and so it is destined to be confirmable only after the fact. What markers, atlases, and algorithms can do is push the odds of this bet up bit by bit; and at the moment the trigger is pulled, the answer always lies on the cell’s side, not in our hands.

References

  1. Jacobson LO, Marks EK, et al. The role of the spleen in radiation injury. Proc Soc Exp Biol Med. 1949;70(4):740-2. DOI
  2. Lorenz E, Uphoff D, Reid TR, Shelton E. Modification of irradiation injury in mice and guinea pigs by bone marrow injections. J Natl Cancer Inst. 1951;12(1):197-201. (PMID: 14874130)
  3. Ford CE, Hamerton JL, Barnes DW, Loutit JF. Cytological identification of radiation-chimaeras. Nature. 1956;177(4506):452-4. DOI
  4. Till JE, McCulloch EA. A direct measurement of the radiation sensitivity of normal mouse bone marrow cells. Radiat Res. 1961;14:213-22. (PMID: 13776896; the 1961 original has no DOI; for the 2011 classic reprint see DOI)
  5. Becker AJ, McCulloch EA, Till JE. Cytological demonstration of the clonal nature of spleen colonies derived from transplanted mouse marrow cells. Nature. 1963;197:452-4. DOI
  6. Siminovitch L, McCulloch EA, Till JE. The distribution of colony-forming cells among spleen colonies. J Cell Comp Physiol. 1963;62:327-36. DOI
  7. Wu AM, Till JE, Siminovitch L, McCulloch EA. A cytological study of the capacity for differentiation of normal hemopoietic colony-forming cells. J Cell Physiol. 1967;69(2):177-84. DOI
  8. Wu AM, Till JE, Siminovitch L, McCulloch EA. Cytological evidence for a relationship between normal hemotopoietic colony-forming cells and cells of the lymphoid system. J Exp Med. 1968;127(3):455-64. DOI
  9. Civin CI, Strauss LC, Brovall C, Fackler MJ, et al. Antigenic analysis of hematopoiesis. III. A hematopoietic progenitor cell surface antigen defined by a monoclonal antibody raised against KG-1a cells. J Immunol. 1984;133(1):157-65. (PMID: 6586833)
  10. Spangrude GJ, Heimfeld S, Weissman IL. Purification and characterization of mouse hematopoietic stem cells. Science. 1988;241(4861):58-62. DOI
  11. Jones RJ, Wagner JE, Celano P, Zicha MS, et al. Separation of pluripotent haematopoietic stem cells from spleen colony-forming cells. Nature. 1990;347(6289):188-9. DOI
  12. Baum CM, Weissman IL, Tsukamoto AS, Buckle AM, et al. Isolation of a candidate human hematopoietic stem-cell population. Proc Natl Acad Sci USA. 1992;89(7):2804-8. DOI
  13. Osawa M, Hanada K, Hamada H, Nakauchi H. Long-term lymphohematopoietic reconstitution by a single CD34-low/negative hematopoietic stem cell. Science. 1996;273(5272):242-5. DOI
  14. Bhatia M, Bonnet D, Murdoch B, Gan OI, Dick JE. A newly discovered class of human hematopoietic cells with SCID-repopulating activity. Nat Med. 1998;4(9):1038-45. DOI
  15. Kiel MJ, Yilmaz OH, Iwashita T, et al. SLAM family receptors distinguish hematopoietic stem and progenitor cells and reveal endothelial niches for stem cells. Cell. 2005;121(7):1109-21. DOI
  16. Morita Y, Ema H, Nakauchi H. Heterogeneity and hierarchy within the most primitive hematopoietic stem cell compartment. J Exp Med. 2010;207(6):1173-82. DOI
  17. Notta F, Doulatov S, Laurenti E, et al. Isolation of single human hematopoietic stem cells capable of long-term multilineage engraftment. Science. 2011;333(6039):218-21. DOI
  18. Sun J, Ramos A, Chapman B, et al. Clonal dynamics of native haematopoiesis. Nature. 2014;514(7522):322-7. DOI
  19. Busch K, Klapproth K, Barile M, et al. Fundamental properties of unperturbed haematopoiesis from stem cells in vivo. Nature. 2015;518(7540):542-6. DOI
  20. Velten L, Haas SF, Raffel S, et al. Human haematopoietic stem cell lineage commitment is a continuous process. Nat Cell Biol. 2017;19(4):271-281. DOI
  21. Rodriguez-Fraticelli AE, Wolock SL, Weinreb CS, et al. Clonal analysis of lineage fate in native haematopoiesis. Nature. 2018;553(7687):212-216. DOI
  22. Laurenti E, Göttgens B. From haematopoietic stem cells to complex differentiation landscapes. Nature. 2018;553(7689):418-426. DOI