Yang Liu

← Primer

HSC · Healing Source Code

先导编辑——最灵活的"搜索替换",为何仍难写进长期造血干细胞 Prime editing — the most flexible "find-and-replace," and why it still struggles to write itself into long-term hematopoietic stem cells

2026-07-12 · prime-editing, 基因编辑, 造血干细胞

文本编辑器里有两个层次的修改。一个是自动更正:它只会把某几类特定的拼写错误换成对的,能力不大,却几乎不会误伤别处。另一个是”查找并替换”:你可以指定任意一段字符,把它换成任意另一段——插入、删除、替换,悉听尊便。基因编辑走到今天,恰好也分出了这样两层。碱基编辑是前者,稳、准、但只能改四类点突变;而它的近亲 prime editing,把”查找并替换”搬进了活细胞,理论上能改写人类已知致病变异中的绝大多数。可就是这样一件近乎万能的工具,当它想写进造血干细胞、并在骨髓里长久驻留下去时,却一再地卡住。

一件近乎万能的工具

要理解 prime editing 的野心,得先看清它前一代工具的边界。碱基编辑不切断 DNA 双链,只靠一个化学反应把单个碱基就地改写,安全性上是一次漂亮的让步;可它的化学决定了它只能做四类改写——C 变 T、G 变 A、A 变 G、T 变 C,全都属于所谓的 transition。基因组里还有另外八类点突变(transversion),以及大量以插入、删除形式存在的致病变异,碱基编辑一概够不着。镰刀型贫血就是个刺眼的例子:它的病根是 β-珠蛋白基因(HBB)里一个 A·T 到 T·A 的 transversion——恰好落在碱基编辑的能力之外。

2019 年,Anzalone 等人在 Nature 上给出了跨过这道边界的办法,并给它起名 prime editing。他们的构造精巧得像一件微缩机械:把一个被”敲掉一半刀刃”的 Cas9(nickase,只切一条链而非双链)和一段 reverse transcriptase 融合在一起,再配上一条特殊的向导 RNA——pegRNA。这条 pegRNA 身兼两职,既告诉编辑器去基因组的哪一处落脚,又在自己的尾巴上直接编码了”要改成什么”。到达目标后,nickase 只在一条链上划开一个口子,露出的末端就成了 reverse transcriptase 的引子;酶以 pegRNA 尾部为模板,把新的遗传信息直接”抄写”进 DNA。整个过程不制造 double-strand break,也不需要外源的供体模板。就在这篇论文里,他们用它在人类细胞中完成了一百七十多种编辑,涵盖定点插入、删除以及全部十二类点突变,并当场把镰刀型贫血那个 transversion 改了回来。一件工具,第一次同时握住了碱基编辑够不着的那八类点突变和成片的插入删除。

从细胞系走向治疗,先要跨过效率这道坎

新工具诞生时往往并不好用,prime editing 也一样。最初的版本效率参差,还常常在目标位点旁留下多余的小 indel。真正让它从”能做”走向”堪用”,是随后两年里两处并不起眼却关键的改良。

第一处在那条 pegRNA 自己身上。Nelson 等人在 Nature Biotechnology 上发现,pegRNA 尾部那段承载模板信息的序列很容易被细胞里的酶从 3’ 端啃噬降解,一旦被啃掉,编辑就失了准头。他们的对策是在尾巴末端接上一个结构化的 RNA 折叠,像给绳头打个结,挡住降解——这样改造出的 epegRNA,在多种人类细胞里把编辑效率提高了三到四倍,而脱靶并未随之升高。第二处则在细胞内部。2021 年,Chen 等人在 Cell 上追问:是什么在暗中拖慢 prime editing?答案指向了细胞自带的错配修复系统(mismatch repair)——它会把刚写进去的新序列当成”错误”抹掉。他们让编辑器短暂地压住这套修复,配上一个重新优化过的 PEmax 蛋白骨架,平均又把效率抬高了数倍,同时压低了那些多余的 indel。至此,prime editing 才真正具备了走向治疗场景的底气。

血液,是它最难攻的一块阵地

有了趁手的工具,顺理成章的下一步,是把它用到最需要基因编辑的地方之一——造血干细胞。2023 年,Everette 等人在 Nature Biomedical Engineering 上做了这件事:他们从镰刀型贫血患者体内取出造血干/祖细胞(HSPC),用 prime editing 直接把致病的 HBB 等位基因改回正常序列,校正频率达到 15% 到 41%。更要紧的是那场耐力测试——把编辑后的人类细胞移植进免疫缺陷小鼠,十七周之后,这些细胞的植入、分化和谱系成熟与健康供者的细胞相当,由它们分化出的红系细胞里,平均约四成表达正常的 β-珠蛋白,超过了公认的疗效门槛,红细胞在缺氧下也不再镰变;而在实验设计所能覆盖的上百个候选位点上,几乎检不到脱靶。这是迄今为止 prime editing 在造血干细胞里最完整的一次概念验证。

然而,把这项成果放回整条时间线上看,恰恰暴露了 prime editing 进入长期造血干细胞的真正难处。它至今仍停在体外编辑加小鼠移植的阶段,而它的近亲碱基编辑,已经在非人灵长类身上跑完了同类的耐力测试——2025 年,Radtke 等人在 Science Translational Medicine 上报告,用碱基编辑改写恒河猴自体造血干细胞的 β-珠蛋白位点,移植后在骨髓和外周血里稳定维持了两成以上的编辑,长达数月。两件本是同门的工具,一个已站到临床门口,一个还在动物阶段徘徊,差距从何而来?

答案藏在 prime editing 更复杂的身板里。它不是一把简单的剪刀,而是 nickase、reverse transcriptase 与那条身兼两职的 pegRNA 拼成的一整套装置——分子体积明显更大,零件更多,这让”怎么把它送进细胞”本身就成了难题。造血干细胞对外来物质本就挑剔,而 prime editing 这样庞大的载荷,无论用病毒还是非病毒方式递送,都比小巧的碱基编辑器更棘手;若指望有朝一日直接在体内(in vivo)编辑,把如此大的机器塞进递送载体的容量,更是横在面前的一堵墙。零件多也意味着要调的旋钮多:pegRNA 的模板长度、引子结合区、尾部结构,每一处都要为具体的目标序列量身设计,任何一环没配好,效率就会塌下来——这在增殖缓慢、数量稀少、经不起反复折腾的造血干细胞里被格外放大。而最根本的一问是植入竞争力:真正能终身供血的,是骨髓里那一小群长期造血干细胞;当这套庞大的编辑装置在它们体内运转过一遍之后,它们还能不能像未经改动的同类那样,在移植后的骨髓里站稳脚跟、长久地一代代传下去——目前的小鼠数据给出了鼓舞人心的迹象,但这恰恰是从动物迈向人体时最需要被反复夯实的一环。

灵活的代价

回头看,prime editing 留下的是一个耐人寻味的反差:它在纸面上几乎是最全能的基因编辑工具,能改写人类已知致病变异中的大多数;可越是全能,它的机器就越复杂,而越复杂,就越难被完好无损地送进那群最金贵、也最沉默的细胞里去。碱基编辑用能力的克制换来了递送与安全上的从容,率先摸到了造血干细胞治疗的门;prime editing 则手握更宽的疆域,却仍要在递送、效率与长期植入这三道坎上,一步步补齐它进入血液的资格。

它的下一程,几乎和碱基编辑当年走过的是同一条路:把庞大的装置拆小、装进能进入造血干细胞乃至直接进入人体的载体,把 pegRNA 的设计从手艺磨成通用的规则,再用足够长的观察确认被编辑过的长期造血干细胞不会在竞争中掉队。当这几步走完,那件在试管里几乎无所不能的”查找并替换”,才算真正学会了如何写进一个人的血液,并让这行改动,伴随他此后的一生。

参考文献

  1. Anzalone AV, et al. Search-and-replace genome editing without double-strand breaks or donor DNA. Nature. 2019;576(7785):149-157. DOI
  2. Nelson JW, et al. Engineered pegRNAs improve prime editing efficiency. Nat Biotechnol. 2022;40(3):402-410. DOI
  3. Chen PJ, et al. Enhanced prime editing systems by manipulating cellular determinants of editing outcomes. Cell. 2021;184(22):5635-5652.e29. DOI
  4. Everette KA, et al. Ex vivo prime editing of patient haematopoietic stem cells rescues sickle-cell disease phenotypes after engraftment in mice. Nat Biomed Eng. 2023;7(5):616-628. DOI
  5. Radtke S, et al. Engraftment and persistence of base-edited hematopoietic stem cells in nonhuman primates. Sci Transl Med. 2025;17(811):eadn2601. DOI

A text editor offers two levels of change. One is autocorrect: it swaps only certain classes of spelling error for the right ones — modest in power, yet it almost never damages anything elsewhere. The other is “find-and-replace”: you can specify any stretch of characters and swap it for any other — insert, delete, replace, as you please. Gene editing, as it stands today, has divided into exactly these two levels. Base editing is the former — steady, precise, but able to fix only four classes of point mutation; while its cousin, prime editing, brings “find-and-replace” into the living cell, in principle able to rewrite the vast majority of known pathogenic variants in humans. And yet this near-universal tool, when it tries to write itself into hematopoietic stem cells and reside for the long term in the marrow, keeps stalling.

A near-universal tool

To grasp prime editing’s ambition, you first have to see the limits of the tool that came before it. Base editing does not cut the DNA double helix; it rewrites a single base in place through a chemical reaction alone — a beautiful concession on the side of safety. But its chemistry dictates that it can perform only four classes of rewrite — C to T, G to A, A to G, T to C — all of them so-called transitions. The genome holds another eight classes of point mutation (transversions), plus a great many pathogenic variants that exist as insertions and deletions, and base editing can reach none of them. Sickle cell disease is a glaring example: its root cause is an A·T-to-T·A transversion in the β-globin gene (HBB) — falling precisely outside base editing’s reach.

In 2019, Anzalone and colleagues gave, in Nature, a way to cross this boundary, and named it prime editing. Their construct is as intricate as a piece of miniature machinery: they fused a Cas9 with “half its blade knocked out” (a nickase, cutting only one strand rather than both) to a reverse transcriptase, and paired it with a special guide RNA — the pegRNA. This pegRNA holds two jobs at once: it tells the editor where in the genome to land, and it encodes, right on its own tail, what to change it into. Once at the target, the nickase opens a cut on just one strand, and the exposed end becomes the primer for the reverse transcriptase; the enzyme, using the pegRNA’s tail as template, “transcribes” the new genetic information directly into the DNA. The whole process creates no double-strand break, and needs no exogenous donor template. In that very paper, they used it to carry out more than one hundred and seventy edits in human cells — spanning targeted insertions, deletions, and all twelve classes of point mutation — and, on the spot, corrected that sickle cell transversion back. For the first time, a single tool held both the eight classes of point mutation base editing could not reach and swathes of insertions and deletions.

From cell lines toward therapy, first cross the hurdle of efficiency

New tools are often not much good at birth, and prime editing was no exception. The earliest versions were uneven in efficiency, and frequently left extra small indels beside the target site. What truly moved it from “can do” to “usable” were two inconspicuous but crucial refinements over the following two years.

The first was in that pegRNA itself. Nelson and colleagues found, in Nature Biotechnology, that the stretch of the pegRNA tail carrying the template information is easily gnawed and degraded from the 3’ end by enzymes in the cell — and once it is chewed away, the edit loses its aim. Their remedy was to attach a structured RNA fold to the end of the tail, like tying a knot at the end of a rope, to block the degradation — and the epegRNA so engineered raised editing efficiency three- to fourfold across a range of human cells, without off-target effects rising along with it. The second was inside the cell. In 2021, Chen and colleagues asked, in Cell: what is quietly slowing prime editing down? The answer pointed to the cell’s own mismatch repair system — which treats the freshly written new sequence as an “error” and erases it. They had the editor briefly suppress this repair, paired it with a re-optimized PEmax protein backbone, and on average raised efficiency several fold again, while lowering those extra indels. Only at this point did prime editing truly have the footing to head toward therapeutic settings.

Blood is the hardest position it has to take

With a serviceable tool in hand, the natural next step is to apply it to one of the places most in need of gene editing — hematopoietic stem cells. In 2023, Everette and colleagues did exactly this, in Nature Biomedical Engineering: they took hematopoietic stem/progenitor cells (HSPCs) from sickle cell disease patients and used prime editing to change the pathogenic HBB allele directly back to the normal sequence, reaching correction frequencies of 15% to 41%. More important still was the endurance test — transplanting the edited human cells into immunodeficient mice; seventeen weeks later, these cells’ engraftment, differentiation, and lineage maturation were on par with cells from a healthy donor, the erythroid cells they gave rise to expressed normal β-globin at an average of roughly 40%, exceeding the recognized therapeutic threshold, and the red cells no longer sickled under hypoxia; while across the hundreds of candidate sites the experimental design could cover, off-target effects were all but undetectable. This is, to date, the most complete proof of concept for prime editing in hematopoietic stem cells.

And yet, placing this achievement back on the whole timeline is exactly what exposes the real difficulty of prime editing entering long-term hematopoietic stem cells. It still stops at the stage of ex vivo editing plus mouse transplantation, whereas its cousin base editing has already completed the same kind of endurance test in nonhuman primates — in 2025, Radtke and colleagues reported, in Science Translational Medicine, that using base editing to rewrite the β-globin locus in autologous rhesus macaque hematopoietic stem cells, the edits held steadily above 20% in the marrow and peripheral blood after transplantation, for months on end. Two tools from the same lineage — one already standing at the door of the clinic, the other still lingering at the animal stage — where does the gap come from?

The answer lies in prime editing’s more complex build. It is not a simple pair of scissors, but a whole apparatus assembled from a nickase, a reverse transcriptase, and that dual-duty pegRNA — markedly larger in molecular size, with more parts, which makes “how to get it into the cell” a problem in itself. Hematopoietic stem cells are already picky about foreign material, and a payload as bulky as prime editing’s is trickier to deliver — whether by viral or non-viral means — than the compact base editor; and if one hopes one day to edit directly in vivo, fitting a machine this large within the capacity of a delivery vehicle is a wall standing right in the way. More parts also means more knobs to tune: the pegRNA’s template length, primer binding site, and tail structure each have to be tailored to the specific target sequence, and if any one link is not set right, efficiency collapses — a problem sharply magnified in hematopoietic stem cells, which proliferate slowly, are few in number, and cannot withstand repeated handling. And the most fundamental question is engraftment competitiveness: the ones that can truly supply blood for a lifetime are that small population of long-term hematopoietic stem cells in the marrow; once this large editing apparatus has run through them once, can they still, like their unaltered counterparts, hold their ground in the transplanted marrow and pass themselves down generation after generation for the long haul — the current mouse data give encouraging signs, but this is precisely the link that most needs to be hammered home again and again in the move from animal to human.

The price of flexibility

Looking back, what prime editing leaves is an intriguing contrast: on paper it is very nearly the most all-around gene-editing tool, able to rewrite most of the known pathogenic variants in humans; yet the more all-around it is, the more complex its machinery, and the more complex it is, the harder it is to deliver intact into that population of the most precious, and most silent, cells. Base editing traded restraint in capability for ease in delivery and safety, and was first to reach the door of hematopoietic stem cell therapy; prime editing holds the broader territory, yet still has to earn its qualification to enter the blood, step by step, over the three hurdles of delivery, efficiency, and long-term engraftment.

Its next leg is almost the same road base editing once walked: break the bulky apparatus down small and pack it into a vehicle that can enter hematopoietic stem cells and even enter the human body directly; turn pegRNA design from a craft into general rules; and, with observation long enough, confirm that edited long-term hematopoietic stem cells will not fall behind in the competition. Once these steps are done, that “find-and-replace” that is nearly omnipotent in the test tube will have truly learned how to write itself into a person’s blood — and let that one line of change accompany them for the rest of their life.

References

  1. Anzalone AV, et al. Search-and-replace genome editing without double-strand breaks or donor DNA. Nature. 2019;576(7785):149-157. DOI
  2. Nelson JW, et al. Engineered pegRNAs improve prime editing efficiency. Nat Biotechnol. 2022;40(3):402-410. DOI
  3. Chen PJ, et al. Enhanced prime editing systems by manipulating cellular determinants of editing outcomes. Cell. 2021;184(22):5635-5652.e29. DOI
  4. Everette KA, et al. Ex vivo prime editing of patient haematopoietic stem cells rescues sickle-cell disease phenotypes after engraftment in mice. Nat Biomed Eng. 2023;7(5):616-628. DOI
  5. Radtke S, et al. Engraftment and persistence of base-edited hematopoietic stem cells in nonhuman primates. Sci Transl Med. 2025;17(811):eadn2601. DOI