Yang Liu

← Primer

HSC · Healing Source Code

表观编辑——不改一个字母,就把基因关掉 Epigenetic editing: switching a gene off without changing a single letter

2026-07-25 · 表观编辑, 造血干细胞, DNA-methylation, 基因编辑, CRISPRoff

你的肝细胞和淋巴细胞,基因组是同一套。它们之所以判若两物,不是因为谁的 DNA 里少了几段,而是因为哪些段落被朗读、哪些被合上封存。细胞很久以前就学会了怎样长久地关掉一个基因,却从来没有动过剪刀:它在 DNA 上盖一枚化学的印章,让这一页保持合拢,并在此后每一次分裂时把印章复印给两个女儿细胞。既然细胞自己有一套如此可靠的沉默机制,我们为什么非要把染色体切断?

细胞早就会的那件事

这枚印章的主角是 DNA methylation。基因起始区常聚集着大片 CpG 位点,当它们被密集甲基化、染色质又被压上抑制性的组蛋白标记,转录机器就读不进去了。关键在于可遗传性:DNA 复制时,新链上的甲基化会被维持性的甲基转移酶照着母链补齐——这不是一次性的压制,而是一份能被复印下去的记忆。基因编辑花了近十年,才把它变成一件指哪打哪的工具。

先学会”按住”

2013 年,Qi 等人在 Cell 上把 Cas9 的两个切割结构域都做了失活处理,得到一个只会认路、不会动刀的 dCas9;向导 RNA 把它领到一个基因上,单靠占位就能挡住 RNA 聚合酶。他们把这套系统称为 CRISPR interference(CRISPRi):在大肠杆菌里没有检测到脱靶效应,而且——这一点当时被写成优点——效果是可逆的。同年稍晚,Gilbert 等人在 Cell 上把 dCas9 与效应结构域融合,把抑制与激活带进了人类细胞。两年后,Thakore 等人在 Nat Methods 上把 dCas9-KRAB 引导到统辖多个珠蛋白基因的 HS2 enhancer 上:H3K9 三甲基化被高度特异地写在那里,多个珠蛋白基因随之沉默,而全局基因表达几乎没有波动。

但这一代工具有个共同的软肋:它们靠的是”一直有人按着”——KRAB 招募来的组蛋白修饰难以跨越细胞分裂,编辑器一旦稀释、降解,基因就慢慢弹回来。要成为治疗,必须让细胞在编辑器消失后继续记得。

让它记住

2016 年秋天,Cell 同一期背靠背发表的两篇论文补上了”记忆”这一层。Amabile 等人的做法后来被称为 hit-and-run:借用胚胎干细胞沉默内源逆转录病毒的那套机器,把一组改造过的转录抑制因子瞬时送进体细胞,让它们在同一位点协同写下抑制性组蛋白标记,并催出从头的 DNA methylation。全基因组分析显示沉默被牢牢限制在靶位点、没有蔓延到邻近基因,而唯一能解除它的是靶向的 DNA 去甲基化。同一期里,Liu 等人给出了另一半工具:把 Tet1 或 Dnmt3a 挂在失活的 Cas9 上,就能在指定位点擦掉或写上甲基化——他们用 dCas9-Tet1 在有丝分裂后的神经元里重新打开了 BDNF。

到 2021 年,Nuñez 等人把这些收拢成了一个分子:CRISPRoff 是 ZNF10 的 KRAB 结构域加上 Dnmt3A 与 Dnmt3L、一起挂在 dCas9 上的单一融合蛋白。瞬时表达之后,沉默维持了至少 50 天,论文的说法是它能被人类细胞维持超过 450 次细胞分裂;在一个内源基因上,超过九成的细胞失去了表达。全基因组筛选显示绝大多数蛋白编码基因都能被这样关掉,包括并不带经典 CpG 岛的那些;配套的 CRISPRon 则能把记忆抹掉。沉默还能穿过分化——干细胞被诱导成神经元的整个过程里,被关掉的基因一直关着。

一次瞬时给药,写下一个可遗传的状态,不产生 double-strand break,原则上还可以被撤销——表观编辑作为治疗手段的形状,至此清楚了。

走出培养皿

把它推向临床想象的,是 2024 年 Cappelluti 等人在 Nature 上的工作。他们把编码抑制因子的 mRNA 装进脂质纳米颗粒,靶向肝细胞里调控胆固醇代谢的 Pcsk9,给小鼠静脉注射一次:循环中的 PCSK9 迅速下降,随后稳定在对照组的四成左右,一直到 330 天实验终止。在另一批小鼠里,给药三个月后他们做了肝部分切除,逼迫肝细胞增殖再生——再生之后,沉默还在,甲基化也还在。优化后的全合一版本在 43 天的观察里把降幅推到约七成半,与直接用 CRISPR-Cas9 敲除相当,却没有制造 DNA 断裂。次年,Tremblay 等人在 Nat Med 上把同一个靶点推到更接近人的体系:食蟹猴单次给药后,循环 PCSK9 降低约九成、LDL 胆固醇降低约七成;而在小鼠体内,一个靶向的表观激活器可以把已沉默的位点重新打开。

递送这一侧则在同步缩小体积。Cappelluti 的团队在体外比过 dCas9、TALE 与锌指蛋白三类 DNA 结合平台,最后选定锌指,依据是它在这轮筛选里效率最好;作者另在讨论中指出,锌指分子体积小、且不依赖半衰期很短的向导 RNA,这些特点有利于递送与编辑复合物的稳定。同年 Neumann 等人在 Science 上报告的 CHARM 更进一步:它不再自带甲基转移酶,而是用一段组蛋白 H3 尾巴与 Dnmt3l 的融合去招募并激活细胞内源的甲基转移酶,由此缩小了转基因尺寸、也降低了细胞毒性,并通过 AAV 全身给药在小鼠全脑范围内沉默了朊蛋白基因。

为什么骨髓是它最该去的地方

造血干细胞是最不能出错的细胞:它要在体内工作一辈子,任何一次错误的改动都会被自我更新放大成一整个克隆。而 double-strand break 带来的 p53 反应、大片段缺失与染色体重排,在这类细胞里分量格外重(留到讲安全性时细说)。表观编辑不切断 DNA,原则上绕开了这一整类风险。

巧的是,血液病里最成功的那个靶点本来就是”关掉就好”的靶点。BCL11A 的红系 enhancer 被 CRISPR 切断之后胎儿血红蛋白重新回升,这正是已获批上市的疗法所依据的机制。2026 年 Wang 等人在 Blood 上给出的解释很耐人寻味:切断这段 enhancer 之所以奏效,是因为它破坏了维持 BCL11A 表达所必需的染色质绝缘结构,最终导致的正是 BCL11A 自身的表观沉默;他们还用反义寡核苷酸耗竭 enhancer 转录出的 RNA,在成人红系细胞里同样放出了胎儿血红蛋白。那把剪刀真正交付的结果,本就是一个表观状态——能不能跳过剪刀,直接把这个状态装上去?

证据到了哪一层

必须把话说清楚:表观编辑在造血干细胞里的证据,至今仍停留在人类原代细胞加异种移植小鼠这一层。

已发表的关键一步来自 2023 年。Saunderson 等人在 PNAS 上把 dCas9 与 DNMT3A-3L 的融合体送进人 CD34⁺ 造血干/祖细胞,靶向 CDKN2B(p15)启动子。尽管递送是瞬时的,新装上的甲基化在体外髓系分化中被保持了下来;移植进小鼠后,它在造血干/祖细胞中长期维持,并被髓系和淋系的后代共同继承。这项研究本为探究衰老相关的异常甲基化,却顺带证明了一件对治疗至关重要的事:人的造血干/祖细胞能”记住”一次表观改写。论文同时记下两个警讯:甲基化沿着靶位点向向导 RNA 之外扩散;而 p15 被高甲基化之后,移植小鼠体内髓系与淋系细胞的总比例并没有变,但髓系内部的构成变了——单核细胞减少、粒细胞增多,而且这些单核细胞呈现出炎症性的转录程序。表观编辑不是无声无息的。

更系统的一次尝试来自 2026 年春天 Sankaran 实验室的一份 bioRxiv 预印本,尚未经过同行评审,读的时候要按这个分量对待。把三种编辑器的 mRNA 分别电转进人 CD34⁺ 细胞后,CRISPRi 一开始能让约九成细胞失去目标蛋白,几天之内却逐渐消退,与组蛋白修饰难以遗传的性质一致;两种基于 DNA methylation 的编辑器则做到约八成到九成五的持续沉默,在富集了真正造血干细胞的 CD34⁺CD45RA⁻CD90⁺ 亚群里同样成立。随后他们用体积更小的 CHARM 关掉血小板整合素 ITGB3,分化出的成熟巨核细胞中约九成不再表达 CD61;由此分化出的培养血小板,在与健康供者血小板混合的体外聚集试验中,形成聚集体的能力明显减弱——作者同时写明,在这个异种移植模型里,对在体血小板活性与血栓形成做功能评估是做不到的。这份沉默经受住了三周无细胞因子的自我更新培养、连续三轮集落再接种,以及移植进 NBSGW 小鼠后十六周的长期植入。最后再电转一次 TET1-dCas9,甲基化被擦掉、CD61 恢复,可逆也被演示了。

而在人体这一侧,表观编辑已经开始被给药,却一例也不在骨髓里:慢性乙型肝炎有 Tune-401(NCT06671093)与 CRMA-1001(NCT07200193),面肩肱型肌营养不良有 AAVrh74 递送的 EPI-321(NCT06907875),PCSK9 那条路线则走到了 STX-1150(NCT07428473),用脂质纳米颗粒递送 mRNA 与向导 RNA,试验记录里明确写着它不改变 DNA 序列。肝脏、肌肉,正是递送最容易抵达的组织;骨髓不在其中。

还没解决的部分

最硬的一道题是特异性,它与 nuclease 的脱靶不是同一类问题。2018 年,Galonska 等人在 Nat Commun 上追踪 dCas9-甲基转移酶留下的足迹:这类融合蛋白在细胞核里存在广泛的、并非由向导 RNA 指定的活性。核酸酶的脱靶还能被逐一定位,游走的甲基转移酶留下的却是弥散的化学痕迹;加上 Saunderson 观察到的位点周边扩散,评估它的安全性需要一整套新的方法学。

其次是持久性的两面。可逆是优点,出了问题可以撤销;但对一个要维持一辈子的造血系统也可能是缺点——目前最长的观察不过是小鼠体内近一年、异种移植十九周,而人的造血干细胞要在骨髓里工作几十年。这枚印章会不会被一点点磨掉,今天没有人能回答。再次是边界:表观编辑只能调节一个基因的开关,无法纠正一个拼错的字母,那片疆域仍然属于 base editing 与 prime editing。最后是递送:它真正诱人的前景是体内给药,而表观编辑器偏偏是庞大的融合蛋白、payload 最大的那一类,能塞进载体的空间格外紧张——缩小体积的努力密集出现在最近两年,原因正在这里。

印章与剪刀

细胞从来不用剪刀关掉一个基因:它盖章,再让复制机器把章一代代复印下去。基因编辑走了十年,才绕回这个更古老也更温和的办法。

对造血干细胞来说这个思路的分量还要重一层:它被要求终身无误运转,而它最成熟的那个靶点,本质上要的就是一个”关”字。证据链已经从培养皿一路延伸到非人灵长类和人体试验,唯独在骨髓这一段仍然是空的。剪刀已经把第一批病人送出了医院;印章还停在门口,等着人证明它按下去之后,那一页在几十年里都不会再被翻开。

参考文献

  1. Qi LS, Larson MH, Gilbert LA, Doudna JA, Weissman JS, Arkin AP, Lim WA. Repurposing CRISPR as an RNA-guided platform for sequence-specific control of gene expression. Cell. 2013;152(5):1173-1183. DOI
  2. Gilbert LA, Larson MH, Morsut L, et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell. 2013;154(2):442-451. DOI
  3. Thakore PI, D’Ippolito AM, Song L, et al. Highly specific epigenome editing by CRISPR-Cas9 repressors for silencing of distal regulatory elements. Nat Methods. 2015;12(12):1143-1149. DOI
  4. Amabile A, Migliara A, Capasso P, Biffi M, Cittaro D, Naldini L, Lombardo A. Inheritable silencing of endogenous genes by hit-and-run targeted epigenetic editing. Cell. 2016;167(1):219-232.e14. DOI
  5. Liu XS, Wu H, Ji X, et al. Editing DNA methylation in the mammalian genome. Cell. 2016;167(1):233-247.e17. DOI
  6. Nuñez JK, Chen J, Pommier GC, et al. Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing. Cell. 2021;184(9):2503-2519.e17. DOI
  7. Cappelluti MA, Mollica Poeta V, Valsoni S, Quarato P, Merlin S, Merelli I, Lombardo A. Durable and efficient gene silencing in vivo by hit-and-run epigenome editing. Nature. 2024;627(8003):416-423. DOI
  8. Tremblay F, Xiong Q, Shah SS, et al. A potent epigenetic editor targeting human PCSK9 for durable reduction of low-density lipoprotein cholesterol levels. Nat Med. 2025;31(4):1329-1338. DOI
  9. Neumann EN, Bertozzi TM, Wu E, et al. Brainwide silencing of prion protein by AAV-mediated delivery of an engineered compact epigenetic editor. Science. 2024;384(6703):eado7082. DOI
  10. Wang K, Wang J, et al. Silencing of BCL11A by disrupting enhancer-dependent epigenetic insulation. Blood. 2026;147(13):1470-1484. DOI
  11. Saunderson EA, Encabo HH, Devis J, et al. CRISPR/dCas9 DNA methylation editing is heritable during human hematopoiesis and shapes immune progeny. Proc Natl Acad Sci U S A. 2023;120(34):e2300224120. DOI
  12. Ye T, Xu W, Barrachina MN, et al. Epigenome editing of human hematopoietic stem cells enables sustained and reversible thrombosis prevention. bioRxiv. 2026;2026.03.27.714536(预印本,未经同行评审). DOI
  13. Galonska C, Charlton J, Mattei AL, et al. Genome-wide tracking of dCas9-methyltransferase footprints. Nat Commun. 2018;9(1):597. DOI
  14. Tune Therapeutics. Phase 1b study of Tune-401 in chronic hepatitis B. ClinicalTrials.gov NCT06671093
  15. nChroma Bio. Phase 1/2 study of CRMA-1001 in chronic hepatitis B. ClinicalTrials.gov NCT07200193
  16. Epicrispr Biotechnologies. First-in-human study of EPI-321 in facioscapulohumeral muscular dystrophy. ClinicalTrials.gov NCT06907875
  17. Monash University / Scribe Therapeutics. Phase 1 study of STX-1150 in participants with elevated LDL-C. ClinicalTrials.gov NCT07428473

Your liver cells and your lymphocytes carry the same genome. What makes them so unlike each other is not that one of them is missing a few stretches of DNA, but which passages are read aloud and which are closed and sealed. Long ago the cell learned how to keep a gene switched off for good, and it never reached for scissors: it presses a chemical stamp onto the DNA, keeps that page shut, and at every subsequent division copies the stamp to both daughter cells. If the cell already has a silencing mechanism this reliable, why do we insist on cutting the chromosome?

What the cell already knew

The protagonist of that stamp is DNA methylation. The start regions of genes often carry dense clusters of CpG sites; when those are heavily methylated and the chromatin is weighed down with repressive histone marks, the transcription machinery can no longer read through. What matters is heritability: when DNA replicates, methylation on the new strand is filled in by maintenance methyltransferases copying the parental strand — not a one-off suppression, but a memory that can be photocopied forward. It took gene editing nearly a decade to turn that into a tool you can aim.

First, learning to hold it down

In 2013, Qi and colleagues, writing in Cell, inactivated both cleavage domains of Cas9 to obtain a dCas9 that knows the way but carries no blade; a guide RNA walks it to a gene, and mere occupancy is enough to block RNA polymerase. They called the system CRISPR interference (CRISPRi): in Escherichia coli no off-target effects were detectable, and — a point written up at the time as a virtue — the effect was reversible. Later that year, Gilbert and colleagues, also in Cell, fused dCas9 to effector domains and brought both repression and activation into human cells. Two years later, Thakore and colleagues, in Nat Methods, directed dCas9-KRAB to the HS2 enhancer that orchestrates multiple globin genes: H3K9 trimethylation was written there with high specificity, multiple globin genes fell silent, and global gene expression barely moved.

But this generation of tools shared one weakness: they depended on something holding the button down the whole time — the histone modifications recruited by KRAB do not travel well across cell division, and once the editor is diluted or degraded, the gene creeps back. To become a therapy, the cell has to keep remembering after the editor is gone.

Making it remember

In the autumn of 2016, two back-to-back papers in the same issue of Cell supplied the missing layer of memory. The approach of Amabile and colleagues came to be called hit-and-run: borrowing the machinery that embryonic stem cells use to silence endogenous retroviruses, they delivered a set of engineered transcriptional repressors transiently into somatic cells, where they acted together at the same locus to write repressive histone marks and elicit de novo DNA methylation. Genome-wide analysis showed the silencing was tightly confined to the target locus without spreading to neighbouring genes, and the only thing that could relieve it was targeted DNA demethylation. In the same issue, Liu and colleagues supplied the other half of the toolkit: tethering Tet1 or Dnmt3a to a catalytically dead Cas9 makes it possible to erase or write methylation at a chosen site — they used dCas9-Tet1 to reopen BDNF in post-mitotic neurons.

By 2021, Nuñez and colleagues had folded all of this into a single molecule: CRISPRoff is one fusion protein combining the KRAB domain of ZNF10 with Dnmt3A and Dnmt3L, all tethered to dCas9. After transient expression, silencing lasted at least 50 days; the paper states that human cells maintain it for more than 450 cell divisions, and at one endogenous gene more than 90% of cells lost expression. A genome-wide screen showed that the great majority of protein-coding genes can be switched off this way, including those without a classical CpG island; the companion CRISPRon can erase the memory. Silencing also survives differentiation — throughout the induction of stem cells into neurons, the switched-off gene stayed off.

One transient dose, a heritable state written down, no double-strand break, and in principle reversible — the shape of epigenetic editing as a therapeutic modality had become clear.

Out of the dish

What pushed it toward clinical imagination was the 2024 work of Cappelluti and colleagues in Nature. They packaged the mRNA encoding the repressors into lipid nanoparticles, targeted Pcsk9 — a gene that governs cholesterol metabolism in hepatocytes — and gave mice a single intravenous injection: circulating PCSK9 fell rapidly and then stabilized at around 40% of control levels, all the way to day 330, when the experiment was terminated. In a second cohort of mice, three months after dosing, they performed a partial hepatectomy to force hepatocytes to proliferate and regenerate — after regeneration, the silencing was still there, and so was the methylation. An optimized all-in-one version pushed the reduction to around 75% over a 43-day observation, comparable to knocking the gene out with CRISPR-Cas9 directly, but without causing DNA breaks. The following year, Tremblay and colleagues, in Nat Med, carried the same target into a system closer to humans: after a single dose in cynomolgus monkeys, circulating PCSK9 fell by approximately 90% and LDL cholesterol by approximately 70%; and in mice, a targeted epigenetic activator could reopen a silenced locus.

On the delivery side, meanwhile, size has been shrinking. Cappelluti’s team compared dCas9, TALE and zinc-finger proteins as DNA-binding platforms in vitro and settled on zinc fingers because they were the best-performing platform in that screen; the authors separately note in their discussion that zinc-finger proteins have a reduced molecular size and are independent of short-lived guide RNAs, features that would favour delivery and the stability of the editing complex. The same year, CHARM, reported by Neumann and colleagues in Science, went further: it no longer carries its own methyltransferase but uses a histone H3 tail-Dnmt3l fusion to recruit and activate the cell’s endogenous methyltransferases, thereby reducing transgene size and cytotoxicity, and it silenced the prion protein gene across the mouse brain after systemic AAV delivery.

Why bone marrow is where it belongs

Hematopoietic stem cells are the cells that can least afford a mistake: they have to work in the body for a lifetime, and any erroneous change is amplified by self-renewal into an entire clone. And the p53 response, large deletions and chromosomal rearrangements that come with a double-strand break weigh especially heavily in these cells (more on that when we get to safety). Epigenetic editing does not cut DNA, and in principle sidesteps that whole class of risk.

As it happens, the most successful target in blood disease is one that only needs to be switched off. Cutting the erythroid enhancer of BCL11A with CRISPR brings fetal hemoglobin back up, and that is exactly the mechanism behind the therapy that has been approved. The explanation offered by Wang and colleagues in Blood in 2026 is intriguing: cutting that enhancer works because it destroys the chromatin insulation required to sustain BCL11A expression, and what it ultimately produces is epigenetic silencing of BCL11A itself; they also used antisense oligonucleotides to deplete the RNAs transcribed from the enhancer, which likewise released fetal hemoglobin in adult erythroid cells. What the scissors actually deliver is an epigenetic state — so could one skip the scissors and install that state directly?

How far the evidence reaches

It has to be said plainly: the evidence for epigenetic editing in hematopoietic stem cells still sits at the level of human primary cells plus xenotransplanted mice.

The key published step came in 2023. Saunderson and colleagues, in PNAS, delivered a fusion of dCas9 with DNMT3A-3L into human CD34⁺ hematopoietic stem/progenitor cells, targeting the CDKN2B (p15) promoter. Despite transient delivery, the newly installed methylation was maintained through myeloid differentiation in vitro; after transplantation into mice, it was maintained long term in hematopoietic stem/progenitor cells and inherited by both myeloid and lymphoid progeny. The study was designed to probe aging-associated aberrant methylation, but it incidentally demonstrated something crucial for therapy: human hematopoietic stem/progenitor cells can remember a single epigenetic rewrite. The paper also records two warning signs: methylation spread beyond the guide RNA location around the target site; and after p15 was hypermethylated, the overall proportions of myeloid and lymphoid cells in transplanted mice were unaffected, but the composition within the myeloid lineage shifted — monocytes decreased and granulocytes increased, and those monocytes showed increased inflammatory transcriptional programs. Epigenetic editing is not silent.

A more systematic attempt came from a bioRxiv preprint out of the Sankaran lab in the spring of 2026 — not yet peer reviewed, and to be read with that weight in mind. After electroporating the mRNA of three editors separately into human CD34⁺ cells, CRISPRi initially made about 90% of cells lose the target protein, but the effect faded within days, consistent with the poor heritability of histone modifications; the two DNA methylation-based editors achieved sustained silencing in about 80% to 95% of cells, which also held in the CD34⁺CD45RA⁻CD90⁺ subset enriched for bona fide hematopoietic stem cells. They then used the smaller CHARM to switch off the platelet integrin ITGB3, and about 90% of the mature megakaryocytes differentiated from those cells no longer expressed CD61; the culture-derived platelets produced from them showed a clearly reduced capacity to form aggregates in an in vitro assay in which they were mixed with healthy donor platelets — the authors also state that in vivo functional assessment of platelet activity and thrombus formation was not feasible in this xenotransplantation model. The silencing survived three weeks of cytokine-free self-renewal culture, serial replating through tertiary colonies, and sixteen weeks of long-term engraftment after transplantation into NBSGW mice. Finally, a second electroporation with TET1-dCas9 erased the methylation and restored CD61, demonstrating reversibility.

On the human side, epigenetic editing has begun to be dosed into people, but not one case is in the bone marrow: chronic hepatitis B has Tune-401 (NCT06671093) and CRMA-1001 (NCT07200193), facioscapulohumeral muscular dystrophy has the AAVrh74-delivered EPI-321 (NCT06907875), and the PCSK9 route has reached STX-1150 (NCT07428473), which delivers mRNA and guide RNA in lipid nanoparticles and whose trial record states explicitly that it does not change the DNA sequence. Liver and muscle are precisely the tissues that delivery reaches most easily; bone marrow is not among them.

What is still unsolved

The hardest question is specificity, and it is not the same class of problem as nuclease off-targeting. In 2018, Galonska and colleagues, in Nat Commun, tracked the footprints left by dCas9-methyltransferases: these fusion proteins show widespread nuclear activity that is not specified by the guide RNA. Nuclease off-targets can still be mapped one by one, whereas a roaming methyltransferase leaves a diffuse chemical trace; together with the spread around the target site observed by Saunderson, assessing its safety will require an entirely new methodology.

Second is the double face of durability. Reversibility is an advantage — if something goes wrong it can be undone; but for a blood system that must be maintained for a lifetime it may also be a drawback: the longest observations so far run only to nearly one year in mice and nineteen weeks in xenografts, while human hematopoietic stem cells have to work in the marrow for decades. Whether this stamp gets rubbed away bit by bit, nobody today can answer. Third are the limits of scope: epigenetic editing can only adjust a gene’s switch, not correct a misspelled letter — that territory still belongs to base editing and prime editing. Last is delivery: its truly enticing prospect is dosing in vivo, and yet epigenetic editors are bulky fusion proteins, the largest payloads of the lot, which leaves the space inside a vector especially tight — the push to shrink them has clustered in the past two years for exactly this reason.

The stamp and the scissors

A cell never uses scissors to switch off a gene: it stamps, and lets the replication machinery copy the stamp forward through the generations. Gene editing took a decade to come back around to this older and gentler method.

For hematopoietic stem cells this line of thinking carries even more weight: they are required to run without error for a lifetime, and their most mature target essentially asks for nothing more than off. The chain of evidence already runs from the dish to non-human primates and human trials — only the bone marrow stretch is still empty. The scissors have already sent the first patients home from the hospital; the stamp is still at the door, waiting for someone to show that once it is pressed down, that page stays shut for decades.

References

  1. Qi LS, Larson MH, Gilbert LA, Doudna JA, Weissman JS, Arkin AP, Lim WA. Repurposing CRISPR as an RNA-guided platform for sequence-specific control of gene expression. Cell. 2013;152(5):1173-1183. DOI
  2. Gilbert LA, Larson MH, Morsut L, et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell. 2013;154(2):442-451. DOI
  3. Thakore PI, D’Ippolito AM, Song L, et al. Highly specific epigenome editing by CRISPR-Cas9 repressors for silencing of distal regulatory elements. Nat Methods. 2015;12(12):1143-1149. DOI
  4. Amabile A, Migliara A, Capasso P, Biffi M, Cittaro D, Naldini L, Lombardo A. Inheritable silencing of endogenous genes by hit-and-run targeted epigenetic editing. Cell. 2016;167(1):219-232.e14. DOI
  5. Liu XS, Wu H, Ji X, et al. Editing DNA methylation in the mammalian genome. Cell. 2016;167(1):233-247.e17. DOI
  6. Nuñez JK, Chen J, Pommier GC, et al. Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing. Cell. 2021;184(9):2503-2519.e17. DOI
  7. Cappelluti MA, Mollica Poeta V, Valsoni S, Quarato P, Merlin S, Merelli I, Lombardo A. Durable and efficient gene silencing in vivo by hit-and-run epigenome editing. Nature. 2024;627(8003):416-423. DOI
  8. Tremblay F, Xiong Q, Shah SS, et al. A potent epigenetic editor targeting human PCSK9 for durable reduction of low-density lipoprotein cholesterol levels. Nat Med. 2025;31(4):1329-1338. DOI
  9. Neumann EN, Bertozzi TM, Wu E, et al. Brainwide silencing of prion protein by AAV-mediated delivery of an engineered compact epigenetic editor. Science. 2024;384(6703):eado7082. DOI
  10. Wang K, Wang J, et al. Silencing of BCL11A by disrupting enhancer-dependent epigenetic insulation. Blood. 2026;147(13):1470-1484. DOI
  11. Saunderson EA, Encabo HH, Devis J, et al. CRISPR/dCas9 DNA methylation editing is heritable during human hematopoiesis and shapes immune progeny. Proc Natl Acad Sci U S A. 2023;120(34):e2300224120. DOI
  12. Ye T, Xu W, Barrachina MN, et al. Epigenome editing of human hematopoietic stem cells enables sustained and reversible thrombosis prevention. bioRxiv. 2026;2026.03.27.714536 (preprint, not peer reviewed). DOI
  13. Galonska C, Charlton J, Mattei AL, et al. Genome-wide tracking of dCas9-methyltransferase footprints. Nat Commun. 2018;9(1):597. DOI
  14. Tune Therapeutics. Phase 1b study of Tune-401 in chronic hepatitis B. ClinicalTrials.gov NCT06671093
  15. nChroma Bio. Phase 1/2 study of CRMA-1001 in chronic hepatitis B. ClinicalTrials.gov NCT07200193
  16. Epicrispr Biotechnologies. First-in-human study of EPI-321 in facioscapulohumeral muscular dystrophy. ClinicalTrials.gov NCT06907875
  17. Monash University / Scribe Therapeutics. Phase 1 study of STX-1150 in participants with elevated LDL-C. ClinicalTrials.gov NCT07428473