Yang Liu

← Primer

HSC · Healing Source Code

免清髓体内编辑——不杀骨髓,怎么让编辑过的细胞占优 Conditioning-Free In Vivo Editing — How Edited Cells Win Without Killing the Marrow First

2026-07-31 · 造血干细胞, in vivo 基因编辑, 范可尼贫血, ADA-SCID, MGMT 药物选择, 镰状细胞病

传统的造血干细胞移植,第一步几乎总是一场蓄意的破坏。白消安或全身放疗把骨髓里原本占据每一个角落的细胞几乎清空,像先把一块地烧成焦土,新来的种子才有地方扎根。这场”烧地”手术本身就足以致命——它是这一百年移植医学里最沉重的成本之一。In vivo 基因编辑最诱人的承诺,正是跳过这一步:不采集、不清髓,直接把编辑器打进血管,让它在体内找到造血干细胞、悄悄改写几个基因,病人甚至可以当天回家。可问题也随之而来——骨髓里的龛位(niche)还挤满了从未被触碰过的原住民细胞,那一小撮被编辑过的干细胞,凭什么能在竞争里笑到最后?

先烧地,还是不烧地

清髓的逻辑其实很直白:骨髓里的龛位数量有限,常驻的造血干细胞牢牢占据着这些位置,新回输的细胞要想长期定居、参与终身造血,几乎必须先腾出空间。这也是为什么经过充分清髓的移植,供者细胞往往能在几个月内占到血液系统的绝大部分——不是因为供者细胞天生更强,而是因为对手已经被清场。

In vivo 编辑反其道而行:载体经静脉注射直接进入体内,在没有清髓、龛位仍被原装干细胞占满的情况下去寻找造血干细胞并完成编辑。这意味着,即便递送效率不错,目前多数报道显示,能够真正被修改的长期造血干细胞占比通常也只是个位数到十几个百分点——其余绝大多数龛位仍属于那些从未被编辑器碰过的细胞。如果编辑过的细胞和它们的邻居在骨髓这个封闭市场里完全势均力敌,那一小撮”改过”的细胞,终其一生也很可能只是被稀释掉的少数派,难以在血液输出里留下可被测量的印记。真正决定这条路线成败的问题,从来不是编辑器切得准不准,而是:没有人为它们清场,这些编辑过的细胞要靠什么赢下这场旷日持久的竞争。

自然已经做过这个实验

答案的第一部分,其实早已被疾病本身写好了剧本。范可尼贫血(Fanconi anemia)患者的造血干细胞携带 DNA 修复通路上的先天缺陷,基因组不稳定性持续累积,最终导致骨髓衰竭。但临床医生很早就注意到一个奇怪的现象:少数患者的血象会在没有任何治疗干预的情况下悄悄好转。2002 年,Gross 等人在 Cytogenetic and Genome Research 上系统描述了这背后的机制——某个造血干细胞偶然获得了一次自发的第二次突变,恰好补偿了原来的缺陷,让这个克隆重新获得了正常的 DNA 修复能力。这个单一的”自我修正”克隆,没有借助任何载体、任何化疗、任何清髓,就凭借相对于依旧带病的邻居们的生存优势,在随后数年里逐渐扩张,在部分嵌合体(mosaic)患者体内成长为血细胞产出的主要来源,一些患者的骨髓衰竭进程因此显著减缓。这一现象被称为”反向嵌合”(reverse mosaicism),常被形容成一场大自然自己完成的基因治疗——它证明了一件对 in vivo 编辑至关重要的事:只要被修改的那一个克隆确实比邻居更能打,它完全可以在完全不清场的骨髓里,靠时间本身完成接管。

把这份运气写进方案里

范可尼贫血的例子是自然的巧合,而腺苷脱氨酶缺陷型重症联合免疫缺陷(ADA-SCID)则是第一次被人为地把这份”运气”设计进了治疗方案。ADA 缺陷会让脱氧腺苷及其代谢产物在细胞内蓄积到有毒的浓度,这种毒性对淋巴细胞格外致命,却唯独放过了那些已经拿到功能性 ADA 基因的造血干细胞——换句话说,疾病本身就替基因矫正后的细胞铺好了竞争优势。2002 年,Aiuti 等人在 Science 上报告,他们把这条逻辑用在了两名患者身上:自体 CD34⁺ 细胞经基因矫正后回输,只配合了远低于清髓强度的预处理(nonmyeloablative conditioning),矫正细胞仍然实现了持续、多系的 engraftment,患者的免疫功能逐步恢复。

这条逻辑后来被推到了更极端的位置。2021 年,Uchiyama 等人在 Molecular Therapy: Methods & Clinical Development 上报告了两名日本 ADA-SCID 患者的长期随访——他们在 2003 至 2004 年间接受基因矫正细胞回输时,完全没有使用任何预处理,不化疗、不放疗。十余年的随访显示,基因标记的克隆依然逐步在多个血细胞系里占据主导地位;作者推测,这很可能是因为患病骨髓本身的微环境(包括代谢毒性和成骨细胞功能受损)已经在被动地为健康克隆腾出空间。这是一个近乎极限的验证:哪怕连非清髓预处理都省去,只要疾病本身已经把天平倾斜向矫正后的细胞,竞争优势就足以独立完成接管——不需要人为烧地,时间和生物学会替你完成剩下的工作。

大多数病,没有这份先天优势

但把目光转向镰状细胞病、β-地中海贫血这类当下 in vivo 编辑最急于攻克的疾病,故事的前提就不成立了。这些病的缺陷主要体现在造血干细胞分化出的红细胞上——干细胞本身活得好好的,并没有像范可尼贫血或 ADA-SCID 那样被自己的病拖累。这意味着,编辑过的干细胞和未被触碰的干细胞在骨髓龛位的竞争中几乎处在同一起跑线上,自然选择不会主动替你完成筛选。

这正是”体外制造选择优势”这套工程学思路被引入的地方——最成熟的一种做法,是给编辑过的细胞装上一个药物抗性基因 MGMT(P140K),它对烷化剂 O6-苄基鸟嘌呤(O6-BG)不敏感,而未编辑的细胞会被同样的药物清除。这套逻辑最早在犬类大动物模型里得到验证:2003 年,Neff 等人在 Journal of Clinical Investigation 上报告,经过反复的 O6-BG/BCNU 给药周期,一只犬体内的基因标记细胞比例被推高到粒细胞的约 98%,是当时报道过的体内标记水平的最高纪录。它的价值在于,这种选择不是一次性的暴力清场,而是一个可以反复启动、逐步加码的药物旋钮。

把这套选择工具直接嫁接到 in vivo 编辑上,已经有了一个具体的证明。André Lieber 实验室开发的 HDAd5/35++ 载体,靶向造血干祖细胞表面的 CD46 受体,只需先用 G-CSF 和 AMD3100(plerixafor)把干细胞短暂动员进外周血,再经静脉注射即可完成体内转导,全程不涉及清髓。2021 年,Li 等人在 Molecular Therapy 上用携带人源 CD46 的镰状细胞病小鼠模型验证了这条路线:单纯依靠体内递送,最初只有约 5% 的红细胞携带治疗性的 γ-珠蛋白表达,不足以改变病情;但研究者随后启动了三轮低剂量的 O6-BG/BCNU 选择性给药,把这个比例逐步推高到超过 95%,小鼠的溶血、脾脏肿大、镰变红细胞等表型随之全面恢复正常。这个实验把整套逻辑压缩进了同一只动物:先在不清髓的前提下完成一次性的体内编辑,再用一个可反复调控的药物选择步骤,让这一小撮编辑过的细胞逐渐赢下与邻居的竞争——至于这套选择机制本身的精细设计与潜在风险,值得单独用一整篇来细说。

或者,干脆不必依赖选择

还有一条并行的思路,是把赌注押在递送效率本身,而不是事后的选择:如果一次注射就能编辑到足够高比例的长期造血干细胞,也许根本不需要额外的选择步骤来放大它。2025 年,Botchkarev 等人在 Nature Biotechnology 上报告了一种包膜工程化的病毒样颗粒(VLP),经静脉注射、全程不使用任何预处理或选择性给药,在人源化小鼠体内于给药后 8 周仍能在长期人造血干祖细胞中维持约 31% 的 B2M 基因编辑;而在血红蛋白病相关的 BCL11A、HBG1/2 位点上,作者报告的约 26% 与 7.5% 编辑效率,测的是给药后第 5 天这个更早的时间点——能否在同样 8 周的长期节点上维持,这篇论文本身尚未给出数据。如果某种疾病所需要的疗效阈值恰好落在这类效率之下,单靠一针足够精准的递送,或许就能绕开人为制造竞争优势这一整套麻烦。

边界

即便是最成熟的药物选择策略,也没有彻底摆脱基因毒性的影子。2012 年,Adair 等人在 Science Translational Medicine 上报告,三名接受 MGMT 修饰造血干细胞联合反复 O6-BG/替莫唑胺周期治疗的胶质母细胞瘤患者,基因标记细胞的比例确实随治疗周期逐步上升,证明这套选择机制在人体内同样成立;但其中两名患者体内扩张的优势克隆,其病毒整合位点恰好邻近原癌基因 PRDM16 与 HMGA2 附近——一个提醒:把清髓换成可反复调用的药物选择,换来的是更温和、可控的暴露,而不是零暴露,克隆动力学的风险仍需被持续监测(这条线留给后面专讲克隆动力学与基因毒性的篇目)。选择本身也只能放大已经存在的少数派,如果一次体内递送连极少数长期造血干细胞都没能真正修改到,再多轮的药物选择也无法凭空造出可供扩张的种子。而无论是范可尼贫血、ADA-SCID 式的天然优势,还是 MGMT 式的人工优势,真正被验证到能在人体内、针对镰状细胞病或地中海贫血这类无天然选择压力的疾病里稳定起效的临床数据,眼下都还在路上。

回到那块被烧过的地——in vivo 编辑真正想省下的,或许从来不是”用一场更温和的火”去清场,而是学会去读懂,在一片从未被烧过的土地上,谁本就占着上风;而当生物学没有主动递上这份优势时,想办法为它悄悄造出一个。

参考文献

  1. Gross M, Hanenberg H, Lobitz S, et al. Reverse mosaicism in Fanconi anemia: natural gene therapy via molecular self-correction. Cytogenet Genome Res. 2002;98(2-3):126-135. DOI
  2. Aiuti A, Slavin S, Aker M, et al. Correction of ADA-SCID by stem cell gene therapy combined with nonmyeloablative conditioning. Science. 2002;296(5577):2410-2413. DOI
  3. Uchiyama T, Takahashi S, Nakabayashi K, et al. Nonconditioned ADA-SCID gene therapy reveals ADA requirement in the hematopoietic system and clonal dominance of vector-marked clones. Mol Ther Methods Clin Dev. 2021;23:424-433. DOI
  4. Neff T, Horn PA, Peterson LJ, et al. Methylguanine methyltransferase–mediated in vivo selection and chemoprotection of allogeneic stem cells in a large-animal model. J Clin Invest. 2003;112(10):1581-1588. DOI
  5. Li C, Wang H, Georgakopoulou A, Gil S, Yannaki E, Lieber A. In Vivo HSC Gene Therapy Using a Bi-modular HDAd5/35++ Vector Cures Sickle Cell Disease in a Mouse Model. Mol Ther. 2021;29(2):822-837. DOI
  6. Botchkarev VV Jr, Harrington S, Stoppato M, et al. In vivo gene editing of human hematopoietic stem and progenitor cells using envelope-engineered virus-like particles. Nat Biotechnol. 2025 Dec 5. DOI
  7. Adair JE, Beard BC, Trobridge GD, et al. Extended survival of glioblastoma patients after chemoprotective HSC gene therapy. Sci Transl Med. 2012;4(133):133ra57. DOI

The first step of a conventional hematopoietic stem cell transplant is almost always a deliberate act of destruction. Busulfan or total-body irradiation empties the bone marrow of nearly every cell that once occupied it — scorching the earth so the new seeds have somewhere to take root. That scorching alone is dangerous enough to be fatal — it is one of the heaviest costs in a century of transplant medicine. The most tempting promise of in vivo gene editing is skipping this step entirely: no harvest, no conditioning, just infuse the editor into the bloodstream and let it find the hematopoietic stem cells and quietly rewrite a few genes — the patient can go home the same day. But then the problem follows: the niches in the marrow are still packed with untouched native cells. What makes anyone think that small edited fraction can win the competition?

Scorch the earth, or don’t

The logic of conditioning is simple: niche space in the marrow is limited, and resident hematopoietic stem cells hold onto it tightly, so cells newly infused need that space cleared before they can settle in for good and contribute to lifelong hematopoiesis. That is why, after full myeloablative conditioning, donor cells typically come to dominate the blood system within a few months — not because donor cells are inherently stronger, but because the competition has been cleared away.

In vivo editing does the opposite: the vector is delivered intravenously directly into the body, seeking out hematopoietic stem cells and editing them while the niches remain full of untouched, native cells and no conditioning has been given. That means even with reasonably efficient delivery, most reports so far put the fraction of long-term hematopoietic stem cells actually modified somewhere in the single digits to the low teens — the great majority of niches still belong to cells the editor never touched. If edited cells and their neighbors are otherwise perfectly evenly matched in this closed marketplace that is the marrow, that small edited minority will likely spend a lifetime being diluted out, never leaving a measurable mark on blood output. The question that actually decides whether this approach succeeds was never how precisely the editor cuts — it’s this: with no one clearing the field for them, what do the edited cells have to win this long, unaided competition?

Nature already ran this experiment

The first part of the answer was written by disease itself, long before anyone thought to ask the question. Patients with Fanconi anemia carry an inborn defect in a DNA-repair pathway; genomic instability accumulates continuously, eventually leading to bone marrow failure. But clinicians noticed early on a curious phenomenon: in a subset of patients, blood counts quietly improve with no treatment at all. In 2002, Gross and colleagues, writing in Cytogenetic and Genome Research, systematically described the mechanism behind it — a single hematopoietic stem cell acquires a spontaneous second mutation that happens to compensate for the original defect, restoring that clone’s DNA-repair capacity. This one self-correcting clone, with no vector, no chemotherapy, no conditioning of any kind, expanded over the following years purely on the strength of its survival advantage over its still-diseased neighbors, becoming the dominant source of blood production in some mosaic patients, and slowing the progression of bone marrow failure in a number of them. The phenomenon, called reverse mosaicism, is often described as gene therapy nature performs on its own — and it demonstrates something essential for in vivo editing: as long as the modified clone genuinely outcompetes its neighbors, it can take over an entirely uncleared marrow, given nothing but time.

Writing that luck into the protocol

Fanconi anemia is a natural coincidence; adenosine deaminase-deficient severe combined immunodeficiency (ADA-SCID) was the first disease where that same luck was deliberately built into a treatment protocol. ADA deficiency lets deoxyadenosine and its metabolites accumulate to toxic concentrations inside cells — a toxicity that is especially lethal to lymphocytes, yet spares precisely those hematopoietic stem cells that have already received a functional ADA gene. In other words, the disease itself hands corrected cells a competitive edge. In 2002, Aiuti and colleagues, reporting in Science, applied this logic to two patients: autologous CD34⁺ cells were gene-corrected and reinfused, accompanied only by nonmyeloablative conditioning, far below the intensity of full conditioning, and the corrected cells still achieved sustained, multilineage engraftment, with the patients’ immune function gradually recovering.

This logic was later pushed to an even more extreme position. In 2021, Uchiyama and colleagues, in Molecular Therapy: Methods & Clinical Development, reported long-term follow-up of two Japanese ADA-SCID patients who received infusions of gene-corrected cells between 2003 and 2004 with no conditioning at all — no chemotherapy, no irradiation. Over a decade of follow-up showed gene-marked clones gradually coming to dominate across multiple blood lineages; the authors proposed this was most likely because the diseased marrow’s own microenvironment — including metabolic toxicity and impaired osteoblast function — was passively clearing space for the healthy clone. It is a near-limit demonstration: even when nonmyeloablative conditioning itself is skipped, as long as the disease has already tilted the balance toward the corrected cells, a competitive advantage alone is enough to complete the takeover — no need to scorch the earth by hand; time and biology do the rest.

Most diseases don’t come with this built-in advantage

But turn to sickle cell disease and β-thalassemia — the diseases in vivo editing is currently most eager to tackle — and this premise falls apart. Their defects show up mainly in the red blood cells that hematopoietic stem cells differentiate into; the stem cells themselves are perfectly healthy, not weighed down by their own disease the way Fanconi anemia or ADA-SCID cells are. That means edited and untouched stem cells start on essentially equal footing in the competition for marrow niches, and natural selection won’t do the sorting for you.

This is exactly where the engineering idea of manufacturing a selective advantage in vivo comes in — the most mature version of which equips edited cells with the drug-resistance gene MGMT(P140K), which is insensitive to the alkylating agent O6-benzylguanine (O6-BG), while unedited cells are cleared out by the same drug. This logic was first validated in a large-animal canine model: in 2003, Neff and colleagues, in the Journal of Clinical Investigation, reported that repeated cycles of O6-BG/BCNU dosing drove the fraction of gene-marked cells in one dog up to about 98% of granulocytes — the highest in vivo marking level reported at the time. Its value lies in the fact that this selection is not a one-time brute-force clearing but a knob that can be turned on repeatedly and dialed up in steps.

Grafting this selection tool directly onto in vivo editing has already produced a concrete proof of concept. The HDAd5/35++ vector developed by André Lieber’s lab targets the CD46 receptor on the surface of hematopoietic stem and progenitor cells; it requires only a brief mobilization of stem cells into peripheral blood using G-CSF and AMD3100 (plerixafor), followed by intravenous administration to achieve transduction in vivo, with no conditioning at any step. In 2021, Li and colleagues, in Molecular Therapy, validated this route in a human CD46-transgenic mouse model of sickle cell disease: relying on in vivo delivery alone, initially only about 5% of red blood cells carried therapeutic γ-globin expression, not enough to change the disease course; but the researchers then initiated three rounds of low-dose O6-BG/BCNU selective dosing, gradually driving that fraction above 95%, and the mice’s hemolysis, splenomegaly, and sickled red cells all fully normalized. This experiment compressed the entire logic into a single animal: a one-time in vivo editing step performed without conditioning, followed by a repeatable, tunable drug-selection step that lets a small edited population gradually win out over its neighbors — the fine design and risks of that selection mechanism itself are worth an entire piece of their own.

Or, simply skip needing selection at all

There is a parallel line of thinking that puts the bet on delivery efficiency itself, rather than selection after the fact: if a single injection can edit a high enough proportion of long-term hematopoietic stem cells, perhaps no additional selection step is needed to amplify it. In 2025, Botchkarev and colleagues, in Nature Biotechnology, reported an envelope-engineered virus-like particle (VLP) that, delivered intravenously with no conditioning or selective dosing of any kind, maintained roughly 31% B2M gene editing in long-term human hematopoietic stem and progenitor cells at 8 weeks after dosing in humanized mice; at the hemoglobinopathy-relevant BCL11A and HBG1/2 loci, the roughly 26% and 7.5% editing efficiencies the authors reported were measured at the earlier day-5 timepoint after dosing — whether those levels hold up at the same 8-week benchmark is not something this paper’s own data addresses. If the efficacy threshold a given disease requires happens to fall below what this kind of delivery can achieve, a single sufficiently precise injection alone might be enough to sidestep the whole business of manufacturing a competitive advantage.

The limits

Even the most mature drug-selection strategy has not fully escaped the shadow of genotoxicity. In 2012, Adair and colleagues, in Science Translational Medicine, reported that in three glioblastoma patients treated with MGMT-modified hematopoietic stem cells combined with repeated O6-BG/temozolomide cycles, the fraction of gene-marked cells did rise progressively over the treatment cycles, confirming that this selection mechanism also works in humans; but in two of those patients, the expanding dominant clone happened to carry viral integration sites near the proto-oncogenes PRDM16 and HMGA2 — a reminder that trading conditioning for a repeatable drug-selection knob buys a gentler, more controllable exposure, not zero exposure, and clonal dynamics still need continuous monitoring (a subject left for a later piece dedicated to clonal dynamics and genotoxicity). Selection itself can only amplify a minority that already exists — if a single in vivo delivery fails to genuinely modify even a tiny fraction of long-term hematopoietic stem cells, no number of additional selection rounds can conjure a population to expand from nothing. And whether it’s the natural advantage of Fanconi anemia and ADA-SCID, or the engineered advantage of MGMT, clinical data proving stable success in humans, specifically for diseases like sickle cell disease or thalassemia that carry no natural selective pressure of their own, is still on its way.

Back to that scorched field — what in vivo editing is really trying to save may never have been about setting a gentler fire to clear the ground. It’s learning to read who already holds the advantage on land that has never been burned at all — and when biology doesn’t hand that edge over on its own, finding a way to quietly manufacture one.

References

  1. Gross M, Hanenberg H, Lobitz S, et al. Reverse mosaicism in Fanconi anemia: natural gene therapy via molecular self-correction. Cytogenet Genome Res. 2002;98(2-3):126-135. DOI
  2. Aiuti A, Slavin S, Aker M, et al. Correction of ADA-SCID by stem cell gene therapy combined with nonmyeloablative conditioning. Science. 2002;296(5577):2410-2413. DOI
  3. Uchiyama T, Takahashi S, Nakabayashi K, et al. Nonconditioned ADA-SCID gene therapy reveals ADA requirement in the hematopoietic system and clonal dominance of vector-marked clones. Mol Ther Methods Clin Dev. 2021;23:424-433. DOI
  4. Neff T, Horn PA, Peterson LJ, et al. Methylguanine methyltransferase–mediated in vivo selection and chemoprotection of allogeneic stem cells in a large-animal model. J Clin Invest. 2003;112(10):1581-1588. DOI
  5. Li C, Wang H, Georgakopoulou A, Gil S, Yannaki E, Lieber A. In Vivo HSC Gene Therapy Using a Bi-modular HDAd5/35++ Vector Cures Sickle Cell Disease in a Mouse Model. Mol Ther. 2021;29(2):822-837. DOI
  6. Botchkarev VV Jr, Harrington S, Stoppato M, et al. In vivo gene editing of human hematopoietic stem and progenitor cells using envelope-engineered virus-like particles. Nat Biotechnol. 2025 Dec 5. DOI
  7. Adair JE, Beard BC, Trobridge GD, et al. Extended survival of glioblastoma patients after chemoprotective HSC gene therapy. Sci Transl Med. 2012;4(133):133ra57. DOI