Yang Liu

← Primer

HSC · Healing Source Code

AAV 的两副面孔——同一个载体,在体外是英雄,在体内却搞不定造血干细胞 The two faces of AAV—one vector, a hero in the dish, helpless against hematopoietic stem cells in the body

2026-07-16 · AAV, 递送, 造血干细胞, 基因编辑, ex-vivo

翻开美国 FDA 那份获批细胞与基因治疗产品的名单,你会看到一串以 “-parvovec” 结尾的通用名:Luxturna 治遗传性视网膜病变,Zolgensma 治脊髓性肌萎缩症,Hemgenix 和 Roctavian 分别对付血友病 B 和 A,Elevidys 送进肌肉,Kebilidi 送进大脑。它们背后是同一辆车——adeno-associated virus,AAV。可是把名单从头看到尾,会发现一个刺眼的空缺:没有一款是冲着骨髓、冲着造血干细胞去的。血友病那两款靠的是让肝细胞分泌凝血因子,绕开了造血系统本身。造血干细胞不在 AAV 的战果里。更奇怪的是,走进任何一间做造血干细胞基因编辑的实验室,你几乎一定会看到 AAV6——它是把序列精确写进 HSPC 的首选工具。同一个载体,在培养皿里是英雄,在人体里却偏偏够不着骨髓。这不是一个失败的故事,而是一堂关于”载体属性与细胞生物学是否匹配”的课。

AAV 凭什么赢下那么多器官

要理解 AAV 在血液系统的失手,得先弄清它在别处凭什么赢。

AAV 是一种很小的单链 DNA 病毒,天然感染人却不引起明确的疾病。作为载体,它有三样让人踏实的优点:免疫原性相对温和;不同的 serotype 各有 tropism,可以挑着送往不同组织;而最要紧的是,它进入细胞核后不倾向于整合进宿主染色体,而是以游离的 episomal 形式待着——不整合,就意味着不会像逆转录病毒那样把自己插进癌基因附近。2016 年,Pillay 等人在 Nature 上用一次无偏的遗传学筛选给这套系统补上了分子基础:他们在单倍体人类细胞里找出 AAV2 感染必需的蛋白——一个此前未被描述的 I 型跨膜蛋白 KIAA0319L,命名为 AAV receptor(AAVR);敲掉它,各种哺乳动物细胞都对 AAV2 高度抵抗,而且他们测试过的所有 serotype 都依赖这个受体。

但 AAV 真正的杀手锏,藏在一个常被忽略的巧合里:它最成功的那些器官,细胞都不太分裂。视网膜的感光细胞、脊髓的运动神经元、成年人的肝细胞、肌纤维——都是终末分化、静止或近乎静止的组织。对它们来说,一个只是游离在核里的 episomal 基因组,可以安安稳稳待上很多年。AAV 不需要整合就能实现”长期表达”,不是因为它有多顽强,而是因为它选对了不会把它稀释掉的细胞。

一个被肝脏切除实验戳穿的秘密

这个巧合并非猜测,它在二十多年前就被一个设计得极漂亮的实验测量过。2001 年,Nakai 等人在 Journal of Virology 上想搞清楚:AAV 在肝脏里的长期表达,靠的是整合进染色体的那部分基因组,还是游离的那部分?他们的办法是逼肝细胞分裂——给注射过 AAV 的小鼠做三分之二的部分肝切除,强行触发肝脏再生。如果表达来自整合基因组,它会随染色体复制传给子代,肝再生不该影响它;如果来自 episomal 基因组,它没有复制机制,每一次分裂都会把它对半稀释。

结果是决定性的:作为对照,来自整合型转座子载体的凝血因子 IX 水平在肝切除后纹丝不动;而 AAV 组的因子 IX 血浆水平下降了约 92%,每个细胞里稳定存在的载体基因组数下降了约 86%。这个实验一次说清了两件事——AAV 在肝脏的长期表达压倒性地来自 episomal 基因组;而这份”长期”完全建立在细胞不分裂这个前提上。

如果说肝切除还是人为的极端,2009 年 Cunningham 等人在 Molecular Therapy 上报告的结果,则展示了同一机制在真实治疗场景里的杀伤力。他们用 AAV2/8 给鸟氨酸氨甲酰基转移酶缺陷的 Spf(ash) 小鼠送去正常的酶:成年鼠的代谢纠正强劲而且终身维持;新生鼠的完全纠正却只是短暂的,随后便滑落下去。原因写在论文结论里,朴素得近乎冷酷——新生鼠的肝脏还在长大,而这份表达的流失,正是肝细胞增殖的后果。

同一个载体,在不分裂的成年肝脏里意味着”终身”,在正在生长的新生肝脏里就只意味着”一阵子”。而造血干细胞,是这个身体里最不可能停止分裂的细胞之一——它一生的职责,就是不停地自我更新,并源源不断地生产出数以万亿计的后代。把一个靠”不被稀释”来维持的载体交给一个”以分裂为业”的细胞,这是一场从生物学根子上就错配的婚姻。

稀释之外,还有几道关

即便抛开稀释这个终局问题,AAV 想在体内碰到造血干细胞,还得先闯过几道关,而每一道都不利。

先是找得到、进得去。骨髓深处血流缓慢、隔着屏障,这是所有全身递送共同的困境;而即使把车开到细胞跟前,AAV 还得靠受体和 tropism 认门,偏偏 AAV 家族对造血细胞普遍不友好。2013 年,Song 等人在 PLoS One 上系统比较了十种 serotype,结论是 AAV6 转导原代人类造血干细胞的效率最高——但”最高”需要被正确理解:那是在十个不算好的选项里挑出的相对最优,而且要靠在 capsid 表面把特定酪氨酸残基突变掉,才能把 CD34⁺ 的转导率推过 70%,而这一切发生在培养皿里。接着是进核与变双链这两道连在一起的关。2004 年,Zhong 等人在 Human Gene Therapy 上发现,在原代小鼠 c-Kit⁺Lin⁻ 造血细胞里,约 85% 的 AAV 基因组滞留在细胞质、根本没进核;而当他们给小鼠注射羟基脲——一种核糖核苷酸还原酶抑制剂,靠阻断 dNTP 的供应把细胞卡在 G1/S 交界,在 AAV 领域早就被发现能增强转导——滞留比例降到约 40%,核内的那一份相应升到约 60%。核转运这道关确实松动了;可 Zhong 等人紧接着发现,进了核的病毒粒子大多没能脱壳,基因组仍裹在衣壳里,第二链合成与转基因表达因此只得到部分的改善。推开一道门,后面还锁着一道。这里要留意作者自己划的那条界:人的造血细胞难转导,Zhong 等人归因于 AAV 受体的表达量不足;而小鼠的这一小群细胞受体和辅受体都有、却照样转导不动,才轮到核转运与脱壳出来解释。两个物种卡在不同的环节上,却指向同一个结局。而单链变双链这一步本身,早在 1996 年就被 Ferrari FK 等人在 Journal of Virology 上指认为 AAV 转导的限速步骤:他们注意到腺病毒的存在能大幅提高转导,又用一系列腺病毒突变体把这份增益定位到 E4 区的第 6 号开放阅读框(E4 ORF6)——单是用质粒表达这一个蛋白,就足以把转导效率提高一百到一千倍,而它的作用点正是合成第二链。

这个三十年前的病毒学发现,今天读起来像是给造血干细胞量身定做的判决。长期造血干细胞的定义性特征之一就是深度静息——它不进细胞周期,不启动 DNA 合成机器;而 AAV 恰恰要靠细胞的 DNA 合成机器,才能把自己变成一个能被读取的模板。HSC 越是”干”,AAV 就越是转不动它。更何况门口还站着免疫:2010 年 Boutin 等人在 Human Gene Therapy 上测得,健康人群中抗 AAV2、AAV1 的总 IgG 阳性率高达 72% 和 67%,抗 AAV6 也有 46%,中和抗体的血清阳性率以 AAV2 的 59% 为最高,serotype 之间还交叉反应显著——相当比例的人在第一次接受治疗之前,血液里就埋伏着能拦下这辆车的抗体(这道关留到讲载体免疫原性时再细说)。于是答案浮出来了:AAV 搞不定体内的造血干细胞,不是因为哪一个环节特别糟,而是因为受体不亲和、进核与脱壳一路受阻、静息又卡住第二链合成、抗体还拦在门口——每个物种被卡住的具体环节甚至不尽相同,可谁也没能走通;而就算前面全部侥幸闯过,一个 episomal 基因组也终究会被这个终身分裂的细胞稀释掉。

反转:把缺点变成优点

故事到这里如果就结束,AAV 大概会被写成造血领域的失败者。可真实发生的事恰恰相反——今天几乎每一个做 HSPC 精准编辑的实验室,都离不开 AAV6。

转折点在 2015 年。Wang 等人在 Nature Biotechnology 上报告了一个后来被反复复制的组合拳:电穿孔把锌指核酸酶的 mRNA 送进细胞,同时用 AAV6 递送同源供体模板。在动员外周血来源的 CD34⁺ HSPC 里,CCR5 和 AAVS1 两个位点分别拿到平均 17% 和 26% 的定点插入,胎肝来源的则是 19% 和 43%。更关键的是,被改动的细胞里包含 CD34⁺CD133⁺CD90⁺ 这个含长期重建造血干细胞的稀有亚群,而且它们能在免疫缺陷小鼠体内长期植入——证明真正的 HSC 被击中了,而不只是那些很快会被淘汰的短命前体。

一年后,Dever 等人在 Nature 上把这套逻辑推到了最有临床分量的靶点:用 Cas9 ribonucleoprotein 配上 AAV6 递送的同源供体,在造血干细胞的 HBB 基因上完成同源重组,并筛出定点整合率超过 90% 的 HSPC;用镰刀型贫血患者的细胞,他们纠正了那个著名的 Glu6Val 突变。到 2021 年,Lattanzi 等人在 Science Translational Medicine 上把这套流程做成了临床规模的产品:患者动员来的 CD34⁺ 细胞实现最高 60% 的 HBB 等位基因纠正,移植进 NSG 小鼠后仍保有 20% 的纠正率和多谱系植入,长期安全性与致瘤性研究未见异常造血或基因毒性——他们据此认为,这些数据支持启动一项针对镰刀型贫血的 phase 1/2 临床试验。

为什么同一辆在体内寸步难行的车,到了体外就所向披靡?因为 ex vivo 这个环境把 AAV 的每一条短板都顺手拆掉了,而且——这是最精妙的部分——把它最致命的那条,直接翻译成了优点。

细胞被从骨髓里取出来,泡在细胞因子里,被主动推出静息、推进细胞周期——1996 年被指认的那道第二链合成的限速关,至少不再卡在”细胞根本不开动 DNA 合成机器”这个死结上。至于 Zhong 等人当年在小鼠细胞里撞见的核转运与脱壳,ex vivo 的成功并不是把他们那个羟基脲效应外推来的——那个效应本身也只换来部分的改善。真正把这条路走通的是经验:AAV6 这个 serotype、开到很高的 MOI、加上细胞因子培养,这套组合在 Wang 等人和 Dever 等人手里被证明足以把序列高效写进 HSPC,至于其中每一道关各自松了多少,并没有被逐条拆开算过。剂量也不再受限于全身分布——在一个培养孔里,MOI 可以开到体内根本达不到的水平,不必担心肝脏截走绝大部分载体。预存的中和抗体?培养皿里没有血清,这一关直接不存在。

而”episomal、不整合、会被稀释”这条在体内致命的属性,在体外被彻底掉了个个儿。因为在 HDR 编辑里,AAV 根本不需要长期存在——它的全部工作,就是在 Cas9 切开染色体的那一瞬间,在旁边提供一份同源的 DNA 模板;细胞的同源重组机器把序列抄写进染色体之后,它的使命就完成了。此后这个 episomal 基因组被分裂稀释干净,不但无害,反而正是我们想要的:改动被永久写在染色体上,载体自己不留痕迹地退场。一个”用完即弃”的模板,恰恰需要的就是”不会赖着不走”的载体。

这里还有一层运气。2018 年,Cromer 等人在 Molecular Therapy 上把整套 CRISPR/Cas9-AAV6 流程的每个组件拆开,逐一测量 CD34⁺ HSPC 的全局转录反应,结果颇为意外:Cas9 mRNA 引发了最大的转录扰动和明显的抗病毒反应,而 AAV6——这个货真价实的病毒——竟然没有引发可检测到的病毒反应。它安静得不像病毒,这也是它在体外好用的原因之一。

代价并没有消失

不过,“好用”不等于”没有代价”。AAV6 在体外带来的麻烦换了一副面孔:不是免疫,而是 DNA 损伤应答。2019 年,Schiroli 等人在 Cell Stem Cell 上用优化过的核酸酶在 HSPC 各亚群里追踪损伤应答,一直做到单细胞分辨率:哪怕只是单个核酸酶造成的一个双链断裂,主导的细胞反应也是 p53 通路激活;而当断裂负荷升高、或叠加上 AAV 递送的修复模板时,p53 的激活会累积,压制编辑后 HSPC 的增殖、产量和植入能力。次年,Ferrari S 等人(Naldini 组)在 Nature Biotechnology 上用条形码追踪编辑后细胞的克隆命运,把严重性量化了:p53 的激活显著收缩了小鼠体内人源移植物的 HSC 克隆库,而短暂抑制 p53 能把多克隆结构恢复回来;当他们再叠加上腺病毒 5 型的 E4orf6/7 蛋白——同样出自 1996 年那篇论文里被点名的 E4 区,只是另一个开放阅读框、另一套本事:它招募细胞周期开关 E2F,强行推动细胞周期、上调同源重组机器——长期人源移植物里的 HDR 效率被推到最高 50%。腺病毒这个当年被拿来解剖 AAV 生活史的老搭档,二十多年后又一次成了 AAV 的助手,只不过这回帮的是另一道工序。

这对 in vivo 意味着什么

把 AAV 这两副面孔并排看,对体内造血干细胞基因治疗的启示是否定性的,但很清晰:ex vivo 能把 AAV 用得如此漂亮,靠的是四件在体内统统办不到的事——把细胞从静息里推出来、把剂量开到任意高、把血清抗体排除在外、只要求载体短暂存在。体内的每一条都反了过来。

所以问题不是”能不能把 AAV 的效率再调高一点”,而是这个载体的核心属性,与体内造血干细胞的核心生物学之间存在方向性的冲突。这也解释了为什么体内造血领域的主流路线并不指望 AAV 独自扛下全场:要么用整合型载体让改动随染色体传下去,要么干脆放弃”让载体长期存在”这个念头,用一次性的碱基编辑把序列直接改写进染色体,再配一辆非病毒的、能开进骨髓的车。AAV 若要在体内造血里翻身,现实路径大概是从 capsid 本身下手——用定向进化搜出一个既能穿透骨髓、又能被静息 HSC 接受的新衣壳,那已是另一篇的题目。

匹配,而不是优劣

回到开篇那份 FDA 名单。AAV 在眼睛、肝脏、肌肉、神经系统的连胜,和它在骨髓前的止步,是同一条生物学规律的两面:它是一个不整合的、游离的、依赖宿主 DNA 合成机器才能启动的载体。在不分裂的细胞里,这三条分别意味着”安全""持久""够用”;在终身分裂、且以静息为立身之本的造血干细胞里,同样这三条就变成了”会被稀释""转瞬即逝""根本启动不了”。

而当我们把细胞取出体外、把它唤醒、只让 AAV 干十几个小时的活,这三条又一次翻转:不整合成了”不留痕迹”,游离成了”用完即弃”,短暂成了”刚刚好”。

递送载体从来没有绝对的好坏。有的只是:你要送的东西、你要送到的细胞、你要它待多久——这三者对不对得上。AAV 在造血系的故事之所以值得讲,不是因为它输了,而是因为它同时示范了错配有多致命,和匹配有多漂亮。

参考文献

  1. Ferrari FK, Samulski T, Shenk T, Samulski RJ. Second-strand synthesis is a rate-limiting step for efficient transduction by recombinant adeno-associated virus vectors. J Virol. 1996;70(5):3227-34. DOI
  2. Nakai H, et al. Extrachromosomal recombinant adeno-associated virus vector genomes are primarily responsible for stable liver transduction in vivo. J Virol. 2001;75(15):6969-76. DOI
  3. Zhong L, et al. Impaired nuclear transport and uncoating limit recombinant adeno-associated virus 2 vector-mediated transduction of primary murine hematopoietic cells. Hum Gene Ther. 2004;15(12):1207-18. DOI
  4. Cunningham SC, et al. AAV2/8-mediated correction of OTC deficiency is robust in adult but not neonatal Spf(ash) mice. Mol Ther. 2009;17(8):1340-6. DOI
  5. Boutin S, et al. Prevalence of serum IgG and neutralizing factors against adeno-associated virus (AAV) types 1, 2, 5, 6, 8, and 9 in the healthy population: implications for gene therapy using AAV vectors. Hum Gene Ther. 2010;21(6):704-12. DOI
  6. Song L, et al. High-efficiency transduction of primary human hematopoietic stem cells and erythroid lineage-restricted expression by optimized AAV6 serotype vectors in vitro and in a murine xenograft model in vivo. PLoS One. 2013;8(3):e58757. DOI
  7. Wang J, et al. Homology-driven genome editing in hematopoietic stem and progenitor cells using ZFN mRNA and AAV6 donors. Nat Biotechnol. 2015;33(12):1256-1263. DOI
  8. Pillay S, et al. An essential receptor for adeno-associated virus infection. Nature. 2016;530(7588):108-12. DOI
  9. Dever DP, et al. CRISPR/Cas9 β-globin gene targeting in human haematopoietic stem cells. Nature. 2016;539(7629):384-389. DOI
  10. Cromer MK, et al. Global Transcriptional Response to CRISPR/Cas9-AAV6-Based Genome Editing in CD34+ Hematopoietic Stem and Progenitor Cells. Mol Ther. 2018;26(10):2431-2442. DOI
  11. Schiroli G, et al. Precise Gene Editing Preserves Hematopoietic Stem Cell Function following Transient p53-Mediated DNA Damage Response. Cell Stem Cell. 2019;24(4):551-565.e8. DOI
  12. Ferrari S, et al. Efficient gene editing of human long-term hematopoietic stem cells validated by clonal tracking. Nat Biotechnol. 2020;38(11):1298-1308. DOI
  13. Lattanzi A, et al. Development of β-globin gene correction in human hematopoietic stem cells as a potential durable treatment for sickle cell disease. Sci Transl Med. 2021;13(598):eabf2444. DOI
  14. U.S. Food and Drug Administration. Approved Cellular and Gene Therapy Products. FDA

Open the U.S. FDA’s list of approved cellular and gene therapy products and you will find a string of generic names ending in “-parvovec”: Luxturna for inherited retinal disease, Zolgensma for spinal muscular atrophy, Hemgenix and Roctavian for hemophilia B and A respectively, Elevidys delivered into muscle, Kebilidi into the brain. Behind them all is the same vehicle—adeno-associated virus, AAV. Yet read the list from top to bottom and a glaring absence emerges: not one of them is aimed at the bone marrow, at hematopoietic stem cells. The two hemophilia products work by having hepatocytes secrete clotting factors, bypassing the hematopoietic system itself. Hematopoietic stem cells are not among AAV’s conquests. Stranger still, walk into any laboratory doing hematopoietic stem cell gene editing and you are almost certain to find AAV6—it is the tool of choice for writing sequences precisely into HSPCs. The same vector: a hero in the dish, yet in the human body unable to reach the marrow. This is not a story of failure but a lesson in whether a vector’s properties match a cell’s biology.

What won AAV so many organs

To understand AAV’s failure in the blood system, one has to first understand what wins it elsewhere.

AAV is a very small single-stranded DNA virus that naturally infects humans without causing any well-defined disease. As a vector it has three reassuring virtues: relatively mild immunogenicity; different serotypes with different tropisms, so one can choose where to ship it; and, most importantly, once inside the nucleus it does not tend to integrate into the host chromosome but sits there as a free, episomal form—not integrating means it will not insert itself near an oncogene the way a retrovirus does. In 2016, Pillay and colleagues supplied the molecular foundation for this system in Nature with an unbiased genetic screen: in haploid human cells they identified the protein required for AAV2 infection—a previously undescribed type I transmembrane protein, KIAA0319L, which they named AAV receptor (AAVR); knock it out and a variety of mammalian cells become highly resistant to AAV2, and every serotype they tested depended on this receptor.

But AAV’s real trump card is hidden in a coincidence that often goes unnoticed: in the organs where it succeeds best, the cells do not divide much. The photoreceptors of the retina, the motor neurons of the spinal cord, the hepatocytes of an adult, muscle fibers—all are terminally differentiated, quiescent or nearly quiescent tissues. For them, an episomal genome merely floating in the nucleus can sit there safely for many years. AAV achieves “long-term expression” without integrating not because it is especially tenacious, but because it picked the cells that would not dilute it away.

A secret exposed by a liver-resection experiment

This coincidence is not a guess; it was measured more than twenty years ago by a beautifully designed experiment. In 2001, Nakai and colleagues set out in the Journal of Virology to settle a question: does AAV’s long-term expression in the liver come from the fraction of genomes integrated into the chromosome, or from the free fraction? Their method was to force hepatocytes to divide—performing a two-thirds partial hepatectomy on mice that had been injected with AAV, forcibly triggering liver regeneration. If the expression came from integrated genomes, it would be passed to daughter cells along with chromosomal replication, and regeneration should not affect it; if it came from episomal genomes, which have no replication mechanism, every division would dilute them by half.

The result was decisive: as a control, factor IX levels from an integrating transposon vector did not budge after hepatectomy; in the AAV group, plasma factor IX levels fell by about 92%, and the number of vector genomes stably present per cell fell by about 86%. This single experiment settled two things at once—AAV’s long-term expression in the liver comes overwhelmingly from episomal genomes; and that “long-term” rests entirely on the premise that the cells do not divide.

If hepatectomy still seems an artificial extreme, the results Cunningham and colleagues reported in Molecular Therapy in 2009 show the same mechanism’s destructive power in a real therapeutic setting. They used AAV2/8 to deliver the normal enzyme to Spf(ash) mice deficient in ornithine transcarbamylase: in adult mice the metabolic correction was robust and sustained for life; in neonatal mice the complete correction was only transient, and then slid away. The reason is written in the paper’s conclusion, plain to the point of cruelty—the neonatal liver is still growing, and this loss of expression is precisely the consequence of hepatocyte proliferation.

The same vector means “for life” in a non-dividing adult liver and only “for a while” in a growing neonatal one. And the hematopoietic stem cell is among the cells in this body least likely to ever stop dividing—its lifelong job is to keep renewing itself and to churn out trillions of descendants without pause. Handing a vector that survives by “not being diluted” to a cell whose profession is division is a marriage mismatched at the biological root.

Beyond dilution, several more gates

Even setting aside dilution as the endgame problem, for AAV to encounter a hematopoietic stem cell in the body it must first get past several gates, and every one of them is unfavorable.

First, finding them and getting in. Deep in the bone marrow the blood flow is slow and a barrier stands in the way—this is the common predicament of all systemic delivery; and even with the vehicle driven up to the cell, AAV still has to recognize the door via receptors and tropism, and the AAV family happens to be generally unfriendly toward hematopoietic cells. In 2013, Song and colleagues systematically compared ten serotypes in PLoS One and concluded that AAV6 transduces primary human hematopoietic stem cells most efficiently—but “most efficiently” needs to be understood correctly: it is the relative best picked from ten not-very-good options, and it takes mutating specific tyrosine residues on the capsid surface to push CD34⁺ transduction past 70%, and all of this happened in a dish. Next come the two linked gates of nuclear entry and conversion to double-stranded DNA. In 2004, Zhong and colleagues found in Human Gene Therapy that in primary murine c-Kit⁺Lin⁻ hematopoietic cells, about 85% of AAV genomes were retained in the cytoplasm, never entering the nucleus at all; and when they injected the mice with hydroxyurea—a ribonucleotide reductase inhibitor that stalls cells at the G1/S boundary by cutting off the dNTP supply, and which had long been found in the AAV field to enhance transduction—the retained fraction dropped to about 40%, and the nuclear fraction rose correspondingly to about 60%. The nuclear-transport gate did loosen; but Zhong and colleagues immediately found that most of the virions that made it into the nucleus failed to uncoat, the genome still wrapped in the capsid, so second-strand synthesis and transgene expression were only partially improved. Push one door open and another is still locked behind it. Note here the line the authors themselves drew: human hematopoietic cells are hard to transduce, which Zhong and colleagues attributed to insufficient expression of the AAV receptor; whereas this small population of murine cells has both the receptor and the co-receptor and still cannot be transduced, which is where nuclear transport and uncoating come in as the explanation. Two species stuck at different steps, pointing to the same outcome. And the single-to-double-strand conversion itself was already identified as the rate-limiting step of AAV transduction back in 1996, by Ferrari FK and colleagues in the Journal of Virology: they noticed that the presence of adenovirus greatly increased transduction, and used a series of adenoviral mutants to map that gain to the sixth open reading frame of the E4 region (E4 ORF6)—expressing this one protein from a plasmid alone was enough to raise transduction efficiency a hundred- to a thousandfold, and its point of action is precisely the synthesis of the second strand.

This thirty-year-old virology finding reads today like a verdict tailor-made for hematopoietic stem cells. One of the defining features of long-term hematopoietic stem cells is deep quiescence—they do not enter the cell cycle, do not start up the DNA synthesis machinery; and AAV depends on exactly that cellular DNA synthesis machinery to turn itself into a readable template. The more “stem” an HSC is, the less AAV can transduce it. And immunity stands at the door besides: in 2010, Boutin and colleagues measured in Human Gene Therapy that in the healthy population, total IgG seroprevalence against AAV2 and AAV1 runs as high as 72% and 67%, against AAV6 46%, with neutralizing-antibody seroprevalence highest for AAV2 at 59%, and significant cross-reactivity between serotypes—meaning a substantial fraction of people, before ever receiving a first treatment, already have antibodies lying in wait in their blood that can stop this vehicle (a gate left for the discussion of vector immunogenicity). So the answer surfaces: AAV cannot handle hematopoietic stem cells in the body not because any one step is especially bad, but because the receptor affinity is poor, nuclear entry and uncoating are blocked all along the way, quiescence stalls second-strand synthesis, and antibodies wait at the door—the specific step at which each species gets stuck is not even the same, yet none of them gets through; and even if all of that were cleared by luck, an episomal genome would in the end still be diluted away by this lifelong-dividing cell.

The reversal: turning the flaws into virtues

If the story ended here, AAV would probably be written up as the loser of the hematopoietic field. But what actually happened is precisely the opposite—today, nearly every laboratory doing precise HSPC editing depends on AAV6.

The turning point came in 2015. Wang and colleagues reported in Nature Biotechnology a combination that would be replicated again and again: electroporation to deliver zinc-finger nuclease mRNA into the cells, together with AAV6 to deliver the homologous donor template. In CD34⁺ HSPCs from mobilized peripheral blood, the CCR5 and AAVS1 loci achieved mean targeted insertion rates of 17% and 26% respectively; from fetal liver, 19% and 43%. More crucially, the modified cells included the rare CD34⁺CD133⁺CD90⁺ subpopulation that contains long-term reconstituting hematopoietic stem cells, and these could engraft long-term in immunodeficient mice—proving that genuine HSCs had been hit, not merely the short-lived progenitors that are soon washed out.

A year later, Dever and colleagues pushed this logic in Nature to the target with the most clinical weight: using Cas9 ribonucleoprotein together with an AAV6-delivered homologous donor, they carried out homologous recombination at the HBB gene in hematopoietic stem cells and sorted out HSPCs with targeted integration rates above 90%; using cells from sickle cell disease patients, they corrected the famous Glu6Val mutation. By 2021, Lattanzi and colleagues had turned this workflow into a clinical-scale product in Science Translational Medicine: patient-mobilized CD34⁺ cells achieved up to 60% correction of HBB alleles, and after transplantation into NSG mice retained a 20% correction rate with multilineage engraftment, while long-term safety and tumorigenicity studies revealed no abnormal hematopoiesis or genotoxicity—on which basis they held that these data support the initiation of a phase 1/2 clinical trial for sickle cell disease.

Why does the same vehicle that cannot move an inch in the body become unstoppable outside it? Because the ex vivo setting casually dismantles every one of AAV’s weaknesses, and—this is the most exquisite part—translates the most fatal one directly into a virtue.

The cells are taken out of the marrow, bathed in cytokines, actively pushed out of quiescence and into the cell cycle—so the second-strand synthesis bottleneck identified in 1996 is at least no longer stuck at the dead end of “the cell simply does not turn on its DNA synthesis machinery.” As for the nuclear transport and uncoating that Zhong and colleagues ran into back then in murine cells, the ex vivo success is not an extrapolation of their hydroxyurea effect—that effect itself only bought a partial improvement. What actually made this path work was empirical: the AAV6 serotype, an MOI turned way up, plus cytokine culture—this combination was shown in the hands of Wang and colleagues and Dever and colleagues to be enough to write sequences efficiently into HSPCs; how much each individual gate loosened along the way was never taken apart and accounted for. Dose, too, is no longer constrained by systemic distribution—in a culture well the MOI can be pushed to levels simply unreachable in the body, with no worry about the liver intercepting the bulk of the vector. Pre-existing neutralizing antibodies? There is no serum in a dish; that gate simply does not exist.

And that property—“episomal, non-integrating, will be diluted”—which is fatal in the body, is turned completely inside out outside it. Because in HDR editing, AAV does not need to persist at all—its entire job is to offer a homologous DNA template alongside the chromosome at the instant Cas9 cuts it open; once the cell’s homologous recombination machinery has copied the sequence into the chromosome, its mission is complete. That the episomal genome is then diluted clean away by division is not merely harmless but exactly what we want: the change is written permanently into the chromosome and the vector exits without leaving a trace. A template meant to be used and discarded needs precisely a vector that will not overstay.

There is another layer of luck here. In 2018, Cromer and colleagues took apart each component of the whole CRISPR/Cas9-AAV6 workflow in Molecular Therapy, measuring the global transcriptional response of CD34⁺ HSPCs to each in turn, and the result was rather unexpected: Cas9 mRNA provoked the largest transcriptional perturbation and a clear antiviral response, while AAV6—a bona fide virus—provoked no detectable viral response at all. It is quiet in a way a virus should not be, and that is one reason it works so well outside the body.

The price has not disappeared

Still, “works well” is not the same as “comes free.” The trouble AAV6 brings ex vivo simply wears a different face: not immunity, but the DNA damage response. In 2019, Schiroli and colleagues tracked the damage response across HSPC subpopulations in Cell Stem Cell using an optimized nuclease, all the way down to single-cell resolution: even a single double-strand break made by a single nuclease provoked p53 pathway activation as the dominant cellular response; and when the break burden rose, or an AAV-delivered repair template was layered on, p53 activation accumulated, suppressing the proliferation, yield and engraftment capacity of edited HSPCs. The following year, Ferrari S and colleagues (the Naldini group) used barcoding in Nature Biotechnology to track the clonal fate of edited cells and quantified the severity: p53 activation markedly contracted the HSC clonal repertoire of human grafts in mice, and transient p53 inhibition could restore the polyclonal structure; and when they further layered on the adenovirus type 5 E4orf6/7 protein—also from the E4 region singled out in that 1996 paper, only a different open reading frame with a different skill: it recruits the cell-cycle switch E2F, forcibly driving the cell cycle and upregulating the homologous recombination machinery—HDR efficiency in long-term human grafts was pushed to as high as 50%. Adenovirus, the old companion once used to dissect AAV’s life cycle, had become AAV’s helper once again more than twenty years later, only this time assisting at a different step.

What this means for in vivo

Setting AAV’s two faces side by side, the lesson for in vivo hematopoietic stem cell gene therapy is negative but very clear: the reason ex vivo can use AAV so beautifully rests on four things that are all impossible in the body—pushing the cells out of quiescence, turning the dose up arbitrarily high, keeping serum antibodies out, and asking the vector to persist only briefly. In the body, every one of them is reversed.

So the question is not “can we tune AAV’s efficiency up a little more,” but that this vector’s core properties are in directional conflict with the core biology of hematopoietic stem cells in the body. This also explains why the mainstream routes in the in vivo hematopoietic field do not count on AAV to carry the whole load alone: either use an integrating vector so the change is passed down with the chromosome, or abandon the idea of “having the vector persist” altogether and use one-shot base editing to rewrite the sequence directly into the chromosome, paired with a non-viral vehicle that can drive into the marrow. If AAV is to turn things around in in vivo hematopoiesis, the realistic path is probably to start with the capsid itself—using directed evolution to search out a new shell that can both penetrate the marrow and be accepted by a quiescent HSC. That is the subject of another piece.

Matching, not better or worse

Back to that FDA list at the opening. AAV’s winning streak in the eye, the liver, muscle and the nervous system, and its halt at the door of the bone marrow, are two sides of the same biological rule: it is a vector that does not integrate, stays free, and depends on the host’s DNA synthesis machinery to get started. In non-dividing cells, those three things mean “safe,” “durable,” and “good enough” respectively; in hematopoietic stem cells, which divide for life and whose very identity rests on quiescence, those same three become “will be diluted,” “gone in a flash,” and “cannot even start.”

And when we take the cells out of the body, wake them up, and ask AAV to work for only a dozen or so hours, those three flip once again: non-integrating becomes “leaves no trace,” episomal becomes “used and discarded,” and brief becomes “just right.”

Delivery vectors are never absolutely good or bad. There is only this: what you want to deliver, the cell you want to deliver it to, and how long you need it to stay—whether those three line up. The story of AAV in the blood system is worth telling not because it lost, but because it demonstrates at once how fatal a mismatch can be, and how beautiful a match is.

References

  1. Ferrari FK, Samulski T, Shenk T, Samulski RJ. Second-strand synthesis is a rate-limiting step for efficient transduction by recombinant adeno-associated virus vectors. J Virol. 1996;70(5):3227-34. DOI
  2. Nakai H, et al. Extrachromosomal recombinant adeno-associated virus vector genomes are primarily responsible for stable liver transduction in vivo. J Virol. 2001;75(15):6969-76. DOI
  3. Zhong L, et al. Impaired nuclear transport and uncoating limit recombinant adeno-associated virus 2 vector-mediated transduction of primary murine hematopoietic cells. Hum Gene Ther. 2004;15(12):1207-18. DOI
  4. Cunningham SC, et al. AAV2/8-mediated correction of OTC deficiency is robust in adult but not neonatal Spf(ash) mice. Mol Ther. 2009;17(8):1340-6. DOI
  5. Boutin S, et al. Prevalence of serum IgG and neutralizing factors against adeno-associated virus (AAV) types 1, 2, 5, 6, 8, and 9 in the healthy population: implications for gene therapy using AAV vectors. Hum Gene Ther. 2010;21(6):704-12. DOI
  6. Song L, et al. High-efficiency transduction of primary human hematopoietic stem cells and erythroid lineage-restricted expression by optimized AAV6 serotype vectors in vitro and in a murine xenograft model in vivo. PLoS One. 2013;8(3):e58757. DOI
  7. Wang J, et al. Homology-driven genome editing in hematopoietic stem and progenitor cells using ZFN mRNA and AAV6 donors. Nat Biotechnol. 2015;33(12):1256-1263. DOI
  8. Pillay S, et al. An essential receptor for adeno-associated virus infection. Nature. 2016;530(7588):108-12. DOI
  9. Dever DP, et al. CRISPR/Cas9 β-globin gene targeting in human haematopoietic stem cells. Nature. 2016;539(7629):384-389. DOI
  10. Cromer MK, et al. Global Transcriptional Response to CRISPR/Cas9-AAV6-Based Genome Editing in CD34+ Hematopoietic Stem and Progenitor Cells. Mol Ther. 2018;26(10):2431-2442. DOI
  11. Schiroli G, et al. Precise Gene Editing Preserves Hematopoietic Stem Cell Function following Transient p53-Mediated DNA Damage Response. Cell Stem Cell. 2019;24(4):551-565.e8. DOI
  12. Ferrari S, et al. Efficient gene editing of human long-term hematopoietic stem cells validated by clonal tracking. Nat Biotechnol. 2020;38(11):1298-1308. DOI
  13. Lattanzi A, et al. Development of β-globin gene correction in human hematopoietic stem cells as a potential durable treatment for sickle cell disease. Sci Transl Med. 2021;13(598):eabf2444. DOI
  14. U.S. Food and Drug Administration. Approved Cellular and Gene Therapy Products. FDA