Yang Liu

← Primer

HSC · Healing Source Code

当载体落错了地方——insertional mutagenesis 致白血病与自失活载体的诞生 When the Vector Lands in the Wrong Place — Insertional Mutagenesis, Leukemia, and the Birth of the Self-Inactivating Vector

2026-07-12 · insertional-mutagenesis, X-SCID, 基因治疗安全

有一类男婴,生下来就没有免疫系统。一个叫 IL2RG 的基因坏掉,让他们造不出成熟的 T 细胞和 NK 细胞,任何一次寻常的感染都可能致命,医学上称之为 X-linked SCID(SCID-X1)。二十多年前,一群医生第一次用基因治疗把这样的孩子从必死的境地里拉了回来——这是人类历史上第一批被基因疗法真正治愈的病人之一。然后,几年之后,其中一些孩子得了白血病。让他们活下来的那把钥匙,同时在他们的基因组里埋下了另一颗种子。这是基因治疗最沉重的一课,也是今天所有”安全”承诺的起点。

曙光

故事要从一个漂亮的成功讲起。2000 年,Cavazzana-Calvo、Fischer 等人在 Science 上报告了巴黎 Necker 儿童医院的一项试验:他们从 SCID-X1 患儿体内取出自身的 CD34⁺ 造血干细胞,在体外用一种 Moloney 鼠白血病病毒衍生的 γ-retroviral vector,把正常的 γc(IL2RG)基因送进细胞,再回输。十个月后,两名患儿体内出现了表达 γc 的 T 细胞和 NK 细胞,数量和功能都接近同龄的健康孩子。一个原本注定夭折的先天缺陷,被”补上一个基因”这件事纠正了。

这在当时是石破天惊的。此前几十年,基因治疗更多是承诺而非疗效;而这两个孩子,是第一个无可争辩地证明”把一个正常基因装回造血干细胞、就能重建一整套免疫”的活的证据。试验继续做下去,前后接受治疗的十名男婴里,九名重建了免疫、离开了无菌隔离舱,回家过上了普通孩子的生活。伦敦的另一支团队(Gaspar、Thrasher 等)也独立开展了几乎相同的试验,同样看到了免疫的恢复。领域一时间沉浸在乐观里。

阴影

转折出现在 2002 年前后。2003 年,Hacein-Bey-Abina 等人在 Science 上报告了一个谁都不愿看到的结果:在巴黎那批患儿中,两个年龄最小的孩子,在治疗将近三年之后,体内出现了成熟 T 细胞的克隆性增殖(当时判为癌前病变)——白血病的确诊,是随后随访里才坐实的事。研究者去追问原因,发现两个白血病克隆有一个惊人的共同点:那枚被用来运送基因的 γ-retroviral vector,都恰好整合在了一个叫 LMO2 的原癌基因的启动子附近,病毒自带的 enhancer 像一只错放的扩音器,把本该在成熟 T 细胞里沉默的 LMO2 硬生生地激活了。载体没有把基因送到一个”安全”的地方,而是落在了一个一旦被吵醒就会推动细胞癌变的开关旁边。

这不是孤例。随访继续,坏消息接踵而至。2008 年,Hacein-Bey-Abina 等人在 Journal of Clinical Investigation 上更新:巴黎试验里九名成功治疗的患儿,已有四人先后患上 T 细胞白血病,发病时间在治疗后的 31 到 68 个月之间。同期,伦敦团队的 Howe、Thrasher 等人也在同一本 Journal of Clinical Investigation 上报告,他们的试验中同样出现了一例因 insertional mutagenesis 引发的 T-cell acute lymphoblastic leukemia。两个独立的中心、同一种载体、同一个 LMO2——事情的性质变了:这不再是运气不好,而是这套技术本身携带的、可复现的风险。领域进入了它的至暗时刻,许多正在进行的试验被叫停。

整合的力量,与它的原罪

要理解这场事故,得先理解 retroviral vector 赖以工作的那件事:整合。这类病毒会把自己携带的基因永久地缝进宿主细胞的基因组,正是这种”缝进去”的能力,才让治疗基因能随造血干细胞一代代传下去、长期供货——这是它的力量所在。可整合是有代价的:它落在基因组的哪个位置,早期是无法控制的。更糟的是,γ-retroviral vector 有一种系统性的偏好,倾向于插进那些正在活跃转录的基因的起始区附近,而这些位置里就潜伏着不少调控增殖的原癌基因。载体长长的末端重复序列(LTR)里带着强力的 enhancer,一旦落在原癌基因隔壁,就可能把它长期开着。这就是它的原罪——赋予疗效的那个机制,和引发癌变的那个机制,是同一个。

但事故的全貌比”一次错误整合”更复杂,也更耐人寻味。Howe 等人在 2008 年那篇论文里把白血病克隆的基因组彻底拆开看,发现 LMO2 被激活只是第一步:真正把细胞推过癌变门槛的,还有一连串后天获得的突变——一个让 NOTCH1 功能获得性激活的点突变、抑癌基因 CDKN2A 所在位点的丢失、以及把 T 细胞受体区域接到 STIL-TAL1 上的染色体易位。换句话说,载体的插入是必要的第一击,却不是充分的全部;白血病是”插入激活 + 多重后天打击”共同作用的结果。这个认识在 Hacein-Bey-Abina 那篇 JCI 里也得到印证:不同患儿的白血病克隆里,除了 LMO2,还分别出现了插在 BMI1、CCND2 等原癌基因附近的载体。基因毒性不是一道简单的”命中或不命中”,而是一张被载体点燃、又被随机突变续火的网络。

这里必须诚实地说明一件不该被事故掩盖的事:即便如此,基因治疗对这些孩子仍然是有意义的。四名患白血病的孩子中,三人经化疗获得了持续缓解,一人不幸去世;而化疗之后,他们体内经过修饰的、多克隆的 T 细胞群落又重新长了回来,免疫功能得以维持。2010 年,Hacein-Bey-Abina 等人在 New England Journal of Medicine 上给出近十年的长期随访:九名患儿中八人存活,中位随访约九年,免疫缺陷在其中大多数人身上被持续地纠正了。这是一份沉重而复杂的资产负债表——它既写着”治愈”,也写着”白血病”,还写着”死亡”。领域没有资格轻描淡写其中任何一栏。

把扩音器拆掉

代价换来的,是一次彻底的安全范式重写。既然祸首是 LTR 里那个会激活邻居的 enhancer,那么最直接的对策就是:把它拆掉。这催生了 self-inactivating(SIN)vector——在载体设计上删去 LTR 中的病毒 enhancer/promoter 序列,让整合进去的前病毒失去自我转录和向外”喊话”的能力;治疗基因改由载体内部一个温和的、组织合适的 internal promoter 来驱动,而不再借用那把粗暴的强 enhancer。有的设计还额外加上 insulator(绝缘子)序列,像给整合位点砌一道隔音墙,进一步阻断它对周边基因的干扰。让载体”自我失活”,正是从那场事故里学到的最核心一课。

与此同时,领域开始用 lentiviral vector 系统性地替换 γ-retroviral vector。慢病毒同样能整合、同样能长期供货,但它的整合偏好不同——不那么偏爱原癌基因起始区那种高危位置;配上 SIN 设计和内部启动子,它的安全裕度明显更高。2010 年,Zhou 等人在 Blood 上报告了一款为 SCID-X1 定制的 SIN lentiviral vector:用 EF1α 内部启动子驱动 γc、并嵌入绝缘子,在体外的 Jurkat LMO2 激活实验中不再转激活 LMO2,在小鼠体内也未致瘤——把”不再吵醒 LMO2”这件事,做成了可以在实验台上验证的指标。

这些改动不是纸上谈兵,它们真的被带回了病床。2014 年,Hacein-Bey-Abina、Thrasher、Williams 等人在 New England Journal of Medicine 上报告了一项国际多中心试验:改用 SIN 型 γ-retroviral vector(删去了 enhancer)治疗九名 SCID-X1 男婴,疗效得以保留——多数患儿的 T 细胞获得功能性恢复;而当研究者去分析这批孩子体内的载体整合位点,发现它们在 LMO2、MECOM 等淋系原癌基因附近的聚集,比早期试验显著减少了。同一年前后,在另一种免疫缺陷病 Wiskott-Aldrich syndrome 上,Aiuti 等人在 Science 上报告用自失活(SIN)慢病毒载体治疗的患儿,造血呈现高度多克隆、多谱系,整合并未向原癌基因附近选择性富集,随访 20 到 32 个月未见异常克隆扩增。安全范式的重心,由此从”祈祷载体别落错地方”,转向”从设计上让它即使落错也不至于点火”,再辅以整合位点分析和长期随访,把每一次整合的后果放到显微镜下持续盯着。

尚未说完的部分

必须克制地承认,风险并没有被”解决”,只是被大幅压低了。2014 年那篇 NEJM 在结论里写得很老实:这种改良载体在长期意义上是否彻底摆脱了致白血病的可能,仍然未知——安全是一个需要用几十年随访去逐渐确认的说法,而不是一次实验就能盖棺定论的结论。整合始终是一个概率事件,SIN 设计和慢病毒把坏结果的概率往下压,却没有、也无法把它压到零。

更深的问题在于,当年那场事故里到底有多少要归咎于载体的 enhancer、有多少要归咎于被运送的 γc 基因本身或 SCID-X1 这种疾病特有的 T 细胞选择压力,至今仍有讨论——毕竟同样用 γ-retroviral vector 的一些其他疾病试验,并没有出现同等频率的白血病。这提醒我们,基因毒性从来不只是载体一个变量,它还纠缠着转基因、疾病背景、乃至被治疗细胞所处的增殖环境。而”整合会不会激活某个原癌基因""大片段缺失、染色体重排如何被及时发现”,这些更细的安全问题,构成了这个板块后面几篇要继续追问的东西。

收束

回头看,SCID-X1 的这段历史,是基因治疗从”能不能治好”走向”能不能安全地治好”的成人礼。它用最惨痛的方式教会了整个领域一件事:让治疗基因永久驻留的那种整合的力量,和让细胞癌变的那种整合的风险,本是一体两面;真正的进步不在于回避整合,而在于学会驯服它——把载体身上那把会误伤旁人的强 enhancer 拆下来,换上一套更克制、更可被验证、也更愿意被长期审视的设计。今天,当一款 ex vivo 造血干细胞基因疗法能被谈论”安全”二字时,它底下垫着的,正是二十年前那几个孩子的名字。这份安全,是用最不该付出的代价换来的,也因此格外不容辜负。

参考文献

  1. Cavazzana-Calvo M, Hacein-Bey S, de Saint Basile G, et al. Gene therapy of human severe combined immunodeficiency (SCID)-X1 disease. Science. 2000;288(5466):669-72. DOI
  2. Hacein-Bey-Abina S, Von Kalle C, Schmidt M, et al. LMO2-associated clonal T cell proliferation in two patients after gene therapy for SCID-X1. Science. 2003;302(5644):415-9. DOI
  3. Howe SJ, Mansour MR, Schwarzwaelder K, et al. Insertional mutagenesis combined with acquired somatic mutations causes leukemogenesis following gene therapy of SCID-X1 patients. J Clin Invest. 2008;118(9):3143-50. DOI
  4. Hacein-Bey-Abina S, Garrigue A, Wang GP, et al. Insertional oncogenesis in 4 patients after retrovirus-mediated gene therapy of SCID-X1. J Clin Invest. 2008;118(9):3132-42. DOI
  5. Hacein-Bey-Abina S, Hauer J, Lim A, et al. Efficacy of gene therapy for X-linked severe combined immunodeficiency. N Engl J Med. 2010;363(4):355-64. DOI
  6. Zhou S, Mody D, DeRavin SS, et al. A self-inactivating lentiviral vector for SCID-X1 gene therapy that does not activate LMO2 expression in human T cells. Blood. 2010;116(6):900-8. DOI
  7. Aiuti A, Biasco L, Scaramuzza S, et al. Lentiviral hematopoietic stem cell gene therapy in patients with Wiskott-Aldrich syndrome. Science. 2013;341(6148):1233151. DOI
  8. Hacein-Bey-Abina S, Pai SY, Gaspar HB, et al. A modified γ-retrovirus vector for X-linked severe combined immunodeficiency. N Engl J Med. 2014;371(15):1407-17. DOI

There is a class of baby boys born with no immune system at all. A broken gene called IL2RG leaves them unable to make mature T cells and NK cells, so that any ordinary infection can be lethal; in medicine it is called X-linked SCID (SCID-X1). More than two decades ago, a group of physicians used gene therapy for the first time to pull such children back from certain death — these were among the first patients in human history to be genuinely cured by a gene therapy. Then, a few years later, some of those children developed leukemia. The very key that let them live had, at the same time, planted another seed in their genomes. This is the heaviest lesson gene therapy ever learned, and the starting point of every promise of “safety” we make today.

First Light

The story has to begin with a beautiful success. In 2000, Cavazzana-Calvo, Fischer, and colleagues reported in Science a trial at the Necker Children’s Hospital in Paris: they took the SCID-X1 patients’ own CD34⁺ hematopoietic stem cells out of their bodies, used a γ-retroviral vector derived from the Moloney murine leukemia virus to deliver a normal copy of the γc (IL2RG) gene into the cells ex vivo, and then reinfused them. Ten months later, two patients had T cells and NK cells expressing γc, in numbers and function close to those of healthy children the same age. A congenital defect that would otherwise have been fatal had been corrected by the simple act of “putting one gene back.”

At the time this was earth-shaking. For decades before, gene therapy had been more promise than result; and these two children were the first incontrovertible, living proof that “putting one normal gene back into hematopoietic stem cells can rebuild an entire immune system.” The trial went on: of the ten baby boys treated, nine rebuilt their immunity, left the sterile isolation chamber, and went home to live ordinary children’s lives. Another team in London (Gaspar, Thrasher, and colleagues) independently ran a nearly identical trial and likewise saw immune recovery. For a while the field was steeped in optimism.

The Shadow

The turn came around 2002. In 2003, Hacein-Bey-Abina and colleagues reported in Science a result no one wanted to see: among that Paris cohort, the two youngest children, nearly three years after treatment, developed an uncontrolled clonal proliferation of mature T cells (judged premalignant at the time) — the diagnosis of leukemia was something only confirmed in the subsequent follow-up. When the researchers asked why, they found a striking commonality between the two leukemic clones: the γ-retroviral vector used to carry the gene had, in each case, integrated right next to the promoter of a proto-oncogene called LMO2, and the enhancer the virus carried acted like a loudspeaker set down in the wrong place, forcibly switching on the LMO2 that should have stayed silent in mature T cells. The vector had not delivered the gene to a “safe” spot; it had landed beside a switch that, once woken, would drive the cell toward cancer.

This was not an isolated case. As follow-up continued, the bad news kept coming. In 2008, Hacein-Bey-Abina and colleagues updated in the Journal of Clinical Investigation: of the nine successfully treated patients in the Paris trial, four had by then developed T-cell leukemia, with onset between 31 and 68 months after treatment. Around the same time, Howe, Thrasher, and colleagues of the London team reported, in the same Journal of Clinical Investigation, that their trial had likewise produced a case of T-cell acute lymphoblastic leukemia caused by insertional mutagenesis. Two independent centers, the same vector, the same LMO2 — the nature of the thing had changed: this was no longer bad luck, but a reproducible risk carried by the technology itself. The field entered its darkest hour, and many ongoing trials were halted.

The Power of Integration, and Its Original Sin

To understand this catastrophe, you first have to understand the one thing a retroviral vector depends on to work: integration. This class of virus permanently stitches the genes it carries into the host cell’s genome, and it is precisely this capacity to “stitch in” that lets the therapeutic gene be passed down through generations of hematopoietic stem cells, supplying the product long-term — that is where its power lies. But integration has a price: where in the genome it lands was, in the early days, impossible to control. Worse, the γ-retroviral vector has a systematic preference, tending to insert near the start regions of actively transcribed genes — and lurking among those spots are quite a few proto-oncogenes that regulate proliferation. The vector’s long terminal repeats (LTRs) carry a powerful enhancer, and once it lands next door to a proto-oncogene it can keep that gene switched on for the long term. This is its original sin — the mechanism that confers the therapeutic effect and the mechanism that triggers cancer are one and the same.

But the full picture of the catastrophe is more complex than “a single misdirected integration,” and more thought-provoking. In their 2008 paper, Howe and colleagues took the leukemic clone’s genome completely apart, and found that the activation of LMO2 was only the first step: what actually pushed the cell across the threshold into malignancy was a whole series of acquired mutations — a point mutation conferring gain-of-function activation of NOTCH1, loss of the locus carrying the tumor-suppressor gene CDKN2A, and a chromosomal translocation joining the T-cell receptor region to STIL-TAL1. In other words, the vector’s insertion was the necessary first blow, but not the sufficient whole; the leukemia was the joint product of “insertional activation plus multiple acquired hits.” This understanding was corroborated in Hacein-Bey-Abina’s JCI paper too: across different patients’ leukemic clones, besides LMO2 there were vectors inserted near proto-oncogenes such as BMI1 and CCND2. Genotoxicity is not a simple “hit or miss,” but a web ignited by the vector and kept burning by random mutation.

Here one must honestly state something that the catastrophe should not be allowed to obscure: even so, the gene therapy still meant something for these children. Of the four children who developed leukemia, three achieved durable remission through chemotherapy, and one, sadly, died; and after chemotherapy, their modified, polyclonal T-cell populations grew back and their immune function was maintained. In 2010, Hacein-Bey-Abina and colleagues gave nearly a decade of long-term follow-up in the New England Journal of Medicine: of the nine patients, eight were alive, with a median follow-up of about nine years, and in most of them the immunodeficiency remained continuously corrected. This is a heavy and complicated balance sheet — it records “cure,” it records “leukemia,” and it records “death.” The field has no right to gloss over any single column of it.

Taking the Loudspeaker Apart

What that price bought was a thorough rewrite of the safety paradigm. Since the culprit was the enhancer in the LTR that activated its neighbors, the most direct countermeasure was: take it out. This gave rise to the self-inactivating (SIN) vector — the design deletes the viral enhancer/promoter sequences from the LTR, so that the integrated provirus loses its ability to transcribe itself and to “shout” outward; the therapeutic gene is instead driven by a mild, tissue-appropriate internal promoter within the vector, rather than borrowing that crude, strong enhancer. Some designs additionally add an insulator sequence, like building a soundproof wall around the integration site, to further block its interference with neighboring genes. Making the vector “self-inactivate” is the single most central lesson learned from that catastrophe.

At the same time, the field began systematically replacing the γ-retroviral vector with the lentiviral vector. Lentivirus can integrate too and can supply the product long-term as well, but its integration preference is different — it is less fond of the high-risk positions at proto-oncogene start regions; combined with the SIN design and an internal promoter, its safety margin is markedly higher. In 2010, Zhou and colleagues reported in Blood a SIN lentiviral vector custom-built for SCID-X1: using the EF1α internal promoter to drive γc and embedding an insulator, it no longer transactivated LMO2 in an in vitro Jurkat LMO2-activation assay, and did not cause tumors in mice — turning “no longer waking LMO2” into a metric you could verify at the bench.

These changes were not paper exercises; they really were carried back to the bedside. In 2014, Hacein-Bey-Abina, Thrasher, Williams, and colleagues reported in the New England Journal of Medicine an international, multicenter trial: switching to a SIN-type γ-retroviral vector (with the enhancer deleted) to treat nine SCID-X1 baby boys, the efficacy was preserved — most patients achieved functional recovery of their T cells; and when the researchers analyzed the vector integration sites in these children, they found that their clustering near lymphoid proto-oncogenes such as LMO2 and MECOM was markedly reduced compared with the early trials. Around the same year, in another immunodeficiency, Wiskott-Aldrich syndrome, Aiuti and colleagues reported in Science that in patients treated with a self-inactivating (SIN) lentiviral vector, hematopoiesis was highly polyclonal and multilineage, the integrations were not selectively enriched near proto-oncogenes, and no abnormal clonal expansion was seen over 20 to 32 months of follow-up. The center of gravity of the safety paradigm thus shifted from “praying the vector doesn’t land in the wrong place” to “designing it so that even if it lands wrong it won’t ignite anything,” supplemented by integration-site analysis and long-term follow-up that keep the consequences of every integration under continuous microscopic watch.

The Part Not Yet Told

One must admit, with restraint, that the risk has not been “solved,” only greatly reduced. That 2014 NEJM paper was very honest in its conclusion: whether this improved vector, in the long term, has entirely escaped the possibility of causing leukemia remains unknown — safety is a claim that has to be confirmed gradually over decades of follow-up, not a conclusion that a single experiment can settle for good. Integration is always a probabilistic event; the SIN design and the lentivirus push the probability of a bad outcome down, but have not, and cannot, push it to zero.

The deeper question is that, of that catastrophe back then, how much is to be blamed on the vector’s enhancer, how much on the delivered γc gene itself, or on the T-cell selective pressure peculiar to a disease like SCID-X1, is still debated to this day — after all, some other trials for other diseases that used the same γ-retroviral vector did not develop leukemia at the same frequency. This reminds us that genotoxicity has never been about the vector as a single variable; it is also entangled with the transgene, the disease context, and even the proliferative environment of the treated cells. And “will an integration activate some proto-oncogene,” “how are large deletions and chromosomal rearrangements to be caught in time” — these finer safety questions form what the later pieces in this board will go on to pursue.

Closing

Looking back, this stretch of SCID-X1 history is gene therapy’s rite of passage from “can it be cured” to “can it be cured safely.” It taught the whole field one thing in the most agonizing way possible: the power of integration that lets a therapeutic gene take up permanent residence, and the risk of integration that lets a cell turn cancerous, are two sides of one coin; real progress lies not in avoiding integration, but in learning to tame it — taking the strong enhancer off the vector that could wound bystanders, and replacing it with a design that is more restrained, more verifiable, and more willing to be scrutinized over the long term. Today, when an ex vivo hematopoietic stem cell gene therapy can be spoken of with the word “safe,” what underlies it is precisely the names of those few children from twenty years ago. This safety was bought at a price that never should have had to be paid, and for that reason it deserves all the more not to be squandered.

References

  1. Cavazzana-Calvo M, Hacein-Bey S, de Saint Basile G, et al. Gene therapy of human severe combined immunodeficiency (SCID)-X1 disease. Science. 2000;288(5466):669-72. DOI
  2. Hacein-Bey-Abina S, Von Kalle C, Schmidt M, et al. LMO2-associated clonal T cell proliferation in two patients after gene therapy for SCID-X1. Science. 2003;302(5644):415-9. DOI
  3. Howe SJ, Mansour MR, Schwarzwaelder K, et al. Insertional mutagenesis combined with acquired somatic mutations causes leukemogenesis following gene therapy of SCID-X1 patients. J Clin Invest. 2008;118(9):3143-50. DOI
  4. Hacein-Bey-Abina S, Garrigue A, Wang GP, et al. Insertional oncogenesis in 4 patients after retrovirus-mediated gene therapy of SCID-X1. J Clin Invest. 2008;118(9):3132-42. DOI
  5. Hacein-Bey-Abina S, Hauer J, Lim A, et al. Efficacy of gene therapy for X-linked severe combined immunodeficiency. N Engl J Med. 2010;363(4):355-64. DOI
  6. Zhou S, Mody D, DeRavin SS, et al. A self-inactivating lentiviral vector for SCID-X1 gene therapy that does not activate LMO2 expression in human T cells. Blood. 2010;116(6):900-8. DOI
  7. Aiuti A, Biasco L, Scaramuzza S, et al. Lentiviral hematopoietic stem cell gene therapy in patients with Wiskott-Aldrich syndrome. Science. 2013;341(6148):1233151. DOI
  8. Hacein-Bey-Abina S, Pai SY, Gaspar HB, et al. A modified γ-retrovirus vector for X-linked severe combined immunodeficiency. N Engl J Med. 2014;371(15):1407-17. DOI