Yang Liu

← Primer

HSC · Healing Source Code

CHIP——每个人的血液里都长着克隆,而基因治疗必须在这片森林里动手 CHIP: everyone's blood grows clones—and gene therapy has to work inside that forest

2026-07-16 · CHIP, 克隆造血, 基因治疗安全性, 造血干细胞, 归因

我们习惯把血液想象成一条均质的河流:骨髓源源不断地送出细胞,它们彼此相同、共同流淌。这个图景在人年轻时大致是对的。但随着年龄增长,河流会悄悄变成一片森林——某一个造血干细胞偶然获得一个突变,这个突变让它比邻居稍微能干一点,于是它的后代一点一点占据了越来越大的份额。携带这支克隆的人体检一切正常,血常规漂亮,自己毫不知情。直到 2014 年,两组人从本来为别的目的攒下的测序数据里,把这片森林打捞了出来。而这件事,恰恰给正在把自体造血干细胞取出、改造、再输回去的基因治疗,提出了一个很不舒服的问题:我们动手的那片组织,从来就不是一张白纸。

从别人的数据里打捞出来的发现

2014 年 11 月,《New England Journal of Medicine》同期发表了两篇论文,它们用的都不是为这个问题专门收集的样本——这正是故事最耐人寻味的地方。

Jaiswal 等人分析了 17,182 人外周血细胞 DNA 的 whole-exome sequencing 数据,这些人并未因血液异常而被选中。他们在 160 个已知在血液肿瘤中反复突变的基因里找体细胞突变,结果是:40 岁以下几乎检不出,频率随年龄陡然上升——70 至 79 岁占 9.5%(2300 人中的 219 人),80 至 89 岁为 11.7%(317 人中的 37 人),90 至 108 岁达 18.4%(103 人中的 19 人)。绝大多数突变落在三个基因上:DNMT3A、TET2、ASXL1。这些变异的中位 VAF 是 0.09——它们只存在于一部分血细胞里,是后天获得而非遗传而来。

几乎同时,Genovese 等人用 12,380 名瑞典人的 whole-exome sequencing 数据照出了几乎一样的图景——那批数据原本是为精神分裂症与双相障碍的遗传学研究测的。65 岁以上有 10% 的人带着可检出的 clonal hematopoiesis,50 岁以下只有 1%,最常见的仍是那三个基因。

一个现象能被两批完全不同目的、不同人群的数据同时照出来,通常说明它一直就在那里,只是过去没有人往那个方向看。次年,Steensma 等人在《Blood》上给了它一个名字:clonal hematopoiesis of indeterminate potential,CHIP——当一个人携带着驱动克隆扩张的体细胞突变,却没有 cytopenia、没有发育异常、也没有血液肿瘤时,他处在的就是这样一个”潜能未定”的状态。他们同时给出了一条可操作的门槛:作为一个工作定义,外周血中的突变等位基因分数须达到 ≥2% 才算 CHIP——因为只要测得足够深,几乎每个人身上都能找出突变来;作者也同时申明,这条线可能随着更多人群数据而修订。这个命名本身包含着一份克制:大多数带着它的人,终其一生都不会发展成骨髓增生异常综合征。

一个大得吓人的相对风险,和一个小得多的绝对风险

这些克隆意味着什么风险,两篇论文给出的数字非常一致,也非常需要被小心地读。

在 Jaiswal 的队列里,携带体细胞突变的人后续发生血液肿瘤的风险是不携带者的 11.1 倍(95% CI 3.9–32.6);Genovese 那边独立地给出 12.9 倍(95% CI 5.8–28.7)。这是个吓人的倍数。但 Jaiswal 那篇紧接着写下了另一半:研究期内,携带突变的人里约 4% 真的发展成了血液肿瘤,折算下来大约每年 0.5%。风险倍数很大,是因为基线本来极低。

真正让这个数字变得有用的,是它并不均匀。当 VAF 达到 0.10 或更高——这支克隆已占据相当一部分血细胞——风险升到近 50 倍(HR 49,95% CI 21–120),年风险约 1%;那些后来真的得了血液肿瘤的人,采样当时的平均 VAF 也显著更高(25.2% vs 12.0%)。决定风险的不只是”有没有克隆”,更是”这支克隆有多大”。Genovese 那边则给出了时间维度的证据:该队列中约 42% 的血液肿瘤,发生在采样时就已存在克隆的人身上,且采样早于确诊 6 个月以上;两例急性髓系白血病患者的骨髓活检显示,肿瘤正是从更早的那支克隆里长出来的。克隆不是白血病,但白血病常常是从克隆里长出来的。

故事本可以就此打住,作为一个纯粹的血液学话题。但 Jaiswal 那篇论文里还藏着一个当时看来古怪的关联:携带突变的人,全因死亡(HR 1.4)、冠心病(HR 2.0)与缺血性卒中(HR 2.6)的风险都升高了。2017 年,Jaiswal 等人回到《New England Journal of Medicine》把它坐实:在两个前瞻性队列的巢式病例对照分析中,CHIP 携带者的冠心病风险是非携带者的 1.9 倍(95% CI 1.4–2.7);在两个早发心肌梗死的回顾性队列中,风险高达 4.0 倍(95% CI 2.4–6.7)。最关键的一步不在人身上——他们把 Tet2 缺失的骨髓移植进易发动脉粥样硬化的小鼠体内,这些小鼠长出了更大的斑块,而 Tet2⁻ᐟ⁻ 的巨噬细胞高表达一系列促动脉粥样硬化的趋化因子和细胞因子。一个从造血干细胞里长出来的克隆,通过它派出的髓系后代,把炎症带到了血管壁上。CHIP 至此不再只是一个癌前状态。

三重含义,落在基因治疗身上

现在把镜头转向自体造血干细胞基因治疗的流程:动员、采集 CD34⁺ 细胞、体外转导或编辑、清髓预处理、回输、engraftment、然后是长达十几年的随访。CHIP 在这条流水线的每一段上都留下了阴影,而且是三种不同性质的阴影。

最先的一重落在起点上:采集到的细胞里,可能本来就有。基因治疗用的是患者自己的细胞,而患者自己的血液系统同样会随年龄长出克隆——甚至可能长得更早。2021 年,Pincez 等人在《Blood》上用 whole-exome sequencing 数据比较了镰刀型贫血患者与对照:在 4 个队列中找到 15 名携带 clonal hematopoiesis 的镰刀型贫血患者,校正年龄、性别与外显子捕获方法后,镰刀型贫血本身与 CHIP 的出现独立相关(OR 13.5,95% CI 3.1–41.9);到 50 岁时,检出克隆性造血的概率在患者中是 7.1%,对照仅 0.7%。作者自己明确提醒这不该被当成定论——样本量小,各数据集测序深度不齐。但方向是清楚的:一个终生处在慢性溶血、反复炎症和造血高压之下的骨髓,克隆出现的时钟可能走得更快。而这些患者,恰恰就是基因治疗的目标人群。

第二重则落在流程本身——它其实是一个筛子。清髓把绝大多数造血干细胞清空,再让少数细胞去重建整个系统,这是一场极端的适者生存,而 CHIP 克隆恰恰在”适应力”上略胜一筹。2019 年,Ortmann 等人在《Cell Reports》上追踪了 81 例因实体瘤或淋巴系统疾病接受自体造血干细胞移植的患者:移植后 CHIP 的发生率高达 22%,平均 VAF 达 10.7%。最要紧的是下一句——大多数这些突变在回输的移植物里就已经存在了,只是 VAF 更低。它们不是移植后新生的,而是被移植这件事放大的。

那么基因治疗会不会做同样的事?2023 年,Spencer Chapman 等人在《Nature Medicine》上用 whole-genome sequencing 追踪了 6 名镰刀型贫血患者治疗前后的造血干细胞。好消息是:治疗前的克隆谱系树高度多克隆;治疗后,无论被修改还是未被修改的细胞,都没有找到克隆性扩张。但与髓系肿瘤或 clonal hematopoiesis 相关的潜在驱动突变(尤以 DNMT3A 和 EZH2 突变的克隆为著)频率升高了——在被修改和未被修改的细胞里都升高。这个细节值得停一下:选择压力并不认识载体。它筛的是细胞的适应力,而携带驱动突变的细胞本来就更能扛。

而第三重最麻烦,因为它伤的不是安全本身,而是我们判断安全的能力:如果一名接受了基因治疗的患者若干年后得了髓系肿瘤,这件事该记在谁头上?

谁干的:归因这门手艺

这不是一个理论问题,它已经反复发生过。2020 年,Hsieh 等人在《Blood Advances》上报告了一名接受 LentiGlobin 治疗镰刀型贫血后发生骨髓增生异常综合征的患者,论文的意图写在标题里——这例 MDS 与 lentiviral vector 无关,他们展示的是如何把 insertional oncogenesis 排除掉。2022 年的案例更能说明归因的难度:Goyal 等人在《New England Journal of Medicine》上报告了一名女性患者在接受 LentiGlobin 约 5.5 年后发生急性髓系白血病;外周血分析显示,白血病母细胞里确实含有 BB305 载体的插入位点。任何人看到这一句,第一反应都会是”载体干的”。但因果调查给出了相反的结论:考虑到插入位点的位置、母细胞里极低的转基因表达、以及对周围基因表达没有影响,这例白血病不太可能由载体插入引起;与此同时,诊断后在母细胞中检出了若干易感急性髓系白血病的体细胞突变。作者的判断因而落在一个复合的解释上:镰刀型贫血这一基础疾病本身、移植过程、以及治疗后疾病控制不足的风险,共同抬高了血液恶性肿瘤的可能。

把它和反面并排,归因这门手艺的逻辑就浮现了。2024 年,Duncan 等人在《New England Journal of Medicine》上报告:接受 eli-cel 治疗脑型肾上腺脑白质营养不良的 67 名患者中有 7 人发生血液肿瘤;在 6 名有数据的患者里,优势克隆携带的载体插入落在 MECOM–EVI1(5 例)或 PRDM16(1 例)上。插入位点、优势克隆、已知致癌基因,三者对齐了——那是载体干的(这条线索属于插入突变的故事,留待那一篇细说)。

于是标准变得清楚:能把责任判给载体,靠的不是”母细胞里有载体”这个事实,而是插入位点是否落在致癌基因上、优势克隆是否由这次插入定义、转基因表达是否真的扰动了邻近基因。这些条件都不满足时,剩下的嫌疑人——本底克隆、清髓用的烷化剂、疾病本身——就得被一一称重。

这也解释了为什么”背景 CHIP”对整个领域首先是一个方法学问题。判断一支治疗后出现的克隆是不是治疗造成的,唯一可靠的办法是知道它治疗前在不在那里、有多大——而这个信息只能来自采集时留下的基线样本。一旦没留,后面十几年的随访再密集,也补不回来。Spencer Chapman 那篇论文之所以有说服力,正因为它拿到了治疗前后配对的样本;Ortmann 之所以能说”这些突变在移植物里就已经存在”,也是同一个道理。基线不是随访的附属品,它是随访的前提。

还没有答案的部分

诚实地说,这个领域眼下的未解比已解要多。VAF 多大才算需要警惕?0.10 这条线来自一个观察性队列里的风险分层,不是一条为基因治疗患者划的安全阈值。哪些基因该被担心?DNMT3A 突变的克隆和 TP53 突变的克隆虽然同被装进 CHIP 这个筐,其生物学与危险性并不相同(这个分野本身需要单独一篇来讲)。清髓、编辑、疾病本身各自贡献了多少风险?在没有大规模、带基线的长期队列之前,这些权重只能靠个案的因果调查一次次拼凑。而当基因治疗从体外走向体内——不再有采集、不再有那管可以冻起来的基线细胞——“治疗前这支克隆在不在”这个问题,将会以一种更棘手的形式重新提出来。

回到开头那片森林。CHIP 教给这个领域最本质的一件事,或许是它改写了”正常人的骨髓”这个默认前提:那里不是一片均质的、等待被写字的白纸,而是一片已经长了年轮、有强有弱、彼此竞争的树林。基因治疗做的事,是把其中一批树取出来、改造、再种回一片刚被清空的土地上——这个动作本身不制造克隆,但它会决定哪些克隆能活下来、长多大。理解这一点不是为了让人对基因治疗却步,而是为了让长期随访里的每一个异常信号,都能被放回它本来的坐标里去读。

参考文献

  1. Jaiswal S, et al. Age-related clonal hematopoiesis associated with adverse outcomes. N Engl J Med. 2014;371(26):2488-2498. DOI
  2. Genovese G, et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N Engl J Med. 2014;371(26):2477-2487. DOI
  3. Steensma DP, et al. Clonal hematopoiesis of indeterminate potential and its distinction from myelodysplastic syndromes. Blood. 2015;126(1):9-16. DOI
  4. Jaiswal S, et al. Clonal hematopoiesis and risk of atherosclerotic cardiovascular disease. N Engl J Med. 2017;377(2):111-121. DOI
  5. Ortmann CA, et al. Functional dominance of CHIP-mutated hematopoietic stem cells in patients undergoing autologous transplantation. Cell Rep. 2019;27(7):2022-2028.e3. DOI
  6. Pincez T, et al. Clonal hematopoiesis in sickle cell disease. Blood. 2021;138(21):2148-2152. DOI
  7. Spencer Chapman M, et al. Clonal selection of hematopoietic stem cells after gene therapy for sickle cell disease. Nat Med. 2023;29(12):3175-3183. DOI
  8. Hsieh MM, et al. Myelodysplastic syndrome unrelated to lentiviral vector in a patient treated with gene therapy for sickle cell disease. Blood Adv. 2020;4(9):2058-2063. DOI
  9. Goyal S, et al. Acute myeloid leukemia case after gene therapy for sickle cell disease. N Engl J Med. 2022;386(2):138-147. DOI
  10. Duncan CN, et al. Hematologic cancer after gene therapy for cerebral adrenoleukodystrophy. N Engl J Med. 2024;391(14):1287-1301. DOI

We are used to picturing blood as a homogeneous river: the bone marrow sends out a steady stream of cells, all alike, all flowing together. That picture is roughly right while we are young. But with age, the river quietly turns into a forest—one hematopoietic stem cell happens to acquire a mutation, that mutation makes it a little more capable than its neighbors, and so its descendants come to occupy a larger and larger share. The person carrying that clone has a perfectly normal check-up, a beautiful blood count, and no idea. Then, in 2014, two groups dredged this forest up out of sequencing data that had been amassed for entirely different purposes. And this turns out to pose a rather uncomfortable question to gene therapy—which is in the business of taking autologous hematopoietic stem cells out, modifying them, and putting them back: the tissue we operate on was never a blank page.

A discovery dredged up from other people’s data

In November 2014, the New England Journal of Medicine published two papers in the same issue, and neither used samples collected for this question—which is precisely what makes the story so intriguing.

Jaiswal and colleagues analyzed whole-exome sequencing data from the peripheral-blood-cell DNA of 17,182 people who had not been selected for any blood abnormality. Looking for somatic mutations across 160 genes known to be recurrently mutated in hematologic cancers, they found: essentially nothing detectable below age 40, and a frequency that climbs steeply with age—9.5% among those aged 70 to 79 (219 of 2300), 11.7% among those 80 to 89 (37 of 317), and 18.4% among those 90 to 108 (19 of 103). The great majority of mutations fell in three genes: DNMT3A, TET2, ASXL1. The median VAF of these variants was 0.09—they were present in only a fraction of blood cells, acquired rather than inherited.

Almost simultaneously, Genovese and colleagues used whole-exome sequencing data from 12,380 Swedes to reveal an almost identical picture—data originally generated for genetic studies of schizophrenia and bipolar disorder. Ten percent of those over 65 carried detectable clonal hematopoiesis, versus only 1% of those under 50, and the most common genes were still those same three.

When a phenomenon shows up in two sets of data collected for completely different purposes, in different populations, it usually means it had been there all along and nobody had thought to look in that direction. The following year, Steensma and colleagues gave it a name in Blood: clonal hematopoiesis of indeterminate potential, CHIP—when someone carries a somatic mutation that drives clonal expansion but has no cytopenia, no dysplasia and no hematologic cancer, that is the state of “indeterminate potential” they are in. They also offered an operational threshold: as a working definition, the mutant allele fraction in peripheral blood must be ≥2% to count as CHIP—because with deep enough sequencing, a mutation can be found in virtually anyone; the authors also stated that this line might need revising as more population data accumulate. The naming itself carries a measure of restraint: most people who have it will never, in their whole lives, go on to develop myelodysplastic syndrome.

A frighteningly large relative risk, and a much smaller absolute one

As for what risk these clones actually represent, the numbers the two papers give are strikingly consistent—and very much need to be read carefully.

In Jaiswal’s cohort, people carrying somatic mutations had an 11.1-fold risk of subsequently developing a hematologic cancer compared with non-carriers (95% CI 3.9–32.6); Genovese’s side independently arrived at 12.9-fold (95% CI 5.8–28.7). That is a frightening multiple. But Jaiswal’s paper immediately writes down the other half: during the study period, about 4% of the mutation carriers actually went on to develop a hematologic cancer, which works out to roughly 0.5% per year. The risk multiple is large because the baseline was extremely low to begin with.

What really makes this number useful is that it is not uniform. When the VAF reaches 0.10 or higher—the clone has taken over a substantial share of the blood cells—the risk rises to nearly 50-fold (HR 49, 95% CI 21–120), an annual risk of about 1%; and those who did go on to develop a hematologic cancer had a significantly higher average VAF at the time of sampling (25.2% vs 12.0%). What determines risk is not just “is there a clone” but “how big is the clone.” Genovese’s side supplied the evidence along the time axis: about 42% of hematologic cancers in that cohort occurred in people in whom a clone was already present at sampling, with sampling more than 6 months before diagnosis; bone marrow biopsies from two patients with acute myeloid leukemia showed that the cancer had in fact grown out of that earlier clone. A clone is not leukemia, but leukemia often grows out of a clone.

The story could have stopped there, as a purely hematologic topic. But Jaiswal’s paper also harbored an association that looked odd at the time: mutation carriers had elevated risks of all-cause death (HR 1.4), coronary heart disease (HR 2.0) and ischemic stroke (HR 2.6). In 2017, Jaiswal and colleagues returned to the New England Journal of Medicine to nail it down: in nested case-control analyses of two prospective cohorts, CHIP carriers had a 1.9-fold risk of coronary heart disease compared with non-carriers (95% CI 1.4–2.7); in two retrospective cohorts of early-onset myocardial infarction, the risk was as high as 4.0-fold (95% CI 2.4–6.7). The most decisive step was not in humans—they transplanted Tet2-deficient bone marrow into atherosclerosis-prone mice, and those mice developed larger plaques, while Tet2⁻ᐟ⁻ macrophages highly expressed a series of pro-atherosclerotic chemokines and cytokines. A clone that grew out of a hematopoietic stem cell had, through the myeloid descendants it dispatched, carried inflammation to the vessel wall. From that point on, CHIP was no longer merely a precancerous state.

Three implications, landing on gene therapy

Now turn the camera to the workflow of autologous hematopoietic stem cell gene therapy: mobilization, collection of CD34⁺ cells, ex vivo transduction or editing, myeloablative conditioning, reinfusion, engraftment, and then a follow-up lasting more than a decade. CHIP casts a shadow over every segment of this assembly line—and three shadows of different kinds.

The first lands at the starting point: it may already be in the cells you collect. Gene therapy uses the patient’s own cells, and the patient’s own blood system likewise grows clones with age—perhaps even earlier. In 2021, Pincez and colleagues used whole-exome sequencing data in Blood to compare sickle cell disease patients with controls: across 4 cohorts they found 15 sickle cell disease patients carrying clonal hematopoiesis, and after adjusting for age, sex and exome capture method, sickle cell disease itself was independently associated with the appearance of CHIP (OR 13.5, 95% CI 3.1–41.9); by age 50, the probability of detecting clonal hematopoiesis was 7.1% in patients versus only 0.7% in controls. The authors explicitly caution that this should not be taken as settled—the sample size is small, and sequencing depth is uneven across the datasets. But the direction is clear: a marrow that spends a lifetime under chronic hemolysis, recurrent inflammation and hematopoietic pressure may run its clock toward clonality faster. And these patients are precisely the target population for gene therapy.

The second lands on the workflow itself—which is, in effect, a sieve. Myeloablation empties out the vast majority of hematopoietic stem cells and then leaves a small number to rebuild the entire system; this is survival of the fittest at its most extreme, and CHIP clones happen to have a slight edge in “fitness.” In 2019, Ortmann and colleagues followed 81 patients in Cell Reports who underwent autologous hematopoietic stem cell transplantation for solid tumors or lymphoid disease: the incidence of CHIP after transplantation was as high as 22%, with a mean VAF of 10.7%. The most important part is the next sentence—most of these mutations were already present in the reinfused graft, just at a lower VAF. They were not newly created after transplantation; they were amplified by the act of transplantation.

So does gene therapy do the same thing? In 2023, Spencer Chapman and colleagues used whole-genome sequencing in Nature Medicine to track the hematopoietic stem cells of 6 sickle cell disease patients before and after treatment. The good news: the clonal phylogenies before treatment were highly polyclonal; after treatment, no clonal expansion was found in either modified or unmodified cells. But the frequency of putative driver mutations associated with myeloid neoplasms or clonal hematopoiesis (most notably clones with DNMT3A and EZH2 mutations) had risen—risen in both modified and unmodified cells. This detail is worth pausing on: selective pressure does not recognize the vector. What it sieves for is the cell’s fitness, and cells carrying driver mutations were always better at holding up.

And the third is the most troublesome, because what it damages is not safety itself but our ability to judge safety: if a patient who received gene therapy develops a myeloid neoplasm some years later, whose account does that go on?

Who did it: the craft of attribution

This is not a theoretical question; it has happened repeatedly. In 2020, Hsieh and colleagues reported in Blood Advances a patient who developed myelodysplastic syndrome after LentiGlobin treatment for sickle cell disease, with the paper’s intent written into its title—this case of MDS was unrelated to the lentiviral vector, and what they demonstrated was how to rule insertional oncogenesis out. The 2022 case illustrates the difficulty of attribution even better: Goyal and colleagues reported in the New England Journal of Medicine a female patient who developed acute myeloid leukemia about 5.5 years after receiving LentiGlobin; peripheral blood analysis showed that the leukemic blasts did contain a BB305 vector insertion site. Anyone reading that sentence would react first with “the vector did it.” But the causality investigation reached the opposite conclusion: given the location of the insertion site, the very low transgene expression in the blasts, and the absence of any effect on the expression of surrounding genes, this leukemia was unlikely to have been caused by vector insertion; at the same time, several somatic mutations predisposing to acute myeloid leukemia were detected in the blasts after diagnosis. The authors’ judgment therefore settled on a composite explanation: the underlying sickle cell disease itself, the transplantation process, and the risk of inadequate disease control after treatment had together raised the likelihood of a hematologic malignancy.

Place that side by side with its opposite and the logic of the craft of attribution emerges. In 2024, Duncan and colleagues reported in the New England Journal of Medicine that among 67 patients who received eli-cel for cerebral adrenoleukodystrophy, 7 developed a hematologic cancer; in the 6 patients for whom data were available, the vector insertions carried by the dominant clone fell in MECOM–EVI1 (5 cases) or PRDM16 (1 case). Insertion site, dominant clone, known oncogene—all three lined up, and that was the vector’s doing (that thread belongs to the story of insertional mutagenesis, to be told in detail in that piece).

The standard thus becomes clear: what lets you assign responsibility to the vector is not the fact that “there is vector in the blasts,” but whether the insertion site falls in an oncogene, whether the dominant clone is defined by that insertion, and whether transgene expression really perturbed the neighboring genes. When these conditions are not met, the remaining suspects—background clones, the alkylating agents used for myeloablation, the disease itself—have to be weighed one by one.

This also explains why “background CHIP” is, for the whole field, first of all a methodological problem. The only reliable way to judge whether a clone that appears after treatment was caused by the treatment is to know whether it was there before treatment, and how big it was—and that information can only come from a baseline sample kept at the time of collection. Once it isn’t kept, no amount of intensive follow-up over the following decade or more can make it up. The reason Spencer Chapman’s paper is persuasive is precisely that it obtained paired samples from before and after treatment; the reason Ortmann could say “these mutations were already present in the graft” is the same. A baseline is not an accessory to follow-up; it is the precondition for it.

The parts that have no answers yet

Honestly, in this field the unresolved currently outnumbers the resolved. How large a VAF warrants concern? The line at 0.10 comes from risk stratification within an observational cohort; it is not a safety threshold drawn for gene therapy patients. Which genes should we worry about? Clones with DNMT3A mutations and clones with TP53 mutations, though both packed into the basket called CHIP, do not share the same biology or the same danger (that distinction itself needs a piece of its own). How much risk do myeloablation, editing and the disease itself each contribute? Until there are large-scale, baseline-anchored long-term cohorts, these weights can only be pieced together, case by case, out of individual causality investigations. And when gene therapy moves from ex vivo to in vivo—no more collection, no more tube of baseline cells to freeze—the question “was this clone there before treatment” will pose itself again in a more intractable form.

Back to the forest we started with. The most fundamental thing CHIP taught this field is perhaps that it rewrote the default premise of “a normal person’s bone marrow”: that place is not a homogeneous blank page waiting to be written on, but a stand of trees that has already grown rings—some strong, some weak, all competing with one another. What gene therapy does is take a batch of those trees out, modify them, and plant them back into ground that has just been cleared—an act that does not itself create clones, but that does determine which clones survive and how large they grow. Understanding this is not meant to make anyone shrink back from gene therapy, but to let every abnormal signal in long-term follow-up be read back in the coordinates it actually belongs to.

References

  1. Jaiswal S, et al. Age-related clonal hematopoiesis associated with adverse outcomes. N Engl J Med. 2014;371(26):2488-2498. DOI
  2. Genovese G, et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N Engl J Med. 2014;371(26):2477-2487. DOI
  3. Steensma DP, et al. Clonal hematopoiesis of indeterminate potential and its distinction from myelodysplastic syndromes. Blood. 2015;126(1):9-16. DOI
  4. Jaiswal S, et al. Clonal hematopoiesis and risk of atherosclerotic cardiovascular disease. N Engl J Med. 2017;377(2):111-121. DOI
  5. Ortmann CA, et al. Functional dominance of CHIP-mutated hematopoietic stem cells in patients undergoing autologous transplantation. Cell Rep. 2019;27(7):2022-2028.e3. DOI
  6. Pincez T, et al. Clonal hematopoiesis in sickle cell disease. Blood. 2021;138(21):2148-2152. DOI
  7. Spencer Chapman M, et al. Clonal selection of hematopoietic stem cells after gene therapy for sickle cell disease. Nat Med. 2023;29(12):3175-3183. DOI
  8. Hsieh MM, et al. Myelodysplastic syndrome unrelated to lentiviral vector in a patient treated with gene therapy for sickle cell disease. Blood Adv. 2020;4(9):2058-2063. DOI
  9. Goyal S, et al. Acute myeloid leukemia case after gene therapy for sickle cell disease. N Engl J Med. 2022;386(2):138-147. DOI
  10. Duncan CN, et al. Hematologic cancer after gene therapy for cerebral adrenoleukodystrophy. N Engl J Med. 2024;391(14):1287-1301. DOI