Yang Liu

← Primer

HSC · Healing Source Code

多少才算够——疗效阈值不是一个技术数字,而是每种病自己的算术 How Much Is Enough — The Correction Threshold Is Not a Technical Number but Each Disease's Own Arithmetic

2026-07-25 · 基因治疗, 造血干细胞, 疗效阈值, 血红蛋白病, 基因编辑

ex vivo 的基因治疗一直在回避一个问题。细胞取到体外,在培养皿里几乎可以改得干干净净;清髓把骨髓腾空,回输进去的几乎全是改好的细胞,“要改多少才够”就被工程手段一路推成了”尽量多”。in vivo 把它摆回桌面:一针推进静脉,载体只摸得到一部分造血干细胞,真正改成的又只是其中一部分——没有筛选,没有腾空,改多少就是多少。多少才算够?

答案几乎与技术无关。同一台编辑机器、同样的效率,放到不同的病上,“够”的定义可以完全不同:决定门槛的不是编辑器,而是被编辑的细胞在体内要做什么事、它的后代能不能活得比没修好的邻居更久。

一个来自移植病房的数字

最早的硬数字来自异基因移植的失败。镰刀型贫血患者接受非清髓移植后,常停在一种不彻底的状态:供者细胞和自己的细胞长期共存,谁也没把谁挤掉。供者细胞少到什么程度,病会回来?

2017 年,Fitzhugh 等人在 Blood 上给出了来自真实病房的回答。他们随访的 67 名患者里,有 3 人在最初顺利 engraftment 之后,供者髓系 chimerism 缓慢下滑,最终疾病复发。把这三条曲线和复发时间点对上,门槛浮现出来:大约 20%。

更有意思的是随后那一步。他们建了个模型,只放进一个变量——供者与患者红细胞寿命的差异——然后问它够不够解释那 20%。够。这个阈值不是骨髓里的算术,而是血液里的算术;顺着模型往下推,红细胞寿命越短、网织红细胞越高的患者,需要的 chimerism 反而越低。

红细胞替干细胞把账放大了

红细胞产出之后命运并不对等:镰变的很快被破坏,正常的活满寿命。于是外周血里被校正的红细胞比例,会显著高于骨髓里的干细胞比例。2016 年,Altrock 等人在 Am J Hematol 上把这条放大关系写成了公式:用 Berkeley 镰刀型贫血小鼠与正常小鼠的骨髓做混合移植,量出正常红细胞比镰变红细胞长寿三到四倍,人身上这个倍数被估在 6 到 10 之间;代进模型,改掉 25% 的造血干细胞可能换来约 65% 的非镰变红细胞,改到 40% 则中位约 80%。这是模型加小鼠实验的推算,不是人体测量,但量级与病房那个 20% 对得上。

人身上最干净的观察来自血红蛋白病(地中海贫血与镰刀型贫血)的移植随访。2011 年,Andreani 等人在 Haematologica 上报告四名长期混合 chimerism 的患者:供者来源的有核细胞分别为 71%、46%、15% 和 25%,同日测到的供者来源成熟红细胞却是 100%、100%、73% 和 90%。同一组人随后在 Chimerism 上补充说明,这种持续的混合 chimerism 约见于十分之一的 β-地中海贫血移植患者,他们已不再需要输血。

百分比不是唯一的刻度

如果治疗思路不是换掉致病的血红蛋白,而是重新放出胎儿血红蛋白(HbF),阈值就得换一套刻度。最容易被误读的数字,恰恰是”HbF 占总血红蛋白的百分之多少”。

2014 年,Steinberg 等人在 Blood 上挑明:能否抑制镰变取决于每个红细胞里的 HbF 浓度,而非血样的平均值。他们估算,一个红细胞约需 10 pg 的 HbF 才算被保护住;总 HbF 为 20% 时,被保护的细胞可以少到 1%、也可以多到 24%,全看分布;只有当总 HbF 接近 30%,被保护细胞才有可能接近 70%。

所以这里的阈值是二维的:多少细胞被改到,以及每个被改到的细胞里装了多少 HbF。这解释了临床报告为什么执着于 pancellular 这个词:2021 年 Frangoul 等人在 N Engl J Med 上报告首批两名用 CRISPR-Cas9 破坏 BCL11A 红系 enhancer 的患者时,特意写明回升的 HbF 是 pancellular 分布的;同年,Esrick 等人在同一本刊上报告六名用 shmiR 在红系特异敲低 BCL11A 的患者,其中可完整评估的那些人把两个坐标都量了出来:F-cell 占未输血红细胞的 58.9% 到 93.6%,每个 F-cell 里 9.0 到 18.6 pg——正好落在 Steinberg 那条保护线附近或之上。

起点不同,门槛也不同。2018 年 Thompson 等人报告的 β-地中海贫血基因治疗里,13 名非 β0/β0 患者中 12 名彻底停输血;9 名 β0/β0 或双拷贝 IVS1-110 的患者却只把年输血量中位数降了 73%,仅 3 人停输。两组用的是同一个 LentiGlobin BB305 载体、同样的清髓 conditioning,不同的是患者自身残余的珠蛋白产出。

当选择压力站在你这边

在另一些病里,门槛不再由起始比例单独决定,而由被修好的细胞能不能扩张出来决定,因为身体会替你筛选。ADA-SCID 和 X 连锁 SCID 就属于这一类:修好的淋巴细胞前体能发育、增殖、存活,没修好的走不到终点。2002 年,Aiuti 等人在 Science 上把两面都写了出来:此前 ADA-SCID 的基因治疗疗效有限,正是因为植入的校正造血干细胞比例太小;他们的改动是在基因转移之外加上非清髓 conditioning,让校正细胞先站住脚,再由淋巴系的选择优势把它放大。两名患者由此得到持久的多谱系 engraftment,淋巴细胞计数上升、免疫功能改善。

最能说明选择压力有多重要的,反而是它缺席的地方。2010 年,Hacein-Bey-Abina 等人在 N Engl J Med 上总结九名接受 γ-retrovirus 基因治疗的 SCID-X1 患儿,其中八人存活,中位随访九年:被转导的 T 细胞最长 10.7 年后仍可检出,胸腺持续产出幼稚 T 细胞——但被转导的 B 细胞一个都没测到。同一次输注,T 细胞谱系轻松跨过门槛,B 细胞谱系没有,因为选择优势只作用在 T 细胞身上。(四人后来发生急性白血病、一人死亡,那是插入突变的代价,留到讲安全性时再说。)

还有一种情况是阈值自己往下走。Fanconi 贫血里被修复的造血干细胞不再受 DNA 交联损伤折磨,天然比邻居更能活。2019 年,Río 等人在 Nat Med 上报告一项不做任何 conditioning 的 FA-A 基因治疗研究:被校正的细胞在体内逐步扩张,插入位点分析显示它们是多能的(multipotent)造血干细胞,而这种重建优势并非载体插入激活了某个基因所致;基因标记最高的那几名患者,骨髓衰竭的进展被止住了。

没有人替你做选择的时候

慢性肉芽肿病(CGD)是反面教材:修好的中性粒细胞并不会因此活得更久,它只是在被感染的组织里能产生活性氧。阈值必须硬凑。1997 年,Malech 等人在 PNAS 上给五名 p47phox 型 CGD 成人回输了经逆转录病毒转导的自体 CD34⁺ 细胞,不做 conditioning:五人都测到了功能被纠正的粒细胞,峰值在三到六周,却只占外周血粒细胞的 0.004% 到 0.05%。基因确实在表达,但那个数量级离临床意义还很远。

差多远?X 连锁 CGD 的女性携带者因 X 染色体随机失活,体内正常中性粒细胞的比例天差地别,构成一场天然的剂量实验。2018 年,Marciano 等人在 J Allergy Clin Immunol 上给出了刻度:93 名有临床资料的携带者中,发生 CGD 型感染的那 14 人,氧化功能正常细胞比例中位数只有 8%;低于 10% 与感染强烈相关(比值比 99),放宽到低于 20% 仍然相关(比值比 12)。2020 年,Kohn 等人在 Nat Med 上报告的慢病毒 CGD 基因治疗正落在这条线上方:九名重症 X-CGD 患者(两人因既往合并症早期去世)中,七名存活者里六人在 12 个月时维持了 16% 到 46% 的氧化酶阳性中性粒细胞,存活者此后没有出现新的 CGD 相关感染。

有些病的阈值根本不是细胞比例

异染性脑白质营养不良(MLD)换了一种问法。缺的是一种可分泌的酶(ARSA),而酶能被邻居细胞摄取——cross-correction 意味着你要的不是每个细胞都被改到,而是足够的总产量和正确的去向。2013 年,Biffi 等人在 Science 上就已在三名症状前的晚婴型 MLD 患儿身上看到造血各谱系与脑脊液中的高水平酶表达;2022 年,Fumagalli 等人在 Lancet 上给出 29 名患者的长期结果:治疗两年后,晚婴型患者外周血单个核细胞的 ARSA 活性比他们自己的基线平均升高 18.7 倍,早少年型升高 5.7 倍;从治疗后三个月起,所有患者的酶活性都落在正常范围之内或之上,脑脊液里的酶活性平均在 6 到 12 个月内回到正常水平。要紧的与其说是被改到的细胞占多大比例,不如说是这些细胞能不能把酶的总量顶回正常区间、并送到该去的地方。

这类病还有第二个阈值,单位不是百分比而是时间:Fumagalli 等人写得很清楚,获益在症状出现前治疗的患者身上最为明显——已经发生的脱髓鞘不会因为酶补上来就自行修回去。

这对体内意味着什么

排在一起看,in vivo 面前的题就清晰了:它不必复刻 ex vivo 那种近乎彻底的编辑效率,只需跨过每种病自己那条线。而这条线并不在同一个高度:镰刀型贫血要的是两成上下,CGD 女性携带者的自然史把线划在一成以上;在那些身体会替你筛选的病里,门槛则不再由起始比例单独决定,而由被修好的细胞能不能扩张出来决定。选哪个适应症先做,就是在选要跨多高的栏。

但这条线很容易被量错。你测哪一种细胞,决定你看到哪个数字:Andreani 那四名患者里,外周血与骨髓的有核细胞、乃至 BFU-E 克隆,三种口径在同一名患者体内彼此一致(其中一人 15%),唯独成熟红细胞冲到 73%——差别不在取材部位,而在被测的谱系有没有经历过差别存活。另一种陷阱在时间轴上:如果编辑发生在寿命短的祖细胞而非真正的长期造血干细胞,早期数字会很好看,然后一路衰减。in vivo 没有回输前的产品放行检测,只能靠体内连续采样把这两种情形分开,而它们在第一个月里长得很像。阈值也不是常数:不再输血、不再发作危象、不再感染、器官损伤不再进展,是四条高度不同的线。Fitzhugh 那 20% 之所以有分量,正因为它是从病重新回来的过程里读出来的,而不是从一次横断面的相关性里算出来的。

再走一步就是下一个念头:如果阈值由”被修好的细胞能不能活得更好”决定,能不能人为给它们造一个优势?这正是 in vivo selection 想做的事,也是它风险的来源——那是另一篇的题目。

一条被疾病自己写下的线

回到那支推进静脉的针管。它的成败不由”能编辑多少”单独决定,而由”这个病要求多少”和”能编辑多少”之间的差额决定。前一半写在载体和编辑器里,是工程;后一半写在红细胞的寿命、淋巴细胞的扩张、中性粒细胞的职责和酶的扩散半径里,是生物学,而且早在任何人想到基因治疗之前就已经写好了。

参考文献

  1. Fitzhugh CD, Cordes S, Taylor T, et al. At least 20% donor myeloid chimerism is necessary to reverse the sickle phenotype after allogeneic HSCT. Blood. 2017;130(17):1946-1948. DOI
  2. Altrock PM, Brendel C, Renella R, Orkin SH, Williams DA, Michor F. Mathematical modeling of erythrocyte chimerism informs genetic intervention strategies for sickle cell disease. Am J Hematol. 2016;91(9):931-937. DOI
  3. Andreani M, Testi M, Gaziev J, et al. Quantitatively different red cell/nucleated cell chimerism in patients with long-term, persistent hematopoietic mixed chimerism after bone marrow transplantation for thalassemia major or sickle cell disease. Haematologica. 2011;96(1):128-133. DOI
  4. Andreani M, Testi M, Battarra M, Lucarelli G. Split chimerism between nucleated and red blood cells after bone marrow transplantation for haemoglobinopathies. Chimerism. 2011;2(1):21-22. DOI
  5. Steinberg MH, Chui DH, Dover GJ, Sebastiani P, Alsultan A. Fetal hemoglobin in sickle cell anemia: a glass half full? Blood. 2014;123(4):481-485. DOI
  6. Frangoul H, Altshuler D, Cappellini MD, et al. CRISPR-Cas9 gene editing for sickle cell disease and β-thalassemia. N Engl J Med. 2021;384(3):252-260. DOI
  7. Esrick EB, Lehmann LE, Biffi A, et al. Post-transcriptional genetic silencing of BCL11A to treat sickle cell disease. N Engl J Med. 2021;384(3):205-215. DOI
  8. Thompson AA, Walters MC, Kwiatkowski J, et al. Gene therapy in patients with transfusion-dependent β-thalassemia. N Engl J Med. 2018;378(16):1479-1493. DOI
  9. 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
  10. 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):355-364. DOI
  11. Río P, Navarro S, Wang W, et al. Successful engraftment of gene-corrected hematopoietic stem cells in non-conditioned patients with Fanconi anemia. Nat Med. 2019;25(9):1396-1401. DOI
  12. Malech HL, Maples PB, Whiting-Theobald N, et al. Prolonged production of NADPH oxidase-corrected granulocytes after gene therapy of chronic granulomatous disease. Proc Natl Acad Sci U S A. 1997;94(22):12133-12138. DOI
  13. Marciano BE, Zerbe CS, Falcone EL, et al. X-linked carriers of chronic granulomatous disease: illness, lyonization, and stability. J Allergy Clin Immunol. 2018;141(1):365-371. DOI
  14. Kohn DB, Booth C, Kang EM, et al. Lentiviral gene therapy for X-linked chronic granulomatous disease. Nat Med. 2020;26(2):200-206. DOI
  15. Biffi A, Montini E, Lorioli L, et al. Lentiviral hematopoietic stem cell gene therapy benefits metachromatic leukodystrophy. Science. 2013;341(6148):1233158. DOI
  16. Fumagalli F, Calbi V, Natali Sora MG, et al. Lentiviral haematopoietic stem-cell gene therapy for early-onset metachromatic leukodystrophy: long-term results from a non-randomised, open-label, phase 1/2 trial and expanded access. Lancet. 2022;399(10322):372-383. DOI

Ex vivo gene therapy has been able to duck a question. Take the cells out of the body and, in a dish, they can be edited almost completely; myeloablative conditioning empties the marrow, so nearly everything infused back has already been fixed, and “how much is enough” got pushed by engineering all the way to “as much as possible.” In vivo puts the question back on the table: one injection into a vein, the vector reaches only some of the hematopoietic stem cells, and only a fraction of those are actually edited — no selection, no emptying, what you edit is what you get. How much is enough?

The answer has almost nothing to do with the technology. The same editing machine at the same efficiency, applied to different diseases, meets completely different definitions of enough: the threshold is set not by the editor but by what the edited cells have to do inside the body, and by whether their descendants outlive the neighbors that were never repaired.

A number from the transplant ward

The earliest hard number came out of the failures of allogeneic transplantation. After nonmyeloablative transplantation, patients with sickle cell disease often settle into an incomplete state: donor cells and their own cells coexist for years, with neither crowding out the other. How few donor cells does it take before the disease comes back?

In 2017, Fitzhugh and colleagues answered from the ward, in Blood. Among the 67 patients they monitored, 3 showed a slow decline in donor myeloid chimerism after initially robust engraftment, and eventually the disease returned. Line those three curves up against the timing of relapse and a threshold appears: about 20%.

The step that came next is the more interesting one. They built a model containing a single variable — the difference in red-cell survival time between donor and recipient — and asked whether it was enough to explain that 20%. It was. This threshold is arithmetic done in the blood rather than in the marrow; the model further predicts that patients with the highest reticulocyte counts and correspondingly shortened red-cell lifespans need the lowest chimerism.

Red cells amplify the stem-cell ledger

Once produced, red cells do not meet equal fates: the sickled ones are destroyed quickly, the normal ones live out their span. So the fraction of corrected red cells in the peripheral blood runs well above the fraction of corrected stem cells in the marrow. In 2016, Altrock and colleagues turned that amplification into a formula in Am J Hematol: transplanting mixtures of Berkeley sickle cell and normal murine bone marrow, they observed normal cells surviving three to four times longer than sickle cells, and estimated the factor in humans at between 6 and 10; fed into the model, 25% altered hematopoietic stem cells could lead to about 65% non-sickling red cells, and 40% to a median of around 80%. This is an inference from a model plus mouse experiments, not a measurement in humans — but the order of magnitude matches that 20% from the ward.

The cleanest human observation comes from transplant follow-up in the hemoglobinopathies (thalassemia major and sickle cell disease). In 2011, Andreani and colleagues reported in Haematologica on four patients with long-term persistent mixed chimerism: donor-derived nucleated cells were 71%, 46%, 15% and 25%, while donor-derived mature erythrocytes, measured on the same days of observation, were 100%, 100%, 73% and 90%. The same group added in Chimerism that persistent mixed chimerism of this kind occurs in approximately one in ten transplanted β-thalassemia patients, who no longer require red blood cell support.

Percentage is not the only scale

If the therapeutic idea is not to replace the faulty hemoglobin but to release fetal hemoglobin (HbF) again, the threshold needs a different scale altogether. The number most easily misread is precisely “what percentage of total hemoglobin is HbF.”

In 2014, Steinberg and colleagues made the point in Blood: whether sickling is inhibited depends on the HbF concentration inside each red cell, not on the average of a blood sample. They estimated that a red cell needs roughly 10 pg of HbF to count as protected; with 20% total HbF, as few as 1% and as many as 24% of cells can carry protective levels, depending entirely on distribution; only when total HbF is near 30% is it possible for the number of protected cells to approach 70%.

So the threshold here is two-dimensional: how many cells get edited, and how much HbF each edited cell carries. That is why clinical reports are so insistent on the word pancellular. When Frangoul and colleagues reported the first two patients treated by disrupting the erythroid enhancer of BCL11A with CRISPR-Cas9 in N Engl J Med in 2021, they specified that the increases in fetal hemoglobin were distributed pancellularly; the same year, Esrick and colleagues reported in the same journal on six patients in whom BCL11A was knocked down in an erythroid-specific manner with an shmiR, and, among those who could be fully evaluated, both coordinates were measured: F-cells were 58.9 to 93.6% of untransfused red cells, with 9.0 to 18.6 pg of HbF per F-cell — right around or above Steinberg’s protective line.

Different starting points, different thresholds. In the β-thalassemia gene therapy reported by Thompson and colleagues in 2018, all but one of the 13 patients with a non-β0/β0 genotype stopped receiving red-cell transfusions; among the 9 patients with a β0/β0 genotype or two copies of the IVS1-110 mutation, the median annualized transfusion volume fell by 73% and only 3 discontinued transfusions. Both groups received the same LentiGlobin BB305 vector and the same myeloablative conditioning; what differed was the patients’ own residual globin output.

When selection pressure is on your side

In another group of diseases the bar is no longer set by the starting fraction alone but by whether the repaired cells can expand, because the body does the selecting for you. ADA-SCID and X-linked SCID belong here: repaired lymphocyte precursors can develop, proliferate and survive, while unrepaired ones never reach the finish line. In 2002, Aiuti and colleagues laid out both sides in Science: hematopoietic stem cell gene therapy for ADA-SCID had shown limited clinical efficacy because of the small proportion of engrafted genetically corrected HSCs; their change was to add nonmyeloablative conditioning on top of gene transfer, letting the corrected cells gain a foothold first and letting the lymphoid selective advantage amplify them afterwards. The two patients treated achieved sustained engraftment with differentiation into multiple lineages, increased lymphocyte counts and improved immune function.

What shows how much selection pressure matters is, paradoxically, where it is absent. In 2010, Hacein-Bey-Abina and colleagues summarized in N Engl J Med nine children with SCID-X1 treated by γ-retrovirus gene therapy, eight of whom were alive after a median follow-up of 9 years: transduced T cells were detected for up to 10.7 years, and sustained thymopoiesis kept producing naive T cells — but transduced B cells were not detected at all. From the same infusion, the T-cell lineage cleared the bar easily and the B-cell lineage did not, because the selective advantage acts only on T cells. (Acute leukemia later developed in four patients and one died; that is the price of insertional mutagenesis, and belongs to the piece on safety.)

There is also the case where the threshold walks downwards by itself. In Fanconi anemia, repaired hematopoietic stem cells are no longer tormented by DNA cross-linking damage and are naturally more viable than their neighbors. In 2019, Río and colleagues reported in Nat Med a gene therapy study in Fanconi anemia subtype A carried out without any conditioning: corrected cells expanded progressively in vivo, insertion-site analyses revealed the multipotent nature of those corrected hematopoietic stem cells, and their repopulation advantage was not due to genotoxic integrations of the therapeutic provirus; in the patients with the highest levels of gene marking, the progression of bone marrow failure was arrested.

When nobody does the selecting for you

Chronic granulomatous disease (CGD) is the counterexample: a repaired neutrophil does not live any longer for having been repaired, it simply becomes able to produce reactive oxygen species in infected tissue. The threshold has to be met the hard way. In 1997, Malech and colleagues reported in PNAS on five adults with the p47phox-deficient form of CGD who received autologous CD34+ cells transduced with a retroviral vector, without marrow conditioning: functionally corrected granulocytes were detected in all five, peaking 3 to 6 weeks after infusion, but accounting for only 0.004 to 0.05% of total peripheral blood granulocytes. The gene was indeed being expressed; that order of magnitude was still far from clinical significance.

How far? Because of random X-chromosome inactivation, female carriers of X-linked CGD differ enormously in the proportion of normal neutrophils they carry, which makes them a natural dose experiment. In 2018, Marciano and colleagues supplied the scale in J Allergy Clin Immunol: among the 93 female subjects for whom clinical data were available, the 14 carriers who had CGD-type infections had a median of only 8% oxidase-normal cells; values below 10% were strongly associated with infections (odds ratio 99), and the association persisted below 20% (odds ratio 12). The lentiviral CGD gene therapy reported by Kohn and colleagues in Nat Med in 2020 lands just above that line: of nine patients with severe X-linked CGD (two of whom died early from pre-existing comorbidities), six of the seven survivors maintained 16 to 46% oxidase-positive neutrophils at 12 months, and the survivors had no new CGD-related infections.

For some diseases the threshold is not a cell fraction at all

Metachromatic leukodystrophy (MLD) reframes the question. What is missing is a secreted enzyme (ARSA), and enzyme can be taken up by neighboring cells — cross-correction means what you need is not for every cell to be edited but for total output, and its destination, to be sufficient. As early as 2013, Biffi and colleagues saw high enzyme expression throughout the hematopoietic lineages and in the cerebrospinal fluid of three presymptomatic children with late-infantile MLD, reported in Science; in 2022, Fumagalli and colleagues gave long-term results for 29 patients in Lancet: two years after treatment, ARSA activity in peripheral blood mononuclear cells was increased above the patients’ own baseline by a mean 18.7-fold in the late-infantile variant and 5.7-fold in the early-juvenile variant; and from 3 months after treatment onward all patients showed reconstituted ARSA activity within or above the normal range, while mean ARSA activity in cerebrospinal fluid reached normal levels by 6 to 12 months. What matters is less what fraction of cells were edited than whether those cells can push total enzyme back into the normal range and deliver it where it needs to go.

Diseases like this carry a second threshold whose unit is time rather than percentage: Fumagalli and colleagues are explicit that treatment benefits were particularly apparent in patients treated before symptom onset — demyelination that has already happened does not repair itself just because the enzyme has arrived.

What this means for in vivo

Laid side by side, the problem facing in vivo becomes clear: it does not have to reproduce the near-complete editing efficiency of ex vivo, it only has to clear each disease’s own line. And that line does not sit at the same height everywhere: sickle cell disease wants something around 20%, the natural history of female CGD carriers draws it above 10%; in diseases where the body does the selecting, the bar is no longer set by the starting fraction alone but by whether the repaired cells can expand. Choosing which indication to do first is choosing how high a hurdle to clear.

But the line is easy to mismeasure. Which cell type you measure decides which number you see: in Andreani’s four patients, nucleated cells from the peripheral blood, nucleated cells from the bone marrow and burst-forming unit-erythroid colonies agreed with one another within the same patient (15% in one of them), while mature erythrocytes alone ran to 73% — the difference lies not in where the sample was taken but in whether the lineage being measured has passed through differential survival. The other trap sits on the time axis: if the editing lands in short-lived progenitors rather than genuine long-term hematopoietic stem cells, the early numbers look excellent and then decay. In vivo has no pre-infusion product release testing, and can only separate the two cases by serial sampling inside the body — and in the first month they look very much alike. Nor is the threshold a constant: no longer transfused, no longer having crises, no longer getting infected, no longer accumulating organ damage are four lines of very different height. Fitzhugh’s 20% carries weight precisely because it was read out of a process in which the disease came back, not calculated from a single cross-sectional correlation.

One step further leads to the next idea: if the threshold is set by whether the repaired cells can live better, could that advantage be manufactured on purpose? That is exactly what in vivo selection is trying to do, and also where its risk comes from — a topic for another piece.

A line written by the disease itself

Back to that syringe going into a vein. Its success is not decided by how much can be edited alone, but by the gap between how much this disease demands and how much can be edited. The first half is written into vectors and editors, and is engineering; the second half is written into the lifespan of a red cell, the expansion of a lymphocyte, the job description of a neutrophil and the diffusion radius of an enzyme, and is biology — written long before anyone thought of gene therapy.

References

  1. Fitzhugh CD, Cordes S, Taylor T, et al. At least 20% donor myeloid chimerism is necessary to reverse the sickle phenotype after allogeneic HSCT. Blood. 2017;130(17):1946-1948. DOI
  2. Altrock PM, Brendel C, Renella R, Orkin SH, Williams DA, Michor F. Mathematical modeling of erythrocyte chimerism informs genetic intervention strategies for sickle cell disease. Am J Hematol. 2016;91(9):931-937. DOI
  3. Andreani M, Testi M, Gaziev J, et al. Quantitatively different red cell/nucleated cell chimerism in patients with long-term, persistent hematopoietic mixed chimerism after bone marrow transplantation for thalassemia major or sickle cell disease. Haematologica. 2011;96(1):128-133. DOI
  4. Andreani M, Testi M, Battarra M, Lucarelli G. Split chimerism between nucleated and red blood cells after bone marrow transplantation for haemoglobinopathies. Chimerism. 2011;2(1):21-22. DOI
  5. Steinberg MH, Chui DH, Dover GJ, Sebastiani P, Alsultan A. Fetal hemoglobin in sickle cell anemia: a glass half full? Blood. 2014;123(4):481-485. DOI
  6. Frangoul H, Altshuler D, Cappellini MD, et al. CRISPR-Cas9 gene editing for sickle cell disease and β-thalassemia. N Engl J Med. 2021;384(3):252-260. DOI
  7. Esrick EB, Lehmann LE, Biffi A, et al. Post-transcriptional genetic silencing of BCL11A to treat sickle cell disease. N Engl J Med. 2021;384(3):205-215. DOI
  8. Thompson AA, Walters MC, Kwiatkowski J, et al. Gene therapy in patients with transfusion-dependent β-thalassemia. N Engl J Med. 2018;378(16):1479-1493. DOI
  9. 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
  10. 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):355-364. DOI
  11. Río P, Navarro S, Wang W, et al. Successful engraftment of gene-corrected hematopoietic stem cells in non-conditioned patients with Fanconi anemia. Nat Med. 2019;25(9):1396-1401. DOI
  12. Malech HL, Maples PB, Whiting-Theobald N, et al. Prolonged production of NADPH oxidase-corrected granulocytes after gene therapy of chronic granulomatous disease. Proc Natl Acad Sci U S A. 1997;94(22):12133-12138. DOI
  13. Marciano BE, Zerbe CS, Falcone EL, et al. X-linked carriers of chronic granulomatous disease: illness, lyonization, and stability. J Allergy Clin Immunol. 2018;141(1):365-371. DOI
  14. Kohn DB, Booth C, Kang EM, et al. Lentiviral gene therapy for X-linked chronic granulomatous disease. Nat Med. 2020;26(2):200-206. DOI
  15. Biffi A, Montini E, Lorioli L, et al. Lentiviral hematopoietic stem cell gene therapy benefits metachromatic leukodystrophy. Science. 2013;341(6148):1233158. DOI
  16. Fumagalli F, Calbi V, Natali Sora MG, et al. Lentiviral haematopoietic stem-cell gene therapy for early-onset metachromatic leukodystrophy: long-term results from a non-randomised, open-label, phase 1/2 trial and expanded access. Lancet. 2022;399(10322):372-383. DOI