Yang Liu

← Primer

HSC · Healing Source Code

切在正确的位置也会出事——大缺失与染色体丢失,以及把它们找出来的办法 Cutting in the right place can still go wrong — large deletions, chromosome loss, and how to find them

2026-10-09 · 基因编辑, 安全性, on-target, chromothripsis, CAST-seq, 造血干细胞

基因编辑的”效率”通常是这样量出来的:在切口两侧设计一对引物,把中间那段 DNA 扩增出来,测序,数一数有多少条序列被改写。这套做法有一个沉默的前提——损伤不会大到把引物自己也带走。可一旦一条等位基因被删掉几千个碱基,引物无处可落,这条等位基因就不再出现在读数里;它没有被记成失败,只是从统计里消失了。

被网眼漏掉的大缺失

2018 年,Kosicki、Tomberg 和 Bradley 在 Nat Biotechnol 上用长读长测序和 long-range PCR 基因分型重新检查了 Cas9 切口周围。他们在小鼠胚胎干细胞、小鼠造血祖细胞和一个人类分化细胞系里发现,断裂经常被修复成跨越许多个 kb 的缺失,还能看到离切口很远的损伤和交换事件。同一年,Giannoukos 等人在 BMC Genomics 上描述 UDiTaS 时点明了问题:靶向 PCR 加 NGS 或酶切分析,在检测大于约 100 bp 的缺失、倒位和易位时有关键局限。

缺失的尺度还可以更大。2019 年,Cullot 等人在 Nat Commun 上报告,在 UROS 位点只做一个 Cas9 nuclease 造成的 double-strand break,就能在细胞系和原代细胞里造成 megabase 级的染色体截短,依赖 p53 的机制;作者认为单 nickase 策略更安全。要追问的是:这些损伤到底有多大、发生在哪些细胞里、又怎样才能被看见。

从一段 DNA 到一整条染色体臂

最极端的版本发生在人类胚胎里。2020 年,Zuccaro 等人在 Cell 上报告,他们在 EYS 位点的父源染色体上引入一个 Cas9 断裂,最常见的修复结果是 microhomology-mediated end joining;但约一半的断裂始终没有被修复,父源等位基因随之变得无法检出,细胞分裂之后会失去染色体的一条或两条臂。2021 年,Alanis-Lobato 等人在 PNAS 上重新分析了 POU5F1 靶向的人类植入前胚胎的单细胞数据,在所分析的胚胎细胞中约 16% 看到了非预期的编辑结果,并见到 6 号染色体的片段性丢失与增加。这两项研究都在胚胎里完成,与造血干细胞疗法并不等价,但它们示范了一件事:只测切口附近,看不到整条染色体臂的去向。

在造血干/祖细胞里,类似的事件有人直接查过。2021 年,Boutin 等人在 Nat Commun 上报告,靶向珠蛋白基因的 Cas9 编辑会造成从切口一直延伸到端粒的 11p 截短和少见的 copy-neutral LOH;在原代造血干/祖细胞来源的克隆中,648 个里有 7 个(1.1%)带有获得性的 megabase 级 LOH,SNP array 的深入分析显示其中存在 copy-neutral LOH,造成 11p15.5 部分 uniparental disomy,涵盖两个 imprinting center。

同年,Leibowitz、Papathanasiou 等人在 Nat Genet 上给出了机制。他们结合活细胞成像与单细胞全基因组测序,在模型细胞里证明:Cas9 切割会在分裂的细胞里制造 micronuclei(微核)和 chromosome bridge,而这两种异常的核结构会启动 chromothripsis——集中在一条或少数几条染色体上的大范围重排。把同样的问题搬到临床相关的细胞里,他们用 Cas9 RNP 切开正常供者 CD34⁺ 造血干/祖细胞里的 BCL11A 红系 enhancer,24 小时后微核形成率升高了 16 倍;含微核的细胞中超过 80% 在被切染色体的无着丝粒片段上出现拷贝数异常。但 chromothripsis 本身在 HSPC 里无法用他们的成像加单细胞测序直接观察,用 PCR 在造血集落里找相应的 LOH 也没有找到——作者指出这种检测灵敏度有限,因为只有约 2.5% 的克隆预计曾形成微核。HSPC 里被证实的是 chromothripsis 的前兆,而不是它的发生率。

数字是多少

在造血干/祖细胞里,把大缺失量化的工作大致分两路。Park 等人 2022 年在 Sci Adv 上联用 clonal genotyping、ddPCR、带 UMI 的单分子实时测序和 long-amplicon 测序,测到长达数千个碱基的缺失,在 Cas9 切点处的比例为:HBB 11.7% 到 35.4%(两个不同的 gRNA,后者来自镰状细胞病患者的 HSPC)、HBG 14.3%、BCL11A 13.2%。另一路是 Turchiano 等人 2021 年在 Cell Stem Cell 上的 CAST-seq:依所用的 designer nuclease 不同,在 0% 到 0.5% 的编辑后 CD34⁺ 细胞里检出易位,而多达 20% 的 on-target 位点带有 gross rearrangement。两组数字的方法、位点和定义不同,不能互相换算,但都提示切口周围的损伤谱比短片段测序呈现的宽。

怎么把它们找出来

没有一种检测能看全,每一种都有自己的视野。长读长测序和 long-range PCR 把读取范围从几百个碱基推到几千个,抓得住 Kosicki 看到的那类缺失,但引物仍然要落在剩下的 DNA 上;ddPCR 的 allelic drop-off 检测可以数出信号掉了的等位基因,Park 等人用它与 long-range PCR 对照,指出短读长 NGS 会漏掉这些缺失等位基因。UDiTaS 走的是另一条路:先用 Tn5 打断并加接头,只需切口一侧的序列特异引物做单向扩增,一次反应里同时量小 indel 和倒位、易位这类结构变异。

要找染色体之间的重排,方法换成了 bait-prey 的逻辑。Frock 等人 2015 年在 Nat Biotechnol 上发表的 LAM-HTGTS,把一个固定的 bait 断裂作为诱饵,通过它与其他断裂发生的易位,在全基因组范围里捞出 prey 断裂,还看到了同源染色体上两个真正靶点之间发生的易位。2016 年 Hu 等人在 Nat Protoc 上发表了完整协议。CAST-seq 沿用了这个思路,专为 on-target 与 off-target 活性产生的染色体畸变设计,并能识别由同源重组介导、与脱靶无关的 homology-mediated translocation,Turchiano 等人正是用它在原代人 CD34⁺ 细胞里做了量化。

最后,有些损伤只有放到染色体或细胞的层面才看得见。Rayner 等人 2019 年在 CRISPR J 上用癌细胞系报告,常用的克隆筛选方法查不出大尺度重排,而简单的细胞遗传学方法可以;Leibowitz 用的 FISH 和 spectral karyotyping 也属于这一类。SNP array 能看出 Boutin 所说的 copy-neutral LOH——拷贝数没变,只是两条等位基因变成了同一条,任何只数拷贝的方法或只测局部序列的方法都会漏掉它。至于各方法的盲区——靠引物扩增的方法需要引物落点,LAM-HTGTS 靠断裂间的易位发现断裂,染色体制片只能看到进入分裂期的细胞——是对方法原理的归纳,不是某一篇文献的结论;实践中的办法是把几种方法叠起来用。

还没有答案的部分

最大的未解是后果。Leibowitz 等人在讨论里写得很直接:现行的编辑细胞输注剂量约为每公斤 3 到 16 × 10⁶ 个 CD34⁺ 细胞,按他们测得的微核频率推算,其中含有约 10⁶ 个微核细胞,而这类事件在临床上的结局是未知的。哪些损伤的细胞会存活并扩增,取决于更多因素——Tsuchida 等人 2023 年在 Cell 上分析了 Cas9 编辑的 T 细胞,发现染色体丢失遍布全基因组,这些细胞可以在培养中持续数周,而 p53 的表达与改良工艺中的保护作用相关。

另一个方向是绕开双链断裂。Fiumara 等人 2024 年在 Nat Biotechnol 上在人造血干/祖细胞里比较了 base editor、prime editor 与 Cas9:前两者也会产生双链断裂和包括缺失、易位在内的 genotoxic byproduct,但频率低于 Cas9;同时 base editor 还改变了基因组范围内的突变图谱,增加了核苷酸变异的负担。风险随编辑方式下降,但没有归零。

监管层面已经在把这些检测写进要求。FDA 在 2026 年 4 月发布了指南草案《Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing》(draft guidance,含 non-binding recommendations),面向 IND 和 BLA 的非临床研究,涉及脱靶编辑与基因组完整性的丧失(loss of genome integrity)。

回到开头那对引物。过去我们只问它们之间的序列被改成了什么;现在要问的是引物之外——那条等位基因还在不在,那条染色体臂还在不在,下一次分裂时它们会去哪里。

参考文献

  1. Kosicki M, Tomberg K, Bradley A. Repair of double-strand breaks induced by CRISPR-Cas9 leads to large deletions and complex rearrangements. Nat Biotechnol. 2018;36(8):765-771. DOI
  2. Giannoukos G, Ciulla DM, Marco E, et al. UDiTaS™, a genome editing detection method for indels and genome rearrangements. BMC Genomics. 2018;19(1):212. DOI
  3. Cullot G, Boutin J, Toutain J, et al. CRISPR-Cas9 genome editing induces megabase-scale chromosomal truncations. Nat Commun. 2019;10(1):1136. DOI
  4. Zuccaro MV, Xu J, Mitchell C, et al. Allele-Specific Chromosome Removal after Cas9 Cleavage in Human Embryos. Cell. 2020;183(6):1650-1664.e15. DOI
  5. Alanis-Lobato G, Zohren J, McCarthy A, et al. Frequent loss of heterozygosity in CRISPR-Cas9-edited early human embryos. Proc Natl Acad Sci U S A. 2021;118(22):e2004832117. DOI
  6. Boutin J, Rosier J, Cappellen D, et al. CRISPR-Cas9 globin editing can induce megabase-scale copy-neutral losses of heterozygosity in hematopoietic cells. Nat Commun. 2021;12(1):4922. DOI
  7. Leibowitz ML, Papathanasiou S, Doerfler PA, et al. Chromothripsis as an on-target consequence of CRISPR-Cas9 genome editing. Nat Genet. 2021;53(6):895-905. DOI
  8. Park SH, Cao M, Pan Y, et al. Comprehensive analysis and accurate quantification of unintended large gene modifications induced by CRISPR-Cas9 gene editing. Sci Adv. 2022;8(42):eabo7676. DOI
  9. Turchiano G, Andrieux G, Klermund J, et al. Quantitative evaluation of chromosomal rearrangements in gene-edited human stem cells by CAST-Seq. Cell Stem Cell. 2021;28(6):1136-1147.e5. DOI
  10. Frock RL, Hu J, Meyers RM, Ho YJ, Kii E, Alt FW. Genome-wide detection of DNA double-stranded breaks induced by engineered nucleases. Nat Biotechnol. 2015;33(2):179-186. DOI
  11. Hu J, Meyers RM, Dong J, Panchakshari RA, Alt FW, Frock RL. Detecting DNA double-stranded breaks in mammalian genomes by linear amplification-mediated high-throughput genome-wide translocation sequencing. Nat Protoc. 2016;11(5):853-871. DOI
  12. Rayner E, Durin MA, Thomas R, et al. CRISPR-Cas9 Causes Chromosomal Instability and Rearrangements in Cancer Cell Lines, Detectable by Cytogenetic Methods. CRISPR J. 2019;2(6):406-416. DOI
  13. Tsuchida CA, Brandes N, Bueno R, et al. Mitigation of chromosome loss in clinical CRISPR-Cas9-engineered T cells. Cell. 2023;186(21):4567-4582.e20. DOI
  14. Fiumara M, Ferrari S, Omer-Javed A, et al. Genotoxic effects of base and prime editing in human hematopoietic stem cells. Nat Biotechnol. 2024;42(6):877-891. DOI
  15. U.S. Food and Drug Administration. Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing; Draft Guidance for Industry (April 2026; docket FDA-2026-D-1255; Federal Register notice of availability, 15 April 2026). FDA · Federal Register

Editing efficiency is usually measured like this: design a pair of primers on either side of the cut, amplify the DNA between them, sequence it, and count how many reads have been rewritten. The method carries a silent assumption — that the damage is not big enough to take the primers themselves away. But once an allele loses a few thousand bases, the primers have nowhere to land and that allele no longer appears in the readout; it is not recorded as a failure, it simply disappears from the statistics.

Large deletions slip through the net

In 2018, Kosicki, Tomberg and Bradley in Nat Biotechnol re-examined the region around Cas9 cuts with long-read sequencing and long-range PCR genotyping. In mouse embryonic stem cells, mouse hematopoietic progenitors and a human differentiated cell line, they found that breaks were frequently repaired into deletions spanning many kb, and that damage and crossover events could be seen far from the cut site. The same year, Giannoukos and colleagues, describing UDiTaS in BMC Genomics, named the problem: targeted PCR combined with NGS or restriction-enzyme analysis has key limitations in detecting deletions, inversions and translocations larger than about 100 bp.

Deletions can be larger still. In 2019, Cullot and colleagues reported in Nat Commun that a single Cas9 nuclease-induced double-strand break at the UROS locus could cause megabase-scale chromosomal truncations in cell lines and primary cells, through a p53-dependent mechanism; the authors argued that a single-nickase strategy would be safer. What remains to be asked is how large this damage actually is, in which cells it occurs, and how it can be seen.

From a stretch of DNA to a whole chromosome arm

The most extreme version occurs in human embryos. In 2020, Zuccaro and colleagues reported in Cell that after they introduced a Cas9 break on the paternal chromosome at the EYS locus, the most common repair outcome was microhomology-mediated end joining; but about half of the breaks remained unrepaired, the paternal allele became undetectable, and after cell division one or both arms of the chromosome were lost. In 2021, Alanis-Lobato and colleagues in PNAS reanalyzed single-cell data from POU5F1-targeted human preimplantation embryos and saw unintended editing outcomes in about 16% of the embryonic cells analyzed, along with segmental loss and gain of chromosome 6. Both studies were done in embryos and are not equivalent to hematopoietic stem cell therapy, but they demonstrate one thing: measuring only near the cut site cannot show what happens to a whole chromosome arm.

In hematopoietic stem/progenitor cells, similar events have been looked for directly. In 2021, Boutin and colleagues reported in Nat Commun that Cas9 editing targeting globin genes can cause 11p truncations extending from the cut site all the way to the telomere, and rare copy-neutral LOH; among clones derived from primary hematopoietic stem/progenitor cells, 7 of 648 (1.1%) carried acquired megabase-scale LOH, and in-depth SNP array analysis showed copy-neutral LOH among them, producing partial uniparental disomy of 11p15.5 that spans two imprinting centers.

In the same year, Leibowitz, Papathanasiou and colleagues supplied the mechanism in Nat Genet. Combining live-cell imaging with single-cell whole-genome sequencing, they showed in model cells that Cas9 cutting generates micronuclei and chromosome bridges in dividing cells, and that these two abnormal nuclear structures can initiate chromothripsis — extensive rearrangement concentrated on one or a few chromosomes. Moving the same question into clinically relevant cells, they cut the BCL11A erythroid enhancer with Cas9 RNP in normal donor CD34⁺ hematopoietic stem/progenitor cells; 24 hours later, micronucleus formation was up 16-fold, and over 80% of the micronucleated cells showed copy number abnormalities in the acentric fragment of the cut chromosome. But chromothripsis itself could not be observed directly in HSPCs with their imaging plus single-cell sequencing, and looking by PCR for corresponding LOH in hematopoietic colonies found none — the authors point out that this assay has limited sensitivity, since only about 2.5% of clones would be expected to have formed a micronucleus. What was demonstrated in HSPCs was a precursor of chromothripsis, not its rate.

What the numbers are

In hematopoietic stem/progenitor cells, work that quantifies large deletions falls roughly into two lines. Park and colleagues in Sci Adv in 2022 combined clonal genotyping, ddPCR, single-molecule real-time sequencing with UMIs, and long-amplicon sequencing, and measured deletions up to several thousand bases long, at Cas9 cut sites, at these proportions: HBB 11.7% to 35.4% (two different gRNAs, the latter from HSPCs of patients with sickle cell disease), HBG 14.3%, BCL11A 13.2%. The other line is the CAST-seq work of Turchiano and colleagues in Cell Stem Cell in 2021: depending on the designer nuclease used, translocations were detected in 0% to 0.5% of edited CD34⁺ cells, while up to 20% of on-target loci carried gross rearrangements. The two sets of numbers differ in method, locus and definition and cannot be converted into one another, but both suggest that the damage spectrum around the cut is wider than short-fragment sequencing presents.

How to find them

No single assay sees everything; each has its own field of view. Long-read sequencing and long-range PCR extend the readable range from a few hundred bases to a few thousand, and catch the kind of deletion Kosicki saw, but the primers still have to land on DNA that remains; the allelic drop-off assay in ddPCR can count alleles whose signal has been lost, and Park and colleagues used it alongside long-range PCR to show that short-read NGS misses these deleted alleles. UDiTaS takes a different route: it first fragments DNA and adds adapters with Tn5, needs only a sequence-specific primer on one side of the cut for unidirectional amplification, and in a single reaction measures both small indels and structural variants such as inversions and translocations.

To find rearrangements between chromosomes, the method switches to a bait-prey logic. LAM-HTGTS, published by Frock and colleagues in Nat Biotechnol in 2015, uses a fixed bait break as the lure and, through translocations between it and other breaks, fishes out prey breaks across the whole genome; it also saw translocations between two true target sites on homologous chromosomes. In 2016, Hu and colleagues published the full protocol in Nat Protoc. CAST-seq follows the same idea, designed for chromosomal aberrations arising from on-target and off-target activity, and can identify homology-mediated translocations that arise through homologous recombination and are unrelated to off-target activity; Turchiano and colleagues used it to quantify these in primary human CD34⁺ cells.

Finally, some damage is visible only at the level of the chromosome or the cell. Rayner and colleagues, in CRISPR J in 2019, reported in cancer cell lines that commonly used clone-screening methods fail to detect large-scale rearrangements while simple cytogenetic methods can; the FISH and spectral karyotyping used by Leibowitz belong to the same category. SNP arrays can reveal the copy-neutral LOH described by Boutin — the copy number is unchanged, only the two alleles have become the same one — and any method that counts only copies, or sequences only a local region, will miss it. As for the blind spot of each method — primer-based amplification needs primer landing sites, LAM-HTGTS finds breaks through translocations between them, and chromosome spreads show only cells that have entered mitosis — these are generalizations from how the methods work, not the conclusion of any one paper; in practice the approach is to stack several methods.

What has no answer yet

The biggest unknown is the consequence. Leibowitz and colleagues wrote quite directly in their discussion: the current infusion dose of edited cells is about 3 to 16 × 10⁶ CD34⁺ cells per kilogram, which, by the micronucleus frequency they measured, contains about 10⁶ micronucleated cells, and the clinical outcome of such events is unknown. Which damaged cells survive and expand depends on more factors — Tsuchida and colleagues in Cell in 2023 analyzed Cas9-edited T cells and found chromosome loss across the whole genome; these cells could persist in culture for weeks, and p53 expression was associated with protection in an improved process.

Another direction is to bypass the double-strand break. In 2024, Fiumara and colleagues in Nat Biotechnol compared base editors, prime editors and Cas9 in human hematopoietic stem/progenitor cells: the first two also produced double-strand breaks and genotoxic byproducts including deletions and translocations, but at lower frequency than Cas9; base editors also altered the genome-wide mutational landscape, increasing the load of nucleotide variants. Risk falls with the editing modality, but does not reach zero.

On the regulatory side, these assays are already being written into requirements. In April 2026 the FDA issued a draft guidance, Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing (containing non-binding recommendations), addressed to nonclinical studies for INDs and BLAs, and covering off-target editing and loss of genome integrity.

Back to the primers at the start. We used to ask only what the sequence between them had been changed into; now the question is what lies beyond them — whether that allele is still there, whether that chromosome arm is still there, and where they will go at the next division.

References

  1. Kosicki M, Tomberg K, Bradley A. Repair of double-strand breaks induced by CRISPR-Cas9 leads to large deletions and complex rearrangements. Nat Biotechnol. 2018;36(8):765-771. DOI
  2. Giannoukos G, Ciulla DM, Marco E, et al. UDiTaS™, a genome editing detection method for indels and genome rearrangements. BMC Genomics. 2018;19(1):212. DOI
  3. Cullot G, Boutin J, Toutain J, et al. CRISPR-Cas9 genome editing induces megabase-scale chromosomal truncations. Nat Commun. 2019;10(1):1136. DOI
  4. Zuccaro MV, Xu J, Mitchell C, et al. Allele-Specific Chromosome Removal after Cas9 Cleavage in Human Embryos. Cell. 2020;183(6):1650-1664.e15. DOI
  5. Alanis-Lobato G, Zohren J, McCarthy A, et al. Frequent loss of heterozygosity in CRISPR-Cas9-edited early human embryos. Proc Natl Acad Sci U S A. 2021;118(22):e2004832117. DOI
  6. Boutin J, Rosier J, Cappellen D, et al. CRISPR-Cas9 globin editing can induce megabase-scale copy-neutral losses of heterozygosity in hematopoietic cells. Nat Commun. 2021;12(1):4922. DOI
  7. Leibowitz ML, Papathanasiou S, Doerfler PA, et al. Chromothripsis as an on-target consequence of CRISPR-Cas9 genome editing. Nat Genet. 2021;53(6):895-905. DOI
  8. Park SH, Cao M, Pan Y, et al. Comprehensive analysis and accurate quantification of unintended large gene modifications induced by CRISPR-Cas9 gene editing. Sci Adv. 2022;8(42):eabo7676. DOI
  9. Turchiano G, Andrieux G, Klermund J, et al. Quantitative evaluation of chromosomal rearrangements in gene-edited human stem cells by CAST-Seq. Cell Stem Cell. 2021;28(6):1136-1147.e5. DOI
  10. Frock RL, Hu J, Meyers RM, Ho YJ, Kii E, Alt FW. Genome-wide detection of DNA double-stranded breaks induced by engineered nucleases. Nat Biotechnol. 2015;33(2):179-186. DOI
  11. Hu J, Meyers RM, Dong J, Panchakshari RA, Alt FW, Frock RL. Detecting DNA double-stranded breaks in mammalian genomes by linear amplification-mediated high-throughput genome-wide translocation sequencing. Nat Protoc. 2016;11(5):853-871. DOI
  12. Rayner E, Durin MA, Thomas R, et al. CRISPR-Cas9 Causes Chromosomal Instability and Rearrangements in Cancer Cell Lines, Detectable by Cytogenetic Methods. CRISPR J. 2019;2(6):406-416. DOI
  13. Tsuchida CA, Brandes N, Bueno R, et al. Mitigation of chromosome loss in clinical CRISPR-Cas9-engineered T cells. Cell. 2023;186(21):4567-4582.e20. DOI
  14. Fiumara M, Ferrari S, Omer-Javed A, et al. Genotoxic effects of base and prime editing in human hematopoietic stem cells. Nat Biotechnol. 2024;42(6):877-891. DOI
  15. U.S. Food and Drug Administration. Safety Assessment of Genome Editing in Human Gene Therapy Products Using Next-Generation Sequencing; Draft Guidance for Industry (April 2026; docket FDA-2026-D-1255; Federal Register notice of availability, 15 April 2026). FDA · Federal Register