Yang Liu

← Primer

HSC · Healing Source Code

Cas9 的免疫反应——预存免疫、体内表达的持久炎症 The Immune Response to Cas9 — Pre-Existing Immunity and the Persistent Inflammation of In Vivo Expression

2026-07-31 · Cas9免疫原性, 预存免疫, 体内基因编辑, AAV基因治疗, T细胞反应

在第一个 CRISPR 疗法真正进入人体之前,大多数人的免疫系统早就”认识”它了——不是因为做过基因编辑,而是因为得过一次咽炎,或者皮肤上长过一个疖子。Cas9 不是人体的发明,它是细菌用来对抗病毒的武器,而人类几乎终生都在和携带这套武器的细菌打交道。当科学家把这件武器借来编辑基因组时,借来的不只是一个蛋白质的功能,还有它作为异物、被免疫系统记住的那部分历史。

两种常见细菌留下的记忆

被改造成基因编辑工具的 Cas9,绝大多数来自两种细菌:化脓性链球菌(Streptococcus pyogenes,提供最常用的 SpCas9)和金黄色葡萄球菌(Staphylococcus aureus,提供体积更小、常用于 AAV 递送的 SaCas9)。这两种细菌不是实验室里的稀客——前者是链球菌性咽炎、猩红热的元凶,后者常年寄居在皮肤和鼻腔里,是各种皮肤感染的常见病原。几乎每个人一生中都会与它们交手很多次,而免疫系统对交手过的蛋白质,向来不健忘。

2019 年,斯坦福大学 Charlesworth 等人在 Nature Medicine 上系统检测了健康人血清和外周血单个核细胞,结果相当刺眼:在被检测的健康供者中,分别有 78% 和 58% 携带能识别 SaCas9、SpCas9 的抗体,同时分别有 78% 和 67% 的人体内存在能识别这两种 Cas9 的 T 细胞。同一年,柏林 Charité 医学院的 Wagner 等人独立证实了这一点,发现健康成年人群体中普遍存在针对 SpCas9 的效应 T 细胞反应,同时也检测到相当比例的 SpCas9 反应性调节性 T 细胞——这提示身体里同时存在想攻击和能压制这两股力量,二者的平衡决定了免疫反应真正会不会被点燃。也是在这一年,FDA 研究者 Simhadri 等人用另一套 ELISA 方法检测了 200 份美国人群血清,得到的抗体阳性率低得多——SaCas9 约一成,SpCas9 仅 2.5%。三项研究给出的数字差异很大,原因并不神秘:检测用的是抗体还是 T 细胞、抗原制备和阈值设定的差异、供者人群的年龄和地域分布,都会把”阳性率”这个数字大幅拖动。但三项研究共同指向同一个方向——这不是一个可以忽略的小概率事件,而是相当一部分人群体内已经存在的、真实的适应性免疫记忆。

ex vivo 为什么躲过了一劫

如果预存免疫如此普遍,为什么已经获批的 CRISPR 疗法——比如靠体外编辑造血干细胞而成的 exa-cel(Casgevy)——没有被这个问题绊倒?答案藏在给药方式里。这类疗法把 Cas9 蛋白和向导 RNA 组装成核糖核蛋白复合物(RNP),电穿孔进细胞后几个小时内就完成切割并迅速降解,细胞回输给患者时,残留的 Cas9 蛋白已经微乎其微。免疫系统要发动一场像样的反应,需要持续足够久、足够多的抗原暴露——而这恰恰是 ex vivo RNP 编辑刻意避开的东西。

2020 年,宾夕法尼亚大学 Stadtmauer 等人在 Science 上报告了一项更直接的检验:他们用 CRISPR 敲除三名晚期癌症患者自身 T 细胞里的内源 TCR 基因和 PD-1,再导入识别 NY-ESO-1 的合成 TCR,回输后随访长达数月。编辑后的细胞在体内实现了稳定植入,没有出现与基因编辑相关的严重不良事件——即便其中部分患者本就携带针对 Cas9 的预存 T 细胞反应。这组结果印证了一个朴素的逻辑:只要抗原暴露的窗口足够短,预存免疫这张早就存在的底牌,大概率不会被真正打出来。

in vivo 把风险搬到了台面上

体内编辑就没有这么幸运了。无论是用 AAV 装载 Cas9,还是让它在细胞内以 DNA 或长效 mRNA 的形式持续表达,抗原暴露的时间窗都被拉长了,免疫系统有充裕的时间被唤醒、扩增、动员效应细胞。

哈佛大学 Church 实验室 2016 年在 Nature Methods 上的一项早期工作已经预警了这一点:他们用拆分式 AAV 递送 SpCas9,在小鼠体内实现基因编辑和转录调控的同时,观察到小鼠针对 Cas9 蛋白和 AAV9 衣壳蛋白各自产生了明确的抗体反应,并检测到相应的 T 细胞克隆扩增——虽然这种免疫反应没有引发像肌肉电穿孔那样广泛的组织损伤,但它清楚地表明,体内持续表达的 Cas9 本身就是一个会被记住的抗原,而不只是搭便车的乘客。

密苏里大学段东升团队 2021 年在 Nature Communications 上用杜氏肌营养不良症犬模型给出了更严峻的答案。用肌肉特异性启动子驱动 AAV-CRISPR、并配合短期泼尼松龙免疫抑制,起初确实在患病犬体内恢复了大量抗肌萎缩蛋白的表达;但这些保护措施都没能挡住随之而来的肌肉炎症,以及针对 Cas9 的体液免疫和细胞毒性 T 淋巴细胞反应。作为对照,携带微小抗肌萎缩蛋白基因、SERCA2a 基因或碱性磷酸酶报告基因的同类 AAV 载体,却能在同一物种体内维持稳定表达而不引发这种炎症——这个对照至关重要,它把矛头精确指向了 Cas9 本身,而不是笼统地归咎于 AAV 衣壳。在小鼠身上看起来温和的反应,放大到体型更接近人类的大型哺乳动物身上,变成了足以清除编辑细胞、动摇疗效持久性的真实威胁。

临床还没撞上这堵墙,但已经看到影子

第一个经全身静脉给药、真正让 Cas9 蛋白进入人体血液的体内 CRISPR 疗法,选择的是一条尽量缩短抗原暴露窗口的路线。2021 年,Gillmore 等人在 New England Journal of Medicine 上报告了 NTLA-2001 治疗转甲状腺素蛋白淀粉样变性的首批数据:脂质纳米颗粒递送编码 Cas9 的 mRNA 和向导 RNA,在肝细胞内完成一次性编辑后,Cas9 表达随 mRNA 降解在体内迅速消退。首批六名患者出现的不良事件均为轻度,未见严重不良事件。更早的 2020 年 3 月,面向遗传性视网膜疾病 CEP290 突变的 EDIT-101(Editas 的 BRILLIANCE 试验,NCT03872479)其实已经完成了被广泛认为是全球第一例体内 CRISPR 基因编辑疗法的人体给药——只是它走的是局部视网膜下注射,而非全身静脉给药。这类疗法把 AAV 直接注射进视网膜下腔,而眼睛本身是相对的免疫豁免部位,这大概率也是它至今没有暴露出明显 Cas9 免疫问题的原因之一。几年后的 2024 年,Pierce 等人在同一期刊上发表了这项试验的同行评审数据,同样显示出良好的耐受性。这些早期信号是令人鼓舞的,但样本量都还很小、随访时间也有限,尤其是这类疗法未来若需要二次给药以提高编辑效率或覆盖新发突变,首次给药后建立起来的抗 Cas9、抗衣壳免疫记忆,会不会成为拦路虎,目前还没有足够长期的人体数据可以回答。

工程师们已经在设计解药

既然问题的根源是 Cas9 携带着能被 T 细胞识别的免疫显性表位,一个直接的思路就是把这些表位从蛋白质上抹掉。2019 年,亚利桑那州立大学 Anderson 实验室在 Nature Communications 上报告了一版经过工程改造的 SpCas9:他们利用表位预测算法,鉴定出两个由 HLA-A*02:01 呈递的免疫显性 T 细胞表位,通过定点突变将其消除,在保留原有编辑活性和特异性的同时,显著降低了这个蛋白被 T 细胞识别的能力。这类 immunosilencing 策略,连同利用不同细菌来源、相互之间预存免疫交叉反应较低的 Cas9 直系同源蛋白轮换使用的思路,正在成为体内基因编辑工具箱里越来越受重视的一部分——目标不是消灭 Cas9 的免疫原性,而是把它压低到不足以中断治疗的水平。

一道还没被写进说明书的风险

预存免疫和体内表达引发的持久炎症,和这个连载里讲过的双链断裂、插入突变致癌完全是两条不同的风险线——它不来自基因组被切开的那一刻,而来自身体如何看待”Cas9”这个外来蛋白质本身。对大多数已经上市或走到中后期临床的疗法而言,靠一次性、短窗口的编辑策略,这堵墙目前还只是影子;但对那些押注长期或反复体内表达的下一代疗法来说,这道题迟早要正面回答。它和递送载体本身的免疫原性——预存抗体、补体反应、二次给药的困境——是同一枚硬币的两面:身体不仅会记住装 Cas9 的那辆车,也会记住车上坐的是谁。

参考文献

  1. Charlesworth CT, Deshpande PS, Dever DP, et al. Identification of preexisting adaptive immunity to Cas9 proteins in humans. Nat Med. 2019;25(2):249-254. DOI
  2. Wagner DL, Amini L, Wendering DJ, et al. High prevalence of Streptococcus pyogenes Cas9-reactive T cells within the adult human population. Nat Med. 2019;25(2):242-248. DOI
  3. Simhadri VL, McGill J, McMahon S, Wang J, Jiang H, Sauna ZE. Prevalence of Pre-existing Antibodies to CRISPR-Associated Nuclease Cas9 in the USA Population. Mol Ther Methods Clin Dev. 2018;10:105-112. DOI
  4. Stadtmauer EA, Fraietta JA, Davis MM, et al. CRISPR-engineered T cells in patients with refractory cancer. Science. 2020;367(6481):eaba7365. DOI(clinicaltrials.gov NCT03399448)
  5. Chew WL, Tabebordbar M, Cheng JKW, et al. A multifunctional AAV–CRISPR–Cas9 and its host response. Nat Methods. 2016;13(10):868-874. DOI
  6. Hakim CH, Kumar SRP, Pérez-López DO, Wasala NB, Zhang D, Yue Y, et al. Cas9-specific immune responses compromise local and systemic AAV CRISPR therapy in multiple dystrophic canine models. Nat Commun. 2021;12:6769. DOI
  7. Gillmore JD, Gane E, Taubel J, et al. CRISPR-Cas9 In Vivo Gene Editing for Transthyretin Amyloidosis. N Engl J Med. 2021;385(6):493-502. DOI(clinicaltrials.gov NCT04601051)
  8. Pierce EA, Aleman TS, Jayasundera KT, et al. Gene Editing for CEP290-Associated Retinal Degeneration. N Engl J Med. 2024;390(21):1972-1984. DOI(clinicaltrials.gov NCT03872479)
  9. Ferdosi SR, Ewaisha R, Moghadam F, et al. Multifunctional CRISPR-Cas9 with engineered immunosilenced human T cell epitopes. Nat Commun. 2019;10:1842. DOI

Long before the first CRISPR therapy ever entered a human body, most people’s immune systems already “knew” it — not because they’d undergone gene editing, but because they’d once had a bout of strep throat, or a boil on their skin. Cas9 is not a human invention; it’s a weapon bacteria use against viruses, and humans spend nearly their entire lives in contact with the bacteria that carry it. When scientists borrowed this weapon to edit genomes, they borrowed not just a protein’s function, but also its history as a foreign body — one the immune system remembers.

The Memory Left by Two Common Bacteria

The Cas9 proteins engineered into gene-editing tools come, overwhelmingly, from two bacteria: Streptococcus pyogenes, which supplies the most widely used SpCas9, and Staphylococcus aureus, which supplies the smaller SaCas9 often favored for AAV delivery. Neither bacterium is a laboratory rarity. S. pyogenes is the culprit behind strep throat and scarlet fever; S. aureus lives year-round on human skin and in the nasal passages, a common cause of skin infections. Almost everyone crosses paths with them many times over a lifetime, and the immune system never forgets a protein it has already fought.

In 2019, Charlesworth and colleagues at Stanford systematically screened healthy human serum and peripheral blood mononuclear cells in Nature Medicine, and the results were striking: among the healthy donors tested, 78% and 58% carried antibodies recognizing SaCas9 and SpCas9 respectively, while 78% and 67% had T cells recognizing the two Cas9 proteins. That same year, Wagner and colleagues at Charité – Universitätsmedizin Berlin independently confirmed the pattern, finding widespread effector T-cell responses against SpCas9 in the healthy adult population, alongside a substantial proportion of SpCas9-reactive regulatory T cells — suggesting that the body harbors both a force that wants to attack and one that can suppress it, and that the balance between them determines whether an immune response actually ignites. Also in 2019, FDA researcher Simhadri and colleagues used a different ELISA method to screen 200 serum samples from the US population and found much lower seropositivity rates — around one in ten for SaCas9 and just 2.5% for SpCas9. The wide spread across the three studies isn’t mysterious: whether antibodies or T cells were assayed, differences in antigen preparation and threshold-setting, and the age and geographic distribution of the donor populations can all pull a “positivity rate” substantially in one direction or the other. But all three studies point the same way — this is not a negligible edge case, but real, adaptive immune memory already present in a meaningful share of the population.

Why Ex Vivo Dodged the Bullet

If pre-existing immunity is this widespread, why hasn’t it tripped up already-approved CRISPR therapies — such as exa-cel (Casgevy), built by editing hematopoietic stem cells ex vivo? The answer lies in how the drug is delivered. These therapies assemble Cas9 protein and guide RNA into a ribonucleoprotein (RNP) complex, electroporate it into cells, and the complex completes its cutting and degrades rapidly within a few hours; by the time the cells are infused back into the patient, residual Cas9 protein is already vanishingly small. Mounting a real immune response requires antigen exposure that is sustained long enough and large enough — and that is precisely what ex vivo RNP editing is designed to avoid.

In 2020, Stadtmauer and colleagues at the University of Pennsylvania reported a more direct test of this in Science: they used CRISPR to knock out the endogenous TCR genes and PD-1 in three advanced-cancer patients’ own T cells, introduced a synthetic TCR recognizing NY-ESO-1, and followed the reinfused cells for months. The edited cells achieved stable engraftment in vivo, with no serious adverse events related to the gene editing — even though some of these patients already carried pre-existing T-cell responses to Cas9. The result confirms a simple logic: as long as the window of antigen exposure is short enough, the pre-existing immunity already sitting in the deck is unlikely to actually get played.

In Vivo Brings the Risk Into the Open

In vivo editing isn’t so lucky. Whether Cas9 is packaged into AAV or expressed persistently inside cells as DNA or long-acting mRNA, the window of antigen exposure stretches out, giving the immune system ample time to be awakened, expand, and mobilize effector cells.

An early study from George Church’s lab at Harvard, published in Nature Methods in 2016, had already flagged this: using split AAV vectors to deliver SpCas9 and achieve gene editing and transcriptional regulation in mice, they observed clear antibody responses against both the Cas9 protein and the AAV9 capsid protein, along with corresponding T-cell clonal expansion. Although this immune response didn’t cause the kind of widespread tissue damage seen with muscle electroporation, it made clear that Cas9 expressed persistently in vivo is itself an antigen that gets remembered — not merely a passenger along for the ride.

Dongsheng Duan’s team at the University of Missouri gave a starker answer in 2021 in Nature Communications, using a canine model of Duchenne muscular dystrophy. Driving AAV-CRISPR with a muscle-specific promoter, combined with short-term prednisolone immunosuppression, did initially restore substantial dystrophin expression in the affected dogs; but these protective measures failed to stop the muscle inflammation, humoral immunity against Cas9, and cytotoxic T-lymphocyte responses that followed. As a control, comparable AAV vectors carrying microdystrophin, SERCA2a, or an alkaline phosphatase reporter gene maintained stable expression in the same species without triggering this inflammation — a control that matters, because it points the finger precisely at Cas9 itself rather than the AAV capsid in general. A response that looked mild in mice, scaled up in a large mammal closer in body size to humans, became a real threat capable of clearing edited cells and undermining the durability of the therapy.

The Clinic Hasn’t Hit This Wall Yet, But It’s Already Seen the Shadow

The first in vivo CRISPR therapy to actually put Cas9 protein into the human bloodstream via systemic intravenous administration chose a route designed to keep the antigen-exposure window as short as possible. In 2021, Gillmore and colleagues reported the first data on NTLA-2001 for transthyretin amyloidosis in the New England Journal of Medicine: lipid nanoparticles deliver mRNA encoding Cas9 together with a guide RNA, a one-time edit is completed inside liver cells, and Cas9 expression subsides rapidly in vivo as the mRNA degrades. The adverse events in the first six patients were all mild, with no serious adverse events. Earlier still, in March 2020, EDIT-101 — Editas’ BRILLIANCE trial (NCT03872479), targeting the CEP290 mutation behind an inherited retinal disease — had already completed what is widely regarded as the world’s first human dosing of an in vivo CRISPR gene-editing therapy, though by subretinal injection rather than systemic intravenous delivery. This class of therapy injects AAV directly into the subretinal space, and the eye itself is a relatively immune-privileged site, which is probably one reason it hasn’t yet surfaced an obvious Cas9 immune problem. Years later, in 2024, Pierce and colleagues published this trial’s peer-reviewed data in the same journal, which likewise showed good tolerability. These early signals are encouraging, but the sample sizes remain small and follow-up limited — and in particular, if these therapies eventually need a second dose to improve editing efficiency or cover newly emerging mutations, whether the anti-Cas9 and anti-capsid immune memory built up after the first dose becomes an obstacle is a question that current long-term human data simply cannot yet answer.

Engineers Are Already Designing a Fix

Since the root of the problem is that Cas9 carries immunodominant epitopes recognizable by T cells, one direct approach is to erase those epitopes from the protein. In 2019, the Anderson lab at Arizona State University reported an engineered version of SpCas9 in Nature Communications: using epitope-prediction algorithms, they identified two immunodominant T-cell epitopes presented by HLA-A*02:01, eliminated them through site-directed mutagenesis, and — while preserving the protein’s original editing activity and specificity — significantly reduced its recognition by T cells. This kind of immunosilencing strategy, together with the idea of rotating among Cas9 orthologs from different bacterial sources that show lower cross-reactivity with each other’s pre-existing immunity, is becoming an increasingly emphasized part of the in vivo gene-editing toolbox — the goal isn’t to eliminate Cas9’s immunogenicity, but to suppress it below the level that would interrupt treatment.

A Risk Not Yet Written Into the Label

Pre-existing immunity and the persistent inflammation triggered by in vivo expression are an entirely separate line of risk from the double-strand breaks and insertional oncogenesis covered earlier in this series — it doesn’t come from the moment the genome is cut, but from how the body regards “Cas9” as a foreign protein in its own right. For most therapies already on the market or in mid-to-late-stage clinical trials, which rely on one-time, short-window editing strategies, this wall is, for now, only a shadow; but for next-generation therapies betting on sustained or repeated in vivo expression, this question will eventually have to be answered head-on. It is the flip side of the same coin as the delivery vehicle’s own immunogenicity — pre-existing antibodies, complement responses, the difficulty of redosing: the body remembers not only the vehicle carrying Cas9, but also who was riding in it.

References

  1. Charlesworth CT, Deshpande PS, Dever DP, et al. Identification of preexisting adaptive immunity to Cas9 proteins in humans. Nat Med. 2019;25(2):249-254. DOI
  2. Wagner DL, Amini L, Wendering DJ, et al. High prevalence of Streptococcus pyogenes Cas9-reactive T cells within the adult human population. Nat Med. 2019;25(2):242-248. DOI
  3. Simhadri VL, McGill J, McMahon S, Wang J, Jiang H, Sauna ZE. Prevalence of Pre-existing Antibodies to CRISPR-Associated Nuclease Cas9 in the USA Population. Mol Ther Methods Clin Dev. 2018;10:105-112. DOI
  4. Stadtmauer EA, Fraietta JA, Davis MM, et al. CRISPR-engineered T cells in patients with refractory cancer. Science. 2020;367(6481):eaba7365. DOI(clinicaltrials.gov NCT03399448)
  5. Chew WL, Tabebordbar M, Cheng JKW, et al. A multifunctional AAV–CRISPR–Cas9 and its host response. Nat Methods. 2016;13(10):868-874. DOI
  6. Hakim CH, Kumar SRP, Pérez-López DO, Wasala NB, Zhang D, Yue Y, et al. Cas9-specific immune responses compromise local and systemic AAV CRISPR therapy in multiple dystrophic canine models. Nat Commun. 2021;12:6769. DOI
  7. Gillmore JD, Gane E, Taubel J, et al. CRISPR-Cas9 In Vivo Gene Editing for Transthyretin Amyloidosis. N Engl J Med. 2021;385(6):493-502. DOI(clinicaltrials.gov NCT04601051)
  8. Pierce EA, Aleman TS, Jayasundera KT, et al. Gene Editing for CEP290-Associated Retinal Degeneration. N Engl J Med. 2024;390(21):1972-1984. DOI(clinicaltrials.gov NCT03872479)
  9. Ferdosi SR, Ewaisha R, Moghadam F, et al. Multifunctional CRISPR-Cas9 with engineered immunosilenced human T cell epitopes. Nat Commun. 2019;10:1842. DOI