Triptolide

作者：唐立成 美工：何国红 罗真真 排版：马超 01 引言 如果把今天的ADC和十年前相比，变化最明显的可能还不是靶点数量增加了多少，而是开发者开始重新审视一个更基础的问题：ADC，究竟应该怎样识别肿瘤、怎样杀伤肿瘤，又该怎样避开正常组织？ 在相似靶点、相似Payload甚至相近的Linker体系上，ADC之间的差异越来越难只靠“换一个靶点”拉开。临床中暴露出的另一面也更加清楚：同一个病灶内部可以同时存在Target-high、Target-low甚至Target-negative细胞；经历ADC治疗后，抗原下调和Payload耐药并不少见；不少看起来很诱人的肿瘤靶点，又因为正常组织表达而难以获得足够的治疗窗口。 ▲ 图1 传统ADC研发面临的三类常见瓶颈：肿瘤异质性、Payload耐药与治疗窗口 于是，一些公司在抗体端做文章，用两种抗原或者两个表位提高覆盖和内吞；另一些公司开始在一个ADC上安装两种作用机制不同的Payload；还有一类更激进的设计，不满足于“看到靶点就结合”，而是尝试让ADC先判断多个条件，再决定是否进入细胞和释放药物。 双抗ADC、Dual-payload ADC以及逻辑门控ADC，正是目前这几条路线中最值得关注的三个方向。 ▲ 图2 从传统ADC到下一代ADC：识别、载荷 与激活方式的三条升级路径 它们的成熟度并不相同。双抗ADC已经出现Ⅲ期阳性数据；Dual-payload ADC到2025—2026年才真正密集进入人体试验；逻辑门控ADC则一部分已经显示初步临床信号，更多设计仍在验证“复杂结构能不能换来更好的治疗窗口”。 02 2026年全球双抗ADC、双Payload ADC和逻辑门控ADC最新临床进展 截至2026年，从临床阶段分布来看，三类下一代ADC的成熟度差异已经比较明显。双抗ADC的管线数量最多，而且从I期到III期形成了相对完整的梯队，说明这一方向不仅早期项目储备充足，也已经有一部分资产进入后期临床验证。相比之下，双Payload ADC仍主要集中在I期和II期，III期尚为空白，当前仍处于验证双重Payload能否在提高疗效的同时控制额外毒性的阶段。逻辑门控ADC的临床管线最少，目前主要分布在I期和II期，说明其技术概念虽然具有较强吸引力，但在分子设计、门控稳定性和治疗窗口方面仍需要更多临床数据支持。整体来看，三条路线的发展节奏并不一致：双抗ADC已经进入“后期验证”阶段，而双Payload和逻辑门控ADC仍更多处在“早期探索向临床过渡”的阶段。 ▲ 图3 三类下一代ADC临床管线统计 （来源：据公开资料整理） 03 双抗ADC ADC最先遇到的问题，是肿瘤本身并没有想象中那么整齐。 即使被归为HER2阳性、TROP2阳性或EGFR阳性的肿瘤，同一个患者不同转移灶之间、同一个病灶的不同区域之间，抗原密度都可能相差很大。治疗本身还会进一步施加选择压力——高表达细胞被优先杀伤，低表达甚至不表达靶点的亚群留下来，最后重新成为优势群体。 这也是双抗ADC最直接的出发点。 传统ADC只有一个识别入口： Target A → Binding → Internalization → Payload release. 双抗ADC则允许一个分子同时识别两个抗原或两个表位。 但“双抗”只是结构描述，并不能解释它到底解决什么问题。目前的开发思路其实已经分化得相当明显。 ▲ 图4 双抗ADC的三种代表性设计策略：双靶 覆盖、双表位与功能互补 01 双靶点-双抗ADC 例如EGFR×HER3、TROP2×HER3、EGFR×MET等组合，本质上希望降低对单一抗原表达的依赖。当肿瘤细胞对其中一个靶点表达不足时，另一个靶点仍可能提供结合和内吞机会。这类结构更接近一个生物学意义上的“OR Gate”——A或者B都可以帮助ADC找到肿瘤。 Iza-bren（BL-B01D1）同时靶向EGFR和HER3。2026年2月，其在既往接受过紫杉类治疗的不可切除局部晚期或转移性三阴性乳腺癌Ⅲ期研究中，同时达到PFS和OS双主要终点；到ASCO 2026，BMS披露Iza-bren已经在三项中国Ⅲ期研究中达到主要终点。食管鳞癌研究中，中位OS为9.8个月，对照化疗为7.2个月，HR为0.64；中位PFS为4.2个月，对照组2.0个月。 这已经不是“一个新型双抗ADC在Ⅰ期看到几个PR”的概念验证。Iza-bren至少说明了一件事：复杂的双靶结构可以被带到大规模随机Ⅲ期，并产生生存获益。 当然，它还没有回答另一个问题——这种优势究竟有多少来自“EGFR×HER3双靶设计”，又有多少来自抗体、Linker、Payload、DAR和患者选择共同作用。判断一项ADC技术是否真的优于单靶设计，最终仍然需要同靶点、同Payload甚至头对头研究来拆分变量。 但从开发成熟度看，双抗ADC已经率先走出了实验室。 02 双表位-双抗ADC 另一条路线是biparatopic ADC。它不识别两个不同蛋白，而是同时结合一个抗原上的两个不同表位。 为什么同一个靶点还要识别两次？关键不是“多认一个地方”，而是抗体与受体之间的空间关系可能改变。双表位结合可能增强avidity，也可能促进靶点聚集和内吞。对于ADC来说，表面结合只是第一步，进入细胞才真正决定Payload能不能被有效送达。一个亲和力很高、但长时间停留在细胞表面的抗体，未必是最好的ADC抗体。 这也是ADC研发里常被低估的一点：最佳ADC抗体不一定是亲和力最高的那一个。 表位位置、受体循环、internalization kinetics，很多时候比单纯把KD继续压低一个数量级更重要。 03 功能互补-双抗ADC 功能互补的双抗ADC真正有意思的地方，不是又增加了一组“双靶点”，是希望通过设计，让两个靶点形成功能协同。以IDE034这个同时靶向B7-H3和PTK7，携带TOP1 Payload的ADC为例，IDE034希望主要在B7-H3和PTK7位于同一个肿瘤细胞表面时获得充分结合和内吞。 这和前面的OR型双抗ADC已经不是同一件事。前者在问“A或者B有没有”？功能互补-双抗ADC试图问“A和B是不是同时存在”？也正因为如此，它同时站在“双抗ADC”和“逻辑门控ADC”的交界线上。 继续增加一个靶点本身并没有多少技术含量。真正困难的是，如何选择一对在肿瘤中高度共表达、在关键正常组织中却很少同时出现的抗原；两条抗体臂的亲和力又要控制在一个合适范围——单臂太强，双抗可能退化成两个单抗的简单叠加；太弱，又可能损失药效。 换句话说，双抗ADC把问题提前到了抗体发现阶段。靶点对怎么选、两个表位是否兼容、哪个Arm负责定位、哪个Arm更利于内吞、单臂亲和力应该做到多高，这些决定不能等到ADC已经偶联完成以后再解决。 04 双Payload ADC 如果双抗ADC主要解决“识别谁”的问题，Dual-payload ADC面对的则是另外一个麻烦：同一种Payload持续施加选择压力之后，肿瘤会产生耐药性。 今天临床和研发最成熟的ADC Payload并不算多。TOP1 inhibitor、MMAE/MMAF、DM1/DM4等少数机制占据了相当高比例。特别是DXd、exatecan及其衍生TOP1 Payload成功之后，大量新ADC都选择了相近机制。 这是一条经过验证的路，却也带来了同质化。 ADC耐药可以发生在很多环节：靶点下调、内吞改变、溶酶体处理异常、drug efflux增加、DNA损伤修复能力提高，甚至肿瘤细胞直接降低对Payload本身的敏感性。如果问题发生在Payload层面，换一个抗体靶点未必解决得了。 ▲ 图5 Dual-payload ADC的三类组合逻辑：双细胞毒、DNA损伤+DDR抑制、细胞毒+免疫刺激 01 双细胞毒素-双Payload ADC 双细胞毒素的双Payload ADC逻辑不难理解。如果一部分肿瘤细胞已经降低了对某个Payload抑制的敏感性，第二种机制仍可能保留活性。以KH815为例，这款TROP2 Dual-payload ADC同时携带TOP1 inhibitor和triptolide相关Payload，后者作用于RNA polymerase II相关机制。它的目的不是简单提高总DAR，而是让同一个TROP2阳性细胞同时面对两种不同压力。 但真正把这种想法做成药，比写在PPT上难得多。两个Payload的疏水性不同，效力不同，膜通透性不同，释放之后的半衰期也不同。一个Payload在皮摩尔水平有效，另一个可能需要更高胞内浓度；如果只是按化学上方便的比例挂到抗体上，并不意味着两者能够在肿瘤细胞里形成理想比例。 Dual-payload真正要解决的是“剂量配比”。只是这个剂量，不再由医生分别开两支药，而是在ADC制造阶段就写进了分子结构里。 02 DNA损伤+DDR抑制-双Payload ADC DNA损伤+DDR抑制的双Payload ADC受到关注，很大程度上不是因为“有两个Payload”，而是这两个Payload之间存在较明确的生物学关系。 以CLIO-8221为例，其中一个负责产生TOP1相关DNA损伤，另一个针对ATR通路。癌细胞遭遇复制压力和DNA损伤后，需要ATR等DNA damage response机制帮助完成修复和存活。于是这个组合实际是在同一个细胞里完成两件事：一边制造损伤，一边削弱修复。 这种设计比“把两种成熟毒素各挂一半”更接近Dual-payload真正需要证明的东西——机制协同。 不过到目前为止，CLIO-8221仍处于Ⅰ期。它能不能在人体中取得比成熟HER2 ADC更好的治疗窗口，还没有答案。漂亮的机制，并不等于临床优势。 03 细胞毒素+免疫调节-双Payload ADC 2026年另一个明显变化，是Dual-payload开始从“两种细胞毒机制”走向“细胞毒+免疫调节”。 这套设计试图同时做两件事：TOP1 Payload直接杀伤肿瘤；STING agonist在局部刺激先天免疫。 这类项目可能比“双毒素组合”更值得长期观察。传统ADC已经在做精准递送。如果同一个递送平台能够同时完成直接杀伤和局部免疫调节，ADC的功能就不再局限于“精准化疗”。 但历史经验也提醒开发者，免疫刺激Payload不好做。STING、TLR等激动剂系统暴露后可能带来明显炎症和安全性问题，而ADC也不可能做到百分之百只在肿瘤中释放。Linker稳定性、未偶联Payload、肿瘤外释放以及Fc相关摄取，都可能重新把局部治疗变成全身暴露。 所以这条路线是否成立，最终看的仍是治疗窗口，不是机制图画得有多漂亮。 05 逻辑门控ADC 双抗ADC提高识别能力，Dual-payload增加杀伤机制，逻辑门控ADC尝试解决的则是ADC长期以来最棘手的问题之一：怎样把肿瘤和正常组织区分得更干净？ 因为现实中的很多优质靶点都不是真正的tumor-specific antigen。它们更常见的状态是：肿瘤表达很多，正常组织也表达一些。传统ADC只能依靠表达量差异寻找治疗窗口。只要抗体还能结合正常细胞，就始终存在on-target/off-tumor毒性。 逻辑门控的思路是，不再只读取一个输入。 这和电子电路中的Boolean Logic有些相似，但目前药物研发中的“逻辑门”更多是一种功能上的类比，并不是电子计算机那种严格数字逻辑。 ▲ 图6 逻辑门控ADC的主要设计类型： OR、AND、条件门及NOT概念 01 OR门：多数双抗ADC代表的逻辑门 如果一个ADC能够识别A或B两个抗原：A OR B → Binding，那它本质上已经具有OR型特征。 Iza-bren、TROP2×HER3以及部分EGFR×MET类分子大体都可以放进这一框架中理解。OR Gate追求覆盖面。A丢失了，还有B。 它特别适合解决肿瘤异质性，但对正常组织选择性未必天然更好。事实上，如果A和B分别存在于两类正常组织，覆盖面扩大反而可能带来新的毒性边界。 所以“BsADC”和“逻辑门控ADC”不能直接画等号。真正让这个概念变得有意思的是AND。 02 AND门：只有两个抗原同时存在，ADC才获得足够活性 AND门设计追求的不是“多打一群细胞”。恰恰相反，它主动缩小目标群体，希望换取更高选择性。 假设A在正常肺组织中有一定表达，B在另一类正常细胞中也存在，但A和B很少同时出现在同一个正常细胞表面，那么：A单阳性时结合不足；B单阳性时结合不足；A+B双阳性时avidity增强并触发高效内吞。 ▲ 图7 AND Gate ADC的核心机制： 双阳性肿瘤获得更强avidity与内吞 这才是AND Gate真正吸引人的地方。它提供了一种不同的靶点开发思路：过去开发者要寻找一个“肿瘤高、正常低”的单一抗原；未来也许可以寻找两个都不够完美、但共表达图谱足够肿瘤特异的抗原。 靶点空间因此有可能扩大。但工程难度也明显增加。真正的AND Gate往往不能让两个单臂都拥有非常高的亲和力。如果任意一条Arm自己就可以牢牢抓住细胞并完成内吞，所谓AND最终还是会退化成OR。 这意味着传统抗体研发中“亲和力越高越好”的直觉，在这里甚至可能是错的。需要优化的不是单个抗体，而是两个Arm之间的平衡。 03 条件门：不读取第二个抗原，而是读取肿瘤环境 条件门属于另一类思路，它没有要求肿瘤同时表达A和B，而是给抗体加上一个Mask。ADC进入血液和正常组织时，抗原结合位点被遮挡；到达蛋白酶活跃的肿瘤微环境后，Mask被切除，抗体重新获得结合能力。于是逻辑可以近似理解为：Tumor antigen AND protease activity → Active ADC。 ▲ 图8 条件激活ADC：利用蛋白酶 或肿瘤微环境控制ADC“开/关” 如果这种条件门能够广泛成立，一些过去因为正常组织表达而被认为治疗窗口不足的靶点，可以通过增加条件激活重新进入开发视野。 04 Not门：仍在概念早期 目前行业讨论逻辑门控时，很容易出现一种倾向：AND之后还有OR，之后可以再加NOT，最后做成多输入计算系统。概念上当然成立。 真正的NOT Gate可以设计成：A AND NOT B → Kill。比如A在肿瘤和正常组织都有表达，而B只存在于正常组织。ADC识别到B后进入抑制状态，于是A+B正常细胞被保护，A+B−肿瘤细胞被杀伤。 这个想法在CAR-T领域比ADC成熟得多。ADC的问题在于，它没有天然的细胞内抑制信号回路。CAR-T可以让一个受体产生激活信号、另一个受体产生抑制信号；ADC只是一个分子。一旦高亲和力结合、内吞、Linker裂解，Payload已经进入细胞，很难再告诉它“刚才算错了，不要释放”。 因此截至目前，真正严格意义上的NOT-gated ADC仍非常早期。现阶段最现实的门控方式主要还是两种：利用avidity实现A AND B，或者利用Mask、蛋白酶、pH等肿瘤环境实现Target AND Tumor Condition。 更复杂的Boolean circuit值得关注，但没必要过早把它包装成已经成型的下一代技术。 06 三优生物助力下一代ADC研发 下一代ADC研发，前端抗体是关键。 ADC常被形容成“抗体+Linker+Payload”。这种拆法很方便，却容易让人误以为抗体只是一个负责把毒素带到肿瘤的导航头。到了下一代ADC，这种理解越来越不够用了。 双抗ADC需要的是一组匹配的抗体。两个靶点是否在同一个肿瘤细胞上共表达，两个Epitope之间的空间距离是否允许有效双价结合，哪个Arm需要更强亲和力，哪个Arm负责促进内吞，都可能改变最终效果。 AND-gated ADC甚至可能要求开发者刻意保留较低的单臂亲和力。对于普通ADC，内吞能力已经重要；对于逻辑门控ADC，内吞本身还必须和“条件成立”绑定。 Dual-payload看起来主要发生在偶联端，前端抗体其实同样重要。如果两种Payload导致分子整体疏水性增加、DAR升高或PK变差，那么抗体本身的稳定性、聚集倾向和表达水平必须留出更多开发空间。 因此，一个适合下一代ADC的抗体，评价标准很难再停留在“特异性好、亲和力高”。Epitope、internalization、trafficking、cross-reactivity、developability，甚至它与第二株抗体组合后的行为，都需要更早进入筛选流程。 三优生物目前以AI-STAL（AI-Super Trillion Antibody Library） 和 SAI-DA（Sanyou AI-Drug Accelerator）为核心，结合ADC分子构建与评价平台，为下一代ADC提供从抗体发现、功能筛选到偶联优化的研发支持。 三优生物ADC分子构建平台链接：https://adcdev.sanyoubio.com.cn ▲ 图9 三优生物下一代ADC发现与开发支持流程 01 三优AI-STAL：先把适合ADC的抗体候选空间打开 ADC开发的第一步仍然是抗体，但“能结合靶点”只是最基础的要求。 三优AI-STAL以超大容量抗体库为核心。AI-STAL目前拥有约11万亿级库容，多样化的库种类，并以候选分子数量、筛选成功率和筛选速度作为其核心指标。 丰富的候选抗体可能性，对于普通ADC，这意味着可以围绕同一靶点获得更多不同Epitope、不同亲和力和不同内吞特征的抗体；对于双抗ADC、双Payload ADC和逻辑门控ADC则更为关键，当候选抗体丰富度不足时，开发者很容易过早锁定一个Pair。候选空间足够大，才有条件真正比较： ·  Epitope是否匹配； ·  单臂亲和力是否平衡； ·  两个靶点是否能形成有效共结合； ·  双抗构建后是否保持表达和稳定性； ·  是否具有足够的internalization。 对于下一代ADC研发，项目需要的不是“最强的一株抗体”，而是最适合这个分子架构的一组抗体。 ▲ 图10 三优生物AI-STAL一图了解 02 三优SAI-DA：把“找抗体”继续推进到“选ADC候选物” AI-STAL解决的是源头分子产生，SAI-DA则承担后续筛选和开发加速。 三优生物新药加速器SAI-DA是一个整合AI设计和湿实验验证的全场景新药研发加速器，覆盖从分子产生到临床前研发的多个环节。其核心思路不是用计算替代实验，而是把部分原本需要大量试错的工作提前到计算和并行筛选阶段。 这一点放在下一代ADC上尤其实际。例如，一个双抗ADC项目可能同时有： ·  5株Target A抗体； ·  5株Target B抗体； ·  数种双抗Format； ·  不同Affinity组合； ·  多种Linker； ·  不同Payload； ·  多个DAR设计。 理论组合数量会很快膨胀。 真正开发时，不可能把所有组合一路推进到动物实验。需要在结合、内吞、细胞杀伤、稳定性以及成药性层面不断淘汰。SAI-DA所代表的AI+湿实验工作方式，更适合处理这种高参数项目：计算用于缩小搜索空间，实验负责验证真正决定药效和安全性的变量。 ▲ 图11 三优生物SAI-DA一图了解 03 三优ADC分子构建平台：从候选进入真正的ADC比较 目前，三优ADC研发平台已经完成600+ ADC相关项目，其中ADC偶联项目超过500个，覆盖bispecific antibody、nanobody、mAb等不同抗体形式，并支持多种偶联方式、Linker-Payload体系及DAR设计。这对于双抗ADC和Dual-payload ADC很关键，因为同一个抗体骨架往往需要同时比较不同：Payload、DAR、Linker和偶联方式。复杂ADC真正的开发价值，也往往来自这种并行比较，而不是一次性完成偶联。 从公司未来管线布局来看，三优也已将以逻辑门控ADC为代表的新一代ADC列为重点方向，并规划双靶点、双Payload及其组合形式。 因此，三优在下一代ADC研发中的核心价值可以概括为一条更直接的路径： AI-STAL提供更丰富的抗体候选→SAI-DA筛出更合适的分子与组合→ADC平台完成偶联、优化和验证。 对于双抗ADC、Dual-payload ADC和逻辑门控ADC而言，开发的重点已经不只是“能不能做出来”，而是能否更快找到真正值得推进的那个分子。这恰恰是三优最擅长的方向。 07 结语 目前来看，ADC的发展并没有出现一条能够取代其他路线的“标准答案”。 三条路线，其实在解决三种不同的失败方式。把双抗、双Payload和逻辑门控放在一起看，会发现它们并不是三个为了“更新”而产生的新概念。 双抗ADC主要针对抗原异质性和单靶逃逸。Dual-payload主要针对单一药理机制带来的耐药选择压力。逻辑门控关注的则是正常组织表达和治疗窗口。 这也意味着三个方向并没有互斥关系。理论上，一枚ADC完全可以同时是“双抗 + AND Gate + Dual-payload”。比如利用两个抗原共同表达决定是否高效内吞，再通过两种具有机制互补性的Payload完成杀伤。 从2026年的临床成熟度看，三条路线的排序也很明显。双抗ADC已经进入“后期验证”阶段，而双Payload和逻辑门控ADC仍更多处在“早期探索向临床过渡”的阶段。 现在谈谁会成为ADC 2.0的最终形态，还太早。更可能发生的是，这些技术先各自证明一部分价值，之后再出现自由组合，成为效果更好，能为病人带来更多改善的“终极ADC”。 SAI-DA 1.0 | Sanyou Bio Supports Next-Generation ADCs: Bispecific ADCs, Dual-Payload ADCs, and Logic-Gated ADCs 01 Introduction Compared with ADCs a decade ago, the most striking change today may not be the sheer increase in the number of targets, but the fact that developers are revisiting a more fundamental question: how should an ADC recognize tumor cells, kill them effectively, and spare normal tissues? As more programs converge on similar targets, payloads, and even linker systems, it has become increasingly difficult to differentiate one ADC from another simply by “switching targets”. Clinical experience has also made the limitations clearer. Within the same lesion, target-high, target-low, and even target-negative cells can coexist; antigen downregulation and payload resistance can emerge after ADC treatment; and many otherwise attractive tumor targets are also expressed in normal tissues, making an adequate therapeutic window difficult to achieve. ▲ Figure 1. Three common bottlenecks in conventional ADC development: tumor heterogeneity, payload resistance, and the therapeutic window Developers are therefore beginning to intervene at different points in the ADC design. Some are engineering the antibody component to recognize two antigens or two epitopes, with the aim of improving coverage and internalization. Others are loading a single ADC with two payloads that act through different mechanisms. A third, more ambitious approach moves beyond the simple rule of “see the target, bind the target” and instead asks the ADC to satisfy multiple conditions before it enters the cell and releases its payload. Bispecific ADCs, dual-payload ADCs, and logic-gated ADCs are three of the most closely watched directions in this next wave of ADC development. ▲ Figure 2. From conventional ADCs to next-generation ADCs: three upgrade paths in recognition, payload delivery, and activation These approaches are not equally mature. Bispecific ADCs have already produced positive Phase III data; dual-payload ADCs only began entering human studies in meaningful numbers in 2025–2026; and logic-gated ADCs have generated some early clinical signals, while many designs are still testing whether greater molecular complexity can translate into a wider therapeutic window.  02 2026 Global Clinical Progress in Bispecific ADCs, Dual-Payload ADCs, and Logic-Gated ADCs As of 2026, the three categories show a clear difference in clinical maturity. Bispecific ADCs have the largest pipeline and now span a relatively complete continuum from Phase I through Phase III, indicating both a substantial early-stage reserve and a growing group of assets in late-stage validation. Dual-payload ADCs remain concentrated in Phase I and Phase II, with no Phase III programs yet, and the key question is still whether two payloads can improve efficacy without adding unacceptable toxicity. Logic-gated ADCs have the smallest clinical pipeline and are currently concentrated in Phase I and Phase II, suggesting that although the concept is compelling, more clinical evidence is still needed around molecular design, gating robustness, and the therapeutic window. Overall, the three routes are moving at different speeds: bispecific ADCs have entered late-stage validation, whereas dual-payload and logic-gated ADCs remain closer to the transition from early exploration into broader clinical testing. ▲ Figure 3. Clinical pipeline counts for three next-generation ADC categories (source: compiled from publicly available information) 03 Bispecific ADCs One of the first problems ADCs encounter is that tumors are far less uniform than they appear on paper. Even within tumors classified as HER2-positive, TROP2-positive, or EGFR-positive, antigen density can vary substantially between metastatic lesions in the same patient and across different regions of the same lesion. Treatment itself adds further selective pressure: cells with high target expression are preferentially eliminated, while target-low or target-negative subclones survive and may eventually become dominant. This is the most straightforward rationale for bispecific ADCs. A conventional ADC has a single recognition entry point: Target A → Binding → Internalization → Payload Release. A bispecific ADC, by contrast, can recognize two antigens or two epitopes within the same molecule. “Bispecific”, however, is only a structural description; it does not explain what problem the molecule is designed to solve. In practice, several distinct design strategies have emerged. ▲ Figure 4. Three representative bispecific ADC strategies: dual-antigen coverage, biparatopic binding, and functional pairing 01 Dual-Target Bispecific ADCs Combinations such as EGFR×HER3, TROP2×HER3, and EGFR×MET are designed primarily to reduce dependence on a single antigen. When tumor cells express one target at insufficient levels, the second target may still support binding and internalization. Functionally, these designs resemble a biological “OR gate”: either A or B can help the ADC recognize the tumor. Iza-bren (BL-B01D1) targets both EGFR and HER3. In February 2026, a Phase III study in patients with unresectable locally advanced or metastatic triple-negative breast cancer previously treated with taxane therapy met both primary endpoints of progression-free survival (PFS) and overall survival (OS). By ASCO 2026, BMS had reported that Iza-bren had met the primary endpoint in three Phase III studies conducted in China. In the esophageal squamous cell carcinoma study, median OS was 9.8 months versus 7.2 months with chemotherapy (HR 0.64), while median PFS was 4.2 months versus 2.0 months. This is no longer the kind of proof of concept in which a new bispecific ADC produces a handful of partial responses in Phase I. At a minimum, Iza-bren shows that a complex dual-target architecture can be taken into large randomized Phase III studies and generate a survival benefit. What it has not yet answered is how much of that advantage comes specifically from the EGFR×HER3 dual-target design, and how much reflects the combined contribution of the antibody, linker, payload, DAR, and patient selection. Determining whether a given ADC technology is truly superior to a single-target design will ultimately require comparisons that isolate these variables, ideally using the same target and payload and, where feasible, head-to-head studies. From a development-maturity perspective, however, bispecific ADCs are the first of these next-generation approaches to have clearly moved beyond the laboratory. 02 Biparatopic Bispecific ADCs Another route is the biparatopic ADC, which does not recognize two different proteins but instead binds two distinct epitopes on the same antigen. Why bind the same target twice? The point is not simply to “recognize one more site”, It is that the spatial relationship between the antibody and receptor may change. Biparatopic binding can increase avidity and may also promote target clustering and internalization. For an ADC, surface binding is only the first step; effective payload delivery ultimately depends on entering the cell. An antibody with very high affinity but prolonged residence at the cell surface is not necessarily the best ADC antibody. This is an often-underappreciated point in ADC development: the best ADC antibody is not necessarily the one with the highest affinity. Epitope location, receptor recycling, and internalization kinetics can matter more than pushing the Kd down by another order of magnitude. 03 Functionally Paired Bispecific ADCs What makes functionally paired bispecific ADCs interesting is not simply the addition of another target, but the attempt to make the two targets perform complementary roles. IDE034, for example, targets B7-H3 and PTK7 and carries a TOP1 payload. The molecule is designed to achieve efficient binding and internalization primarily when B7-H3 and PTK7 are present on the same tumor cell. That is fundamentally different from the OR-type bispecific ADCs described above. Those molecules ask, “Is A or B present”? A functionally paired design asks, “Are A and B present together”? This is why such molecules sit at the intersection of bispecific ADCs and logic-gated ADCs. Simply adding another target has little value by itself. The real challenge is to identify a pair of antigens that are frequently co-expressed in tumors but rarely co-expressed in critical normal tissues, while also balancing the affinities of the two arms. If either arm is too strong, the bispecific may behave like two independent monospecific antibodies; if both are too weak, efficacy may be lost. In other words, bispecific ADCs push key decisions upstream into antibody discovery. Which target pair should be chosen? Are the two epitopes compatible? Which arm should provide localization, and which should promote internalization? How strong should the monovalent affinity be? These questions cannot be postponed until after conjugation. 04 Dual-Payload ADCs If bispecific ADCs mainly address the question of “what to recognize”, dual-payload ADCs confront a different problem: sustained selective pressure from a single payload can drive tumor resistance. The number of payload classes with substantial clinical and development validation remains limited. TOP1 inhibitors, MMAE/MMAF, and DM1/DM4 account for a large share of the field. In particular, the success of DXd, exatecan, and related TOP1 payloads has encouraged many new ADC programs to adopt similar mechanisms. That is a validated development path, but it has also contributed to increasing homogeneity across the field. ADC resistance can arise at multiple points, including target downregulation, altered internalization, abnormal lysosomal processing, increased drug efflux, enhanced DNA damage repair, or reduced tumor-cell sensitivity to the payload itself. If the resistance mechanism sits at the payload level, switching to another antibody target may not solve the problem. ▲ Figure 5. Three dual-payload strategies: dual cytotoxic payloads, DNA damage plus DDR inhibition, and cytotoxicity plus immune stimulation 01 Dual-Cytotoxic Dual-Payload ADCs The rationale for a dual-cytotoxic dual-payload ADC is straightforward: if a subset of tumor cells becomes less sensitive to one payload, a second mechanism may retain activity. KH815 is a representative example. This TROP2-targeted dual-payload ADC carries both a TOP1 inhibitor and a triptolide-related payload acting through an RNA polymerase II-related mechanism. The goal is not simply to increase total DAR, but to expose the same TROP2-positive cell to two distinct pharmacologic pressures. Turning that concept into a viable drug is much harder than drawing it on a slide. The two payloads may differ in hydrophobicity, potency, membrane permeability, and half-life after release. One may be active at picomolar concentrations while the other requires higher intracellular exposure. Attaching them at a chemically convenient ratio does not mean that they will reach the optimal functional ratio inside tumor cells. The real challenge for dual-payload ADCs is dose ratio. The difference is that this “dose” is no longer prescribed as two separate drugs by a physician; it is encoded into the ADC structure during manufacturing. 02 DNA Damage + DDR Inhibition Dual-Payload ADCs Dual-payload ADCs that combine DNA damage with DNA damage response (DDR) inhibition are attracting attention not simply because they carry two payloads, but because the two mechanisms have a clear biological relationship. CLIO-8221 is a useful example. One payload produces TOP1-associated DNA damage, while the other targets the ATR pathway. When cancer cells experience replication stress and DNA damage, ATR and related DDR mechanisms help them repair the damage and survive. The combination therefore performs two linked actions in the same cell: creating damage while weakening the repair response. This is closer to what dual-payload ADCs ultimately need to demonstrate—mechanistic synergy—than simply attaching two established cytotoxins to the same antibody. CLIO-8221 is still in Phase I. Whether it can achieve a better therapeutic window in humans than established HER2 ADCs remains unanswered. An elegant mechanism does not automatically translate into a clinical advantage. 03 Cytotoxic + Immunomodulatory Dual-Payload ADCs Another notable shift in 2026 is the move from “two cytotoxic mechanisms” toward “cytotoxicity + immune modulation”. This design aims to do two things at once: the TOP1 payload directly kills tumor cells, while a STING agonist locally activates innate immunity. These programs may warrant closer long-term attention than simple “two-toxin” combinations. Conventional ADCs already provide targeted delivery; if the same delivery platform can combine direct tumor-cell killing with local immune modulation, the role of an ADC is no longer limited to “targeted chemotherapy”. Experience to date also shows why immune-stimulatory payloads are difficult to develop. Systemic exposure to STING or TLR agonists can trigger substantial inflammation and safety liabilities, and no ADC can release its payload exclusively within the tumor. Linker instability, unconjugated payload, extra-tumoral release, and Fc-mediated uptake can all convert an intended local effect into systemic exposure. Ultimately, the viability of this strategy will be determined by the therapeutic window, not by how compelling the mechanism looks on paper. 05 Logic-Gated ADCs Bispecific ADCs expand recognition and dual-payload ADCs expand killing mechanisms. Logic-gated ADCs try to solve another long-standing challenge: how to distinguish tumor tissue from normal tissue more cleanly. Many attractive ADC targets are not truly tumor-specific antigens. More often, they are highly expressed in tumors but are also present to some degree in normal tissues. Conventional ADCs therefore rely on quantitative differences in expression to create a therapeutic window. As long as the antibody can still bind normal cells, on-target/off-tumor toxicity remains a risk. The idea behind logic gating is to read more than one input. The analogy to Boolean logic in electronic circuits is useful, but in drug development “logic gates” are primarily a functional analogy rather than strict digital logic. ▲ Figure 6. Major logic-gated ADC designs: OR, AND, conditional gates, and the emerging NOT concept 01 OR Gates: A Common Logic in Many Bispecific ADCs If an ADC can recognize either antigen A or antigen B—A OR B → binding—it already has an OR-like functional logic. Iza-bren, TROP2×HER3 molecules, and some EGFR×MET ADCs can broadly be understood in this way. OR gates are designed for coverage: if A is lost, B may still remain available. This can be useful for tumor heterogeneity, but it does not automatically improve normal-tissue selectivity. If A and B are expressed in different normal tissues, broader coverage may in fact introduce additional toxicity liabilities. For this reason, “bispecific ADC” and “logic-gated ADC” should not be treated as synonyms. The concept becomes more interesting when the logic shifts to AND. 02 AND Gates: Sufficient Activity Only When Both Antigens Are Present AND-gate designs are not intended to “hit more cells”. In fact, they deliberately narrow the target population in an effort to improve selectivity. Suppose antigen A is expressed to some extent in normal lung tissue and antigen B is present in another normal-cell population, but A and B rarely coexist on the same normal cell. In that case, A-only cells may bind insufficiently, B-only cells may also bind insufficiently, while A+B double-positive cells gain stronger avidity and trigger efficient internalization. ▲ Figure 7. Core mechanism of an AND-gated ADC: stronger avidity and internalization in double-positive tumor cells This is the real appeal of an AND gate. It suggests a different approach to target discovery. Instead of searching for a single antigen that is “high in tumor and low in normal tissue”, developers may be able to pair two individually imperfect markers whose co-expression pattern is sufficiently tumor-selective. That could expand the accessible target space, but it also increases the engineering challenge. A true AND gate generally cannot allow either arm to have very high monovalent affinity. If one arm alone can firmly bind a cell and drive internalization, the AND gate effectively degrades into an OR gate. This means that the familiar antibody-development instinct that “higher affinity is always better” may be wrong in this setting. What needs to be optimized is not one antibody in isolation, but the balance between the two arms. 03 Conditional Gates: Reading the Tumor Environment Instead of a Second Antigen Conditional gating takes a different approach. Rather than requiring the tumor to express both A and B, the antibody is equipped with a mask. In circulation and normal tissues, the antigen-binding site remains blocked; in a protease-rich tumor microenvironment, the mask is cleaved and binding activity is restored. Functionally, the logic can be approximated as: Tumor Antigen AND Protease Activity → Active ADC. ▲ Figure 8. Conditionally activated ADCs: using proteases or the tumor microenvironment to switch ADC activity on and off If this type of conditional gating proves broadly applicable, targets previously considered difficult to develop because of normal-tissue expression may be revisited by adding a tumor-dependent activation step. 04 NOT Gates: Still at an Early Conceptual Stage Discussion of logic gating can easily become a progression from AND to OR to NOT and eventually to multi-input systems. Conceptually, that progression is possible. A true NOT gate could be designed as A AND NOT B → kill. For example, antigen A might be expressed in both tumor and normal tissues, while antigen B is present only in normal tissue. If engagement of B switches the ADC into an inhibited state, A+B normal cells would be protected while A+B− tumor cells would remain susceptible. This concept is more mature in CAR-T than in ADCs. CAR-T cells can use one receptor to deliver an activating signal and another to deliver an inhibitory signal. An ADC, by contrast, is a molecule without a natural intracellular inhibitory circuit. Once high-affinity binding, internalization, linker cleavage, and payload entry have occurred, there is no easy way to tell the molecule, “that was the wrong cell—do not release”. For that reason, strictly defined NOT-gated ADCs remain at a very early stage. The most practical approaches today are still avidity-based A AND B designs, or target AND tumor-condition strategies using masks, proteases, pH, or other features of the tumor microenvironment. More complex Boolean circuits are worth following, but it is premature to present them as an established next-generation technology. 06 Sanyou Bio Supports Next-Generation ADC Development For next-generation ADCs, the quality of the upstream antibody is critical. ADCs are often described as “antibody + linker + payload”. That shorthand is convenient, but it can imply that the antibody is merely a navigation module that carries a toxin to the tumor. For next-generation ADCs, that view is increasingly inadequate. Bispecific ADCs require a matched pair of antibodies. Whether the two targets are co-expressed on the same tumor cell, whether the spatial relationship between the two epitopes permits effective bivalent engagement, which arm requires higher affinity, and which arm promotes internalization can all affect the final molecule. AND-gated ADCs may even require deliberately lower monovalent affinity. Internalization is already important for conventional ADCs; in a logic-gated ADC, internalization must also be coupled to the correct condition being met. Dual-payload development may appear to be mainly a conjugation challenge, but the upstream antibody remains equally important. If two payloads increase overall hydrophobicity, raise DAR, or impair pharmacokinetics, the antibody itself must provide sufficient stability, low aggregation propensity, and expression performance to accommodate the added burden. Accordingly, an antibody suitable for a next-generation ADC can no longer be judged only by specificity and affinity. Epitope, internalization, trafficking, cross-reactivity, developability, and even its behavior when paired with a second antibody all need to enter the screening process earlier. Sanyou Bio currently combines AI-STAL (AI-Super Trillion Antibody Library) and SAI-DA (Sanyou AI-Drug Accelerator) with its ADC molecular construction and evaluation platforms to support next-generation ADC programs from antibody discovery and functional screening through conjugation optimization. Sanyou Bio ADC Molecular Construction Platform: https://adcdev.sanyoubio.com.cn ▲ Figure 9. Sanyou Bio workflow supporting next-generation ADC discovery and development 01 Sanyou AI-STAL: Expanding the ADC Antibody Candidate Space Antibody discovery remains the first step in ADC development, but “binding the target” is only the minimum requirement. Sanyou’s AI-STAL is built around an ultra-large antibody library. AI-STAL currently has a library capacity of approximately 11 trillion molecules across multiple library formats, with the number of candidate molecules, screening success rate, and screening speed among its key performance metrics. A rich antibody candidate space allows conventional ADC programs to explore antibodies against the same target with different epitopes, affinities, and internalization profiles. It is even more important for bispecific ADCs, dual-payload ADCs, and logic-gated ADCs. If antibody diversity is limited, developers can be forced to lock onto a particular pair too early. A sufficiently broad candidate pool makes it possible to compare whether: • the epitopes are compatible; • the monovalent affinities are appropriately balanced; • the two targets can support effective co-engagement; • the bispecific format retains expression and stability; • internalization is sufficient. For next-generation ADC development, the objective is not to find the “strongest single antibody”, but the set of antibodies that best fits the intended molecular architecture. ▲ Figure 10. Sanyou Bio AI-STAL at a glance 02 Sanyou SAI-DA: From “Finding Antibodies” to “Selecting ADC Candidates” AI-STAL addresses upstream molecule generation, while SAI-DA supports downstream screening and development acceleration. Sanyou Bio’s SAI-DA is an all-scenario drug R&D accelerator that integrates AI design with wet-lab validation across multiple stages from molecule generation through preclinical development. Its purpose is not to replace experiments with computation, but to move part of the trial-and-error process upstream into computational design and parallel screening. This is particularly practical for next-generation ADCs. A single bispecific ADC project, for example, may simultaneously involve: • 5 Target A antibodies; • 5 Target B antibodies; • several bispecific formats; • multiple affinity combinations; • several linkers; • different payloads; • multiple DAR designs. The number of theoretical combinations can expand very quickly. In real development, it is impossible to advance every combination into animal studies. Candidates must be progressively eliminated based on binding, internalization, cell-killing activity, stability, and developability. The AI-plus-wet-lab workflow represented by SAI-DA is well suited to these high-dimensional projects: computation narrows the search space, while experiments validate the variables that ultimately determine efficacy and safety. ▲ Figure 11. Sanyou Bio SAI-DA at a glance 03 Sanyou ADC Molecular Construction Platform: From Candidates to Comparative ADC Evaluation To date, Sanyou’s ADC R&D platform has completed more than 600 ADC-related projects, including over 500 ADC conjugation projects. It supports multiple antibody formats, including mAbs, bispecific antibodies, and nanobodies, as well as multiple conjugation methods, linker-payload systems, and DAR designs. This is particularly important for bispecific and dual-payload ADCs, because the same antibody backbone often needs to be compared across different payloads, DARs, linkers, and conjugation strategies. The value of developing a complex ADC frequently lies in this parallel comparison rather than in completing a single conjugation successfully. Looking ahead, Sanyou has identified next-generation ADCs—particularly logic-gated ADCs—as a strategic focus and is planning dual-target, dual-payload, and combined designs. Sanyou’s role in next-generation ADC development can therefore be summarized as a direct progression: AI-STAL provides a broader antibody candidate pool → SAI-DA selects more suitable molecules and combinations → the ADC platform completes conjugation, optimization, and validation. For bispecific ADCs, dual-payload ADCs, and logic-gated ADCs, the challenge is no longer simply whether a molecule can be built, but how quickly the right molecule can be identified and advanced. This is where Sanyou’s integrated discovery and development capabilities are most directly relevant. 07 Conclusion At present, there is no single “standard answer” that is poised to replace all other ADC approaches. These three directions address different modes of failure rather than existing simply because the field wants something “new”. Bispecific ADCs mainly address antigen heterogeneity and single-target escape. Dual-payload ADCs address the selective pressure created by a single pharmacologic mechanism. Logic gating focuses on normal-tissue expression and the therapeutic window. The three approaches are therefore not mutually exclusive. In principle, a single ADC could simultaneously be bispecific, AND-gated, and dual-payload—for example, using co-expression of two antigens to determine whether efficient internalization occurs and then delivering two mechanistically complementary payloads. Based on clinical maturity in 2026, the hierarchy is already fairly clear: bispecific ADCs have entered late-stage validation, whereas dual-payload and logic-gated ADCs remain closer to the transition from early exploration into clinical proof of concept. It is still too early to know which approach will define the ultimate form of ADC 2.0. A more likely scenario is that each technology first proves value independently, after which the most effective elements are combined into more capable ADC designs that deliver greater clinical benefit. 推荐阅读 “全程智能”羊驼免疫抗体发现平台 研发前置数字质检，提升湿实验成功率 三优生物原创新药创新工场Clicklinks全新上线 抗体免疫原性风险评估与智能优化 GEOTA智生：开启AI抗体从头设计新路径 2026抗肥胖新靶点全景图：破解“减重不保肌”难题 Peptide-Generator：开启AI多肽从头生成新路径 从180天到14天: 如何重新定义创新药研发速度 AI驱动蛋白质从头设计，加速创新药物发现 兔单抗分子产生系统解决方案 SAI-DA 1.0 | 智优人源化加速器 SAI-DA 1.0 | GEOTA平台 三优CPO为抗衰老生物药研发全程护航 AI-STAL 2.0 | 智能化多肽亲和力成熟解决方案 AI-STAL 2.0 | 全程智能多肽分子产生系统解决方案 Clicklinks | 新药圈你好，我是原创新药创新工场 关于三优生物 三优生物是一家以“让天下没有难做的创新药”为使命，以“成为全球创新药研发的引领者”为愿景，专注于创新药的研发及商业转化的高新科技企业。 公司以智能超万亿分子库（AI-STAL）与三优智能新药加速器（SAI-DA）为基石，以创新药管线为驱动，致力于透彻解决创新药的源头创新问题。 公司致力于打造全球顶尖的原创新药创新工场，协同各方共同加速全球创新药的研发进程。 公司总部位于中国上海，在亚洲、北美洲、欧洲等多地建立了业务中心，形成了全球化的业务网络，现有投产及布局的研发及GMP场地20000多平方米。 公司已与全球2000多家药企、生技公司等建立了良好的合作关系，已赋能1500多个新药研发项目；已完成100多个合作研发项目，其中10多个协同研发项目已推至IND及临床研发阶段。 公司已申请180多项发明专利，其中30多项发明专利已获得授权，并获得了国家级高新技术企业、上海市专精特新、ISO9001、ISO27001等10余项资质及体系认证。

摘   要：线粒体自噬是一种选择性地清除受损和去极化的线粒体以维持线粒体活性和细胞内环境稳定的细胞进程。线粒体自噬参与类风湿关节炎(RA)的病理发展,在滑膜炎症、骨代谢、铁死亡等多个过程中起重要作用。目前,许多中药成分已被证明可以通过调节线粒体自噬减缓RA的进展。本文综述了线粒体自噬在RA中的作用和机制,以及中医药干预线粒体自噬治疗RA的研究,探讨了线粒体自噬作为RA治疗新策略的潜力,为RA的理论研究和临床治疗提供依据。 关键词：类风湿关节炎;线粒体自噬;铁死亡;槲皮素;人参皂苷 引用格式： [1]王煜,邓慧榕,张宇翔,等.线粒体自噬在类风湿关节炎病理机制中的作用及中医药干预研究进展[J].中国中医基础医学杂志,2026,32(3):705-710. DOI:10.19945/j.cnki.issn.1006-3250.2026.03.003. 中 医  ZGZYJC 类风湿关节炎(RA)是一种慢性自身免疫性疾病,以滑膜增生、血管翳形成、骨与软骨破坏为主要特征,持续的骨侵蚀和关节损伤可能导致身体畸形、残疾甚至过早死亡[1]。全球约1.0%的人口患有RA,女性发病率是男性的3～4倍。迄今为止,RA的发病机制仍不清楚。目前的治疗策略以调节炎症反应、缓解关节症状、延缓病程进展为主[2]。然而,现有的药物治疗仍存在不良反应多、停药后易复发、个体疗效差异明显等缺点。目前,开发能够有效阻止或逆转RA病理进展的治疗方案成为临床研究的迫切需求。深入探讨RA的发病机制,确定新的治疗靶点,将有助于开发更加个性化、有效且安全性更高的治疗方案。 线粒体是一种存在于真核细胞中重要的细胞器,在氧化磷酸化、能量转换、细胞凋亡、炎症反应等生命活动的各个方面起重要作用,同时也是活性氧(ROS)的主要来源[3]。线粒体功能障碍可导致炎症反应途径的关键刺激因子,如ROS、促凋亡因子的显著释放[4]。为了防止损伤线粒体的累积,细胞进化出复杂的质量控制机制以维持线粒体的稳态,即线粒体自噬。最近,越来越多的证据表明线粒体自噬与RA的发病密切相关[5]。本文综述了线粒体自噬在RA病理机制中的作用,并强调了中医药对RA靶点治疗的有效性。1 线粒体自噬及相关作用通路 线粒体自噬属于选择性自噬,是一种通过清除受损或功能障碍的线粒体以维持细胞内线粒体形态和功能稳定的细胞进程[6]。线粒体损伤、功能障碍或数量过多是启动线粒体自噬途径的先决条件,这些异常的线粒体被包裹在双层膜结构的自噬体中,通过细胞内的运输机制与溶酶体融合,利用溶酶体中的水解酶将线粒体降解成氨基酸、脂肪酸等小分子物质,最终被细胞重新回收利用[7]。 目前研究最多的线粒体自噬途径为PINK1/Parkin介导的泛素依赖性途径,即线粒体外膜的泛素化。PTEN诱导激酶1(PINK1)和Parkin蛋白是该通路的核心调控蛋白。正常情况下,PINK1进入线粒体内膜后被蛋白酶降解,而当线粒体受损时,PINK1在线粒体外膜积累,并募集Parkin至受损线粒体。PINK1磷酸化激活Parkin,促使其对线粒体外膜蛋白进行泛素化修饰。这些泛素链进一步被p62、视神经病变诱导反应蛋白(OPTN)等多种自噬受体识别,招募自噬起始因子,最终诱导线粒体自噬产生。除泛素依赖性途径外,含Fun14 结构域的蛋白1(FUNDC1)和BCL2相互作用蛋白3(BNIP3)/NIP3样蛋白X(NIX)等线粒体自噬受体蛋白也可通过其LIR基序直接与微管相关蛋白1轻链3(LC3)结合,将自噬膜锚定到线粒体表面,从而介导线粒体清除[8]。FUNDC1可以在缺氧反应中调节线粒体自噬。BNIP3通路主要由环境刺激引发,在组织中广泛表达。NIX作为BNIP3的同源蛋白,介导哺乳动物红细胞成熟过程中的线粒体清除。研究发现,当NIX缺失导致网织红细胞线粒体清除障碍时,BNIP3的高表达可以发挥代偿作用[9]。2 线粒体自噬在RA病理机制中的作用2.1 维持滑膜炎症 2.1.1 成纤维样滑膜细胞 成纤维样滑膜细胞(FLS)是RA滑膜炎发生的关键因素。在RA中,一方面,FLS对内源性模式识别受体配体和炎性细胞因子产生应答。另一方面,FLS通过与T细胞、B细胞等免疫细胞的相互作用来激活免疫反应,进而刺激滑膜炎症的发生[10]。此外,异常激活的FLS渗入细胞外基质并释放基质金属蛋白酶,降解骨与软骨组织[11]。同时由于血管生成失调、滑膜增殖,RA关节常处于极度缺氧状态。缺氧引起了广泛的线粒体损伤,导致线粒体呼吸减少、ROS产生过多、ATP损失,进一步加剧免疫炎症反应[12]。Deng等[13]研究发现,在缺氧条件下RA-FLS的活力及增殖能力随着氧气浓度的降低而增加,RA-FLS中BNIP3表达明显升高,线粒体自噬水平增强,抑制BNIP3水平后细胞增殖能力下降,线粒体自噬减少。因此认为,在缺氧条件下,RA-FLS通过激活线粒体自噬来清除细胞内积累的ROS,从而维持细胞内氧化还原平衡并促进FLS异常增殖,且这一过程主要由BNIP3表达水平调控。基于此,推测在今后的治疗中可通过靶向抑制FLS的线粒体自噬水平以诱导FLS凋亡,从而延缓RA的病情进展。Hong等[14]的实验研究证实了上述观点,并进一步发现RA-FLS中NOD样受体家族蛋白3(NLRP3)水平升高,而线粒体自噬对NLRP3炎性小体活化具有显著抑制作用。这表明线粒体自噬和NLRP3炎性小体之间的相互作用在RA发展中同样具有重要意义,但具体机制亟待研究。 2.1.2 驻留组织记忆T细胞 RA患者在治疗结束后普遍存在病情反复的现象。最近的研究表明,这种复发特征与RA中存在介导局部记忆的机制相关,特别是组织驻留记忆T细胞(TRM)在关节滑膜等病变部位的长期驻留。TRM是记忆T细胞的亚群之一,在抗原刺激后会在外周组织中长期驻留并存活,不通过血液或淋巴循环,即可提供快速的免疫保护[15]。然而,越来越多的证据表明,异常激活的TRM也可攻击自身抗原,驱动局部组织的损伤。与健康人相比,RA患者的CD4+T细胞及CD8+T细胞更容易分化为TRM。研究发现,RA患者的关节滑液中含有以CD8+T细胞占优势的T细胞群,这些细胞聚集在炎症关节的滑膜中,在临床缓解期持续存在。当被抗原激活时,这些T细胞通过分泌趋化因子招募外周循环中其他效应T细胞,从而触发局部炎症的级联反应,最终导致疾病复发[16]。这一证据表明,TRM可能是RA关节炎反复发作的驱动因素。进一步研究发现,RA患者T细胞中线粒体自噬高度活跃。这种异常的自噬可驱动TRM分化成熟,同时维持TRM长期存活。抑制线粒体自噬将显著减少滑膜组织中TRM的数量,进而减少滑膜炎症的发生率[17]。因此,线粒体自噬在T细胞亚群分化中具有重要调控作用。2.2 参与骨代谢 RA的病理特征包括滑膜增生、软骨退化及骨侵蚀。成骨细胞的骨形成和破骨细胞的骨吸收之间的不平衡是这些病理变化产生的主要原因。新的研究发现,异常的自噬水平会打破骨代谢平衡[18],导致RA的发生发展。 雷帕霉素靶蛋白(mTOR)信号通路参与前成骨细胞、破骨细胞和软骨细胞的增殖和分化。有学者通过对RA患者关节骨组织的免疫荧光染色分析发现,骨小梁边缘磷酸化S6核糖体蛋白(p-S6)表达显著增加,且主要存在于前成骨细胞中。提示mTOR 复合物1(mTORC1)信号通路在前成骨细胞中被异常激活。其活性与成骨细胞增殖呈正相关,而与分化能力呈负相关。在炎性因子刺激后,前成骨细胞中Parkin积聚在受损的线粒体上,使用mTORC1抑制剂雷帕霉素处理后发现,成骨分化标志物表达上调,Parkin在线粒体上的聚集更加明显。敲低Parkin后,雷帕霉素的促分化作用被显著抑制。以上结果表明,mTORC1/Parkin信号轴可能通过调控线粒体自噬来影响成骨细胞分化,这为理解RA骨代谢异常提供了新的分子机制[19]。 ROS是一种包括过氧化物、超氧化物等在内的含氧自由基,在氧化磷酸化过程中作为线粒体代谢的副产物产生。ROS在细胞中通常以低水平存在,对于维持细胞稳态和功能是必需的[20]。然而,ROS的异常产生导致炎性细胞因子和趋化因子的过度表达,引起蛋白质、脂质和DNA等细胞大分子物质的氧化,并改变其生理功能。在低氧环境中,线粒体ROS在破骨细胞中积累,影响破骨细胞转录因子的活性[21]。其水平增加或抗氧化活性降低会导致成骨细胞形成缺陷。Lee等[22]在敲除PINK1的小鼠模型中发现,小鼠成骨细胞中异常线粒体数量及ROS水平明显增加,分化能力受到抑制。说明PINK1介导的线粒体自噬受到抑制,导致异常线粒体清除不足,从而诱导成骨细胞分化过程中线粒体ROS过度产生和钙摄取异常。此外,肿瘤坏死因子超家族成员11(TNFSF11)是一种主要负责RA患者骨侵蚀发展的蛋白质,自噬受体OPTN可功能性下调TNFSF11表达水平,敲除RA-FLS中的OPTN将导致破骨细胞形成增加[23]。综上所述,线粒体自噬相关蛋白参与了RA骨形成,调控相关蛋白的表达水平、促进线粒体自噬、诱导异常线粒体自我清除、减少ROS产生可能是治疗 RA骨破坏的有效方法。2.3 抑制铁死亡 铁死亡是一种新型的铁依赖性细胞死亡,在形态学上表现出与坏死相同的特征,包括细胞质和细胞器肿胀、线粒体膜密度增加、外膜破裂等[24]。目前的研究表明,铁积累、脂质过氧化和氧化还原系统失调是导致铁死亡的主要原因[25]。在铁死亡过程中,过量积累的铁离子通过芬顿反应催化产生ROS 促进脂质过氧化,并损伤细胞膜。随后,由于抗氧化防御机制的破坏,氧化应激进一步损伤细胞膜和细胞内重要的生物分子,这种氧化损伤和抗氧化防御之间的失衡是铁死亡的重要因素[26]。谷胱甘肽依赖性过氧化物酶 4(GPX4)是抗氧化防御系统的重要组分,通过利用底物谷胱甘肽将磷脂氢过氧化物还原为相应的醇,有效限制脂质 ROS的积累,达到减少氧化以清除自由基和其他有害物质的目的[27]。因此认为GPX4清除过氧化物的能力不足和脂质过氧化物的过度积累最终导致铁死亡发生。 研究发现RA患者普遍存在血清铁、铁蛋白等铁代谢指标的升高,且铁的浓度与关节炎症的严重程度呈正相关[28]。在实验研究中发现,应用铁螯合剂能有效减少铁沉积,并改善铁过载诱导的氧化损伤和免疫功能紊乱。Fang等[29]通过研究证实铁死亡可以调节CD4+T细胞及CD8+T细胞的活性。活化的CD8+T细胞产生干扰素-γ(IFN-γ)促进RA-FLS脂质过氧化水平的升高,并抑制谷胱甘肽、GPX 4的表达。这表明CD 8+T细胞可能通过IFN-γ通路促进RA-FLS铁死亡的发生。研究发现线粒体自噬通过减少异常ROS的产生以减轻氧化应激,从而抑制铁死亡发生。Lin等[30]对骨关节炎大鼠给予过氧化物酶体增殖物激活受体γ(PPAR-γ)激动剂干预后,发现其关节软骨破坏情况较患病大鼠有所改善,同时发现铁死亡标志物的表达下调。此外,体内和体外数据表明,PPAR-γ激活能明显恢复PINK1/Parkin依赖的线粒体自噬,并有效改善线粒体功能[31]。这一发现表明,激活的PPAR-γ可通过促进PINK1/Parkin依赖的线粒体自噬途径,抑制软骨细胞铁死亡,从而缓解骨关节炎的临床症状。3 中医药调控线粒体自噬治疗RA3.1 中药单体 槲皮素是一种黄酮类化合物,广泛存在于洋葱、苹果等各种水果和蔬菜中,具有抗高血压、抗炎、抗血管生成、抗癌和神经保护等作用。研究发现槲皮素通过减少炎性细胞因子和介质,降低氧化应激,抑制FLS增殖、迁移和侵袭,从而抑制滑膜炎症[32]。此外,该药物与甲氨蝶呤同用能增强甲氨蝶呤的治疗效果并减少其对肝脏的不良反应[33]。槲皮素对线粒体具有保护作用,可通过促进线粒体自噬以减轻线粒体ROS应激来抑制NLRP 3炎性小体活化[34]。 人参皂苷是人参的重要活性成分,为三萜皂苷化合物,目前已报道了近200种人参皂苷成分具有抗炎、抗氧化、抗病毒等作用[35]。Zhang等[36]研究发现人参皂苷Rg1、人参皂苷Rg3、人参皂苷化合物 K(CK)等6种单体对RA有一定的治疗作用。其中,CK的干预效果最好,不仅在低浓度下能显著抑制RA单核巨噬细胞和内皮细胞的增殖,而且能促进异常增殖细胞的凋亡。在胶原诱导性小鼠的实验中,CK通过降低T细胞、M1型巨噬细胞数量,减少炎性因子分泌,下调免疫应答,在减轻关节肿胀、发红、功能障碍及病理改变方面也具有显著优势。近期研究发现,在关节炎大鼠模型中,Rg1通过腺苷酸活化蛋白激酶(AMPK)通路激活PINK1/Parkin介导的线粒体自噬,减少软骨细胞线粒体损伤和衰老,从而缓解临床症状 [37]。 姜黄素是姜黄的主要成分,属于苯丙类化合物,可发挥抗氧化、抗血管生成等药理作用,且无明显不良反应。姜黄素的软骨保护作用已通过多种实验证实,包括体外(软骨细胞实验)和体内(动物模型)研究[38]。在胶原诱导性关节炎大鼠实验中发现,姜黄素和雷公藤内酯醇起协同作用[39],通过抑制白介素-17/核因子(NF)-κB通路减少炎症因子水平、诱导细胞凋亡,从而缓解RA的严重程度。临床研究发现RA患者每日服用姜黄素250～1500 mg,连续治疗8～12周,可改善辅助性T细胞1型、辅助性T细胞17型、调节性T细胞及B细胞等免疫细胞的功能异常,进而减轻疾病的临床症状[40]。近年来,姜黄素调节线粒体自噬的能力受到广泛关注,Jin等[41]的研究表明,姜黄素对白介素-1β诱导的软骨细胞线粒体功能障碍起保护作用。姜黄素可通过激活腺苷酸活化蛋白激酶(AMPK)信号通路,促进磷酸化的PINK1在受损线粒体表面的聚集,进而促进线粒体自噬。这一过程有效清除了功能异常的线粒体,改善软骨细胞的能量代谢和线粒体功能,具体表现为ATP生成能力增强、线粒体膜电位恢复,同时ROS和钙离子浓度下降。这些分子层面的改善最终减轻了关节炎的临床表现,提示姜黄素可能通过调控线粒体自噬发挥软骨保护作用。 栀子苷是从栀子中提取出的环烯醚萜苷类化合物,具有抗炎、抗血管新生等作用[42]。研究表明,栀子苷可以通过刺激多条炎症信号传导通路,如促分裂原活化蛋白激酶(MEK)/丝裂原活化蛋白激酶(MAPK)等的活化降低RA-FLS中炎性因子的表达水平,并维持促炎/抗炎因子的动态平衡。另一方面,栀子苷也可以通过阻断相关通路的过度活化,有效抑制RA-FLS的异常增殖、侵袭和迁移[43]。邓然[44]的研究进一步揭示,栀子苷通过特异性下调c-Jun氨基末端激酶(JNK)/BNIP3通路,抑制RA-FLS的线粒体自噬功能,以降低其增殖活性,改善RA大鼠的滑膜异常增生及局部缺氧微环境。3.2 中药复方 桂枝芍药知母汤最早记载于《金匮要略》,由桂枝、芍药、知母、麻黄、生姜等组成,是治疗RA的传统中药复方[45]。研究发现,该方在胶原诱导性关节炎大鼠中具有骨保护作用,其特点是抑制破骨细胞形成、减轻炎症和抑制滑膜组织中FLS活化[46]。桂枝芍药知母汤通过抑制NF-κB通路活化,阻遏破骨前体细胞生成。进一步研究发现,关节炎大鼠在桂枝芍药知母汤干预后,对自噬具有正调控作用的蛋白水平显著增加,而负调控的蛋白水平受到抑制[47]。透射电镜显示,破骨前体细胞中线粒体自噬溶酶体的荧光强度显著增加,且PINK1和Parkin的表达水平增加。同样,敲低PINK1后,药物对破骨细胞生成的抑制作用基本消除。因此认为,桂枝芍药知母汤通过促进PINK1/Parkin通路介导的线粒体自噬抑制NF-κB轴,从而抑制破骨细胞的产生。 清络通痹方是国医大师周仲瑛教授根据数十年临床诊疗经验总结而成的治疗痹证的经典方剂,由雷公藤、生地黄、三七、青风藤、僵蚕等药物配伍组成,具有祛风通络、散结止痛之效。现代研究表明,该方通过调控免疫-炎症反应、抑制骨质破坏以及调节相关信号通路以达到治疗RA的目的[48]。体内实验研究发现,雷公藤可通过激活蛋白激酶R样内质网激酶(PERK)/活化转录因子4(ATF4)通路上调LC3等线粒体自噬标志物及相关蛋白水平,引发异常的线粒体自噬,而清络通痹方能有效逆转这一过程,从而降低雷公藤的毒性反应。另一方面,方中生地黄、三七能够协同降低自噬体和自噬溶酶体含量,疗效优于单味药物治疗[49]。充分体现了中医复方配伍“减毒增效”的独特优势。 补阳还五汤出自清代医家王清任所著的《医林改错》,由黄芪、当归、地龙等药物组成,具有补气活血、化瘀通络之效,尤其适用于气虚络瘀型RA患者。现代实验研究表明,适量使用补阳还五汤可通过下调病变关节中中性粒细胞标志物蛋白及其mRNA的表达水平,抑制中性粒细胞活性;同时降低血清和关节液中多种炎性因子的表达水平,以减轻病变关节骨破坏程度,从而缓解RA模型小鼠关节肿胀、疼痛等症状,展现出显著的抗炎效果[50]。为进一步探讨该方剂在低氧条件下对RA-FLS线粒体自噬的调控作用,展俊平等[51]通过实验研究发现,经补阳还五汤干预后,RA模型小鼠BNIP3和p62蛋白表达水平上调,磷酯酰肌醇3-激酶(PI3K)、蛋白激酶B (AKT)表达下调,线粒体膜电位下降。提示,补阳还五汤可能通过调控BNIP3介导的PI3K/AKT信号通路,抑制自噬相关因子的表达,进而改善低氧微环境诱导的细胞凋亡,这可能是其发挥治疗作用的重要机制之一。4 结语 RA作为以滑膜炎症和骨破坏为特征的疾病,目前临床尚缺乏能有效阻止疾病进展或逆转关节破坏的治疗方法。在此背景下,基于中医药多靶点调控优势探索新型治疗策略成为当前研究重点。线粒体作为重要的供能细胞器,参与多种重要的生理过程,线粒体功能障碍可导致心力衰竭、神经退行性变、代谢综合征及骨关节病等多系统疾病的发生发展[52]。在RA病理过程中,线粒体动力学和质量控制的异常、蛋白质稳态的功能失调、ATP产生受抑制等通常同时发生并在病理条件下相互作用[53]。自噬作为线粒体质量控制的方法之一,通过对异常线粒体的自噬降解维持线粒体稳态。最新研究证据表明,线粒体自噬功能紊乱与RA的发病机制存在显著关联,这为开发靶向线粒体质量控制的新型治疗策略提供了理论依据。 槲皮素、人参皂苷、姜黄素、栀子苷等中药成分及中药复方均能通过不同途径调节线粒体自噬,具体表现为:降低炎症因子水平、抑制滑膜细胞增殖,从而达到抗炎、抗氧化应激、保护骨及软骨的目的,治疗前景广阔。综合现有的研究成果,发现目前仍存在以下亟待解决的问题:一,线粒体自噬与RA病理相关的实验与临床研究仍然较少;二,现有的研究多集中在RA-FLS的调控机制方面,对RA免疫系统、炎症反应等重要病理环节中线粒体自噬的作用尚未阐明;三,传统中医疗法,如针刺、艾灸等对RA线粒体自噬的调控作用及其分子机制尚属空白。 综上所述,虽然目前关于线粒体自噬与RA相关性的研究仍存在不足之处,但通过中医药调控线粒体自噬相关通路治疗RA的研究方向具有重要价值,有望为RA的治疗及新药研发提供思路。 利益冲突:本文不存在任何利益冲突。 参考文献： [1] RADU A F,BUNGAU S G.Management of rheumatoid arthritis:an overview[J].Cells,2021,10(11):2857. [2] SINGH J A.Treatment guidelines in rheumatoid arthritis[J].Rheum Dis Clin North Am,2022,48(3):679-689. [3] NUNNARI J,SUOMALAINEN A.Mitochondria:in sickness and in health[J].Cell,2012,148(6):1145-1159. [4] MILLS E L,KELLY B,O’NEILL L A J.Mitochondria are the powerhouses of immunity[J].Nat Immunol,2017,18(5):488-498. [5] MA C,WANG J,HONG F F,et al.Mitochondrial dysfunction in rheumatoid arthritis[J].Biomolecules,2022,12(9):1216. [6] VARGAS J N S,HAMASAKI M,KAWABATA T,et al.The mechanisms and roles of selective autophagy in mammals[J].Nat Rev Mol Cell Biol,2023,24(3):167-185. [7] XU Y,SHEN J,RAN Z H.Emerging views of mitophagy in immunity and autoimmune diseases[J].Autophagy,2020,16(1):3-17. [8] GATICA D,LAHIRI V,KLIONSKY D J.Cargo recognition and degradation by selective autophagy[J].Nat Cell Biol,2018,20(3):233-242. [9] ZHANG J,LOYD M R,RANDALL M S,et al.A short linear motif in BNIP3L (NIX) mediates mitochondrial clearance in reticulocytes[J].Autophagy,2012,8(9):1325-1332. [10] YOSHITOMI H.Regulation of immune responses and chronic inflammation by fibroblast-like synoviocytes[J].Front Immunol,2019,10:1395. [11] CHENG L Y,WANG Y Y,WU R H,et al.New insights from single-cell sequencing data:synovial fibroblasts and synovial macrophages in rheumatoid arthritis[J].Front Immunol,2021,12:709178. [12] DENG R,BU Y H,LI F,et al.The interplay between fibroblast-like synovial and vascular endothelial cells leads to angiogenesis via the sphingosine-1-phosphate-induced RhoA-F-Actin and Ras-Erk1/2 pathways and the intervention of geniposide[J].Phytother Res,2021,35(9):5305-5317. [13] DENG R,WANG Y,BU Y H,et al.BNIP3 mediates the different adaptive responses of fibroblast-like synovial cells to hypoxia in patients with osteoarthritis and rheumatoid arthritis[J].Mol Med,2022,28(1):64. [14] HONG Z Y,WANG H,ZHANG T J,et al.The HIF-1/BNIP3 pathway mediates mitophagy to inhibit the pyroptosis of fibroblast-like synoviocytes in rheumatoid arthritis[J].Int Immunopharmacol,2024,127:111378. [15] GAO A Q,ZHAO W P,WU R H,et al.Tissue-resident memory T cells:the key frontier in local synovitis memory of rheumatoid arthritis[J].J Autoimmun,2022,133:102950. [16] CHANG M H,LEVESCOT A,NELSON-MANEY N,et al.Arthritis flares mediated by tissue-resident memory T cells in the joint[J].Cell Rep,2021,37(4):109902. [17] LIU T,WANG M,LI L,et al.Mitophagy drives maldifferentiation of tissue-resident memory T cells in patients with rheumatoid arthritis[J].Scand J Rheumatol,2025,54(2):69-78. [18] WANG S,DENG Z T,MA Y C,et al.The role of autophagy and mitophagy in bone metabolic disorders[J].Int J Biol Sci,2020,16(14):2675-2691. [19] CHEN Q Y,FAN K,SONG G B,et al.Rapamycin regulates osteogenic differentiation through Parkin-mediated mitophagy in rheumatoid arthritis[J].Int Immunopharmacol,2022,113(Pt B):109407. [20] MUSAOGULLARI A,CHAI Y C.Redox regulation by protein S-glutathionylation:from molecular mechanisms to implications in health and disease[J].Int J Mol Sci,2020,21(21):8113. [21] AGIDIGBI T S,KIM C.Reactive oxygen species in osteoclast differentiation and possible pharmaceutical targets of ROS-mediated osteoclast diseases[J].Int J Mol Sci,2019,20(14):3576. [22] LEE S Y,AN H J,KIM J M,et al.PINK1 deficiency impairs osteoblast differentiation through aberrant mitochondrial homeostasis[J].Stem Cell Res Ther,2021,12(1):589. [23] LEE W S,KATO M,SUGAWARA E,et al.Protective role of optineurin against joint destruction in rheumatoid arthritis synovial fibroblasts[J].Arthritis Rheumatol,2020,72(9):1493-1504. [24] DONDELINGER Y,PRIEM D,HUYGHE J,et al.NINJ1 is activated by cell swelling to regulate plasma membrane permeabilization during regulated necrosis[J].Cell Death Dis,2023,14(11):755. [25] STOCKWELL B R,FRIEDMANN ANGELI J P,BAYIR H,et al.Ferroptosis:a regulated cell death nexus linking metabolism,redox biology,and disease[J].Cell,2017,171(2):273-285. [26] KUANG F M,LIU J,TANG D L,et al.Oxidative damage and antioxidant defense in ferroptosis[J].Front Cell Dev Biol,2020,8:586578. [27] URSINI F,MAIORINO M.Lipid peroxidation and ferroptosis:the role of GSH and GPx4[J].Free Radic Biol Med,2020,152:175-185. [28] YAZAR M,SARBAN S,KOCYIGIT A,et al.Synovial fluid and plasma selenium,copper,zinc,and iron concentrations in patients with rheumatoid arthritis and osteoarthritis[J].Biol Trace Elem Res,2005,106(2):123-132. [29] FANG F,HUA M Q,YU G M.Study of the mechanism of fibroblast-like synoviocytes-derived exosomes inducing macrophages M1 polarization and CD8+T cells immune regulation ferroptosis and autophagy in rheumatoid arthritis[J].Immunol Lett,2024,270:106936. [30] LIN Q S,LI S,JIN H J,et al.Mitophagy alleviates cisplatin-induced renal tubular epithelial cell ferroptosis through ROS/HO-1/GPX4 axis[J].Int J Biol Sci,2023,19(4):1192-1210. [31] XUE X,DAI T M,CHEN J Y,et al.PPARγ activation suppresses chondrocyte ferroptosis through mitophagy in osteoarthritis[J].J Orthop Surg Res,2023,18(1):620. [32] SINGH P,ARIF Y,BAJGUZ A,et al.The role of quercetin in plants[J].Plant Physiol Biochem,2021,166:10-19. [33] DE FIGUEIREDO COSTA A C,DE SOUSA L M,DOS SANTOS ALVES J M,et al.Anti-inflammatory and hepatoprotective effects of quercetin in an experimental model of rheumatoid arthritis[J].Inflammation,2021,44(5):2033-2043. [34] HAN X J,XU T S,FANG Q J,et al.Quercetin hinders microglial activation to alleviate neurotoxicity via the interplay between NLRP3 inflammasome and mitophagy[J].Redox Biol,2021,44:102010. [35] RATAN Z A,HAIDERE M F,HONG Y H,et al.Pharmacological potential of ginseng and its major component ginsenosides[J].J Ginseng Res,2021,45(2):199-210. [36] ZHANG M,REN H W,LI K,et al.Therapeutic effect of various ginsenosides on rheumatoid arthritis[J].BMC Complement Med Ther,2021,21(1):149. [37] CHEN H,ZHAO D,LIU S,et al.Ginsenoside Rh1 attenuates chondrocyte senescence and osteoarthritis via AMPK/PINK1/Parkin-mediated mitophagy[J].International Immunopharmacology,2025,159:114911. [38] BUHRMANN C,BROCKMUELLER A,MUELLER A L,et al.Curcumin attenuates environment-derived osteoarthritis by Sox9/NF-kB signaling axis[J].Int J Mol Sci,2021,22(14):7645. [39] ZHANG C F,WENG Y Y,WANG H B,et al.A synergistic effect of triptolide and curcumin on rheumatoid arthritis by improving cell proliferation and inducing cell apoptosis via inhibition of the IL-17/NF-κB signaling pathway[J].Int Immunopharmacol,2024,142(Pt A):112953. [40] MOHAMMADIAN HAFTCHESHMEH S,KHOSROJERDI A,ALIABADI A,et al.Immunomodulatory effects of curcumin in rheumatoid arthritis:evidence from molecular mechanisms to clinical outcomes[J].Rev Physiol Biochem Pharmacol,2021,179:1-29. [41] JIN Z Z,CHANG B H,WEI Y L,et al.Curcumin exerts chondroprotective effects against osteoarthritis by promoting AMPK/PINK1/Parkin-mediated mitophagy[J].Biomed Pharmacother,2022,151:113092. [42] BU Y H,WU H,DENG R,et al.The anti-angiogenesis mechanism of Geniposide on rheumatoid arthritis is related to the regulation of PTEN[J].Inflammopharmacology,2022,30(3):1047-1062. [43] SUN M H,DENG R,WANG Y,et al.Sphingosine kinase 1/sphingosine 1-phosphate/sphingosine 1-phosphate receptor 1 pathway:a novel target of geniposide to inhibit angiogenesis[J].Life Sci,2020,256:117988. [44] 邓然.JNK介导BNIP3过度磷酸化对缺氧下类风湿关节炎成纤维样滑膜细胞线粒体自噬/凋亡调节机制及栀子苷的干预[D].合肥:安徽中医药大学,2022. [45] 李楠,杨海芯,曾珊,等.桂枝芍药知母汤对类风湿关节炎Th17/Treg细胞失衡及JAK2/STAT3信号通路的影响[J].中华中医药杂志,2023,38(6):2567-2571. [46] WEI S J,ZHANG Q,XIANG Y J,et al.Guizhi-Shaoyao-Zhimu decoction attenuates bone erosion in rats that have collagen-induced arthritis via modulating NF-κB signalling to suppress osteoclastogenesis[J].Pharm Biol,2021,59(1):262-274. [47] YAO H,XIANG L,HUANG Y C,et al.Guizhi Shaoyao Zhimu granules attenuate bone destruction in mice with collagen-induced arthritis by promoting mitophagy of osteoclast precursors to inhibit osteoclastogenesis[J].Phytomedicine,2023,118:154967. [48] 刘天阳,周学平,黄传兵,等.清络通痹方调控类风湿关节炎“免疫-骨侵蚀”的机制[J].南方医科大学学报,2023,43(10):1706-1714. [49] ZHANG L L,ZHOU J,FENG Z,et al.Qingluo Tongbi formula alleviates hepatotoxicity induced by Tripterygium wilfordii hook.F.by regulating excessive mitophagy through the PERK-ATF4 pathway[J].Front Pharmacol,2022,13:918466. [50] 展俊平,孟庆良,孟婉婷,等.补阳还五汤对类风湿关节炎小鼠MPO,NE mRNA表达及TNF-α,IL-1β,IL-6,IL-17的影响[J].中国实验方剂学杂志,2018,24(24):158-163. [51] 展俊平,黄硕,孟庆良,等.缺氧微环境下补阳还五汤通过抑制BNIP3-PI3K/Akt通路抑制类风湿关节炎滑膜成纤维细胞的线粒体自噬[J].南方医科大学学报,2025,45(1):35-42. [52] ELDEEB M A,THOMAS R A,RAGHEB M A,et al.Mitochondrial quality control in health and in Parkinson’s disease[J].Physiol Rev,2022,102(4):1721-1755. [53] PRASUN P.Mitochondrial dysfunction in metabolic syndrome[J].Biochim Biophys Acta Mol Basis Dis,2020,1866(10):165838. 文章编号： 1006-3250(2026)03-0705-06 原文下载： 王煜,邓慧榕,张宇翔,等.线粒体自噬在类风湿关节炎病理机制中的作用及中医药干预研究进展 .发表于《中国中医基础医学杂志》2026年第32卷第3期 编  辑：黄晓华     审  核：于    峥