From traditional animal experiments to new generation alternative technologies: opportunities and challenges
-
摘要:
当前生物医药研究正处于从传统动物实验向动物实验与新替代技术共存的多元化体系转型的关键时期。传统动物实验对推动人类健康事业发展起着至关重要的作用，但随着科技进步与人文理念发展，动物模型存在的物种间生物学差异、研发效率偏低、成本较高、伦理争议等短板日益凸显。在此背景下，类器官、器官芯片等体外仿生模型与人工智能驱动的虚拟模型等新型替代技术凭借贴合人体生理病理特征、实验通量高等优势成为生物医药研究的新兴助力，但类器官标准化质量控制体系的建立、虚拟模型数据偶联与算法可解释性的优化以及风险控制盲区的全覆盖等方面仍存在诸多待解难题。本文在系统分析传统动物实验优势与局限性的基础上，探讨类器官、器官芯片、虚拟模型等替代技术在弥补传统动物模型局限性中的价值以及未来发展面临的挑战，为促进多元化模型研究技术融合发展、助力人类健康事业进步提供参考。
Abstract:Biomedical research is currently in a critical period of transition from traditional animal experiments toward a diversified system combining animal experiments and novel alternative technologies. Conventional animal experiments have historically played a pivotal role in advancing human health. However, with advances in science and technology and evolving ethical considerations, their limitations, including interspecies biological differences, low R&D efficiency, high costs, and ethical controversies, have become increasingly prominent. In this context, emerging alternative technologies such as in vitro biomimetic models (e.g., organoids and organ-on-a-chip systems) and artificial intelligence-driven virtual models have shown great promise in biomedical research, offering advantages such as high physiological and pathological fidelity to human systems and high experimental throughput. Nevertheless, significant challenges remain to be addressed, including the establishment of standardized quality control systems for organoids, optimization of data integration and algorithm interpretability in virtual models, and identification of unaddressed risks. Based on a systematic analysis of the advantages and limitations of traditional animal experiments, this paper explores the value of organoids, organ-on-a-chip systems, and virtual models in addressing the limitations of conventional animal models and the challenges they face in future development, so as to promote the integration and advancement of research technologies based on diverse models and contribute to the advancement of human health.
-
Keywords:
- animal experiments /
- organoids /
- organ-on-a-chip /
- virtual models
-
长久以来，传统动物实验凭借实验动物与人类相近的组织结构与生理功能、可模拟各类疾病表型等优势，在医药研发、外科技术革新及国防卫生等领域发挥了重要作用，为人类攻克重大疾病提供了关键的数据支撑。然而，随着研究持续深入，传统动物实验的局限性日益凸显。一方面，物种间基因调控差异、生理特性及疾病机制的显著差异导致动物实验预测人类对药物的反应出现不同程度的偏差，全球约86%的候选药物因动物模型与人体的适配性差异在临床试验中失败[1]，大大增加了新药研发的时间和经济成本；另一方面，动物实验存在实验周期长、测试成本高及伦理争议等问题，加上全球监管体系对动物实验的逐步限制（如美国FDA已提出让动物研究成为临床前研究“例外而非常态”的目标）等，开发新型替代技术的呼声越来越高[2]。在新一轮工业革命的契机下，类器官及器官芯片等体外仿生模型与人工智能（artificial intelligence，AI）驱动的虚拟模型等新兴技术迅速崛起，展现出与人体生理病理特征匹配度高、检测通量高、伦理风险低等优势，成为动物模型的主流替代技术。
当前，生物医药研究正处于从传统动物实验向动物实验与新替代技术共存的多元化体系转型的关键阶段。整合实验动物的资源管控优势，结合新型替代技术高度模拟人类表型及虚拟模型的仿真预测能力，可推动人类疾病模型转型，有效提升生物医药研发质量与效率。不容忽视的是，这些新型替代技术仍面临诸多挑战，如类器官与器官芯片亟待搭建标准化质量控制体系、探索多器官之间的通讯策略，虚拟模型则受限数据偶联、算法可解释性及数据安全等因素制约[3]。本文系统分析传统动物实验的不可替代之处，探讨类器官、器官芯片、虚拟模型等新技术的发展现状，剖析其未来方向与挑战，为生物医药研究的创新发展提供切实有效的参考。
1 传统动物实验
动物实验几十年来一直是生物医学研究的基础手段，为科学家提供了理解各种生命规律、疾病机制和演变的方法。实验动物涵盖从低等到高等的多个物种，形成完整的进化梯度，为研究不同生理机制提供了丰富选择。尽管没有任何一种实验动物能完全重现人类所有特性，但每一类实验动物都能从某一角度去模拟人类的部分特征（如从解剖结构、生理特征、发育规律、行为表型等方面实现模拟），丰富多样的实验动物群体可以被看作一个庞大的模式矩阵，在不同维度上组合互补，从而高度模拟人类特征[4-14]（表 1）。此外，每种动物都有其独特的进化史，进化程度停滞或分化差异也能够为人类提供突破性的借鉴，如蝙蝠进化出独特的抗癌特性、水熊虫进化出超强的DNA修复能力，这些优势特性可以为生物医药研发带来全新的理念。
表 1 各类实验动物的特点与应用Table 1 Characteristics and applications of various laboratory animalsClassification Representative animal model Characteristic Major application Invertebrates Caenorhabditis elegans, Drosophila melanogaster Short life cycle and well-characterized genetic background Gene function screening and mechanistic studies of neurodegenerative diseases [4] Fish Zebrafish Transparent embryos and rapid embryonic development High-throughput preliminary screening for embryotoxicity [5] Amphibians and reptiles Xenopus laevis Relatively short life cycle Studies of early embryonic development, gene regulation, and primitive neural reflexes [6]> Mammals Mice and rats Well-defined genetic background, mature genetic engineering technologies, short breeding cycle,large litter size, and well-established industrial breeding infrastructure Widely used for mechanistic studies of cancer, metabolic disorders, and neurological diseases, as well as antibody discovery and drug development [7] Rabbits Herbivorous physiology, relatively large body surface area, unique anatomical features, and complex physiological systems Studies of cardiovascular disease mechanisms,antibody production, skin injury and wound healing, surgical training, and basic medical physiology research [8-9] Miniature pigs Anatomical structures, skin, and cardiovascular system highly resemble their human counterparts,with high tissue compatibility Research on diabetic complications and osteosarcoma metastasis, surgical training, and organ transplantation [10] Dogs Long lifespan, human-like anatomy, and complex physiological systems complex physiological systems Studies of higher nervous system functions and long-term drug toxicity testing [11] Ferrets Brain development and neurological and respiratory disease phenotypes closely resemble their human counterparts Research on neurological and respiratory diseases and preclinical drug development [12] Rhesus macaques and Cynomolgus macaques Nervous and immune systems closely resemble their humans counterarts Gold-standard models for neurological disease research and vaccine evaluation [13-14] 动物实验规模自21世纪起呈现爆发式增长态势，全球每年约有1亿只动物被用于实验[15]，推动了生物医学飞速发展。实验动物主要用于药品及化妆品毒性测试[16]、生物学特性研究及人类疾病动物模型的构建。类器官等替代技术可高度还原人类基因特征，在新药药效测试、毒理学评估等方面已展现出替代动物实验的巨大潜力；但在全身代谢、整体免疫应答、高级神经调控、生殖发育等领域，动物实验仍具有极强的不可替代性。然而，我们也必须正视动物实验暴露出的一些问题。
其一是科学可靠性问题。动物与人类的相似性并不完全重叠，两者的差异部分也是导致动物实验结果难以在临床试验中完全再现的重要原因。在全球范围内，平均有86%的候选药物在人类临床试验中失败[1, 17]。很多对动物有效的药物在人体无效甚至致畸致毒[1]，历史上著名的“反应停”（沙利度胺，孕妇服用后导致胎儿短肢畸形）事件，就与所选用实验动物不合适密不可分。研发沙利度胺的医药公司在申请该药上市前使用妊娠大鼠进行动物实验来评估妊娠期用药风险，却忽视了大鼠与非人灵长类动物及人类之间的物种差异：大鼠体内缺少一种可将沙利度胺代谢为毒性产物的酶，因此给药后不会引起胎鼠肢体畸形。这一事件让业界深刻认识到动物与人类的物种差异会削弱毒理研究结果的可靠程度与临床参考意义。在疾病造模方面，各类实验动物模型与人类疾病在发病机制、进程、表型上均存在一定的差异，难以完全还原人类疾病的复杂病理特征[10, 13, 18-21]（表 2）。
表 2 实验动物在生物医学研究中的局限性Table 2 Limitations of experimental animals in biomedical researchDisease Commonly used animal model Limitation Neurodegenerative disease Caenorhabditis elegans, Drosophila melanogaster, zebrafish, rodents (primarily mice), non-human primates Biological processes differ substantially from those in humans; lack of homologs for certain mutant nuclear proteins; unable to fully recapitulate human pathophysiology; non-human primate models are costly and time-consuming[18] Metabolic disease Rodents, rabbits, miniature pigs Differences in blood glucose homeostasis compared with humans; disease mechanisms and progression do not fully reflect human metabolic disorders[19] Cancer Rodents (primarily mice), zebrafish, Drosophila melanogaster Small animal models have limited blood supply due to body size; substantial differences in physiology, immune responses, and genetics compared with humans[20] Mechanical trauma Rodents (primarily mice) Tissue repair capacity differs from that of humans; anatomical and physiological differences limit translational relevance; lack of standardized experimental designs and protocols[21] Organ transplantation Miniature pigs, non-human primates Risk of immune rejection; potential transmission of infectious diseases; ethical concerns[10] Psychiatric disorder Rodents, non-human primates Unable to fully model higher-order human emotions and complex psychiatric phenotypes; high cost and long experimental duration[13] 其二是伦理与动物福利问题。尽管实验动物在一些领域研究中不可或缺，但动物实验仍存在伦理问题和特定局限。在实验处理方式、麻醉方式、组织取材或指标监测等环节，都有可能使动物遭受身体、心理和生理上的痛苦与困扰[22]。国际动物福利规范的要求不断提高，非人灵长类动物具备复杂的认知、情感与社交能力，使得相关实验需要更高的伦理审查标准；人源化动物虽然可以进一步模拟人类生理环境，但存在“跨物种边界”伦理风险。因此，从伦理角度，学界提出了3R原则，试图减少（reduction）、优化（refinement）并取代（replacement）动物实验，推动实验体系革新[23]。
其三是动物实验周期长、成本高昂，易延缓药物研究进程[24]。例如，非人灵长类动物繁殖周期长、繁殖率低，人工饲养成本极高，导致实验用非人灵长类动物供给侧严重失衡，难以满足各类科研项目的需求；构建稳定遗传的基因工程动物需数月甚至数年，流程较为烦琐，并且依赖多代繁育及基因表型鉴定，拖慢整体研发进度；慢性疾病（如神经退行性疾病、代谢性疾病等）模型制备及实验周期普遍较长；药物非临床研究质量管理规范（good laboratory practice，GLP）级动物设施建设、SPF级实验环境维护、专业技术人员配置等投入巨大。
2 动物实验替代技术
为了突破传统动物实验在种属差异、周期成本、伦理争议等方面的瓶颈，多种替代技术逐步兴起，以类器官为主，结合器官芯片、数字孪生虚拟模型，共同构建起更加精准、高效的生物医学研究体系。
2.1 类器官及器官芯片
2.1.1 类器官
类器官是一类由干细胞在三维培养体系中通过自组织形成的结构，能够展现出对应器官的特异性功能与结构特征[25-26]。类器官通常由胚胎干细胞、诱导多能干细胞和成体干细胞建立（图 1），既可以用于研究器官发育、组织再生、疾病发生过程，也可用于高通量药物筛选。目前类器官已经在消化、呼吸、神经、泌尿及代谢系统等疾病中得到应用[27-33]（表 3）。类器官技术在一些罕见病如渐冻症和孤独症研究领域也展示出广阔的应用前景[34]，可以有效解决人源样本来源受限和物种差异的难题。
图 1 类器官的建立与应用Fig. 1 Establishment and application of organoids
下载:
全尺寸图片
表 3 iPSC/hESC来源多疾病类器官模型的优势与局限对比Table 3 Comparison of advantages and limitations of iPSC/hESC-derived organoid models for multiple diseasesDisease Cell source Advantage Limitation Cardiovascular disease iPSC-derived cardiomyocytes[27] Highly valuable for investigating disease mechanisms; well suited for integration with image-based deep learning or machine learning analyses Limited numbers of transplanted cardiomyocytes; potential risk of graft-associated ventricular arrhythmias Cancer Cells derived from various tumor tissues[28] Effective for studying human cancer syndromes; serves as a valuable complement to conventional cancer cell line models Successful engraftment is required, posing technical challenges; variable growth rates may compromise study reproducibility and reliability Cystic fibrosis iPSCs[29] Facilitates investigation of genotype-phenotype relationships and enables identification of regulatory factors independent of airway fluid secretion Cellular phenotypes are influenced by the origin of the donor cells Polycystic kidney disease H9 hESC line[30] Suitable for investigating cyst initiation and progression in ADPKD; supports multi-omics analyses to elucidate disease mechanisms and enables high-throughput drug screening Findings still require further in vivo validation to confirm physiological relevance Nonalcoholic fatty liver disease Human iPSCs[31] Faithfully recapitulates key pathological features of nonalcoholic steatohepatitis, including steatosis, inflammation, and fibrosis; exhibits high reproducibility and provides a physiologically relevant platform for drug development Transcriptomic profiles of lipid metabolism-related genes are relatively consistent, but their functional activities remain uncertain; batch effects may influence phenotypic analyses Alzheimer’s disease iPSC-derived brain cells[32] Recapitulates key features of the human brain and provides functional brain models incorporating the blood-brain barrier Limited consistency and reproducibility; clonal variability has been reported even among clones derived from the same parental iPSC line Hypothyroidism hESCs[33] Can be used to screen drugs or cell-based therapies that restore thyroid hormone secretion; provides a platform for studying thyroid development Limited differentiation efficiency and prolonged culture time before measurable outcomes; vascularization remains limited iPSC: Induced pluripotent stem cell; hESC: Human embryonic stem cell; ADPKD: Autosomal dominant polycystic kidney disease. 近年来类器官在抗肿瘤药物敏感性筛选、单基因遗传病机制研究、药物毒性评价方面的应用已相对成熟，可部分替代传统动物实验。患者来源肿瘤类器官的体外药敏检测结果与临床疗效一致性可达80%~95%[35]，已规模化应用于结直肠癌、肺癌、胰腺癌等，可将传统动物实验长达数月的药物筛选周期缩短至数周，同时大幅降低测试成本[36-38]。在药物毒性评价方面，类器官相较于传统动物的优势在于药物代谢过程与人类的高度相似性，例如用肝类器官研究替代大鼠/犬肝毒性实验，其结果更接近药物在人体的代谢与转运过程。值得一提的是，美国FDA已认可将肝类器官用于新药肝毒性筛查，其对肝毒性化合物识别率达到87%，还成功识别出多种动物实验未预警却在临床阶段暴露出肝毒性风险的药物[1]。
2.1.2 器官芯片
与类器官的组织培养模拟不同，器官芯片是一种体外微生理系统，通过准确整合活体器官的关键微环境参数，包括生化因素、物理因素、多个器官之间的系统相互作用或器官基质，实现“合成器官群”。目前，呼吸肺、蠕动肠道、节律心脏、折叠脑，甚至身体轴（如肝-胰岛轴、肠-肾轴和肠道-微生物-脑轴）已被开发出来，用于模拟人体生理病理生态系统，包括器官生成、宿主免疫反应以及药物-器官相互作用[39-40]。器官芯片技术的重大进展促进了类体内三维器官模型的构建，这些器官模型能够重现从早期发育到成熟阶段的人类健康和疾病状态，模拟生理与病理特征。
器官芯片可用于研究人类胚胎、胎盘和生殖器官的早期发育，有助于揭示生命的起源。有研究团队以工程化方式模拟早期胚胎植入过程，使干细胞在芯片特定微结构内定向排列，这种器官芯片可附着于模拟子宫壁的三维凝胶支架上，形成在持续营养供应下的体外胚胎模型。令人惊讶的是，器官芯片模型能够以超过90%的效率生成类似胚胎的结构，这远远超出了传统技术（不到5%）[41]。这一突破凸显了微流控芯片的独特优势，进一步加深了对人类胚胎学和生殖相关研究的认识。迄今为止，研究人员已利用器官芯片技术构建出多种人体组织模型，如大脑、肺、肠、肝脏、肾脏、胰岛、心脏和软骨等[42-46]。
2.1.3 器官芯片与类器官的协同应用
通过器官芯片可以提升类器官的成熟度与稳定性，弥补类器官营养分布不均、代谢废物堆积等问题。微流控技术能够实现精准流体灌注、机械力加载、氧气梯度调控、多细胞共培养分区，使类器官形成更复杂的组织结构，提升类器官的生理功能和体内相似度。将器官芯片与类器官技术协同应用，可以实时定量监测类器官电信号、代谢产物、细胞凋亡与药物响应情况，将类器官从静态终点模型升级为动态模型，显示出生物医学应用的巨大潜力。越来越多的研究人员在多器官芯片构建技术与细胞类型依赖性相关领域深耕。例如，通过血管内皮芯片与患者肿瘤类器官共培养能够动态模拟药物向肿瘤富集的过程，可直观观测到药物的杀伤效应，相较单独的肿瘤类器官，能够更加精准地评估载体靶向递送效率[47]。再者有研究人员在肝脏、肿瘤等器官芯片中加入免疫类器官，通过协同应用研究免疫与代谢的交互作用，在炎症代谢紊乱与微环境代谢重编程方面进行了深入研究[48]。
2.2 虚拟模型
动物实验种间差异大、周期长、成本高、伦理受限，而类器官、器官芯片虽解决了物种差异，但仍然受限于标准化不足、免疫缺失、通量低等问题。在此背景下，以计算模拟为核心的虚拟模型逐步兴起，虚拟模型通过整合多组学数据与AI，构建可模拟生物过程的数字化系统，突破体外模型微环境还原度不足、通量低、可重复性不佳等问题，可实现个性化预测与分子机制推演。虚拟模型的出现预示未来药物敏感性评估、毒性筛查等实验研究领域将逐步向数字化、软件化仿真转型。
目前虚拟模型研究主要为AI虚拟细胞（artificial intelligence virtual cell，AIVC）与虚拟类器官（virtual organoid，VO）。AIVC是一个以单细胞测序、空间转录组及多组学等数据为输入，依托AI算法构建的具有时空演化特征的动态细胞数字孪生模型，用于模拟细胞功能状态、信号网络及其在不同扰动下的动态。AIVC通过支持功能反应的机制性预测模拟，为生命科学研究带来了颠覆性的前景。与传统计算细胞模型相比，AIVC通过大规模多模态生物数据挖掘潜在生物学规律，构建具有预测能力的细胞状态空间[49]，它的出现正在将生物医学从依赖实验室的验证模式转向数据和预测驱动的仿真-验证闭环体系[50]。
在临床前药物研究中，虚拟细胞已展示出多样应用：通过模拟基因敲除、过表达或突变，预测相关基因在疾病进展中的功能作用，并结合成簇规律间隔短回文重复序列（clustered regularly interspaced short palindromic repeats，CRISPR）验证，帮助筛选潜在药物靶点[51]；深入分析药物-细胞相互作用过程，辅助药物作用机制推断[50]；进行高通量的计算机药物疗效预测，评估选择性和潜在不良反应[9]。中国科学院李鑫团队基于1.26亿单细胞数据构建了全球首个跨物种生命基础模型GeneCompass，破译了通用基因调控机制，模型可应用于药物靶点的发现[52]。随着关注度的提高，虚拟细胞领域不断出现技术突破。同济大学刘琦团队深耕虚拟细胞算法体系，自主构建了AlphaCell虚拟细胞世界模型，可高精度模拟细胞动态变化，并预测细胞对外部微环境的响应规律[53]；团队联合微软亚洲研究院开发的CausCell框架创新性引入因果解耦建模思路，突破了传统AI黑箱不可解释的局限，实现基因扰动、信号通路调控效应的可解释预测[54]。近日，美国伊利诺伊大学厄巴纳-香槟分校Thornburg等[55]构建了四维空间动力学全细胞仿真模型，完整数字化复刻JCVI-syn3A最小细菌的全生命周期，融合多尺度混合计算框架精准还原代谢、基因表达与染色体动态过程，可定量预测细胞随机异质性与基因扰动表型，为全细胞虚拟仿真体系建立了分子级时空建模范式。
VO则是在虚拟细胞基础上进一步升级，通过整合类器官的结构数据、生理功能数据，结合AI算法构建与真实类器官高度相似的虚拟模型，能够实现对器官生理状态、病理变化及药物干预效果的快速模拟与预测。相较于虚拟细胞的单细胞层面模拟，VO可容纳多种细胞类型，能够模拟细胞间通讯、旁分泌信号传递、肿瘤微环境相互作用等复杂的生物学调控机制[35]。更重要的是，VO可以直接对应患者来源的类器官，构建出患者特异性数字模型，这将是基础研究到精准医疗的关键转化桥梁。例如，已有研究人员利用VO模拟患者胰腺癌模型进行化疗药物与靶向药物的预测[56]。但需要注意的是，VO在整合多模态数据时仍面临算法瓶颈：首先，不同来源数据的分辨率、时空尺度差异大，而目前又缺乏通用的对齐方法；其次，数据的整合往往有着极大的数据量，导致模型训练成本高昂，同时AI“黑箱”特性也影响模型的可信度；最后，目前类器官研究中高质量的纵向数据还十分稀缺。
虚拟模型的使用应始终与生物实验紧密结合。短期内，虚拟模型仅能提供预测性佐证，所有仿真衍生假说必须经过实验验证，并利用实验反馈不断完善模型，提升其与现实的一致性。未来的方向应是构建“预测-实验验证-模型修正-再预测”的环路，这一闭环过程将推动药物筛选和机制分析的真正精准度，使海量数据高效转化为知识发现。
3 未来发展方向
动物实验未来的发展并非被完全替代，而是在伦理约束与技术革新的双重驱动下，实现自身优化与协同赋能的转型，最终与类器官、虚拟模型等技术形成互补，共同支撑生物医药研究的高质量发展，既兼顾科学研究的需求，又回应伦理与社会诉求，实现科学价值、伦理价值与临床价值的完美统一。
立足于服务人类健康事业发展的长远目标，结合全球政策导向、伦理诉求、技术革新及生物医药研究的核心需求，动物实验未来将精准应用于替代技术无法模拟的领域，如机体特有生物表型挖掘、完整器官的血管与神经网络调控、内分泌与生殖系统的相互作用、组织衰老等生理过程，承担“源头发现”与“决赛角逐”的角色。
类器官与器官芯片作为当下最具应用前景的体外替代技术，其核心痛点为缺少标准化的质量控制评价体系，细胞来源、培养基成分和处理方案的细微差异可能导致技术差异不可控，难以实现跨实验室数据互通互认。与此同时，配套的外部因素如支架材料的生物活性参数、细胞类型和数量比例与原生组织之间的差异，以及空间组织不完整等局限性，在一定程度上限制了类器官与器官芯片在科研工作中的普遍应用[57]。因此，建立全球标准的类器官、器官芯片质量控制体系是类器官健康持续发展的重要方向。欧盟HCA Organoid项目已提出构建人类类器官图谱，提供标准化、大规模的数据基线。未来类器官研究应加速与器官芯片技术融合，构建兼具类器官结构保真与微环境精准调控能力的一体化体外仿生平台；而器官芯片研究则应聚焦单器官芯片高仿真与多器官联动方面，从而更贴近人体生理病理微环境。AI技术也将在类器官、器官芯片领域发挥重要作用，将AI系统嵌入实验流程中，用机器学习模型加速构建培养筛选体系，例如从现有数据预测合适的水凝胶组成和培养参数，从明场和荧光图像中提取多尺度特征量化细胞形态、状态和分化结果等[58]，这将提高类器官、器官芯片相关实验的可靠性和重复性，并扩展其应用。
作为AI赋能的新型体外替代技术，虚拟模型未来的发展核心是实现精准化、协同化、规范化、临床化。虽然虚拟模型展现出巨大的潜力，但其需要高质量多模态的数据支持，同时模型的可解释性和透明度直接影响其在临床环境中的推广度[59]。未来需要通过数据驱动与技术迭代弥补这些短板，实现虚拟模型的高质量发展。从定位来看，虚拟模型研究的核心价值不在于取代类器官技术，而是从数字维度对体外模型进行延伸与赋能，两者虚实结合、协同互补，最终服务于提高药物研发效率、提升临床转化成功率、推动精准医学落地这一根本目标。未来药物筛选类实验的前景将是：依托类器官或实体动物产生真实数据，利用该数据构建虚拟模型，借助AI优化实验技术参数，再将仿真推演得到的候选方案回归到类器官或实体动物水平进行验证。虚拟模型应逐步通过标准化建设推动行业发展与临床转化，最终成为生物医药研究高质量发展的核心支撑，助力新药研发质效双升、精准医疗落地应用。
未来在临床药物研发中，类器官、器官芯片、虚拟模型及动物实验将协同互补、各司其职：类器官聚焦临床相关性验证，可挖掘验证疾病靶点、评估药物药效并开展患者特异性药敏测试；器官芯片承担候选药物中通量验证与药物毒性早期筛查的角色；虚拟模型依托多模态数据与AI算法，实现靶点成药性预测、化合物高通量虚拟筛选及药效毒性精准预测；而动物实验作为临床前的最终验证，负责确认药物体内药效、评估毒性并验证作用机制。虚拟模型前置高效筛选，类器官与器官芯片分层验证，动物实验最终把关，同时实验结果反向校准虚拟模型，持续提升研发效率与精准度，推动药物从基础研究向临床转化。
与此同时，应加强法律监管，在新的替代技术广泛应用于生物医学领域的同时规避新的伦理及法律风险。替代技术未来任重而道远，若想在GLP体系中彻底取代动物实验，还需要在模型制备、性能验证、实验操作、数据管理等维度建立可量化、可复现、可追溯的标准化参数：模型制备需明确细胞来源质控、制备流程及形态结构标准；模型性能需满足生理功能、病理模拟、重复性及药物响应与人体和动物实验的一致性要求；实验操作需规范设备试剂质控、流程设计及人员环境管理，符合GLP合规要求；数据管理需遵循ALCOA+原则，规范采集、处理、存储与归档，最终才能实现对动物实验的有效替代。总体来看，全球监管趋势对AI驱动的替代方法持谨慎支持态度：一方面认识到其在加速药物开发和减少动物使用方面的潜力，另一方面通过立法和指南确保其可靠性和透明度[60]。2025年，美国FDA发布了关于药物开发中使用AI模型的首份草案指导，提出了基于风险的模型可信度评估框架[61]。近年来欧洲化学品管理局依托《化学品注册、评估、授权与限制法规》，持续推动非动物替代实验方法的应用，部分毒理评估框架采用了计算建模和生物信息学预测工具。我国《“十五五”规划纲要（草案）》已将类器官、器官芯片列为国家前沿攻关技术。后续可吸收国外成熟监管经验，立足本土生物医药产业实情，制定特色化监管规则；依托规范化制度约束，助力科研人员合规开展研究，提升替代技术在生物医学及创新疗法研发中的应用成功率。
4 小结
生物医学正朝着精准化、高效化与智能化方向迭代发展，其稳步推进与落地应用亟需技术创新、伦理治理与监管体系的全方位支撑。在技术层面，通过加强生物医学研究者、临床医生与计算生物学家之间的合作，依托高质量生物医学数据优化智慧治疗策略，推动疾病进展预测、疗效评估和生物标志物发现更趋准确，不断优化替代技术的种类和标准化质量控制体系，实现生物医学研究质效提升；从政策法规及伦理层面，则需建立前瞻性的动态监测机制，确保多重替代技术体系的安全性和可持续发展，实现各类模型融合共生，最终服务于人类卫生健康事业的持续健康发展。
-
图 1 类器官的建立与应用
Fig. 1 Establishment and application of organoids
下载:
全尺寸图片
表 1 各类实验动物的特点与应用
Table 1 Characteristics and applications of various laboratory animals
Classification Representative animal model Characteristic Major application Invertebrates Caenorhabditis elegans, Drosophila melanogaster Short life cycle and well-characterized genetic background Gene function screening and mechanistic studies of neurodegenerative diseases [4] Fish Zebrafish Transparent embryos and rapid embryonic development High-throughput preliminary screening for embryotoxicity [5] Amphibians and reptiles Xenopus laevis Relatively short life cycle Studies of early embryonic development, gene regulation, and primitive neural reflexes [6]> Mammals Mice and rats Well-defined genetic background, mature genetic engineering technologies, short breeding cycle,large litter size, and well-established industrial breeding infrastructure Widely used for mechanistic studies of cancer, metabolic disorders, and neurological diseases, as well as antibody discovery and drug development [7] Rabbits Herbivorous physiology, relatively large body surface area, unique anatomical features, and complex physiological systems Studies of cardiovascular disease mechanisms,antibody production, skin injury and wound healing, surgical training, and basic medical physiology research [8-9] Miniature pigs Anatomical structures, skin, and cardiovascular system highly resemble their human counterparts,with high tissue compatibility Research on diabetic complications and osteosarcoma metastasis, surgical training, and organ transplantation [10] Dogs Long lifespan, human-like anatomy, and complex physiological systems complex physiological systems Studies of higher nervous system functions and long-term drug toxicity testing [11] Ferrets Brain development and neurological and respiratory disease phenotypes closely resemble their human counterparts Research on neurological and respiratory diseases and preclinical drug development [12] Rhesus macaques and Cynomolgus macaques Nervous and immune systems closely resemble their humans counterarts Gold-standard models for neurological disease research and vaccine evaluation [13-14] 表 2 实验动物在生物医学研究中的局限性
Table 2 Limitations of experimental animals in biomedical research
Disease Commonly used animal model Limitation Neurodegenerative disease Caenorhabditis elegans, Drosophila melanogaster, zebrafish, rodents (primarily mice), non-human primates Biological processes differ substantially from those in humans; lack of homologs for certain mutant nuclear proteins; unable to fully recapitulate human pathophysiology; non-human primate models are costly and time-consuming[18] Metabolic disease Rodents, rabbits, miniature pigs Differences in blood glucose homeostasis compared with humans; disease mechanisms and progression do not fully reflect human metabolic disorders[19] Cancer Rodents (primarily mice), zebrafish, Drosophila melanogaster Small animal models have limited blood supply due to body size; substantial differences in physiology, immune responses, and genetics compared with humans[20] Mechanical trauma Rodents (primarily mice) Tissue repair capacity differs from that of humans; anatomical and physiological differences limit translational relevance; lack of standardized experimental designs and protocols[21] Organ transplantation Miniature pigs, non-human primates Risk of immune rejection; potential transmission of infectious diseases; ethical concerns[10] Psychiatric disorder Rodents, non-human primates Unable to fully model higher-order human emotions and complex psychiatric phenotypes; high cost and long experimental duration[13] 表 3 iPSC/hESC来源多疾病类器官模型的优势与局限对比
Table 3 Comparison of advantages and limitations of iPSC/hESC-derived organoid models for multiple diseases
Disease Cell source Advantage Limitation Cardiovascular disease iPSC-derived cardiomyocytes[27] Highly valuable for investigating disease mechanisms; well suited for integration with image-based deep learning or machine learning analyses Limited numbers of transplanted cardiomyocytes; potential risk of graft-associated ventricular arrhythmias Cancer Cells derived from various tumor tissues[28] Effective for studying human cancer syndromes; serves as a valuable complement to conventional cancer cell line models Successful engraftment is required, posing technical challenges; variable growth rates may compromise study reproducibility and reliability Cystic fibrosis iPSCs[29] Facilitates investigation of genotype-phenotype relationships and enables identification of regulatory factors independent of airway fluid secretion Cellular phenotypes are influenced by the origin of the donor cells Polycystic kidney disease H9 hESC line[30] Suitable for investigating cyst initiation and progression in ADPKD; supports multi-omics analyses to elucidate disease mechanisms and enables high-throughput drug screening Findings still require further in vivo validation to confirm physiological relevance Nonalcoholic fatty liver disease Human iPSCs[31] Faithfully recapitulates key pathological features of nonalcoholic steatohepatitis, including steatosis, inflammation, and fibrosis; exhibits high reproducibility and provides a physiologically relevant platform for drug development Transcriptomic profiles of lipid metabolism-related genes are relatively consistent, but their functional activities remain uncertain; batch effects may influence phenotypic analyses Alzheimer’s disease iPSC-derived brain cells[32] Recapitulates key features of the human brain and provides functional brain models incorporating the blood-brain barrier Limited consistency and reproducibility; clonal variability has been reported even among clones derived from the same parental iPSC line Hypothyroidism hESCs[33] Can be used to screen drugs or cell-based therapies that restore thyroid hormone secretion; provides a platform for studying thyroid development Limited differentiation efficiency and prolonged culture time before measurable outcomes; vascularization remains limited iPSC: Induced pluripotent stem cell; hESC: Human embryonic stem cell; ADPKD: Autosomal dominant polycystic kidney disease. -
[1] Atkins J T, George G C, Hess K, et al. Pre-clinical animal models are poor predictors of human toxicities in phase 1 oncology clinical trials[J]. Br J Cancer, 2020, 123(10): 1496-1501. DOI: 10.1038/s41416-020-01033-x. [2] Kwon D. The age of animal experiments is waning. Where will science go next?[J]. Nature, 2026, 650(8103): 812-814. DOI: 10.1038/d41586-026-00563-3. [3] Schuhmacher A, Hinder M, Brief E, et al. Benchmarking R&D success rates of leading pharmaceutical companies: an empirical analysis of FDA approvals (2006-2022)[J]. Drug Discov Today, 2025, 30(2): 104291. DOI: 10.1016/j.drudis.2025.104291. [4] Zhang Q, Dang W, Wang M C. Lysosomes signal through the epigenome to regulate longevity across generations[J]. Science, 2025, 389(6767): 1353-1360. DOI: 10.1126/science.adn8754. [5] Fang F, Zhu Y, Xu W, et al. Insights into the developmental and cardiovascular toxicity of bixafen using zebrafish embryos and larvae[J]. Environ Res, 2024, 262(Pt 2): 119916. DOI: 10.1016/j.envres.2024.119916. [6] Cui H, Liang W, Shi Z, et al. CANTAC-seq analysis reveals E2f1 and Otx1 coordinate zygotic genome activation in Xenopus tropicalis[J]. Nat Commun, 2025, 16(1): 10101. DOI: 10.1038/s41467-025-65146-8. [7] Batsios G, Taglang C, Udutha S, et al. Lactylation fuels nucleotide biosynthesis and facilitates deuterium metabolic imaging of tumor proliferation in preclinical models of H3K27M-mutant gliomas[J]. Sci Transl Med, 2026, 18(836): eadw0834. DOI: 10.1126/scitranslmed.adw0834. [8] Bogoni F, Brunner M S, Almer G, et al. Homocysteine leads to aortic stiffening in a rabbit model of atherosclerosis[J]. Acta Biomater, 2025, 201: 412-428. DOI: 10.1016/j.actbio.2025.06.003. [9] Kuai M, Shi Z, Li B, et al. Rabbit monoclonal antibodies: synergistic innovation and breakthrough based on B-cell development mechanism and single B-cell technology[J]. Colloids Surf B Biointerfaces, 2026, 257: 115179. DOI: 10.1016/j.colsurfb.2025.115179. [10] Tao K S, Yang Z X, Zhang X, et al. Gene-modified pigto-human liver xenotransplantation[J]. Nature, 2025, 641(8064): 1029-1036. DOI: 10.1038/s41586-025-08799-1. [11] Ren W, Yu S, Guo K, et al. Disrupted human-dog interbrain neural coupling in autism-associated Shank3 mutant dogs[J]. Adv Sci (Weinh), 2024, 11(41): e2402493. DOI: 10.1002/advs.202402493. [12] Velazquez E M, Basu Thakur P, Brock N, et al. Establishment and validation of a ferret model for systemic antibiotic treatment during influenza A virus infection[J]. Lab Anim, 2025, 54(7): 188-194. DOI: 10.1038/s41684-025-01574-9. [13] Zhang J, Li M, Wang M, et al. Single-cell spatiotemporal transcriptomic and chromatin accessibility profiling in developing postnatal human and macaque prefrontal cortex[J]. Nat Neurosci, 2026, 29(3): 746-758. DOI: 10.1038/s41593-025-02150-7. [14] Yan X, Arcoverde Cerveira R, Ols S, et al. Biochemical and hematological reference intervals in Rhesus and Cynomolgus macaques and implications for vaccine and drug development[J]. Lab Anim, 2025, 54(6): 141-155. DOI: 10.1038/s41684-025-01547-y. [15] Ericsson Aaron C, Crim Marcus J, Franklin Craig L. A brief history of animal modeling[J]. Mo Med, 110(3): 201-205. [16] McNamee P, Hibatallah J, Costabel-Farkas M, et al. A tiered approach to the use of alternatives to animal testing for the safety assessment of cosmetics: eye irritation[J]. Regul Toxicol Pharmacol, 2009, 54(2): 197-209. DOI: 10.1016/j.yrtph.2009.04.004. [17] Mullard A. Parsing clinical success rates[J]. Nat Rev Drug Discov, 2016, 15(7): 447. DOI: 10.1038/nrd.2016.136. [18] Khan E, Hasan I, Haque M E. Parkinson's disease: exploring different animal model systems[J]. Int J Mol Sci, 2023, 24(10): 9088. DOI: 10.3390/ijms24109088. [19] Kottaisamy C P D, Raj D S, Prasanth Kumar V, et al. Experimental animal models for diabetes and its related complications: a review[J]. Lab Anim Res, 2021, 37(1): 23. DOI: 10.1186/s42826-021-00101-4. [20] Li Z, Zheng W, Wang H, et al. Application of animal models in cancer research: recent progress and future prospects[J]. Cancer Manag Res, 2021, 13: 2455-2475. DOI: 10.2147/CMAR.S302565. [21] Ramirez S, Mukherjee A, Sepulveda S, et al. Modeling traumatic brain injury in human cerebral organoids[J]. Cells, 2021, 10(10): 2683. DOI: 10.3390/cells10102683. [22] Kiani A K, Pheby D, Henehan G, et al. Ethical considerations regarding animal experimentation[J]. J Prev Med Hyg, 2022, 63(2 Suppl 3): E255-E266. DOI: 10.15167/2421-4248/jpmh2022.63.2S3.2768. [23] Morton D B. A model framework for the estimation of animal 'suffering': its use in predicting and retrospectively assessing the impact of experiments on animals[J]. Animals, 2023, 13(5): 800. DOI: 10.3390/ani13050800. [24] Van Norman G A. Limitations of animal studies for predicting toxicity in clinical trials: part 2: potential alternatives to the use of animals in preclinical trials[J]. JACC Basic Transl Sci, 2020, 5(4): 387-397. DOI: 10.1016/j.jacbts.2020.03.010. [25] Park G, Rim Y A, Sohn Y, et al. Replacing animal testing with stem cell-organoids: advantages and limitations[J]. Stem Cell Rev Rep, 2024, 20(6): 1375-1386. DOI: 10.1007/s12015-024-10723-5. [26] Wang X. Stem cells in tissues, organoids, and cancers[J]. Cell Mol Life Sci, 2019, 76(20): 4043-4070. DOI: 10.1007/s00018-019-03199-x. [27] Hnatiuk A P, Briganti F, Staudt D W, et al. Human iPSC modeling of heart disease for drug development[J]. Cell Chem Biol, 2021, 28(3): 271-282. DOI: 10.1016/j.chembiol.2021.02.016. [28] Aboul-Soud M A M, Alzahrani A J, Mahmoud A. Induced pluripotent stem cells (iPSCs): roles in regenerative therapies, disease modelling and drug screening[J]. Cells, 2021, 10(9): 2319. DOI: 10.3390/cells10092319. [29] Berical A, Lee R E, Lu J, et al. A multimodal iPSC platform for cystic fibrosis drug testing[J]. Nat Commun, 2022, 13(1): 4270. DOI: 10.1038/s41467-022-31854-8. [30] Tran T, Song C J, Nguyen T, et al. A scalable organoid model of human autosomal dominant polycystic kidney disease for disease mechanism and drug discovery[J]. Cell Stem Cell, 2022, 29(7): 1083-1101.e7. DOI: 10.1016/j.stem.2022.06.005. [31] Ouchi R, Togo S, Kimura M, et al. Modeling steatohepatitis in humans with pluripotent stem cellderived organoids[J]. Cell Metab, 2019, 30(2): 374-384.e6. DOI: 10.1016/j.cmet.2019.05.007. [32] Wang Y, Liang G, Liang S, et al. Dantrolene ameliorates impaired neurogenesis and synaptogenesis in induced pluripotent stem cell lines derived from patients with Alzheimer's disease[J]. Anesthesiology, 2020, 132(5): 1062-1079. DOI: 10.1097/ALN.0000000000003224. [33] Romitti M, Tourneur A, de Faria da Fonseca B, et al. Transplantable human thyroid organoids generated from embryonic stem cells to rescue hypothyroidism[J]. Nat Commun, 2022, 13(1): 7057. DOI: 10.1038/s41467-022-34776-7. [34] Schutgens F, Clevers H. Human organoids: tools for understanding biology and treating diseases[J]. Annu Rev Pathol Mech Dis, 2020, 15: 211-234. DOI: 10.1146/annurev-pathmechdis-012419-032611. [35] Zhou Z, Cong L, Cong X. Patient-derived organoids in precision medicine: drug screening, organoid-on-a-chip and living organoid biobank[J]. Front Oncol, 2021, 11: 762184. DOI: 10.3389/fonc.2021.762184. [36] Yao Y, Xu X, Yang L, et al. Patient-derived organoids predict chemoradiation responses of locally advanced rectal cancer[J]. Cell Stem Cell, 2020, 26(1): 17-26.e6. DOI: 10.1016/j.stem.2019.10.010. [37] Anderko R R, Bowman A E, Murthy P, et al. Pancreatic cancer organoids recapitulate chemotherapy response and identify a potent cytotoxic T-cell population[J]. Clin Cancer Res, 2026, 32(3): 596-610. DOI: 10.1158/1078-0432.CCR-25-1110. [38] Hu Y, Sui X, Song F, et al. Lung cancer organoids analyzed on microwell arrays predict drug responses of patients within a week[J]. Nat Commun, 2021, 12(1): 2581. DOI: 10.1038/s41467-021-22676-1. [39] Tao T, Deng P, Wang Y, et al. Microengineered multiorganoid system from hiPSCs to recapitulate human liver-islet axis in normal and type 2 diabetes[J]. Adv Sci (Weinh), 2022, 9(5): e2103495. DOI: 10.1002/advs.202103495. [40] Moysidou C M, Owens R M. Advances in modelling the human microbiome-gut-brain axis in vitro[J]. Biochem Soc Trans, 2021, 49(1): 187-201. DOI: 10.1042/BST20200338. [41] Zheng Y, Xue X, Shao Y, et al. Controlled modelling of human epiblast and amnion development using stem cells[J]. Nature, 2019, 573(7774): 421-425. DOI: 10.1038/s41586-019-1535-2. [42] Tuerxun K, He J, Ibrahim I, et al. Bioartificial livers: a review of their design and manufacture[J]. Biofabrication, 2022, 14(3). DOI: 10.1088/1758-5090/ac6e86. [43] Zhu Y, Zhang X, Sun L, et al. Engineering human brain assembloids by microfluidics[J]. Adv Mater, 2023, 35(14): e2210083. DOI: 10.1002/adma.202210083. [44] Zhu Y, Sun L, Wang Y, et al. A biomimetic human lungon-a-chip with colorful display of microphysiological breath[J]. Adv Mater, 2022, 34(13): e2108972. DOI: 10.1002/adma.202108972. [45] Moossavi S, Arrieta M C, Sanati-Nezhad A, et al. Guton-chip for ecological and causal human gut microbiome research[J]. Trends Microbiol, 2022, 30(8): 710-721. DOI: 10.1016/j.tim.2022.01.014. [46] Occhetta P, Mainardi A, Vo t t a E, e t a l. Hyperphysiological compression of articular cartilage induces an osteoarthritic phenotype in a cartilage-on-achip model[J]. Nat Biomed Eng, 2019, 3(7): 545-557. DOI: 10.1038/s41551-019-0406-3. [47] Mitchell C S, Yoo H I, Gracia Diaz C, et al. Human multi-organ-on-a-chip platforms for next-generation drug delivery strategies[J]. Theranostics, 2026, 16(12): 6463-6484. DOI: 10.7150/thno.130030. [48] Shroff T, Aina K, Maass C, et al. Studying metabolism with multi-organ chips: new tools for disease modelling, pharmacokinetics and pharmacodynamics[J]. Open Biol, 2022, 12(3): 210333. DOI: 10.1098/rsob.210333. [49] Bunne C, Roohani Y, Rosen Y, et al. How to build the virtual cell with artificial intelligence: priorities and opportunities[J]. Cell, 2024, 187(25): 7045-7063. DOI: 10.1016/j.cell.2024.11.015. [50] Gupta R, Srivastava D, Sahu M, et al. Artificial intelligence to deep learning: machine intelligence approach for drug discovery[J]. Mol Divers, 2021, 25(3): 1315-1360. DOI: 10.1007/s11030-021-10217-3. [51] Worth A P, Louisse J, Macko P, et al. Virtual cell based assay simulations of intra-mitochondrial concentrations in hepatocytes and cardiomyocytes[J]. Toxicol In Vitro, 2017, 45(Pt 2): 222-232. DOI: 10.1016/j.tiv.2017.09.009. [52] Yang X, Liu G, Feng G, et al. GeneCompass: deciphering universal gene regulatory mechanisms with a knowledge-informed cross-species foundation model[J]. Cell Res, 2024, 34(12): 830-845. DOI: 10.1038/s41422-024-01034-y. [53] Chuai G, Chen X, Yang X, et al. Towards building a world model to simulate perturbation-induced cellular dynamics by AlphaCell[J]. bioRxiv: 2026.03.02.709176. DOI: 10.64898/2026.03.02.709176. [54] Gao Y, Dong K, Shan C, et al. Causal disentanglement for single-cell representations and controllable counterfactual generation[J]. Nat Commun, 2025, 16(1): 6775. DOI: 10.1038/s41467-025-62008-1. [55] Thornburg Z R, Maytin A, Kwon J, et al. Bringing the genetically minimal cell to life on a computer in 4D[J]. Cell, 2026, 189(9): 2582-2597.e27. DOI: 10.1016/j.cell.2026.02.009. [56] Xu J, Yang X, Li Y, et al. ODFormer: a virtual organoid for predicting personalized therapeutic responses in pancreatic cancer[J]. bioRxiv: 2025.07.08.663664. DOI: 10.1101/2025.07.08.663664. [57] Tong T, Qi Y, Rollins D, et al. Rational design of oral drugs targeting mucosa delivery with gut organoid platforms[J]. Bioact Mater, 2023, 30: 116-128. DOI: 10.1016/j.bioactmat.2023.07.014. [58] Bai L, Wu Y, Li G, et al. AI-enabled organoids: construction, analysis, and application[J]. Bioact Mater, 2024, 31: 525-548. DOI: 10.1016/j.bioactmat.2023.09.005. [59] Nazer L H, Zatarah R, Waldrip S, et al. Bias in artificial intelligence algorithms and recommendations for mitigation[J]. PLoS Digit Health, 2023, 2(6): e0000278. DOI: 10.1371/journal.pdig.0000278. [60] Deshmukh A D, Wagner J K. FDA draft guidelines for AI and the need for ethical frameworks[J]. JAMA Pediatr, 2025, 179(9): 937-938. DOI: 10.1001/jamapediatrics.2025.1979. [61] U.S. Food & Drug Administration. FDA proposes framework to advance credibility of AI models used for drug and biological product submissions[EB/OL]. (2025-01-06)[2026-03-10].https://www.fda.gov/newsevents/press-announcements/fda-proposes-frameworkadvance-credibility-ai-models-used-drug-andbiological-product-submissions
