From Monkeys to Microchips: A New Paradigm in Drug Evaluation under FDA's Regulatory Shift
In the field of pharmaceutical R&D, every major policy change heralds a reshaping of the industry landscape and the emergence of new opportunities. Since April this year, the FDA’s series of frequent adjustments have undoubtedly accelerated the pace of this transformation.
In its regulatory adjustments, the FDA has announced a phased elimination of mandatory animal testing requirements in the development of monoclonal antibodies (mAbs) and other pharmaceuticals, shifting instead toward a suite of more effective and human-relevant New Approach Methodologies (NAMs). The agency specifically highlighted applications such as AI-based computational toxicity models and cell line- and organoid-based in vitro toxicity assays. To interpret these new international regulatory developments and their potential implications, we interviewed two former senior FDA reviewers: Dr. Du Tao, Chairman of Eglin Pharma, and Dr. Li Changqing, Chief Medical Officer.
The shift did not occur overnight. As early as 2022, the U.S. Senate passed a significant bill related to the pharmaceutical industry, the FDA Modernization Act 2.0. This legislation aimed to eliminate the mandatory requirement for animal testing in new drug development and for biosimilars. This year, lawmakers have introduced version 3.0 of the bill to urge the FDA to take action.
To ensure the orderly advancement of policies, the FDA has released a document titled “Roadmap to Reducing Animal Testing in Preclinical Safety Studies.” The roadmap clearly states that initial efforts will focus on monoclonal antibodies (mAbs), subsequently expanding to other biologics, and ultimately encompassing new chemical entities (NCEs) and medical countermeasures (MCMs). This will be a long-term process.
Under current standards, new drugs must undergo safety testing in two animal species before clinical trials, including observation of potential toxic side effects and performance at different doses.
However, animal models have several inherent and significant limitations in evaluating the safety and efficacy of biologic macromolecule drugs such as monoclonal antibodies (mAbs): substantial differences exist between animal targets and human counterparts; experiments cannot accurately predict human immune responses to mAbs; failures in predicting specific toxicities pose serious safety risks; and there is generally low concordance between preclinical animal studies and human clinical outcomes for biologics. Meanwhile, animal experimentation has long been ethically controversial and constitutes a major portion of preclinical expenditures. Under current regulations, mAbs must rely on non-human primates for long-term repeated-dose toxicity studies, with the price of experimental cynomolgus monkeys reaching as high as RMB 150,000 per animal in 2022.
Previously, when monoclonal antibodies (mAbs) lacked cross-reactivity in animal models, the FDA had effectively accepted the use of in vitro data (such as data derived from organoids) combined with a human dose-escalation strategy starting at very low doses to support clinical trials. This has also laid a solid foundation for the implementation of new regulations.
The FDA’s initiative to replace animal testing with new approaches is regarded as a “paradigm shift” in the field of drug evaluation. Through this move, the agency aims to enhance drug safety, accelerate the assessment process, reduce the use of animals, lower research and development costs, and ultimately benefit patients by reducing drug prices.
Technologies used to replace animal testing can currently be broadly categorized into In Silico NAMs and advanced in vitro NAMs.
In silico NAMs refer to a class of novel approaches that leverage computer simulations and computational methods to predict the biological effects of drugs or other chemical substances. These methods encompass a variety of technologies, including artificial intelligence (AI), machine learning (ML), quantitative structure-activity relationship (QSAR) modeling, physiologically based pharmacokinetic (PBPK) modeling, and quantitative systems pharmacology (QSP) modeling. Their core function is to predict in vivo drug behavior—including pharmacokinetics, pharmacodynamics, and potential toxicity risks—by applying mathematical models and algorithms based on chemical structures, physicochemical properties, biological data, physiological knowledge, and existing experimental data.
“The reliability of such models is highly dependent on the quality, quantity, and representativeness of the training data. To gain regulatory acceptance, compelling evidence must be provided to demonstrate their reliability and predictive performance in specific use scenarios, while clearly defining the model’s intended scope of application and conducting adequate and rigorous validation,” commented Dr. Li Changqing.
As another key pillar of the FDA’s New Approach Methodologies (NAMs) strategy, advanced in vitro models aim to better mimic the structure and function of human organs in a laboratory setting. Notably, this includes the use of lab-grown human “organoids” and “organ-on-a-chip” systems.
Organoids refer to miniature three-dimensional cell cultures formed in vitro through self-organization processes from stem cells (including induced pluripotent stem cells, iPSCs, or adult stem cells), capable of mimicking partial structural features and cellular composition of specific organs, such as the liver, intestine, and kidney.
Microphysiological Systems (MPS), also known as "organ-on-a-chip," typically refer to engineered platforms built using microfluidic technology that can simulate the dynamic microenvironment in vivo, construct tissue barrier interfaces, or be interconnected to mimic inter-organ interactions.
These in vitro models offer greater human relevance, enabling the detection of toxic effects that may be easily overlooked in animal studies, particularly in the assessment of immunotoxicity. New technologies can not only simulate the physiological functions of human organs but also investigate drug mechanisms of action at the microscopic level, thereby providing a more direct window into understanding human responses.
However, in vitro models also have their drawbacks. Issues such as the lack of stable and consistent sources, poor reproducibility, imperfect validation standards, high costs, low throughput, and inter-donor variability persist. This means that in the short term, in vitro models cannot independently conduct hazard identification or risk assessment, nor can they fully replace animal experiments.
The FDA has established mechanisms and programs to support the development and integration of new approach methodologies (NAMs), including issuing strategic plans, establishing new working groups, launching fitness-for-purpose certification programs, encouraging the submission of NAM data as supplementary information alongside traditional animal study data, developing relevant guidance documents, and participating in the establishment of global harmonized standards for NAMs.
Furthermore, new methods currently lack key specifications such as standardized operating procedures, quality control standards, key performance indicators, data reporting templates, and reference materials. This makes it difficult to directly compare data generated by different laboratories and technical platforms, posing significant challenges for validation efforts and regulatory assessment. To address these issues, the FDA plans to build an open and collaborative ecosystem through cooperation with other government agencies and industry consortia, as well as participation in international harmonization initiatives. To fully realize the potential of in vitro models, concerted efforts from academia, industry, technology vendors, and regulatory authorities are required to support infrastructure development.
In the long term, through the synergistic application of various novel methods and the establishment of more robust data linkages, it is expected to develop a holistic preclinical evaluation strategy that is more comprehensive, accurate, and human-relevant than single-method approaches or traditional animal testing. This will significantly enhance the efficiency and speed of drug development, reduce R&D costs, improve drug safety, and promote ethical progress and technological innovation.
Dr. Li Changqing pointed out that under the new evaluation paradigm, the status of data has been elevated to an unprecedented level. Whether derived from human clinical trials, real-world evidence, New Approach Methodologies (NAMs) experiments, or large-scale toxicology databases, high-quality, structured, and interpretable data will become the core “currency” driving the R&D process. The ability to effectively generate, manage, integrate, and interpret these complex datasets will directly impact the efficiency and success rate of drug development.
In the short to medium term, companies that actively invest in research on new methodologies and collaborate closely with the FDA to successfully pass validation and regulatory review will significantly accelerate development and reduce costs for certain drugs, particularly monoclonal antibodies (mAbs). Conversely, companies that continue to rely on traditional pathways or encounter obstacles in the validation of New Approach Methodologies (NAMs) may perceive limited changes in the near term. This disparity has the potential to reshape the competitive landscape of the industry over a certain period.
Since the beginning of this year, the FDA’s stance on the use of AI has become increasingly clear.
At the beginning of the year, the FDA released a draft industry guidance titled “Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products: Guidance for Industry and Other Interested Parties (Draft Guidance),” which outlines the agency’s regulatory reasoning regarding the application of AI technologies in new drug development.
Previously, AI was extensively applied in preclinical areas such as drug discovery, virtual screening, novel target identification, and ADMET prediction, primarily due to the abundance and public availability of relevant data, which facilitated easy access, high AI acceptance, and the maturity of underlying technologies. During the preclinical stage, regulatory barriers remain relatively low; as long as a suitable drug candidate is identified and demonstrated to be effective and safe, there is no need to address the issue of “how the conclusions were derived.”
Once entering the clinical phase, on one hand, ethical requirements for human trials mandate that models must be able to explain “why these specific patients were selected,” ensuring that the selection logic is transparent and reasonable, particularly when it involves patient treatment decisions. On the other hand, due to privacy protection and corporate confidentiality concerns, clinical trial data are not readily accessible, yet the quality and comprehensiveness of such data directly impact the validity of the conclusions. These factors impose exceptionally high demands on companies seeking to enter this field.
The guideline highlights the key focus areas of regulatory authorities and the principles for the clinical use of AI. It is foreseeable that, with technological advancements and the gradual refinement of the regulatory environment, the application of AI in clinical research is expected to increase steadily.
Moreover, a recent development at the FDA further demonstrates its positive stance toward AI. On May 8, FDA Commissioner Martin Makary announced that by the end of June, a generative artificial intelligence system would be deployed agency-wide to assist in regulatory reviews. This AI system will be “unified, secure, and embedded within the FDA’s data platform.” Review experts stated that with AI assistance, “tasks that previously took three days can now be completed in minutes.”
This AI tool can automatically extract core information from hundreds of pages of technical documents, perform standardized terminology comparisons, identify potential consistency deficiencies, and generate preliminary draft reports for reviewers’ reference. Furthermore, the system can integrate with the FDA’s existing unified document management platform, facilitating access to structured historical data to assist in assessing the reasonableness of current submission data.
Against the backdrop of the FDA’s vigorous promotion of new approaches, the application of domestic AI technology in clinical settings started relatively late, yet pioneers have already emerged in this field. As a leading figure among the first in China to venture into AI-driven clinical applications, Dr. Du Tao pointed out, “The trend of applying artificial intelligence in the pharmaceutical sector is becoming increasingly prominent. Therefore, the intelligent transformation of the traditional pharmaceutical industry is not an optional choice, but an urgent and inevitable requirement.”
It is now evident that the influence of AI has gradually permeated the entire pharmaceutical industry workflow, from drug discovery to market access. As the most time-consuming, costly, and highly regulated phase, clinical development is a critical stage determining success or failure. Dr. Du Tao believes that the application of AI technology is transforming this phase along two dimensions: first, by enhancing efficiency through intelligent design; and second, by necessitating adherence to higher standards due to regulatory requirements. Innovative practices at Eaglet Pharma demonstrate that integrating international regulatory expertise with AI technology can provide pharmaceutical companies with intelligent solutions that comply with regulatory standards. Notably, Eaglet Pharma has pioneered the development of an intelligent system for drafting clinical protocols that meets international regulatory standards. This system, set to be launched soon, is poised to become a crucial technological support helping domestic pharmaceutical companies align with international regulations and overcome bottlenecks in regulatory submissions.