The past 12 months have shown an incredible focus on New Approach Methodologies (NAMs) to reduce our reliance on animals and incorporate human-relevant biology into drug development and preclinical studies. 2025 saw the US FDA’s roadmap to reduce animal testing and the UK’s strategy to phase out animals. Global regulators have clearly communicated that NAMs data should be integrated into regulatory submissions as soon as possible.
However, one of the major objectives for NAMs is to replace animals in preclinical testing. While many of the most advanced NAMs precisely recreate one or a handful of human tissues, none adequately recreates an entire living biological system. After research animals are euthanised, their tissues are processed. Toxicologists regularly review data from thirty to over seventy organ systems and tissues. All of these tissues acted systemically in the presence of the drug. Few (if any) NAMs can begin to approach that complexity.
Very fairly, proponents of continued animal testing take issue with that. They posit NAMs’ failure to identify systemic toxicity risks means animals will never be fully replaced in preclinical testing. Perhaps there are simply too many individual tissues and too many interconnected systems to be accurately modelled.
But one major question remains.
How many individual tissues (systemic or otherwise) are relevant for preclinical and clinical decision-making?
The Numbers
Lessons learned from the fate of AstraZeneca’s drug pipeline: a five-dimensional framework by Cook, et al. in Nature Reviews Drug Discovery 2014 explores this question, using data from AstraZeneca’s pipeline. Their results might be surprising. Their Figure 2a evaluates causes of preclinical and clinical safety failures. These results are summarised in the below Figure 1 and 2. Notably, sources of failure differed significantly between preclinical and clinical. However, the top 8 systems (cardiovascular, liver, musculoskeletal, genetic, respiratory, renal, gastrointestinal, and CNS) accounted for 79% and 97% of preclinical and clinical safety failures, respectively. The remaining sources are labelled “Other (for example, immunity)”. Further details on the “other” category are not provided. However, this likely represents diverse sources, with no one cause exceeding 3% of total failures.
The NAMs Exist
For virtually all organ systems mentioned above, validated NAMs are currently commercially available. While their ability to detect every failure mode identified in the Cook, et al. paper has not been established, their validation and benchmarking protocols often provide evidence of sensitivity to a wide range of human-relevant toxicity mechanisms. When paired together, these NAMs may provide a near comprehensive picture of safety.
For example, consider the following technologies.
Emulate, Inc. sells liver, brain, gut, and kidney in vitro chips (and have guided lung and bone marrow chips). Per the Cook, et al. paper, those four systems account for 64% of clinical failures. Notably, Emulate has already demonstrated their ability to identify and evaluate the toxicity of major drug metabolites. Emulate’s liver chip has also had its qualification plan accepted in the FDA ISTAND program to qualify NAMs. Axion BioSystems was just accepted into the FDA ISTAND program for their Human iPSC-Cardiomyocyte MEA Assay for Prediction of Clinical Cardiovascular Repolarization Risk. Alongside other direct cardiotoxicity screening tools, such as Genome Biologics’ vascularised and innervated cardiac organoids, these 5 organ systems account for up to 88% of the sources of clinical safety failures.
Further, multi-organ and specialised NAMs are emerging. Hesperos’ human-on-a-chip pairs four human tissues (such as heart, liver, neuron, skeletal muscle) to detect interconnected drug interactions between critical organ systems. Toxys’ ToxTracker and ReproTracker are extensively validated systems for evaluating genotoxic and reproductive toxicity. Within the larger “Other” group for preclinical safety, immunity appears to play a large role. Lonza’s Cell-based assays for immunogenicity and immunotoxicity offer human-relevant in vitro immune evaluations for certain therapeutics.
When paired with organ agnostic evaluations, NAMs might provide excellent coverage. For example, Integral Molecular’s Membrane Proteome Array for biologics specificity profiling detects potential off-target binding for more than 94% of the human membrane proteome, derived from 37 regulatory-relevant tissues. It is also the first technology accepted into the ISTAND program and is the nearest to receiving full qualification status.
The half-dozen or so companies listed above represent the tip of the iceberg for available and validated NAMs. Dozens more, ranging from specialised skin and eye irritation models to AI-based digital twins, provide increasingly complex and complementary methods to probe human-relevant biology.
It is conceivable, if appropriately designed, that a NAMs-focused preclinical safety package could be submitted that has the power to evaluate over 85% of preclinical and 95% of clinical organ systems responsible for drug failure. When paired with emerging PK and dose-focused NAMs, this might, one day, enable truly animal-free submissions.
How Good Do NAMs Need to Be?
So, if enough NAMs exist to address a major proportion of safety questions, why aren’t they being prioritised (or even just more aggressively integrated) into existing preclinical packages? One of the major drivers is cautious regulatory uptake. Programs like ISTAND will gradually qualify and integrate NAMs, eventually making them preclinical mainstays. However, this is likely to be completed in a piecemeal fashion. So, the bigger question may ultimately be, “How many NAMs will need to be qualified (and how good will they have to be) for regulators to more frequently allow NAMs-only IND and CTA submissions?”
Interestingly, the answer may have less to do with NAMs and more to do with the limitations of animals.
For example, to quote a paper by Van Norman (JACC Basic Transl Sci, 2019), “the presence of toxicity in a species sometimes added evidentiary weight to the risk of toxicity in another, but the reverse was not true: negative toxicity tests in animals did not significantly increase the probability that a toxic test would also be negative in humans, and a lack of toxicity in any species would not reliably indicate a probable lack of toxicity in any other species, including comparisons of primate to human toxicity tests.” This clearly indicates the need for models that can reliably identify when drugs are safe. Rather than saying “NAMs must perfectly recreate animal outputs,” we instead need models with better predictive outcomes. NAMs won’t need to be perfect replacements. They just need to move the needle.
However, to fully complete this transition, more data is needed. Dozens more individual studies will need to be completed with NAMs before there are enough data to conclusively compare their predictive ability to animals. However, many of these studies are already complete and, with improving regulatory frameworks for NAMs validation and qualification, most of the remaining studies are likely to be completed within the next 5 years.
As NAMs qualification accelerates, regulators will be increasingly equipped to evaluate how much (if at all) NAMs improve human clinical trial outcomes. Soon, a NAMs-only future might become the reality. With the right scientific, regulatory, and economic incentives, that future could be closer than we realise.
One response
[…] uncertainty remains real. NAMs cover a substantial proportion of safety risks, but important gaps remain including in metabolism, immune complexity, and systemic […]