Where NAMs Beat Animal Models

by Michael Phelan
on October 26, 2026

In our last post, we asked whether NAMs can really replace animals. The answer was yes, but not easily. A single repeat-dose study answers dozens of questions at once. Most NAMs answer one.

That post was about where animals still hold the advantage. This one is about where they don’t.

The usual case for NAMs is ethical. It is a strong case, but it is not the one I will make here. On the science alone, NAMs beat animals in five places:

  1. They are built on human biology.
  2. They are designed around a defined question.
  3. They show you mechanism, not just outcome.
  4. They can be engineered for reproducibility.
  5. The ones headed for regulatory use are being validated in a way most animal models never were.

Animals still lead in a few areas. But the gaps NAMs face are engineering problems that can be solved. The gaps animals face are biological, and cannot. That is why NAMs-first is viable now, and NAMs-only is the future.

Human biology, not a proxy for it

Every animal study contains a hidden step. The result is generated in one species and then extrapolated to another. NAMs built on human cells, tissues and targets remove that step.

For biologics, the problem is well known. Many antibodies have no pharmacologically relevant species beyond the non-human primate, and some have none. TGN1412 and the cytokine release syndrome seen with T cell-engaging therapies show what happens when the species gap goes unnoticed.

But this is not only a biologics problem. Small molecules face it too, through metabolism.

Drug-metabolising enzymes differ between species in which isoforms are present, how much of each is expressed and what they act on. A widely cited review found that CYP2E1 translates reasonably well across species. The species-specific forms of CYP1A, 2C, 2D and 3A do not. Those are the families that clear most marketed small-molecule drugs.

The dog, the default non-rodent species, is no exception. Dog and human CYPs differ in both catalytic properties and relative abundance. A drug can produce a metabolite in humans that the toxicology species barely forms, or never forms at all. That gap is the reason FDA’s guidance on safety testing of drug metabolites (MIST) exists.

Transporters matter as much as enzymes. In 1993, fialuridine killed five patients in a hepatitis B trial. The nonclinical package had found no liver toxicity in mice (6 months), rats (1 month), dogs (3 months) or monkeys (1 month). The leading explanation is a human nucleoside transporter, hENT1, that carries the drug into mitochondria in a way other species do not. A human cell model with the right biology later reproduced the mechanism: HepaRG cells lost mitochondrial DNA and respiratory function after extended dosing. HepG2 cells, which lack key parts of that biology, did not.

That last point matters, and we will come back to it. Human-relevant only counts if the model expresses the biology in question.

Human NAMs can also capture human diversity. Multiple donors, both sexes and known genotypes, including CYP2D6 poor and ultra-rapid metabolisers, can be built into a study by design. An inbred colony is deliberately uniform. Your patient population is not.

Context of Use is the defining strength

If NAMs have one advantage above all others, it is the Context of Use.

An animal study is an open-ended question. Dose the animal, then see what happens. Sometimes it answers the question that matters most for your molecule. Often it answers dozens that matter less, and leaves the critical one ambiguous.

A NAM works the other way around. You start from the question. Will this compound inhibit BSEP at clinical exposures? Does it activate T cells in human blood? Does it prolong repolarisation in human cardiomyocytes? Then you choose the tool built to answer it.

A defined Context of Use also means defined performance. You know what the NAM answers, how well it answers it, and where it stops being reliable. No one set those terms in advance for a repeat-dose study.

This is why the earlier criticism, that most NAMs answer only one question, misses the point. No single NAM should cover a full assessment. Complete coverage comes from stacking and combining Contexts of Use across several NAMs, each with a known job, and integrating them in a transparent Weight of Evidence.

The package becomes a set of targeted answers rather than one broad observation.

NAMs show you why

An animal study tells you that something happened. A well-designed NAM can tell you why.

The reason is access. In an animal, most endpoints are measured once, at necropsy, through blood chemistry and histopathology. In a NAM, the biological system sits in a plate or a chip. You can measure it directly, repeatedly and with almost any tool in the modern lab:

  • Bulk and single-cell sequencing to see which pathways respond and in which cells
  • Flow cytometry to track activation, proliferation and death across cell populations
  • Immunostaining to localise injury within a tissue structure
  • qPCR for targeted gene expression
  • Real-time and live-cell microscopy to watch injury unfold over hours or days
  • Secreted biomarkers and metabolomics sampled from the medium over time

Just as important, you can rerun the experiment while changing one variable at a time. Concentration. Exposure duration. Donor genotype. A CYP inhibitor or inducer. A transporter knocked down. A cell type added or removed. Each run tests a hypothesis about mechanism, rather than inferring one after the fact.

Mechanism is what lets a sponsor decide whether a finding is relevant to humans, design a liability out during lead optimisation, and plan clinical monitoring.

But the ability to rerun an experiment under controlled conditions is only valuable if the system is reproducible. That brings us to quality.

Not all NAMs are created equal

Reproducibility is where the comparison gets interesting, because both sides have problems. The difference is which problems can be fixed.

Animals carry variables you cannot remove

Animal studies are standardised to reduce variability. The evidence suggests standardisation does not remove it.

In a landmark 1999 study, three laboratories ran identical mouse experiments with tightly matched protocols, strains and equipment. They still found significant laboratory effects and genotype-by-laboratory interactions.

Standardisation may even make things worse. An analysis of 440 preclinical studies found single-laboratory studies generally failed to predict effect sizes accurately. Larger sample sizes made the estimates less accurate, not more. The authors identified within-study standardisation as a major cause of poor reproducibility.

The hidden variables are biological, and many are hard to control:

GLP toxicology studies are more tightly controlled than most of these examples. But microbiome, stress and circadian timing affect metabolism and toxicity whether or not a study is GLP. These variables cannot be eliminated. They are part of the animal.

NAMs can be engineered for reproducibility

NAMs are not reproducible by default. Many still rely on a single donor. Cell sources drift. Matrix lots vary. Methods that work in one lab can fail in another.

The difference is that these variables can be identified, controlled, measured and reported. Donors can be characterised and banked. Reference compounds can be run with every batch. Acceptance criteria can be set before the study starts. NIST and US federal partners have published a step-by-step framework for building measurement quality into NAM protocols using standard quality tools.

This is where NAMs separate. A NAM run under a fit-for-purpose quality system generates decision-grade data. The same biology without one might generate interesting scientific results but they are not fit for regulators.

Not all NAMs are created equal. Before asking whether a NAM is biologically relevant, ask whether it is reproducible.

NAMs are being validated. Most animal models never were.

Virtually all NAMs destined for regulatory use are pursuing validation or qualification. FDA’s ISTAND program, OECD test guidelines, EURL ECVAM and ICCVAM all require developers to define a Context of Use, show reproducibility and demonstrate predictive performance. FDA’s 2026 draft guidance on NAMs in drug development sets the same expectations.

I have been through that process. When we brought the first technology into ISTAND, every claim had to be defined, justified and supported with data. That is how it should be.

Most animal models in a standard nonclinical package never went through anything comparable. They were not validated against human outcomes before they became standard. They became standard through decades of use, and were then written into guidelines.

The closest thing animal models have to validation is retrospective concordance analysis. As we showed in a previous post, those analyses produce figures anywhere from near 0% to near 100%, depending on the metric, endpoint and dataset. The often-quoted 96% for dogs is a negative predictive value for pulmonary outcomes. For the same endpoint, the positive predictive value was zero.

Animal models should be validated to the same standard we ask of NAMs. They almost certainly will not be. Prospective validation would be enormously expensive. It would use large numbers of animals. And the incumbent method has little incentive to put its own credibility to the test.

That has a consequence. Every validated NAM adds to a growing body of evidence with known performance. Each qualification raises the bar. Animal models stay where they are, with performance that is assumed rather than measured.

If animal models never receive formal validation, it is inevitable they will be outpaced and outclassed by NAMs. Over time, a tool with known performance will beat a tool with assumed performance.

Faster, earlier and cheaper

The practical advantages follow from the scientific ones. NAMs return answers in days or weeks rather than months. That means decisions can be made before a candidate is locked in and before a GLP study is booked. Fewer late surprises means fewer expensive restarts. For startups and small biotechs, where one failed tox study can consume a funding round, that matters most of all.

NAMs-first is viable now

NAMs-first does not mean no animals. It means every question in the package starts with the best human-relevant tool available. Animals are used only for the questions that remain, where NAM evidence does not yet meet the risk tolerance of the first-in-human study.

The regulatory signals already support it. The rabbit pyrogen test has left the European Pharmacopoeia. FDA’s draft guidance on monoclonal antibodies opens the door to reducing primate studies. FDA’s 2026 draft guidance sets general expectations for NAMs in drug development. ISTAND acceptances are accumulating. The TGA now has a dedicated NAMs page, and China’s CDE has launched its Pioneer Program.

As we argued last time, the sensible path is risk-tiered. Start with the lowest-risk programs. Replace study by study as the evidence meets the bar.

Where animals still lead, for now

None of this means NAMs win everywhere. Animals still lead in a few places, and the gaps are real: systemic absorption, distribution and elimination; toxicity that emerges from interacting organs; chronic, carcinogenicity and reproductive endpoints.

But these are engineering and validation problems, and there is a visible path to solving them. Multi-organ systems, PBPK and QSP modelling, and longer-duration cultures are all advancing.

The animal’s core limitation is different. A dog will never express human CYPs. A monkey’s T cells will never respond exactly like a human’s. No amount of investment changes the species. The places where animals lead are shrinking. The places where NAMs lead are permanent.

Closing the gap

Getting from NAMs-first to NAMs-only will take deliberate work across the field:

  1. Education. Sponsors, reviewers, ethics committees and CROs need a working understanding of what NAMs can and cannot do.
  2. Packages designed around Context of Use and Weight of Evidence. Start from the questions the package must answer, not from the guideline checklist.
  3. Quality systems built for regulatory use. Fit-for-purpose quality for every NAM that informs a decision, including in silico tools.
  4. Validation against human data. Clinical reference sets, not animal results, should be the benchmark.
  5. Integrated systemic models. Multi-organ systems and modelling that link in vitro results to human exposure.
  6. Regulatory harmonisation. FDA, EMA, MHRA, TGA and CDE converging on shared expectations.
  7. Published case studies. NAMs-first packages that reached the clinic, shared openly.

NAMs already beat animals where it matters most: human biology, precise questions, mechanism, reproducibility and validation. NAMs-first is viable now. NAMs-only is the future. The question for every sponsor is how soon they want to get there.

Sources

Categories:

Uncategorized

Tags:

No Tag

No responses yet

Leave a Reply