A development team can finish an animal efficacy study, receive a statistically significant result and still be unable to answer a basic question: did the compound work because it engaged the intended biology at a relevant exposure? The study may have been well executed. The decision it was expected to support may not have been well defined.
That distinction is becoming more consequential. Regulators are taking a broader view of acceptable nonclinical tools, including human-derived systems and computational methods. But a larger menu of models does not automatically create stronger evidence. The opportunity in preclinical drug development is to assemble studies around the uncertainty a program needs to resolve—not around a familiar checklist of experiments.
Regulatory flexibility raises the bar for explaining why a study belongs.
In April 2025, the US Food and Drug Administration published its roadmap for reducing animal testing in preclinical safety studies. Rather than prescribing a single replacement, it described a stepwise transition toward scientifically validated methods, including advanced in vitro systems and computational approaches. It set a direction for regulatory science, not an across-the-board exemption from animal studies.
The discussion moved from ambition toward implementation in 2026. FDA's March draft guidance sets out general principles for validating new approach methodologies, or NAMs, when they are used to support drug development submissions. Its scope is primarily regulatory nonclinical evaluation; it does not purport to dictate the choice of discovery assays. In September, FDA announced a direct final rule on nonclinical testing terminology, recognizing that nonclinical evidence can come from more than animal tests. The agency explicitly said the change does not alter evidentiary standards. The rulemaking process was still subject to its stated comment and withdrawal procedures as of this article's research cut-off.
There are important distinctions between therapeutic areas and products. FDA's May 2026 draft guidance proposes more streamlined safety assessments for certain oncology biologics and conjugated products. It should not be read as permission to apply the same approach to an unrelated small molecule. Product class, pharmacology and the intended regulatory question still determine what evidence is needed.
Another development is easy to overlook. The ICH M15 guideline reached Step 4 in February 2026, setting harmonized principles for planning, evaluating and documenting model-informed drug development evidence. M15 concerns computational modeling and simulation. It is not a blanket validation of NAMs or an instruction to dispense with animal studies. Taken together, these developments point toward a more explicit account of what each source of evidence can, and cannot, support.
Choosing a method is becoming inseparable from defining the decision it is meant to inform.
The change is not simply a reduction in study count. For sponsors and CRO teams, it shifts attention toward scientific justification, the credibility of a method for its intended use and the consistency of the resulting package. A shorter program may be a better one—but only if the omitted work does not leave a consequential question unanswered.
A more human-relevant model is useful only for the question it can answer.
Consider a team studying an inflammatory pathway. A biochemical assay might establish potent inhibition of the isolated target. A human-cell system may show whether that mechanism produces the expected response in a particular cellular setting. An in vivo disease model can reveal a phenotype within a more integrated biological system. None of those results is a substitute for the others, and none is a direct prediction of patient benefit.
Model relevance has several dimensions. The target must be present and functional; the disease process being modeled must correspond to the claim; and the selected endpoint must reflect the biological change that matters. A complex model with poorly understood controls may tell a team less than a simpler, well-characterized assay. Conversely, an elegant human-derived system may miss distribution, endocrine feedback or immune interactions that affect a candidate in an intact organism.
One practical principle is the context of use: specify the method's intended purpose, biological setting, inputs, outputs and limitations before interpreting its results. The European Medicines Agency's qualification framework explicitly evaluates methods for a specific use. Qualification for one question does not imply that a method is valid for every compound, endpoint or disease.
This is also why an animal model should not be dismissed solely because it is an animal model. Its value depends on how closely its relevant biology and exposure-response characteristics connect to the intended human question. A model can be reproducible yet poorly predictive for a particular intervention. Scannell and colleagues have argued that predictive validity of decision tools deserves far more attention in drug R&D. That argument applies to new technologies as much as to established ones.
For an early discovery program, the right question may be whether a mechanism is engaged at all. Later, it may be whether the same mechanism can be reached at a feasible exposure and whether the apparent benefit survives more disease-relevant testing. The model changes because the question has changed—not because the protocol has acquired another line item.
A positive phenotype is less informative when the route to it is unknown.
A compound can be powerful in a plate and unremarkable in a disease model. Perhaps it never reaches sufficient free concentration at the site of action. Perhaps the assay measured a metabolite or a nonspecific response. Or the target is adequately engaged but does not control the disease feature as expected. Those explanations lead to different next experiments and very different medicinal chemistry decisions.
This is where chemistry and preclinical biology meet. Material identity, purity and batch consistency affect what is actually dosed. Formulation changes the concentration available for absorption. DMPK determines what can be said about exposure, while bioanalysis determines whether those concentrations are trustworthy. The ICH M10 guidance describes method-validation expectations for assays used in nonclinical and clinical regulatory studies. An accurate concentration is still not proof of target engagement; where possible, the program also needs a pharmacodynamic measure linked to the mechanism.
There is published evidence that these links are sometimes missing in development records. A 2023 analysis of investigator's brochures for CNS-active drugs found gaps in pharmacokinetic reporting for animal disease or behavioral models within its selected sample. That does not quantify the problem across all therapeutic areas. It illustrates why later interpretation becomes difficult when exposure and study-design context have not travelled with an efficacy claim.
In practice, each result places conditions on the others. A negative efficacy readout with inadequate exposure points toward formulation or molecular-property work before the biology can be judged. A negative result with sufficient target engagement raises a different question about the target hypothesis or chosen endpoint. A strong phenotype without measured exposure may justify follow-up, but it cannot resolve which mechanism produced the effect.
A good evidence package makes those branches visible. It does not force every project into the same sequence of studies. It shows which explanation the available data support, which alternatives remain plausible and what the next experiment must distinguish.
Work backward from the choice the team actually has to make.
Imagine an early-stage small-molecule inhibitor intended to modulate a defined inflammatory pathway. The sponsor has a characterized compound, a reproducible cellular assay and a candidate disease model. Before committing to a broad in vivo campaign, the team needs to decide whether the compound is suitable for a mechanistic proof-of-concept study—or whether the chemistry should return to optimization. This is a hypothetical planning example, not a report of a DempoChem client project.
The first task is to agree on what would change that decision. Evidence could include a selective cellular response, acceptable material and formulation behavior, measurable exposure at the relevant time, a pharmacodynamic marker tied to the pathway and an endpoint that the disease model can reasonably inform. Controls, randomization, blinding, the experimental unit and a defensible sample-size rationale should be planned rather than retrofitted to the report. The ARRIVE study-plan resources are useful prompts for robust animal-study design and transparent reporting, not a guarantee of predictive validity.
At the decision meeting, suppose the pharmacodynamic marker moves but the disease endpoint does not. If exposure was adequate and the model is relevant, the program may need to revisit the biological hypothesis or endpoint. If exposure was inadequate, changing the molecule or formulation may be more informative than immediately repeating the efficacy study with more animals. If a phenotypic benefit appears but the compound's exposure remains unknown, the team has learned something, yet not enough to interpret the dose-response relationship with confidence.
Exposure and mechanism evidence change what an efficacy result can tell us
Three selected scenarios from the hypothetical small-molecule program. Read each path from the observed result, through its evidence context, to a question for the next experiment. PD means pharmacodynamics; PK means pharmacokinetics.
- Observed resultNo phenotypic benefitEvidence contextExposure is inadequate
The biological hypothesis remains unresolved.
Next questionCan relevant exposure be reached?Consider molecular-property or formulation work, with exposure measurements before repeating efficacy testing.
- Observed resultNo phenotypic benefitEvidence contextExposure is adequate; a mechanism-linked PD marker changes
If the model is relevant, the hypothesis or endpoint may need revisiting.
Next questionDoes the mechanism affect this endpoint?Consider orthogonal testing of the biological hypothesis, model relevance and endpoint sensitivity.
- Observed resultApparent phenotypic benefitEvidence contextExposure is unknown
The phenotype does not establish which mechanism produced it.
Next questionWhat connects exposure to the effect?Consider matched PK and mechanism-linked PD measurements, with controls that distinguish alternative explanations.
Source: DempoChem editorial synthesis of the hypothetical planning example, informed by ICH M10, Scannell et al. (2022) and EMA context-of-use principles. Conceptual only; not experimental data, a validated decision algorithm or a regulatory requirement.
Figure 02
Different methods answer different questions; none closes the evidence chain alone
The comparison is qualitative. Model suitability, performance and regulatory acceptability are context-specific; entries describe typical questions rather than validated claims for a particular test.
| Evidence source | What it can help establish | What it cannot establish alone |
|---|---|---|
| Biochemical and cellular assays | Target activity, cellular response and potential mechanism under defined conditions. | Whole-organism exposure, organ-level toxicity or patient benefit. |
| Human-derived 3D or microphysiological systems | Selected human tissue mechanisms or responses within a qualified context of use. | The full multi-organ distribution and clinical performance of a candidate. |
| In vivo disease models | Phenotypic response under specified species, model, dose and endpoint conditions. | Human therapeutic benefit without evidence of biological and exposure relevance. |
| DMPK and bioanalysis | Concentration–time profiles, metabolite information and measured exposure. | That a measured concentration engages the target or produces a clinically relevant effect. |
| Nonclinical safety assessments | Characterized hazards and selected exposure-related safety signals. | An absence of all risk or the suitability of a replacement method outside its validated scope. |
Source: DempoChem editorial synthesis based on FDA NAMs draft guidance (2026), ICH M10 (2022), ICH M15 (2026) and EMA qualification guidance (2026)
The table is deliberately not a ranking. It clarifies the scope of each evidence source without pretending that heterogeneous models can be scored on a common quality axis. For regulatory submissions, the relevant development-stage requirements and appropriate scientific advice still govern the nonclinical package.
Three questions before authorizing the next study
Thinking this way can change a study budget without promising cheaper or faster development. It may remove a redundant experiment. It may also justify adding an assay that makes an expensive in vivo result interpretable. The relevant measure is whether the work reduces a consequential uncertainty at a defensible cost.
A connected program matters more than a longer service menu.
For sponsors outsourcing early development, the practical challenge often lies between work packages. A chemistry provider delivers material against one specification. A pharmacology laboratory selects a dose and endpoint. An analytical team measures a concentration. A toxicology study is planned against a different set of assumptions. Each assignment can be completed correctly while the project still lacks a coherent answer.
Integration does not mean that one CRO must perform everything. It means agreeing in advance on the test article, batch identity, dosing formulation, sample handling, analytical method, key decision points and interpretation plan. The source of a study result matters less than whether its evidence can be connected to the rest of the program and critically challenged.
DempoChem's approach links compound sourcing and custom synthesis with partner-coordinated preclinical services, including pharmacology, DMPK and bioanalysis and relevant safety workstreams. That is a coordination model, not a claim that every study is conducted in-house or that a particular method is regulatory-qualified. Its potential value is in making the material, experimental and analytical interfaces explicit at project definition.
The broader industry direction is not necessarily toward fewer experiments in every program. It is toward evidence selected for its intended use, documented well enough to be challenged, and assembled so that the next scientific choice is better informed. New approach methodologies enlarge the toolkit. Better decisions will come from knowing where each tool earns its place.
Sources & methodology
- Approach
- Qualitative editorial analysis of primary regulatory publications, ICH guidance and selected peer-reviewed research available by 13 October 2026. The study-design example is hypothetical; its decision map and the method comparison table are explanatory syntheses, not validated scores or experimental results. The map connects selected efficacy readouts and exposure/PD evidence conditions to possible follow-up questions; it is not an exhaustive decision tree.
- Definitions
- New approach methodologies (NAMs) encompass a range of non-animal and human-relevant test systems in the regulatory context discussed. Context of use means the stated purpose, setting and limitations of a method. Decision-ready evidence here means an internally coherent data package sufficient to inform a defined program choice; it is not an FDA-recognized certification or regulatory status.
- Limitations
- Regulatory documents differ in legal force. The March and May 2026 FDA guidances are drafts, and the September 2026 terminology action does not remove evidentiary requirements. Requirements vary by modality, indication, jurisdiction, development stage and agreed regulatory strategy. An illustrative framework cannot establish that a specific NAM replaces a required study.
- Roadmap to Reducing Animal Testing in Preclinical Safety Studies
U.S. FDA · April 2025
- General Considerations for the Use of New Approach Methodologies in Drug Development (draft guidance)
U.S. FDA · March 2026
- FDA Updates Regulations to Advance Innovative Alternatives to Animal Testing
U.S. FDA · September 2026
- Oncology Pharmaceuticals: Streamlined Nonclinical Safety Studies for Biologics and Conjugated Products (draft guidance)
U.S. FDA · May 2026
- ICH M15: General Principles for Model-Informed Drug Development (Step 4)
ICH · February 2026
- Qualification of novel methodologies for medicine development
European Medicines Agency · July 2026 revision
- M10 Bioanalytical Method Validation and Study Sample Analysis
U.S. FDA / ICH · November 2022
- Predictive validity in drug discovery: what it is, why it matters and how to improve it
Nature Reviews Drug Discovery · 2022
- Translatability of preclinical to early clinical tolerable and pharmacologically active dose ranges for central nervous system active drugs
Translational Psychiatry · 2023
- Introducing the ARRIVE study plan
NC3Rs · 2024
