The role of in vitro screening models in drug discovery is hard to overstate. By simulating the interactions between drugs and biomolecules, these models return early information on the physicochemical properties, metabolic pathways, drug interactions, and transmembrane permeability of candidate molecules. Those predictions help explain a mechanism of action, and they also guide the design of the in vivo testing that follows.
What in vitro work cannot do is reproduce a living organism. A dish has no circulation, no hepatic clearance, no immune traffic, and no intercellular signalling at the scale of a tissue. The gap between a controlled culture system and a whole animal shows up in absorption, in first-pass metabolism, in exposure at the target organ, and in the compensatory responses that a body mounts against a pharmacologically active molecule.
Because of that gap, the results of in vitro activity assays may not fully reflect the true efficacy of a drug in vivo. In vitro screening therefore keeps an indispensable place in early development without ever replacing the animal study. Regulatory agencies worldwide require the submission of complete in vivo data before a new drug can be approved for market release, and that requirement covers both safety and efficacy.
Registration pathways treat those data as the bridge between a laboratory observation and a first-in-human dose. The package normally has to show exposure in a species that metabolises the molecule, a dose at which the pharmacological effect appears, and a dose at which toxicity appears. When any of the three is missing, a programme either repeats work or enters the clinic holding an assumption where a measurement should be.
Testing directly in human subjects is not an option. Ethically, giving an insufficiently validated drug to people carries health risks and erodes public trust in pharmaceutical development. Economically, human trials demand substantial resources and carry considerable uncertainty, which puts financial strain on a programme from the outset.
So the work moves to animal models that can evaluate how a candidate acts on the human body and on disease progression. An animal model reproduces the pathogenesis and pathophysiological sequence of a human disease closely enough to serve as a test environment. Within it, safety, efficacy, and pharmacokinetic properties can be assessed together, and the resulting package supports the later clinical trial.
This article looks at how animal models are applied in drug development and where their advantages and limits sit. Because of length, the topic runs in two parts. Part 1 covers establishment and classification. Part 2 turns to classic animal models.
2. Establishment of Animal Models
Models are built by three routes: chemical, physical, and genetic. The choice follows the disease mechanism rather than convenience, and each route carries its own assumptions about what the model actually represents.
2.1 Chemical Intervention
Chemical intervention means administering a specific substance, by injection or by feeding, to induce a defined pathophysiological change. Two decisions dominate: which substance, and at what dose. Get either wrong and the pathological state becomes unstable or unreproducible, which then contaminates every downstream comparison.
The MPTP model of Parkinson's disease is a standard example. MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine) is a neurotoxin that selectively damages dopaminergic neurons in the substantia nigra, which lowers dopamine levels and reproduces the pathological features of Parkinson's disease.
The procedure starts with strain and age selection, and C57BL/6 mice are commonly used. The MPTP dose is then calculated from the experimental objective and body weight. Administration is typically intraperitoneal, once daily, for several days to weeks.
Two constraints shape that step. Dose is usually taken from a published range and then confirmed in a small pilot, because strain, vendor, and age all shift sensitivity. Timing matters as much as dose: too short and the lesion stays incomplete, too long and compensatory mechanisms begin to blur the readout.
During that period the animals need close observation for behavioural change, particularly motor coordination and postural balance, since these are the first signs that the lesion has taken hold. Biochemical assays and histological analysis then confirm the pathophysiological picture rather than relying on behaviour alone.
The MPTP model has clear advantages. It reproduces the hallmarks of Parkinson's disease, including the loss of dopaminergic neurons in the substantia nigra and the fall in dopamine. Establishment is straightforward, operationally convenient, and inexpensive. Because MPTP acts selectively, the model is also stable and reproducible across batches, which is what makes it usable for ranking compounds.
Its limits matter just as much. The model does not fully recapitulate the complex pathogenesis and pathological progression of human Parkinson's disease, and the administration itself can produce side effects and toxic reactions. Anyone using it for mechanism work or for efficacy claims needs to state which part of the human disease the model is standing in for.
2.2 Physical Intervention
Some diseases come from an external physical cause: a fracture, a sprain, an exercise-induced injury. Experimenting on people to study them is infeasible on ethical grounds and risky on practical ones, which is where physical intervention models earn their place.
The methods vary, and surgical or mechanical force application is the most common. In bone fracture healing, for instance, researchers create a fracture in the leg of a mouse or a rat with specialised instruments, reproduce the trauma of a human fracture, and then follow the healing process. It is a scaled-down version of an accident, staged so that recovery can be watched and measured.
Heart valve disease offers a second example. Catheterisation techniques can simulate valve stenosis or regurgitation in an animal heart. Under precise surgical control, a catheter is advanced into the heart to mimic a valvular lesion, and cardiac function is then tracked. The logic resembles introducing a known fault into a machine and recording how the system responds.
Endpoint choice follows from what the model can hold constant. A fracture model can be read mechanically or radiographically, and a valve model through pressure gradients or ventricular dimensions. Those readouts answer different questions, so mixing them within one study usually costs more than it returns.
These methods do more than mimic disease states; they make efficacy and safety measurable. In a fracture model, dosing a new compound and recording the rate and quality of bone healing gives a direct readout. In a valve disease model, the same compound can be judged on whether cardiac function improves and complications fall.
Physical intervention is often combined with other techniques, including gene editing and drug induction, to reproduce a more complete disease picture. Cancer research is a familiar case: a specific mutation is first introduced by gene editing, then a physical or chemical stimulus such as radiation or a carcinogen triggers tumour formation. Neither step alone would give the same model.
2.3 Genetic Intervention
Genetic intervention modifies an animal's genome to mimic a human disease. The core tool is gene editing, CRISPR/Cas9 being the common choice, which allows a precise change at a defined locus. The technology works like molecular scissors: it cuts and replaces DNA sequence at a chosen site, which produces an animal carrying a specific mutation.
Take a hereditary disease caused by a known mutation, such as autism spectrum disorder. CRISPR/Cas9 can introduce the identical mutation into the genome of an animal, mice or dogs for instance, and the resulting animals then show disease characteristics similar to those of human patients. That gives research a platform where the cause is defined rather than inferred.
The canine model of autism spectrum disorder is a well-known case of genetic intervention succeeding. The team led by Professor Yongqing Zhang in China introduced the Shank3 gene mutation into dogs using CRISPR/Cas9, creating a canine model of autism spectrum disorder. The mutant dogs reproduced core clinical manifestations of autism in humans, including social deficits, which gave scientists a tool for exploring pathogenesis and treatment strategy in a species with richer social behaviour than a rodent.
Species choice inside this route follows the phenotype. A mouse can carry a mutation and still lack the circuit or the behavioural repertoire through which the disease shows itself, which is where species with more complex social behaviour become relevant.
Genetic intervention also combines with transgenesis and chemical induction to approach the complexity of human disease. A mutation can be introduced by editing, then a chemical agent used to drive a specific physiological or pathological change, so that the model carries both the predisposing genotype and the acquired lesion.
3. Classification of Models
Once a model exists, the useful question is what kind of claim it can support. The three categories below differ in how much of the human disease mechanism they actually share, and that difference sets the strength of the conclusion. The same animal can belong to more than one category depending on the endpoint in use, so classification is a statement about the claim rather than about the species.
3.1 Homologous Animal Models
Homologous models show a high degree of similarity to human disease mechanisms. Gene expression and physiological characteristics are close, and more importantly so are aetiology, progression, and drug response. Because the mechanism is shared, the model can stand in for the human pathophysiological process with reasonable confidence. Two characteristics define the category:
- High Fidelity Simulation: homologous models replicate the causes, symptoms, and treatment responses of human disease, which keeps the development programme relevant and efficient.
- Predictive Power: because mechanism and drug response are close to human, these models can predict efficacy and side effects with reasonable accuracy, which is what gives them weight in supporting a clinical trial.
Bacterial infection models, built by reproducing the human course of bacterial infection, belong here. They are central to understanding pathogenesis and to judging antibacterial drugs. In antibiotic development, a bacterial infection model is where inhibitory and bactericidal effects of different compounds get compared, and where the basis for clinical dosing is established.
The category earns its name only when the mechanism is shared, not when the symptoms merely resemble each other. A homologous claim therefore needs support on more than one axis: the same molecular lesion, the same cell populations responding, and a comparable exposure-response.
3.2 Isomorphic Animal Models
Isomorphic models share symptoms with a human disease and respond to the same treatments, while the underlying cause may differ. Arthritis-related degenerative damage illustrates the point. Injecting iodoacetate into an animal joint induces arthritis, and the model can then be used to study the anti-degenerative effect of a compound. Human osteoarthritis, by contrast, forms through a far more complex route and is not caused by iodoacetate. The category has three features:
- Symptom Similarity: symptoms closely resemble those of the human disease, so analogous pathophysiological changes can be observed and read as a reference for development.
- Treatment Parallels: because the symptoms match, the same treatment strategies often apply to the model and to patients, which makes efficacy and safety assessment practical.
- Etiological Differences: the fundamental cause may differ from that in humans. Results therefore need cautious analysis, and findings should not be extrapolated directly from the model to patients.
Returning to the arthritis example, intra-articular iodoacetate produces joint swelling, pain, and limited mobility, all of which parallel human osteoarthritis and allow a compound's anti-degenerative potential to be assessed. Human osteoarthritis develops through genetics, environment, and age acting together. A programme using the isomorphic model has to weigh those factors explicitly, or the accuracy of its conclusion suffers.
Isomorphic models are most useful when the question is symptomatic: does this compound reduce swelling, restore mobility, slow structural damage. They are least useful when the question is causal, since a treatment aimed at the human aetiology may have nothing to act on in the model.
3.3 Predictive Animal Models
Predictive models are used when no animal model corresponds directly to a human disease. They simulate the symptoms and responses by other means, which is rather like planning a route from a hand-drawn sketch when no real map exists.
Mental illness is the clearest case. These conditions are complex in aetiology, and an animal's mental state cannot be observed or interpreted directly, so deciding whether an animal truly has a mental illness is difficult. The workaround is pharmacological: certain drugs induce or worsen mental symptoms in animals, and the resulting state becomes the test condition.
Imagine developing a drug for depression. With no suitable model available, researchers might inject mice with a substance known to induce depressive symptoms and then observe behavioural change. If the new compound relieves those symptoms, that is preliminary evidence it might also work in human depression. The word preliminary carries weight here: the prediction still has to be confirmed in clinical trials.
Validation is how these models are judged. A predictive model is only as good as the compounds that have already passed through it. If known actives fail and known inactives pass, the prediction carries little weight, however plausible the behavioural readout looks.
4. Species Selection and Sample Size Requirements
Two decisions are made before any dosing starts, and both are easy to get wrong. Species sets how far the result can travel toward humans; sample size sets how much of the result is signal.
4.1 Species Selection
Species selection resembles choosing materials for a complex project: the model has to fit the question, and the fit is rarely generic. Not all animals are suitable for drug development, and the factors that decide suitability include physiological structure, the metabolic system, and susceptibility to the disease of interest.
Priority generally goes to species that are physiologically and pathologically closest to humans. Mice and rats are used constantly, for good reasons: short breeding cycles, ease of genetic manipulation, and low cost. For particular diseases, more specialised models are needed. Rabbits and monkeys are preferred for ophthalmological work because their ocular structures resemble the human eye.
Non-human primates (NHPs) sit at the far end of that gradient. As social animals with clear hierarchies and complex behaviour, they are phylogenetically, anatomically, physiologically, and biomedically closer to humans than the rodent models in standard use. They serve as advanced experimental animals in medical and life science research, and they hold an irreplaceable place in vaccine development for human disease prevention and in the study of brain function and neurological disorders. As medical research deepens, lower-order models prove insufficient in some areas, which creates demand for species that are evolutionarily closer to humans.
Practice bears this out. In cancer treatment research, mouse models have supported the screening of numerous anti-tumour drugs. In neurodegenerative diseases, fruit fly and nematode models have clarified the molecular mechanisms behind disease onset. Different species answer different questions, and matching the species to the question is what separates a useful study from an expensive one.
Availability and cost sit inside the same decision. A species that is physiologically ideal but limited in supply constrains group sizes and timelines, and a model that cannot be repeated across sites is hard to carry through a registration package. Selection weighs scientific fit against the practical terms of running the study.
4.2 Sample Size Requirements
The number of animals used rests on five factors: statistical principles, experimental objectives, animal characteristics, prior experience, and relevant regulations. None of the five acts alone. A stringent statistical plan raises the number, a screening objective lowers it, and regulations can override both. Fixing the number first and justifying it afterwards is the usual way these studies go wrong.
Statistical Principles
A result has to be effective, reliable, and reproducible, and statistics is how those three are reconciled. Sample size is estimated from the anticipated effect size, the experimental error, and the desired level of confidence. The intuition is the coin flip: more tosses give a closer estimate of the probability of heads.
Power, the probability of detecting an effect that is genuinely present, links those three quantities. It is normally set before the study begins, and reporting a value computed after the fact tells the reader very little.
Experimental Objectives
The objective sets the number. Toxicity studies typically need larger samples because the conclusion has to hold for safety, while a preliminary screening phase can proceed on smaller groups. Treating the two as interchangeable is a common source of underpowered safety data, and the error usually only becomes visible after the study has been reported.
Animal Characteristics
Species, age, sex, and health status all shift the required number, because animals with different characteristics respond differently to the same treatment. Cooking times and heat levels differ by ingredient for the same reason. Adjusting sample size to the characteristics of the animals in hand is part of the design, not an afterthought.
Prior Experience
Earlier work informs the number. Where a disease or drug class already has established models and protocols, those provide a basis for deciding how many animals are needed, and historical control data often carry more information than a fresh power calculation, provided the earlier studies used comparable endpoints and a comparable dosing schedule.
Relevant Regulations
Countries and regions set specific rules and guidelines on the number and condition of animals used in experiments. Compliance with those rules is mandatory when the experimental animal numbers are set, and it applies at the design stage rather than at review. Reporting follows the same logic: the number of animals, the reason for it, and any deviation from the protocol all belong in the record that reviewers read.
