Why animal models are still needed for discoveries in mental health research

Animal models have been essential for neuroscience discovery for over a century. Recently, animal models have been called into question for their translatability to human psychiatric conditions. Furthermore, the development of New Approach Methods (NAMs) including computer models (artificial intelligence, AI), human derived induced pluripotent stem cells (iPSCs), organoids, and microphysiological systems (e.g., brain-on-chips) have been proposed as replacements for animal models in research. Indeed, the National Institutes of Health (NIH) has adopted a new initiative to reduce animal use in research described in a news release in April 2025,¹ and the UK government is also committing to reduce animal use in science.² Thus, given the challenges with translatability to human conditions and the development of novel technologies, are animal models still needed to make discoveries in mental health research?

The short answer is yes. Animal models are still essential in psychiatry research because they allow for the understanding brain–behaviour relationships and mechanisms that cannot ethically or practically be studied directly in humans or recapitulated in reduced systems or by computer models.

Pros and cons of animal models

It is widely recognized that no animal model can completely replicate the complexity of neuropsychiatric and neurodevelopmental disorders. However, convergence across domain-relevant criteria and disease endophenotypes (e.g., prepulse inhibition of acoustic startle response measuring sensory gating deficits, fear conditioning models) has provided valuable insights, particularly when analyses span multiple levels; from genomic and epigenomic mechanisms to cellular, circuit, network, and behavioural outcomes. Much of what we know in neuroscience about homeostasis, reward systems, reinforcement, and motivation stems from animal research and has translated well in human studies. Experimental studies involving animals have been instrumental in elucidating the physiological and neurophysiological foundations of hunger, thirst, and sexual motivation.^3,4 Studies in rats and primates identified dopamine's role in reward, motivation, and reinforcement (e.g., Olds and Milner 1954; Schultz 1997 ^5,6), which was instrumental in understanding how we learn, how we are motivated, providing insights into addiction, schizophrenia, and mood disorders. Other fundamental contributions to the development of mental illnesses come from Canadian psychologist Donald O. Hebb's studies demonstrating that rats raised in an enriched environment show better adult learning, and that this learning requires neuronal assemblies firing together, leading to growth or metabolic processes that enhance the efficiency colloquially stated as “neurons that fire together wire together”. This led to further discoveries about how early life experiences influence stress and resiliency. For example, rat studies show that variations in maternal grooming ^7–9 alter gene expression and stress reactivity in offspring through epigenetic mechanisms. These gene expression changes can translate to humans as similar epigenetic mechanisms have been found both in cocaine self-administering mice and human cocaine users.^10,11 Additionally, studies of early life stress influences plasticity in development and later life responses to stress.^12,13 These findings highlight how early environment shapes vulnerability to mental illness. In the absence of these and other investigations in animals, our understanding of these fundamental processes would likely be considerably diminished.

The number of animal models of psychiatric disorders has grown substantially. However, there have been challenges in the ability to translate discoveries in animal models to effective treatments for people.¹⁴ Limited success with efficacious new treatments for neuropsychiatric disorders has largely been blamed on the poor translatability of animal models of psychiatric disorders and the challenge of translating symptom specific mechanisms into animal models. Although animal models do not capture the human disorder in its entirety, these model systems remain essential for dissecting the complexity of the brain, including its diverse cell types, distinct gene expression profiles, and intricate neural circuits that collectively underlie behavioural states. Rodents and humans share ∼85%–90% of protein-coding genes. Many genes involved in neurodevelopment, neurotransmission, and stress responses are highly conserved. Additionally, core molecular pathways responsible for synaptic plasticity, oxidative stress, neuroinflammation, and circadian regulation function very similarly in humans and rodents. This similarity allows for gene editing in rodent models (e.g., CRISPR, transgenic mice) to introduce human mutations (e.g., “humanized mouse models”), improving disease relevance or to introduce elements to help infer causal effects of neural circuits and behaviour that is not achievable in humans (e.g., optogenetics, chemogenetics). Of course, there are some human-specific genes and alternative splicing patterns (e.g., in neurexins ¹⁵ or serotonin receptor subtypes ^16,17) that do not exist in rodents, which may alter how psychiatric risk variants manifest. However, the usefulness and effectiveness of preclinical studies in informing future drug treatments are maximized when animal models of psychiatric disorders employ behavioural constructs that demonstrate strong cross-species transferability while also incorporating ethological relevance within a well-controlled experimental design.

Improving preclinical models

There have been improvements in animal models that have overcome typical translational pitfalls. These translational pitfalls broadly include the following: (1) The sensory world of rodents is different—where in humans, visual input is dominant, in rodents, smell, hearing, and tactile whisker sensations drive behaviours. (2) While there is a large degree of similarity between the complex anatomical organization and ordering of cortical and subcortical connections of human and rodent brains,^18–21 mouse and human brains should not be equated inappropriately. For example, the association cortex makes up a large portion of the human neocortex, which is relatively small in rodents. In rodents, the hippocampus serves as the primary associative cortex, integrating inputs mainly from secondary sensorimotor regions and producing species-specific motor patterns that are difficult to align with the proposed functions of the human hippocampus.²² Regardless, there are many conserved processes that occur between human and rodent hippocampus, reflecting their shared role in learning, memory, and navigation. For example, hippocampal long term potentiation,²³ adult neurogenesis,^24,25 spatial and contextual representation,^26–28 pattern separation and completion,^29,30 oscillatory dynamics and temporal coding,^31,32 and stress and emotional regulation ^33,34 occur similarly in rodents and humans, although they may contribute to behaviour in different degrees. (3) It is important to consider the ecology of rodents,³⁵ especially mice which are a small species which survive by freezing or flight reactions that do not need much cognitive processing. Thus, translational interpretation of studies of memory or emotionality should take this into account. (4) Confounding neuronal properties with system properties may lead to issues. For example, experimentally induced changes in the brain (e.g., knockout of a protein, silencing of a circuit) could lead to compensatory changes or plasticity elsewhere in the brain. Many techniques have improved on this using inducible or conditional knockout strategies, or temporary inhibition or excitation strategies (e.g., optogenetics, chemogenetics) to prevent long lasting compensatory changes. However, this should still be a factor in the consideration of experimental design. (5) Parallel experimental designs in rodents and humans that consider similar species-appropriate outcomes may also improve translatability of rodent models. Related to this, coordinated tests of reproducibility and generalizability of pre-clinical studies across different labs will improve translatability, as recently done with psilocybin's effect on mouse behaviour ³⁶ or the Stroke Preclinical Assessment Network.³⁷ (6) Reverse translation strategies of mimicking clinically successful treatments (e.g., modeling contingency management (rewarding voluntary abstinence) and community reinforcement approaches in addiction ³⁸) to identify new biomedical treatments and elucidate neurocircuits affected.^39,40 (7) The laboratory environment itself may be a limitation in psychiatric models. Rodents live and are tested in highly controlled, low-complexity settings that are different from dynamic, unpredictable environments in which human psychiatric symptoms develop. Thus, increasing the complexity of the environment may be beneficial for stress and social neurobiological studies. Acknowledging these pitfalls and improving experimental design has led to better animal models improving translatability. That said, there are still issues with experimental design and reproducibility of data that the animal research community need to grapple with to better enhance the translatability of findings.

Limitations with NAMs

Large cohort genomic studies of individuals with neuropsychiatric disorders have led to a wealth of genomic data on the genetics of these disorders.^41,42 Human inducible pluripotent stem cells and induced neuron-like cells can take advantage of these genetic studies by taking patient somatic cells and reprograming them to iPSCs, which can then be grown into monolayers of a single neuronal-like cell type or into complex three-dimensional organoids. Subsequently, these cellular models can be used for phenotypic characterization of electrophysiological characteristics or protein and/or gene expression changes. For example, cerebral organoids generated from schizophrenia patient-derived iPSCs were used for transcriptomic analysis and found differentially expressed genes in synapses and nervous system development. However, they observed no changes in basal electrical activity, but a diminished response to depolarization.⁴³ It is important to consider that schizophrenia manifests in late adolescence and early adulthood while cerebral organoids are younger than 1 year, and these models may be more amenable to examining neurobiology related to neurodevelopment, especially given that gene expression and epigenomic studies indicate that cerebral organoids model fetal cortical development.⁴⁴ For neuropsychiatric disease, efforts into creating more mature organoid tissue will be required to develop later pathological events such as circuit dysfunction or abnormal neuron–glia interactions.

Importantly, there are several considerations that limit interpretation of data derived from organoids. First, they are necessarily studied under highly artificial conditions, often lacking well-established critical disease risk factors, and that they exclude any environmental or social factors that may impinge on genetic background influencing disease onset or disease progression. Secondly, organoids exclude hormonal, microbiome, immune, and metabolic systems influencing brain function and behaviours. There is no way to model these inter-organ factors in a dish. Thirdly, mental illnesses arise from complex, network-level dysfunctions that involve many brain regions interacting dynamically (e.g., prefrontal cortex–amygdala–hippocampus loops). Cerebral organoids are derived from a single neuronal phenotype (although some have developed protocols to generate several cell types of the human brain),⁴⁵ and thus dynamics in brain network function cannot be examined. Finally, psychiatric disorders are defined by changes in behaviour, including anxiety, social withdrawal, compulsions, anhedonia, and so on. Cell cultures and organoids do not exhibit behaviour. Thus, organoids can model specific cell types or circuits, but not the integrated, systemic biology that shapes cognition, emotion, and motivation in a living organism. Organoids are useful to study how genetics may influence protein dynamics and localization and can inform studies that are possible in vivo, in the intact brain. Thus, organoids may provide a complementary approach to animal models. Findings could be triangulated across cell culture/organoid, animal, and human/clinical approaches. Ideally, grant applications in psychiatric research demonstrating this approach should be prioritized for funding.

Artificial intelligence is revolutionizing neuroscience research by allowing researchers to gain insights into complex datasets and developing predictive models for neurological disorders. Modern neuroscience techniques deliver numerous multidimensional variables, and the use of machine learning strategies allow for solutions to understand the data. Examples include using and understanding complex neural activity patterns,^46,47 making sense of -omic information,⁴⁸ and creating predictive models of behaviour.^46,49–51 Additionally, breakthroughs in protein folding prediction ^52–54 allow for billions of protein combinations to predict structures, accelerating drug discovery research.^53,55 Ironically, many computational models are built using processes discovered in animal models (e.g., deep neural networks, neuromorphic computing) and trained or benchmarked using existing animal data. Progress in machine learning occurs at pace where it is likely that we will be able to build powerful causal models that predict across all scales within the not-too-distant future. However, AI depends on high-quality biological data, which often comes from animals. AI and machine learning models require large, validated datasets of neural recordings, imaging, or gene expression to learn from. Much of these data still originate from controlled animal studies where variables can be tightly managed. Thus, AI is useful for analyzing large data sets from animal or human studies to detect subtle patterns, something that may help advance animal models in neuropsychiatry. Indeed, there is value in developing large-scale, open-access behavioural data repositories for rodent models that can capture high dimensional behavioural data. Animal models are still required to validate AI-generated predictions about disease mechanisms or drug targets. Thus, AI models can simulate correlates of these behaviours but not test biological causation. Only animal models allow researchers to connect neural activity and molecular changes to observable, quantifiable behaviour. Given the current gaps in our understanding of brain function in both health and disease, using AI models trained on limited data is likely to yield unforeseen errors. Furthermore, biology operates on principles rather than strict rules, and AI systems struggle to handle uncertainty and redundancy. While AI can effectively model simple chemical or physical processes, it is far less capable of capturing the adaptive behaviour of biological systems, such as the ability to switch between multiple redundant brain or cellular signaling pathways in response to environmental changes. Taken together, AI models can predict and simulate, but animal models are necessary to demonstrate and validate biological and behavioural phenomena in a living system.

Drug development still requires systemic testing. Organoids and AI can be used for high throughput drug screening to identify potential drug targets.^53,56 Before any new psychiatric medication reaches human trials, it must be tested for safety, efficacy, and mechanism in animals. A whole system is required for testing the pharmacokinetics of how a compound is absorbed, metabolized, and excreted. Furthermore, whole animals are required for knowing whether and the time course at which it crosses the blood-brain barrier or whether the compound has unintended effects on other organs. No NAM currently replicates the whole-body pharmacokinetics and interactions of a living organism. As such, regulatory agencies (including the Food and Drug Administration, European Medicines Agency, Pharmaceutical Drugs Directorate, Health Canada) are open to NAMs, but they still require animal data for most new psychiatric therapeutics. As NAMs become more predictive, policies may evolve. However, we are not yet at a point where they can fully replace in vivo testing. Without this step, human trials would be unsafe and less informed.

The stake of moving prematurely into clinical trials could be high. Gene therapy was gaining significant momentum in the late 1990s as a cure for genetic diseases; yet an early clinical trial led to the death of a participant due to a severe immune reaction to the adenoviral vector.⁵⁷ This casted a lasting dark cloud over the entire field for decades, as the trial was advanced before full pre-clinical studies were able to assess the immunogenic effects of adenoviral vectors in animal models. Thus, systemic testing in animal models is still a necessary step to ensure safety in humans.

Animal welfare

The welfare of animals used in research is of utmost importance to researchers. In Canada, guidelines for humane and ethical animal use in research are set by the Canadian Council for Animal Care (CCAC). This national organization is governed by a Board of Directors, that includes researchers, veterinarians, and representatives from government, industry, and the broader community. Additionally, a series of member organizations ranging from the SPCA to the U15 Group of Canadian Research Universities advise the CCAC. Within research institutions, Animal Care Committees implement and enforce these guidelines. Each procedure is reviewed by animal care committee members for its scientific merit, justification for humane animal use, animal welfare considerations, and compliance with the “3Rs”; Replacement, Reduction, and Refinement. The 3Rs has sometimes led to underpowered studies, leading to challenges in translation. However, with a publication requirement of reporting ARRIVE guidelines (Animal Research: Reporting of In vivo Experiments),⁵⁸ there is now a requirement of sufficiently powered studies. Post-approval monitoring ensures that approved procedures are carried out appropriately and that animal welfare standards are maintained. The CCAC conducts comprehensive assessments of institutional animal care and use programs every 5–6 years, with interim visits as needed, to ensure institutes meet the criteria for maintaining their “Good Animal Practice (GAP)^R” certification. Institutes that do not maintain GAP^R certification risk losing their ability to hold tri-council (CIHR, NSERC, SSHRC) funding. Taken together, researchers and institutions are held to the highest standards by the CCAC to ensure humane and ethical animal practices in research and teaching. This is important for both animal welfare considerations as well as the integrity and quality of the scientific data.

Conclusions

Animals remain essential in psychiatry research because they allow scientists to investigate brain–behaviour relationships and mechanisms that cannot ethically or practically be studied directly in humans. Although emerging technologies promise greater access to human biology without relying on animals, our current understanding of biological complexity still depends on animal research. In the future, it may be possible to leverage the knowledge gained from such studies to train artificial intelligence and mine large-scale databases for drug discovery, reducing or even eliminating the need for animal testing. Indeed, several companies, such as Recursion, are already working to harness large cell atlases and imaging datasets toward this goal. However, these technologies are nascent and unproven. We also do not know enough about the biology of brain–body functional relationships to know where real breakthroughs will come. The future may be to use NAMs and AI for early-stage screening, mechanistic insights, and reduction of animal use. This “3Rs” approach is the guiding principle for ethical biomedical research and would allow for the use of refined animal models only when necessary to understand systems-level effects and behaviour. However, until translational alternatives become available, and the value of these alternatives has been proven, animal use in neuropsychiatric research is still necessary for discovery.

Author contributions

Conceptualization: SLB

Funding acquisition: SLB

Project administration: SLB

Writing – original draft: SLB

Writing – review & editing: SLB