2.1 A mosaic of idealized “models of” and “models for”

Although the idea that many scientific models are constructed representations is well established (Teller Reference Teller2001), applying this conception to model organisms demands a closer examination. Classical accounts hold that models represent their targets through analogy—so that a model’s validity depends on the relevant similarities it shares with its target (Godfrey-Smith Reference Godfrey-Smith2006; Teller Reference Teller2001). In scientific practice, models are employed to capture specific aspects of the world for particular investigative purposes (Giere Reference Giere2004; Parker Reference Parker2020). This point is especially crucial for model organisms, which gain epistemic legitimacy not by virtue of universal validity but in relation to specific research questions or hypotheses.

In biological research, model organisms are employed to capture specific aspects of biological phenomena while enabling experimental intervention. Researchers deliberately idealize these organisms—simplifying certain variables or interactions—to render complex systems more comprehensible or to isolate key causal mechanisms. As a result, the representation of biological phenomena by model organisms is not merely a passive mirroring of reality but also instead fulfills a dynamic, dual function: Model organisms both represent a target phenomenon and serve as experimental setups for manipulation or intervention (Morrison and Morgan Reference Morrison, Morgan, Mary and Morrison1999). In other words, they function dually as models of the target phenomena—providing a constructed representation—and as models for experimental inquiry—serving as tools for manipulation and measurement (Fox Keller Reference Fox Keller2000; Ratti Reference Ratti2020). More specifically, a model organism sets the conditions under which a phenomenon is realized, delineating the variables that can be manipulated and measured (Baetu Reference Baetu2014). In this context, the choice of a model organism is guided by the scientific objective, with judgments of relevance concurrently shaping both the representation of the phenomenon and the experimental protocols that yield pertinent data (Peschard and Fraassen Reference Peschard and van Fraassen2014).

Importantly, biologists rarely rely on a single model organism. In many scientific investigations, multiple model organisms, each based on different assumptions, are used to capture distinct facets of the target system (Weisberg Reference Weisberg2007). Knowledge is advanced through cross-model comparisons, a strategy Baetu (Reference Baetu2014) refers to as mosaicism. This mosaic approach reinforces the epistemic foundation of model organisms research by revealing which aspects of a phenomenon are robust across different models.

Altogether, these aspects—the use of idealization, the integration of intervention with representation, and the practice of mosaicism—are fundamental for understanding the unique epistemic status of model organisms as scientific models.

Despite general agreement that model organisms frequently serve a representational role, debates persist in philosophy of science over whether representation is always involved (Germain Reference Germain2014; Levy and Currie Reference Levy and Currie2015; Sartori Reference Sartori2023; Weber Reference Weber2014). For instance, organisms can be studied for what has been called “preparative experimentation” (Weber Reference Weber2004). Through the studies on model organisms, practical knowledge, protocols, and tools are produced. This expertise on rearing conditions or genetic engineering, for example, shapes how experiments are conducted and interpreted. In some research teams, developing and validating such tools is the sole purpose of model organism-based research, rather than modeling an external target.

Thus, model organisms can also serve as practical instruments for tool development, even though they most often act as representations of a target. In this regard, the notion of representation has been refined in two distinct ways in the context of model organisms by the philosophers Leonelli and Ankeny: the representational target and the representational scope.

2.2. Representational target and scope

Within the context of model organisms, the “representational target” refers to the phenomena that can be investigated through the use of an organism (Ankeny and Leonelli Reference Ankeny and Leonelli2011, Reference Ankeny and Leonelli2020). Model organisms serve to represent a diverse array of targets, ranging from specific biological processes to diseases or even entire human patients (Green Reference Green2024). Some scholars differentiate between models of human diseases—often labeled “surrogates”—and models of general phenomena, known as “exemplars” (Baetu Reference Baetu2014; Bolker Reference Bolker2009). The distinction is based on the fact that in models of general phenomena, the phenomenon is both instantiated in and represented by the organism, whereas in disease models, the organism does not fully manifest the target but represents only certain of its features.

Defining the representational target of model organisms raises a subsequent question regarding the generalizability of findings. This idea is captured by the notion of “representational scope,” often termed external validity, which refers to the degree to which a result can be applied to broader contexts or populations. In model organism research, external validity is frequently underpinned by evolutionary considerations (Schaffner Reference Schaffner, Machamer, Grush and McLaughlin2001; Weber Reference Weber2004). While known phylogenetic relationships provide a rationale for extrapolating findings, they do not guarantee that a biological process will generalize without limitation: Species-specific differences can and do constrain the applicability of results. Importantly, the evolutionary argument does not negate these differences but rather explains why generalization may be plausible in the first place.

Depending on the research objectives, the scope of a given model organism may be narrow, applicable only to individuals within the same species, or broader. When the scope is confined to one species, two scenarios emerge. In one case, the target phenomenon is unique to that species, so the findings pertain solely to its members. In the other case, while the target exhibits species-specific characteristics, the study nevertheless yields insights that contribute to a more general understanding. For instance, to understand the principles underlying information coding in the central nervous system, studies are conducted using various species. For each model organism, the representational scope is inherently limited by species-specific peculiarities. Neurotransmitter actions may vary significantly: In arthropods, glutamate functions as an inhibitory agent, whereas in mammals it serves an excitatory role. Nonetheless, by comparing and integrating findings from multiple models, each with its own constrained scope, researchers are able to build a broader theoretical framework for neural information coding that applies across species (Laurent Reference Laurent2020). In other words, although the unique characteristics of any single species may constrain the scope, the transfer of results across different models enables the construction of a broader conceptual understanding. This target is not represented by any single model organism in isolation but is constructed from the collective knowledge generated by multiple models, each with its own limited scope. Such an integrative approach ultimately influences the interpretation of experimental results obtained within the restricted scope of each model organism.

Among both scientists and philosophers, it is frequently assumed that the representational scope of a model organism can be partly determined by a priori considerations. Such considerations typically take into account, on one hand, the phylogenetic distance between the model and the target species, and on the other hand, the nature of the phenomenon or mechanism under investigation. Among philosophers, these arguments have been notably developed by Weber (Reference Weber2004). According to this perspective, it is plausible to hypothesize that exocytosis—a cellular process in which vesicles fuse with the plasma membrane to release their contents—constitutes a low-level, highly conserved mechanism with broad generalizability. In contrast, complex phenomena such as elaborate behavioral patterns (e.g., food intake-related obesity), or emotional states, might be expected to exhibit a narrower scope, thereby confining their study to mammalian models.

However, we contend that neither a phylogenetic proximity nor a low-/high-level distinction is sufficient to determine representational scope. To illustrate, while exocytosis mechanisms show remarkable similarities between unicellular eukaryotes and human neuronal synapses (Emes and Grant Reference Emes and Grant2011), comparative studies reveal that mouse and human synapses differ in the composition of key proteins (Bayés et al. Reference Bayés, Collins, Croning, Van De Lagemaat, Choudhary and Grant2012). This indicates that even for low-level processes presumed to be broadly conserved, important variations can constrain generalization. Conversely, high-level complex phenomena can, in fact, be studied using organisms that are often considered “simpler.” For instance, genes associated with obesity in humans have also been identified in C. elegans, and manipulating these genes in worms exposed to a high-fructose diet results in weight gain (Ke et al. Reference Ke, Reed and Yang2021). Another example is provided by studies that have investigated the patterns of neuronal activity underlying emotional states in zebrafish larvae and drosophila (Athéa et al. Reference Athéa, Heck and Forest2024). Consequently, classifying phenomena as low-level (and likely to be conserved) or high-level (and unlikely to be conserved) does not provide a sufficiently reliable method for determining the generalizability of results.

Another version of this argument might maintain that we should instead rely on gene conservation as the basis for extrapolating findings. We contend that this approach can also be misleading. For instance, consider a gene from Salpingoeca rosetta, a unicellular organism that is closely related to animals, which encodes what is typically known as a synaptic protein. In multicellular organisms, synaptic proteins are integral components of nerve cells; they are localized at synapses (the specialized junctions where neurons communicate) and are crucial for transmitting signals. However, in S. rosetta, although the gene retains protein domains usually associated with synaptic function, it is expressed in the nucleus where it performs functions unrelated to neural communication (Burkhardt et al. Reference Burkhardt, Grønborg, McDonald, Sulur, Wang and King2014). This example illustrates that even when gene sequences appear conserved, the actual cellular context and the resultant functions may differ markedly. Consequently, a more reliable assessment of representational scope requires analyzing the conservation of multiple genes together with a detailed understanding of the corresponding protein-level mechanisms.

These considerations lead us to propose that the representational scope of model organisms is more effectively assessed a posteriori and through systematic comparisons across different models. In practice, extrapolation is evaluated by comparing and ranking results obtained from a spectrum of models—from C. elegans and other invertebrates to primates, and even to reduced models such as cell lines—under varying experimental conditions. This holistic strategy of multiple comparisons ensures that the external validity of findings is continuously interrogated and reevaluated (Baetu Reference Baetu2016). Ultimately, the external validity of results from any single model organism depends on the extent to which these results can be coherently integrated with those from other models, thereby constructing a robust and comprehensive understanding of the target phenomenon.

3.1 Between specimens and artifacts

A first distinguishing characteristic of model organisms is that they are actual, living specimens—tangible exemplars of the biological world. To our knowledge, the literature does not explain in what sense their status as specimens has epistemic relevance. We argue that it does: The very fact that a model organism is a living specimen fundamentally affects the validity of the knowledge it generates. Unlike computational or physical models, which may omit unknown yet crucial parameters, a living specimen inherently embodies all its biological attributes, both those that are known and those that remain undiscovered. This inherent completeness ensures the minimal validity of model organisms.

It is evident that this inherent completeness is necessary, yet insufficient. Even if all variables are present within a model organism, overlooking certain factors can undermine their interpretative contribution, regardless of their causal significance. Nonetheless, because model organisms serve as natural exemplars rather than as reduced models, successive empirical studies can incrementally reveal these hidden variables. In this way, model organisms function as idealized models that retain the authenticity of a complete exemplar.

Beyond their status as natural specimens, model organisms are also artifacts, defined as objects intentionally created or modified by human intervention to fulfill specific research objectives. For example, standard laboratory mouse strains are produced through inbreeding to achieve a level of genetic uniformity that does not occur in wild populations. Another example is provided by animal models of genetic diseases. In these models, the mutation known to cause the disease in humans is introduced, typically into mice, to replicate the condition, thereby enabling the study of its underlying mechanisms. The artifactual nature of model organisms underscores their dual character: They are simultaneously natural exemplars, embodying the full complexity of the organism, and products of experimental design, subject to the practical constraints of biological manipulation (Ankeny and Leonelli Reference Ankeny and Leonelli2011; Bolker Reference Bolker2017; Rader Reference Rader2004).

3.2 Plastic entities

Model organisms are, in part, artifacts precisely because they can be genetically modified in ways that do not occur naturally. These modifications are employed either to construct the model—as in the case of mouse models of genetic diseases, where the introduced mutation serves as the etiological cause of the pathology—or to use the model, for instance by altering a specific gene to study its function. In both cases, however, the intervention can never be “ideal” in Woodward’s sense (Reference Woodward2010) because of the inherent plasticity of living organisms.

For instance, in genetically modified organisms, a phenomenon sometimes referred to as genetic compensation may occur. Here, the observed phenotype results not solely from the intentional genetic alteration but also from compensatory changes in the expression of other genes (Jakutis and Stainier Reference Jakutis and Stainier2021). This dynamic response indicates that a model organism is not merely an artifact of an “ideal intervention”; it is a self-modifying entity whose internal state can shift in unforeseen ways. Such plasticity complicates our idea that model organisms are inherently valid because they are natural exemplars: If we cannot ascertain which variables are modified through compensatory processes, then generalizing findings becomes problematic. Yet, there are instances where similar systemic effects occur in both the model organism and the target. Consider genetically modified organisms designed to model specific genetic diseases. For instance, a mutation in the FMR1 gene leads to its silencing, thereby causing Fragile X syndrome, a condition associated with intellectual disability and features of autism spectrum disorder. Genetically modified mice have been developed to represent this disease by mimicking the loss of FMR1 gene expression (Santos et al. Reference Santos, Kanellopoulos and Bagni2014). In these models, one might expect compensatory mechanisms to obscure the disease phenotype; however, the genetic compensation occurs in both the mouse models and in human patients. In both cases, this compensation is better named “systemic effects” because it does not prevent the pathology. This similarity suggests that such compensatory responses are not unique to the model organism but are shared features of the pathology, thereby reinforcing rather than detracting from the model’s validity. Thus, the self-modification capacities do not necessarily compromise the model organism’s epistemic reliability.

Equally important is phenotypic plasticity, defined as the ability of a single genotype to give rise to different phenotypes under varying environmental conditions (Pigllucci Reference Pigllucci1996; Nicoglou Reference Nicoglou2024). For example, a rodent maintained under standardized laboratory conditions is markedly different from one living in the wild. Empirical studies have shown that enriched housing environments can alter brain structure and connectivity (Bennett et al. Reference Bennett, Diamond, Krech and Rosenzweig1964). These findings underscore that the conditions under which model organisms are maintained critically shape their biological responses, thereby reinforcing that model organisms function as both natural exemplars and as artifacts of experimental design.

3.3 An experimental system

Moreover, model organisms must be understood in the context of the experimental conditions in which they are studied, as these conditions shape the phenomena that the organisms bring forth. In many experiments, variables are controlled to a significant extent, which in turn restricts the range of phenomena that emerge and are ultimately represented by the model organism. In other words, contrary to Weber’s suggestion, we argue that the representation provided by a model organism is inseparable from the experimental settings and the parameters measured.

Consider, for example, investigations into how the sensory cortex processes environmental information. In one series of studies, the neural activity of a rat’s somatosensory cortex was recorded while its whiskers were mechanically stimulated. Initial experiments conducted on anesthetized rats yielded one pattern of activity; subsequent recordings in head-fixed awake animals produced different results; and further studies correlating cortical responses with whisker touch in freely moving rats revealed yet another pattern (Erchova et al. Reference Erchova, Lebedev and Diamond2002; Nicolelis and Fanselow Reference Nicolelis and Fanselow2002). These variations indicate that the biological process being represented is deeply influenced by the experimental conditions, such that the phenomenon is not an invariant property of the organism but is coconstituted by the methodological framework. This perspective aligns with the view in the philosophy of experiment that phenomena are, in many cases, brought into being through experimental practices rather than simply being discovered (Hacking Reference Hacking1983, Reference Hacking and Pickering1992).

Even when the phenomenon is considered in its full complexity, researchers focus on selected elements by controlling or omitting certain variables. This process of idealizing the target results in a model that is not solely the organism but a composite of the organism, the specific experimental settings, and the chosen measured parameters. The way in which a phenomenon is understood depends on this intricate interplay between the organism and its experimental context, as well as on the variables selected for observation. In the end, the study of the organism provides an idealized explanation of the biological process that is being targeted.

In summary, organisms can only function as models when they are embedded within an experimental protocol and a broader body of background knowledge that justify experimental choices (Hacking Reference Hacking1983; Shelley Reference Shelley2010). While the living organism constitutes the central component, it is part of an entire experimental system. Thus, the construction and validation of a model depend on background assumptions (such as fundamental concepts of cell biology), methodological constraints, and the experimental conditions imposed. In other words, the relationship between the model and the instruments or protocols used to manipulate it is constitutive (Rheinberger Reference Rheinberger1997). Regarding Rheinberger’s terminology, we find ourselves largely in agreement with his characterization of “experimental systems” as the smallest integral working units of research—that is, as complete experimental arrangements designed to produce knowledge, and characterized by technical, epistemic, social, and institutional aspects. However, we diverge from his perspective on model organisms. Indeed, he does not regard model organisms as epistemic things and reduces them to mere “organic instruments.” In contrast, based on the arguments developed in the preceding text, we regard model organisms as full-fledged experimental systems that oscillate between being epistemic entities and technical objects. As Rheinberger states, “[A]n epistemic object is first and foremost a question-generating machine” (Reference Rheinberger1997, 32), which we maintain is precisely the fundamental role that model organisms fulfill.

As a result, a model organism is a dynamic entity on three levels: the biological organism, the experimental procedures (including procedural adjustments), and the conceptual framework used to justify experimental choices (defining the target, selecting concepts to interpret results, etc.). These components may be modified during the course of a research project to address new questions or to enhance the reliability (internal validity) or generalizability (external validity) of the results. In every case, the model’s selection and construction are guided by the choice of the components to be represented, a choice that depends on the scientific question.

3.4 A way to do science

Finally, the use of model organisms embodies a distinctive mode of scientific practice. The standardization of model organisms; through controlled breeding, uniform housing conditions, and harmonized experimental protocols; is a collective enterprise shaped by the scientific community. This standardization facilitates the sharing of critical resources, such as reference genomes and neural connectomes, through centralized platforms like WormBase or FlyWire (Devineni Reference Devineni2024; The C. elegans Sequencing Consortium* 1998). Thus, the utility of a model organism is not solely a function of its biological properties but also of the infrastructural, methodological, and communal support that underpins its use in research (Ankeny and Leonelli Reference Ankeny and Leonelli2011, Reference Ankeny and Leonelli2020).

3.5 Summary: The six characteristics of model organisms

In summary, our analysis distinguishes six interrelated features that elucidate the unique epistemic status of model organisms and set them apart from other types of scientific models:

1. 1. Diversity of representational targets. Model organisms are employed to investigate a wide spectrum of phenomena, from basic biological processes to clinical conditions. Each target is typically explored through a range of models.

2. 2. A representational scope grounded in evolutionary justification. The generalizability of findings from a model organism is supported by evolutionary arguments, rendering it a priori plausible to extrapolate results across species. However, as we have demonstrated, this alone is not sufficient to justify such extrapolations.

3. 3. A distinctive mode of representation. Unlike other scientific models, model organisms function as both natural exemplars, embodying the full complexity of biological systems, and as products of artifactual, idealized experimental design.

4. 4. A self-modifying capacity. Model organisms are not static entities, but actively interact with, and adapt to, their experimental environments.

5. 5. A dynamic experimental system. Living organisms are deeply embedded within evolving experimental and theoretical frameworks. All these components (the biological organism, as mentioned in point 4, the experimental procedures, and the conceptual framework) can evolve during the course of a research project.

6. 6. Unique practical dimensions. The use of model organisms brings with it distinct practical methodologies and research practices. These include standardized breeding, controlled housing conditions, and collaborative infrastructures, all which contribute to the epistemic strength of the models.

Collectively, these six characteristics not only delineate the distinct nature of model organisms but also provide a framework for understanding their role as scientific models.