HPLS (2014) 36(1):114–128 DOI 10.1007/s40656-014-0007-0 ORIGINAL PAPER

From replica to instruments: animal models in biomedical research Pierre-Luc Germain

Received: 28 November 2013 / Accepted: 16 March 2014 / Published online: 10 July 2014 Ó Springer International Publishing AG 2014

Abstract The ways in which other animal species can be informative about human biology are not exhausted by the traditional picture of the animal model. In this paper, I propose to distinguish two roles which laboratory organisms can have in biomedical research. In the more traditional case, organisms act as surrogates for human beings, and as such are expected to be more manageable replicas of humans. However, animal models can inform us about human biology in a much less straightforward way, by being used as measuring devices—what I call their instrumental role. I first characterize this role and provide criteria for it, before illustrating it with some examples from biomedical research, especially cancer research. In such an instrumental role, phenotypes are not expected to phenocopy human phenomena, but instead have the purely instrumental value of detecting or measuring differences. I argue that the instrumental role is more prevalent than might first be suspected, and that some characteristics of contemporary biomedical research are increasingly shifting the use of laboratory organisms to the instrumental role. Finally, in light of the distinction proposed, I discuss the meaning of the expression ‘‘animal model’’. Keywords

Animal models  Modelling  Measurement  Biomedical research

1 Introduction Laboratory animals have long been used in biological research aimed at learning about human biology, and as a consequence they have been represented as P.-L. Germain (&) Universita` degli Studi di Milano and European Institute of Oncology (IEO), Campus IFOM-IEO, Via Adamello, 16, 20139 Milan, Italy e-mail: [email protected]

123

From replica to instruments

115

surrogates for humans.1 To be sure, there is surrogacy to the extent that laboratory research is often pursued in replacement of clinical research which would be more suitable were it not for ethical and economical limitations. However, what I will call ‘‘the surrogacy view’’ is generally taken to mean something stronger, namely that it is the animal itself, rather than the larger laboratory context in which it is embedded, that is used as a surrogate for a human being. This is well illustrated by the following passage from Bolker (2009): In carrying out research in surrogates with the intention of benefiting other species (particularly our own), we make a series of assumptions. One is that the surrogate will respond to manipulations in the same way as the target would, if it were examined directly. (Bolker 2009, p. 490) The surrogacy view implies that the laboratory animals (or parts thereof) are used instead of humans in experimental designs which are, in other respects, roughly the same: that both the interventions performed and the endpoints measured on the animal are the same, in some sense, as those one would use with humans. When animal models are indeed used in this way, extrapolation rests on similarity between humans and the model. In this context, considerations of phylogenetic proximity are a proxy for similarity. For instance, the 1985 report of the Committee on Models for Biomedical Research of the National Research Council (NRC) discusses evolution as the source of homologies, adding that these are ‘‘of heuristic value in the search for analogues’’ (Committee on Models for Biomedical Research 1985, p. 17) On their view, To be successful in terms of yielding information and interpretation that permits prediction, any model must be an effective analogy (i.e., it must reflect an agreement, likeness, or correspondence in function) to the object or process being modelled. (Committee on Models for Biomedical Research 1985, p. 73) Full identity between the model and target systems is obviously not required, and the similarity should instead be with respect to the features under investigation.2 Most philosophers appear to endorse this conception of modelling through local/ partial similarity, most explicitly Steel (2008), Shelley (2010), Piotrowska (2012), LaFollette and Shanks (1996), Shanks et al. (2009). There are, however, major differences between their treatment of extrapolation: Steel (2008) offers a mechanism-based approach, while Shelley’s (2010) account rests on analogy between structures and LaFollette and Shanks (1996) require full identity of all causally-relevant features. Nevertheless, they all converge in requiring the model to be a (partial) replica of the target system. And as all of these authors noted, the 1

The 1998 report of the Committee of the Institute for Laboratory Animal Research on New and Emerging Models in Biomedical and Behavioral Research writes begins with the following definition: ‘‘A biomedical model is a surrogate for a human being, or a human biologic system, that can be used to understand normal and abnormal function from gene to phenotype and to provide a basis for preventive or therapeutic intervention in human diseases.’’ (Committee on New and Emerging Models in Biomedical and Behavioral Research 1998, p.10).

2

In fact, Keuck (2012) noted that in many cases, complete recapitulation of a disease, for instance, might not even be desirable.

123

116

P.-L. Germain

similarity between humans and any animal will always be incomplete, and there will always be causally relevant disanalogies.3 Hence many critics of animal experiments (most famously LaFollette and Shanks 1996) have argued that the animal is never predictive in a strong sense—a claim which hardly any reasonable scientist would deny (see for instance the aforementioned NRC report). While this view is certainly not wrong, it captures only a part (and perhaps not even the largest part) of the way experimental organisms are used in biomedical research. In this paper, I argue that there is another important way in which they are used, namely as observational instruments or measuring devices. Although scientists tend to call them ‘‘animal models’’ just the same, their epistemic status differs substantially from philosophers’ accounts of models. As such, there can be legitimate disagreement about whether they ought to be called models, and indeed part of the rationale for this paper is to explore the distinction between experimentation on material models and what we might call ‘‘direct experimentation’’. I will therefore leave it to my conclusion to come back to the question whether animals-as-instruments are models at all.

2 Two roles for laboratory organisms Historians of science have long emphasized the technological aspect of animal models, especially model organisms (Rheinberger 1997; Rader 2004; Gaudillie`re 2006; Gayon 2006; Landecker 2007).4 However, as Weber (2005, pp. 169–173) remarked, it would seem from these accounts that ‘‘instrument’’ is used in a rather superficial, metaphorical way, or that the analogy refers mostly to the social transactions involving model organisms; indeed their standardization is, from a social and historical point of view, very reminiscent of that of technological devices. If the notion of instrument is to be more than a metaphor, or at least a useful metaphor, the analogy has to extend further into the epistemological realm. Building on Weber’s treatment of the issue, I will propose a strict sense of the animal model qua instrument, explore the relevance of this view for contemporary biomedical research, and discuss its consequences for the problem of extrapolation. In his criticism of the analogy, Weber lays out the basic criteria for a proper (more than metaphorical) instrument: In a typical instrument, a causal input (e.g., the Earth’s magnetic field in the case of the compass) leads to an observable signal (the motion of the compass needle). What the user of an instrument is interested in is the process or object that is responsible for the causal input (the orientation of the Earth’s magnetic field). (Weber 2005, pp. 170–171) 3

Maugeri and Blasimme (2011) discuss how scientists actively correct these disanalogies.

4

Animal models are models which have the property of being animal, i.e. of involving animals, or parts of animals. In contrast, model organisms are organisms with the additional feature of being models, generally because they are expected to be representative of a broader class of organisms (see especially Ankeny and Leonelli 2011, as well as Gayon 2006). The present paper is not about model organisms, but about animals used in the study of human diseases—about animal models and their frontiers.

123

From replica to instruments

117

It is important to note that the sense in which ‘instrument’ is understood here is rather narrow: a scalpel is clearly an instrument (made to fulfill a function), but does not seem to fit Weber’s description. The meaning of instrument that Weber has in mind here is the more narrow meaning in which scientists often speak of instruments as measuring devices (by which I mean to encompass both measuring devices, that have an ordinal or quantitative output, and detection devices, that have a binary output). It is in this sense that I will use the word ‘instrument’ as well in this paper. However this does not exclude the possibility of other instrumental roles, for instance the use of organisms as ‘factories’ for materials. It is uncontroversial that laboratory organisms are sometimes used as detection devices, and many authors (including Weber 2005, p. 171) have given the example of animals brought into mines to detect toxic gases. Likewise, when Boyle put a mouse in his pump and saw it die, he was not trying to study the mouse, but was rather using it to learn something about the environment inside the pump (he was giving a shot at the Herculean task of detecting a vacuum). My aim, here, will therefore not be limited to making a claim about the existence of animal models as instruments. Rather, after having characterized this role more precisely, I will argue for its relevance to understand contemporary biomedical research, and will discuss its relationship with more traditional animal models acting as replicas of humans. In order to further characterize the instrumental role of animal models, it is appropriate to start with an obvious example. In the 1920s, Bernhard Zondek and Selmar Ascheim developed a pregnancy test known as the Ascheim–Zondek test, or A–Z test (Zondek 1928; see Olszynko-Gryn 2013 for a historical analysis). In this test, juvenile mice were injected with the urine of a female patient, and after 2 days were dissected. If the injection caused a maturation of the mouse’s ovarian follicles, then the patient was pregnant. Clearly, the animal had the function of detecting a signal in order to learn something about the woman. The rationale for this method was grounded on the conservation of the hormone governing the maturation of the follicles in both species (Zondek 1928, p. 1088), and therefore on a partial similarity between humans and mice. Nevertheless, the mature design of the test presented in the 1928 paper is clearly the result of an induction largely independent of questions of similarity. Indeed, the authors measured the accuracy of their test on a number of women in various conditions before concluding that the swelling of the reproductive organs is not a trustworthy signal: it can occur in the mouse, in different situations, even if the woman is not pregnant. Rather, they concluded that the relevant signal was the presence of small blood spots (the ‘‘Blutpunkte’’ of ‘‘Reaktion II’’) on the follicles (Zondek 1928, p. 1089), a phenomenon whose presence or absence in the pregnant woman is not even discussed in the paper. The reason is simple: although similarity between the two species might have prompted the test in the first place, its instrumentalisation only requires that the urine of pregnant women gives a reproducible signature in the mouse, independently of the nature of this signature. The blood spots, and the general reaction produced in the mouse, only serve as a map on which meaningful differences can be located. This is an important particularity of typical measuring devices, which can further refine the instrumental role of animal models and turn it into a more meaningful

123

118

P.-L. Germain

epistemological category. If an instrument were merely something that produces a signal informative about its target system, then the vast majority of examples of animal models could be represented in this way, thus undermining an important part of the utility of the concept. Take, for instance, the traditional example of an animal being used to test the carcinogenicity of a chemical compound, say tar. Tar (the input) is regularly applied to mice, and after some time the incidence of cancer (the signal) is measured. If there is a significantly higher incidence of cancer in the treatment versus the control group, then this is evidence for the carcinogenicity of tar. Hence the signal is used to learn something about the input—tar. The test relies on carcinogenicity in the mouse to extrapolate carcinogenicity in humans, and therefore informs us of the carcinogenic effect of the substance on both organisms. As the NRC committee puts it, ‘‘the modelling relationships are reciprocal’’ (Committee on Models for Biomedical Research 1985, p. 19). This is a feature which is generally not shared by typical measuring devices, because there is a decoupling between the output of the instrument—the signal—and the information this signal provides about the target system (from which the input comes). In fact, in many cases the signal and information regarding the target system each have no clear counterpart in the other system: it is not a valid conclusion to say, of the mice used in the A–Z test, that they are pregnant, nor can one conclude that a pregnant woman has maturing follicles or the blood spots. To further probe this particularity, it is useful to rely on the analog versus digital distinction (Goodman 1968), which was recently applied to traits and phenotypes by Meunier (2011). The following example, while simplistic with respect to Meunier’s discussion, is sufficient for our present purposes. A digital watch, for instance, tells us the time in a single and definite way: there are not different ways of reading the watch to learn about the time, and one will not get more information by looking at it more closely. In other words, it is straightforward to say whether two watches are giving the same time. A pressure gauge, by contrast, is analog: between any two points is always another one. As a consequence, two observers may have a different reading of the gauge, depending on how closely they look, and it is not straightforward to say whether two such gauges give the same measurement. Being digital or analog is a property not of an object, but of the whole system of representation (a digital watch may be used as an analog for other purposes than knowing the time). Importantly, digital does not imply quantitative, and in fact Goodman discussed it primarily for non-quantitative settings, such as diagrams (e.g. directed graphs) or alphabetical representations (which are typically digital in the sense that there is no ambiguity in answering the question ‘‘what is written’’). What it does however imply is ‘‘notational’’, which links the readout to a representational system.5 While the digital/analog distinction does not perfectly fit the distinction I wish to discuss here, I simply wanted to note that measuring devices, in the sense described above, necessarily imply either (or both) a reduction of dimensionality, and/or some extent of digitalization. Once more, the A–Z test illustrates this. The biologist, 5

Discussing the intricacies of these distinctions is not the purpose of this paper, and readers are invited to refer to Meunier (2011) for a detailed discussion in the context of experimental biology.

123

From replica to instruments

119

during the dissection of the mouse, might notice a variety of phenomena: perhaps the mouse’s ovaries have a peculiar shape, a greenish taint, or the mouse has a particularly pungent smell. Or perhaps the follicles have particularly many of these small blood stains—but there is no way in which the woman is more pregnant, and from these facts the biologist will never draw any conclusion for the purpose of the pregnancy test. We can therefore characterize two different usages of laboratory organisms. The first is the more traditional animal model as conceived by the surrogacy view of modelling, the paradigmatic example of which is the dissection of an animal, observation of a phenomenon, and inference to the same phenomenon in humans. The inference relies on the animal being a more manageable replica, at least with respect to the phenomenon of interest, of a human being. I will refer to this role as the surrogate role, involving a model-as-replica. On the other hand, the animal can be used as a measuring instrument as I described. This is in many respects similar to what Gayon (2006) labelled as ‘‘tool’’ (‘‘organisme-outil’’), although his term seems to be more encompassing than the role I have tried to define here. For these reasons, I will instead refer to the role I described as the instrumental role. In the remainder of this paper, I will pursue three main goals. I will attempt to identify the epistemological implications of the instrumental role, especially regarding the problem of extrapolation. But I will also argue for the relevance of the instrumental role in contemporary biomedical research. More specifically, I will try to show that it covers many cases of animal models that are not normally perceived in this way, and will argue that the instrumental role is becoming increasingly more important, suggesting some tentative explanations for this phenomenon. In order to reach these goals, I will present two more examples of the instrumental role of animal models, going from the more obvious case to the less apparent one. Finally, I will conclude with a discussion on the nature of these systems and on the meaning and boundaries of the concept of ‘‘animal model’’.

3 Xenograft models As transplantation experiments exerted much fascination at the end of the nineteenth and beginning of the twentieth century, it was noticed early on that normal somatic cells could not survive xenotransplantation (or xenografting)—transplantation into a foreign species. Normal adult tissues quickly resorbed, while tissues of embryonic origin or malignant cancers could be engrafted successfully. This inspired a measuring device very similar to the A–Z test: it was proposed that ‘‘transplantability constitute[s] a biological test for cancer’’ (Greene 1948, p. 1364). Harry S. N. Greene, a Yale pathologist famous for his transplantation experiments, suggested that ‘‘the proof of malignancy lies in behavior’’, and therefore that the ‘‘study of the transplants allows a more precise classification than is warranted from the morphologic features of the biopsy specimen’’ (ibid.). The mouse was therefore not only a device for the detection of cancer, but a measuring device for the classification of cancers. The test however lacked sensitivity, and by the time it was correctly calibrated (most importantly by tinkering with the mouse’s immune

123

120

P.-L. Germain

system and altering the site of injection), histopathology had developed other cheap and reliable ways of detecting and classifying cancer. Nevertheless, contemporary xenograft experiments, especially in the field of cancer stem cells, are surprisingly similar to Greene’s test. According to the Cancer Stem Cell (CSC) model, cancer progression is driven by a small subpopulation of tumour cells with stem-cell-like properties, which are (exclusively) capable of infinite replication. While the bulk of tumour cells stop proliferating after a few rounds of replication, the CSCs can divide asymmetrically into a long-lived CSC (hence sustaining the CSC pool) and a short-lived but highly prolific daughter.6 The initial discovery of CSC in leukaemia prompted the search for similar cells in other forms of cancer. The basic strategy is to divide the population of tumour cells into subpopulations according to some markers, and assess whether and to what extent these subpopulations contain CSCs by serially transplanting them into immunodeficient mice. Scientists first transplant cancer cells into a first mouse, harvest the cells of the newly grown tumour, and transplant them into a second mouse. If the second mouse develops tumours, then the initial population contained CSCs. Obviously, practice is in reality fraught with challenges and complexities.7 The use of immuno-compromised mice is meant to reduce immune reaction to the foreign (trans-specific) tissue. Otherwise one would not be able to see the difference between an immune rejection of the transplant and an intrinsic incapacity of the cells to form tumours. The idea here is not that the immuno-compromised mouse is more similar to humans—it is not, for very few cancer patients lack a functional immune system. As Maugeri and Blasimme note, ‘‘the impairment of the immune system in NOD/SCID mice does not make these creatures more human but rather less murine’’ (Maugeri and Blasimme 2011, p. 612). The focus on removing the murine specificity rather than mimicking human specificity is best understood as controlling errors (Weber 2005), or more specifically controlling for artifacts (in this case, immune rejection of foreign tissues) rather than striving for a greater similarity. It might be argued that these xenografted mice are not animal models, but rather in vitro models in a ‘‘very complex medium’’, so to say. However, it is important to note that there are three very different ways in which the mouse can be said to be a ‘‘mere’’ test-tube. Until the 1940s, viruses were kept and grown in the lab by serial infection of host animals. In this context, the animal does not accomplish any epistemic task, but merely serves as a practical mean of keeping stocks of viruses. Although the animal can be said to have an instrumental role in a very broad sense, it is not an observational instrument, for it does not serve the purpose of making differences visible. In xenograft experiments, however, the mouse does serve such a purpose, and hence it can be said to be an instrument in the much narrower sense that I described earlier. Importantly, the same can sometimes be said with respect to in vitro models: while the medium in the dish is often designed for the mere 6

For a scientific scientific review of the CSC model, see Visvader and Lindeman (2012) or Valent et al. (2012). For philosophical discussions, see Blasimme (2013).

7

See for instance the debate of melanoma-initiating cells: Schatton et al. (2008), Quintana et al. (2008). Both examples are discussed in more detail in Germain (2014).

123

From replica to instruments

121

practical aim of maximizing cell growth and survival, it equally often has the purpose of making certain things visible (live stainings being the most obvious example). The xenograft experiments, as well as the A–Z test, seem to share a relevant characteristic allegedly absent from most cases of animal models: in both cases, the model includes a material input from the target system. In the next section, I will show with a last example that the hybridity of the material system—to the material transfer from the target to the model system—is only indirectly relevant. But this is also suggested by the fact that a particular feature of such hybrid systems is that they often oscillate between technical object and epistemic thing (Rheinberger 1997, p. 32). While the mouse clearly has an instrumental role in the experimental design just described, it must be emphasized that this is not intrinsic to the material system, but comes from the way it is used. Indeed, xenografts models can be used in an analogical way: if someone studies the way the transplanted tumours prompt the formation of blood vessels in the neighbouring mouse tissues, and infers that a similar process of angiogenesis occurs in humans, the xenograft model is used as a replica. In such a context, the organism’s phenotype can be considered as an analog representation in the philosophical sense (Meunier 2011). When this is the case, the mouse does not simply serve the purpose of making visible unobservable differences, but has the additional aim of mimicking the human micro-environment of the cancer cells.

4 Genetically engineered models Santoriello et al. (2010) genetically engineered a zebrafish line which, due to the selective over-expression of an oncogene (HRAS) in melanocytes, developed melanoma regularly after some weeks. Like most oncogenes, genes of the RAS family were first discovered in a transforming virus before a homologue was found in human cells (see Morange 1993, pp. 48–50). HRAS mutations, as well as the specific mutation used (G12 V), are very frequent in cancer, especially in skin cancers. It was therefore to be expected that the fish would develop a cancer akin to human melanoma (strictly speaking, however, the model should not be considered a model of ‘‘melanoma’’, but a tool in the investigation of HRAS-mutated tumours of melanocytic origin). If there was doubt about the previous examples, there should be no question as to whether this zebrafish line is a bona fide animal model. Yet I want to argue that it is continuous with the xenograft example in several respects. This first becomes obvious when one considers the genetic engineering involved more closely. The zebrafish line was created by inserting chunks of DNA into the genome of the organism—but where did these DNA chunks come from? The oncogene is a mutated version of the human HRAS, and at some point, hidden in the intricacies of laboratory histories, it was cloned from a human sequence extracted and isolated from tumour cells (Kraus et al. 1984). It was then cloned through plasmids, joined to a reporter and to a promoter that would allow conditional activation, before being inserted into the zebrafish genome (Santoriello et al. 2009).

123

122

P.-L. Germain

To what extent, then, is the model the result of human material being brought into the fish? If the mouse was simply an environment for the transplanted cells, so can the fish be an environment for the transplanted molecule. But would it make the slightest difference if the inserted DNA came directly from human cells, or if it was entirely synthetic, but designed after the sequence of the human gene? In the absence of any physical difference between the sequences, we are forced to conclude that it would make no difference. When ‘‘going molecular’’, the distinction between modelling and direct experimentation breaks down. Like the xenograft example, this genetically engineered model can be used as a replica: one observes the mechanisms by which the fish develops tumours, and hypothesizes that carcinogenesis (or any sub-phenomenon, such as migration, angiogenesis, etc.) proceeds in the same way in humans (or at least in human HRAS-mutated tumours of melanocytic origin). This extrapolation, however, is warranted only insofar as the fish is similar to the human, in full accordance with the similarity view. As a matter of fact, however, the model was used in a quite different way, for which the tumours—and the very fact that it formed tumours—were rather superfluous. Already at the larval stage, the fish display a hyper-pigmentation due to an overproliferation of melanocytes (see Fig. 1). Because of the speed at which this phenotype could be observed, and of the ease with which chemicals could be administered to the animals, it could be used in a high-throughput platform to test thousands of compounds.8 Although ultimately the goal of this screening was to find a drug for melanoma, its proximate goal was more precisely to find compounds that have an effect on the effect of the HRAS mutation. This becomes obvious when we look at the operationalization of the model. The typical experiment in molecular biology has two controls: a negative control and a positive control. For instance, an experiment to measure or detect the presence of some DNA stretch in a sample will typically contain, on top of the sample to be tested, a sample that is known to contain the DNA (the positive control), and a sample that is known not to, for instance water (the negative control). Detecting the DNA in the negative control, or not detecting it in the positive control, indicates that the experiment did not proceed correctly (due, for instance, to sample contamination). The drug screening, however, contained an additional control. For each compound, melanocytes were counted on four groups of fish: untreated wild-type fish, treated wild-type fish, untreated mutant fish, and treated mutant fish. If a compound supresses the effect of the HRAS mutation, we would expect the ‘‘mutant treated’’ group to depart from the ‘‘mutant untreated’’ group, and recapitulate the ‘‘wild-type untreated’’ group. In analogy with a basic experimental design such as the DNA detection example, the ‘‘mutant untreated’’ and ‘‘wild-type untreated’’ groups would represent, respectively, the negative and positive controls. However, the experimental design contained an additional control group, which had the particular purpose of controlling for effects that counteracted the mutation and 8

The information on the drug screening aspects of this zebrafish model comes from the few months I could spent under precious tutoring provided by Cristina Santoriello, at the IFOM-IEO Campus, Milano. The results of the screen have not yet been published, but very similar screens have been published on other mutant larval phenotypes, and they are reviewed in White et al. (2013).

123

From replica to instruments

123

Fig. 1 The larval phenotype of the zebrafish model (photograph by author)

yet were independent from it. For a compound could rescue the hyperpigmentation phenotype by means completely independent from the mutation, which would most likely be irrelevant to the cancer. If a compound could rescue the phenotype, in other words bring back, in the mutant, the melanocyte count to that of the untreated wild-type, without significantly reducing the melanocyte count in the (treated) wild-type, then this compound had an effect on (the effect of) the HRAS mutation. It could then be studied further as a potential drug. The important point here is that the compounds identified are not those having an effect on the zebrafish itself, nor on its tumour: in fact we could run the same test if the mutant never developed cancer. What matters is that the compound rescues the specific mutant phenotype—whatever it is—and does not simply has an inverse effect. This procedure effectively reduces the observed effect of the compound to the causal pathway connecting the mutation (the difference-maker) to the phenotype (the difference). The genetic engineering plays here the same role that is played by the material transfer from target to model system in the xenograft example as well as in the case of the A–Z test. Both ensure that the difference-maker is shared between the model and the target system. Maugeri and Blasimme (2011) argued that scientists can control or restore the analogy between model and target systems by modifying the animal model. However, representing the present example in this way might be misleading, for it ignores the very particular way in which the model was actually used. Because one knows that the difference-maker (the mutation) is shared by the model and target system (as a consequence of having been engineered in that way), and because the experimental setting (with the double control) allows one to reduce the effect of a compound to a modification of the effect of the difference-maker, one is able to establish a causal interaction between the compound and the pathway through which the mutation has its effect, and to extrapolate this interaction to the

123

124

P.-L. Germain

target system in an imperfect but quite robust way. Compared to models as replica, in this context the strength of this extrapolation is considerably less dependent on the similarity between man and fish, and most importantly between human and fish tumours. In fact, the tumours are not even used, and the phenotype could have been virtually anything, for it has a purely instrumental value (on the instrumental value of phenotypes, see Meunier 2012). It is simply a detection device, signalling the causal relevance of the compound with respect to the immediate effects of the mutation.

5 The rise of the instrumental role The last example shows that the instrumental role is not limited to some anomalous or borderline cases like xenotransplantation, but is also very likely to be common in many other fields, especially (but not exclusively) where genetically engineered models are used. In fact, I would claim that this use is becoming more and more important, because of three intertwined reasons which I will now briefly discuss. The first reason has to do with reductionism: the zebrafish example yields information that is barely above the molecular level. In such a context, organisms are but a way to probe molecules. This general strategy does not need to assume that phenotypes are reducible to the molecular, but simply that relevant knowledge—in the case of cancer, means of intervention—can be gained at the molecular level. It also assumes a relative generality of molecular interactions: if a compound A interacts with a pathway B in a given context, then it does so in most contexts. This assumption is pervasive in contemporary biology, especially molecular biology (for instance in the use of interaction databases), which does not mean that it goes untested, but simply that it is hypothesized by default (and productively so). The second reason is the scientists’ increasing capacity to craft their animal model (Maugeri and Blasimme 2011). As I showed with the zebrafish example, genetic engineering can also fulfill the same role as the material input from the target system: that of ensuring that the difference in which we are interested is the consequence of a same difference-maker. Obviously, this in the first place requires having an actual difference-maker in the target system. This is only possible if the target system is narrow and homogeneous enough, which brings us to the third reason. As opposed to biology in general, biomedical research is interested in gaining knowledge about humans and human pathologies, and this narrow target class enables the crafting of animal models specifically geared toward extrapolation. Finally, the fourth reason is the distributed nature of biomedical research. Much of the philosophical criticisms of animal models (LaFollette and Shanks 1996; Shanks, Greek and Greek 2009; Knight 2011) seems to rest on a simplistic perception of biomedical research inspired by massive screenings such as those carried out by the early Cancer Chemotherapy National Service Center (CCNSC). In such a research pipeline, models are simply sieves: drugs passing the in vitro models are tested on animals, before being finally tested in clinical trials. Each step in this linear and unidirectional path is taken to be independent from the others. This view of drug discovery is unrepresentative of contemporary research, and I would in fact

123

From replica to instruments

125

argue that it is even inappropriate to understand the last three decades of the CCNSC itself. Instead, most of biomedical research proceeds in a highly non-linear way, with scientists shifting between different models, transferring samples and materials from one to the other, introducing insights from the clinic into their bench work, and so on. More importantly, in this more elaborate way of knowing, it is not a single laboratory organism that is expected to predict clinical outcomes, but a network of interacting model systems. That is to say, different models are not simply juxtaposed, with their findings adding up to increasing credibility, but instead the very meaning of one model depends on the other. As a consequence, each of these models is not expected to act as a surrogate for a human patient, and its scope is much narrower. They are certainly not models of a given disease anymore, and taken in isolation it is not even clear that they are models for a given disease: like other instruments, they are small nodes of a research network, each meant to detect and/or measure highly specific phenomena. The reductionism mentioned above is not so much the reflection of a reductionist conception of disease, but rather of this changing way of doing biomedical research. It is because of this context that it makes sense to design experiments in order to offer molecular knowledge that can be more easily transplanted and extrapolated across different material systems.

6 Conclusion The English language distinguishes two usages of the word model which have been discussed in philosophy of biology especially by Keller (2000, 2002): ‘‘model of’’, and ‘‘model for’’. Models-of have a structural relationship of representation with some target object, while models-for are for doing something: they afford actions, or are ‘‘for the study of’’ (Gayon 2006, p. 28). Of course, models are always both of and for, and their ability to afford actions depends on some kind of similarity of structure with some object of the real world. Nevertheless, depending on the context one aspect or another of modelling can emerge as more prominent, and to some extent this aligns with what I have tried to emphasize here in the distinction between the replica and instrumental roles of laboratory organisms in biomedical research. As I have shown, animal models can function as replica, relying solely on a similarity with the target system, but they can also function as instruments or measuring and detecting devices, in which case they are informative in the same way other instruments are. The instrumental role dominates when the model’s function of being for the study of human disease detaches itself from the model’s function of being a model of the target system. The example of the genetically engineered zebrafish shows that the instrumental role is not limited to some exotic examples, and is much more prevalent than one might at first suppose. In fact, as I have argued in the previous section, there are reasons to think that it is becoming increasingly important in biomedical research. Throughout the paper, I have avoided the important philosophical question of whether so-called ‘‘animal models’’ fulfilling an instrumental role ought to be considered models at all, independently of how scientists call them. The hybrid

123

126

P.-L. Germain

nature of these systems, as well as the peculiarity of the way they relate to human biology, raise some legitimate doubts as to whether they are bona fide animal models. A major problem here is that the notion of model is notoriously difficult to define, and the more so in the context of material models such as animal models. Accounts of modelling as affording surrogate reasoning (Suarez 2004; Weisberg 2007) imply the existence of non-surrogative means of research, what we could call direct experimentation, which are particularly difficult to define in biology. The first reason for this difficulty is that (supra-molecular) entities and classes in biology are not natural kinds, which makes it largely arbitrary to say whether two tokens are part of the same class. The second and most important reason is that an experimental system is always different from the scientific object it is meant to teach us about.9 A key problem of xenograft models, for instance, is precisely that the tumour they form is not human, despite the fact that the cancer cells are of human origin. The question then is not how similar, but how different a biological object should be to qualify as a model? Unless we are able to draw a line, we have to acknowledge that all experimental systems serve as material models insofar as they relate to a ‘‘natural’’ or non-experimental setting. Alternatively, one could choose to restrict the notion of ‘‘animal model’’ to models-as-replica, used as surrogates for humans in experimental designs which are roughly the same in other respects. Such a strategy would have two important implications. First, as mentioned earlier the two roles discussed in this paper are not properties of the material systems, nor of their relation to humans, but of the way they are used. The same xenografts models of cancer can be used both as replica, or as instruments, depending on the precise experimental context. No system, then, is per se an animal model, and a given system could, in the course of a single project aimed at understanding human pathologies, be intermittently an animal model or not. The second implication of such a characterization of animal models would be that it encompasses only a proportion (and as I argued, a decreasing proportion) of the organisms that are used in biomedical research. This would require a reassessment of the way several questions have been framed, starting with the discussion of the cost-effectiveness of animal experimentation. Acknowledgments In addition to the participants of the Second European Advanced Seminar in the Philosophy of the Life Sciences (EASPLS 2012), whose contributions are partly found in the present issue, I would like to thank the participants of the workshop ‘‘Animal Models, Model Animals? Meanings and Practices in the Biomedical Sciences’’ (Centre for the History of Science, Technology and Medicine, of the University of Manchester, 2012) at which this paper was also presented. I also wish to thank Cristina Santoriello for her precious tutoring, and Marina Mione for welcoming me in her lab. Finally, I am grateful to Mae¨l Lemoine, Jean Gayon, Giuseppe Testa and my FOLSATEC colleagues for interesting discussions on these topics.

9

‘‘The experimental conditions ‘contain’ the scientific objects in the double sense of this expression: they embed them, and through that very embracement, they restrict and constrain them.’’ (Rheinberger 1997, p. 29).

123

From replica to instruments

127

References Ankeny, R. A., & Leonelli, S. (2011). What’s so special about model organisms? Studies In History and Philosophy of Science Part A, 42(2), 313–323. Blasimme, A., Maugeri, P., & Germain, P.-L. (2013). What mechanisms can’t do: Explanatory frameworks and the function of the p53 gene in molecular oncology. Studies in History and Philosophy of Biological and Biomedical Sciences, 44(3), 374–384. Bolker, J. A. (2009). Exemplary and surrogate models: Two modes of representation in biology. Perspectives in Biology and Medicine, 52(4), 485–499. doi:10.1353/pbm.0.0125. Committee on Models for Biomedical Research. (1985). Models for Biomedical Research: A New Perspective. Washington: National Academy Press. Committee on New and Emerging Models in Biomedical and Behavioral Research, Institute for Laboratory Animal Research. (1998). Biomedical models and resources: Current needs and future opportunities. Washington: National Academy Press. Gaudillie`re, J.-P. (2006). ‘‘Produire et utiliser les souris inbred: complexe biome´dical, cancer et obe´site´ aux E´tats-Unis d’Ame´rique apre`s 1945’’, dans Les Organismes Mode`les dans la Recherche Me´dicale, sous la direction de Gachelin G., Presses Universitaires de France, pp. 163–180. Gayon, J. (2006). ‘‘Les organismes mode`les en biologie et en me´decine’’, dans Les Organismes Mode`les dans la Recherche Me´dicale, sous la direction de Gachelin G., Presses Universitaires de France, pp. 9–44. Germain, P.-L. (2014). Living instruments and theoretical terms. In M. C. Galavotti, S. Hartmann, M. Weber, W. Gonzalez, D. Dieks, & T. Uebel (Eds.), New Directions in the Philosophy of Science. Berlin: Springer. Goodman, N. (1968). Languages of art: An approach to a theory of symbols (2nd ed.). Indianapolis, IN: Bobbs-Merrill. Greene, H. S. N. (1948). Identification of malignant tissues. JAMA, the Journal of the American Medical Association, 137(16), 1364–1366. Keller, E. F. (2000). Models of and models for: Theory and practice in contemporary biology. Philosophy of Science, 67, S72–S86. Keller, E. F. (2002). Making sense of life: explaining biological development with models, metaphors, and machines. Cambridge, MA: Harvard University Press. Keuck, L. K. (2012). Relevant similarity in the light of biomedical experimentation. In K. Hagen, F. Schnieke, Angelika, & Thiele (Eds.), Large animals as biomedical models: Ethical, societal, legal and biological aspects (pp. 69–83). Bad Neuenahr-Ahrweiler: Europa¨ische Akademie. Knight, A. (2011) The costs and benefits of animal experiments. Basingstoke: Palgrave Macmillan. Kraus, M. H., Yuasa, Y., & Aaronson, S. A. (1984). A position 12-activated H-ras oncogene in all HS578T mammary carcinosarcoma cells but not normal mammary cells of the same patient. Proceedings of the National Academy Sciences, 81(17), 5384–5388. LaFollette, H., & Shanks, N. (1996). Brute science. Dilemmas of animal experimentation. London: Routledge. Landecker, H. (2007). Culturing life: How cells became technologies. Cambridge: Harvard University Press. Maugeri, P., & Blasimme, A. (2011). Humanised models of cancer in molecular medicine: The experimental control of disanalogy. History and Philosophy of the Life Sciences, 33, 603–622. Meunier, R. (2011). Thick and thin characters: Organismal form and representational practice in embryology and genetics. Ph.D. thesis, Universita` degli Studi di Milano. Meunier, R. (2012). Stages in the development of a model organism as a platform for mechanistic models in developmental biology: Zebrafish, 1970–2000. Studies in History and Philosophy of Biological and Biomedical Sciences, 43(2), 522–531. Morange, M. (1993). The discovery of cellular oncogenes. History and Philosophy of the Life Sciences, 13(1), 45–58. Olszynko-Gryn, J. (2013). The demand for pregnancy testing: The Aschheim–Zondek reaction, diagnostic versatility, and laboratory services in 1930s Britain. Studies in History and Philosophy of Biological and Biomedical Sciences. doi:10.1016/j.shpsc.2013.12.002. Piotrowska, M. (2012). From humanized mice to human disease: Guiding extrapolation from model to target. Biology and Philosophy, 28(3), 439–455.

123

128

P.-L. Germain

Quintana, E., Shackleton, M., Sabel, M. S., Fullen, D. R., Johnson, T. M., & Morrison, S. J. (2008). Efficient tumour formation by single human melanoma cells. Nature, 456(7222), 593–598. Rader, K. (2004). Making mice: Standardizing animals for American biomedical research, 1900–1955. Princeton: Princeton University Press. Rheinberger, H.-J. (1997). Toward a history of epistemic things: Synthesizing proteins in the test tube. Stanford: Stanford University Press. Santoriello, C., Deflorian, G., Pezzimenti, F., Kawakami, K., Lanfrancone, L., d’Adda di Fagagna, F., et al. (2009). Expression of H-RASV12 in a zebrafish model of Costello syndrome causes cellular senescence in adult proliferating cells. Disease Models & Mechanisms, 2(1–2), 56–67. Santoriello, C., Gennaro, E., Anelli, V., Distel, M., Kelly, A., Ko¨ster, R. W., Hurlstone, A., Mione, M. (2010). Kita driven expression of oncogenic HRAS leads to early onset and highly penetrant melanoma in zebrafish. PloS One, 5(12), e15170. Schatton, T., Murphy, G. F., Frank, N. Y., Yamaura, K., Waaga-Gasser, A. M., Gasser, M., et al. (2008). Identification of cells initiating human melanomas. Nature, 451(7176), 345–349. Shanks, N., Greek, R., & Greek, J. (2009). Are animal models predictive for humans? Philosophy, Ethics, and Humanities in Medicine, 4, 2. Shelley, C. (2010). Why test animals to treat humans? On the validity of animal models. Studies in History and Philosophy of Biological and Biomedical Sciences, 41(3), 292–299. Steel, D. P. (2008). Across the boundaries: Extrapolation in biology and social science. New York: Cambridge University Press. Suarez, M. (2004). An inferential conception of scientific representation. Philosophy of Science, 71(5), 767–779. Valent, P., Eaves, C., Bonnet, D., De Maria, R., Lapidot, T., Copland, M., et al. (2012). Cancer stem cell definitions and terminology: The devil is in the details. Nature Reviews Cancer, 12(11), 767–775. Visvader, J., & Lindeman, G. (2012). Cancer stem cells: Current status and evolving complexities. Cell Stem Cell, 10(6), 717–728. Weber, M. (2005). The philosophy of experimental biology. Cambridge, MA: Cambridge University Press. Weisberg, M. (2007). Who is a modeler? The British Journal for the Philosophy of Science, 58(2), 207–233. White, R., Rose, K., & Zon, L. (2013). Zebrafish cancer: The state of the art and the path forward. Nature Reviews Cancer, 13(9), 624–636. Zondek, B. (1928). Die Schwangerschaftsdiagnose aus dem Harn durch Nachweis des Hypophysenvorderlappenhormons. Die Naturwissenschaften, 51, 1088–1090.

123

From replica to instruments: animal models in biomedical research.

The ways in which other animal species can be informative about human biology are not exhausted by the traditional picture of the animal model. In thi...
278KB Sizes 1 Downloads 8 Views