The structure of non-human animal cognitive neuroscience: an epistemological critique

1 Francisco José Correia Graça Almeida The structure of non-human animal cognitive neuroscience: an epistemological critique Estrutura da neurociência cognitiva animal: uma crítica epistemológica março, 2019 2 Francisco José Correia Graça Almeida The structure of non-human animal cognitive neuroscience: an epistemological critique Estrutura da neurociência cognitiva animal: uma crítica epistemológica Mestrado Integrado em Medicina Área: Neurociências Básicas Tipologia:XXXXXXXXXXX Trabalho efetuado sob a Orientação de: Professor Doutor Vasco Miguel Clara Lopes Galhardo Trabalho organizado de acordo com as normas da revista: Reviews in the Neurosciences março, 2019 3 4 5 6 The structure of non-human cognitive neuroscience: an epistemological critique Francisco Almeida1 1 Department of Biomedicine, Faculty of Medicine, University of Porto, Porto, Portugal Corresponding Author: Francisco Almeida Address: Department of Biomedicine FACULDADE DE MEDICINA DA UNIVERSIDADE DO PORTO Al. Prof. Hernâni Monteiro, 4200 - 319 Porto, PORTUGAL e-mail: francisco[email protected]m 7 Abstract Every scientific practice rests on implicit unrevised theoretical assumptions. Neuroscience, in particular, focuses on a very controversial object of studythe brainand is therefore prone to tacitly embrace philosophical positions in its everyday workings. It is thus of the utmost importance to develop a critique of the structure of neuroscientific investigation so as to understand what the uncovered pillars of the field are, what pitfalls they may implicate and how we can correct them. In this paper, I gather the first critiques in animal cognitive neuroscience and hope to establish the first step in a continuous process of revision. By applying a conceptual division of neuroscience into cognitive, behavioural and neurobiological theories, I point out the main problems in articulating the three, based on actual scientific practice rather than purely theoretical reasoning. I conclude by proposing developments on behavioural theory and set an initial critique on assumptions on both cognitive and neurobiological theories. Keywords behaviour; reductionism; cognitive; neurobiology; epistemology 8 Introduction A critical analysis of the fundamental assumptions and workings of a scientific field is necessary to understand its limitations and how we can tackle them. This entails coming out of the everyday working paradigm and question it from an objective standpoint. What do we actually do when we perform an experiment in cognitive neuroscience? What is its meaning and how does it relate to the real world? In essence, advances within the field, be them technological or in understanding, must be accompanied by an epistemological critique to guarantee their validity and usefulness. Cognitive neuroscience is an intersection of cognitive theory, behavioural theory and neurobiological theory. By the first, we mean the psychological processes we think are processed in the brain and are somewhat transversally present in all species of study within cognitive neuroscience. The second refers the range of behaviours specific to each species and what these behaviours mean in terms of their life goals. Lastly, neurobiological theory is the body of knowledge about the biology of neurons and how they interact, from circuits to genes. Each of them has specific problems worthy of a dedicated analysis. Perhaps due to the technical and ethical constraints on human cognitive neuroscience research, analysing the relationships between these three fields has been more common than in non-human studies. For this reason, I will focus in non-human research. Cognitive theory presupposes what has been coined as “cognitive ontologies”(R. A. Poldrack et al., 2016), the cognitive elements at work in the brain, be it fear, memory, etc. Each of them is important for our way of thinking as scientists as well to the hypotheses that can be generated(J. E. LeDoux 2017). Therefore, we must analyse if they provide appropriate concepts and if they have any correlation with what is actually going on in the brain. They also have a relationship with behavioural theory, in as much as behaviours are determined by 9 internal cognitive states. How this relationship is established in practice is an important point for consideration. Moreover, behavioural tasks in the laboratory must be subject to an analysis both in terms of their comparison to natural behaviour, as studied by ethology, and as what concerns their design and which cognitive processes they reflect. Finally, behaviour is correlated with neural evidence, which must itself be analysed in terms of the causal status of interventionist techniques and the statistical treatment of acquired data. In this article I seek to pinpoint the major challenges in articulating these theories. However, rather than studying their interaction from a purely theoretical point view, I will focus on how actual experimental practice interlinks them in the laboratory. Behaviour and meaning in non-human cognitive neuroscience experiments The brain is an ever-changing, dynamic, always active object. For researchers to make sense of it, restrictions have to be made in order to create a space of relevance and significance for the observed behaviours, neural activities and manipulations. This is usually done by confining an animal to a behavioural task, that is, a stereotyped environment where an animal is expected to perform a determined behaviour. This task is designed according to a behavioural theory, a description of what the animal will do in these circumstances. This is then linked to a cognitive theory, that is, the description of which psychological process is in use during this task. While this division of labour between these two fields is clear in theory, we should note that, in experimental practice, we take a certain task to be synonymous with a certain process, e.g., fear learning or reward behaviour. Finally, neurobiological theory joins in by being linked, in accordance with the experimental results, to this chain from behaviour to cognition. Reductionism is a philosophical theory which amounts to assuming that we can reduce psychological constructs to their neural correlates up to the most fundamental level of molecular biology(John Bickle 2006). While this is a matter of intense philosophical discussion, we rarely 16 2018). Previous tasks led to an over-representation of amygdala neurons in encoding valence and representing sensory responses that automatically and immediately caused a behavioural response. Ranges of behaviours may be latent and hidden if not allowed visibility by the task and, therefore, their computation is also hidden. The space of possible behaviours must therefore be extended and well defined in order to have solid mechanistic hypothesis(M. Jazayeri et al., 2017). From this we learn that the behavioural task defines both the behaviour of the animal, its possibilities of action, as well as the logic of our analysis of neural activity. It creates the world for the animal, an artificial one, thereby posing restrictions to what our neurobiological theory may develop. In other words, the design of the task reflects the structure of our behavioural and cognitive theories, which will themselves reflect onto the neurobiological evidence. Naturally, new and contrasting information from neural evidence demands an accommodation in cognitive theory. However, only by designing new tasks will we be able to probe deeper into how the brain works, overcoming the analytical restrictions of a particular behavioural paradigm. Only progressive design of new environments will contribute to understand evermore complex computations. Further, new experimental methods, such as specific reversible manipulations (optogenetics, chemogenetics) and longitudinal imaging of neural activity in freely moving animals (calcium imaging miniscopes) will allow us to correlate and influence neural activity in complex tasks, with multiple behaviours. Developments on techniques must therefore be accompanied by the development of behavioural tasks and theory. Comparing neural activity in a particular area or circuit with different behaviours is consequently a fundamental goal for today. This can be done inside a particular task, but also across tasks, as we discuss next. The very short story of dopamine: a case for battery tests 17 Dopamine was once thought to be the “pleasure molecule”. This was supported, according to Berridge, by the following two reasons: “ 1mesolimbic dopamine systems that are activated by many reward-related stimuli, and 2most so-called pleasure electrodes for deep brain stimulation that supported behavioural selfadministration (i.e., animals or people were willing to work to stimulate the electrodes, such as by pressing a button).” (K. C. Berridge et al., 2015) The deconstruction of this hypothesis came from a decomposition of reward into two putative distinct cognitive ontologies: “wanting” and “liking”. Dopamine depletion does not affect orofacial liking experiments and Parkinson’s disease patients still report normal hedonic feelings (K. C. Berridge et al., 2015). The behavioural observation of this “liking” behaviour allowed a dissection of reward into different components. Studies from reinforcement learning also enabled us to hypothesize the role of dopamine in reward-prediction errors(W. Schultz 2016), rather than in pleasure itself. Thus, what seemingly was inseparable, namely, motivation and hedonism, came to be understood as a rather more complex process. Behavioural testing across different tasks, coupled to new technological methods made it possible. Dopamine has now even even implicated in aversion(C. M. V. Weele et al., 2018). The lesson to draw from here is the fact that for us to understand the mechanisms of the brain, we must see it as working in very different behaviours, with circuit activity overlapping across different cognitive processes and behavioural tasks. Our manipulations may affect a certain behaviour in a certain task, but the reason why it does so remains occult if we do not compare it with other behaviours in other tasks. Thus, battery tests, that is, multiple behavioural tasks should be applied as frequently as possible to study a specific correlation to a certain behaviour. Given that, for 18 pragmatic reasons this is not usually carried out and while it is true that tasks could be compared across studies, it would be important to compare the activity of the same neurons in the same animal across tasks. Indeed, comparing neural activity of the same cells across tasks allows us to study the eventual segregation of neurons due to emotional valence and specificity within a component of a task. Behavioural Data and Computational Opportunities Acquisition and analysis of large amounts of behavioural data has opened the possibility to construct computational models of behaviour which may apply to broader contexts than localized tasks, thereby promising a framework for task-neutral behavioural descriptions. Moreover, computational models of behaviour may provide bridging opportunities with computational models of neural activity, thus conjoining brain and behaviour(A. Gomez-Marin et al., 2014). Nonetheless, pure behavioural data has to be given meaning within a theoretical background. A theory of behavioural universals means the ability to extrapolate behaviour of the same species (perhaps even between species) across different contexts. However, scientific explanation is dedicated to phenomena, not data (J. Bogen et al., 1988) and this is especially true to behaviour. This means that computational models must be oriented by a framework in which a functional-analytic decomposition is at work (low-level information is relevant according to the contribution to a higher-order function). Determining the relevant functions, that is, which descriptions of behaviour are appropriate seems to be context-specific and dependent upon cognitive theory and animal goals. Tasks based on ethological approaches would thus enable an explanation of the computational models of behaviour, which work at the level of description. At this point, it might be interesting to direct the interested reader to a discussion (see (O. Shagrir et al., 2017) ) of Marr’s computational level(D. Marr 2010) and Gibson’s theory of 19 affordances(J. J. Gibson 2014). Briefly, in the first one, a computation performed by the brain receives its meaning from the structure of the physical world. On the other hand, to Gibson, behaviour is always relational and relative to the environment and the opportunities it opens, with the computations not being necessarily a representation of the physical world. From this second point of view, ethological explanations may thus enable behavioural generalizations which will support computational models and consequently the study of mechanistic correlates across evolution. Thus, computational models represent an important opportunity to improve granularity of behavioural descriptions and tighten correlations with brain activity. Convergence of behavioural science, cognitive science and neuroscience will be important to test the possibility of generalized descriptions of behaviour and improve biological realism. Phenomenal decomposition and binary assessments of behaviour Until now, we have seen how behavioural tasks can be designed to accommodate more complex behaviours, and therefore, approach a more natural environment, unveiling in the process more complex cognitive ontologies. We also saw how battery tests may be important to pinpoint the cognitive interpretation of a behavioural alteration and how big behavioural data and computational models may open the avenue for a generalized theory of behaviour. Here we retake our analysis of the interaction between behaviour and cognition. How do we relate a cognitive ontology (motivation, reward, learning, fear, anxiety) with a specific behaviour? How many ontologies are present in a behavioural task and which of them is present at any time? Which one is affected by a manipulation? Phenomenal decomposition is the name given to the division of a certain cognitive process into its components or subtypes(William Bechtel 2008). For instance, memory has been classically divided into various types of memory (short-term, long-term, etc). In our example above, 20 reward was divided into motivation and hedonic pleasure. These subdivisions, which interact with each other, but remain separable, are formed both on psychological grounds and in neural evidence. Behavioural tasks are usually seen as reflecting a very broad cognitive category, e.g. anxiety, without taking into account the processes which might have contributed to the behavioural performance. Let us take an illustrating example. A commonly used behavioural task to measure anxiety is the novelty-supressed feeding paradigm, where a starved animal is presented with food in an anxiogenic environment and the latency to feed is seen as a measure of anxiety itself. Now, this behaviour is based on a conflict between the motivation to eat and the fear of traversing an open space in order to do it. When this task was first designed, the authors wrote the following: “Conflict paradigms rely upon eliciting some predictable response by an animal, inhibiting that response by introducing a response contingent aversive component and overcoming the behavioral effects of the aversive component by drug treatment. It has been important to show that anxiolytics do not alter an animal's sensation of aversion (e.g., they have little or no analgesic action which would make animals insensitive to pain and the analgesic morphine is without anticonflict action). It is equally important to demonstrate that the drug effect is not to enhance the reward value of the goal towards which the behaviour is directed. That is, the drug should not alter responses in the absence of the aversive component.”(D. R. Britton et al., 1981) While the authors underline the necessity of controlling for possible alterations in the reward value by the drug, we should note that, especially when this task has been used widely in depression models(E. J. Nestler et al., 2010), we have to be sure about the intact motivation of 21 the animal to search for food. Tasks which are based on the effects of drugs, which are thought to act upon certain cognitive processes, must also show that the drug is quite specific in influencing the cognitive process (anxiety in this case) without affecting others. Therefore, this task, while assuming a cognitive ontology for itself, may not be reflecting a necessary phenomenal decomposition of behaviour, consequently confounding results. Another important point is how we understand the results of performance in a task. Due to experimental easiness, classically, we usually look at behaviours in a binary way. The animal is either learning or not. It’s motivated or not. It’s anxious or not. While this facilitates our analysis, we see that the behaviour of the animals is dependent upon a gradation. Indeed, research to establish a mechanistic account of gradual accumulation of evidence for decisionmaking in rodents is already underway(C. D. Brody et al., 2016). Moreover, even in classical paradigms we see the necessity of this gradual account. For instance, rats with reduced neurogenesis don’t learn fear conditioned responses in single-trials, but do in multiple trials(M. R. Drew et al., 2010). Rats with chronic pain reinstate an avoidance behaviour after a pain predictive cue when in conflict with small rewards, but approach for larger ones, their motivation depending on the value of the reward(Neil Schwartz et al., 2017). The ethological approach takes into consideration multiple variables to explain a certain behaviour. From the effort of approaching to the value of the reward, all these ontologies must be taken into consideration as a dynamic interplay. Thus, behavioural tasks should also take into account the possible variation of behaviour depending upon, for example, the perceived weight of a stimulus. What seemingly could be attributed to an absolute incapacity may turn out to be a redefinition of the motivation to act or of the limitations of learning in a certain condition. 22 To sum up, detailed analysis of behaviour and its relation to phenomenal decomposition, the complexification of the environment and the comparison of behaviour in battery tests are important challenges for future research and for an epistemological account of neuroscience. Conclusion These considerations are important to expand our knowledge of the mechanisms of the brain and what they mean in terms of psychological processes. Explanation of a neuronal mechanism is relevant as pure knowledge, but its practical use, be it for drugs or for understanding human brains must have meaning in our psychological theory. Preclinical studies in the area of cognitive neuroscience do not predict success well in human trials and a number of reasons regarding scientific methodology have been pointed out(K. K. Tsilidis et al., 2013). In fact, human studies and animal models are separated by a methodological abyss. However, a critique directed at the working epistemology of neuroscience is also fundamental. While attempts have been made at providing a framework to optimize and guide research(A. J. Silva et al., 2013), they remained based on reductionist assumptions. Focusing on behaviour in order to bring it closer to biological realism is an essential step to make translation more plausible. Otherwise, all research will be valid only in its specific laboratorial conditions, wasting resources and animal lives with very low usefulness. Analysis of the structure of both these areas and how they may be bridged represents a challenge and a necessity for the future. It should be noted that the challenges herein analysed do not exhaust the problems. Biological realism as here defined is not a panacea, but only the definition of one possible path. Other areas are worth a critique, for example, the analysis of causality in neuroscience and animal models of human diseases. This paper attempts to lay down first steps in a critique which must be made by neuroscience and philosophy in conjunction. Since it is based on concepts derived from experimental 23 practice, it is limited to an historical and example-based construction. This means that further analysis of emergent and past practices is needed in conjunction with conceptual and theoretical frameworks from philosophy. Acknowledgments I thank Prof. Vasco Galhardo, Jorge Félix Cardoso and Alexandra Rodrigues for carefully reading and commenting on the manuscript. 24 References 1. Arakawa, H. and Iguchi, Y. (2018). Ethological and multi-behavioral analysis of learning and memory performance in laboratory rodent models. Neurosci Res. 135, 1-12. 2. Bechtel, W. (2008).Mental mechanisms: Philosophical perspectives on cognitive neuroscience (New York, NY: Taylor & Francis Group/Lawrence Erlbaum Associates). 3. Berridge, K.C. and Kringelbach, M.L. (2015). Pleasure systems in the brain. Neuron. 86, 646664. 4. Bickle, J. (2006). Reducing mind to molecular pathways: Explicating the reductionism implicit in current cellular and molecular neuroscience. Synthese. 151, 411-434. 5. Bogen, J. and Woodward, J. (1988). Saving the phenomena. Philosophical Review. 97, 303352. 6. Brette, R. (2019). Is coding a relevant metaphor for the brain? Behav Brain Sci. 1-44. 7. Britton, D.R. and Britton, K.T. (1981). A sensitive open field measure of anxiolytic drug activity. Pharmacol Biochem Behav. 15, 577-582. 8. Brody, C.D. and Hanks, T.D. (2016). Neural underpinnings of the evidence accumulator. Curr Opin Neurobiol. 37, 149-157. 9. Buzsaki, G. and Llinas, R. (2017). Space and time in the brain. Science. 358, 482-485. 10. Craver, C.F. (2009).Explaining the Brain (Oxford: Oxford University Press). 11. Drew, M.R., Denny, C.A. and Hen, R. (2010). Arrest of adult hippocampal neurogenesis in mice impairs singlebut not multiple-trial contextual fear conditioning. Behav Neurosci. 124, 446454. 12. Fetsch, C.R. (2016). The importance of task design and behavioral control for understanding the neural basis of cognitive functions. Curr Opin Neurobiol. 37, 16-22. 13. Friston, K.J., Price, C.J., Fletcher, P., Moore, C., Frackowiak, R.S. and Dolan, R.J. (1996). The trouble with cognitive subtraction. Neuroimage. 4, 97-104. 14. Gibson, J.J. (2014).The ecological approach to visual perception (New York: Psychology Press). 15. Gomez-Marin, A., Paton, J.J., Kampff, A.R., Costa, R.M. and Mainen, Z.F. (2014). Big behavioral data: psychology, ethology and the foundations of neuroscience. Nat Neurosci. 17, 1455-1462. 16. Gouvea, T.S., Monteiro, T., Motiwala, A., Soares, S., Machens, C. and Paton, J.J. (2015). Striatal dynamics explain duration judgments. Elife. 4. 17. Jazayeri, M. and Afraz, A. (2017). Navigating the Neural Space in Search of the Neural Code. Neuron. 93, 1003-1014. 18. Juavinett, A.L., Erlich, J.C. and Churchland, A.K. (2018). Decision-making behaviors: weighing ethology, complexity, and sensorimotor compatibility. Curr Opin Neurobiol. 49, 42-50. 19. Kim, J.J. and Jung, M.W. (2018). Fear paradigms: The times they are a-changin'. Curr Opin Behav Sci. 24, 38-43. 20. Krakauer, J.W., Ghazanfar, A.A., Gomez-Marin, A., MacIver, M.A. and Poeppel, D. (2017). Neuroscience Needs Behavior: Correcting a Reductionist Bias. Neuron. 93, 480-490. 21. Kyriazi, P., Headley, D.B. and Pare, D. (2018). Multi-dimensional Coding by Basolateral Amygdala Neurons. Neuron. 99, 1315-1328.e1315. 22. LeDoux, J.E. (2017). Semantics, Surplus Meaning, and the Science of Fear. Trends Cogn Sci. 21, 303-306. 23. Marr, D. (2010).Vision: A computation investigation into the human representational system and processing of visual information (Cambridge: MIT Press). 24. Mau, W., Sullivan, D.W., Kinsky, N.R., Hasselmo, M.E., Howard, M.W. and Eichenbaum, H. (2018). The Same Hippocampal CA1 Population Simultaneously Codes Temporal Information over Multiple Timescales. Curr Biol. 28, 1499-1508.e1494. 25 25. Mello, G.B., Soares, S. and Paton, J.J. (2015). A scalable population code for time in the striatum. Curr Biol. 25, 1113-1122. 26. Mobbs, D. and Kim, J.J. (2015). Neuroethological studies of fear, anxiety, and risky decisionmaking in rodents and humans. Curr Opin Behav Sci. 5, 8-15. 27. Mobbs, D., Trimmer, P.C., Blumstein, D.T. and Dayan, P. (2018). Foraging for foundations in decision neuroscience: insights from ethology. Nat Rev Neurosci. 28. Nestler, E.J. and Hyman, S.E. (2010). Animal models of neuropsychiatric disorders. Nat Neurosci. 13, 1161-1169. 29. Poldrack, R. (2010). Subtraction and Beyond: The Logic of Experimental Designs for Neuroimaging. In: Foundational Issues in Human Brain Mapping. (Cambridge: MIT Press), pp. 147-159. 30. Poldrack, R.A. (2010). Mapping Mental Function to Brain Structure: How Can Cognitive Neuroimaging Succeed? Perspect Psychol Sci. 5, 753-761. 31. Poldrack, R.A. and Yarkoni, T. (2016). From Brain Maps to Cognitive Ontologies: Informatics and the Search for Mental Structure. Annu Rev Psychol. 67, 587-612. 32. Schultz, W. (2016). Dopamine reward prediction-error signalling: a two-component response. Nat Rev Neurosci. 17, 183-195. 33. Schwartz, N., Miller, C. and Fields, H.L. (2017). Cortico-Accumbens Regulation of ApproachAvoidance Behavior Is Modified by Experience and Chronic Pain. Cell Reports. 19, 1522-1531. 34. Shagrir, O. and Bechtel, W. (2017). Marr's Computational Level and Delineating Phenomena. In: Explanation and Integration in Mind and Brain Science. (Oxford: Oxford University Press), pp. 190-214. 35. Silva, A.J., Bickle, J. and Landreth, A. (2013).Engineering the Next Revolution in Neuroscience: The New Science of Experiment Planning (Oxford: Oxford University Press). 36. Soares, S., Atallah, B.V. and Paton, J.J. (2016). Midbrain dopamine neurons control judgment of time. Science. 354, 1273-1277. 37. Tsilidis, K.K., Panagiotou, O.A., Sena, E.S., Aretouli, E., Evangelou, E., Howells, D.W., Al-Shahi Salman, R., Macleod, M.R. and Ioannidis, J.P. (2013). Evaluation of excess significance bias in animal studies of neurological diseases. PLoS Biol. 11, e1001609. 38. Weele, C.M.V., Siciliano, C.A. and Tye, K.M. (2018). Dopamine tunes prefrontal outputs to orchestrate aversive processing. Brain Res.