By Tobias Donner.
These reflections are triggered by the recent NeuroView article Interactionist Neuroscience by David Badre, Michael Frank, and Christopher Moore, published in Neuron in December 2015.
The authors highlight an ongoing revolution in neuroscience: the systematic, structural and functional dissection of neural circuits at a meso-scale in non-human species. This revolution has been afforded by a battery of new technical and genetic tools that are particular well suited for (but not restricted to) application in rodents. Indeed, this is a very exciting development in neuroscience, which has been praised in many recent articles, and which is receiving major support from funding initiatives such as the US BRAIN Initiative or the EU Human Brain Project. Despite sharing the enthusiasm about this revolution, the authors of this NeuroView article make the convincing case that its potential can only fully be realized when closely integrated with human neuroscience – an approach they call interactionist neuroscience.
The authors lay out three complementary strategies for achieving this goal:
#1. Computational theory: This entails building bridges between biophysically detailed circuit models and algorithmic models of cognitive processes.
#2. Integrated experimentation: This entails explicit alignment of human and animal studies, through systematic experiments conducted in parallel in the different species.
#3. Integrated training for postdocs and graduate students across levels in neuroscience: This entails structural (i.e., supported by funding agencies and implemented within the curricula of graduate schools) training in different labs, working at different levels of analysis.
The authors’ assessment resonates well with our lab’s position. We notice that neglect of animal research is widespread in our own field, human neuroscience (“why bother about all these mechanistic details?”). An analogous ignorance about work in humans may well hold for wide parts of animal neuroscience. These isolationist tendencies become most salient in informal discussions with scientists from the respective fields. But they are also evident in a selective (often close to complete) neglect of the relevant literature of the other field in the reference lists of research articles.
We find these tendencies to be uninspiring and obstructive for the advance of neuroscience. We feel that an interactionist approach is one of the most exciting challenges of neuroscience. This challenge brings with it the problem of bridging across measurement scales in the brain. It also brings with it the more fundamental problem of bridging from the functional/computational to the algorithmic to the implementation levels that already David Marr referred to in his Tri-Level Hypothesis. Achieving these types of integration is crucial for understanding how cognitive function emerges from the specific neural circuits that evolution has selected – which, in turn, is an obvious prerequisite for understanding, and ultimately curing, the failures of these circuits that produce disorders of cognition in neurological and psychiatric disorders.
As one humble contribution towards realizing this integration, we regularly read papers from rodent and monkey physiology groups in our journal club (in fact, we almost read more animal than human neuroscience work). As a somewhat bolder contribution, basically all our ongoing and planned experimental projects are driven by hypotheses derived from computational models that bridge across scales, i.e., strategy #1 from above – our research topic decision-making happens to be a showcase for this type of theoretical neuroscience. And, as a third attempt towards increasing integration, we are now starting up systematic collaborations with monkey and rodent physiologists, that are specifically tailored towards strategy #2: Parallel, comparative experiments performed in humans and non-human species.
It is important to note that such parallel experiments do not by themselves guarantee a successful pursuit of strategy #2. Establishing firm links between species and levels of analyses in such experiments is far from straightforward. The challenge starts with designing behavioral tasks that can be learned by animals while still capturing the core cognitive process one would like to dissect in humans. Maximizing the similarity in task design between species, requires compromises on all sides, as well as the willingness to learn about the constraints and interests of the other. Another challenge concerns the selection of measurements. Simply reporting changes in single-unit spiking activity in a region of the monkey brain and an increase in the BOLD-fMRI or EEG signal in an analogous region of the human brain leaves open the fundamental question about the complex relationship between these two different types of signals. A key approach to mastering this challenge is (as the authors of the NeuroView article also allude to) the use of common ‘reference signals’ in animal and human experiments. These reference signals then serve as vehicles for linking the measurements from the different species, which should then exploit the complementary tools and measurement scales.
Neural oscillations form a class of such reference signals. Neural oscillations have remarkably similar functional properties across different spatial scales, from local field potentials to extracranial EEG or MEG signals. Neural oscillation patterns assessed during a single experimental protocol in different species can form a bridge between distant levels of analysis: Conjoint invasive recordings of spiking activity in animals make the link to the activity of individual cells, whereas assessment of large-scale interactions patterns based on non-invasive EEG or MEG recordings in humans make a link to whole-brain network states. Another reference signal that is currently receiving a lot of attention, from us and many other labs, is a marker of ‘brain state’: fluctuations of pupil diameter at constant light levels, which can easily be monitored conjointly with various brain signals, in rodents, monkeys, and humans. Here, invasive approaches in rodents and monkeys play a important role in elucidating the cellular and circuit mechanisms of such changes in brain state, whereas experiments in humans can assess the functional impact of brain state in complex cognitive processes.
Despite our excitement about these developments going on in our and other labs, they are really just first steps towards a fully integrated, interactionist neuroscience. There is need for integrative computational models of cognitive processes, that bridge all the way from microcircuit dynamics to whole-brain networks assessed with neuroimaging, to the algorithmic level and behavior. Models linking microcircuits and large-scale networks (‘circuits of circuits’) have recently begun to be developed for the analysis of ongoing (‘resting-state’) activity in the brain. These types of models would have to be extended for making predictions for task-related brain dynamics. Parallel experimentation with the same task in different species should happen systematically, not only sporadically, in a few sub-projects of a lab’s research program.
The most important, and perhaps most difficult development, will be the systematic implementation of strategy #3: a radical change in neuroscience training, moving from the dependence on single-lab-mentorship towards a structural multi-level, multi-mentors training for both graduate students and postdocs. This implementation goes beyond the effort of individual PIs – entire faculties and funding agencies will have to recognize the importance of the challenge and implement radical changes to meet them. For example, funding agencies could boost this development by making such multi-lab collaboration (and multi-lab training for each student/postdoc funded by the grant) a formal requirement for grant applications. Similar instruments could be installed in graduate schools. Implementing such changes will take time, and it will require influential individuals taking the lead in pushing for such changes.
Hopefully, neuroscience students in 2030 will look at the article by Badre, Frank & Moore wondering how the study of the brain could ever have been so disintegrated.