Emotions present us with a puzzle. On the one hand, they seem ideal to study as internal brain processes: Interposed between stimuli and behavior, emotions (together with other processes) should explain how the former cause the latter. And there are beautiful model systems in which to study various emotions, such as fear. A simple dark expanding disk on the ceiling of a mouse’s cage will cause the animal to freeze or flee, behaviors that can be quantified. Research shows that a brain structure called the superior colliculus processes this visual input, converting noisy sensory evidence to a behavioral decision to flee.
On the other hand, we usually think of internal brain processes as an element of cognition, and many people think emotions are not the same as cognition. In fact, it is hard to get any consensus at all on what people mean by ‘emotion.’ The term is used more frequently in psychology than in neuroscience, and more frequently in studies of people than in studies of animals. Indeed, neither of the above cited studies on fear in mice uses the words ‘fear’ or ‘emotion’ in their title or abstract. A primary reason is that emotions are typically conflated with feelings, the conscious experience of an emotion, and we do not currently have good criteria for studying conscious experiences in animals.
The two of us recently took stock of this conundrum and proposed a path forward in a new book, The Neuroscience of Emotion. One of us (David J. Anderson) investigates neural circuits in flies and mice and uses tools such as optogenetics, whereas the other (Ralph Adolphs) studies people, using tools such as functional magnetic resonance imaging (fMRI). Both of us think we are studying emotion, but our methods and criteria could not be more different. The book was our prescription for how to bring this work together and provide the neuroscience of emotion with a common framework — one useful for those working with animals as well as for those working with people. We plan to test some of the ideas put forward in the book as part of a new project funded by the Simons Collaboration on the Global Brain (SCGB). (Other investigators on the project include Markus Meister, Pietro Perona and Doris Tsao.)
We should study animal emotions
A fundamental problem with the study of emotion is that the people who seem to have the loudest claim on ‘really’ studying emotion are the ones who have the worst criteria. Human research on emotion typically relies on verbal reports from research participants, a measure that is difficult to validate and is not generalizable to other species. Not only are verbal reports necessarily about conscious experiences of emotions, but they filter those experiences through the concepts and words we have for describing emotions. The study of human emotion is thus the study of what people think or say about how they feel.
By contrast, studying emotions in a mouse is on much more solid footing. Scientists infer emotional state not from verbal reports but from objectively quantifiable behaviors, such as freezing or fleeing in the dark-disk looming example mentioned above. Oddly, many scientists working on emotion in animals are reluctant to use the word ‘emotion’ to describe what they study, and some even advocate getting rid of the word altogether in animal research. This is problematic for using animal models to investigate antidepressants.
This disconnect between emotion research in people and in animals is unproductive and unnecessary. Fortunately, the solution is remarkably simple: Get rid of conscious experience. No doubt the meaning of ‘emotion’ to nonscientists is about the conscious experience of feeling an emotion. But to most people, the meaning of ‘vision’ is also about the conscious experience of seeing, and the nonscientist’s meaning of ‘memory’ is about conscious recollection. In neither of those cases do neuroscientists seem to have a problem using these terms with a more specific meaning. Most neuroscientists studying vision in animals do not study conscious experiences. Even studies of high level vision in humans need not rely on conscious experience. The same is true for memory. We should do exactly the same thing with emotions: Analyze them without relying on conscious experience.
Once the field takes this step, we can be in agreement with Charles Darwin’s observations about the striking homologies in emotional behaviors across species, even though we may not agree with his particular interpretations of them. But Darwin had a less sophisticated neuroscience, and so it was difficult for him to distinguish homology — a shared ancestral trait — from convergence — traits that evolved independently on a similar trajectory. Moreover, Darwin could say nothing about the mechanisms underlying emotional behaviors. This is another reason to study emotions in animals: Right now, the tools available for circuit-level measures and manipulations of emotion states are providing a wealth of findings but are difficult or unavailable for use in people. Indeed, systems neuroscience is undergoing an explosion in data, with the availability of high-density recording arrays and efforts for data sharing underway. The time is ripe to include emotion in the inventory of central brain states that we need to study — in people and animals.
The new neuroscience of emotion
Getting more concrete about the study of emotion requires us to begin listing specifics. In our book, we spend some time arguing for emotions as a particular type of functional state, with particular features or operating characteristics, such as scalability (the state can be more or less intense), temporal persistence (it lasts for some time) and generalizability (the same emotion state can be triggered by many different stimuli and involves learning).
Although the list of behavioral operating characteristics of emotions is still preliminary, it gives us a necessary purchase on the investigation: We know what to quantify in behavior in order to have evidence for an emotion state. In the looming example discussed above, for instance, we measure freezing behavior, which reflects temporal persistence. Switching from freezing to fleeing to fighting reflects a change in emotion scale. These behavioral features in turn constrain the neural computations that generate the behavior, giving neuroscientists something to look for in the brain. For looming, we look for a neural state that persists for the duration of the freezing behavior.
The looming response is a particularly useful measure because it seems to occur across many species. An ecologically similar visual cue (a dark overhead shadow) also induces a fear-like state in flies, and scientists have quantified scalability, persistence and generalizability in this model. It seems justified to classify the fly’s internal state as an ‘emotion primitive’ given that these properties distinguish the state from a simple reflex.
Stepping back and thinking about emotions as states that are identified based on functional or computational criteria is likely to force us to revise what we mean by ‘emotion’ and almost certain to inspire us to revise the actual emotion categories we currently have. It is unlikely that shame, guilt, pride, awe and love are going to emerge as the right way to classify brain function, even in people. But this should come as no big surprise: None of the words we have for emotions are based on scientific criteria (let alone neuroscience criteria), so why should we expect them to correspond to the ones we will apply scientifically?
Constructing an alternative ontology of emotions is difficult, but in principle it is no different from any other data-driven ontology construction. New categories or dimensions to describe emotions should emerge from knowledge of the mechanisms that generate emotions. Ultimately, types of different emotions are delineated by their similarity. But whereas in the past we attempted to gauge the similarity of two emotions based on how we feel when we consciously experience them, we will now quantify similarity in the kinds of stimuli that induce emotions, the kinds of behaviors caused and the kinds computations that are constitutive of the emotion.
The acid test for a useful framework or approach is that it should generate testable hypotheses. We have sketched out some such experiments and last year won funding from the SCGB to test some of these ideas. Namely, what collaborative experiments could one do across mice and humans?
Optogenetic stimulation of subcortical structures such as the amygdala can modulate fear in mice. Electrical stimulation of the amygdala in people, though necessarily much cruder, also seems capable of inducing fear. Are the evoked brain states in the two species similar? To find out, we need to combine the experimental amygdala activation with recordings. Our initial approach has been to use fMRI, because it provides a whole-brain field of view. Opto-fMRI has been done in mice and monkeys, and we are in the process of setting it up at the California Institute of Technology. We recently developed an approach to do concurrent electrical stimulation with fMRI in people. So we can now begin to ask to what extent the central brain states that are evoked by amygdala stimulation are similar in mice and people. More intense stimulation triggers stronger and additional behaviors, and we can ask study participants to describe these fears. What changes when we activate participants’ brains?
Given that human studies of emotion have historically been entrenched in concepts from psychology, the field is wide open for new discoveries using modern neuroscience tools. How do emotion states cause behavior? What are their dynamics as they are processed through time? How do they generalize over stimuli and context? We have the model systems and tools available to investigate these questions; now we need to jettison the entrenched historical concepts and move forward.
Ralph Adolphs and David J. Anderson, both at the California Institute of Technology in Pasadena, are investigators with the Simons Collaboration on the Global Brain.