A longstanding goal of neuroscience is to figure out how the brain reacts to what is happening in the outside world. But it has been challenging to map the complete neural pathways transforming sensory input into appropriate actions. In recent years, researchers have made significant progress towards understanding complete neural circuits in the nervous systems of small invertebrate model organisms, such as the nematode C. elegans and the fruit fly Drosophila. By combining a variety of new technologies, scientists have been able to delineate the processes by which chemosensory (i.e. smell), thermosensory (i.e. heat), mechanosensory (i.e. touch), or visual stimuli generate specific motor outputs in these animals. In each of these cases, the stereotypy and small size of the invertebrate brain, and the identifiability of its neurons, were essential for determining neural circuit architectures and dynamics.
The circuits revealed in invertebrates, however, are unlikely to translate one-to-one to the more complex vertebrate versions like our human brain, which are not only much bigger but also fundamentally different in structure. However, efforts to map the neural activity and connectivity of most vertebrate brains are compromised by the insufficient resolution necessary to decipher the precise nature of the way individual neurons interact. Thus, little is known about how our many (>1010) neurons and their seemingly endless, redundant, and complex neural circuits collaborate to transform sensory signals into behavior.
As a proxy for the human brain, researchers have traditionally focused on measuring neural activity in mammalian vertebrate models. But in rodents, we literally have only patchy knowledge about how the intact brain processes information from the outside world. This is largely due to technical limitations, as most experimental methods are restricted to measuring only small brain regions and a few thousand neurons at a time. To understand how the entire brain deals with its constant flood of incoming information, it is important that neural activity be measured throughout the brain within well-defined behavioral contexts. And it is necessary to establish a strong theoretical framework for interpreting the big data sets that these experiments can produce.
In their study published in this week’s issue of Cell, (PDF), Naumann et al. made use of one of the smallest and translucent vertebrate model organism, the larval zebrafish. They recorded from almost all of its ~100.000 neurons to understand how a visually guided behavior, the optomotor response, is implemented at the brain scale. They chose this behavior, an ancient and conserved reflex in response to moving visual patterns, because it offered an appealing compromise between simplicity (reflex) and complexity (engages thousands of neurons distributed throughout the brain). While previous studies in zebrafish had reported whole-brain activity maps for similar behaviors, Naumann et al. moved beyond these maps to build a functional brain-wide circuit model informed by experiments and analyses at the behavioral, brain, circuit, and neuronal levels.
A critical step in this process was the choice of stimuli Naumann et al. used to dissect the system. Realizing that animals must combine sensations in both eyes to guide behavior, they showed different combinations of monocular and binocular motion patterns, which allowed them tease apart information channels and interactions that had gone unnoticed in previous studies. These new properties included asymmetries in motion processing between the eyes, separation of the behavior into distinct turning and swimming modules, and stabilization of the behavior via reciprocal inhibition. Ultimately, they were able to distill a set of algorithms that related each stimulus to behavior. For example, the optomotor response was driven asymmetrically by visual motion to each eye or turning in the incorrect direction was suppressed strongly by motion moving medially (towards the fish midline). These rules or algorithms became cornerstones of their brain-wide circuit model.
Armed with the detailed description of the behavior and required computations, neural tracing of the links between brain areas processing sensory information, and whole-brain imaging of neural activity, Naumann et al. were ready to build a functional model to explain the behavior. They were confronted, however, with the great heterogeneity and redundancy characteristic of all vertebrate brains. Processing seemed distributed over thousands of cells of many different classes. To make heads or tails of this complexity, they first reduced the neuron population into 16 critical – and bilaterally symmetrical – cell types, based on relative abundance and similarities in stimulus response profiles. But when they then wired these cell types to motor centers based on the revealed circuit architecture, they found it difficult to assign any consistent role to any of the neurons; connections between types appeared to compensate for each other, such that in one model a neuron would have an excitatory connection but in another model it would have an inhibitory one. To find some method in the madness, they instead looked for collective patterns of connectivity across all of the cell types in their models. Specifically, with the critical help of James Fitzgerald and Haim Sompolinsky from Harvard’s Swartz program, they revealed significant covariations in connectivity, suggesting that individual animals could implement the circuit very differently at the level of single neurons, as long as a critical relative relationship between cell types is maintained. By statistically modeling realistic brain circuits based on observed cell types, Fitzgerald and Sompolinsky could show that while the actual connectivity between neurons is neither determined nor important, certain dimensions of connectivity are.
Beyond the optomotor response, Naumann et al. establish a roadmap for digesting large-scale neural data from vertebrate brains and raise fundamental questions broadly relevant to systems neuroscience. This work is particularly timely and appealing in light of the US BRAIN initiative’s concerted push towards whole-brain descriptions of neural function and behavior.