Understanding how functional neuronal circuits are established during development is a fundamental challenge in neuroscience. During development, the complex neuronal circuitry of the brain arises from limited information contained in the genome. After the genetic code instructs the birth of neurons, the emergence of brain regions, and the formation of axon tracts, it is believed that neural activity plays a critical role in shaping circuits for behavior. Current Artificial Intelligence technologies are modeled after the same principle: connections in an initial weight matrix are pruned and strengthened by activity-dependent signals until the network can sufficiently generalize a set of inputs into outputs. In Nature Communications, Gregor Schuhknecht, Florian Engert and I challenge these learning-dominated assumptions by showing that neuronal activity contributes minimally, if at all, to the development of a visually-guided swimming behavior in larval zebrafish.
We began by examining the role of visual experience on the maturation of the optomotor response (OMR), a visually-guided behavior zebrafish utilize to maintain their position in a moving stream. Here, we imagined the fish learn important visual features by seeing and interacting with the world. To address this, we dark-reared zebrafish, leaving them no visual experience, and raised fish under a strobe light, providing irregular visual experience. Surprisingly, we found that these environmental modifications had little to no effect on the zebrafish’s ability to develop normal visually-guided behaviors.
Recent findings in neuroscience have suggested that spontaneous neural activity patterns, like retinal waves, might imprint essential visual information into the developing brain. To test if such experience-independent neural activity could underpin visual behaviors, we raised zebrafish under anesthesia. The rationale was straightforward: without any neural activity, spontaneous or otherwise, the conventional mechanisms of spike-time dependent learning would not be possible.
Remarkably, we found that fish raised without structured neural activity were still capable of performing swimming maneuvers and accurately responding to visual cues in our OMR assays. Brain-wide imaging allowed us to confirm that the functional neuronal circuits underlying this behavior came ‘online’ fully tuned and without the requirement for activity-dependent plasticity. These results led to the compelling conclusion that elaborate sensory-guided behaviors can be wired up by activity-independent developmental mechanisms.
In addition to this controversial result, we also found that after shorter periods of silenced activity OMR performance stayed above 90% accuracy, which calls into question the importance of “critical periods” for the formation of visual circuits. Many experiments with developmental perturbations, such as eye sutures, sensory deprivations, and genetic silencing of activity have suggested that periods exist when neuronal activity is required for proper brain development. Yet, these experiments generally relied on regional (and not global) plasticity perturbations, and, crucially, they introduced a competitive imbalance between different modalities or mixed input channels. We hypothesize that the popular “critical periods” can be explained by the fact that perturbed visual inputs, such as those seen in seminal experiments by Hubel and Wiesel, are outcompeted by other modalities, such as motor or auditory signals, which have been demonstrated to contribute to primary visual processing.
From an evolutionary perspective, a robust hardcoding of essential, basic behaviors is expected to be present in all animals. Hardcoding can be seen in turtles heading out to sea after hatching, newborn zebras galloping alongside their herd, and freshly-hatched iguanas escaping from snakes. We acknowledge that, once the animal engages with the world, activity-dependent plasticity is necessary to shape and update neural circuits and to maintain them. Nevertheless, given that complex behaviors can certainly emerge from activity-independent developmental processes, open questions remain as to (1) which behaviors and circuits require neuronal activity for maturation and (2) why activity is utilized for refinement of neuronal circuits. Further understanding and incorporating these neurodevelopmental processes may prove instructive to computational neuroscience and machine learning communities in the future.
In summary, we have developed a reversible developmental block of sensory stimuli and neuronal activity in a vertebrate that, when removed, reveals a functional and appropriately-tuned brain. In this way, we were able to separate “nature” and “nurture” for the first six days of an animal’s life, thereby providing a toolkit for studying the extent of “innate” neuronal wiring in a vertebrate model system. Using this approach, we demonstrate the remarkable degree to which activity-independent developmental mechanisms precisely pattern neuronal circuits.