Neural Networks: So Much for “Hard-Wired”

Actually, it all seems quite simple: The brain commands, and the body responds. It’s no coincidence that we often refer to the brain as the “thinking organ,” equating it with the “little head” or the “skull” – and tap our foreheads when someone isn’t quite right in the head. Yet many everyday observations cannot be explained by the simple model of a “thinking center” from which commands are relayed hierarchically from top to bottom. This principle of “top-down” organization of neural networks is contrasted by the “bottom-up” principle, in which information is first registered by the sensory organs and then – usually through several processing steps – reported “upward.”

The latter is the case, for example, when a bright red poppy in a field catches our Attention. This happens without us having planned to look at this flower and without us having specifically sought it out. In addition to color contrasts, bottom-up processes can also detect movements, unexpected touches, or unusual sounds. It is believed that these processes evolved in many animals to quickly detect predators and escape them. The top-down principle, on the other hand, supports goal-directed behavior – such as when a predator sneaks up on its prey. 

Investigating the underlying principles is complicated – because the neural networks involved seem to have a life of their own. They are neither mere recipients of commands from the brain nor simple reporting systems that send information “upward.”

At the Max Planck Institute for Neurobiology in Martinsried near Munich, Herwig Baier and his research group are studying this phenomenon in zebrafish, one of the most popular model organisms. He focuses on bottom-up processes, whose benefits he explains as follows: “To survive, animals often have to make decisions based on incomplete sensory information. To catch prey or escape predators, zebrafish larvae must be able to focus on a specific object and ignore distracting stimuli.” When the larvae are presented with two objects simultaneously, dynamic neural computations ensure that only one stimulus is selected and their behavior is directed toward it.

Baier and his team are developing a behavioral paradigm to observe this neural decision-making process in free-swimming larvae. One goal is to identify the neural circuits in which salience – that is, the highlighting of individual stimuli – is encoded and which govern the selection of the corresponding behavior. Ultimately, the goal is to uncover the causal relationships between neural activity and visual object recognition.

Among the numerous techniques in the neuroscientists’ lab, two are particularly helpful for tracking and manipulating the Attention of zebrafish larvae: Two-photon microscopy, when combined with behavioral tests, allows researchers to identify the neurons involved in this process. Optogenetics, on the other hand, provides a toolkit for non-invasively disrupting or redirecting attention using light.

Baier and his colleagues have thus identified two neural networks whose activity patterns predict the relative salience of competing visual objects. Depending on the situation, two different strategies appear to be employed: If a stimulus is detected by only one eye, a neural network in the inner Retina makes a decision based on the “all-or-nothing” (winner-take-all) principle.

The decision-making process is significantly more complex when a stimulus is presented to both eyes. Here, the researchers were able to demonstrate that information flows back and forth between a neural node in the hindbrain (Nucleus isthmi) and the roof of the Midbrain (Tectum). Both hemispheres of the brain are involved. The calculation of the salience of both stimuli is now influenced by their relative position to one another. Attention is then directed either solely at one stimulus (winner-take-all), or an average value is calculated and attention is, so to speak, split in two.

Also in Munich, Laura Busse is investigating the principles of top-down regulation. The professor at the Faculty of Biology at Ludwig Maximilian University focuses primarily on mice as her research subjects. She combines electrophysiological measurements in the thalamus and Visual cortex of the animals with optogenetic methods, which allow for the manipulation of neural circuits. She also employs the CRISPR/Cas9 technique, which enables the targeted modification of genetic information.

The focus of her research is on how physical activities – especially movement – affect neural information processing in the early Visual system Busse and her colleagues confirmed that the animals’ visual Perception does not simply mirror their environment on a 1:1 basis, but is highly context-dependent. In this context, not only can different sensory stimuli compete with the visual system for attention, but so can past experiences and goals that the mice have been trained to pursue through the administration of rewards.

The state of the brain at a given moment thus has a fundamental influence on how information is processed in the Visual cortex Specifically, as early as 2014, Busse demonstrated in a highly acclaimed study how the movement of mice in a running wheel caused a sort of “decoupling” of V1 neurons and, furthermore, led to increased activity in a “switching station” of the diencephalon, the dorsal Lateral geniculate nucleus (dLGN). As running speed increased, the animals’ pupils also dilated – an important marker of arousal and cognitive processes. Other research groups have observed a similar phenomenon in human subjects: visual performance improved when participants stood instead of sitting, or when they moved instead of remaining still.

Since then, Busse and her team have uncovered further details and described how a feedback loop between the cortex, the dLGN, and the visual sector of the thalamic reticular Nucleus (visTRN) – which is also located in the thalamus – influences the processing of spatial information in the thalamus. Put simply, the Cortex sharpens Perception in the dLGN and suppresses distracting signals with the help of the visTRN. “All of this can be precisely visualized using the new methods, and you can see how the cerebral cortex effectively modulates its own input,” says Busse.

Although this is purely basic research, unraveling these complex regulatory processes could also contribute to our understanding of human diseases – or at least provide new hypotheses about what might go wrong in mental illnesses such as schizophrenia. “One hypothesis is that changes in the inhibitory activity of certain neural circuits could also play a role here,” explains Busse.

For example, certain optical illusions are not perceived by schizophrenia patients, unlike healthy subjects. These patients are less sensitive to the context in which a stimulus is embedded. Of course, mouse models would not do justice to such a complex disease, but they could provide an explanation for some specific deficits, according to Busse. Certain stimuli are apparently modulated more strongly in healthy individuals: “One could say that patients see the truth, while others perceive a modulated form of reality.” Who would have thought that the question of how neural networks are organized would lead to such astonishing insights?