On the relationship between structure and function in cortical networks - with Michael Reimann - #46

Detailed reconstructed neuronal morphologies are now widely available, but how do you wire them up in a network? The shape of the neurons provides a key constraint: synapses can only occur when presynaptic axon terminals are physically close to postsynaptic dendrites. Today's guest has used this and other principles to construct large-scale cortical model networks and to study the relationship between the anatomical connectivity and function of such networks. 

---

For neurons to function, the appropriate ion-channel proteins must be present where they are needed, including in distal dendrites and axon terminals. This requires energy, and the guest's group has developed a mechanistic mathematical model to investigate how neurons can reduce this cost by optimizing where proteins are synthesized within the cell—locally or at more distant sites. The model's predictions agree with experimental findings, suggesting that energy optimization is a fundamental operating principle of neurons.

The prominent and colorful neuroscientist Valentino Braitenberg was born 100 years ago. He co-founded the Max Planck Institute of Biological Cybernetics in Tübingen in Germany, where he made seminal contributions to neuroanatomy, synthetic psychology, and theories for cerebellar, fly vision and cortical function. He was celebrated at the recent Braitenberg*100 symposium which I attended together with today's guest. Ad Aertsen is an outstanding computational neuroscientist and worked with Braitenberg back in the days.

A bursting neuron can maintain its firing-pattern identity throughout an animal's life, even though the ion-channel proteins underlying this identity are turned over on the timescale of days.   Today's guest has proposed that neuronal identities are stored in the specific protein production rules, which are regulated by intracellular calcium signaling.   And how can animals reliably perform a learned task for weeks, even when the underlying neural representation drifts over time, so-called representational drift?

Most neural network models till date have assumed all neurons to be identical, or at least that all neurons within a population are identical. In reality, no two neurons are completely the same. Is this due to unavoidable "biological noise" that the nervous system has to cope with, or can it be a useful feature included by design? The guest co-wrote the recent paper "How heterogeneity shapes dynamics and computation in the brain" addressing this question.

Fruit flies need a short-term (working) memory to keep their direction when they navigate their way to the fruit by smelling. Mean-field ring models was theoretically suggested to encode stimulus orientations 30 years and was observed in fruit-fly compass neurons 10 years ago. But how does odor input come into the picture to set the compass course?   The group of the guest has studied the question with a host of different experimental and theoretical methods.

Starting with the work of pioneers like Wilson and Cowan in the 1970s, mean‑field models have become a dominant tool for modeling neural activity at the level of neuronal populations. Despite their popularity, most mean‑field models have been heuristic and not systematically derived from the underlying 'microscopic' dynamics of individual neurons. Today's guest has made important contributions towards remedying this situation.  

While some models aim to explain qualitative features of brain activity, other aim to reproduce experimental data quantitatively. If so, model parameters must be adjusted to make the model predictions fit the experimental data. A complication is that in most neurobiological applications, there is not a unique best fit: many parameter combinations give equally good model fits. Recently, the guest, together with colleagues, made the tool AutoMIND to fit spiking network models to data.  

Reproducibility is key for scientific progress. If research results cannot be reproduced and trusted, other researchers cannot build on them. Reproducibility is a challenge also in computational neuroscience, and today's guest has worked on how this can be remedied, for example, through standardized model description and model sharing. He also recently organised a workshop celebrating a decade with the (reproducible) Potjans-Diesmann neural network model, which has become an important community tool.