Can a single neuron in the brain really solve complicated problems all by itself?

Ilenna Jones was the lead author on these studies. She is a Neuroscience Ph.D. candidate in Dr. Konrad Kording’s lab at Penn.

or technically,

Might a single neuron solve interesting machine learning problems through successive computations on its dendritic tree? & Do biological constraints impair dendritic computation?

See Original Abstracts on Pubmed: Paper 1 Paper 2

Authors of the studies: Ilenna Simone Jones & Konrad Kording

Figure 1: A Purkinje neuron found exclusively in the cerebellum. Illustration by Ramon y Cajal.

In the late 1800s, a scientist named Ramon y Cajal turned his microscope to the brain and discovered neurons, the cells of the brain. At the time, cameras had not yet been invented, so instead he drew what he saw. He compiled a collection of beautiful illustrations of the many different shapes and variations of neurons, which are still cited and referenced to this day (see Figure 1). In doing so he gave birth to the field of modern neuroscience.

Cajal’s drawings demonstrated the anatomical complexity and variety of neurons throughout the brain. He observed that neurons are composed of several parts, including branched fibers called dendrites that converge onto a cell body, and a single thin fiber that departs the cell body called an axon. Since Cajal’s time, neuroscientists have learned that neurons receive electrical activity from other neurons through their dendrites and send electrical activity through their axons. These electrical signals form the basis of brain activity and allow us to sense, interpret, and respond to cues in our environment.  

Much of neuroscience research has focused on the activity of populations and networks of neurons, but how much can a single neuron do? Does a neuron’s extensive tree of dendrites allow it to perform complex calculations and send new information to other neurons? Or does a neuron simply act like a relay station that transfers the signals it receives without analyzing it? These are the questions that Neuroscience Graduate Group student Ilenna Jones wanted to answer. 

In her first paper, Ilenna used a computerized version of a neuron and asked it to perform various complex tasks. By modifying the number and organization of dendrites on her “virtual neuron,” she found that neurons with complex branching patterns performed tasks better than neurons with simpler branching patterns. This finding suggests that the shape of a neuron actually influences how much it can do! Neurons with densely layered, tree-like dendritic structures can perform sophisticated calculations, as opposed to neurons with more simple dendritic structures which cannot. 

In her second paper, Ilenna next wondered whether making her “virtual neuron” more realistic would change how they performed the same tasks. To do this she included even more of the biological properties found in real neurons, including how dendrites receive and respond to electrical signals from other neurons. She expected that by ‘humanizing’ her virtual neuron it would impair its ability to perform complex calculations, leading to worse task performance. This is a reasonable prediction because in many cases adding more rules for a computer model to follow can push it farther from the ‘idealized case' where it performs very well. But to her surprise, adding these new, realistic characteristics to her neuron actually improved performance in many cases! 

Thanks to Ilenna, we now know that dendritic complexity can allow individual neurons to act as mini-computers that receive information, perform calculations on it, and send new information to many other neurons. Moreover, because neurons come in many shapes and sizes across the brain, it’s likely that different types of neurons can perform completely different calculations depending on their shape. Her findings are significant because it opens up a whole new perspective as to how neurons process information. Understanding what individual neurons are capable of will help neuroscientists study the brain more closely and ultimately help us understand how the brain works!

Want to learn more about the details of Ilenna’s computational modeling of neurons? You can check out the full papers here and here!

About the brief writer: Joe Stucynski

Joe is a graduate student in Dr. Franz Weber’s and Dr. Shinjae Chung’s labs at Penn. He is broadly interested in what makes us sleep how the brain transitions between states.

Citations:

1. Purkinje Neuron Picture: https://upload.wikimedia.org/wikipedia/commons/b/bb/PurkinjeCellCajal.gif