To keep information in mind, you may store it among synapses

The human brain is an amazing organ that holds information in its working memory, which has been a mystery to researchers for decades — a team of neuroscientists at the Massachusetts Institute of Technology (MIT) may have recently unraveled part of that mystery, with a clue that may tell us how the brain holds information in working memory.

An MIT News release noted the study published in PLOS Computational Biology by conducted a team of researchers at The Picower Institute for Learning and Memory. The researchers compared measurements of brain cell activity in an animal performing a working memory task with the output of several computer models that detailed two theories of how the brain stores information.

According to the release, the researchers found the results supported the theory that a network of neurons stores information via temporary changes in the pattern of their connections, known as synapses, going against previous beliefs that memory is maintained by neurons remaining active.

"You need these kinds of mechanisms to give working memory activity the freedom it needs to be flexible,” Earl K. Miller, senior author of the study and Picower Professor of Neuroscience in MIT’s Department of Brain and Cognitive Sciences, said in the news release. “If working memory was just sustained activity alone, it would be as simple as a light switch. But working memory is as complex and dynamic as our thoughts.”

Co-lead author Dr. Leo Kozachkov, who earned his doctorate from MIT last year for his work in theoretical modeling, including his work on this study, noted in the news release that matching the computer models to viable real-world information was key to the study.

“Using artificial neural networks with short-term synaptic plasticity, we show that synaptic activity (instead of neural activity) can be a substrate for working memory,” Kozachkov said in the MIT News release. “The important takeaway from our paper is: These ‘plastic’ neural network models are more brainlike, in a quantitative sense, and also have additional functional benefits in terms of robustness.”

Working with co-lead author John Tauber, an MIT graduate student, Kozachkov said he wanted to determine how working memory information might be held in mind and explain how it actually occurs. He had to start with measurements of the electrical “spiking” actions of neurons located in the prefrontal cortex of an animal as it performs a memory game.

MIT noted in the release that the researchers witnessed what Miller’s lab had seen several times: Neurons repeatedly spike when the original image is seen and only fire intermittently during the gap, spiking again when the image is seen again during the memory game.

Researchers were able to set up software “decoders” to unravel the memory information from the spiking activity. They proved to be very accurate when the spikes peaked but not during the gap period, which suggested that spiking doesn’t represent information during the gap.

Furthermore, researchers proposed that the changes in the strength of the synapses might store information during the gap period. The MIT team tested that theory by modeling neural networks using two versions of the primary theory.

MIT noted that as was done with the animal, researchers trained the machine learning networks to conduct the same working memory skill and to output neural activity that a decoder could interpret. Networks that provided short-term synaptic plasticity to code data spiked when the actual brain spiked and performed the same when it didn’t spike.

According to the MIT release, networks that demonstrated frequent spiking as the way to maintain memory were firing all the time, including when the natural brain was not. Decoder results showed that the accuracy fell during the delay period in the models but often was abnormally high in the frequent spiking models.