Read on Twitter
1/n Our next preprint is out! “Hebbian plasticity in parallel synaptic pathways: A circuit mechanism for systems memory consolidation”, with Michiel Remme, Urs Bergmann, @denis_alevi, Susanne Schreiber, Richard Kempter ( https://www.biorxiv.org/content/10.1101/2020.12.03.408344v1)
2/n The intricate dynamics of systems memory consolidation has inspired a variety of phenomenological models that emphasize different features to different degrees. But what are the cellular and network mechanisms that mediate the reorganization of memory?
3/n We introduce the Parallel Pathway Theory as a mechanistic basis for systems memory consolidation. We suggest that memory transfer arises naturally from Hebbian plasticity in parallel synaptic pathways — two ubiquitous features in the brain.
4/n The mechanism: Start with a memory stored in an indirect path (blue) between an input and output area. During consolidation, reactivate memory cues in the input area. STDP in a parallel pathway (red) will learn the same association. Such motifs exist e.g. in the hippocampus.
5/n A simple case: We simulated a single spiking CA1 neuron receiving input through Schaeffer collateral (SC) and perforant path (PP-CA1) synapses. SC input was the same as PP-CA1 input, but delayed by 5ms. Consolidation copied a stored weight configuration from SC to PP-CA1.
6/n We show in simulations and analytically that this memory transfer is robust with respect to differing input representation such as place-cell input through SC and grid-cell input through PP-CA1, even when the inputs target different neuronal compartments of neurons in CA1.
7/n The mechanism can be reiterated to consolidate associations beyond PP-CA1. We show that this can explain experiments by @MiguelRemondes, @erin_schuman, who lesioned PP-CA1 at different times during a Morris water maze task ( https://www.nature.com/articles/nature02965).
8/n Parallel pathways are widespread in the brain and could mediate consolidation all the way from hippocampus to cortex. A hierarchical reiteration with exponentially decreasing pathway learning rates supports power law forgetting as in humans and cascade models. @StefanoFusi2
9/n The mechanism was inspired by hippocampal, declarative memory, but it could well apply to any other form of memory! @seanescola
10/n The idea could unify the various phenomenological theories out there by providing a mechanistic basis. We suggest explanations for why certain memories stay hippocampus-dependent and how our theory could provide an inroad into understanding memory transformations.
11/n For more details on what we did, how the theory relates to other theories and some maths, check out the preprint! https://www.biorxiv.org/content/10.1101/2020.12.03.408344v1
