Interregional synaptic maps among engram cells underlie memory formation

A specific memory is thought to be encoded by a sparse population of neurons 1,2 . These neurons can be tagged during learning for subsequent identification 3 and manipulation 4,5,6 . Moreover, their ablation or inactivation results in reduced memory expression, suggesting their necessity in mnemonic processes. However, a critical question of sufficiency remains: can one elicit the behavioral output of a specific memory by directly activating a population of neurons that was active during learning? Here we show that optogenetic reactivation of hippocampal neurons activated during fear conditioning is sufficient to induce freezing behavior. We labeled a population of hippocampal dentate gyrus neurons activated during fear learning with channelrhodopsin-2 (ChR2) 7,8 and later optically reactivated these neurons in a different context. The mice showed increased freezing only upon light stimulation, indicating light-induced fear memory recall. This freezing was not detected in non-fear conditioned mice expressing ChR2 in a similar proportion of cells, nor in fear conditioned mice with cells labeled by EYFP instead of ChR2. Finally, activation of cells labeled in a context not associated with fear did not evoke freezing in mice that were previously fear conditioned in a different context, suggesting that light-induced fear memory recall is context-specific. Together, our findings indicate that activating a sparse but specific ensemble of hippocampal neurons that contribute to a memory engram is sufficient for the recall of that memory. Moreover, our experimental approach offers a general method of mapping cellular populations bearing memory engrams.

Existing protein tagging and detection methods are powerful but have drawbacks. Split protein tags can perturb protein solubility or may not work in living cells. Green fluorescent protein (GFP) fusions can misfold or exhibit altered processing. Fluorogenic biarsenical FLaSH or ReASH substrates overcome many of these limitations but require a polycysteine tag motif, a reducing environment and cell transfection or permeabilization. An ideal protein tag would be genetically encoded, would work both in vivo and in vitro, would provide a sensitive analytical signal and would not require external chemical reagents or substrates. One way to accomplish this might be with a split GFP, but the GFP fragments reported thus far are large and fold poorly, require chemical ligation or fused interacting partners to force their association, or require coexpression or co-refolding to produce detectable folded and fluorescent GFP. We have engineered soluble, self-associating fragments of GFP that can be used to tag and detect either soluble or insoluble proteins in living cells or cell lysates. The split GFP system is simple and does not change fusion protein solubility.

The identification of synaptic partners is challenging in dense nerve bundles, where many processes occupy regions beneath the resolution of conventional light microscopy. To address this difficulty, we have developed GRASP, a system to label membrane contacts and synapses between two cells in living animals. Two complementary fragments of GFP are expressed on different cells, tethered to extracellular domains of transmembrane carrier proteins. When the complementary GFP fragments are fused to ubiquitous transmembrane proteins, GFP fluorescence appears uniformly along membrane contacts between the two cells. When one or both GFP fragments are fused to synaptic transmembrane proteins, GFP fluorescence is tightly localized to synapses. GRASP marks known synaptic contacts in C. elegans, correctly identifies changes in mutants with altered synaptic specificity, and can uncover new information about synaptic locations as confirmed by electron microscopy. GRASP may prove particularly useful for defining connectivity in complex nervous systems.