Synaptic Mechanisms in Long-Term Memory

Memory refers to the ability of an organism to recall the outcomes of sensory experiences. Memories can be classified in terms of time (such as working, short-term and long-term memories) or type (such as habituation, habits, cued and contextual memories, etc). Impressive progress has been made in mapping memory circuits, in particular those involved in fear memory. The concept of engram neurons, first proposed by Richard Semon1, is a generally used framework for memory encoding especially for fear memories2. This concept posits that a limited number of synaptically connected neurons (a ‘circuit’) becomes marked (‘engrammed’) during memory acquisition and that the ‘engram’ stores the memory at least temporally and possibly chronically.

The engram concept has proved highly useful for understanding the mechanism by which contextual and cued inputs produce a short-term memory. However, the engram concept has limitations in that it doesn’t easily explain the need for multiple phases during long-term memory formation (acquisition, consolidation, storage, recall, and extinction), some of which require protein synthesis and may reside in distinct brain regions. A cascade of engrams is possible to explain these phases but would not account for the second limitation of the engram concept, namely the absence of sufficient numbers of neurons in brain to encode separate engrams for each memory item. This second limitation could be resolved by the assumption that a given neuron and circuit is involved in many different engrams. However, then the engram concept becomes indistinguishable from the notion of broadly distributed memory traces that are mediated by subsets of neurons in defined neural circuits, rendering it rather non-specific as a conceptual framework.

Independent of how strictly one applies the engram concept, it seems likely that a memory trace is mediated by synaptic modifications in defined neural circuits, and that these modifications involve changes in gene expression because long-term memory requires protein synthesis. Our laboratory aims to study memory as a synaptic modification and to use studies of circuits and gene expression changes as an entry point for identifying the synaptic mechanisms of memory formation. Thus, we make no attempts to understand and to map memory as a whole but aim to uncover synaptic features of memory formation. Thus, our memory formation research is an extension of our synapse formation studies that are identifying key mechanisms in synapse restructuring.


1. Semon, R. (1922) Die mnemischen Empfindungen in ihren Beziehungen zu den Originalempfindungen. Erste Fortsetzung der ‘Mneme’. Zweite Auflage, Verlag von Wilhelm Engelmann in Leipzig.

2. Josselyn, S.A., Tonegawa, S. (2020) Memory engrams: Recalling the past and imagining the future. Science 367, eaaw4325.

Long-term fear memory: A single-trial adversive memory

The circuits mediating fear memory acquisition and recall via the basolateral (BLA) and central nucleus of the amygdala (CEA) are well characterized, as are the contributions of the hippocampus, nucleus reuniens, zona incerta, superior colliculus, and cortex to fear memory formation, including studies from our own laboratory1-4. After acquisition, short-term memories are consolidated into long-term memories. Memory consolidation is generally divided into cellular consolidation that is thought to be protein synthesis-dependent and systems consolidation that involves hippocampal memory-replay during sleep5-8. Because cellular memory consolidation involves gene expression changes, studies of gene-expression changes may provide an approach to dissect the mechanism of memory consolidation5-8. To gain insight into the synaptic pathways underlying memory consolidation and storage, we performed in recent years single-cell transcriptomics studies in collaboration with the Quake Lab at Stanford on persistent gene expression changes induced by long-term fear-memory formation9,10. The challenge of such experiments rests in the poor signal-to-noise ratio of mapping the transcriptome of individual activated cells. For example, only less than half of cells that are generally measured as being activated by memory formation and/or memory recall (optimistically called ‘engram cells’) are specific for memory formation. Thus, broad gene expression changes are usually observed during single-cell transcriptomics experiments that render interpretations of biological pathways difficult.

To deal with this challenge, we designed our experiments with a focus on long-term persistent gene-expression changes, aiming to exclude effects of sensory inputs or salience and to identify specific transcriptional memory signals9. For this purpose, we crossed TRAP2 mice that express tamoxifen-inducible CreERT2 under control of the Fos promoter with Ai14 mice that express tdTomato in a Cre-dependent manner. These mice were then subjected to four conditions. In the test condition (‘fear recall’), mice were fear-conditioned on day 0, injected with 4-hydroxytamoxifen 30 minutes before returning them to the conditioning chamber for fear memory recall on day 16, and analyzed by single-cell or spatial transcriptomics of the prefrontal cortex and amygdala on day 25, 9 days after memory recall. In the three control conditions, mice were left in the home cage for the entire period but still injected with 4-hydroxytamoxifen on day 16 (‘home cage’) or were treated with the same manipulations as the test condition, except that either the electric shock was omitted during fear conditioning on day 0 (‘no fear’) or that the mice were not placed into the fear-conditioning chamber on day 16 (‘no recall’). This experimental design thus marks cells by tdTomato expression (presumably ‘engram cells’) that are activated during fear recall but measures gene expression changes 3.5 weeks after fear memory formation and 1.5 weeks after memory recall. With this design, our experiments allow definition of persistent, as opposed to transient, gene expression changes9,10.

Our results revealed persistent memory-associated gene expression changes in the prefrontal cortex and the BLA that were observed in subsets of neurons and in astrocytes9,10. In both brain regions, a series of synaptic, signaling, and metabolism genes was induced by memory recall. The gene-expression changes partly overlapped between the two brain regions, although region- and cell-type-specific changes were also detected. In BLA neurons, large changes in neuropeptide expression were observed, suggesting a persistent neuromodulatory mechanism of memory formation. Astrocyte-specific manipulations showed that the gene expression changes in astrocytes were essential for long-term fear memory10. Overall, these experiments established a definable transcriptional program associated with long-term memory that persists long after memory acquisition and consolidation were completed.


1. Xu, W., Morishita, W., Buckmaster, P.S., Pang, Z.P., Malenka, R.C., and Südhof, T.C. (2012) Distinct Neuronal Coding Schemes in Memory Revealed by Selective Erasure of Fast Synchronous Synaptic Transmission. Neuron 73, 990-1001.

2. Xu, W. and Südhof, T.C. (2013) A Neural Circuit for Memory Specificity and Generalization. Science 339, 1290-1195.

3. Polepalli, J., Wu, H., Goswami, D., Halpern, C.H., Südhof, T.C., and Malenka, R. C. (2017) Specification of excitatory synapses on parvalbumin-positive interneurons by neuroligin-3 regulates hippocampal network activity and contextual fear memory extinction. Nature Neurosci. 20, 219-229.

4. Zhou, M., Liu, Z., Melin, M.D., Ng, Y.H., Xu, W., and Südhof, T.C. (2018) A Central Amygdala to Zona Incerta Projection is Required for Acquisition and Remote Recall of Conditioned Fear Memory. Nature Neuroscience 21, 1515-1519.

5. Tonegawa, S., Morrissey, M.D., Kitamura, T. (2018) The role of engram cells in the systems consolidation of memory. Nat. Rev. Neurosci. 19, 485-498. 

6. Ma, H., Khaled, H.G., Wang, X., Mandelberg, N.J., Cohen, S.M., He, X., Tsien, R.W. (2023) Excitation-transcription coupling, neuronal gene expression and synaptic plasticity. Nat. Rev. Neurosci. 24, 672-692.

7. Shrestha, P., Klann, E. (2022) Spatiotemporally resolved protein synthesis as a molecular framework for memory consolidation. Trends Neurosci. 45, 297-311.

8. Redondo, R.L., Morris, R.G. (2011) Making memories last: the synaptic tagging and capture hypothesis. Nat. Rev. Neurosci. 12, 17-30.

9. Chen, M., Jiang, X., Quake, S.R., and Südhof, T.C. (2020) Persistent transcriptional programs are associated with remote memory. Nature 587, 437-442.

10. Sun, W., Liu, Z., Jiang, X., Michelle B. Chen, M.B., Dong, H., Liu, J., Südhof, T.C., Quake, S.R. (2024) Spatial and single-cell transcriptomics reveal neuron-astrocyte interplay in long-term memory. Nature, in press.

Social transmission of food preference (STFP): A single-trial appetitive memory

We have used STFP as a paradigm to elucidate the phases, circuits, and molecular mechanisms underlying an ecologically relevant appetitive memory1,2. If a mouse given a choice between foods with different flavors, for example cinnamon and cocao, it will naturally prefer one of the two (‘innate food preference’; in the case of panel A of the Figure, cocao over cinnamon in our colony). In STFP, a demonstrator mouse that has eaten scented food-

-(usually with an innately non-preferred odor) communicates that odor to a recipient test mouse during a social interaction. When the test mouse is subsequently given the choice between foods that are scented with the innately non-preferred odor to which it was socially exposed vs. the preferred odor to which it was not socially exposed, it will now mainly consume the food with the non-preferred odor (see panel B of the Figure)3,4. The food odor memory lasts for months. Thus, STFP represents a one-trial long-term appetitive memory.  

STFP is known to involve multiple brain regions apart from the olfactory bulb as the primary sensory organ, including the hippocampus since it is a form of contextual memory, as well as the prefrontal and orbitofrontal cortex and basal ganglia5-9.  However, it is unknown how memory formation proceeds from STFP training (the social exposure to the demonstrator mouse) to memory storage and retrieval. We have begun to dissect these stages with the hypothesis that they involve distinctive steps and that at each step, synaptic mechanisms form the basis for memory formation.

In initial studies, we identified a specific form of long-term synaptic potentiation (LTP) at granule cell->mitral synapses in the olfactory bulb (which are part of the granule cell<->mitral cell dendrodendritic synapses) and showed that this form of LTP makes an essential contribution to STFP memory acquisition1. This unusual form of LTP is mediated by Ca2+-dependent secretion of IGF1 induced by neuromodulator-dependent excitation of mitral cells that we had previously discovered and that uses synaptotagmin-10 as a Ca2+-sensor10. These results suggested that STFP memory acquisition involves, among others, synaptic restructuring of the reciprocal mitral cell<->granule cell dendrodendritic microcircuit by an unusual type of synaptic plasticity.

More recently, we discovered that STFP memory acquisition additionally requires another reciprocal synaptic circuit formed by projections from mitral cells4. Specifically, mitral cells in the main olfactory bulb (MOB) project to the accessory olfactory nucleus (AON). AON neurons project back to the MOB by forming synapses on inhibitory granule cells that in turn engage in dendrodendritic synapses with the mitral cells (the MOB mitral cell->AON->MOB granule cell<->mitral cell circuit)4. Interestingly, we found that the AON->granule cell synaptic connections require the presynaptic secretion of C1ql3, a secreted ligand for the postsynaptic adhesion-GPCR Bai3 (gene symbol Adgrb3; see section on synaptic adhesion-complexes). The AON->granule cell synaptic connections also required postsynaptic Bai3 expression, demonstrating that the C1ql3-Bai3 complex is essential for these connections and for STFP memory acquisition4.

In current studies, we are pursuing three goals. First, to dissect the circuit mechanisms of STFP memory formation following memory acquisition, such as memory consolidation, storage, and retrieval, and to relate these circuits to the known involvement of the hippocampus and other brain regions (see diagram in panel C of Figure). Second, to map the gene expression changes associated with different STFP memory formation stages and relate them to those involved in fear memory formation. Third, to elucidate the synaptic mechanisms involved in the various stages of STFP memory formation, with the long-term goal of gaining insight into the memory storage mechanism.


1. Galef, B.G. (2012) A case study in behavioral analysis, synthesis and attention to detail: social learning of food preferences. Behav. Brain Res. 231, 266-271. 

2. Wrenn, C.C. (2004) Social transmission of food preference in mice. Curr. Protoc. Neurosci. Chapter 8: Unit 8.5G.

3. Liu, Z., Chen, Z., Shang, C., Yan, F., Shi, Y., Zhang, J., Qu, B., Han, H., Wang, Y., Li, D., Südhof, T.C., and Cao, P. (2017) IGF1-dependent synaptic plasticity of mitral cells encodes olfactory memory during social learning. Neuron 95, 106-122.

4. Wang, C.Y., Liu, Z., Ng, Y.H., and Südhof, T.C. (2020) A synaptic circuit required for acquisition but not recall of social transmission of food preference. Neuron 107, 144-157.

5. Noguer-Calabús, I., Schäble, S., Kalenscher, T. (2022) Lesions of nucleus accumbens shell abolish socially transmitted food preferences. Eur. J. Neurosci. 56, 5795-5809.

6. Alvarez, P., Wendelken, L., Eichenbaum, H. (2002) Hippocampal formation lesions impair performance in an odor-odor association task independently of spatial context. Neurobiol. Learn. Mem. 78, 470-476.

7. Ross, R.S., Eichenbaum, H. (2006) Dynamics of hippocampal and cortical activation during consolidation of a nonspatial memory. J. Neurosci. 26, 4852-4859. 

8. Carballo-Márquez, A., Vale-Martínez, A., Guillazo-Blanch, G., Martí-Nicolovius, M. (2009) Muscarinic receptor blockade in ventral hippocampus and prelimbic cortex impairs memory for socially transmitted food preference. Hippocampus 19, 446-455.

9. Loureiro, M., Achargui, R., Flakowski, J., Van Zessen, R., Stefanelli, T., Pascoli, V., Lüscher, C. (1019) Social transmission of food safety depends on synaptic plasticity in the prefrontal cortex. Science 364, 991-995.

10. Cao, P., Maximov, A., and Südhof, T.C. (2011) Activity-Dependent IGF-1 Exocytosis is Controlled by the Ca2+-Sensor Synaptotagmin-10. Cell 145, 300-311.