Learning induces persistent chromatin loops underlying robust gene expression during memory recall

Long-term memories are stored in neuronal ensembles called engrams, but the existence of persistent molecular traces in nuclei of engram neurons remain unknown. Using activity-dependent nuclear tagging in vivo, we profiled higher-order chromatin folding and DNA methylation in thousands of single hippocampal neurons up to a month after contextual fear conditioning (CFC). We find CFC-induced chromatin loop plasticity genome-wide, including persistently gained and lost loops with enduring traces in vivo. DNA methylation showed minimal CFC-induced persistence at promoters and enhancers. Persistently gained and lost loops connect distinct enhancers and promoters in excitatory and inhibitory subtypes and correlate with robust CFC-upregulated and -downregulated gene expression, respectively, upon recall. Synaptic genes associated with post-traumatic stress disorder and autism anchor neuronal subtype-specific persistent loops, suggesting relevance to neuropsychiatric dysfunction. We harness the power and sensitivity of multi-modal single-cell measurements to find enduring chromatin traces linked to robust gene expression during fear memory recall. Structured AbstractO_ST_ABSIntroductionC_ST_ABSLong-term memory is thought to be stored in specific neural circuits called engrams (1). Mechanisms centered at the synapse have been proposed, but what is altered within each engram cell that persists for the duration of the memory is unknown. Neural circuits fire action potentials in response to experiences. Such electrochemical signals are converted to molecular signaling pathways which travel from the synapse to the nucleus to activate new gene expression programs. However, experience-dependent RNA and protein molecules as well as synaptic plasticity phenomena are short-lived, lasting for only a few hours to several days (2). Thus, the extent to which chromatin and gene expression changes induced by behavior persist and functionally contribute to memory encoding, consolidation, and retrieval remains an important answered question. DNA is folded in the mammalian nucleus into higher-order long-range chromatin looping interactions (3). Loops form mechanistically through extrusion in which cohesin subunits form a ring that shuttles along chromatin and extrudes out the intervening DNA until it stalls at boundaries occupied by architectural proteins such as CTCF (4, 5). CTCF-independent looping mechanisms have also been uncovered (6-10). A subset of loops regulates gene expression by bringing distant enhancers into spatial proximity with their target promoters. Loops connect activity-dependent enhancers to their distal target genes to govern gene expression during in vitro neural stimulation and in vivo behavior paradigms (11-14). However, the extent to which chromatin changes persist on time scales to support long-term memory storage in vivo remains unclear. RationaleMultiple studies support the idea that higher-order chromatin architecture is plastic and can undergo activity- and experience-dependent remodeling linked to gene expression. Electron microscopy measurements provide direct evidence of chromatin architecture plasticity within minutes, with some changes persisting for up to an hour after potassium chloride stimulation of primary hippocampal neurons (15). Perturbing loops by selectively eliminating either CTCF or cohesin subunits impairs memory encoding in multiple behavior models (11, 14, 16-18), disrupts long-term potentiation (17, 18), and alters dendritic morphology (13, 19, 20). Cohesin-mediated loops are necessary for the establishment of new gene expression programs in post-mitotic neurons, including the upregulation of genes encoding axon guidance, dendritic spine morphology, and synaptic plasticity during neuron maturation in vivo as well as activity-dependent regulation of secondary response genes during neural stimulation in vitro (12, 13). The state of histone modifications before learning influences which neurons are recruited into memory traces, suggesting chromatin carries long-lasting yet adaptable information during memory encoding and long-term storage (21). Moreover, a subset of enhancers retains chromatin accessibility after learning in vivo (11). Together, these observations suggest that chromatin might provide a durable regulatory scaffold that persists long after the initial experience to govern gene expression programs required for the aspects of long-term memory storage. ResultsHere, we employ activity-dependent nuclear labeling in TRAP2 (targeted recombination in active populations) mice to isolate nuclei from neurons stimulated during contextual fear conditioning (CFC) (22, 23). Using single-nucleus methyl 3C-sequencing (snm3C-seq3), we simultaneously profiled DNA methylation and 3D genome folding in the same single tagged neurons from the hippocampus in a time course up to 28 days after CFC and during fear memory recall. Our multi-modal inquiry offered ability to group neurons by their neuronal subtype-specific DNA methylation profile, and thereby computationally generate pseudobulk Chromosome Conformation Capture heatmaps per each sorted hippocampal neuron subtype. We find CFC-induced chromatin loop plasticity genome-wide, including persistently gained and persistently lost loops with enduring structural traces up to 28 days after training. Persistent loops connect distinct distal enhancers to target genes encoding synaptic plasticity and neurotransmitter signaling pathways unique to each excitatory and inhibitory neuron cell type. By contrast to loops, we observe negligible persistence of CFC-induced changes in DNA methylation at enhancers and promoters. Persistently gained and lost loops show upregulation and downregulation of gene expression upon recall, respectively, with the functional effect focused on genes with mRNA levels influenced during training. We uncovered a strong enrichment for genes associated with post-traumatic stress disorder and autism spectrum disorder anchoring the base of persistent loops, suggesting the relevance of enduring chromatin architecture traces to dysregulation of fear, anxiety, and stress in neuropsychiatric disorders. ConclusionOur findings reveal that learning induces neuronal subtype-specific changes in chromatin loops, a subset of which remain durable for at least a month in vivo. Persistent chromatin loops are linked to expression of disease-associated synaptic genes during fear memory recall, thus highlighting relevance to neuropsychiatric and neurodevelopmental disorders. Our work opens up new avenues for investigating enduring chromatin traces for their role in governing cell-wide mRNA levels of genes that impact synaptic plasticity during fear learning and long-term memory storage.