Identification of a functional connectome for long-term fear memory in mice

Abstract

Long-term memories are thought to depend upon the coordinated activation of a broad network of cortical and subcortical brain regions. However, the distributed nature of this representation has made it challenging to define the neural elements of the memory trace, and lesion and electrophysiological approaches provide only a narrow window into what is appreciated a much more global network. Here we used a global mapping approach to identify networks of brain regions activated following recall of long-term fear memories in mice. Analysis of Fos expression across 84 brain regions allowed us to identify regions that were co-active following memory recall. These analyses revealed that the functional organization of long-term fear memories depends on memory age and is altered in mutant mice that exhibit premature forgetting. Most importantly, these analyses indicate that long-term memory recall engages a network that has a distinct thalamic-hippocampal-cortical signature. This network is concurrently integrated and segregated and therefore has small-world properties, and contains hub-like regions in the prefrontal cortex and thalamus that may play privileged roles in memory expression.

Figure 1. Overview of experimental approach.

Mice were fear conditioned, and fear memory was tested after either a short- or long retention delay. In order to identify neurons activated by memory recall, following testing brain sections were stained for the activity-dependent gene c-fos. Fos expression was subsequently quantified in 84 brain regions, and a complete set of inter-regional correlations computed in order to identify collections of brain regions where Fos expression co-varies across mice. The most robust correlations were then used to generate functional networks for long-term fear memory, and network properties and hubs were characterized using graph theoretical approaches.

Figure 2. Fos is induced by contextual fear memory recall.

A. Experimental design. Mice were trained, and fear memory was assessed either 1 (short delay) or 36 (long delay) days later. Ninety minutes following this test, brains were removed and expression of the activity-regulated gene, c-fos, was evaluated immunohistochemically. B. Percent freezing at the 1 day or 36 day retention test in trained (black bars) or control (open bars) mice. C–D. Task PLS analysis of Fos expression in trained vs. control mice tested 1 or 36 days following training. These analyses identified LVs (left graph) that strongly differentiated the trained vs. control conditions at both the (C) short and (D) long retention delays. Salience scores (right) identify regions that maximally differentiate between these conditions at both the (C) short and (D) long retention delays. The hatched line reflects a salience score of 3, above which the contribution of the regions is considered reliable. At the short delay, Fos expression in the hippocampus contributed strongly to this contrast, whereas at the longer retention delay, Fos expression in multiple brain regions contributed to the contrast.

Figure 3. Generation of long-term fear memory networks.

A. Matrices showing inter-regional correlations for Fos expression at the short (upper) and long (lower) retention delays. Axes are numbered, and correspond to brain regions listed in Table S1. Colors reflect correlation strength (scale, right). B. Network graphs were generated by considering only the strongest correlations (Pearson's r≥0.83). In these graphs, regions are grouped by major brain subdivision and node size is proportional to the number of connections (degree) while the weight of the connection is proportional to correlation strength.

Figure 4. Functional connectivity changes as a function of memory age.

A. Color-coded matrices showing inter-regional correlations for Fos expression within the neocortex at the short (1 day) and long (36 days) retention delays. The inset matrices correspond to inter-regional correlations for subregions of the somatosensory (SS) cortex. Mean inter-regional correlation coefficients were greater at the long delays for (B) cortical regions and (C) somatosensory regions (for this, and other comparisons, see Table S3). D. Color-coded matrices showing inter-regional correlations for Fos expression between regions of the prefrontal cortex and other cortical, thalamic and hippocampal regions at the short (1 day) and long (36 days) retention delays. E. Mean correlation coefficients were greater at the long delay, suggesting that medial prefrontal cortex regions play increasingly important roles in memory expression as a function of memory age. F. Color-coded matrices showing inter-regional correlations for Fos expression between hippocampal and cortical regions at the short (1 day) and long (36 days) retention delays. G. Correlation strength increased over time, suggesting that the hippocampus plays a sustained role in memory expression. In all graphs error bars indicate 95% confidence intervals.

Figure 5. Fear memory networks have small-world structure.

A. Histogram showing degree distribution for fear memory networks corresponding to the 1 day (upper) and 36 day (lower) retention delay. B. Mean clustering coefficient for fear memory vs. random network. At both short (upper) and long (lower) retention delays the fear memory network was more clustered. C. Mean global efficiency for fear memory vs. random network. At both short (upper) and long (lower) retention delays global efficiency (or integration) was equivalent in the fear memory vs. random networks. Error bars represent 95% confidence intervals.

Figure 6. The long-term fear memory network is clustered and resilient.

A. Network summarizing functional connections after memory recall at the long delay. Brain regions were categorized into discrete (color-coded) clusters with similar connectivity to the rest of the network using the Markov Clustering Algorithm. This network contains a densely-interconnected core (green and blue clusters). B. Proportion of brain regions in green, blue and remaining clusters that have previously been implicated in remote memory expression (darkly-shaded portion of bars). Remote memory brain regions are over-represented in the green cluster compared to remaining network (for complete list of regions see Table S5). C. Assortativity for the fear memory vs. random network. Higher assortativity in the fear memory network indicates that highly-connected nodes tend to be connected to one another. Error bars represent 95% confidence interval. D. Consequence of random node deletion on the size of the largest connected component in the fear memory network (grey circles) vs. matched random control network (white circles). In both networks, as nodes were successively removed the size of the largest connected component declined. E. Consequence of targeted node deletion on the size of the largest connected component in the fear memory network (grey circles) vs. matched random control network (white circles). In this simulation, nodes were removed in order of descending degree value. The fear memory network was more resilient to successive deletion of high degree nodes. In both graphs, component size is shown as a proportion of the largest original component.

Figure 7. Identification of hub regions in the long-term memory network.

Brain regions ranked in descending order for (A) degree and (B) betweenness. C. Venn diagram shows the overlap between brain regions ranked above the 80^th percentile for degree and betweenness in each of the primary, high and low confidence networks. D. The three putative hub regions (Cg-a, PrL and Re) that were ranked above the 80^th percentile for degree and betweenness in all three networks were used as seeds in a multi-seed PLS analysis. These analyses identified a LV (upper graph) that strongly differentiated the patterns of correlations between the seed and other brain regions in the trained (closed bars) vs. control (open bars) conditions. The salience scores (lower graph) identify brain regions whose activity correlates strongly with all three seed regions in the trained (but not control) mice. The hatched lines reflect a salience of ±3, above or below which the contribution of the regions is considered reliable.

Figure 8. Fear memory networks are altered in α-CaMKII^+/− mice.

A. Matrices showing inter-regional correlations for Fos expression at the 1 day (upper) and 36 day (lower) retention delays for α-CaMKII^+/− mice. Axes are numbered, and correspond to brain regions listed in Table S1. Colors reflect correlation strength (scale, right). B. Network graphs generated for Pearson's r≥0.83. In these graphs, regions are grouped by major brain subdivision and node size is proportional to the number of connections (degree) while the weight of the connection is proportional to correlation strength. C. Comparison of mean degree (observed connections per node [K]/all possible connections for that node [N]) for α-CaMKII^+/− mice at the short (1 day; closed circles) and long (36 days; open circles) retention delays. The number of connections per node is greater at the 1 day delay for almost all correlation (r) thresholds, reflecting greater network density at the shorter delay. D. Individual correlations differ between WT and α-CaMKII^+/− matrices. These were determined by permutation testing, using a false discovery rate of 5% to account for multiple comparisons. Correlations where WT>α-CaMKII^+/− are shown in white, and where α-CaMKII^+/−>WT are shown in black. Notably, at this short retention delay correlation strength within the somatosensory cortex (outlined in purple) and between hippocampal regions and other brain regions (outlined in green) was stronger in α-CaMKII^+/− mice.