Plasticity impairment alters community structure but permits successful pattern separation in a hippocampal network model

Abstract

Patients who suffer from traumatic brain injury (TBI) often complain of learning and memory problems. Their symptoms are principally mediated by the hippocampus and the ability to adapt to stimulus, also known as neural plasticity. Therefore, one plausible injury mechanism is plasticity impairment, which currently lacks comprehensive investigation across TBI research. For these studies, we used a computational network model of the hippocampus that includes the dentate gyrus, CA3, and CA1 with neuron-scale resolution. We simulated mild injury through weakened spike-timing-dependent plasticity (STDP), which modulates synaptic weights according to causal spike timing. In preliminary work, we found functional deficits consisting of decreased firing rate and broadband power in areas CA3 and CA1 after STDP impairment. To address structural changes with these studies, we applied modularity analysis to evaluate how STDP impairment modifies community structure in the hippocampal network. We also studied the emergent function of network-based learning and found that impaired networks could acquire conditioned responses after training, but the magnitude of the response was significantly lower. Furthermore, we examined pattern separation, a prerequisite of learning, by entraining two overlapping patterns. Contrary to our initial hypothesis, impaired networks did not exhibit deficits in pattern separation with either population- or rate-based coding. Collectively, these results demonstrate how a mechanism of injury that operates at the synapse regulates circuit function.

Keywords: cellular networks; memory impairment; patterned structure; plasticity; traumatic brain injury.

FIGURE 1

Modeling STDP impairment in a network model of the hippocampus. (A) The hippocampus consists of several regions connected in a predominantly feedforward topology with information passed from the DG to CA3 to CA1. These three regions are represented in the network model. (B) According to classical STDP, synapses between neurons with causal spikes (positive spike timing) are strengthened, but synapses between neurons with acausal spikes (negative spiking timing) are weakened. With STDP impairment, peak strengthening, or potentiation, is decreased. (C) At baseline, each region has a distinct pattern of firing activity. (D) After STDP impairment, firing rate significantly decreased in areas CA3 and CA1. (E) The power in the theta band, which is important for information processing and hippocampal function, also significantly decreased after injury. *p < 0.01.

FIGURE 2

Modularity methods. (A) Networks can consist of interconnected modules or communities, where similar nodes are grouped with one another. (B) The matrix shows a network representation of community structure where neurons are grouped by module membership. (C) The original empirical matrix is rewired to produce the null matrix, which is a random directed graph with the same input and output degree distributions as the original matrix. The process of community detection maximizes modularity Q to find the optimal community partition. The same parameters are applied to the null matrix and module quality Q is measured for both matrices. Hypothesis testing compares the values of Q between the network of interest and the null model to verify the significance of the identified modular structure. The network is reordered based on community membership. From the reordered matrix, module size and composition can be analyzed. Created with BioRender.com.

FIGURE 3

Hippocampal model networks have significant community structure compared randomized control networks. (A) A representative baseline network organized by anatomical structure (CA3 vs. CA1). (B) A representative network reorganized by module. (C) The number of modules is significantly higher in the randomized networks than at baseline (p < 1e-5). (D) Modularity, Q, is significantly lower for randomized networks (p < 1e-5). Randomized controls rewired connections in the original network while preserving the degree distribution. (E) There was no significant change in modularity over time at baseline. *p < 0.01.

FIGURE 4

Spike-timing-dependent plasticity (STDP) impairment decreases modularity in the CA3-CA1 network. (A) A representative network organized by community assignment shows five modules at baseline. (B) The same representative network has five communities after STDP impairment, but individual node assignments can change resulting in different module size characteristics. (C) Histograms of module size across all 10 networks show that there are more modules at the extreme ends of the size range after STDP impairment. (D) Module quality Q decreased significantly with injury (*p < 0.01). (E) The average number of modules per network did not change after injury. (F) The range of module size increased significantly after injury (*p < 0.01).

FIGURE 5

Module characterization by underlying neuron type reflect hippocampal anatomy. (A) At baseline, one subgroup of modules is comprised primarily of CA3 excitatory neurons (within circle). Predominantly CA1 modules contain most of the inhibitory neurons from both CA3 and CA1. Therefore, there is a significant relationship between the percentage of inhibitory neurons and the percentage of CA1 neurons in these modules (inset) (R² = 0.75; linear hypothesis test; p < 1e-5). As the percentage of inhibitory neurons increases, the percentage of CA1 neurons decreases (inset). (B) After STDP impairment, there remains a subgroup of modules comprised of CA3 excitatory neurons (within circle). However, a new subgroup of small modules develops. These are made up of excitatory neurons from both CA1 and CA3. The appearance of these small excitatory modules eliminates the relationship between inhibitory tone and the percentage of CA1 neurons (inset) (R² = 0.03; linear hypothesis test; p > 0.1). *p < 0.01.

FIGURE 6

Networks successfully encode patterned responses although STDP impairment decreases the signal-to-noise ratio. (A) Training consisted of stimulating sets of 200 neurons in the DG. Baseline networks were trained once with STDP impairment and once under control STDP conditions. Networks were tested before and after training to compare the activity response in each region. (B) Firing rates after training are normalized by the response to stimulation in the untrained network. The gray dashed line is the reference point for activity in untrained baseline networks. The activity of target neurons increases significantly from baseline while the average activity of off-target neurons remains the same or decreases. (C) Networks with STDP impairment exhibit the same paradigm as baseline networks with higher activity in target neurons than in off-target neurons. (D) The signal-to-noise ratio (on-target divided by off-target response) decreases significantly after injury in each region (paired Student’s t-test, p < 0.02 with significance determined by Bonferroni correction).

FIGURE 7

There is no pattern separation deficit in circuits with STDP impairment. (A) Pattern separation occurs when the output patterns differ more than the input patterns do. In this study, we stimulated two patterns with 50% overlap in the population of input neurons. (B) For each region, the output populations consisted of 200 target neurons for each pattern. The percent overlap in baseline networks was below 20% for the DG and CA1. Similar to baseline networks, STDP impaired networks had low percentage overlap in the DG and CA1 with higher overlap in area CA3. (C) The difference between the Hamming distance of the input population and the output population measures pattern separation where a higher value indicates greater pattern separation. With STDP impairment, the distance between output populations was greater in the DG and CA3 than at baseline (paired Student’s t-test, p < 0.02 with significance determined by Bonferroni correction). (D) The rate difference between common neurons shows that common neurons responded preferentially to one pattern or the other. Common target neurons from the DG in one representative network are shown. P1 = pattern 1; P2 = pattern 2. (E) The distance between the rate response to pattern 1 vs. pattern 2 was computed as the Spearman distance. The rate distance for CA3 outputs was significantly different between baseline and STDP impaired networks (paired Student’s t-test, p < 1e-5). *p < 0.01.

FIGURE 8

Target output neurons have low promiscuity among network communities. (A) Both trained and untrained networks with STDP impairment have lower modularity than untrained baseline networks. (B) Neurons that change their community affiliation frequently have high flexibility. If their affiliation shifts between unique communities, those neurons also have high promiscuity. (C) Target neurons are more likely to fall in the first or fifth quintiles of the flexibility distribution. (D) Target neurons have low promiscuity, most likely falling into the first two quintiles of the distribution. *p < 0.01.