Perineuronal Nets on CA2 Pyramidal Cells and Parvalbumin-Expressing Cells Differentially Regulate Hippocampal-Dependent Memory

Abstract

Perineuronal nets (PNNs) are a specialized extracellular matrix that surrounds certain populations of neurons, including (inhibitory) parvalbumin (PV)-expressing interneurons throughout the brain and (excitatory) CA2 pyramidal neurons in hippocampus. PNNs are thought to regulate synaptic plasticity by stabilizing synapses and as such, could regulate learning and memory. Most often, PNN functions are queried using enzymatic degradation with chondroitinase, but that approach does not differentiate PNNs on CA2 neurons from those on adjacent PV cells. To disentangle the specific roles of PNNs on CA2 pyramidal cells and PV neurons in behavior, we generated conditional knock-out mouse strains with the primary protein component of PNNs, aggrecan (Acan), deleted from either CA2 pyramidal cells (Amigo2 Acan KO) or from PV cells (PV Acan KO). Male and female animals of each strain were tested for social, fear, and spatial memory, as well as for reversal learning. We found that Amigo2 Acan KO animals, but not PV Acan KO animals, had impaired social memory and reversal learning. PV Acan KOs, but not Amigo2 Acan KOs, had impaired contextual fear memory. These findings demonstrate independent roles for PNNs on each cell type in regulating hippocampal-dependent memory. We further investigated a potential mechanism of impaired social memory in the Amigo2 Acan KO animals and found reduced input to CA2 from the supramammillary nucleus (SuM), which signals social novelty. Additionally, Amigo2 Acan KOs lacked a social novelty-related local field potential response, suggesting that CA2 PNNs may coordinate functional SuM connections and associated physiological responses to social novelty.

Keywords: extracellular matrix; hippocampus; inhibitory neurons; pyramidal neurons; reversal learning; social memory.

Figure 1.

Cell type-specific Acan deletion results in a lack of Acan immunofluorescence and PNNs, as identified by WFA staining, around CA2 pyramidal cells or PV interneurons. A, Immunostaining for Acan shows the presence of Acan around CA2 pyramidal cells of Amigo2 Cre− tissue but the lack of such staining in Amigo2 Cre+ tissue. B, Cre+ Amigo2 Acan KO animals lack WFA stain on CA2 pyramidal cells (i) but retain WFA stain in PV-expressing cells (ii), whereas Cre− animals show normal WFA stain (i). C, Cre+ PV Acan KO animals lack WFA stain on PV-expressing cells (i) but retain expression on CA2 pyramidal cells (ii), as shown by colocalization with the CA2 marker, PCP4. Scale bars: A, 100 μm; B, 250 μm (i) and 50 μm (ii); and C, 50 μm (i) and 250 μm (ii).

Figure 2.

Amigo2 Acan KOs, but not PV Acan KOs, have impaired social recognition memory. A, In the sociability assay (i), animals can interact with a novel social stimulus within an enclosure or an inanimate empty enclosure, and the amount of time the animal spends within the interaction zone of each is measured. For each of the Amigo2 Acan KO (ii) and PV Acan KO (iii) strains, both Cre− and Cre+ animals showed a preference for interacting with the social stimulus (Amigo2 Acan KO: main effect of chamber: F_(1,46) = 21.86, p < 0.0001, main effect of genotype: F_(1,46) = 0.16, p = 0.70, interaction: F_(1,46) = 0.0013, p = 0.97; multiple comparisons: Cre−: p = 0.0057, Cre+: p = 0.0022; PV Acan KO: chamber: F_(1,51) = 14.43, p = 0.0004, genotype: F_(1,51) = 0.39, p = 0.54, interaction: F_(1,51) = 0.018, p = 0.89; multiple comparisons: Cre−: p = 0.02, Cre+: p = 0.0072), and difference scores (time interacting with novel social stimulus minus time interacting with the empty chamber) were similar between Cre− and Cre+ animals for each strain (Amigo2 Acan KO: t_(33.9) = 0.034, p = 0.97; PV Acan KO: t₍₅₁₎ = 0.14, p = 0.89). B, In the preference for social novelty assay (i), animals have a choice between a novel and familiar mouse, and the amount of time the animal spends within the interaction zone of each is measured. Among Amigo2 Acan KO animals (ii), Cre− animals preferred the novel mouse but Cre+ animals did not (chamber: F_(1,46) = 41.10, p < 0.0001, genotype: F_(1,46) = 4.29, p = 0.044, interaction: F_(1,46) = 11.45, p = 0.0015; multiple comparisons: Cre−: p < 0.0001, Cre+: p = 0.0605), and the difference score (novel minus familiar) was greater for Cre− than that for Cre+ animals (t_(29.2) = 3.21, p = 0.0032). However, both Cre− and Cre+ PV Acan KO animals (iii) preferred the novel mouse (chamber: F_(1,51) = 72.23, p < 0.0001, genotype: F_(1,51) = 0.64, p = 0.43; multiple comparisons: Cre−: p < 0.0001, Cre+: p < 0.0001), and difference score was similar between genotypes (t_(48.8) = 0.73, p = 0.47). C, In the direct interaction test of social recognition memory assay (i), animals were exposed to a novel animal in Trial 1 and the same, now-familiar, animal in Trial 2. For the Amigo2 Acan KO strain (ii), Cre− controls spent less time with the stimulus animal on Trial 2 than Trial 1, but Cre+ KOs spent equivalent time with the stimulus mouse in each trial (trial: F_(1,37) = 40.69, p < 0.0001, genotype: F_(1,37) = 3.22, p = 0.081; interaction: F_(1,37) = 20.69, p < 0.0001; multiple comparisons: Cre−: p < 0.0001, Cre+: p = 0.85), and difference score (Trial 1 minus Trial 2) was significantly greater for Cre− than Cre+ animals (t_(32.5) = 5.58, p < 0.0001). For the PV Acan KO strain (iii), both Cre− controls and Cre+ KOs spent significantly less time with the stimulus animal on Trial 2 than Trial 1 (trial: F_(1,34) = 115.4, p < 0.0001, genotype: F_(1,34) = 0.33, p = 0.57; interaction: F_(1,34) = 0.064, p = 0.80; multiple comparisons: Cre−: p < 0.0001, Cre+: p < 0.0001), and difference scores were similar between genotype (t₍₃₄₎ = 0.25; p = 0.80). Repeated-measures two-way ANOVAs with Bonferroni’s multiple-comparisons tests were used for all grouped comparisons, and results of multiple-comparisons tests are shown on graphs. Two-tailed unpaired t tests, with or without Welch's correction for unequal variance, were used for difference score comparisons, and results are shown on graphs. *p < 0.05, **p < 0.01, ****p < 0.0001. Analyses of data separated by sex are shown in Extended Data Figure 2-1.

Figure 3.

Peak theta frequency increases during novel social stimulus exploration in control animals but not Cre+ Amigo2 Acan KO animals, possibly due to impaired SuM inputs. A, LFP was recorded from CA2 while animals explored a novel context and were then habituated to a new clean cage before introducing a novel animal while recording LFP. One hour later, the same, now-familiar, stimulus animal was reintroduced to the cage and LFP was recorded. B, Power spectral densities (i, ii), peak theta frequency (iii), and area under the curve of 6–12 Hz filtered LFP power spectral density (iv) recorded from CA2 in Cre− (i) and Cre+ (ii) Amigo2 Acan KO animals during investigation in each condition. iii, In Cre− animals, peak theta frequency significantly increased during novel social investigation compared with the novel context and then decreased upon re-exposure to the now-familiar stimulus animal but was still significantly increased compared with novel context exposure. In contrast, peak theta frequency was not significantly increased in Cre+ animals during novel social investigation (genotype: F_(1,8) = 12.77, p = 0.0073; stimulus: F_(2,16) = 18.23, p < 0.0001; interaction: F_(2,16) = 7.53, p = 0.005). iv, Area under the curve of the theta-filtered power spectral density varied significantly across stimulus but did not differ between Cre− and Cre+ animals (genotype: F_(1,8) = 1.35, p = 0.28; stimulus: F_(2,16) = 16.32, p = 0.0001; interaction: F_(2,16) = 1.40, p = 0.28). v, Representative theta-filtered (6–12 Hz) LFP traces recorded during investigation of a novel context (black) or a novel social stimulus (red). C, vGluT2, PCP4 and WFA costaining of CA2 neurons shows vGluT2-positive terminals surrounding PCP4-expressing CA2 neurons, with WFA indicating the presence of PNNs in Cre− animals and the lack of WFA indicating the absence of PNNs around CA2 pyramidal cells of Cre+ animals. The red and green merged CA2 images show vGluT2 and PCP4 only. In the DG images, vGluT2 and DAPI only are shown. Measurements of vGluT2 fluorescent intensity from CA2 and DG reveal significantly less vGluT2 in CA2 of Cre+ Amigo2 Acan KO animals compared with Cre− control animals but no significant difference in DG (genotype: F_(1,15) = 6.29, p = 0.024, subfield: F_(1,15) = 23.39, p = 0.0002, interaction: F_(1,15) = 3.92, p = 0.066). RM two-way ANOVA with Bonferroni’s multiple-comparisons tests used for all comparisons. Results of multiple-comparisons tests shown on graphs. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Scale bars in C represent 25 μm.

Figure 4.

Morris water maze data for Amigo2 Acan KO (A) and PV Acan KO (B) animals. For each strain, data are split into that from the acquisition (top) and reversal (bottom) phases of the assay. For each phase, training data show the latency to reach the hidden platform over each day of training, and probe data show the percent of time that animals spent swimming in the quadrant of the pool where the platform was located during training and the opposite quadrant. Difference scores reflect the time in the target quadrant minus the time in the opposite quadrant. Ai, In the acquisition phase, both Cre− and Cre+ Amigo2 Acan KO animals showed a significant reduction in latency to reach the platform from the first to the last day of training (training day: F_(1,36) = 39.87, p < 0.0001; genotype: F_(1,36) = 3.54, p = 0.07; interaction: F_(1,36) = 0.06, p = 0.81). However, considering all days of training, Cre+ animals showed an overall greater latency to reach the platform than Cre− animals (training day: F_(3,108) = 14.62, p < 0.0001; genotype: F_(1,36) = 5.93, p = 0.02; interaction: F_(3,108) = 0.45, p = 0.72). During the acquisition probe trial, both Cre− and Cre+ animals spent significantly more time in the target quadrant than the opposite quadrant (quadrant: F_(1,36) = 38.79, p < 0.0001, genotype: F_(1,36) = 0.28, p = 0.60; interaction: F_(1,36) = 0.027, p = 0.87). The difference score (time in target quadrant minus time in opposite quadrant) was not significantly different between Cre− and Cre+ animals (t_(35.3) = 0.19, p = 0.85). Aii, In the reversal phase of the assay, both Cre− and Cre+ animals showed a significant reduction in latency to reach the platform from the first day to the last day of training (training day: F_(1,36) = 35.84, p < 0.0001; genotype: F_(1,36) = 0.11, p = 0.74; interaction: F_(1,36) = 4.033, p = 0.052), and no difference between genotypes was detected considering all days of reversal training (training day: F_(7,252) = 9.04, p < 0.0001; genotype: F_(1,36) = 1.77, p = 0.19; interaction: F_(7,252) = 1.15, p = 0.33). However, on the reversal probe trial, whereas Cre− spent significantly more time in the target quadrant than the opposite quadrant, Cre+ animals did not (quadrant: F_(1,36) = 19.61, p < 0.0001, genotype: F_(1,36) = 0.63, p = 0.43; interaction: F_(1,36) = 3.64, p = 0.065), and the difference score was significantly greater for Cre− animals than Cre+ animals (t_(32.8) = 2.06, p = 0.047). Bi, For PV Acan KO animals, in the acquisition phase, no differences were detected in latency to reach the platform from the first day to the last day of training (training day: F_(1,54) = 79.80, p < 0.0001; genotype: F_(1,54) = 0.081, p = 0.78; interaction: F_(1,54) = 0.000045, p = 0.99) or over the entirety of training (training day: F_(5,270) = 31.30, p < 0.0001; genotype: F_(1,54) = 1.12, p = 0.29; interaction: F_(5,270) = 0.76, p = 0.58). During the acquisition probe trial, both Cre− and Cre+ animals spent significantly more time in the target quadrant than the opposite quadrant (quadrant: F_(1,54) = 106.9, p < 0.0001, genotype: F_(1,54) = 0.034, p = 0.85; interaction: F_(1,54) = 0.48, p = 0.49), and the difference score was not significantly different between Cre− and Cre+ animals (t₍₅₄₎ = 0.69, p = 0.49). Bii, In the reversal phase of the assay, both Cre− and Cre+ animals showed a significant reduction in latency to reach the platform from the first day to the last day of training (training day: F_(1,54) = 226.5, p < 0.0001; genotype: F_(1,54) = 0.20, p = 0.66; interaction: F_(1,54) = 0.12, p = 0.74), and no difference between genotypes was detected considering all days of reversal training (training day: F_(6,324) = 61.70, p < 0.0001; genotype: F_(1,54) = 0.55, p = 0.46; interaction: F_(6,324) = 0.71, p = 0.64). On the reversal probe trial, both Cre− and Cre+ animals spent significantly more time in the target quadrant than the opposite quadrant (quadrant: F_(1,54) = 44.52, p < 0.0001; genotype: F_(1,54) = 0.29, p = 0.59; interaction: F_(1,54) = 0.073, p = 0.79), and the difference score was similar between genotypes (t₍₅₄₎ = 0.27, p = 0.79). RM two-way ANOVAs with Bonferroni’s multiple-comparisons tests were used for all comparisons, and multiple-comparisons test results are shown on graphs. Two-tailed unpaired t tests, with or without Welch's correction for unequal variance, were used for difference score comparisons. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Analyses of data separated by sex are shown in Extended Data Figure 4-1.

Figure 5.

Cre+ Amigo2 Acan KOs persisted in using non-spatial search strategies during reversal learning training in the MWM among both males (A) and females (B). Ai, Bi, The search strategies used during each trial for each animal were categorized using Rtrack, a machine learning-based water maze analysis package (Overall et al., 2020; 4 trials per day, 8 d of training) and plotted as the percent of animals using each search strategy for each trial of each day. Data are shown blocked into days and trials within each day. Aii, Bii, The percent of trials in which animals used spatial search strategies on the first day (Day 1) and the last day (Day 8) of reversal training were directly compared. Among males (A) and females (B), Cre− animals showed a significant shift toward using spatial search strategies from Day 1 to Day 8, but Cre+ animals showed no significant change in the percent of trials using spatial search strategies (males: main effect of day: F_(1,17) = 10.31, p = 0.0051, main effect of genotype F_(1,17) = 2.61, p = 0.12; interaction: F_(1,17) = 2.84, p = 0.11; multiple comparisons: Cre−: p = 0.01, Cre+: p = 0.51; females: main effect of day: F_(1,18) = 17.26, p = 0.0006, main effect of genotype F_(1,18) = 0.0022, p = 0.96, interaction: F_(1,18) = 2.28, p = 0.15; multiple comparisons: Cre−: p = 0.005, Cre+: p = 0.08). RM two-way ANOVAs with Bonferroni’s multiple-comparisons tests were used for each sex. *p < 0.05, **p < 0.01.

Figure 6.

PV Acan KOs have impaired contextual fear learning. A, Animals were trained to associate a context and a tone with a shock stimulus on Day 1. On subsequent days, animals were exposed to either the context or the tone in the absence of shock, and amount of time that the animal spent freezing was measured. B, Among Amigo2 Acan KOs, Cre+ animals did not differ from Cre− animals in percent of time freezing in either the context or in response to the cue when tested on Days 2–3 or 16–17 of the assay (Context 1: t₍₄₇₎ = 1.30, p = 0.20; Cue 1: t₍₄₇₎ = 1.47, p = 0.15; Context 2: t₍₄₇₎ = 0.62, p = 0.54; Cue 2: t₍₄₇₎ = 1.19, p = 0.24). C, For the PV Acan KO animals, Cre+ animals showed significantly less freezing in response to the context compared with Cre− controls on each of Day 2 and Day 16 (Context 1: t₍₅₄₎ = 2.22, p = 0.03; Context 2: t₍₅₄₎ = 2.27, p = 0.027). However, no differences in freezing between Cre+ and Cre− PV Acan animals were detected in response to the cue (Cue 1: t₍₅₄₎ = 0.34, p = 0.73, Cue 2: t₍₅₄₎ = 0.64, p = 0.52). Two-tailed unpaired t tests were used for all comparisons. *p < 0.05.