Conditional GSK3β deletion in parvalbumin-expressing interneurons potentiates excitatory synaptic function and learning in adult mice

Abstract

Glycogen synthase kinase 3β (GSK3β) has gained interest regarding its involvement in psychiatric and neurodegenerative disorders. Recently GSK3 inhibitors were highlighted as promising rescuers of cognitive impairments for a gamut of CNS disorders. Growing evidence supports that fast-spiking parvalbumin (PV) interneurons are critical regulators of cortical computation. Albeit, how excitatory receptors on PV interneurons are regulated and how this affects cognitive function remains unknown. To address these questions, we have generated a novel triple-transgenic conditional mouse with GSK3β genetically deleted from PV interneurons. PV-GSK3β^−/− resulted in increased excitability and augmented excitatory synaptic strength in prefrontal PV interneurons. More importantly, these synaptic changes are correlated with accelerated learning with no changes in locomotion and sociability. Our study, for the first time, examined how GSK3β activity affects learning capability via regulation of PV interneurons. This study provides a novel insight into how GSK3β may contribute to disorders afflicted by cognitive deficits.

1. Introduction

Glycogen synthase kinase 3 (GSK3) is a highly conserved serine/threonine kinase containing two isoforms, GSK3α and GSK3β, which are both implicated in major neurological disorders such as schizophrenia, bipolar disorder, and Alzheimer’s disease (Beaulieu et al., 2009; Beurel et al., 2015; Hur and Zhou, 2010; Jope and Roh, 2006; Liang and Chuang, 2007; Peineau et al., 2008a). GSK3 was first discovered to be a critical regulator of glucose metabolism, but has been recognized as a key player in neuronal processes, particularly neurodevelopment and synaptic plasticity (Hur and Zhou, 2010; Peineau et al., 2008a; Peineau et al., 2007). GSK3β is highly enriched in the brain, as well as the synapse, and scientific research has largely focused on this isoform regarding its associations with CNS disorders (Peineau et al., 2008a). Several signaling pathways converge on GSK3β, which subsequently regulates numerous downstream targets, including those involved in cognition (Hur and Zhou, 2010; Liang and Chuang, 2007; Peineau et al., 2008a). Under basal conditions, GSK3β is constitutively active and only becomes inactive by upstream regulators in response to stimuli. Given that GSK3β is constitutively active in resting cells, it has been speculated to provide tonic inhibition over cognitive processes such as long-term potentiation (LTP) (Peineau et al., 2008a).

Emergent evidence suggests that GSK3 inhibitors are strong candidates to not only prevent, but also restore cognitive functions in a wide array of nervous system disorders afflicted with cognitive disabilities such as schizophrenia, autism spectrum disorders, Alzheimer’s disease, Parkinson’s disease, Fragile X syndrome, Down syndrome, stroke, and traumatic brain injury (Beurel et al., 2015). Cognitive impairments in several animal models were ameliorated by the use of a GSK3 inhibitor (King et al., 2014). Although GSK3β has been intimately linked with cognition and CNS disorders, the underlying mechanisms remain to be elucidated. Furthermore, it’s still unclear how GSK3β regulates neuronal signaling in different contexts, including temporal and spatial regulation, as well as cell-type specificity. We recently reported that lithium, a direct GSK3β inhibitor (Klein and Melton, 1996), selectively increased GluN2A total expression that corresponded with increased amplitudes of GluN2A-mediated evoked postsynaptic excitatory currents in the prefrontal cortex (PFC) (Monaco et al., 2018). The N-methyl-d-aspartate receptor (NMDAR) changes were paralleled with an increase in GSK3β serine 9 phosphorylation, collectively suggesting that GSK3β inhibition led to increased GluN2A expression at the post-synaptic density. The maturation (Zhang and Sun, 2011) and maintenance of GABAergic interneurons (Kinney et al., 2006) depends upon NMDARs, particularly the GluN2A subunit (Kinney et al., 2006). Parvalbumin (PV) interneurons have a significantly higher GluN2A/GluN2B ratio in comparison to pyramidal neurons (Xi et al., 2009) with GluN2A antagonism reducing PV and GAD67 immunoreactivity (Kinney et al., 2006). This evidence suggests that GluN2A plays a dominant role in PV interneuronal development and function, and PV interneurons are particularly vulnerable to GluN2A subunit disruption. Given that PV interneurons in the PFC are fundamental regulators of cognitive information processing and their dysregulation are speculated to be the underlying etiology of cognitive deficits in neurological interneuronopathies, we hypothesize that GSK3β is critical for proper PV interneuron function. Herein, we sought to characterize the role of GSK3β as an effector of PV interneuron physiological functioning in the PFC and ultimately the behavioral implications. We generated the first transgenic mouse line with conditional GSK3β knockout in PV interneurons and provided insight on how this mutation augments physiological activity and cognition.

2. Materials and methods

Detailed experimental procedures can be found in the Supplemental Information. Briefly, experimental mice were bred in-house utilizing a combination of three transgenic mouse lines: B6.Cg-Gt(ROSA) 26Sor^{tm14(CAG-tdTomato)Hze}/J, B6;129P2-Pvalb^tm1(cre)Arbr/J (Low et al., 2000) and glycogen synthase kinase3β Flx/Flx mouse (BL6/129 background) (Urs et al., 2012). GSK3β^{floxed/floxed} mice were gifted from Dr. Marc Caron at Duke University (Urs et al., 2012), while the PV-cre (RRID:IMSR_JAX: 017320) and tdTomato (RRID:IMSR_JAX: 007914) mouse strains were purchased from The Jackson Laboratory. The PV-Cre knock-in mice express Cre recombinase in PV-expressing neurons only. Expression of tdTomato and GSK3β genetic deletion (hereinafter referred to as PV-GSK3β KO; PV-GSK3β^−/−) occurs following Cre-mediated recombination. For all experiments, male and female mice were used and combined based on genotype.

2.1. Ex vivo whole-cell electrophysiology

Slices containing the medial PFC (mPFC; 300 μm) were collected. All experiments were conducted with the Axon MultiClamp 700B amplifier, and data were acquired using pCLAMP 9.2 software. Action potential recordings in current clamp were obtained to measure intrinsic membrane properties. To record evoked AMPA- and NMDA-EPSCs we performed whole-cell recordings using a CsCl intracellular solution and voltage clamp to hold the membrane at −60 mV and + 60 mV for AMPA- and NMDA-EPSCs, respectively, in the presence of picrotoxin (50 μM) to block GABAergic neurotransmission. To record evoked excitatory/inhibitory (E/I) balance we performed whole-cell recordings using a CsCl intracellular solution and voltage clamp to hold the membrane at −60 mV and 0 mV for EPSCs and IPSCs, respectively. Recorded pyramidal neurons were located in layer V of the mPFC. For electrophysiological data, Clampfit 9.2 was used for analysis. To isolate NMDA current data, we measured amplitude and tau at a window 15 ms after the EPSC peak amplitude, to exclude any contributions from AMPA.

2.2. Behavioral testing

Mice were acclimated to the behavioral testing suite within the animal facility 30 min prior to all behavioral paradigms in their home cages. Preceding cognitive tests, mice were handled by the experimenter daily for approximately a week. In addition, animals were adjusted to a reverse light-dark schedule and food restricted until mice were at 85% of the ad libitum food weight.

2.3. Data analysis

Data were analyzed using a combination of SPSS Statistics and GraphPad Prism. All data underwent Shapiro-Wilk’s test of normality, Levene’s test for equality of variances, Mauchly’s test of sphericity, and Box’s test of equality of covariance matrices when appropriate. If the assumption of normality was violated for independent-samples t-test, a Mann-Whitney U nonparametric test was used for analysis. If the assumption of sphericity was violated, a Greenhouse-Geisser correction was applied. The ROUT method with Q set at 1% was used to detect and remove outliers from datasets. The independent variables sex, genotype, as well as the interaction between the two were analyzed for each dataset. Male and female mice were combined based on genotype when there was no significant main effect of sex. All data were presented as a mean ± standard error (SEM). Level for significance was set at p ≤ .05 for all comparisons. Note that one asterisk symbolizes p ≤ .05, two asterisks p ≤ .01, three asterisks p ≤ .001, and four asterisks p ≤ .0001.

3. Results

3.1. Validation of a novel conditional mouse model: Parvalbumin-specific GSK3β genetic deletion

GSK3β^flox/flox; PV^+/+; Rosa26^+/lsl-tdTom (control) and GSK3β^flox/flox; PV^+/cre; Rosa26^+/lsl-tdTom (PV-GSK3β KO) mice were generated. Fig. 1A illustrates the two experimental groups produced from the final cross, which combines three transgenic mouse lines. Two breeding lines were separately maintained until the final cross, in which mice were bred for Flox homozygosity. All mice were confirmed with genotyping (Fig. S1) prior to experimental testing.

Fig. 1.

Parvalbumin interneurons from PV-GSK3β−/− mice are labeled with tdTomato and exhibit reduced GSK3β expression in the mPFC. (A) A simplified diagram depicting the genetic differences between control and PV-GSK3β−/− mice. Two breeding crosses were separately maintained until the final cross, producing control and littermate PV-GSK3β KO mice. Mice were selectively crossed for Flox homozygosity and tdTomato heterozygosity. PV-GSK3β−/− mice are only differentiated from controls by one genetic characteristic, expression of PV-dependent Cre recombinase. Genetic deletion of GSK3β and tdTomato expression occurs following Cre-mediated recombination in PV-expressing cells. (B) Representative cells from control and PV-GSK3β KO mice comparing (from left to right) tdTomato, PV, GSK3β, DAPI, and merged immunofluorescence. Scale bars = 10 μm. (C-E) Relative fluorescence units of tdTomato (C), PV (D) and GSK3β (E) expression in a control and KO cell. Data were analyzed either with an unpaired t-test or Mann-Whitney test. (F) A representative image of tdTomato, PV, and merged immunofluorescence. Scale bars = 40 μm (G) A majority of PV + INs overlapped with td-Tomato expression (87.8%), with a small population of PV+ neurons devoid of td-tomato (12.2%). No neurons expressed td-Tomato only, indicating no off-target effects of the Cre recombinase. PV interneurons were the only cells in the PFC to ectopically express the tdTomato fluorescent-reporter protein. (H) Electrophysiological confirmation of tdTomato expression in fast-spiking prefrontal PV interneurons. Medial prefrontal cortex mouse slice preparation containing fluorescent and corresponding differential interference contrast images captured by a CCD camera, as indicated by the arrows. Fluorescent image identifies a tdTomato-expressing PV interneuron (left). DIC images demonstrate localization within the tissue (middle) and successful dimple formation on the representative cell. Scale bar = 20 μm. (I) Representative action potential trace from a fluorescently labeled neuron demonstrates fast-spiking activity, indicative of PV-expressing interneurons. In entirety, 100% of fluorescently recorded neurons demonstrated fast-spiking activity, confirming that Cre-mediated recombination is restricted to PV interneurons. Data shown are mean ± SEM for each group. ** p ≤ .01, **** p ≤ .0001, n.s. = not significant.

A representative neuron from a PV-GSK3β KO mouse expressed tdTomato (yellow) colocalized with PV (green), and GSK3β immunolabeling was absent (red; Fig. 1B, bottom panel). However, a control PV neuron had no tdTomato fluorescence and GSK3β was immunolabeled throughout the soma (Fig. 1B, top panel). PV interneurons from PV-GSK3β KO mice (3.72 ± 0.29 F.U.) demonstrated significantly higher tdTomato expression compared to controls (0.36 ± 0.03 F.U.), U = 0.00, p < .0001; Fig. 1C). There were no significant differences in PV expression between wild-type control and PV-GSK3β KO mice (t(46) = 0.25, p = .801; Fig. 1D). However, PV-GSK3β^−/− mice had significantly reduced GSK3β expression (0.43 ± 0.04 F.U.) compared to control (0.70 ± 0.08 F.U.; t(43) = 2.97, p = .005; Fig. 1E). A majority of PV interneurons overlapped with td-Tomato expression (87.8%). However, no neurons stained for td-Tomato only, suggesting that our transgenic animal model did not exhibit off-target Cre recombinase expression (Fig. 1F, G). Expression of tdTomato was restricted to PV interneurons based on firing patterns, in which 100% of the electrophysiological recordings demonstrated fast-spiking activity (Fig. 1H, I). These data validate that PV+ interneurons from the PV-GSK3β KO mice are labeled with tdTomato and exhibit reduced GSK3β expression.

3.2. PV-GSK3β^−/− increased PV neuronal excitability

GSK3β has recently been highlighted as a promising candidate in several CNS disorders (Beurel, Grieco, 2015), particularly regarding its upstream regulation over cognitive processes (King, Pardo, 2014); however, little is known regarding cell-type specificity. Moreover, the role of GSK3β within PV interneurons, to our knowledge, has not previously been addressed. Considering that GABAergic PV interneurons are fundamental for cortical computation and their disruption is a commonality across a spectrum of CNS disorders associated with cognitive impairments, we sought to elucidate how GSK3β affects this neuronal subtype in particular. First, we investigated whether the intrinsic membrane properties of PV interneurons were changed following PV-GSK3β genetic deletion. We found that GSK3β knockout PV interneurons had a more depolarized resting membrane potential (−67.2 ± 1.08 mV) compared to control (−73.4 ± 1.12 mV), a significant difference of +6.23 mV (95% CI, 3.14 to 9.33, t(81) = 4.01, p < .0001; Fig. 2A, B, C). PV-GSK3β^−/− PV interneurons also exhibited a significant increase in input resistance (171 ± 6.28 MΩ) and tau (6.54 ± 0.20 ms) compared to controls (144 ± 6.20 MΩ; 5.54 ± 0.27 ms), a significant difference of +27.2 MΩ (95% CI, 9.67 to 44.8, t(81) = 3.09, p = .003) and + 1.00 ms (95% CI, 0.35 to 1.65 ms, t(81) = 3.04, p = .003 (Fig. 2D, E). Rheobase for controls (140 ± 8.17 pA) and PV-GSK3β^−/− (119 ± 9.93 pA) was significantly different (U = 554, p = .034; Fig. 2F). PV-GSK3β^−/− PV interneurons also demonstrated increased total spikes across all injected currents (2321 ± 118.1) compared to control (1802 ± 111.7), a significant difference of +519.4 (95% CI, 195.7 to 843.2), t(81) = 3.192, p = .002; Fig. 2G). Mean firing frequency was higher in PV-GSK3β^−/− PV interneurons (119 ± 5.97 Hz) compared to control (92.4 ± 5.97 Hz), a mean difference of +26.9 (95% CI, 10.2 to 43.7, p = .002; Fig. 2H). Action potential threshold, amplitude, duration, and afterhyperpolarization amplitude were not significantly different between genotypes (p ≥ .05, Supplemental Table 1). The changes reported here suggest that GSK3β regulates excitability, and potentially could increase sensitivity to incoming stimuli within PV interneurons. Furthermore, PV-GSK3β^−/− PV interneurons are more likely to hold a charge and summate, as suggested by increased tau.

Fig. 2.

PV-GSK3β−/− mice demonstrated increased excitability and sensitivity to synaptic inputs in mPFC PV interneurons. (A-B) Representative traces of fast-spiking PV interneurons from control (A) and PV-GSK3β−/− (B) mPFC ex vivo slices. The circle indicated where significant genotypic differences were reported; including resting membrane potential (RMP), input resistance (IR), and tau (arrow, inset). A magnified image depicting the changes of RMP, IR, and tau between genotypes (grey, control; black, PV-GSK3β−/−). (C–H) The quantification measurements of PV interneuron intrinsic membrane properties for which genotypic differences were observed. (C) PV-GSK3β−/− PV interneurons demonstrated a more depolarized RMP (control n = 41, PV-GSK3β−/−n = 42). Input resistance (D) and tau (E) were increased in PV-GSK3β−/− PV interneurons (control n = 41, PV-GSK3β−/− n = 42). (F) PV-GSK3β−/− PV interneurons had a lower rheobase (control n = 36, PV-GSK3β−/− n = 42). The total number of spikes (G), as well as the firing frequency with increasing current steps (H), was increased in PV-GSK3β−/− PV interneurons. A two-way mixed ANOVA was run to determine the effects of genotype and the current step on firing frequency. The two-way interaction was not significant [F(12, 960) = 1.26, p = .285]; however there was a significant main effect for injected current on firing frequency [F(12, 960) = 105, p < .0001]. All pairwise comparisons were performed with a post-hoc analysis and a Bonferroni adjustment, adjusted p-values are reported. The between-subjects effect, genotype, was significant. Mean firing frequency was higher in PV-GSK3β−/− PV interneurons (119 ± 5.97) compared to control (92.4 ± 5.97), a mean difference of +26.9 (95% CI, 10.2 to 43.7), p = .002. All other data were analyzed either with an unpaired t-test or Mann-Whitney test. Data are means ± SEM. * p ≤ .05, ** p ≤ .01, **** p ≤ .0001.

3.3. PV-GSK3β^−/− increased synaptic strength of excitatory neurotransmission in PV interneurons

PV interneurons are considerably more sensitive to NMDAR dysregulation compared to other neuronal subtypes (Kinney et al., 2006; Wang and Gao, 2012). Additionally, our previous study demonstrated a connection between GSK3β and NMDARs within the prefrontal cortex (Monaco et al., 2018). Little is known regarding how excitatory receptors, such as AMPA and NMDA, are regulated within interneurons. Considering that disinhibition of PV interneurons results in a loss of spatial tuning, reducing spatial selectivity and therefore computational cortical processing (Rao et al., 2000), we aimed to elucidate whether PV-selective GSK3β knockout affected excitatory synaptic neurotransmission. To address this, we activated presynaptic afferents while simultaneously recording AMPA- and NMDA-mediated excitatory postsynaptic currents (eEPSCs) from postsynaptic tdTomato-labeled PV interneurons.

The NMDA/AMPA ratio was not significantly different (p ≥ .05, Fig. S2A), indicating that the relative expression of NMDA and AMPA were comparable. No differences in AMPA or NMDA paired-pulse ratio was observed (p ≥ .05, Fig. S2B, C), suggesting the probability of synaptic release is unaffected by genotype. However, we found that PV-GSK3β^−/− PV interneurons exhibited a significant increase in AMPA charge for the first pulse (2.92 ± 0.33 pC) as well as total charge, across both pulses (6.12 ± 0.69 pC) compared to control (1.75 ± 0.14 pC, 3.76 ± 0.31 pC, U = 553, p = .017; Fig. 3B; U = 560, p = .021, Fig. 3C). The increased AMPA charge in PV-GSK3β^−/− PV interneurons was likely caused by the augmented AMPA-EPSC amplitude (139 ± 12.5 mV, 153 ± 14.8 mV) compared to control (98.1 ± 9.70 mV, 97.7 ± 8.77 mV, U = 555, p = .012, Fig. 3D; U = 517, p = .006, Fig. 3E], which was not due to changes in stimulation intensity (p ≥ .05, not shown). AMPA decay time was not different between groups (p ≥ .05, Fig. 3F).

Fig. 3.

PV-specific GSK3β deletion augmented AMPA charge and amplitude in mPFC PV interneurons, with no change in decay kinetics. (A) Sample traces of evoked AMPA-EPSCs. (B, C) AMPA charge for the first pulse (B; control n = 39, PV-GSK3β−/− n = 41) and total across both pulses (C; control n = 39, PV-GSK3β−/− n = 41) was greater in PV-GSK3β−/− PV interneurons. (D, E) PV-GSK3β−/− PV interneurons demonstrated a larger AMPA-mediated amplitude for both the first (D; control n = 40, PV-GSK3β−/− n = 41) and second pulse (E; control n = 39, PV-GSK3β−/− n = 41). (F) AMPA decay time was not different between genotypes (control n = 40, PV-GSK3β−/−n = 38). All data were analyzed either with an unpaired t-test or Mann-Whitney test. Data are means ± SEM. * p ≤ .05, ** p ≤ .01, n.s. = not significant, p ≥ .05.

Analogous to AMPA, PV-GSK3β^−/− PV interneurons had an increased NMDA charge for the first pulse (43.9 ± 2.19 pC) and total (77.4 ± 3.87 pC) compared to control (36.5 ± 1.80 pC, 65.0 ± 3.20 pC), U = 411, p = .012, Fig. 4A; U = 413, p = .013, Fig. 4B]. However, there were no changes in NMDA-EPSC amplitudes or decay time between genotypes (p ≥ .05, Fig. 4C-F). Our results indicate that PV-GSK3β^−/− PV interneurons in the mPFC have increased excitatory synaptic strength, with a more dramatic effect on AMPA receptor-mediated currents.

Fig. 4.

PV-specific GSK3β deletion potentiated NMDA charge in mPFC PV interneurons. (A) Sample traces of evoked NMDA-EPSCs. (B, C) NMDA charge for the first pulse (B; control n = 33, PV-GSK3β−/− n = 38) and total across both pulses (C; control n = 33, PV-GSK3β−/− n = 38) was greater in PV-GSK3β−/− PV interneurons. (D, E) No significant differences were reported in NMDA-mediated amplitude for both the first (D; control n = 35, PV-GSK3β−/− n = 39) and second pulse (E; control n = 35, PV-GSK3β−/− n = 40). (F) NMDA decay time was not different between genotypes (control n = 34, PV-GSK3β−/− n = 40). In the representative traces, the red solid line indicates 15 ms after peak EPSC amplitude. Measurements for NMDA-only current were collected at 15 ms time window in order to exclude contamination by AMPA-mediated current, which ended on average at 10 ms. The red dash line indicates the peak of NMDA amplitudes. All data were analyzed with a Mann-Whitney test. Data are means ± SEM. * p ≤ .05, n.s. = not significant, p ≥ .05. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

PV-GSK3β^−/− PV interneurons demonstrated a stronger, positive correlation between AMPA amplitude and charge likely due to the significant increase in amplitude reported [r(39) = 0.666, p < .0001; Fig. S3A). As AMPA amplitude increased in control, the NMDA/AMPA ratio decreased as expected [r(31) = −0.49, p = .004; Fig. S3B). However, this moderate correlation was lost in PV-GSK3β^−/− PV interneurons, suggesting changes in NMDA thereby balancing the ratio [r(35) = −0.08, p = .620; Fig. S3B). NMDA amplitude in PV-GSK3β^−/− PV interneurons moderately correlated with NMDA decay [r(37) = 0.45, p = .004; Fig. S4A] and strongly correlated with charge [r(35) = 0.72, p < .0001; Fig. S4B] and NMDA/AMPA ratio [r(35) = 0.73, p < .0001; Fig. S4C], suggesting a subunit change compared to control. These findings can also explain the reported AMPA changes, but the lack of change in the mean NMDA/AMPA ratio.

3.4. PV-GSK3β^−/− did not affect pyramidal neuron activity overall

Pyramidal neurons are the predominant computational output of the PFC and are direct targets of PV interneurons, which shape their excitatory activity. Because of the reported changes in PV interneurons, we further investigated the effects on pyramidal neurons regarding excitatory currents and E/I balance. No differences were reported in pyramidal NMDA/AMPA ratio, EPSC amplitudes (p ≥ .05, Fig. S5), or paired-pulse ratios between genotypes (p ≥ .05, Fig. S6). However, we observed NMDA-EPSCs amplitude strongly correlated with NMDA/AMPA ratio in PV-GSK3β^−/− pyramidal cells [r(27) = 0.52, p = .004; Fig. S7], suggesting changes in NMDA subunit expression. We also reported no differences in pyramidal E/I balance between genotypes (p ≥ .05, Fig. S8).

3.5. PV-GSK3β^−/− mice demonstrated facilitated learning on a cognitive task

Cognitive behaviors regulated by the PFC depend on inhibition. Because genetic deletion of GSK3β from PV interneurons altered the synaptic activity of PV interneurons in the PFC, cognitive acquisition and function were assessed utilizing a discrete paired-trial variable-delay T-maze task. Cognitive performance on a T-maze task includes two phases, learning and working memory. Learning was assessed by the total number of days to reach criterion, whereas working memory performance was assessed by the percent correct at each delay.

Days to criterion (DTC) was normally distributed for controls (p = .105), but not PV-GSK3β^−/− (p = .005), as assessed by Shapiro-Wilk’s test. Visual inspection of the histogram demonstrated that DTC for PV-GSK3β^−/− mice was positively skewed, suggesting facilitated learning compared to controls. Therefore, we assessed the relationship between DTC and genotype using a survival curve analysis. The survival distributions for the two genotypes were significantly different, χ² (1) = 3.92, p = .048 (Fig. 5A). A two-way mixed ANOVA was used to determine the effects of genotype and delay time on working memory performance. The two-way interaction between genotype and delay time as well as the main effect of genotype was not significant (p ≥ .05, Fig. 5B).

Fig. 5.

PV-GSK3β−/− mice demonstrated a faster learning curve on a spatial delayed non-match-to-sample T-maze task. The T-maze task measures two types of cognitive abilities, learning and working memory. (A) For the acquisition phase, each trial of the task was comprised of a sample (forced run) and a choice phase separated by a 10s delay. The performance was scored out of ten and animals needed to reach 70% correct on 3 consecutive days, or 2 consecutive days of ≥90% in order to move to testing (Days to a criterion; DTC). DTC was plotted as a survival curve for the first 7 days, which was the average maximum amount of days required for completion. Once subjects completed the criterion of training, they were counted as dropping out. A Gehan-Breslow-Wilcoxon test was to determine if there were differences in survival distribution for the two genotypes. The survival distributions for the two genotypes were significantly different, χ2 (1) = 3.916, p = .048. (B) For working memory performance, each trial of the task was comprised of a sample phase (forced run) and choice phase separated by a 5 s, 15 s, 30s, or 60s delay. The performance was measured by correct responses at each delay, averaged across three days of testing. A two-way mixed ANOVA was run to determine the effects of genotype and delay time on working memory performance. The two-way interaction between genotype and delay time was not statistically significant (F_{(3, 102)} = 1.95, p = .127]. All pairwise comparisons were performed for statistically significant main effects of delay time (F_{(3, 102)} = 8.32, p < .0001] and post-hoc analysis with a Bonferroni adjustment, adjusted p-values are reported. Mean performance was lower at the 30-s delay (72.0 ± 2.55%) and 60-s delay (68.8 ± 1.99%) compared to the 5-s delay (81.8 ± 2.03%), a mean difference of −9.83% (95% CI, −18.6 to −1.08, p = .020) and −13.1 (95% CI, −20.1 to −6.03, p < .0001). Mean performance was lower at the 60-s delay (68.8 ± 1.99%) compared to the 15-s delay (79.3 ± 2.13%), a mean difference of −11.2% (95% CI, −19.2 to −3.13, p = .003). No genotypic differences were observed; all mice demonstrated that working memory function begins to decline at delays of 30-s or longer (control n = 21, PV-GSK3β−/−n = 19 for 5, 15, and 60s; control n = 20, PV-GSK3β−/−n = 16 for 30s). Data are means ± SEM. * p ≤ .05, ** p ≤ .01, **** p ≤ .0001.

In summary, PV-GSK3β−/− mice exhibit facilitated acquisition on a cognitive task; 52.6% of PV-GSK3β−/− mice reached criterion after 4 days, compared to only 14.3% of control mice. By day 6, 73.7% of PV-GSK3β−/− mice reached criterion compared to 52.4% of control mice. No genotypic differences were observed across different time delays in working memory performance. Our data suggest that PV-specific GSK3β deletion is likely regulating synaptic plasticity that augments learning, but not transient memory.

Non-cognitive behaviors, such as locomotor activity, were assessed because alterations in these behaviors could affect the outcome and interpretation of other behavioral assays. Movement of an individual mouse was tracked for 60 min. A two-way mixed ANOVA was run to determine the effects of genotype and time on locomotor activity, which was not statistically significant (F_{(11, 319)} = 0.66, p = .661), indicating normal locomotion in both mice (Fig. S9A). The PFC is also a major regulator of social behaviors, which can become disrupted in psychiatric disorders, and has been linked to negative symptoms such as social deficits (Bicks et al., 2015). There is no general consensus on how PV interneurons affect social behaviors; because we are interested in PFC-regulated behaviors and genetic deletion of GSK3β in PV interneurons may change their functionality, we tested social preference to assess social motivation (Moy et al., 2004). A two-way mixed ANOVA was run to determine the effects of genotype and social condition (i.e., a novel object or a novel mouse) on sniffing time. The two-way interaction between genotype and sniffing time was not statistically significant (F_{(1, 27)} = 0.56, p = .461). Mean sniffing time of the object was lower (82.6 ± 3.55 s) than the novel mouse (115 ± 6.00 s), a mean difference of −32.2 s (95% CI, −42.9 to −21.5, p ≤ .0001). Our findings suggest that all adult mice, regardless of genotype, exhibited typical sociability (Fig. S9B).

4. Discussion

Our study was the first to investigate the connection between GSK3β and GABAergic fast-spiking PV interneurons. Hypofunctioning of PV interneurons are a leading hypothesis for the pathogenesis of neuropsychiatric disorders, which commonly exhibit cognitive impairments due to circuit disinhibition (Ferguson and Gao, 2018b; Lewis et al., 2012). What remains to be characterized is how excitatory receptors and subsequent neurotransmission in PV interneurons is regulated and what can lead to their disruption. GSK3β inhibitors have been demonstrated to function as rescuers of cognitive impairments in a gamut of conditions, including several CNS disorders (King et al., 2014). Considering that GSK3β has been suggested to regulate GluN2A (Monaco et al., 2018), which are crucial for PV interneurons, we elucidated how this kinase affected the intrinsic and synaptic properties of these neurons and ultimately the behavioral implications. Based on our findings reported herein, we suggest that GSK3β regulates excitatory homeostatic balance in PV interneurons.

GSK3β research has been framed in the context of global deletion and dopamine, but has been overlooked regarding GABAergic-dependent processes. Therefore, our study provides novel insight into the effects of PV-specific GSK3β genetic deletion on physiological properties in the PFC microcircuit and the behavioral consequences with this newly generated transgenic mouse. This study is the first to validate a PV-GSK3β^−/− mouse model, which exhibited cell-specific reduced expression of GSK3β within fluorescently labeled PV cells. PV expression did not change as a result of GSK3β deletion, suggesting that neuronal health and cell-fate were unaffected by Cre-expression and cell-specific GSK3β knockout.

Physiologically we found that mPFC PV interneurons lacking GSK3β exhibited increased excitability and sensitivity to incoming stimuli. PV interneurons in PV-GSK3β^−/− mice were more likely to fire, hold a charge, and summate. PV interneuron excitatory synaptic strength and AMPA-EPSC amplitude were also increased in PV-GSK3β^−/− mice. The augmented AMPA-EPSC amplitudes strongly suggest increased AMPA postsynaptic receptor expression, as the PPR appeared to be unaltered. This corresponds with our reported intrinsic property changes of increased sensitivity to presynaptic inputs and capacitance. If PV interneurons in PV-GSK3β^−/− mice express a larger pool of AMPAR at the postsynaptic density they are likely more responsive to subthreshold EPSPs, and thus have a greater probability to summate and generate an action potential. The overall increase in excitatory synaptic strength is supported by the Hebbian theory, which describes the phenomena of synaptic strengthening from repeated synaptic firing. Additionally, the increase in AMPA-EPSC amplitude parallels our previous findings where NMDA-EPSC amplitude was increased following GSK3β inhibition with lithium treatment (Monaco et al., 2018). Both studies demonstrate the negative regulatory role GSK3β plays regarding the two major glutamatergic receptors. GSK3β has been shown to regulate AMPA, one of the major excitatory receptors. AMPA receptor internalization requires activation of the GSK3β/KLC2 pathway, and GSK3β inhibition increased the surface expression of GluR1. GSK3β inhibition occludes AMPAR endocytosis, which is a crucial step in LTD, therefore signifying its role in modulating AMPA synaptic strength (Du et al., 2010). Interestingly, lithium was reported to increase surface expression of GluR1 and GluR2. Therefore, GSK3β deletion would prevent AMPAR internalization; in our case leading to a larger pool of AMPARs expressed at the postsynaptic site in PV interneurons (Peineau et al., 2008a). We also reported a lack of a negative correlation between AMPA amplitude and the NMDA/AMPA ratio in PV-GSK3β^−/− PV interneurons, as seen in controls. However, PV-GSK3β^−/− PV interneurons demonstrated stronger, positive correlations on several NMDAR associations. Collectively, these data suggest changes in NMDAR functioning following PV-specific GSK3β deletion.

Because of the aforementioned changes in PV interneurons, the effects on pyramidal neurons regarding excitatory currents and E/I balance were investigated. However, in contrast to PV interneurons, conditional GSK3β deletion in PV interneurons had limited effects on pyramidal neurons. We observed no changes in NMDA/AMPA ratio, EPSC amplitudes, PPRs, or E/I balance in pyramidal neurons. Nevertheless, we found NMDA-EPSCs amplitude strongly correlated with NMDA/AMPA ratio in PV-GSK3β^−/− pyramidal cells, suggesting changes in NMDA subunit expression. Both layer III and V pyramidal neurons receive convergent inputs from limbic structures including the hippocampus and amygdala, as well as neuromodulatory regions such as the ventral tegmental area, locus coeruleus, dorsal raphe, and basal forebrain. Albeit, the densest source of afferent orientates from the mediodorsal thalamus (MD), with layer III PV interneurons receiving dense projections. Layer III PV interneurons directly synapse on layer V pyramidal neurons, which regulate transmission to downstream motor centers that generate a behavioral response (Ferguson and Gao, 2015, 2018a; Hoover and Vertes, 2007; Kuroda et al., 2004). Our recent study reported a comparable laminar effect between layer II/III and layer V pyramidal neurons following GSK3β inhibition. A GSK3β inhibitor rescued spine number deficits in both layer II/III and layer V (Xing et al., 2016). We would not predict overall differences between layer III and V pyramidal neurons because of previous reports within our lab regarding GSK3β laminar effects and both are regulated by PV interneurons that are affected by genetic deletion of GSK3β. Furthermore, layer II/III pyramidal cells synapse onto layer V pyramidal neurons. Therefore any changes at layer II/III would likely become reflected downstream (Tritsch and Sabatini, 2012; Xing et al., 2016).

T-maze is a cognitive-behavioral paradigm that depends on the PFC (Kellendonk et al., 2009). PV-GSK3β^−/− mice demonstrated enhanced acquisition on the T-maze task compared to controls based on the number of days to reach criterion. A majority of PV-GSK3β^−/− mice (52.6%) reached criterion after 4 days, compared to only 14.3% of control mice, a difference of 38.4%. This difference was visually apparent in the DTC histogram; control mice exhibited a normally distributed curve, whereas PV-GSK3β^−/− mice exhibited a positively skewed distribution that dramatically peaked at day 4 and subsequently plateau. Furthermore, a significant survival curve quantified the shifted learning exhibited by PV-GSK3β^−/− mice. This facilitated learning may reflect a change in behavioral plasticity (Mery and J.G., 2010). No genotypic differences were observed across different time delays. Therefore, our data suggest that PV-specific GSK3β deletion is likely regulating synaptic plasticity that augments learning over several days, but not transient memory processes. No genotypic differences were reported for locomotor activity; therefore, genotype effects can be explained by reasons other than deficits in ambulatory movement. PV-GSK3β^−/− mice also demonstrated normal sociability, suggesting that GSK3β does not regulate sociability preference in the context of PV interneurons.

Hebbian LTP in PV interneurons could explain the facilitated learning behavior. Potentiation of disynaptic inhibition may help to sharpen memory traces by inhibiting surrounding pyramidal neurons, allowing for contrast during a cognitive task. Hebbian LTP between an afferent excitatory pyramidal neuron and its efferent pyramidal neuron allows for strengthening of synaptic connections relevant for the task. While LTP between the excitatory presynaptic neuron and the post-synaptic interneuron allows for inhibition of erroneous signaling and aids in spatial tuning (Kullmann and Lamsa, 2007). Disinhibition broadens the perceptive fields of pyramidal neurons because recurrent excitation begins recruiting superfluous pyramidal neurons that previously were shut down by inhibition. Disinhibition also impairs memory precision because pyramidal neurons on the edges of a tuning field become more susceptible to activation thereby increasing noise by opening up the receptive field to near-target distractors (i.e., errors) (Murray et al., 2014). By increasing excitatory conduction in PV interneurons, we may be preventing activation of near-target distractors that allows for sharpened ‘memory fields.’

GSK3β has been reported to localize within dendritic spines (Brandon and Sawa, 2011; Peineau et al., 2008a) and regulate synaptic plasticity, including both long-term depression (LTD) and long-term potentiation (LTP). LTD activates GSK3β; conversely, GSK3β becomes inhibited following LTP induction (Peineau, Bradley, 2008a, Peineau, Taghibiglou, 2007). Transgenic animals overexpressing GSK3β have diminished LTP and cognitive deficits, which were reversed with lithium administration, thereby suggesting that GSK3β hyperactivity was responsible for the cognitive impairments (Hernandez et al., 2002; Peineau et al., 2008a). Peineau et al. (2008) suggested that GSK3β could impart tonic inhibition over LTP due to its constitutive activity under basal conditions and therefore its inhibition could enhance LTP (Peineau et al., 2008b). Indeed, our study found that PV-selective GSK3β genetic deletion resulted in increased excitatory synaptic strength in PV interneurons and facilitated learning.

A caveat with a constitutive PV-GSK3β KO is that GSK3β is known to play a major role in early neural development, particularly neuronal proliferation and migration (Brandon and Sawa, 2011; Hur and Zhou, 2010; Mao et al., 2009). Therefore, removing this protein entirely could have negative effects on proliferation as well as the migration of PV interneurons throughout the brain. However, we did not observe an alteration in PV cell number, nor distribution within the PFC. The second restriction is that PV interneurons are prominent in other brain regions such as the hippocampus (Celio, 1990). Therefore, we cannot exclude changes in network circuitries that also influence prefrontal cortical function. Future studies are warranted in order to determine if similar effects hold up within the hippocampus.

Although our main hypothesis focuses on how GSK3β alters glutamatergic signaling, dopaminergic changes are likely to be altered. The dopamine D2 receptor (D2R) is a G-protein coupled receptor and an upstream regulator of the β-arrestin-2/Akt/GSK3 complex. D2R activation promotes the assembly of β-arrestin-2, Akt, PP2A, and GSK3. GSK3 stabilizes complex formation while Akt becomes dephosphorylated by PP2A, which further perpetuates GSK3 activity. β-arrestin-2 functions as an endocytic adaptor, binding to both the GPCR and endocytic proteins such as clathrin. Binding of β-arrestin-2 recruits endocytic proteins, triggering receptor internalization and desensitization (Beaulieu et al., 2005; Ma and Pei, 2007; O’Brien et al., 2011). GSK3β inhibitors, such as lithium, destabilize this complex and thereby prevent receptor endocytosis (O’Brien et al., 2011). Therefore, we expect that genetically deletion of GSK3β would increase D2 postsynaptic receptor expression, which is warranted for further investigation. Interestingly, our recent findings demonstrated that ablation of GSK3β in D2R+ neurons increased NMDAR-mediated current modulated by dopamine, augmented NMDAR protein expression, and prevented cognitive deficits induced by MK-801 treatment (Li et al., 2019). Therefore, our predictions of the PV-GSK3β KO model are corroborated by similar studies conducted within our lab.

GABAergic PV+ interneurons in the mPFC express D2 receptors only (Anastasiades et al., 2019; Gee et al., 2012; Seong and Carter, 2012). Application of a D2 agonist, quinpirole, increased mPFC PV+ interneuron excitability in adult animals. Thus, a D2-dependent mechanism drives interneuron firing activity (Tseng and O’Donnell, 2007). A specific loss of GSK3β in PV interneurons would likely disassociate the β-arrestin-2/AKT/GSK3β complex and block D2 receptor internalization. Stabilization of the D2 receptor population within PV interneurons is likely linked to the excitability changes by preventing endocytosis-mediated receptor internalization.

In contrast, D2Rs demonstrate an inhibitory effect on pyramidal cell excitability that is dependent on GABAA neurotransmission (Tseng and O’Donnell, 2007). Our reported PV effects are more substantial compared to those in pyramidal because pyramidal GSK3β protein expression remains intact. However, we predict that the application of a D2 agonist would reveal robust differences between pyramidal neurons from the two genotypes. D2R agonism would augment GABA transmission and dampen NMDA-mediated EPSCs in pyramidal cells, exhibiting a larger shift towards inhibition in PV GSK3β KO mice. Co-application of a D2 agonist with picrotoxin would then block the latter effects.

Our study, for the first time, examines how GSK3β activity affects GABAergic interneurons, which has important implications for neuropsychiatric diseases and other neurological conditions. Recently GSK3 inhibitors have been highlighted as promising rescuers of cognitive impairments in several CNS disorders (King et al., 2014). Lithium treatment has been clinically used for over 60 years and has been suggested as the first line of treatment based on updated guidelines and research (Alda, 2015). Mechanisms linked to neuronal plasticity are implicated to underlie the beneficial effects of lithium and growing evidence points to GSK3β as the mediator of lithium’s mechanism of action; however, this remains to be elucidated (Won and Kim, 2017). Surmounting research has emphasized that GABAergic hypofunctioning leads to cognitive impairments in psychiatric disorders. What remains to be explored is the function of GSK3β in GABAergic-dependent processes and how this affects cognition in brain regions such as the PFC.

In conclusion, we report that PV-specific GSK3β^−/− demonstrated increased sensitivity to presynaptic inputs and enhanced excitatory synaptic strength in PV interneurons. Whereas pyramidal neurons exhibited possible augmented NMDAR functioning. PV-GSK3β^−/− mice exhibited facilitated learning behavior, with no effect on ambulatory or sociability behavior. Collectively, our data suggest potentiated learning that is associated with the physiological underpinnings. Our study provides novel evidence that GSK3β is an important regulator of excitatory receptors in PV interneurons and provides an essential framework for which prospective studies can continue to build upon.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pnpbp.2020.109901.