Introduction

Glycogen synthase kinase (GSK)-3 is a highly expressed serine–threonine protein kinase present in all cells but is particularly abundant in the central nervous system (CNS) (Woodgett, 1990). GSK3β is mainly a cytoplasmic protein. However, it has been shown to accumulate within the nucleus in response to certain stimuli, although the function of nuclear GSK3β is not well understood (Bijur et al, 2000; Bijur and Jope, 2003). There are two GSK3 isoforms, GSK3α and GSK3β, encoded by separate genes. Although they share significant homology within the kinase domain, their function and/or expression is not fully redundant based on the finding that deletion of GSK3β, but not GSK3α, results in embryonic lethality (Hoeflich et al, 2000). GSK3 was first identified as a kinase that phosphorylates and negatively regulates glycogen synthase (Embi et al, 1980; Frame and Cohen, 2001), but in the many systems where it has been studied, one major consequence of active GSK3β is cell death (Jacobs et al, 2012). In the CNS, loss of GSK3 results in brain abnormalities, neuronal hyperproliferation, and excess survival (Kim et al, 2009). In addition, hyper-activation of GSK3β in neuronal cells increases cell death in response to cell stress (Bijur et al, 2000). Increased neuronal cell death caused by enhanced GSK3 activity has been thought to contribute to human neurological disorders, such as Huntington disease, Alzheimer’s disease, and bipolar disorder (Carmichael et al, 2002; Cai et al, 2012; Cole, 2013). Increased GSK3β activity has been shown to mediate tau hyperphosphorylation, β-amyloid-induced neurotoxicity, and mutant presenilin-1 pathogenic effects in Alzheimer’s disease (Jope and Johnson, 2004). Inhibitors of GSK3 have shown promise in mouse models of Alzheimer’s disease (Noble et al, 2005; Rockenstein et al, 2007; Eldar-Finkelman and Martinez, 2011) and some have been tested in clinical trials for Alzheimer’s disease (Lovestone et al, 2015). Lithium, commonly used in the treatment of bipolar disorder, stabilizes mood in part through the inhibition of GSK3 activity (Jope, 2011). GSK3β inactivation has also been implicated mechanistically in the antidepressant effect of ketamine (Beurel et al, 2011; Liu et al, 2013; Chiu et al, 2015). Thus, determining the mechanisms that regulate GSK3 activity in CNS is highly relevant.

Unlike most kinases, GSK3 is constitutively active. A widely studied mechanism for GSK3 inhibition is through phosphorylation of Ser9 on GSK3β and Ser21 on GSK3α primarily by Akt (Cross et al, 1995). The phosphorylated N-terminus of GSK3β at Ser9 acts as a competitive inhibitor by folding into the active site and preventing substrate binding (Frame et al, 2001). Phosphorylation of Ser9/Ser21 can be detected constitutively in most cells, but it can be further increased by insulin, growth factors, and cellular stress signaling (Doble and Woodgett, 2003). Studies with double knockin (KI) mice in which Ser9 in GSK3β and Ser21 in GSK3α were mutated to Ala have shown the relevance of this mechanism for inactivation of muscle glycogen synthase by insulin (McManus et al, 2005), but not for the inactivation of GSK3 through WNT/β-catenin signaling (McManus et al, 2005). WNT signaling promotes sequestration of GSK3 in multivesicular endosomes and the generation of lipoprotein receptor-related protein pseudo-substrates (Niehrs and Acebron, 2010). Interestingly, the loss of inhibitory serine phosphorylation in GSK3 Ser9Ala/Ser21Ala double KI mice did not affect neuronal survival in vitro or in vivo (Hongisto et al, 2008; Eom and Jope, 2009) further supporting the existence of an alternative neuronal protection mechanism to regulate GSK3β in the brain.

Recently, we identified another direct inhibitory phosphorylation pathway for GSK3β, but not for GSK3α (Thornton et al, 2008). p38 MAPK inactivates GSK3β by phosphorylation of Thr390 in human GSK3β or Ser389 in mouse GSK3β (Thornton et al, 2008). This phosphorylation by p38 MAPK results in a similar degree of GSK3β activity inhibition as Ser9 phosphorylation by Akt (Thornton et al, 2008). Thr390/Ser389 is located at the very end of the GSK3β C-terminus, which is also very flexible in structure, similar to the N-terminus where Ser9 is located (Dajani et al, 2001). Thus, the mechanism of inactivation upon phosphorylation at the C-terminus is also likely through folding into the active site and competing for substrate binding. However, these two mechanisms for GSK3β inactivation are independent, respond to different stimuli, and target different pools of GSK3β (Thornton et al, 2016). While phosphorylation of Ser9 is ubiquitously present in most tissues, Ser389 phosphorylation is more restricted to specific tissues (eg, brain, thymus, and spleen) under physiological conditions (Thornton et al, 2008). In addition, Ser9 phosphorylation of GSK3β occurs primarily in the cytosol, but we have recently shown that Ser389 phosphorylation targets nuclear GSK3β (Thornton et al, 2016). Importantly, phosphorylation of GSK3β at Ser389 is specifically triggered in response to DNA double-strand breaks (DSB) induced by external stimuli as well as by naturally generated DSB arising from V(D)J recombination and class switch recombination in T and B lymphocytes (Thornton et al, 2016). The major function of Ser389 GSK3β phosphorylation in these cells is to prevent cell death, primarily through necroptosis caused by DSB (Thornton et al, 2016).

Although phosphorylation of GSK3β at Ser389 is selectively abundant in the brain under physiological conditions (Thornton et al, 2008), the function of this mechanism of GSK3β regulation in the brain is unknown. Here we show that phosphorylation of GSK3β at Ser389 is a mechanism to inactivate the nuclear pool of GSK3β within the brain. Failure to inactivate nuclear GSK3β in Ser389Ala mutant mice causes the death of a subset of neurons in the cortex and hippocampus. While spatial memory and anxiety/depression-like behaviors are not impaired, Ser389Ala GSK3β KI mice display augmented fear-conditioning behavior, a phenotype commonly observed in stress-associated diseases such as post-traumatic stress disorder (PTSD). Thus, inactivation of GSK3β through Ser389 phosphorylation could be a novel approach for the treatment of PTSD-like symptoms.

Materials and methods

Mice

C57BL/6 GSK3β KI mice have been previously described (Thornton et al, 2016). Wild-type mice (WT) were purchased from Jackson Laboratories (Bar Harbor, ME). Mice deficient for p38α MAPK specifically in neurons (p38αΔ-N) were generated as described (Colié et al, 2017). p38α floxed homozygous mice were crossed with p38α floxed homozygous CAMKII-CRE positive mice to generate p38α floxed homozygous mice plus (p38αΔ-N) or minus (WT) a single copy of the CAMKII-CRE transgene. All procedures were approved by the University of Vermont Animal Care and Use Committee.

Behavioral Tasks

A cohort of male and female WT and GSK3β KI mice (n=5 females, 5 males per genotype) were examined in the behavioral tasks described below. The order of testing was acoustic startle, open field, zero maze, auditory fear conditioning, terminating with forced swim. A separate cohort was used for the water maze experiment (n=8 females, 7 males per genotype). The following behavioral tasks were performed: (1) open field and zero maze, (2) water maze, (3) forced swim, and (4) auditory fear conditioning. Experimental detail is provided in the ‘Extended Methods’ in the Supplementary Information section.

Western Blot Analysis

Whole-cell extracts were prepared in Triton lysis buffer and used for western blot analysis as we previously described (Derijard et al, 1994; Rincon et al, 1997). Nuclear and cytosolic extracts were prepared as previously described (Schreiber et al, 1989; Tugores et al, 1992). For co-immunoprecipitation, nuclear extracts were prepared as previously described (Jamil et al, 2010). ExactaCruz F Immunoprecipitation Kit (Santa Cruz Biotechnology) was used for immunoprecipitation/western blot analysis with rabbit antibodies. Anti-actin, anti-GAPDH, and anti-GSK3β were purchased from Santa Cruz Biotechnology. Anti-phospho-Ser9 GSK3β, anti-phospho-Mcl1, and anti p38α MAPK were purchased from Cell Signaling Technology (Danvers, MA). Anti-phospho-S389 GSK3β and anti-TUJ1 was purchased from Millipore (Billerica, MA). Anti-phospho-Thr390 GSK3β rabbit polyclonal antibody has been described (Thornton et al, 2016). Anti-rabbit HRP and anti-mouse HRP (Jackson ImmunoResearch Laboratories) and anti-goat-HRP (Santa Cruz Biotechnology) were used as secondary antibodies. NIH ImageJ was used to quantify band intensity.

GSK3β Kinase Activity Assays

To determine whether S389 phosphorylation regulated a distinct component of GSK3β activity, kinase assays were conducted. GSK3β was immunopreciptitated from cerebral cortex protein lysates and kinase activity was determined as previously published (Thornton et al, 2008).

Immunolabeling

The primary antibodies used included rabbit anti-phospho-Ser389 or anti-phospho-Thr390 GSK3β polyclonal antibodies, Anti-NeuN (Millipore), anti-TUJ1, anti-rabbit Alexa Fluor 568 (Invitrogen) or Cy3 (Jackson ImmunoResearch). DAPI was used as a nuclear stain. Images of cells were acquired on a Zeiss LSM-510. Staining of human tissue was performed as previously described (Long et al, 2011). Images were acquired with identical exposure settings using a SPOT RT digital camera. Fluoro-Jade C (Millipore) staining was performed according to the manufacturer's instructions.

Statistical Analysis

Data are expressed as means±standard error of the mean. Behavioral analysis utilized mixed model ANOVAs followed by LSD protected t-tests. Initial analysis included sex. However, we observed no sex differences; therefore, sex was left out of the analysis that follows. Results were analyzed by SPSS software version 22 (IBM; Armonk, NY). P<0.05 was considered statistically significant.

Results

Ser389 Phosphorylation of GSK3β Selectively Targets the Nuclear Pool of GSK3β in the Brain

The regulation of GSK3β by Ser9 phosphorylation occurs in all tissues as a default pathway to restrain GSK3β activity (Jope and Johnson, 2004). In contrast, phospho-Ser389 GSK3β is highly abundant in the brain (Thornton et al, 2008). To determine the distribution of phospho-Ser389 GSK3β within the brain, western blot analysis was performed using lysates from different regions of the brain. While phospho-Ser9 GSK3β was abundant in all regions of the brain examined, higher levels of phospho-Ser389 GSK3β were detected in the cortex and hippocampus (Figure 1a). To address whether phosphorylation of Ser389 and phosphorylation of Ser9 represented alternative pathways to control two independent spatial pools of GSK3β we performed immunoprecipitation analysis. Whole-cell extracts from the cerebral cortex were used to immunoprecipitate phospho-Ser9 GSK3β, and both the immunoprecipitate and flow-through were examined by western blot analysis. As expected, phospho-Ser9 was abundantly present in the immunoprecipitates, but not in the flow-through (Figure 1b). In contrast, phospho-Ser389 was not detected in the phospho-Ser9 immunoprecipitates, but was abundantly present in the flow-through (Figure 1b), indicating that the phospho-Ser9 GSK3β pool is not phosphorylated at Ser389. Analysis of total GSK3β revealed that only a fraction of GSK3β was actually phosphorylated on Ser9 since large amounts of GSK3β were still present in the flow-through after phospho-Ser9 immunoprecipitation (Figure 1b). Thus, phosphorylation of GSK3β on Ser389 targets an independent pool of GSK3β in the brain that is not phosphorylated on Ser9.

Figure 1
figure 1

Ser389/Thr390 phosphorylation of nuclear GSK3β in regions of the brain. (a) Whole-cell lysates from different regions of wild-type (WT) mouse brain (BS, brain stem; Cb, cerebellum; Ctx, cerebral cortex; Hip, hippocampal formation; Str, striatum; Thl, thalamus) were examined for P-Ser389GSK3β, P-Ser9 GSK3β and total GSK3β by western blot analysis. GAPDH is shown as a loading control. The ratio of P-Ser389 to P-Ser9 are shown for comparison. (b) P-Ser9 GSK3β was immunoprecipitated from mouse cerebral cortex extracts. P-Ser389GSK3β, P-Ser9 GSK3β, and total GSK3β in the immunoprecipitate (P-S9 IP) or the flow-through (Flow Thr) left were examined by western blot analysis. (c) The presence of P-Ser389 (red) and DAPI nuclear staining (blue) in mouse cerebral cortex were examined by immunostaining and microscopy. Cells in white boxes are shown enlarged to show cellular localization. (d) The presence of P-Thr390 (red) and DAPI nuclear staining (blue) in the post-mortem human cerebral cortex were examined by immunostaining and microscopy. Cells in white boxes are shown enlarged to show cellular localization. (e) Western blot analysis using nuclear (N) and cytosolic (C) extracts from the mouse cortex were analyzed by western blot for P-Ser389 GSK3β, P-Ser9 GSK3β and total GSK3β. GAPDH and Histone are shown as controls for the cytoplasmic and nuclear fractions, respectively.

PowerPoint slide

We have recently shown that Ser389 phosphorylation primarily regulates nuclear GSK3β in lymphocytes (Thornton et al, 2016). We, therefore, examined the subcellular distribution of phospho-Ser389 GSK3β in mouse cerebral cortex by immunostaining and confocal microscopy. The immunostaining revealed a punctate nuclear distribution of phospho-Ser389 GSK3β in a subset of cells (Figure 1c). Similarly, immunofluorescence staining of post-mortem human brain tissue for phospho-Thr390 GSK3β (human equivalent of mouse Ser389) revealed a primarily nuclear staining (Figure 1d). In addition, western blot analysis using nuclear and cytosolic extracts from the cerebral cortex further confirmed the presence of phospho-Ser389 GSK3β in the nuclear fraction, whereas it was practically undetectable in the cytosol (Figure 1e). Thus, unlike phosphorylation of Ser9, phosphorylation of Ser389 targets specifically the nuclear pool of GSK3β in the brain.

Phosphorylation of GSK3β on Ser389 is Required to Restrain the Activity of Nuclear GSK3β in the Brain

To investigate the role of GSK3β Ser389 phosphorylation in the brain, we used GSK3β Ser389Ala knockin (GSK3β KI) mice where Ser389 was replaced with Ala to prevent the phosphorylation of this residue and block the regulation of GSKβ by this mechanism (Thornton et al, 2016). Western blot analysis confirmed the absence of phospho-Ser389 GSK3β in the whole brain from homozygous GSK3β KI mice while normal levels were present in heterozygous GSK3β KI mice (Figure 2a). Ser9 phosphorylation of GSK3β was not affected by the Ser389Ala mutation (Figure 2a), further demonstrating that these two pathways are independent of each other. No discernible differences in gross brain anatomy were observed between GSK3β KI mice and WT mice (Figure 2b). Immunostaining analysis with NeuN and class III β-tubulin/Tuj1, two standard markers for neurons, did not show drastic differences in the density of neurons within the cerebral cortex and hippocampus of WT and GSK3β KI mice (Figure 2c). Thus, blocking Ser389 phosphorylation does not have a major effect on neuronal homeostasis in the brain.

Figure 2
figure 2

Increased nuclear GSK3β activity in the brain of GSK3β Ser389Ala KI mice. (a) Whole-cell lysates from brains of WT, heterozygous (Het), and homozygous GSK3β KI mice were analyzed by western blot analysis for P-Ser389 GSK3β, P-Ser9 GSK3β, and total GSK3β. (b) Brains from WT and GSK3β KI mice. (c) Immunostaining for NeuN and Tuj1 (class II β-tubulin), as markers of mature neurons in WT and GSK3β KI cerebral cortex and hippocampal formation. The location of CA1, CA3, and the dentate gyrus (DG) is shown. The white boxes indicate the regions that correspond to the magnified images that follow. (d) GSK3β kinase activity using lysates from the cerebral cortex of WT and GSK3β KI mice (n=9). (e) GSK3β kinase activity in nuclear and cytosolic extracts from the hippocampus of WT and GSK3β KI mice (n=3). (*) indicates P-value <0.05 as determined by t-test.

PowerPoint slide

To show that phosphorylation on Ser389 contributes to restrain the kinase activity of GSK3β in the brain, we measured GSK3β kinase activity in brain tissue from WT and GSK3β KI mice. Despite the already high kinase activity, we could detect a significant increase in kinase activity in extracts from GSK3β KI mice compared with the activity in WT mice (Figure 2d). To address the role of Ser389 phosphorylation of GSK3β on the activity of the nuclear vs the cytosolic pool of GSK3β, we examined GSK3β activity in nuclear and cytosolic extracts from the brains of WT and GSK3β KI mice. Consistent with the nuclear localization of phospho-Ser389 GSK3β, nuclear GSK3β activity was significantly elevated in GSK3β KI mice compared with WT mice (Figure 2e). In contrast, cytosolic GSK3β activity was not increased in GSK3β KI mice (Figure 2e). Thus, phosphorylation of GSK3β on Ser389 is a mechanism to inactivate nuclear GSK3β in the brain.

Phosphorylation of GSK3β on Ser389 is Required for Survival of a Neuronal Subset in the Cortex and Hippocampus

Enhanced GSK3β activity has been shown to promote neuronal cell death (Bijur et al, 2000). Although the gross anatomy of the GSK3β KI brains was not altered (Figure 2b), histological examination of hematoxylin & eosin (H&E)-stained sections revealed the presence of some highly eosinophilic (darker) neurons with pyknotic nuclei in certain brain regions of GSK3β KI mice (Figure 3a). Eosinophilic neurons were particularly prominent in the cingulate and frontal cortices (Figure 3a—cortex 1 and Supplementary Figure S1), the CA1 region (Figure 3a), and dentate gyrus (Supplementary Figure S2) of the hippocampal formation, the amygdala and entorhinal cortex (Supplementary Figure S3) and the bed nucleus of the stria terminal and hypothalamus (data not shown). However, eosinophilic neurons were absent in other brain regions such as the parietal cortex (Figure 3a, cortex 2) and CA3 region of the hippocampus (Figure 3a) in GSK3β KI mice. The morphology of these cells is characteristic of what in human brain pathology is defined as ‘dark neurons’ and represents neurodegenerative cells (Garman, 2011). The restricted distribution of dark neurons suggests that neurons within a defined neuronal circuit are affected.

Figure 3
figure 3

Impaired neuronal survival in GSK3β Ser389Ala KI mice. (a) H&E-stained section of WT (low magnification). Box 1 indicates the region for ‘Cortex 1’ and box 2 indicates the region for ‘Cortex 2’. The location of CA1, CA3, and the dentate gyrus (DG) are shown. Higher magnification of H&E-stained sections of WT and GSK3β KI cortex (1 and 2), and the hippocampal CA1 and CA3 sub-regions. (b) Immunostaining for phospho-Ser389 GSK3β in WT and GSK3β KI cerebral cortex and CAI hippocampal subregion. (c) Fluoro-Jade C (green) and DAPI (blue) staining of sections from WT and GSK3β KI cerebral cortex and hippocampal CA1 subregion. (d) Immunostaining of WT cortex for P-Ser389 (red), γH2AX (green), and DAPI nuclear staining (blue). Cells with co-staining for both P-Ser389 and γH2AX (white) are indicated by the asterisks (*).

PowerPoint slide

To examine whether the distribution of Ser389 phosphorylated GSK3β in WT mice correlated with the subregions where the dark neurons preferentially accumulate in GSK3β KI mice, we performed immunostaining for phospho-Ser389 GSK3β. The highest levels of phospho-Ser389 GSK3β immunostaining were found in the cerebral cortex and hippocampal regions of WT mice that correlated with the regions in GSK3β KI mice with a higher frequency of dark neurons (Figure 3b). The specificity of the phospho-Ser389 GSK3β staining was demonstrated by its absence in the brain of GSK3β KI mice (Figure 3b).

To demonstrate that the dark cells present in GSK3β KI brains represent degenerative neurons, brains sections from WT and GSK3β KI mice were stained with Fluoro-Jade C, the gold-standard dye used to visualize neurodegeneration (Schmued et al, 2005). Fluoro-Jade staining was practically undetectable in brains from WT mice as expected (Figure 3c). In contrast, clear Fluoro-Jade staining was present in regions of the cerebral cortex and hippocampus in GSK3β KI brain (Figure 3c), and these regions correlated with the areas where the dark cells were more abundant. We have previously shown that in response to endogenous DNA DSB phospho-Ser389 GSK3β co-localizes with γH2AX, a established marker for DSB (Thornton et al, 2016). A number of recent studies have revealed the presence of DSB in neurons even during normal brain activity (Suberbielle et al, 2013; Madabhushi et al, 2015). To determine whether cells positive for phospho-Ser389 GSK3β contain DSB, we performed co-immunolabeling for phospho-Ser389 GSK3β and γH2AX. Phospho-Ser389 GSK3β and γH2AX co-localized in cells within the cingulate cortex indicating that cells positive for phospho-Ser389 GSK3β also contain DSB (Figure 3d and Supplementary Figure S4). Thus, inactivation of nuclear GSK3β through phosphorylation on Ser389 contributes to the fitness of a subset of neurons potentially undergoing DSB distributed in selective areas of the brain.

Conditional Deletion of p38a MAPK Results in Degeneration of a Neuronal Subset

Although Ser9 phosphorylation of GSK3β is mediated by Akt, Ser389 phosphorylation of GSK3β is mediated by p38 MAPK (Thornton et al, 2008, 2016). We investigated whether the failure to inactivate GSK3β through p38 MAPK could have a similar effect on neuronal survival. We used genetically modified mice with the p38α MAPK gene specifically disrupted in neurons (p38αΔ−N) through the expression of Cre recombinase under the control of the calcium/calmodulin-dependent protein kinase II (CAMKII) gene promoter (Colié et al, 2017). Western blot analysis confirmed the reduced expression of p38α MAPK in both the cortex and the hippocampus (Figure 4a). The gross brain morphology was not altered in p38αΔ−N mice (Figure 4b). Western blot analysis revealed a marked decrease of phospho-Ser389 GSK3β in lysates from p38αΔ−N mice compared with WT mice, indicating that phosphorylation of GSK3β at Ser389 in the brain is p38α MAPK dependent (Figure 4c). Examination of H&E staining revealed a higher frequency of dark neurons in the cortex of p38αΔ-N mice compared with WT mice (Figure 4d). In addition, increased numbers of Fluoro-Jade-positive cells were detected in the cortex of p38αΔ−N mice (Figure 4e and Supplementary Figure S5). Together these results show that loss of p38α MAPK correlates with decreased phosphorylation of GSK3β Ser389 and decreased survival of a subset of neurons similar to that found in GSK3β KI mice.

Figure 4
figure 4

Neurodegeneration in neuronal-specific p38α MAPK KO mice. (a) Cell lysates from the cortex and hippocampus of WT and p38αΔ−N mice were analyzed by western blotting for p38α MAPK using actin as a loading control. (b) Representative pictures of brains from WT and p38αΔ-N mice. (c) Whole-cell lysates from brains of wild-type WT and p38αΔ-N mice were analyzed by western blot analysis for P-Ser389 GSK3β, p38α MAPK, P-Ser9 GSK3β, and total GSK3β. (d) H&E stained sections of WT and p38αΔ−N cerebral cortex and CA1. Boxed regions show the cortex area magnified (middle panels). (e) Fluoro-Jade C (green) and DAPI (blue) sections from WT and p38αΔ−N cerebral cortex.

PowerPoint slide

Failure to Inactivate GSK3β through Ser389 Phosphorylation in the Brain Promotes Death by Necroptosis

The best-characterized mechanism for GSK3β to mediate cell death is by phosphorylation of β-catenin in the cytosol, leading to the rapid degradation of this protein by APC complex (Filali et al, 2002; Liu et al, 2002). We examined whether the levels of β-catenin in the brain of GSK3β KI mice were decreased. Western blot analysis showed no difference in the levels of β-catenin between WT and GSK3β KI brains (Figure 5a). GSK3β can also promote cell death by the degradation of the pro-survival factor Mcl-1 through its phosphorylation on Ser159 (mouse Ser140) (Maurer et al, 2006). Since Mcl-1 deficiency has also been shown to promote neuronal cell death in response to DSB (Arbour et al, 2008), we examined the levels of Mcl-1 in the brain in WT and GSK3β KI mice by western blot analysis. Interestingly, unlike β-catenin, Mcl-1 levels were strikingly decreased in GSK3β KI mice compared with WT mice (Figure 5a). To further show that increased GSK3β activity in these mice we examined phosphorylation of Mcl-1on Ser140, the target for GSK3β. Despite the very low levels of Mcl-1 detected in GSK3β KI mice, the ratio of phosphorylated Mcl-1 was higher than in WT mice (Figure 5b). Thus, failure to inactivate nuclear GSK3β through this pathway results in increased phosphorylation and degradation of Mcl-1 in brain. Since mitochondrial Mcl-1 protects from apoptosis, we examine the presence of active caspase 3 by western blot analysis. However, the level of active caspase 3 was very low in brain, and there was no difference between WT and GSK3β KI mice (Figure 5c), suggesting that the neurodegeneration observed in GSK3β KI mice may not be through apoptosis.

Figure 5
figure 5

Decreased Mcl-1 and increased necroptosis in the brains of GSK3β S389Ala KI mice. (a) β-catenin, Mcl-1, P-Ser389 GSK3β and total GSK3β were examined by western blot analysis in brain lysates from WT and GSK3β-KI mice. GAPDH is shown as a loading control. (b) Graph showing the ratio of P-Mcl-1 to Mcl-1 from the densitometry of western blots examining P-Mcl and Mcl in brain lysates from WT and GSK3β-KI mice. (c) Western blot analysis for caspase 3 in brain lysates from WT and GSK3β-KI mice. GAPDH is shown as a loading control. (d) The presence of RIPK3 (red) and DAPI nuclear staining (blue) in mouse cerebral cortex was examined by immunostaining and microscopy. (e) Western blot analysis for phospho-MLKL in brain lysates from WT and GSK3β-KI mice. GAPDH is shown as a loading control. The panel to the right displays the densitometry of phospho-MLKL.

PowerPoint slide

In B cells from GSK3β KI mice, decreased levels of nuclear Mcl-1 have been associated with increased necroptosis (Thornton et al, 2016), an alternative pathway of death independent of mitochondria but dependent on the serine-threonine kinases RIPK1 and RIPK3 (Tait et al, 2014). Necroptosis has been found in some neurons where RIPK3 selectively translocates to the nucleus (Tait et al, 2014; Vitner et al, 2014). We, therefore, examined RIPK3 in brains from GSK3β KI mice by immunostaining and confocal microscopy analysis. Increased nuclear staining for RIPK3 was detected in the cerebral cortex of GSK3β KI mice compared with WT mice (Figure 5d and Supplementary Figure S6), supporting a greater magnitude of necroptosis in the GSK3β KI mice. In necroptosis, activated RIPK3 phosphorylates MLKL, one of the mediators of necroptosis that translocates to the membrane causing membrane rupture and death (Rodriguez et al, 2015). Analysis of phospho-MLKL by western blot analysis showed increased phospho-MLKL levels in GSK3β KI mice compared with WT mice (Figure 5e). Thus, taken together these data suggest that failure to inactivate GSK3β through Ser389 phosphorylation leads to increased necroptosis in some subregions of the brain.

Spatial Memory and Anxiety/Depression-like Behaviors are not Altered in GSK3β KI mice

The presence of pyknotic cells by H&E staining and fluoro-jade labeling in the hippocampus and cortex of GSK3β KI mice is consistent with localized neuronal degeneration. Given the hippocampal neurodegeneration detected, we assessed performance in the water maze, a hippocampal-dependent learning task (Morris et al, 1982). WT and GSK3β KI mice were trained to find a submerged platform over 8 days, with four trials per day. Performance improved over sessions (Day f(2,56)=80.75, P<0.0001) though there was no effect of genotype (Genotype f(1,28)=1.338, P>0.05) or interaction (Day × Genotype f(2,56)=1.37, P >0.05), indicating that hippocampal spatial learning was similar in WT and GSK3β KI mice (Figure 6a). Spatial memory in the water maze can also be measured by utilizing a probe trial in which the platform is removed and the time spent in the general area of the platform is quantified. GSK3β KI and WT mice spent similar time searching in the platform quadrant (t(28)=0.45, P>0.05) (Figure 6b). Additionally, the time spent searching in the appropriate quadrant was above chance in each group (WT t(14)=3.61, P<0.01, KI t(28)=3.36, P<0.01) (Figure 6b). Further, when the platform was moved to a new location, mice again performed similarly. Over four trials, distance travelled was reduced (f(3,84)=9.97, P <0.0001), but there was no difference between the WT and GSK3β KI mice (f(1,28)=.44, P >0.05) or interaction (f(3,84)=0.63, P>0.05) (Figure 6c).

Figure 6
figure 6

Hippocampal-dependent spatial memory and anxiety- and depression-like behaviors are not affected in GSK3β Ser389 KI mice. (a) The water maze is a spatial navigation task dependent on hippocampal integrity. Mice were trained to find a hidden platform over four trials each day for 8 days. (b) A probe trial, in which the platform is removed, can be used to assess memory for the platform location (dotted line represents chance performance). (c) The platform was moved to a new location for reversal trials to measure cognitive flexibility. (d) Locomotor activity and (e) center time during the open field test. (f) Time in the open quadrant and (g) and entries into the open quadrant during the zero-maze test. (h) Immobility time measured during a forced swim exposure. Data are mean±SEM.

PowerPoint slide

Some reports indicate reduced hippocampal volume, indicative of neuronal atrophy, in the hippocampus of individuals with stress-associated diseases such as depression (Sheline et al, 1999). Therefore, the neuronal loss observed in the GSK3β KI mice could also affect anxiety and depression-like behavior. The open field test assesses both locomotor activity and anxiety-like behavior. When subjected to a 10-min open field test, GSK3β KI and WT mice traveled similar distances (t(18)=0.61, P>0.05) (Figure 6d), indicating that locomotion was not affected in GSK3β KI mice. Further, each group entered the center of the arena equivalently (t(18)=0.71, P>0.05), and spent equal time exploring the center of the open field (t(18)=1.70, P>0.05) (Figure 6e), indicating similar anxiety levels between WT and GSK3β KI mice. The zero maze is another test for anxiety in which mice explore a raised circular track with enclosed portions in opposing quadrants over 5 min. Anxiety is defined as reduced open quadrant time or reduced open quadrant entries. Both open quadrant entries (t(18)=0.63, P=0.54) (Figure 6f) and open quadrant time (t(18)=0.85, P=0.41) (Figure 6g) were similar between GSK3β KI and WT mice providing further confirmation that anxiety-like behavior was similar between strains.

The forced swim test is a commonly used model of depression that measures behavioral despair through immobility, with increasing immobility time suggestive of increased depression (Bogdanova et al, 2013). An analysis of forced swim behavior by minute demonstrated that immobility increased in the later portions of the 6 min test (Minute f(5,90)=24.54, P<0.0001), but there was no difference between GSK3β KI mice and WT mice (f(1,18)=0.23, P>0.05) (Figure 6h). Thus, the regional neuronal degeneration observed in the GSK3β KI mice has no impact on locomotor activity, overall anxiety, or depression-like behavior.

Failure to Inactivate GSK3β through Ser389 Phosphorylation Causes an Exaggerated and Persistent Conditioning Fear

Reduced hippocampal volume has also been reported in individuals with PTSD (Gilbertson et al, 2002). Fear conditioning is highly conserved behavior that is defined by a fear response (eg, freezing in rodents) to a previously non-threatening stimulus (conditioned stimulus) that has been paired with an aversive stimulus (unconditioned stimulus). Dysregulated fear conditioning is a hallmark of stress exposure models used in rodents to model disorders such as PTSD (Izquierdo et al, 2006; Baran et al, 2009; Hoffman et al, 2014). Fear conditioning is dependent on the amygdala (LeDoux, 2003; Davis, 2006; Kim and Jung, 2006; Duvarci and Pare, 2014), but it is also influenced by the hippocampus through reciprocal connections between the two regions (Maren, 2001). We therefore investigated the response of GSK3β KI mice to auditory fear conditioning relative to WT mice. In this behavioral test, mice were given a series of pairings of a tone immediately followed by a mild foot shock in a distinctive context (ie, training chamber). Both GSK3β KI mice and WT mice displayed an increasing fear response to tone presentations subsequent to the first tone (f(5,90)=27.99, P<0.0001) and there was no difference between the two groups (f(1,18)=1.06, P>0.05) or interaction (f(5,90)=1.70, P>0.05) (Figure 7a). Thus, acquisition of conditioned fear proceeded equivalently in GSK3β KI and WT mice. In order to test conditioned fear memory mice were given a pair of tests in a counterbalanced fashion beginning 24 h after conditioning and separated by 24 h. To test for an association between the training chamber and foot shock (contextual fear conditioning), mice were placed in the training chamber and allowed to explore without presentation of the tone. Interestingly, GSK3β KI mice clearly froze more than WT mice when placed into the training chamber (t(18)=3.18, P<0.01) (Figure 7b). The association between the tone and the foot shock (auditory fear conditioning) was probed in a novel chamber where a 2-min baseline exploratory period was followed by the presentation of the tone that had previously been paired with foot shock. GSK3β KI mice also showed greater conditioned fear when the tone was presented in a novel context (t(18)=2.68, P<0.05) (Figure 7c). Together these results suggest that the inability to phosphorylate Ser389 in GSK3β KI mice results in exaggerated fear conditioning to both contextual and auditory cues.

Figure 7
figure 7

GSK3β S389Ala KI mice show an exacerbated fear response. (a) Acquisition of fear conditioning during a training session with five tone-shock pairs. (b) Contextual fear expression upon return to the training context. (c) Tone fear expression during tone presentation in a novel context. (d) Fear expression immediately after exposure to the training context in comparison to expression upon placement into the novel context. (e) Time-dependent changes in tone fear expression during, and immediately after, the 3 min tone presentation in the novel context. Data are mean±SEM, Significance was determined by t-test (*P<0.05, **P<0.01).

PowerPoint slide

One component of PTSD is a pathological process by which overlapping features in a safe environment cause transfer of fear from a traumatic event (generalization) (American Psychiatric Association, 2000). Thus, to further examine the exaggerated fear behavior exhibited by GSK3β KI mice, we compared the first 2 min of behavior after mice were returned to the training chamber with the pre-tone baseline period in the novel chamber. During the first 2 min of re-exposure to the training context, GSK3β KI mice froze more than WT mice (t(18)=2.87, P<0.01) (Figure 7d). Although discrimination between the novel chamber and training chamber was evident in both groups (GSK3β KI, t(9)=4.25, P<0.01; WT, t(14)=3.86, P<0.01) there was a clear transfer of fear in the GSK3β KI mice (Figure 7d). GSK3β KI mice spent more time freezing immediately after introduction to the novel context than WT mice (t(9)=2.53, P<0.05) (Figure 7d). Thus, the enhanced fear conditioning observed in GSK3β KI mice is concurrent with an increased generalization of the fear conditioning experience.

Persistence of conditioned fear expression is also a feature of stress-associated disorders such as PTSD (Wessa and Flor, 2007). To examine the persistence of conditioned fear in GSK3β KI mice freezing was analyzed during each minute of the 3-min tone block in the novel chamber, as well as the final minute of the test following termination of the tone. The tone elicited substantial freezing above and beyond that elicited by the novel context alone (Figure 7e). GSK3β KI and WT mice froze similarly during the first 2 min of tone presentation. However, while the tone fear decayed rapidly in WT mice, fear was markedly persistent in GSK3β KI mice (t(18)=3.62, P<0.01) (Figure 7e). This persistent fear was even evident during the minute after tone was terminated (t(18)=2.22, P<0.05) (minute 4; Figure 7e). Together, these results show that the overall fear conditioning response is highly amplified and prolonged in GSK3β KI mice, a compelling phenotype given the focal hippocampal neurodegeneration observed GSK3β KI brains.

Discussion

Although GSK3 was identified almost three decades ago, only during the last few years has it become one of the most prominent therapeutic targets, primarily in the areas of neurological diseases and cancer, as shown by the interest of the pharmaceutical industry and the number of published patents (Palomo and Martinez, 2016). Most approaches to regulating GSK3 activity are based on the best-characterized mechanism for GSK3 inactivation through Ser9 (GSK3β) or Ser21 (GSK3α) phosphorylation by Akt (Stambolic and Woodgett, 1994; Cross et al, 1995). However, we have previously shown that GSK3β, but not GSK3α, can also be inactivated by phosphorylation at the C-terminal Ser389(mouse)/Thr390(human) by p38 MAPK, and the magnitude of inhibition by this mechanism is comparable to Akt-mediated phosphorylation of Ser9 (Thornton et al, 2008). Interestingly, in contrast to Ser9 phosphorylation, Ser389/Thr390 phosphorylation of GSK3β is detectable only in specific tissues, with the brain containing the highest level of phospho-Ser389 GSK3β (Thornton et al, 2008). Our findings here show that GSK3β Ser389 phosphorylation has a distinct role in restraining GSK3β activity and maintaining neuronal viability that is independent of Ser9 phosphorylation. While GSK3β Ser9 phosphorylation is detected in both the cytoplasm and the nucleus, GSK3β Ser389 phosphorylation is primarily nuclear. Ser9 phosphorylation and Ser389 phosphorylation are detected in different pools of GSK3β and Ser9 phosphorylation cannot compensate for the loss of Ser389 phosphorylation in the brain. The neurodegenerative phenotype found in Ser389Ala GSK3β KI mice has not been reported for Ser9Ala GSK3β KI mice (Hongisto et al, 2008; Eom and Jope, 2009). Thus, GSK3β Ser389 and Ser9 inhibitory phosphorylation regulate distinct pools of GSK3β that control different signaling pathways in the brain.

The fact that Ser389 and Ser9 phosphorylation target different physical pools of GSK3β suggests that they respond to different stimuli. Our recent studies have revealed that inactivation of GSK3β by phosphorylation on Ser389 is specifically triggered by DSB generated by external stimuli or endogenous processes (eg, V(D)J recombination of T-cell receptor genes in thymocytes) (Thornton et al, 2016). This explains why phospho-Ser389 GSK3β was not detected in most tissues under normal physiological conditions (Thornton et al, 2008). Although DNA DSB are difficult to detect in the normal brain it was believed that DSB were generated but a highly efficient DNA repair system prevented DSB from accumulating in the brain (McKinnon, 2009). A number of recent studies have provided strong evidence for the existence of DSB in the brain. Normal neuronal activity results in the generation of DSB although the mechanism of DSB induction was not examined (Suberbielle et al, 2013). Moreover, a subset of early-response genes in neurons requires DSB formation for their expression that is critical for the experience-driven changes associated with memory and learning (Madabhushi et al, 2015). It has been shown that neural stem/progenitor cells contain recurrent DSB clusters within long neural genes (Wei et al, 2016). Here we show that in the region where we detect enriched dark neurons in GSK3β KI mice, phospho-Ser389 GSK3β co-localizes with DSB marker γH2AX. It is possible that cells within the regions where dark neurons are detected in the GSK3β KI mice may define a circuit that contains cells more prone to DSB, thus more sensitive to failure to inactivate GSK3β through phosphorylation on Ser389. This could explain why only specific regions and not all neurons are affected in the GSK3β KI mice. In the context of the stress-associated fear conditioning phenotype that we show here, it is interesting that exposure to stress hormones can also lead to DSB accumulation (Hara et al, 2011). Thus, the constitutive phosphorylation of GSK3β on Ser389 in the brain is likely maintained by the ongoing generation of DSB in neurons to attenuate the abundant GSK3β activity present in the brain.

Normally DSB are associated with cell death. However, we have shown that phosphorylation of GSK3β at Ser389 leads to increased cell survival and protecting cells from cell death caused by DSB generated by external stimuli and internal processes (Thornton et al, 2016). While a number of studies have addressed DSB in the brain, none have addressed how cells survive this constant generation of DSB. Here we show that inactivation of GSK3β through phosphorylation at Ser389 can be a novel mechanism to provide protection to prevent neuronal cell death during DSB repair. The increased cell death found in certain regions of the brain in Ser389Ala GSK3β KI mice has not been reported in Ser9Ala GSK3β KI mice, indicating that this is a unique mechanism to control survival in the brain. Interestingly, similar to the death induced by endogenously generated DSB in B cells, the death observed in the brains of GSK3β Ser389Ala KI mice seems to be through necroptosis more than apoptosis. Necroptosis has also been reported to occur in the brain in other studies (Vitner et al, 2014; Zhao et al, 2015). Since DSB have been reported to promote necroptosis (Biton and Ashkenazi, 2011; Tenev et al, 2011), failure to inactivate GSK3β through Ser389 phosphorylation in response to DSB likely contributes to the increased necroptosis found in the brains of GSK3β Ser389Ala KI mice.

Despite the presence of hippocampal neuronal degeneration in the GSK3β Ser389Ala KI mice, no impact on spatial memory or learning was observed. Locomotor activity was also normal. Failure to phosphorylate GSK3β at Ser389 did not affect anxiety or depression-like behavior either. However, we show that GSK3β Ser389Ala KI mice display overgeneralized and persistent fear. This suggests that there may be undetermined specificity to the subset of affected neurons in GSK3β KI mice. Overgeneralization of fear and failure to extinguish responses to fear-associated cues is common in PTSD (Wessa and Flor, 2007). These altered fear responses are also found in rodent stress exposure models used to model this human disorder (Izquierdo et al, 2006; Baran et al, 2009; Hoffman et al, 2014). A brain circuit with the amygdala at its core and involving the hippocampus and several cortical areas (eg, frontal, entorhinal, piriform, infralimbic) is involved in fear learning and extinction. Thus, disruption of this circuit caused by dispersed neurodegeneration in GSK3β KI mice may be the underlying cause of the persistent and generalized fear phenotype. Mice with alanine substitutions blocking N-terminal Ser9 inhibitory phosphorylation of both GSK3β and GSK3α have been shown to have a complex behavior phenotype with an increased anxiety and depression-like phenotype (Polter et al, 2010). The relative contributions of the individual isoforms to this complex behavior phenotype have not been reported.

Taken together our data show that phosphorylation of GSK3β at Ser389 by p38 MAPK promotes the survival of a subset of neurons critical for modulating the conditioned fear response. It is important to note that there is a common polymorphism (around 48% allele frequency) in the human GSK3β gene that results in the expression of a GSKβ protein lacking the Thr390 regulatory site (Rs6438552). The T allele is associated with altered splicing such that exon ll (containing the coding region for Thr390) is deleted (Kwok et al, 2005). This allele is associated with enhanced GSK3β activity and has been linked to neurodegenerative disorders such as Parkinson’s disease, bi-polar disorder, and major depressive disorder (Kwok et al, 2005; Kalinderi et al, 2011; Liu et al, 2012; Lin et al, 2013; Yuan et al, 2013). Therefore, increased expression of GSK3β lacking the Thr390 inhibitory phosphorylation site in humans may impair neuronal survival and contribute to neurological disease. While interest in targeting GSK3 to treat neurological disease has been growing in recent years, the selectivity of this pathway for nuclear GSK3β (in the context of DSB) and the PTSD-like behavior observed make it a more appealing target to manipulate cell survival for specific neurodegenerative diseases. The design of new therapies that specifically target C-terminal Thr390 phosphorylation of nuclear GSK3β could facilitate the development strategies with potentially greater specificity.

Funding and disclosure

This work was supported by NIH grant R01 AI051454 (to MR) and the Consolider grant CSD2010-00045 from the Spanish Ministerio de Economia y Competitividad (MINECO) (to ARN). The authors declare no conflict of interest.