Phosphatidylinositol 3-kinase (PI3K) has been implicated in synaptic plasticity and other neural functions in the brain. However, the role of individual PI3K isoforms in the brain is unclear. We investigated the role of PI3Kγ in hippocampal-dependent synaptic plasticity and cognitive functions. We found that PI3Kγ has a crucial and specific role in NMDA receptor (NMDAR)-mediated synaptic plasticity at mouse Schaffer collateral–commissural synapses. Both genetic deletion and pharmacological inhibition of PI3Kγ disrupted NMDAR long-term depression (LTD) while leaving other forms of synaptic plasticity intact. Accompanying this physiological deficit, the impairment of NMDAR LTD by PI3Kγ blockade was specifically correlated with deficits in behavioral flexibility. These findings suggest that a specific PI3K isoform, PI3Kγ, is critical for NMDAR LTD and some forms of cognitive function. Thus, individual isoforms of PI3Ks may have distinct roles in different types of synaptic plasticity and may therefore influence various kinds of behavior.


PI3K is a dual-specificity kinase that phosphorylates both lipids and proteins in numerous intracellular signaling events1,2. The lipid kinase PI3K produces the phospholipid second messengers phosphatidylinositol-3,4,5-trisphosphate [PI(3,4,5)P3] and phosphatidylinositol-3,4-bisphosphate [PI(3,4)P2], whereas the protein kinase PI3K catalyses serine-threonine phosphorylation. Once activated by various cell-surface receptors, PI3K modulates important cellular functions, affecting cell survival, proliferation, transformation, migration and adhesion3,4. PI3Ks are grouped into three classes (I, II and III) on the basis of their structural and biochemical features, and, of these classes, class I PI3Ks have been the most extensively studied to date. Class I PI3Ks are heterodimers of four 110-kDa catalytic subunits and two distinct families of regulatory subunits, categorized further into class IA and class IB. The class IA PI3Ks (PI3Kα, PI3Kβ and PI3Kδ) are dimers of p110α, p110β or p110δ catalytic subunits and p50–55/p85 regulatory subunits, and are usually activated by a receptor tyrosine kinase. The class IB PI3Ks (PI3Kγ) are dimers of the p110γ catalytic subunit (encoded by Pik3cg) and either p101 or p84 (also known as p87) subunits, and are usually activated by the βγ subunit of G protein–coupled receptors (GPCRs)5,6,7. Notably, unlike other class I PI3Ks, which are ubiquitously expressed, PI3Kγ is only detected in the immune system, cardiovascular system and brain8. In the cardiovascular system, PI3Kγ is involved in cardiac contractility by regulating sarcoplasmic reticulum Ca2+ cycling in the heart9. In the immune system, PI3Kγ is expressed in leukocytes and has a key role in rheumatoid arthritis10,11. Indeed, PI3Kγ has become a major focus of interest as it is a promising drug target for the treatment of inflammatory disease and cardiovascular therapy10. Given its neuronal localization, it is also very likely that this kinase has a specific role in the brain. However, its central functions are unknown.

Mounting evidence has revealed that PI3K has a pivotal role in various neural functions, such as synaptic plasticity, and in a number of brain areas, such as hippocampus and amygdala. For instance, PI3K has been implicated in the induction, maintenance and expression of either NMDAR- or group I metabotropic glutamate receptor 1 (mGluR1)-dependent long-term potentiation (LTP) in the hippocampus12,13,14,15,16. In the amygdala, PI3K is activated following tetanic stimulation of the cortico-lateral amygdala, leading to the induction of LTP17. In addition, PI3K is required for various forms of LTD in the hippocampus18,19,20,21. Selective activation of mGluR1 induces LTD in CA1 by recruiting the PI3K-Akt-mTOR signaling pathway18, and neuropeptide orexins also cause LTD via both GPCR and PI3K activation21. Although PI3K has not been found to be involved in homosynaptic NMDAR LTD, insulin induces NMDAR LTD via PI3K signaling20, and PI3K modulates synapse-specific LTD at Schaffer collateral–commissural (SC-CA1) synapses19. Consistent with these observations, PI3K is also involved in a myriad of brain functions, such as learning and memory. Activation of PI3K modulates spatial memory, inhibitory avoidance conditioning, contextual fear memory retrieval and extinction in the hippocampus22,23,24. In the amygdala, fear conditioning causes activation of PI3K, and this activity is necessary for fear memory consolidation17.

The question of how PI3K is involved in such diverse forms of synaptic plasticity and related cognitive and emotional functions remains unanswered. One possible explanation is that different isoforms of PI3K have specific functions in synaptic plasticity and behaviors. This conjecture is further strengthened by the observation that each isoform of class I PI3Ks has different roles in a vast array of cellular functions3, each with its own distinctive signaling5,6 and varying bodily distribution8,25. To determine whether there may be isoform-specific functions of PI3K in neuronal function, we employed both genetic and pharmacological approaches to investigate the role of PI3Kγ in the hippocampus. We found that PI3Kγ is required specifically for NMDAR LTD at SC-CA1 synapses and that its deficiency alters particular aspects of cognitive processing in the brain, such as behavioral flexibility.


PI3Kγ mRNA and protein are expressed in mouse brain tissue

Before examining the potential functions of PI3Kγ in hippocampal synaptic plasticity and behavior, we first investigated whether PI3Kγ mRNA and protein exist in mouse brain, as there have been conflicting reports about the presence of PI3Kγ in nervous tissue25,26. We used Pik3cg−/− mice, in which the p110γ catalytic subunit of PI3Kγ is disrupted throughout the body27. The Pik3cg−/− mice were generally healthy and indistinguishable from wild-type littermates in overall phenotype. In addition, no anomalies were observed in the gross morphology of Pik3cg−/− brain (Supplementary Fig. 1a). Using reverse transcription PCR, we first verified that p110γ mRNA existed in brain tissues of wild-type mice and that it was eliminated in Pik3cg−/− mice (Supplementary Fig. 2a). Similarly, p110γ protein was observed in wild-type mice and was ablated in Pik3cg−/− mice (Supplementary Fig. 2b), whereas p110α expression remained intact.

To determine the localization of p110γ, we performed a subcellular fractionation using total hippocampi from wild-type mice. Notably, p110γ was mainly observed and enriched in the synaptosomal fraction covering both presynaptic and postsynaptic compartments, whereas p110α was present mainly in S2 (the crude cytosolic fraction) (Supplementary Fig. 2c). We also found that mRNAs of the two regulatory subunits of PI3Kγ, p101 and p84, were expressed in the brain, including the hippocampus (Supplementary Fig. 2d).

Basal synaptic properties are normal in Pik3cg−/− mice

The first step in examining the role of PI3Kγ in hippocampal synaptic plasticity was to identify the electrophysiological features triggered by the genetic deletion of PI3Kγ. We monitored baseline synaptic properties at SC-CA1 synapses from wild-type and Pik3cg−/− mice. Basal synaptic strength, paired-pulse ratio, AMPA receptor (AMPAR)/NMDAR ratio and miniature excitatory postsynaptic currents (mEPSCs) were indistinguishable between wild-type and Pik3cg−/− mice (Fig. 1a–c), indicating that the genetic deletion of p110γ did not affect AMPAR-mediated basal synaptic transmission and presynaptic transmitter release. We also found that neuronal intrinsic excitability did not differ between wild-type and Pik3cg−/− mice (Fig. 1d).

Figure 1: Synaptic and intrinsic properties of CA1 neurons in wild-type and Pik3cg−/− mice.
Figure 1

(a) Input-output relationship in wild-type (WT) and Pik3cg−/− mice (six slices from six mice for wild type; six slices from four mice for Pik3cg−/−). Scale bars represent 150 pA and 20 ms. (b) Paired-pulse ratio at SC-CA1 synapses (six slices from five mice for wild type; six slices from five mice for Pik3cg−/−). Scale bars represent 100 pA and 20 ms. (c) Top, recording traces of AMPAR/NMDAR ratio in wild-type and Pik3cg−/− mice (wild type, 143 ± 29%, seven slices from six mice; Pik3cg−/−, 156 ± 18%, seven slices from five mice). Bottom, recording traces of mEPSCs from CA1 neurons in wild-type and Pik3cg−/− mice (wild type: frequency, 1.2 ± 0.3 (Hz); amplitude, 12.2 ± 1.1 (pA); six slices from three mice; Pik3cg−/−: frequency, 0.9 ± 0.2 (Hz); amplitude, 11.6 ± 1.2 (pA); six slices from three mice). Scale bars represent 100 pA and 40 ms for AMPAR/NMDAR ratio, 10 pA and 1.5 s for mEPSCs. (d) Action potential responses to fixed current injections in hippocampal CA1 pyramidal neurons in wild-type and Pik3cg−/− mice (seven slices from six mice for wild type; seven slices from five mice for Pik3cg−/−). Scale bars represent 40 mV and 200 ms. Error bars represent mean ± s.e.m.

We then investigated NMDAR-mediated responses and found that both NMDAR-mediated current-voltage and input-output relationships were normal in Pik3cg−/− mice (Supplementary Fig. 3a,b). We also investigated the GluN2B receptor–mediated component of the synaptic current, as distinct GluN2 subunits may determine the polarity of hippocampal synaptic plasticity28,29. We found that the GluN2B receptor–mediated component of EPSCs, identified by the bath-application of a selective GluN2B inhibitor Ro 25-6982 (5 μM), was not different between genotypes (Supplementary Fig. 3c,d). Thus, these data suggest that the ablation of Pik3cg does not affect NMDAR-mediated synaptic transmission at SC-CA1 synapses.

NMDAR LTD is selectively impaired in Pik3cg−/− mice

We sought to investigate the physiological role of PI3Kγ in hippocampal synaptic plasticity. To this end, we examined a variety of synaptic plasticity models and, most notably, NMDAR LTD was absent in Pik3cg−/− mice (Fig. 2a). To determine whether the NMDAR LTD deficit in Pik3cg−/− mice would be replicated at the single neuronal level, we recorded from individual CA1 pyramidal neurons using whole-cell patch-clamp recording. In parallel with the field recordings, NMDAR LTD was absent in Pik3cg−/− mice (Fig. 2b) and the LTD induced by low-frequency stimulation (LFS, 300 pulses at 1 Hz, clamped at −40 mV) under our recording conditions was dependent on NMDAR activation (Supplementary Fig. 4a).

Figure 2: NMDAR LTD is absent in Pik3cg−/− mice.
Figure 2

(a) NMDAR LTD at SC-CA1 synapses in wild-type and Pik3cg−/− mice (wild type, 83 ± 2%, 13 slices from eight mice; Pik3cg−/−, 96 ± 5%, 11 slices from nine mice; P < 0.05). (b) NMDAR LTD from whole-cell recording in wild-type and Pik3cg−/− mice (wild type, 76 ± 4%, six slices from four mice; Pik3cg−/−, 109 ± 7%, five slices from four mice; P < 0.01). (c) mGluR LTD in wild-type and Pik3cg−/− mice (wild type, 75 ± 4%, eight slices from six mice; Pik3cg−/−, 79 ± 4%, eight slices from five mice). (d) Depotentiation in wild-type and Pik3cg−/− mice (wild type, 111 ± 3%, five slices from three mice; Pik3cg−/−, 105 ± 2%, three slices from two mice). (e) LTP in wild-type and Pik3cg−/− mice (wild type, 133 ± 4%, six slices from two mice; Pik3cg−/−, 130 ± 2%, seven slices from two mice). Scale bars depict 1 mV and 30 ms for slice field recording, 100 pA and 40 ms for whole-cell recording. Statistical analysis between two groups was performed by comparing the average amplitude of responses over a 5-min period (75–80 min for field recording, 50–55 min for whole-cell recording). Statistical significance was determined using two-tailed unpaired Student's t test.

We then looked into other forms of synaptic depression and found that both mGluR LTD, induced by bath application of the group I mGluR agonist (S)-3,5-dihydroxyphenylglycine (DHPG, 100 μM), and depotentiation were normal in Pik3cg−/− mice, suggesting that PI3Kγ is not associated with all forms of synaptic depression (Fig. 2c,d). We also found that LTP (900 pulses at 10 Hz, 100 pulses at 40 Hz and 100 pulses at 100 Hz) and late-phase LTP were intact in Pik3cg−/− mice (Fig. 2e and Supplementary Fig. 4b–d). Taken together, these results indicate that NMDAR LTD is specifically affected in Pik3cg−/− mice, suggesting that PI3Kγ is important for NMDAR LTD at hippocampal SC-CA1 synapses.

NMDAR LTD is attenuated by a PI3Kγ selective inhibitor

Recently, a PI3Kγ selective inhibitor AS-605240 (5-quinoxalin-6-ylmethylene-thiazolidine-2,4-dione) was developed and used to establish the distinct role of PI3Kγ among class I PI3Ks in inflammatory and autoimmune disorders30. To further examine the requirement of PI3Kγ in NMDAR LTD, we used this inhibitor to pharmacologically suppress PI3Kγ. First, we found that AS-605240 (100 nM) did not alter basal synaptic transmission at SC-CA1 synapses (Supplementary Fig. 4e). Next, we examined the effect of AS-605240 on NMDAR LTD at SC-CA1 synapses. Consistent with the deficit we found in NMDAR LTD from Pik3cg−/− mice, bath application of AS-605240 (100 nM) also eliminated NMDAR LTD (Fig. 3a), demonstrating that the impairment of NMDAR LTD cannot be attributable to a nonspecific effect of genetic interference. Moreover, whole-cell recording verified this finding, revealing that NMDAR LTD at the single neuronal level was prevented by treatment with AS-605240 (Fig. 3b). To further delve into the specificity of the NMDAR LTD deficit by acute inhibition of PI3Kγ, we carried out mGluR LTD, depotentiation and LTP (100 pulses at 100 Hz) in the presence of AS-605240. Consistent with our results in Pik3cg−/− mice (Fig. 2c–e), mGluR LTD, depotentiation and LTP were not affected by treatment with AS-605240 (Fig. 3c–e). PI3Kγ was also expressed in the cerebellum, where mGluR LTD is a prominent form of synaptic plasticity31. Thus, we examined whether cerebellar LTD can be affected by AS-605240. With bath application of AS-605240, mGluR LTD was normally induced, indicating that PI3Kγ is not required for cerebellar LTD (Supplementary Fig. 4f).

Figure 3: NMDAR LTD is blocked by pharmacological inhibition of PI3Kγ.
Figure 3

(a) The effect of AS-605240 on NMDAR LTD (vehicle, 89 ± 5%, eight slices from six mice; AS-605240, 104 ± 2%, ten slices from five mice; P < 0.05). (b) NMDAR LTD with AS-605240 at single neuronal level (vehicle, 70 ± 8%, eight slices from seven mice; AS-605240, 101 ± 3%, seven slices from six mice; P < 0.01). (c) mGluR LTD with AS-605240 (vehicle, 79 ± 5%, six slices from four mice; AS-605240, 87 ± 2%, four slices from three mice). (d) Depotentiation with AS-605240 (vehicle, 107 ± 3%, six slices from three mice; AS-605240, 108 ± 5%, five slices from three mice). (e) The effect of AS-605240 on LTP (vehicle, 145 ± 9%, nine slices from five mice; AS-605240, 129 ± 6%, seven slices from six mice). (f) Application of AS-605240 during induction phase of LTD (vehicle, 71 ± 13%, seven slices from seven mice; AS-605240, 107 ± 6%, six slices from six mice; P < 0.05). (g) Treatment of AS-605240 after LFS (vehicle, 73 ± 7%, six slices from five mice; AS-605240, 73 ± 9%, five slices from four mice). (h) Postsynaptic infusion of AS-605240 during NMDAR LTD (vehicle, 71 ± 9%, seven slices from seven mice; AS-605240, 105 ± 9%, seven slices from six mice; P < 0.05). Scale bars depict 1 mV and 30 ms for slice field recording, and 100 pA and 40 ms for whole-cell recording.

The molecules involved in synaptic plasticity may be required for different temporal phases of synaptic plasticity, such as induction, maintenance and expression. To dissect the possible roles of PI3Kγ in distinct temporal phases of LTD, we bath applied AS-605240 during the baseline period and then washed it out 5 min after LFS. Notably, bath application of AS-605240 during the induction phase blocked NMDAR LTD. However, the application of AS-605240 5 min after LFS did not affect NMDAR LTD (Fig. 3f,g). Next, to establish the locus of PI3Kγ activation in NMDAR LTD, we inhibited PI3Kγ located postsynaptically by dissolving the inhibitor in the recording pipette solution. We found that postsynaptic infusion of AS-605240 was effective in preventing NMDAR LTD (Fig. 3h), implying that PI3Kγ probably functions postsynaptically. In summary, these results indicate that acute pharmacological inhibition of PI3Kγ blocks NMDAR LTD, but not other types of synaptic plasticity, and that postsynaptic PI3Kγ is involved in the induction phase of NMDAR LTD.

Homosynaptic NMDAR LTD specifically involves the PI3Kγ

Previous studies of PI3K in synaptic plasticity have largely focused on the functional role of PI3K in hippocampal NMDAR LTP12,13,14,17,20,32. In contrast, there are only a few studies that have explored the relationship between PI3K and hippocampal NMDAR LTD19,20,32,33. It has been reported that the broad-spectrum PI3K inhibitor LY294002 (10 μM) does not affect homosynaptic NMDAR LTD19,33. We therefore re-examined the effect of this inhibitor on NMDAR LTD. Consistent with previous reports, the application of LY294002 had no substantial effect on NMDAR LTD (Fig. 4a–d). We also applied a class IA PI3Kα selective inhibitor (3-(4-morpholinothieno[3,2-d]pyrimidin-2-yl)phenol, 100 nM) to determine the effect of isoform-specific blockade of class I PI3K on NMDAR LTD34. In this experiment, the PI3Kα inhibitor did not interfere with NMDAR LTD, which provides further evidence for an isoform-specific role of the class I PI3K in NMDAR LTD (Fig. 4b–d).

Figure 4: Specificity of PI3Kγ in the induction of NMDAR LTD.
Figure 4

(a) Effects of LY294002 on NMDAR LTD (70 ± 7%, 12 slices from ten mice). (b) Effects of the class IA PI3Kα inhibitor on NMDAR LTD (61 ± 5%, nine slices from eight mice). (c) Control experiments with vehicle treatment (64 ± 4%, 12 slices from ten mice). (d) Summary graph and statistical analysis among different inhibitors groups (F2,30 = 0.65, P > 0.5, one-way ANOVA). (e) Two-pathway LTD experiment at SC-CA1 synapses with AS-605240 (control pathway, 97 ± 5%; conditioned pathway, 105 ± 10%; four slices from four mice). (f) Two-pathway LTD experiment at SC-CA1 synapses with class IA PI3Kα inhibitor (control pathway, 84 ± 6%; conditioned pathway, 69 ± 5%; six slices from five mice). Scale bars represent 100 pA and 40 ms. Error bars represent mean ± s.e.m.

It has previously been documented that PI3K regulates the input specificity of NMDAR LTD such that inhibition of PI3K by LY294002 enables LFS to induce heterosynaptic LTD19. Before investigating the possible involvement of PI3Kγ in heterosynaptic LTD, we carried out a two-pathway LTD experiment at SC-CA1 synapses to determine whether LY294002 treatment could induce heterosynaptic LTD at the nonconditioned pathway. In the vehicle-treated group, homosynaptic LTD was readily induced. Also, as reported previously, bath application of LY294002 (10 μM) enabled heterosynaptic LTD at the nonconditioned pathway without affecting LTD at the conditioned pathway (Supplementary Fig. 4g,h). In contrast, when AS-605240 (100 nM) was treated throughout the experiment, LTD was suppressed at the conditioned pathway and the nonconditioned pathway was unaltered (Fig. 4e). This suggests that the ability of LY294002 to permit the induction of heterosynaptic LTD is a result of a different isoform of PI3K. Notably, the PI3Kα-selective inhibitor (100 nM) also enabled heterosynaptic LTD at the nonconditioned pathway while having no effect on the induction of homosynaptic LTD at the conditioned pathway (Fig. 4f). Taken together, these results suggest that class IA PI3Ks are involved in the regulation of heterosynaptic LTD, whereas class IB PI3Kγ is specifically involved in homosynaptic NMDAR LTD.

Recovery of NMDAR LTD impairment in Pik3cg−/− mice

To further substantiate the involvement of PI3Kγ in NMDAR LTD, we determined whether re-introduction of purified mouse p110γ catalytic subunit would be sufficient to rescue the impairment of LTD in Pik3cg−/− mice. To achieve this goal, we infused either purified mouse p110γ or heat-inactivated p110γ into CA1 neurons through a recording pipette. As with the previous results, NMDAR LTD was induced in wild-type mice and was impaired in Pik3cg−/− mice. In addition, protein infusion through a recording pipette did not affect basal synaptic transmission (data not shown).

When p110γ (10 nM) was infused into the CA1 pyramidal neurons of Pik3cg−/− mice, LTD was fully restored. In contrast, heat-inactivated p110γ (10 nM) was ineffective at rescuing the impairment of LTD (Fig. 5a,b). Taken together, these results indicate that the impairment of LTD can be restored by the re-introduction of the catalytic subunit of PI3Kγ in CA1 neurons, confirming the postsynaptic direct involvement of PI3Kγ in hippocampal NMDAR LTD.

Figure 5: Recovery of NMDAR LTD impairment in Pik3cg−/− mice.
Figure 5

(a) The effect of wild-type mouse p110γ administration through a recording pipette into CA1 pyramidal neurons on the impairment of NMDAR LTD in Pik3cg−/− mice (wild type, 71 ± 3%, five slices from four mice; Pik3cg−/−, 102 ± 2%, four slices from three mice; Pik3cg−/− with wild-type p110γ, 76 ± 5%, seven slices from four mice; Pik3cg−/− with heat-inactivated (HI) p110γ, 105 ± 5%, four slices from two mice). (b) Summary graph and statistical analysis among the groups (F3,16 = 12.65, P < 0.001, one-way ANOVA with Tukey's multiple comparison test; wild type versus Pik3cg−/−, P < 0.01; wild type versus Pik3cg−/− with heat-inactivated p110γ, P < 0.01; Pik3cg−/− versus Pik3cg−/− with wild-type p110γ, P < 0.01; Pik3cg−/− with wild-type p110γ versus Pik3cg−/− with HI p110γ, P < 0.05). *P < 0.05, **P < 0.01. Error bars represent mean ± s.e.m.

Signaling mechanisms involved in PI3Kγ-mediated NMDAR LTD

Previous studies of other cell types have suggested that PI3Kγ is mainly activated by βγ subunit of G proteins following activation of GPCRs, and so we looked for any involvement of the βγ subunit in NMDAR LTD at SC-CA1 synapses. Hippocampal slices were incubated overnight with or without pertussis toxin (PTX, 5 μg ml−1) before the experiment. However, the treatment of PTX did not lead to the inhibition of NMDAR LTD (Supplementary Fig. 4i). We also bath applied N-ethylmaleimide (NEM, 200 μM) during the baseline period, which is in the range of concentrations previously shown to inhibit PTX-sensitive G proteins in slice preparations35. Consistent with the results of the PTX experiment, NEM application did not affect the induction of NMDAR LTD (Supplementary Fig. 4j). These results were surprising, as βγ subunits of G proteins have been reported to be an upstream activator of PI3Kγ in other cell types. Together, our observations suggest that βγ subunits of G proteins are not involved in NMDAR LTD at SC-CA1 synapses. It is also unlikely that βγ subunit is an upstream molecule activating PI3Kγ on NMDAR stimulation.

What is the biochemical signaling pathway in which PI3Kγ is involved in hippocampal NMDAR LTD? Previous work has shown that glycogen synthase kinase-3β (GSK-3β) regulates the induction of hippocampal NMDAR LTD and, in this regard, GSK-3β activity is regulated by the PI3K/Akt pathway32. Accordingly, we first set out to determine whether the activation of GSK-3β by NMDAR is affected in Pik3cg−/− mice (Fig. 6a,b). As seen previously32,36, dephosphorylation of the Ser9 residue (that is, activation of GSK-3β) was observed in wild-type mice on NMDAR stimulation. In addition, consistent with previous work, the same stimulation tended to induce a decrease in Akt phosphorylation on Thr308, inhibiting Akt activity33, although the level of this decrease did not reach statistical significance (P = 0.1286). However, we found that treatment of NMDA still caused a reduction in the phosphorylation of Ser9 of GSK-3β and of Thr308 of Akt in the hippocampal CA1 region of Pik3cg−/− mice (Fig. 6a,b). These findings indicate that activation of the Akt–GSK-3β pathway occurs independently of PI3Kγ, which is consistent with the proposed role of protein phosphatase 1 in this effect33.

Figure 6: Signaling mechanisms involved in PI3Kγ-mediated NMDAR LTD.
Figure 6

(a) A representative sample of western blottings of GSK-3β phosphorylation on Ser9 and Akt phosphorylation on Thr308 by NMDAR activation in wild-type and Pik3cg−/− mice. (b) Quantifications of GSK-3β and Akt phosphorylation by NMDAR activation (phosphorylated divided by total) in wild-type and Pik3cg−/− mice (GSK-3β, n = 3 per group, the effect of NMDA, F1,8 = 54.26, P < 0.001, two-way ANOVA; Akt, n = 3 per group, the effect of NMDA, F1,8 = 5.29, P > 0.05, two-way ANOVA). (c) A representative sample of western blots of p38 MAPK phosphorylation without NMDA (0 min) and with NMDA (100 μM) treatment for 2 min, 5 min and 10 min in hippocampal slices of wild-type and Pik3cg−/− mice. (d) Quantification of p38 MAPK activation (phospho-p38 divided by total p38; n = 6 per group, the effect of genotype, F1,40 = 7.07, P < 0.05, two-way ANOVA).

Several studies have also implicated p38 MAPK in NMDAR LTD via the Rap1 signaling pathway37,38. Furthermore, there is the possibility that active Rap1 directly binds to the catalytic subunit of PI3K, including PI3Kγ39. PI3Kγ, in turn, might activate p38 MAPK during the induction of NMDAR LTD40. To test whether the activity of p38 MAPK is involved in PI3Kγ-mediated NMDAR LTD, we examined the phosphorylation level of p38 MAPK after NMDAR activation in hippocampal slices of wild-type and Pik3cg−/− mice (Fig. 6c,d). Consistent with previous studies37,38, NMDAR activation induced an increase in phosphorylation of p38 MAPK in wild-type mice. However, the activation of p38 was completely abolished in Pik3cg−/− mice (Fig. 6c,d), which implies that PI3Kγ might contribute to NMDAR LTD via the p38 MAPK pathway.

Behavioral flexibility is reduced in Pik3cg−/− mice

In light of this selective physiological deficit and the accompanying biochemical changes, we sought to determine whether the disruption of PI3Kγ might alter any cognitive functions in the brain. In Pik3cg−/− mice, basal locomotion and anxiety level were normal (Supplementary Fig. 5a–c). In addition, the formation of immediate memory to a novel context was comparable between wild-type and Pik3cg−/− mice (Supplementary Fig. 5d). We then used the Morris water maze task to test hippocampal-dependent spatial memory. In this experiment, both wild-type and Pik3cg−/− mice showed similar performances in the training sessions (day 1–5, except day 3) as well as in a probe trial on day 6 (Fig. 7a,b). We then performed reversal learning by moving the hidden platform to the opposite quadrant of the pool, and both genotypes again equally learned the new target location (Fig. 7a,c). However, Pik3cg−/− mice spent more time in the previous target quadrant during the probe trial on day 9 (Fig. 7c). This difference indicates that, although Pik3cg deletion did not affect the acquisition of spatial memory itself, it might lead to impaired behavioral flexibility when the location of the platform is changed during the reversal learning.

Figure 7: Behavioral flexibility is reduced in Pik3cg−/− mice.
Figure 7

(a) Average escape time traveled to the platform (mean ± s.e.m.) in the Morris water maze (n = 11 for Pik3cg−/−, n = 14 for wild type, the effect of genotype, F1,23 = 1.89, P > 0.2, repeated-measures two-way ANOVA). (b) Averages ± s.e.m. for the percentage of time spent in the initial training quadrant (TQ; 57 ± 10% for Pik3cg−/−, 49 ± 9% for wild type, P > 0.05), the opposite quadrant (OQ) and two adjacent quadrants (AQ) during probe trials given on day 6 of the experiment. (c) Averages ± s.e.m. for the percentage for time spent in the new training quadrant (TQ; 39 ± 16% for Pik3cg−/−, 39 ± 8% for wild type, P > 0.9), the previous training quadrant (PQ; 24 ± 4% for Pik3cg−/−, 15 ± 2% for wild type, P < 0.05) and the two adjacent quadrants (AQ) during probe trials given on day 9 of the experiment. (d) Average number of correct choices (mean ± s.e.m.) from wild-type and Pik3cg−/− mice in delayed nonmatch to place T-maze task (n = 8 for Pik3cg−/−, n = 7 for wild type, the effect of genotype, F1,13 = 6.55, P < 0.05, repeated-measures two-way ANOVA).

To investigate whether other forms of behavioral flexibility dependent on hippocampal functions are also affected, we performed the delayed nonmatch to place T-maze task41. In this task, mice learn and memorize new spatial information because the location of rewards alternates between two places on every trial, making the behavioral flexibility necessary for correct choice41. As in the Morris water maze task, learning was relatively reduced in Pik3cg−/− mice, failing to reach a stable plateau in behavioral performance throughout the experiment (Fig. 7d). This difference was not observed in a spatial version of T-maze task in which the location of the food reward was fixed (n = 6 for wild type, n = 5 for Pik3cg−/−), suggesting that it was not a result of a difference in the motivation level between wild-type and Pik3cg−/− mice.

Contextual fear memory is altered in Pik3cg−/− mice

We then trained Pik3cg−/− and wild-type mice for contextual fear conditioning, which is a well-studied memory procedure that is largely dependent on various hippocampal subregions. Unexpectedly, we found that contextual fear memory was reduced in Pik3cg−/− mice (Supplementary Fig. 6a). Notably, however, when we exposed both genotypes to the fear-conditioning chamber 24 h before conditioning (with pre-exposure) so that the training intensity was increased, the disparity in freezing level between wild-type and Pik3cg−/− mice from the previous experiment disappeared (Supplementary Fig. 6b). This result was replicated when we infused AS-605240 (1 mM) into CA1 area of dorsal hippocampus in naive mice 15 min before fear conditioning (Supplementary Fig. 7) and AS-605240 infusion impaired contextual fear memory (Supplementary Fig. 6c). Again, when the pre-exposure was included 24 h before conditioning, AS-605240 did not affect fear conditioning (Supplementary Fig. 6d).

Previous studies have found that PI3K is activated after contextual fear memory retrieval and that the inhibition of PI3K activity blocks fear memory retrieval22. To examine whether PI3Kγ is also involved in this process, we infused AS-605240 15 min before the retrieval test. Infusion of AS-605240 did not alter memory retrieval (Supplementary Fig. 6e). Consistent with a previous report22, however, administration of LY294002 (10 mM) disturbed contextual memory retrieval (Supplementary Fig. 6f). These results suggest that PI3Kγ is not involved in contextual memory retrieval, whereas other class I PI3Ks may be critical for this process. Consistent with a role for a PI3K species other than PI3Kγ in contextual fear memory retrieval, we found that there was an increase in PI3K activity 10 min after contextual fear memory retrieval (assessed by the phosphorylation of Akt on Ser 473) that was insensitive to AS-605240 (no retrieval, 100%; vehicle, 136.0 ± 25.1%; AS-605240, 142.3 ± 34.7%; n = 5).

Finally, we investigated whether the extinction of contextual fear conditioning might be affected in Pik3cg−/− mice. After contextual fear conditioning, we exposed Pik3cg−/− and wild-type mice to the same context for 7 d in the absence of electric shock to induce fear memory extinction. In this protocol, however, extinction of contextual fear memory was similar between the two genotypes and Pik3cg−/− mice learned to dissociate the fear with the context (Supplementary Fig. 6g).


We investigated the role of PI3Kγ, the neural function of which has hitherto been unexplored, in hippocampal synaptic plasticity and cognition. Using genetic and pharmacological manipulations, we discovered that PI3Kγ was selectively required for NMDAR LTD, but was not necessary for other forms of plasticity. In addition, the impairment of NMDAR LTD in Pik3cg−/− mice was accompanied by an altered activation of p38 MAPK on NMDAR activation. Furthermore, the deficit in NMDAR LTD was associated with reductions in behavioral flexibility, suggesting that PI3Kγ may serve to mediate certain types of cognitive functions in the brain.

PI3Kγ and NMDAR LTD at hippocampal SC-CA1 synapses

We verified that p110γ mRNA and protein were present in various brain regions and confirmed the selective ablation of p110γ mRNA and protein in Pik3cg−/− mice. p110γ was mainly detected in the crude synaptosomal membranes (P2) and purified synaptosomal fractions, whereas p110α existed mostly in the crude cytosolic S2 fraction. Similarly, differential subcellular location of each PI3K isoform has been reported in cultured mouse sympathetic superior cervical ganglia and dorsal root ganglion neurons25. The exact subcellular locations of PI3K isoforms in our study differed from those reported previously25, which is likely a result of the different types of nervous tissues that were used. A common theme, however, was that isoforms of PI3K have restricted regions of localization, making it tempting to speculate that these different localizations may also underlie the distinct physiological functions of PI3Ks in neurons. Hence, the restricted expression of PI3Kγ may offer some clues to the specific role of this kinase in the brain.

In hippocampal physiology, we discovered that NMDAR LTD was selectively impaired in Pik3cg−/− mice. Consistent with this result, pharmacological inhibition of PI3Kγ reproduced this result, further implicating the highly specific role of PI3Kγ in NMDAR LTD. Furthermore, PI3Kγ acted postsynaptically and was critical in the induction, but not the expression, phase of NMDAR LTD. Most notably, the infusion of wild-type p110γ into CA1 pyramidal neurons was sufficient to rescue the impairment of NMDAR LTD in Pik3cg−/− mice. Prior to our findings, the evidence for a role of any isoform of PI3Ks in NMDAR LTD was scarce. One study suggested that PI3K is involved in insulin-induced regulation of NMDAR-mediated LTD, although the isoform involved was not identified20. In apparent contrast with our findings, PI3K has been reported to be not critical for homosynaptic NMDAR LTD at the SC-CA1 pathway, but instead, PI3K may have a modulating role in the input-specific induction of NMDAR LTD19. Similarly, the inhibition of PI3K does not affect NMDAR LTD in the hippocampus32,33. We found that the selective PI3Kγ inhibitor suppressed homosynaptic NMDAR LTD, whereas, consistent with these previous reports, the broad-spectrum PI3K inhibitor LY294002 was ineffective at blocking NMDAR LTD. Moreover, although both the PI3Kα-specific inhibitor and LY294002 could enable heterosynaptic LTD, the PI3Kγ inhibitor did not reveal heterosynaptic LTD. The most likely explanation for this discrepancy is that only PI3Kγ is involved in the induction of NMDAR LTD. Consistent with this idea, LY294002 inhibits class IA PI3K (PI3Kα, PI3Kβ and PI3Kδ) much more potently than the class IB PI3K (PI3Kγ) (half maximal inhibitory concentration values are 0.73, 0.31, 1.06 and 6.60 μM for PI3Kα, PI3Kβ, PI3Kδ and PI3Kγ, respectively). It is therefore likely that LY294002 was applied at an insufficient concentration to block the activity of PI3Kγ19,32,33.

It seems likely that a class IA PI3K isoform(s) is involved in regulating heterosynaptic LTD18 and in the LTP-induced inhibition of LTD32,33, whereas the class IB isoform PI3Kγ is specifically involved in the induction of NMDAR LTD. Thus, distinct PI3K isoforms may have diametrically opposed roles in NMDAR LTD. Taken together with the role of PI3K, of unknown isoform(s), in LTP12,13,14, these findings highlight the complex roles of PI3K in synaptic plasticity.

Signaling pathways in PI3Kγ-mediated NMDAR LTD

How does PI3Kγ induce NMDAR LTD and which signaling pathway mediates this process in the hippocampal SC-CA1 pathway? With the above discussion in mind, we found that the activation of GSK-3β and the inhibition of Akt on NMDAR stimulation, which has been shown to be critical for NMDAR LTD in hippocampal SC-CA132, were not changed in Pik3cg−/− mice. This is fully consistent with a class IA PI3K being involved in the inhibition of NMDAR LTD, via activation of Akt leading to inhibition of GSK-3β, and a distinct role for PI3Kγ in the induction of NMDAR LTD via an independent pathway. A body of evidence has shown that Rap1–p38 MAPK signaling functions to regulate AMPAR removal during hippocampal NMDAR LTD37,38. According to this hypothesis, NMDARs can selectively activate p38 MAPK by means of Rap1, and p38 MAPK, in turn, triggers AMPAR removal. If this is the case, PI3Kγ might serve to activate p38 MAPK during the induction of NMDAR LTD, as PI3Kγ can be activated by Rap1 through its Ras-binding domain and PI3Kγ has been reported to stimulate various MAPKs, such as p38 MAPK, Jun–N-terminal kinase and extracellular signal–regulated kinase, independent of phosphoinositide signaling8,39,40,42. Consistent with this view, the increase in p38 MAPK phosphorylation on NMDAR stimulation was evident in wild-type mice, but was abolished in Pik3cg−/− mice, implicating p38 MAPK as a potential downstream target of PI3Kγ. Undoubtedly, further work will be required to dissect the detailed signaling pathways of PI3Kγ in the induction of NMDAR LTD.

How might PI3Kγ, unlike other class I PI3Ks, specifically regulate NMDAR LTD at hippocampal SC-CA1 synapses and which upstream regulators link NMDAR activation with PI3Kγ? PI3Kγ is usually activated by GPCRs via βγ subunits of G proteins in non-neuronal cells. However, we found that the βγ subunit of G proteins was not involved in NMDAR LTD, as neither PTX nor NEM affected NMDAR LTD. For this reason, different upstream molecules must regulate the activity of PI3Kγ after NMDAR stimulation. One possibility is that PI3Kγ can be activated by Rap1 by means of its Ras-binding domain following NMDAR stimulation. Another possibility is that PI3Kγ might form a complex with postsynaptic density proteins such as an NMDAR subunit or NMDAR multiprotein complexes. In fact, several lines of evidence have shown that the regulatory subunit of class IA PI3K p85 is able to interact directly with NMDARs via its SH2 domains binding to phosphotyrosine residues that are present on GluN2B subunits43,44. Whichever scenario is valid, however, it seems likely that the upstream signaling of PI3Kγ may be quite different from the prevailing model that has been established largely in non-neuronal cells.

Behavioral deficits associated with disruption of PI3Kγ signaling

In behavioral studies, the specific loss of hippocampal NMDAR LTD by genetic deletion of PI3Kγ was accompanied by impairments in reversal learning in the Morris water maze task and in a delayed non-match to place T-maze task. Meanwhile, considerable evidence suggests the central role of NMDAR LTD in numerous forms of cognitive functions36,41,45,46,47,48,49. Particularly, in the hippocampus, NMDAR LTD has been reported to be involved in behavioral flexibility41, episodic-like memory49, immediate memory of a novel context45 and novelty-detection in hippocampus46,47. In addition, NMDAR LTD may contribute to the impairment of acute stress-induced spatial memory retrieval48. Consistent with a previous report41, we observed an association between NMDAR LTD and behavioral flexibility. These findings indicate that PI3Kγ-mediated NMDAR LTD might function to diminish previous memory traces, thereby facilitating the updating of new information.

In the case of contextual fear conditioning, fear memory was reduced by both the genetic deletion and pharmacological inhibition of PI3Kγ. But this reduction was recovered when the training intensity was increased by providing pre-exposure 24 h before conditioning, indicating that memory formation in Pik3cg−/− mice could be normal, provided that certain conditions are satisfied. Besides, contextual memory retrieval and extinction were not affected by the inhibition of PI3Kγ. Although the mechanism behind the alteration in fear memory is unclear, it is plausible that the impairment in hippocampal NMDAR LTD might have affected the learning threshold for fear memory formation, thus Pik3cg−/− mice might be impaired in this form of one-experience learning in parallel with behavioral flexibility49.

In conclusion, our results provide, to the best of our knowledge, the first examination of the isoform-specific role of PI3Ks, focusing on the class IB PI3Ks, and PI3Kγ in hippocampal synaptic plasticity and cognitive behavior. Our data suggest that PI3Kγ may have a pivotal role in the induction of hippocampal NMDAR LTD, possibly through the activation of p38 MAPK. In addition, genetic deletion of PI3Kγ altered behavioral flexibility, implying that PI3Kγ has a physiological role in vivo. Future studies are required to investigate the molecular mechanisms connecting NMDAR activation via PI3Kγ to AMPAR endocytosis.



All of the behavioral experiments were performed on male C57BL/6J mice, male Pik3cg−/− and wild-types littermates having C57BL/6J genetic background (backcrossings were executed over ten generations). These experiments were conducted in accordance with the regulations of the Animal Care and Use Committee of Seoul National University.


For electrophysiology, 4–5-week-old mice (male and female) were used. In extracellular recordings, transverse hippocampal slices (400 μm) were prepared and slices were maintained in an interface chamber at 28 °C. The artificial cerebrospinal fluid (ACSF) contained 124 mM NaCl, 2.5 mM KCl, 1 mM NaH2PO4, 25 mM NaHCO3, 10 mM glucose, 2 mM CaCl2 and 2 mM MgSO4, and oxygenated with 95% O2 and 5% CO2. Extracellular field excitatory postsynaptic potentials (fEPSPs) were recorded in stratum radiatum of CA1 using a glass pipette filled with ACSF (1 MΩ). The Schaffer collateral pathway was stimulated every 30 s using bipolar electrodes (MCE-100, Kopf Instruments). fEPSPs were amplified (GeneClamp 500, Molecular Devices) and digitized (PCI-6221, National Instruments) for measurement. Data were monitored and analyzed using the WinLTP program (http://www.winltp.com/). For LTP and LTD, the stimulation intensity was adjusted to give fEPSP slopes of 40% of maximum, and two successive responses were averaged and expressed relative to the normalized baseline. After a stable baseline was recorded, high-frequency stimulation LTP (100 Hz, 1 s) and LFS LTD (1 Hz, 900 s) were induced. In some experiments, 900 stimuli at 10 Hz or 100 stimuli at 40 Hz were delivered. For depotentiation, 100 Hz, 1-s stimulation was first delivered, followed 5 min later by 900 stimuli at 5 Hz. For mGluR LTD, DHPG was bath applied for 15 min after baseline recording.

For whole-cell recordings, coronal hippocampal slices (300 μm) were prepared. The recording pipettes (35 MΩ) were filled with an internal solution containing 130 mM CsMeSO3, 5 mM NaCl, 10 mM HEPES, 4 mM MgATP, 0.3 mM Na3GTP, 0.5 mM EGTA, 1 mM MgCl2 and 5 mM QX-314 (280300 mOsm, pH 7.2 with CsOH). Spiking activity was measured with an internal solution containing 145 mM potassium gluconate, 5 mM NaCl, 10 mM HEPES, 2 mM MgATP, 0.1 mM Na3GTP, 0.2 mM EGTA and 1 mM MgCl2 (280300 mOsm, pH 7.2 with KOH). Pyramidal neurons in CA1 were voltage clamped at −70 mV and EPSCs were evoked at 0.05 Hz. Three successive EPSCs were averaged and expressed relative to the normalized baseline. Picrotoxin (100 μM) was present to block GABAA receptor–mediated currents. Access resistance was 1530 MΩ and only cells with a change in access resistance <20% were included in the analysis. Whole-cell patch recordings were performed using Axopatch 200B (Molecular Devices) and monitored and analyzed using Clampex 9.2 (Molecular Devices). For LFS LTD, 300 pulses (1 Hz, clamped at −40 mV) were delivered 20 to 30 min after formation of whole-cell configuration. NMDAR-mediated EPSCs were recorded in the presence of CNQX (20 μM), picrotoxin (100 μM) and glycine (1 μM), and input-output of NMDAR-mediated EPSCs was obtained at −30 mV. For the analysis of mEPSCs, tetrodotoxin (1 μM) was included in the ACSF. The amplitudes and frequencies of mEPSCs were analyzed with MiniAnalysis (Synaptosoft). Drugs were purchased from Tocris (CNQX, D-AP5, DHPG, glycine, TTX, Ro 25-6981, LY294002), Sigma (picrotoxin) and Calbiochem (AS-605240, PI3Kα inhibitor). For cerebellar LTD, the experiment was performed as described previously50. Whole-cell recordings were obtained from the third to fifth lobules of cerebellar Purkinje neurons and a depolarization-induced LTD protocol was used that consisted of a brief burst of five stimuli of parallel fiber at 100 Hz, the onset coincident with a step depolarization of the Purkinje cell to 0 mV for 75 ms and such pairing repeated 30 times at 2-s intervals. Data were plotted as fEPSP slope and EPSC amplitude, respectively. Statistical analysis between two groups was performed by comparing the average amplitude of responses over a 5-min period (75–80 min for field recording, 50–55 min for whole-cell recording).

Data analysis.

Statistical significance of data was determined using two-tailed unpaired Student's t test, one-way ANOVA and two-way ANOVA.

Subcellular fractionation and western blotting.

Mouse hippocampi were homogenized using a Dounce homogenizer with 0.32 M sucrose solution. The 'total' homogenized suspension was centrifuged at 1,400g for 10 min at 4 °C. The pellet was resuspended with 0.32 M sucrose solution and centrifuged again at 1,400g for 10 min. The resultant supernatant was considered to be 'S1' and the pellet to be 'P1'. The S1 supernatant was centrifuged at 10,000g for 20 min at 4 °C, resulting in a second supernatant (S2) and a second pellet (P2). The P2 pellet (crude synaptosomal fraction) was resuspended in a 0.32 M sucrose solution and layered over a sucrose step gradient (0.85, 1.0, 1.2 M) of approximately 10 ml in a Beckman centrifuge tube. The tube was centrifuged at 82,500g for 2 h. The fraction from the layer between the 1.0 M and 1.2 M sucrose was collected and centrifuged at 17,000g to precipitate the purified synaptosomal fraction. The membranes were incubated with antibodies to β-tubulin (Sigma, 1:10,000), PSD-95 (Affinity BioReagents, 1:2,000) and GAPDH (Ambion, 1:10,000) or diluted in the 5% bovine serum albumin (wt/vol) with antibody to p110γ (Cell Signaling, 1:250) or p110α (Cell Signaling, 1:1,000).

NMDAR activation in hippocampal slices.

Hippocampal slices (400 μm) were incubated in ACSF for 1 h before NMDA application (50 or 100 μM). To measure the change of p38 MAPK phosphorylation, we applied NMDA for 2, 5 and 10 min. For Akt and GSK-3β phosphorylation, we used the CA1 of hippocampal slices, and treated them with NMDA for 3 min. At the indicated times, slices were added to cold lysis buffer (10 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1% NP-40 (vol/vol), 0.1% SDS (wt/vol), pH 7.5) to which protease inhibitor cocktail (Roche) and phosphatase inhibitor cocktail (Roche) were added. Slices were homogenized in a dounce homogenizer before centrifugation at 17,000g for 20 min at 4 °C to remove unsolubilized cells and debris. The samples were subjected to SDS-PAGE and western blotting with antibodies to phospho–p38 MAPK (Cell Signaling, 1:1,000), p38 MAPK (Cell Signaling, 1:1,000), phospho-Akt (Thr308, Cell Signaling, 1:5,000), Akt (Cell Signaling, 1:1,000), phospho–GSK-3β (Ser9, Cell Signaling, 1:5,000) and GSK-3β (Cell Signaling, 1:1,000).


Brain tissues from wild-type and Pik3cg−/− mice were used to make cDNA using random hexamers. We performed PCR for Pik3cg for 32 cycles using the following primers: sense, 5′-CTCTGCCAAAGGAGAACCAG-3′; antisense, 5′-CATCACTGTCCAGCAAGGC-3′. For semiquantitative reverse transcription PCR, oligo-dT primer was used to make cDNAs. PCR was performed for 30 cycles with the following primers sets: p84 (sense, 5′- GGACGGACGGCGGACTTTC-3′; antisense, 5′- GTGGGGCTGTCAGTGTAAATG -3′) and p101 (sense, 5′- AAGCCGGAGGAGC TAGACTC -3′; antisense, 5′- GCAGAGCCCCACTGAATGTC-3′).

Recombinant mouse p110γ protein preparation.

To produce the recombinant mouse p110γ protein, we subcloned the cDNA (Open Biosystems) of the mouse Pik3cg gene into pFLAG-CMV2 (Invitrogen). The constructed plasmid was transfected into HEK293T cell using the calcium phosphate precipitation method. After 40 h, FLAG-tagged p110γ protein in the cell lysate was immunoprecipitated using anti-FLAG M2 affinity gel (Sigma) and eluted from the resin by adding 3× FLAG peptide (Sigma). The desalting was performed using the Amicon Ultra-0.5 50K device (Millipore) to remove 3× FLAG peptide and exchange the buffer containing 25 mM Tris (pH 8.0), 50 mM NaCl, 0.5 mM MgCl2 and 50% glycerol (vol/vol). Heat-inactivated protein was produced by heating mouse p110γ at 95 °C for 30 min.

Open field task.

Mice were exposed for 15 min to the open field box under dim light. The open field was a square opaque white box (40 × 40 × 40 cm) in which mice were monitored with a tracking program (EthoVision 3.1, Noldus).

Exposure to a novel environment.

Mice were placed for 1 h on two consecutive days in the open field task box under dim light in a separate room. The activities of mice (total distance) in the open field were monitored and analyzed by a tracking program (EthoVision 3.1, Noldus).

Elevated plus maze task.

The elevated plus maze was a white Plexiglas maze, elevated 58 cm above the floor. At the beginning of each experiment, mice were placed in the center of the maze under the fluorescent light. All arms were 150 cm × 150 cm (length × width) and 20-cm walls enclosed two opposing arms (closed arms). The other two arms (open arms) had only rails adjacent to the center. Mice were monitored for 5 min with the tracking program (EthoVision 3.1, Noldus).

Morris water maze task and reversal learning.

The water maze was a circular opaque gray tank, 140 cm in diameter, 100 cm in height, filled with water (2223 °C) to a depth of 30 cm. For training, a submerged Perspex platform (circle, 10 cm in diameter) was placed 35 cm away from the edge in a fixed location. The single training session consisted of four trials with four different random starting positions that were equally distributed around the perimeter of the maze in four quadrants. The interval between trials was 60 s. After mounting the platform, animals were allowed to remain there for 20 s, and were then placed in a holding cage for 1 min until the start of the next trial. If the mouse did not find the platform within 60 s, it was placed manually on the platform and returned to a cage after 20 s. The probe trial consisted of a 60-s free-swim period without the platform on day 6. In reversal learning, the platform was placed at the quadrant opposite the location on days 1–5, and the mice were then retrained in four sessions during 2 d (on days 7–8). On day 9, each mouse was subjected to a single 60-s probe trial. For all probe trials, mice were placed in the center of the maze. All trials were recorded and tracked with software (EthoVision 3.1, Noldus).

Delayed nonmatch to place T-maze task.

This task was modified and performed as previously described41. Mice were group-housed and food-deprived by feeding them 75% of average ad libitum daily intake after completing the task. For this task, mice were habituated for 5 min to the apparatus (Plexiglass, long arm = 70 cm × 10 cm, short left and right arms = 50 cm × 10 cm, start box = 10 cm × 10 cm) with one trial a day for two consecutive days with reward. Mice were tested on four trials per day, each trial consisting of two runs: a forced run and a choice run.

Contextual fear conditioning and extinction.

In this experiment, there were two protocols, depending on the existence of pre-exposure 24 h before conditioning. After handling (3 min, 4 d), contextual fear conditioning was tested in a standard chamber (Freeze Frame, Coulbourn). In the pre-exposure protocol, mice were exposed to the fear chamber for 5 min before contextual fear conditioning. After 24 h, mice were placed in the chamber for 5 min. After 2 min they were presented with three unsigned foot shocks (2 s, 2.4 mA, 1 min apart). Mice were returned to the chamber for memory retrieval after 24 h. For memory extinction, the same protocol as described above (with pre-exposure) was used to acquire contextual fear memory. Extinction consisted of two 5 min exposures to the conditioning chamber in the absence of foot shock with 15-min interval for 7 d.

Surgery and drug infusion into the CA1 region.

Prior to surgery, mice were anesthetized with intraperitoneal injection of ketamine (50 mg per kg of body weight) plus xylazine (5 mg per kg) and placed in a stereotaxic apparatus. Two 26-gauge guide cannulae were placed bilaterally. The stereotaxic coordinates for the CA1 regions of the dorsal hippcampi were −2 mm (anterior-posterior), 1.5 mm (lateral) and 1.7 mm (ventral) from bregma. After 2 weeks of recovery, when mice were placed in the chamber for fear conditioning, mice were anesthetized with gaseous isoflurane, and AS-605240 (1 mM, Calbiochem), LY294002 (10 mM, Tocris) and vehicle (10% DMSO in saline) were infused via bilateral cannulae (30-gauge infusion needle) 15 min before fear conditioning or retrieval.


  1. 1.

    et al. PI 3-kinase is a dual specificity enzyme: autoregulation by an intrinsic protein-serine kinase activity. EMBO J. 13, 522–533 (1994).

  2. 2.

    , & Differential regulation of lipid and protein kinase activities of phosphoinositide 3-kinase gamma in vitro. Biol. Chem. 380, 1337–1340 (1999).

  3. 3.

    & Signaling through the lipid products of phosphoinositide-3-OH kinase. Nature 387, 673–676 (1997).

  4. 4.

    & Phosphoinositide kinases. Curr. Opin. Cell Biol. 8, 153–158 (1996).

  5. 5.

    et al. The role of phosphoinositide-3 kinase and PTEN in cardiovascular physiology and disease. J. Mol. Cell Cardiol. 37, 449–471 (2004).

  6. 6.

    & Structure and function of phosphoinositide 3-kinases. Biochim. Biophys. Acta 1436, 127–150 (1998).

  7. 7.

    , , & Signaling through Class I PI3Ks in mammalian cells. Biochem. Soc. Trans. 34, 647–662 (2006).

  8. 8.

    , & PI3Kγ inhibition: towards an 'aspirin of the 21st century'? Nat. Rev. Drug Discov. 5, 903–918 (2006).

  9. 9.

    , , & Phosphoinositide 3-kinase gamma regulates cardiac contractility by locally controlling cyclic adenosine monophosphate levels. Trends Cardiovasc. Med. 16, 250–256 (2006).

  10. 10.

    , , , & Important roles of PI3Kγ in osteoclastogenesis and bone homeostasis. Proc. Natl. Acad. Sci. USA 107, 12901–12906 (2010).

  11. 11.

    et al. PI3Kγ regulates cartilage damage in chronic inflammatory arthritis. FASEB J. 23, 4288–4298 (2009).

  12. 12.

    & Long-term potentiation in dentate gyrus of the rat is inhibited by the phosphoinositide 3-kinase inhibitor, wortmannin. Neuropharmacology 39, 643–651 (2000).

  13. 13.

    et al. Phosphatidylinositol 3-kinase is required for the expression but not for the induction or the maintenance of long-term potentiation in the hippocampal CA1 region. J. Neurosci. 22, 3359–3365 (2002).

  14. 14.

    , , & Phosphatidylinositol 3-kinase regulates the induction of long-term potentiation through extracellular signal–related kinase-independent mechanisms. J. Neurosci. 23, 3679–3688 (2003).

  15. 15.

    , , & Signaling mechanisms mediated by the phosphoinositide 3-kinase/Akt cascade in synaptic plasticity and memory in the rat. Eur. J. Neurosci. 23, 3375–3384 (2006).

  16. 16.

    et al. Persistent transcription- and translation-dependent long-term potentiation induced by mGluR1 in hippocampal interneurons. J. Neurosci. 29, 5605–5615 (2009).

  17. 17.

    et al. A role for the PI-3 kinase signaling pathway in fear conditioning and synaptic plasticity in the amygdala. Neuron 31, 841–851 (2001).

  18. 18.

    & Activation of the phosphoinositide 3-kinase-akt-mammalian target of rapamycin signaling pathway is required for metabotropic glutamate receptor-dependent long-term depression. J. Neurosci. 24, 6352–6361 (2004).

  19. 19.

    et al. Phosphatidylinositol 3 kinase regulates synapse specificity of hippocampal long-term depression. Nat. Neurosci. 5, 835–836 (2002).

  20. 20.

    , , , & Insulin modulates hippocampal activity–dependent synaptic plasticity in a N-methyl-D-aspartate receptor and phosphatidyl-inositol-3-kinase–dependent manner. J. Neurochem. 94, 1158–1166 (2005).

  21. 21.

    et al. Orexins/hypocretins control bistability of hippocampal long-term synaptic plasticity through co-activation of multiple kinases. Acta Physiol. (Oxf) 198, 277–285 (2009).

  22. 22.

    et al. PI3 kinase signaling is required for retrieval and extinction of contextual memory. Nat. Neurosci. 8, 925–931 (2005).

  23. 23.

    et al. Phosphatidylinositol 3-kinase: a molecule mediating BDNF-dependent spatial memory formation. Mol. Psychiatry 8, 217–224 (2003).

  24. 24.

    et al. LY294002, an inhibitor of phosphoinositide 3-kinase given into rat hippocampus impairs acquisition, consolidation and retrieval of memory for one-trial step-down inhibitory avoidance. Behav. Pharmacol. 12, 629–634 (2001).

  25. 25.

    , , , & Differential mRNA expression and subcellular locations of PI3-kinase isoforms in sympathetic and sensory neurons. J. Neurosci. Res. 56, 44–53 (1999).

  26. 26.

    , , , & Tissue distribution and subcellular localization of a G protein–activated phosphoinositide 3-kinase. An immunohistochemical study. Cell Mol Biol. 44, 973–983 (1998).

  27. 27.

    et al. Function of PI3Kγ in thymocyte development, T cell activation, and neutrophil migration. Science 287, 1040–1046 (2000).

  28. 28.

    et al. Role of NMDA receptor subtypes in governing the direction of hippocampal synaptic plasticity. Science 304, 1021–1024 (2004).

  29. 29.

    Plasticity of NMDA receptor NR2B subunit in memory and chronic pain. Mol Brain 2, 4 (2009).

  30. 30.

    et al. Blockade of PI3Kγ suppresses joint inflammation and damage in mouse models of rheumatoid arthritis. Nat. Med. 11, 936–943 (2005).

  31. 31.

    , & Beyond parallel fiber LTD: the diversity of synaptic and nonsynaptic plasticity in the cerebellum. Nat. Neurosci. 4, 467–475 (2001).

  32. 32.

    et al. LTP inhibits LTD in the hippocampus via regulation of GSK3 beta. Neuron 53, 703–717 (2007).

  33. 33.

    et al. A systematic investigation of the protein kinases involved in NMDA receptor–dependent LTD: evidence for a role of GSK-3 but not other serine/threonine kinases. Mol Brain 2, 22 (2009).

  34. 34.

    et al. Synthesis and biological evaluation of 4-morpholino-2-phenylquinazolines and related derivatives as novel PI3 kinase p110α inhibitors. Bioorg. Med. Chem. 14, 6847–6858 (2006).

  35. 35.

    et al. N-ethylmaleimide blocks depolarization-induced suppression of inhibition and enhances GABA release in the rat hippocampal slice in vitro. J. Neurosci. 17, 941–950 (1997).

  36. 36.

    , , & Long-term depression in the CNS. Nat. Rev. Neurosci. 11, 459–473 (2010).

  37. 37.

    , , , & Ras and Rap control AMPA receptor trafficking during synaptic plasticity. Cell 110, 443–455 (2002).

  38. 38.

    et al. Rap2-JNK removes synaptic AMPA receptors during depotentiation. Neuron 46, 905–916 (2005).

  39. 39.

    , & Rap1 signaling: adhering to new models. Nat. Rev. Mol. Cell Biol. 2, 369–377 (2001).

  40. 40.

    , , , & Phosphatidylinositol 3-kinases regulate ERK and p38 MAP kinases in canine colonic smooth muscle. Am. J. Physiol. Cell Physiol. 279, C352–C360 (2000).

  41. 41.

    et al. Transgenic mice lacking NMDAR-dependent LTD exhibit deficits in behavioral flexibility. Neuron 58, 104–117 (2008).

  42. 42.

    , , & The emerging mechanisms of isoform-specific PI3K signaling. Nat. Rev. Mol. Cell Biol. 11, 329–341 (2010).

  43. 43.

    , , & Phosphorylation-dependent interaction of the N-methyl-D-aspartate receptor epsilon 2 subunit with phosphatidylinositol 3-kinase. Genes Cells 4, 657–666 (1999).

  44. 44.

    , , , & Proteomic analysis of NMDA receptor-adhesion protein signaling complexes. Nat. Neurosci. 3, 661–669 (2000).

  45. 45.

    et al. A role in learning for SRF: Deletion in the adult forebrain disrupts LTD and the formation of an immediate memory of a novel context. Neuron 50, 127–143 (2006).

  46. 46.

    & Hippocampal long-term depression and long-term potentiation encode different aspects of novelty acquisition. Proc. Natl. Acad. Sci. USA 101, 8192–8197 (2004).

  47. 47.

    & Novelty acquisition is associated with induction of hippocampal long-term depression. Proc. Natl. Acad. Sci. USA 96, 8739–8744 (1999).

  48. 48.

    et al. Hippocampal long-term depression mediates acute stress-induced spatial memory retrieval impairment. Proc. Natl. Acad. Sci. USA 104, 11471–11476 (2007).

  49. 49.

    et al. Forebrain-specific calcineurin knockout selectively impairs bidirectional synaptic plasticity and working/episodic-like memory. Cell 107, 617–629 (2001).

  50. 50.

    , , , & Long-term depression of mGluR1 signaling. Neuron 55, 277–287 (2007).

Download references


This work was supported by the Creative Research Initiatives Program of the Korean Ministry of Science and Technology and World Class University project. J.-I.K. and H.-R.L. are supported by a BK21 fellowship and Seoul Science fellowship. S.-E.S. and J.B. are supported by a BK21 fellowship. P.H.B. is a Career Investigator of the Heart and Stroke Foundation of Ontario and has support from a Canadian Institutes of Health Research grant (62954). B.-K.K. is a Yonam Foundation Scholar. G.L.C. and M.Z. are WCU International Scholars.

Author information

Author notes

    • Jae-Ick Kim
    •  & Hye-Ryeon Lee

    These authors contributed equally to this work.


  1. National Creative Research Initiative Center for Memory, Department of Biological Sciences, College of Natural Sciences, Seoul National University, Seoul, Korea.

    • Jae-Ick Kim
    • , Hye-Ryeon Lee
    • , Su-eon Sim
    • , Jinhee Baek
    • , Nam-Kyung Yu
    • , Jun-Hyeok Choi
    • , Hyoung-Gon Ko
    • , Yong-Seok Lee
    • , Soo-Won Park
    • , Chuljung Kwak
    •  & Bong-Kiun Kaang
  2. Department of Brain and Cognitive Sciences, College of Natural Sciences, Seoul National University, Seoul, Korea.

    • Su-eon Sim
    • , Clarrisa A Bradley
    • , Sang Jeong Kim
    • , Min Zhuo
    • , Graham L Collingridge
    •  & Bong-Kiun Kaang
  3. Department of Physiology, Seoul National University College of Medicine, Seoul, Korea.

    • Sung-Ji Ahn
    •  & Sang Jeong Kim
  4. Department of Anatomy, College of Medicine, Korea University, Seoul, Korea.

    • So Yoen Choi
    •  & Hyun Kim
  5. Department of Physiology, Faculty of Medicine, University of Toronto, Toronto, Ontario, Canada.

    • Kyoung-Han Kim
    • , Peter H Backx
    •  & Min Zhuo
  6. National Creative Research Initiative Center for Synaptogenesis, Department of Biological Sciences, Korea Advanced Institute of Science and Technology, Daejeon, Korea.

    • Eunjoon Kim
  7. Department of Applied Biology, College of Ecology and Environment, Kyungpook National University, Sangju-si, Kyeongbuk, Korea.

    • Deok-Jin Jang
  8. Department of Anatomy, School of Medicine, Kyungpook National University, Daegu, Korea.

    • Kyungmin Lee
  9. MRC Center for Synaptic Plasticity, School of Physiology and Pharmacology, Bristol, UK.

    • Graham L Collingridge


  1. Search for Jae-Ick Kim in:

  2. Search for Hye-Ryeon Lee in:

  3. Search for Su-eon Sim in:

  4. Search for Jinhee Baek in:

  5. Search for Nam-Kyung Yu in:

  6. Search for Jun-Hyeok Choi in:

  7. Search for Hyoung-Gon Ko in:

  8. Search for Yong-Seok Lee in:

  9. Search for Soo-Won Park in:

  10. Search for Chuljung Kwak in:

  11. Search for Sung-Ji Ahn in:

  12. Search for So Yoen Choi in:

  13. Search for Hyun Kim in:

  14. Search for Kyoung-Han Kim in:

  15. Search for Peter H Backx in:

  16. Search for Clarrisa A Bradley in:

  17. Search for Eunjoon Kim in:

  18. Search for Deok-Jin Jang in:

  19. Search for Kyungmin Lee in:

  20. Search for Sang Jeong Kim in:

  21. Search for Min Zhuo in:

  22. Search for Graham L Collingridge in:

  23. Search for Bong-Kiun Kaang in:


J.-I.K. and H.-R.L. designed, performed and analyzed most of the electrophysiology and behavioral experiments, and wrote the manuscript. S.S., J.B., N.-K.Y., J.-H.C., H.-G.K., Y.-S.L., S.-W.P., C.K., S.-J.A., S.Y.C., H.K., K.-H.K., D.-J.J., K.L. and S.J.K. conducted the biochemical, electrophysiological and behavioral studies. Y.-S.L., P.H.B., C.A.B., D.-J.J., K.L., E.K., M.Z. and G.L.C. aided in the interpretation of data and contributed to editing the manuscript. B.-K.K. supervised the project, designed the experiments and wrote the manuscript.

Competing interests

The authors declare no competing financial interests.

Corresponding author

Correspondence to Bong-Kiun Kaang.

Supplementary information

PDF files

  1. 1.

    Supplementary Text and Figures

    Supplementary Figures 1–7

About this article

Publication history






Further reading