Strictly regulated agonist-dependent activation of AMPA-R is the key characteristic of TAK-653 for robust synaptic responses and cognitive improvement

Agonistic profiles of AMPA receptor (AMPA-R) potentiators may be associated with seizure risk and bell-shaped dose-response effects. Here, we report the pharmacological characteristics of a novel AMPA-R potentiator, TAK-653, which exhibits minimal agonistic properties. TAK-653 bound to the ligand binding domain of recombinant AMPA-R in a glutamate-dependent manner. TAK-653 strictly potentiated a glutamate-induced Ca2+ influx in hGluA1i-expressing CHO cells through structural interference at Ser743 in GluA1. In primary neurons, TAK-653 augmented AMPA-induced Ca2+ influx and AMPA-elicited currents via physiological AMPA-R with little agonistic effects. Interestingly, TAK-653 enhanced electrically evoked AMPA-R-mediated EPSPs more potently than AMPA (agonist) or LY451646 (AMPA-R potentiator with a prominent agonistic effect) in brain slices. Moreover, TAK-653 improved cognition for both working memory and recognition memory, while LY451646 did so only for recognition memory, and AMPA did not improve either. These data suggest that the facilitation of phasic AMPA-R activation by physiologically-released glutamate is the key to enhancing synaptic and cognitive functions, and nonselective activation of resting AMPA-Rs may negatively affect this process. Importantly, TAK-653 had a wide safety margin against convulsion; TAK-653 showed a 419-fold (plasma Cmax) and 1017-fold (AUC plasma) margin in rats. These findings provide insight into a therapeutically important aspect of AMPA-R potentiation.


TAK-653 bound to the LBD of purified recombinant AMPA-R proteins and induced Ca 2+ influx in hGluA1i CHO cells in a glutamate-dependent manner.
We previously reported that structural interference at Ser750 in the channel-closed state of GluA2o LBD might be involved in the molecular mechanisms underlying the lower agonistic effect of AMPA-R potentiators with dihydropyridothiadiazine 2,2-dioxide derivatives and discovered TAK-137, 9-(4-phenoxyphenyl)-3,4-dihydropyrido [2,1-c] [1,2,4]thiadiazine 2,2-dioxide ( Fig. 1A) 16 . Based on the structure and agonistic-effect relationship studies on this novel chemical series, we designed TAK-653, a dihydropyrazinothiadiazine 2,2-dioxide derivative with cyclohexyl group as bulky terminal substituents to induce steric repulsion at Ser750 in GluA2o; the structural bulkiness of cyclohexyl group is higher than that of the terminal phenyl group of TAK-137, suggesting that TAK-653 may exhibit a lower agonistic effect. TAK-653 bound to the intradimer interface formed by the ligand binding core (Fig. 1B). The 3D overlay analysis showed that the peripheral cyclohexyl rings of TAK-653 caused steric interference at Ser750 in the channel-closed state of GluA2o LBD (Toyofuku et al., in preparation). Binding affinity of TAK-653 to the GluA2o LBD was measured by a scintillation proximity assay (SPA) using [ 3 H]-HBT1, a radio-labeled LBD-binding AMPA-R potentiator, and a His-tagged GluA2o LBD protein (His-LBD) 21 . TAK-653 inhibited binding between [ 3 H]-HBT1 and His-LBD with an IC 50 value of 0.26 μM (Fig. 1C). Binding of [ 3 H]-TAK-653 to His-LBD was also measured by SPA. Binding between [ 3 H]-TAK-653 and His-LBD was robustly increased in a glutamate-dependent manner, whereas binding of [ 3 H]-TAK-653 was not detected when another His-tagged protein such as macrophage migration inhibitory factor (MIF) was used as a control (Fig. 1D). TAK-653 did not inhibit the specific binding of [ 3 H]-AMPA to His-LBD, but rather mildly increased [ 3 H]-AMPA binding with EC 50 and E max values of 1.5 ± 0.2 μM and 19.9 ± 3.4%, respectively (Fig. 1E), suggesting that TAK-653 has no binding affinity for the agonist binding site of AMPA-R. These results suggest that TAK-653 selectively binds to the LBD of AMPA-R in a glutamate-dependent manner due to structural interference at Ser750 (GluA2o LBD) in the channel-closed state.
In the functional assays, TAK-653 robustly increased Ca 2+ influx only in the presence of glutamate (3 mM) in hGluA1i CHO cells; the EC 50 was 3.3 μM (Fig. 1F). Ser750 in GluA2o LBD corresponds to Ser743 in GluA1i LBD, thus the introduction of an S743A mutation into GluA1i was expected to reduce the steric interference with the peripheral cyclohexyl rings of TAK-653 and to facilitate the binding of TAK-653 to GluA1i especially when glutamate concentration is low. In fact, the maximum responses of TAK-653 in CHO cells expressing S743A GluA1i were higher than those in CHO cells expressing wild-type GluA1i in a Ca 2+ influx assay at each concentration of glutamate tested (Fig. 1G). Next, to improve the prediction of efficacious concentration for use in humans, we assessed potential of TAK-653 in human and rat GluA1i. In a Ca 2+ influx assay using human and rat GluA1i-expressing CHO cells, TAK-653 did not show species differences; the fold difference in the EC 50 value between rat and human receptors was 1.1 (Fig. 1H). TAK-653 did not show prominent subunit selectivity for homomeric AMPA-R in a Ca 2+ influx assay using CHO cells expressing GluA1-4i and TARP γ-2 or GluA1-4o and TARP γ-2 (Table S1) (Table S2). These observations further suggest that TAK-653 selectively binds to AMPA-R in a glutamate-dependent manner due to structural interference at Ser750 in GluA2o LBD (S743A in GluA1i LBD).
TAK-653 induced Ca 2+ influx, whole-cell currents, and brain-derived neurotrophic factor (BDNF) production in an AMPA-dependent manner in primary hippocampal neurons. In our previous studies, recombinant AMPA-R on hGluA1i CHO cells was less sensitive than physiological AMPA-R on primary neurons in the detection of agonistic effects of AMPA-R potentiators 16,21 . Thus, we next characterized TAK-653 using rat primary hippocampal neurons. A single application of AMPA dose-dependently increased intracellular Ca 2+ levels ( Fig. 2A). Thus, our experimental conditions are suitable for the characterization of AMPA-R potentiators using cultured primary neurons. The response of 5 μM AMPA plus 10 μM HBT1 was defined as 100% in the Ca 2+ influx assay using rat primary hippocampal neurons 21 . TAK-653 robustly increased   16 .
In whole-cell patch clamp recordings using rat primary hippocampal neurons, a single application of AMPA dose-dependently increased AMPA-R-mediated currents with an EC 50 value of 45 μM (Fig. 2C). Thus, AMPA at a low dose of 1 μM was used to characterize the AMPA-R potentiation activity of TAK-653. The response of TAK-653 with AMPA (1 μM) was normalized against currents induced by 1 μM AMPA. TAK-653 dose-dependently augmented AMPA (1 μM)-elicited currents with an EC 50 value of 4.4 μM (Fig. 2D). Next, we assessed agonistic effect of TAK-653. In the absence of agonist, the response of TAK-653 was normalized against currents induced by 100 μM AMPA. As a result, TAK-653 even at 30 μM produced 1.7% of the AMPA (100 μM)-elicited currents, while TAK-137 and LY451646 showed 6.4% and 39%, respectively, response at 30 μM ( Fig. 2E and Table 1) 16,20 . Thus, the agonistic effects of TAK-653 were lower than those of TAK-137 and LY451646 in rat primary hippocampal neurons.
Activation of AMPA-R increased BDNF mRNA levels in the mouse hippocampus 22,23 . We examined the effect of TAK-653 on BDNF production in rat primary hippocampal neurons. TAK-653 robustly increased BDNF protein levels in the presence of a low concentration of AMPA (1 μM) (Fig. 3A). TAK-653 alone slightly increased BDNF protein levels at 1 μM. Next, we investigated the effect of TAK-653 on BDNF mRNA levels in the mouse hippocampus. TAK-653 at 3 and 10 mg/kg, p.o. significantly increased BDNF mRNA levels in the AMPA (3.5 mg/kg, i.v.)-treated mice, while TAK-653 alone did not increase BDNF mRNA levels under these experimental conditions (Fig. 3B). These results demonstrate that TAK-653 stimulates BDNF production through agonist-dependent AMPA-R activation in both in vitro and in vivo.
TAK-653, but not LY451646 or AMPA, improved cognitive functions in multiple domains in rats and working memory in monkeys. The effect of TAK-653 on visual learning and memory was assessed using the novel object recognition (NOR) test in rats. TAK-653 at 0.03, 0.1 and 0.3 mg/kg, p.o. significantly improved the novelty discrimination index (NDI) (Fig. 4A). TAK-653 may enhance visual learning and memory at ≥ 0.03 mg/kg, p.o. in normal rats. In our previous study in rats, LY451646 at ≥ 1 mg/kg, p.o. improved visual learning and memory in the NOR test, although it induced seizure at 10 and 30 mg/kg, p.o. 16 . AMPA induced seizure at 30 mg/kg, i.p. and abnormal behavior such as head turning at 10 mg/kg, i.p. in rats, thus we assessed its effects on cognitive function at 3 mg/kg, i.p. and less. Surprisingly, AMPA at 0.3 to 3 mg/kg, i.p. did not improve cognitive performance in the NOR test (Fig. 4B).
The effect of TAK-653 on MK-801-induced working memory deficit in rats was evaluated using the radial arm maze (RAM) task. MK-801 (0.08 mg/kg, s.c.) disrupted the performance of well-trained rats and TAK-653 at 0.1, 0.3, 3 and 10 mg/kg, p.o. significantly ameliorated the MK-801-induced deficits (Fig. 4C). Thus, TAK-653 may enhance working memory performance over a broad dose range in hypoglutamatergic conditions. By contrast, neither LY451646 at 0.03 to 1 mg/kg, p.o. nor AMPA at 0.1 to 3 mg/kg, i.p. improve working memory in the RAM test (Fig. 4D,E).
We also characterized another AMPA-R potentiator, PF-04958242. PF-04958242 showed significant agonistic effects; PF-04958242 potently produced a Ca 2+ increase in the absence of AMPA in rat primary hippocampal neurons (Fig. S1A). Similar to LY451646, PF-04958242 improved visual learning and memory at ≥ 0.1 mg/kg, p.o. in the NOR test, while PF-04958242 did not improve working memory in the RAM test ( Fig. S1B and S1C).
The effect of TAK-653 on working memory was also assessed using the delayed match-to-sample (DMTS) paradigm in monkeys. C max at 0.06 mg/kg p.o. in fasted monkeys corresponds to that at 0.1 mg/kg p.o. in rats, thus 0.06 mg/kg, p.o. of TAK-653 was used in the monkey study. TAK-653 at 0.06 mg/kg, p.o. significantly increased DMTS accuracy at a 16-s delay interval (Fig. 4F). The significant improvement in accuracy returned to the vehicle level at 48 h later, reflecting that the favorable effect was not attributable to a training effect. The beneficial effect of TAK-653 on task accuracy maintained 24 h after administration probably due to its T max value (8.0 ± 0.0 h) and t 1/2 value (9.4 ± 3.8 h) at 0.03 mg/kg, p.o. in monkeys. TAK-653 may also improve working memory in monkeys at similar plasma concentrations as observed in rats for improving visual learning, recognition memory and working memory.
We investigated the effects of TAK-653 on attention in the rat 5-choice serial reaction time task (5CSRTT). Rats with poor performance could be a useful model of ADHD 24 . Thus, sub-population analyses with a median split of the population into high and poor performing animals, were also performed based on their correct Table 1. Effects on intracellular Ca 2+ level and AMPA-R-mediated currents in the presence or absence of AMPA in rat primary hippocampal neurons. *% of 5 μM AMPA + 10 μM HBT1 response. # % of AMPA (100 μM)-induced steady-state response. EC 50 value was calculated from dose-response curve in the presence of 5 μM AMPA for intracellular Ca 2+ levels or 1 μM AMPA for AMPA-R currents using a nonlinear regression. prior to the trial showed no significant improvement in the whole population, but median split analysis (median value, 51) revealed that TAK-653 significantly increased correct responses and decreased omissions in the poor performing rats (Fig. S2). TAK-653 did not affect the premature responses. Thus, TAK-653 may enhance sustained attention in the poor performing rats. The effect of TAK-653 on sociability deficits was evaluated using the social approach-avoidance test in the poly-I:C mouse, a developmental immune activation animal model of schizophrenia 26,27 . The sniffing index was significantly decreased in vehicle-treated poly-I:C mice compared with vehicle-treated control mice. TAK-653 at 0.3 mg/kg, p.o. significantly improved the sniffing index (Fig. 4G). TAK-653 may ameliorate abnormal social interaction. and memory in NOR test using naïve rats. AMPA (0.3, 1 and 3 mg/kg, i.p.) was administered 0.5 h prior to the acquisition and the retention trials. NDI data were presented as the mean ± SEM (n = 10). # P ≤ 0.05 (versus vehicle-treated group; two-tailed Williams' test). (C-E) Effect of TAK-653 (C), LY451646 (D) or AMPA (E) on working memory in RAM test using rats. At 1.5 h, 1 h or 0 h before administration of vehicle or MK-801 (0.08 mg/kg, s.c.), TAK-653, LY451646 or AMPA, respectively, was administered to rats. Thirty minutes after dosing of MK-801, rats were placed on the maze, and then the entry into the arm was recorded. The mean errors were indicated as the mean ± SEM (n = 6-18). ***P ≤ 0.001 (versus vehicle-vehicle group; Welch's test); # P ≤ 0.05 (versus vehicle-MK-801 group; two-tailed Shirley-Williams' test). (F) Effect of TAK-653 on working memory in DMTS test using monkeys. TAK-653 at 0.06 mg/kg was orally administered to monkeys at 6 h prior to DMTS testing. Each plot at 0, 4, 8 or 16 s interval were presented as the mean ± SEM within 96 trials per session (n = 3) (left). TAK-653 significantly ameliorated DMTS accuracy at 16 s delay interval (right). * P ≤ 0.05 (versus vehicle group; paired t test). (G) Effect of TAK-653 on sociability deficits in the social approach-avoidance test in poly-I:C mouse. TAK-653 at 0.3 mg/kg was orally administered to mice at 1 h prior to test session. Sniffing index data were presented as the mean ± SEM (n = 7). *P ≤ 0.05 (versus control mice group; Student's t test); # P ≤ 0.05 (versus control mice group; Student's t test).

TAK-653 had a low risk of receptor desensitization or sensitization in vivo.
Down-regulation of AMPA-Rs or sensitization of the AMPA-R system following chronic simulation are a concern with AMPA-R activators 28,29 . TAK-653 at 0.3 mg/kg, p.o. improved social interaction in mice (Fig. 4G) and in vivo plasma exposure level at 0.1 mg/kg, p.o. were similar between rats and mice (Table S3), thus we assessed AMPA-R function after 14 days of TAK-653 administration at 0.3 mg/kg p.o. in mice. AMPA-R activation is known to induce the expression of BDNF and growth arrest and DNA-damage-inducible beta (Gadd45b) mRNA 30 . Preadministration of TAK-653 at 0.3 mg/kg, p.o. for 14 days did not affect AMPA-induced BDNF and Gadd45b mRNA expression in the mouse hippocampus (Fig. 5A,B), suggesting a lower risk of receptor desensitization or sensitization after repetitive dosing.
TAK-653 enhanced AMPA-R-mediated synaptic responses more potently than AMPA or LY451646 in prefrontal cortex (PFC) slices. Desensitization of AMPA-R by agonistic effects may be a cause for reduced (or lack of) efficacy or extremely narrow bell-shaped dose-responses in cognitive improvement mediated by AMPA or LY451646; however, this hypothesis is inconsistent with their seizure risk at higher doses. To understand the underlying mechanism of action, we examined AMPA-mediated synaptic responses.
In the presence of bicuculline (20 μM), CGP52422 (10 μM), and APV (50 μM) to block GABA A , GABA B and NMDA receptors, respectively, electrical stimulation of layer I elicited the AMPA-R-mediated polysynaptic www.nature.com/scientificreports/ EPSPs of layer V pyramidal neurons into burst firing while the somatic membrane was maintained at approximately − 65 mV with constant current (Fig. 6A). Bath application of 0.3 to 30 μM TAK-653 for 10 min enhanced the suprathreshold polysynaptic EPSPs with a significant increase in evoked spikes (Fig. 6B) and EPSP duration (Fig. 6C) in a concentration-dependent manner. The enhancing effects of TAK-653 were maintained for at least 10 to 20 min after elimination. By blocking the synaptic responses of AMPA-R with NBQX (10 μM), the effect of TAK-653 completely disappeared (Fig. 6A). Similar to TAK-653, application of TAK-137 at 3 μM also robustly enhanced polysynaptic EPSP duration and evoked spikes (Fig. S3). Compared with TAK-653 and TAK-137, LY451646 at 0.3 to 30 μM induced moderate potentiation of suprathreshold polysynaptic EPSPs with a smaller number of spikes (Fig. 6D) and EPSP duration (Fig. 6E). Following a switch from LY451646 (30 μM) to TAK-653 (10 μM) (Fig. 6F), a clear increase in spike number (Fig. 6G) and EPSP duration (Fig. 6H) were observed, indicating no pronounced desensitization of AMPA-R by LY451646. Both TAK-653 and LY451646 did not increase the membrane potentials of recorded neurons, while AMPA (0.3 μM) increased the membrane potentials. Under these conditions, AMPA reduced polysynaptic EPSP duration and evoked spike number when the membrane was held at − 65 mV with a current injection (Fig. 6I). Thus, TAK-653 might augment the AMPA-R-mediated polysynaptic network interactions more robustly than LY451646 or AMPA.  Table S4). The exposure margins of TAK-653, calculated using the area under the plasma drug concentration-time curve (AUC plasma ) values and plasma C max values, were 1017-fold (AUC plasma ) and 419-fold (plasma C max ), respectively ( Table 2 and S5). Exposure margins of TAK-137 and LY451646 by this protocol were 122-and 4.0-fold (AUC plasma ), respectively, and 42-and 3.4-fold (plasma C max ), respectively. Thus, TAK-653 may have a wider exposure margin than LY451646 in rats.

Discussion
In neuropsychiatric disorders, functional enhancement of AMPA-R has the potential to improve cognitive deficits in several cognitive domains including executive function, attention, and working memory 32 . Enhancement of AMPA-R-mediated neurotransmission may also lead to rapid antidepressant action 9,33-35 . Thus, AMPA-R potentiators could be promising therapeutic drugs for multiple CNS disorders. However, seizure liability and narrow bell-shaped dose-responses might have restricted the development of AMPA-R potentiators as therapeutic drugs. We hypothesized that the agonistic effects of some AMPA-R potentiators are associated with their seizure risks and bell-shaped dose-response effects. In fact, TAK-137, an AMPA-R potentiator with lower agonistic effects than LY451646, showed lower risks of seizure and bell-shaped dose-response 16 . In this study, we asked whether an AMPA-R potentiator with lower agonistic effects than TAK-137 (i.e. an AMPA-R potentiator with virtually no agonistic effect) may have lower seizure risks or if the beneficial effect is lost. To answer this question, we discovered a novel AMPA-R potentiator TAK-653 with extremely lower agonistic effects. [ 3 H]-TAK-653 bound to the intradimer interface formed by the ligand-binding core of AMPA-R subunits in a glutamate-dependent manner. This intradimer interface is known to undergo conformational changes upon glutamate binding 36,37 . Co-crystallization studies suggested that TAK-653 exhibited glutamate-dependent structural interference at Ser750, located in the intradimer interface of GluA2o with larger steric repulsion in the absence of glutamate. In fact, the introduction of a mutation to Ser743 (S743A) in GluA1i (corresponding to Ser750 in GluA2o) to lower the steric interaction between AMPA-R and TAK-653 reduced the glutamate threshold for AMPA-R activation by TAK-653. Thus, like TAK-137, steric interaction at Ser743 in the channel-closed state may be related to the lower agonistic effects of TAK-653 16 . Both TAK-137 and TAK-653 had similar potential in the augmentation of agonist-elicited AMPA-R function, while TAK-653 had a lower agonistic effect than TAK-137 on both Ca 2+ influx and whole-cell currents using primary neurons.
Quite interestingly, like TAK-137, TAK-653 produced a potent cognitive improvement in a wide range of cognitive domains. Neither LY451646 nor AMPA, however, improved working memory in the RAM test and AMPA did not improve recognition memory in the NOR test in rats. The effects of LY451646 in the RAM test were evaluated at around 1 mg/kg, p.o., a dose at which LY451646 produced efficacy in the NOR test. Therefore, brain exposure of LY451646 under the conditions for the RAM test was high enough to explore its therapeutic potential for working memory. The effects of AMPA were investigated at a broader range of doses from 0.3 to 3 mg/kg, i.p. because AMPA produced abnormal behavior at 10 mg/kg, i.p. Note that all three compounds induced convulsions at higher exposure, thus AMPA-R activation by all of them was clear. Therefore, facilitation of physiologically active AMPA-Rs, but not increasing the number of active AMPA-Rs by stimulating resting receptors, is key to producing potent cognitive improvements through AMPA-R activation. In line with this Table 2. Exposure margin against seizure after acute treatment in rats (plasma C max and AUC plasma ). www.nature.com/scientificreports/ hypothesis, PF-04958242, with prominent agonistic activity for AMPA-R on rat primary hippocampal neurons, improved visual learning and memory but it did not improve working memory. In isolated primary neurons, AMPA dose-dependently increased intracellular Ca 2+ levels and AMPA-R-mediated currents (Fig. 2). Under these conditions, TAK-653 and LY451646 showed similar activity in the potentiation of AMPA-elicited Ca 2+ increases and TAK-653 was 2.3 times less potent than LY451646 in the potentiation of AMPA-elicited currents (Table 1 and fig. S4). To our surprise, TAK-653 produced more robust potentiation of the AMPA-R component of the synaptic responses, especially the spike number, compared with LY451646 in the PFC neuronal network (Fig. 6). TAK-653 could enhance polysynaptic EPSPs after LY451646 exposure, thus it is unlikely that LY451646 reduced the number of AMPA-Rs by rapid desensitization during the 10 min for slice preparation. Furthermore, AMPA (0.3 μM) reduced polysynaptic EPSP duration and evoked spikes. Most neurons transform thousands of synaptic inputs into specific patterns of action potential output 38,39 . Nonspecific stimulation of resting AMPA-R by agonistic effects could activate multiple neural circuits simultaneously. The resulting continuous and repetitive synaptic inputs on single neurons connected through their multiple pathways may cause sublinear summation, such as through the activation of postsynaptic conductance 40,41 , with the reduced signal-to-noise ratio of an evoked response. TAK-653 might remarkably increase firing rate via strictly regulated agonist-dependent activation of AMPA-R, which may lead to potent cognitive improvement (Table S7). In line with this observation, similar to TAK-653, TAK-137, which improves cognitive functions in multiple paradigms, also robustly potentiated the AMPA-R component of the synaptic responses.
As previously reported, TAK-137 has little agonistic activity. However, compared with TAK-137, TAK-653 may have even lower agonistic activity. TAK-653 has a wider exposure margin than TAK-137 between cognitive improvement and seizure in rats, thus, this small difference in the agonistic effects of TAK-653 and TAK-137 in vitro likely related to their seizure susceptibility at higher dose ranges. Besides seizure liability, a bell-shaped dose-response might have limited the development of AMPA-R potentiators as drugs 42,43 . Given the heterogeneity of human metabolic profiles that may be associated with larger variations in pharmacokinetic profiles of drugs, bell-shaped dose-responses may be a significant disadvantage even with deliberate design of dose selection 44 . TAK-653 may overcome these issues because it showed cognitive improvements at a wider dose range and had a low risk of receptor desensitization or sensitization after 14 days of repetitive dosing.
Multiple findings have suggested that ketamine can produce antidepressant activity through a rapid disinhibition of pyramidal neurons and subsequent glutamate burst, followed by postsynaptic AMPA-R activation 45 . Based on the enhancement of AMPA-R-mediated EPSPs, TAK-653 could robustly enhance the activity of postsynaptic AMPA-Rs. These data further support the potential antidepressant action of TAK-653, although detailed preclinical studies to assess antidepressant effects of TAK-653 are needed.
In summary, we found that agonistic effects may significantly impair the function of AMPA-R potentiators in synaptic transmission and interfere with cognitive improvement. In fact, TAK-653, an AMPA-R potentiator with minimal agonistic activity, substantially enhanced synaptic AMPA-R responses and potently improved cognitive functions in multiple tasks at a wider dose range and with a broader safety margin against seizure. TAK-653 could be effective for multiple CNS diseases with cognitive dysfunction, and depression. TAK-653 is currently being developed for the treatment of depressive disorders.

Materials and methods
The care and use of the animals and the experimental protocols used in this research were approved by the Experimental Animal Care and Use Committee of Takeda Pharmaceutical Company Limited and conducted in accordance with the guidelines. The animal care and use program is accredited by the American Association for Accreditation of Laboratory Animal Care (AAALAC) International's Council on Accreditation. The AAALAC sets standards that call for the humane care and use of laboratory animals by enhancing animal well-being, improving the quality of research and advancing scientific knowledge relevant to humans and animals. All experiments were carried out in compliance with the ARRIVE guidelines.
Novel object recognition test. NOR tests were performed using male Long-Evans rats at 6 weeks old as previously described 46 , with some modifications. On day 1, rats were allowed to habituate to the empty test box (a gray-colored polyvinyl chloride box (40 × 40 × 50 cm)) for 10 min individually. Testing comprised two 3-min trials called the acquisition trial and the retention trial that were separated by 48 h inter-trial intervals (ITIs). On day 2, in the acquisition trial, rats were allowed to explore two identical objects (A1 and A2) for 3 min. On day 4, in the retention trial, rats were again allowed to explore a familiar object (A3) and a novel object (B) for 3 min. The object exploration was defined as rats' licking, sniffing or touching the object with forelimbs while sniffing. TAK-653 or AMPA was orally or intraperitoneally, respectively, 2 h or 0.5 h prior to the acquisition and the retention trials. The NDI was calculated using the following equation: novel object interaction/total interaction × 100 (%).
Radial arm maze test. RAM tests were performed using 9-week-old male Long-Evans rats as previously described with a minor modification 31 . Each arm was 50 cm long, 10 cm wide and 40 cm high, and the maze was elevated 50 cm above the floor. Long-Evans rats were food-restricted to 85% of free-feeding body weight throughout the experimental period. Rats were well trained to collect pellets placed on the edge of each arm. The learning criterion for the testing session was defined as 2 errors or fewer for 2 consecutive days. In the testing session, each rat was placed on the maze facing the fixed arm at the start of the trial. The entry of rats into each arm was recorded in sequence until all pellets in the 8 arms were consumed, or 5 min had elapsed. TAK-653, LY451646 or AMPA was administered 1.5 h, 1 h or 0 h prior to the administration of vehicle or MK-801, respectively. Thirty min after dosing of vehicle or MK-801, rats were placed on the maze. Details are described in the Supplementary Information.
Delayed match-to-sample tasks. DMTS tasks were performed using 4-6 years old male cynomolgus monkeys (Macaca fascicularis) weighing 4-6 kg as previously described 16 , using a Cambridge Neuropsychological Test Automated Battery (CANTAB) system (CeNes, Cambridge, UK). Monkeys were maintained at 80% of free-feeding body weight throughout the experiment. Details are described in the Supplementary Information.
Maternal immune activation induction. Poly-I:C (5 mg/kg) dissolved in sterile pyrogen-free 0.9% NaCl (control) solution was administered to pregnant female C57BL/6J mice on gestation day 15 (GD15) via the intravenous route at the tail vein with a volume of 5 ml/kg.
Social approach and avoidance test. The apparatus was a three-chamber gray acryl box (53 × 19 × 21.5 cm, outer chamber: 19.3 × 19 × 21.5 cm), and dividing walls were made from clear acryl plates with gates. In the two outer chambers, transparent acryl cylinders with small holes (cylinder: 8.2 cm φ × 20.5 cm, hole: 1.3 cm φ) were placed to avoid direct physical interaction between a target animal and a test animal. Target animals were C57BL/6 J mice of the same age as test animals and no previous contact with them. On the test day, TAK-653 (0.3 mg/kg, p.o.) or vehicle was administered 1 h before the testing session. Each test animal was introduced in the middle chamber of test apparatus for 3 min with the gates being closed by partitions. Then the partitions were removed gently for the test animal to explore all three chambers freely for 5 min. Sniffing index was calculated as the index of sociability using the following equation: Sniffing index = (Sniffing time to cylinder containing target mouse (s) − Sniffing time to empty cylinder (s))/ (Total sniffing time (s)).

Evaluation of convulsion.
The data were extracted from the studies conducted in accordance with the Good Laboratory Practice Regulation. Male Sprague-Dawley rats obtained at 4 weeks old showed no observable abnormalities in clinical signs during the predose period. The dose levels in the first study and second study were set at "100, 15 and 5 mg/kg, p.o." and "50, 15 and 5 mg/kg, p.o.", respectively. The dosing suspensions were administered into the stomach via a catheter in the morning at the dose volume of 5 mL/kg. Cage side observations to confirm tonic or chronic convulsions were conducted before and at 1, 2, 4, 8 and 24 h after dosing in the first study and before and at 1, 4 and 24 h after dosing in the second study.
Measurement of plasma concentration of compounds. The Sprague-Dawley rats treated with orally administered TAK-653 were decapitated at each time point, and trunk blood was collected into 1.5-mL centrifuge tubes. Plasma was separated from the blood samples by centrifugation. The concentrations of TAK-653 in the plasma were determined using liquid chromatography/tandem mass spectrometry (LC/MS/MS).