Excitatory drive of cortical fast-spiking GABAergic interneurons is set by D-serine acting on NMDA receptors

N-methyl-D-aspartate receptors (NMDARs) populate GABAergic interneurons, where they play a critical role in shaping circuit motifs and memory. However, we are largely ignoring whether and how NMDARs at GABAergic interneurons are gated by signals released in their surrounding microenvironment. Here we explore the dynamics of the co-agonist site occupancy by D-serine and glycine at glutamatergic synapses onto parvalbumin positive GABAergic interneurons in the adolescent prefrontal cortex, an area central to complex cognitive operation. We discovered that D-serine but not glycine is required for maintaining the activity of NMDAR at the GABAergic interneurons and that the identity of the co-agonist is not determined by the synaptic regime. Our study extends the physiological implications of D-serine in brain physiopathology by uncovering its control of inhibitory synaptic networks through NMDARs. report that synaptic activity We further show that d-serine-decient mice, a model of NMDAR hypofunction that exhibits schizophrenia-like phenotypes display attenuated ring pattern of the interneurons and no long-term potentiation. These results provide new insights on the cell-type dependent mode of regulation of NMDAR by their co-agonists. They also widen the physiological landscape of d-serine in brain functioning opening new opportunities for precise therapeutic interventions. nulled action CBIO on ring frequency (n = 5 cells CBIO and Ctrl; P 0.2094, F =2.233 two-way ANOVA). Boxplot graphs show that the rheobase was no longer modied by CBIO upon block of NMDAR (P=0.7381, Mann-Whitney test). f FS-PV + interneurons from SR -/- mice (dark orange, n = 6 cells) showed a marked reduction in their ring activity compared to WT mice (n = 9 cells) (P = 0.2097, F (1,13) =5.958 two-way ANOVA) that is associated with an increase of the rheobase (P = 0.0244, Mann-Whitney test) thus phenocopying the effect of iMK801. RgDAAO = 8, orange circles and traces) acutely d-serine (P = 0.1341, F (1,17) =2,474 two-way ANOVA). Representative traces of NMDA-EPSCs recorded from −70 mV to +40 mV for control and RgDAAO conditions are illustrated. b NMDA-EPSCs (+40 mV) and AMPA-EPSCs (-70 mV) were abolished in the presence of D-AP5 (green trace, n = 4 cells) and NBQX (brown trace, n = 4 cells) respectively. Treatment of WT slices with RgDAAO (orange circles and traces) markedly reduced NMDAR/AMPAR ratio by ~50% from 1.10 ± 0.10 (n = 12 cells) to 0.49 ± 0.05 (n = 10 cells) but concurrent application of BsGO (yellow symbols and traces) showed no further effect (0.54 ± 0.06, n=5 cells) (P = 0.0015, Kruskal-Wallis test; Ctrl vs RgDAAO: P = 0.0018, RgDAAO vs BsGO: P > 0.9999, post-hoc Dunn’s test). c Time course shows that bath application of CBIO (purple symbols and traces) on WT slices to promote the synaptic availability of d-serine increased NMDA-EPSCs by 22.16 ± 4.58% (n = 7 cells: P = 0.0156, Wilcoxon test). Bath application of ALX5407 (fuschia) to increase the synaptic availability of glycine potentiated EPSCs (n = 6 cells; +49.23 ± 11.69%, P = 0.0313, Wilcoxon test). The greatest increase of the NMDA-EPSCs in ALX5407 than in CBIO likely reected that GlyT1 normally prevent glycine access to synaptic NMDARs.

for information processing within the brain 1, 2 . The fast-spiking parvalbumin-containing (FS-PV + ) neurons represent the largest population of cortical inhibitory neurons 1 and by providing the major somatic inhibitory drive to excitatory pyramidal cells (PCs) play a critical role in circuit physiology 3,4 . As a consequence, malfunction of these GABAergic interneurons is central to several neuropsychiatric disorders [5][6][7] .
FS-PV + interneurons are primarily recruited by glutamatergic synaptic input from the PCs. In particular, FS-PV + interneurons express N-methyl-D-aspartate receptors (NMDARs) that are pivotal for circuit entrainment and for generating network oscillations that encode numerous cognitive processes. Selective ablation of NMDARs during early postnatal development results in synaptic dysconnectivity with PCs and in corrupted network oscillations and behavioural de cits [8][9][10] . Thus, hypofunction of NMDAR onto PV + interneurons has emerged as a cause of various neurodevelopmental disorders including schizophrenia (SCZ) and autism spectrum disorders (ASD) 10,11 . However, how NMDAR at FS-PV + are physiologically regulated by extracellular signals is unknown. NMDARs are heterotetrameric molecular devices typically composed of two GluN1 subunits and two GluN2 subunits 12,13 . Yet, activation of the canonical NMDARs requires simultaneous binding of l-glutamate and a co-agonist, glycine or d-serine [12][13][14] . Paradoxically, more than two decades after the discovery of the action of glycine and d-serine at NMDAR [15][16][17] , we are still ignoring how the functions of GABAergic interneurons are regulated by NMDAR co-agonists and the physiopathological relevance of such modulation remains unaddressed. In this study, we examined the potential contributions of d-serine and glycine in controlling the activity and functions of NMDARs at FS-PV + interneurons in the prelimbic area (PrL) of the prefrontal cortex (PFC), a brain area central to complex cognitive operation 18,19 which dysfunction is invariably implicated in the physiopathology of SCZ 20 and then explored the physiological conditions that specify their actions. By combining cellular electrophysiology with the use of unique pharmacological interventions and genetic manipulations, we report that d-serine is the sole co-agonist controlling NMDAR at FS-PV + interneurons under different synaptic activity levels. We further show that d-serine-de cient mice, a model of NMDAR hypofunction that exhibits schizophrenia-like phenotypes display attenuated ring pattern of the interneurons and no long-term potentiation. These results provide new insights on the cell-type dependent mode of regulation of NMDAR by their co-agonists. They also widen the physiological landscape of d-serine in brain functioning opening new opportunities for precise therapeutic interventions.
Strikingly, FS-PV + interneurons from SR -/mice showed a marked reduction in their intrinsic excitability in comparison to wild-type (WT) littermate mice ( Fig. 1f; Supplementary Fig. 3). Altogether, our ndings support the view that the activity of the GABAergic FS-PV + interneurons highly depends on the occupancy of the NMDAR co-agonist binding site by d-serine and that these NMDAR are tonically activated under basal conditions. Co-agonism by d-serine but not glycine selectively controls NMDA-EPSCs in interneurons. Since the level of occupancy of the co-agonist site of NMDAR by d-serine is critical for driving FS-PV + interneurons ring activity, we hypothesized that it could also determine their synaptic coupling with PCs. We then evaluated the physiological consequences of such modulations by recording NMDA-EPSCs from L5 FS-PV + cells (Fig. 2). As before, NMDA-EPSCs were pharmacologically isolated using NBQX (20µM), picrotoxin (50µM) and strychnine (10 µM). To further precise the identity of the co-agonists in controlling the activity of NMDAR at cortical GABAergic interneurons, we used enzymatic scavengers that speci cally blunt the action of endogenous d-serine or glycine by degrading them in the extracellular space. Selective and acute depletion of d-serine with RgDAAO (0.2 U/mL) [29][30][31]46 did not affect the voltage-activity dependency of the NMDA-EPSCs (Fig. 2a) but yielded to a ~50% reduction in the NMDA/AMPA ratio tied to its speci c eroding action on NMDA-EPSCs (Fig. 2b). Analysis of the paired-pulse ratio (PPR) of NMDA-EPSCs showed no difference between control and RgDAAO groups ( Supplementary Fig. 5) thus con rming that d-serine acts speci cally on postsynaptic NMDARs expressed by FS-PV + cells. As shown in Fig. 2b, NMDA-EPSCs persisted in the presence of RgDAAO suggesting that glycine which is left intact in these experiments might also contribute to regulate synaptic NMDARs [29][30][31] . The putative role of glycine in the persistent d-AP5 sensitive NMDA-EPSCs in the presence of RgDAAO is excluded by addition of BsGO (0.2 U/mL) that selectively degrades glycine 29-31 but did not further impact the amplitude of NMDA-EPSCs (Fig. 2b). As expected, CBIO that elevates synaptic levels of d-serine increased NMDA-EPSCs ( Fig. 2c) suggesting that the coagonist site of synaptic NMDAR on FS-PV + cells was not saturated by the ambient levels of the endogenous co-agonist. Intriguingly, blockade of GlyT1 with ALX5407 that increases glycine levels increased the amplitude of NMDA-EPSCs as well (Fig. 2c). Therefore, rendering glycine available at the synapse enables NMDAR modulation but is not su cient to in uence the ring activity (Fig. 1c). CV -2 analysis revealed that CBIO and ALX5407 exerted their potentiating effects at purely postsynaptic NMDARs ( Supplementary Fig. 6). In conclusion, these results revealed that NMDAR at FS-PV + interneurons in adolescence are primarily gated by d-serine and not glycine at low activity regime and point that glycine transporters (e.g GlyT1) are e ciently maintaining the levels of glycine below effective concentrations within the synaptic cleft as already observed at excitatory synapses between PCs in the hippocampus 29,31 or amygdala 30 .
Selective loss-of-function of d-serine impairs synaptic plasticity of PC excitatory output synapse onto FS-PV + cells. We next explored whether the identity of the co-agonist driving NMDAR at PV + interneurons might be determined by the level of synaptic activity 30,31 . To do so, we analyzed the role of the co-agonist in controlling NMDAR-dependent short-and long-term excitatory synaptic plasticity at the PC to FS-PV + connections ( Fig. 3a, b). To study short-term plasticity, we examined the effects of realistic trains of stimulation of the synapse at low physiological 2, 20 or 50 Hz trains typical of PFC PV + cells ring rates during attentional processing [47][48][49] . At 2 and 20Hz individual NMDA-EPSCs were clearly discernible for each pulse of stimulus current and exhibit amplitude attenuation which was precipitated by increasing frequencies 50,51 (Fig. 3a). Acute d-serine depletion by treating the slices with RgDAAO accelerated and increased the depression rate at 2 and 20 Hz in comparison to control slices. At 50Hz, NMDAR-EPSCs were not discernible and exhibited temporal summation of the rst three trains before substantial depression appeared (Fig. 3a). Such pattern is most likely related to the slow kinetics of the NMDAR channels 12,13 thus occluding single NMDA-EPSCs to return to baseline before the next stimulation. Anyway, no difference was observed at 50 Hz between the RgDAAO treated and control groups indicating that the synaptic depression became independent of the levels of the co-agonist at γ ranges. Most importantly, these data show that the presence of a threshold level of d-serine is necessary for NMDAR to maintain persistent states of activity at theta and beta frequencies (~2-20 Hz).
Although excitatory output synapses onto FS-PV + interneurons have been shown to undergo various forms of activity-dependent long term plasticity 52-57 whether NMDAR-dependent long-term potentiation (LTP) of excitatory synapses onto GABAergic interneurons would then rely on a speci c co-agonist is still unknown. Hebbian NMDAR-dependent LTP requires coincident glutamate release and postsynaptic depolarization to relieve channel blockade by Mg 2+ ions. We therefore tested a Hebbian protocol consisting in pairing afferent tetanus high-frequency stimulation (HFS) with postsynaptic depolarization of the FS-PV + cells to 0 mV. This protocol invariably induced a D-AP5 sensitive long-term potentiation (LTP) of the excitatory input synapse onto FS-PV + interneurons (Fig. 3b) that appeared not to depend on somatodendritic voltage-gated Na + channels activation since 5 mM QX314 was included in the whole-cell patch pipette. The mean paired pulse ratio (PPR) of evoked EPSCs at 30 ms apart was not signi cantly changed upon this Hebbian LTP at FS-PV + cells ( Supplementary Fig. 7a). In addition, analysis of CV -2 con rmed the postsynaptic locus of the LTP (Supplementary Fig. 7b). These results suggest that this form of LTP is expressed with no apparent change in presynaptic function (e.g transmitter release) and further support that it does not involve presynaptic NMDARs. More importantly, the same protocol failed to induce LTP in the presence of RgDAAO thus phenocopying the blocking action of D-AP5 (Fig. 3b). In conclusion, excitatory input synapses onto FS-PV + neurons can undergo homosynaptic LTP that relies exclusively on the activation of postsynaptic NMDAR by d-serine.
NMDA-EPSCs at FS-PV + interneurons are not altered in a genetic model of NMDAR hypofunction. To con rm the pivotal role of d-serine, we next anticipated that the activity of NMDAR at FS-PV + interneurons would be ultimately affected in the SR -/mice (Fig. 4a), a translational model of NMDAR hypofunction 41,42,44,58,59 . Unexpectedly, no signi cant differences were found in the voltage dependency and peak amplitude of NMDA-EPSCs recorded from FS-PV + neurons in SR -/-PFC (Fig. 4b, c). In addition, NMDA/AMPA ratio was unaffected in SR -/slices when compared to WT slices (Fig. 4c). However, we noticed that NMDA-EPSCs from FS-PV + cells in SR -/mice systematically displayed a longer rising time ( Fig. 4d) but no change in decay kinetics. As rise-time of EPSCs re ects the opening rate of the NMDAR channel, we envision that the mode of activation of the NMDARs in SR -/mice might likely differ from WT mice because of increased dwell time of neuromodulatory signals in the extracellular space. We then explored the underlying mechanisms that could contribute to maintain fully functional NMDAR in SR -/mice in the absence of d-serine and that could result in slowering rise-time. We had previously found that BsGO did not affect NMDA-EPSCs in WT slices (Fig. 2b). Conversely, addition of BsGO (0.2 U/mL) to SR -/slices remarkably reduced the NMDA-EPSCs by ~60% without altering AMPA-EPSCs ending in a net decrease in the NMDA/AMPA ratio (Fig. 4c). These data demonstrate that glycine could serve as a spare co-agonist for synaptic NMDARs at excitatory synapses onto GABAergic interneurons but solely in the chronic absence of d-serine. Building on these observations, we inferred that the slowing kinetics of risetime of NMDA-EPSCs is a consequence of increased glycine spillover and increased dwell time of the coagonist in the extracellular space consequent to the reduced uptake of glycine by astrocytes in SR -/mice rather than a change in glutamate residency in the synaptic cleft.
Co-agonism by glycine incompletely rescues the absence of d-serine during synaptic plasticity. We then wondered whether this compensation by glycine in SR -/mice we observed during basal synaptic transmission could fully meet or not the needs for NMDAR functioning when increasing the synaptic regime. As shown above, at 50 Hz the synaptic attenuation and summation does not depend anymore on the degree of occupancy of the NMDAR co-agonist binding site, we thus restricted our evaluation of NMDA-EPSCs to 2 and 20 Hz in PFC slices isolated from SR -/-. Interestingly, we found no difference with control WT littermates at 2 Hz but revealed that short-term plasticity in SR -/slices was altered at 20 Hz (Fig. 5a). These results thus indicated that glycine could sustain NMDAR activity at low ring rate below 20 Hz but beyond, the system became invariably unstable in the absence of d-serine. Finally, we considered the potential contribution of glycine in enabling LTP of the PC to FS-PV + excitatory synaptic contacts in the SR -/mice. As previously reported, HFS invariably induced NMDAR-dependent LTP in WT mice but this protocol was ineffective in triggering LTP in SR -/mice (Fig. 5b). This effect was not associated with changes in presynaptic parameters ( Supplementary Fig. 7c, d) but most likely re ect a loss of neuromodulation of postsynaptic NMDARs. Therefore glycine, contrary to d-serine and despite its action on NMDARs at low synaptic regime, is not engaged in the formation of LTP at excitatory synapses onto FS-PV + interneurons. Overall these results argue that glycine could compensate for the chronic absence of d-serine (e.g., SR -/-) to support the demand of NMDAR activity only at low but not at high synaptic regimes in striking contrast to reported observations gathered on synapses between excitatory neurons 30, 46,60,61 .

Discussion
One key conclusion from the present study is that d-serine but not glycine modulates the excitatory drive of cortical GABAergic PV + interneurons by acting on NMDAR. We demonstrate that speci c limited functions of d-serine induce NMDAR hypofunction impairing burst ring and long term synaptic plasticity. Here, inhibitory neurons were targeted using PV-tdTomato reporter mice and then no distinction between the chandelier and basket cells types neurons was made. However, even if both cell populations are present in the mouse mPFC, only the basket cells are found in deep layers 62 . Furthermore, the unique fastspiking electrophysiological features of the PV + neurons we recorded here indicate that we are most likely indeed dealing with basket cells. These studies concluded that, i) both glycine and d-serine cooperate to regulate NMDAR functions at the excitatory PC-to-PC synapses, ii) the identity of the co-agonist is determined by the level of synaptic activity 29-31, 46,63 , and iii) under pathological conditions of chronic d-serine absence as occurring in the SR -/mice, glycine can fully replace d-serine and support the activity of NMDARs at excitatory neurons 30,60,61 especially in the mPFC 46 . Owing to the fact that SR -/mice displayed impaired gamma oscillations 58 and the critical roles of PV + interneurons in circuit's dynamics and their key implication in SCZ pathogenesis as the putative cell locus of NMDAR hypofunction, we decided to focus on this celltype. Here, we probed the level of activation of the NMDAR co-agonist site at FS-PV + interneurons by endogenously released coagonist under different patterns of synaptic activity and pharmacological interventions. We found that the site is unsaturated in physiological conditions which is consistent with the notion that the basal concentration of endogenous co-agonist is insu cient to saturate the NMDAR and that enhancing its recruitment can improve the output functions of the GABAergic neurons. We show that the identity of the co-agonist (eg d-serine) at GABAergic neurons is not dictated by the level of afferent excitatory inputs activity and that glycine can only sparely replace d-serine. Indeed, compensation by glycine of NMDAR gating in SR -/mice or under acute loss-of function of d-serine (e.g RgDAAO) is incomplete and de cits of excitatory drive of PV + interneurons already appear at low beta frequency range (~20 Hz). Since glutamate is fully saturating NMDARs during synaptic transmission 68 , dserine availability is the limiting factor for optimal NMDAR activation at PV + interneurons. Accordingly, despite normal synaptic transmission at low synaptic regimes (<2Hz), HFS-induced LTP is occluded in SR -/mice just phenocopying the effect of RgDAAO treatment. Although here we used a HFS to induce LTP in the presence of the GABAR blocker picrotoxin and strychnine to disable unwanted feed-forward inhibition as a minimal experimental paradigm, it is tempting to speculate that even using bursts repeated at the theta-frequency (~5Hz) would have not recruited glycine and would have fail to rescue LTP. Indeed, a recent study reported that SR -/mice had de cits in frontal cortex evoked power across the beta (20-30 Hz) and the gamma (30-80 Hz) frequencies 58 . A role for glycine in wild-type mice (e.g when dserine is present) was unmasked only when measuring NMDA-EPSCs after blocking the GlyT1 with ALX5407 (Fig. 2) which seems at odds with the non-implication of the amino acid in the control of burst ring of PV + cells. This apparent discrepancy could be related to the fact that bursting activities are recorded at the resting membrane potential under current clamp thus better preserving the physiological conditions. NMDA synaptic currents were collected at + 40 mV to relieve the potential-dependent Mg 2+ block of the channel and in response to non-minimal perisomatic stimulations. These experimental conditions are prompted to place the recorded cell and its local environment in an upper 'non physiological' excitatory state that may have unsilenced some NMDARs normally non active at physiological membrane potentials and thus may favor the engagement of glycine. Henceforth, it is tempting to envision the existence of at least two populations of NMDARs that are differently distributed along the somatodendritic arbor of the PV + interneurons that would be accessible for gating by d-serine and to a lesser degree by glycine depending on the physiological needs. We also show that the effect of blocking GlyT1 with ALX5407 is greater than the effect of inhibiting DAAO with CBIO on NMDA-EPSCs amplitude. Although, the mechanisms of inhibition is different (transporter versus an enzyme) and the pharmacological effects may relate to kinetics of GlyT1 vs DAAO, these results predict that the levels of dserine in the synaptic cleft are near saturation while the levels of glycine are insu cient for the normal activation of NMDARs. Altogether, by reporting that only pharmacological or genetic-induced manipulations of d-serine levels did alter the ring activities of the PV + cells in a NMDAR-dependent manner, we demonstrate that d-serine but not glycine is the permissive factor engaged in the tonic control of NMDAR in physiological conditions. The mechanisms underlying the selective action of d-serine at PV + inhibitory neurons remain to be addressed. Although the NMDARs subunits composition may differ between excitatory and inhibitory neurons 12 , both neuromodulators display nearly similar sensitivity to NMDARs independently of their subunits composition 12,13 . The partial compensatory action of glycine at GABAergic neurons in SR -/mice may rather re ect differences in the dynamics and availability of the co-agonists between excitatory synapses onto principal cells vs inhibitory neurons, as already observed for glutamate 69 . Astrocytes by expressing the GlyT1 are critically involved in regulating the ambient levels of glycine at the proximity of excitatory neurons [29][30][31] . Hence, the synaptic microenvironment of GABAergic and glutamatergic neurons would be largely in uenced by the morphological interposition of individual synapse with astrocytes [69][70][71] . Yet, excitatory neurons establish close contact with astrocytes lea ets while GABAergic neurons notably the PV + interneurons establish sparse contact with glia 71 although still contributing to inhibitory synaptic signalization. On line with these observations, pharmacological blockade of GlyT1 by ALX5407 to build up ambient glycine levels did not alter the ring activity of GABAergic interneurons while inhibition of DAAO by CBIO to boost d-serine synaptic availability positively increases ring activity of PV + interneurons and their synaptic coupling with PCs. Accordingly, it is tempting to speculate that dserine which behave as the strict co-agonist for NMDAR at PV + interneurons would be preferentially formed by the postsynaptic GABAergic neurons themselves or by their apposed presynaptic glutamatergic partner without much involving glia. Delineation of the precise role(s) of these respective partners in the dynamics and synaptic disposition of d-serine (and glycine) in inhibitory network signalization has to be determined in future studies. The presence of active DAAO in the forebrain has been questioned for decades and therefore its role in d-serine catabolism remains an opened issue [34][35][36] . Our previous work indicated that the potentiating effect of CBIO on excitatory synaptic transmission in the hippocampus circuitry was nulled in SR -/mice thus demonstrating that the effect of the compound by inhibiting DAAO was caused by an elevation in d-serine 31 . Here, our pharmacological experiments using CBIO further evidence that DAAO is present in the forebrain and also is the enzyme involved in dserine catabolism but most importantly, show that DAAO by controlling d-serine disposition play a critical role in controlling the activity of GABAergic interneurons as well, Importantly, in addition to unearth NMDAR hypofunction at PV + interneurons in SR -/mice, our results indicate how NMDAR hypofunction impairs PV neuronal functions by demonstrating impaired burst ring. Burst ring of PV + interneurons promote the reliability of somatic inhibitory drive to PCs thus playing a cardinal role in the temporal precision of local pyramidal ring and in the generation of coherent oscillations enabling the synchronization of large PC populations and proper behaviors [1][2][3][4][5][6][7][8][9][10] . Recent ndings suggest that NMDARs in PV interneurons enhances the probability of GABA release in the mPFC 26 . Therefore, we predict that the presence of persistent deprived burst ring in PV neurons as observed in SR -/mice would result in abnormal PC disinhibition and alter PV-to-PV cross-inhibition during behavioral tasks that ne-regulate local oscillations. We have recently reported that the ring activity of layer 5 PCs in the mPFC of adolescent mice is unaffected in SR -/mice 46 . Therefore, our new data further support the idea that the co-agonist site of NMDAR at the PV + GABAergic neurons is more sensitive to dserine withdrawal as compared to PCs. Accordingly, we further highlight that the level of occupancy of the NMDAR co-agonist binding site depends on the cell type re ecting how synapses on PCs and on interneurons adopt different morphology and functional properties [69][70][71] . Altogether, our observations thus offer a mechanistic explanation why SR -/mice show impairments in gamma oscillations and social interactions 58,59 .
Lastly, besides transforming our vision of brain circuit's physiology, our study is of main clinical relevance by fully justifying current therapeutics strategies targeting the co-agonist site of NMDARs by d-serine in the management of SCZ and others brain disorders 11 . Indeed, NMDAR hypofunction at PV + interneurons has been invariably proposed to play a pivotal role in the pathogenesis of many psychiatric disorders, such as SCZ 5,11 and studies suggest that these neurons should be considered as viable targets to more e cient antipsychotics. Yet, DAAO inhibitors are seen as one of the most promising pharmacological approach to counteract the cognitive and negative symptoms caused by NMDAR hypofunction 72 . Our study connects these different emerging concepts by reporting that loss-of-function of d-serine induces selective NMDAR hypofunction at PV + interneurons and that DAAO inhibition with CBIO increases ring activity of PV + interneurons but not of PCs. Although PV + interneurons may be central to NMDAR hypofunction in SCZ, dysfunctions are likely spreading to other inhibitory cell-types including the SST and CCK neurons since the latter are expressing NMDAR 22,23 as well. Additional studies would be needed to elucidate the roles of the co-agonists at these different cell populations and the physiopathological relevance of these modulations.
In summary, our study sheds new light on brain circuit's physiology by uncovering the general principles of NMDARs regulation at GABAergic interneurons by upstream signals released in their surrounding microenvironment. Building on these ndings, we identify that d-serine is critical for network computations by engaging NMDARs of PV + interneurons and that loss of its functions would recapitulate the synaptic de cits in disease like SCZ and could be exploited for the future development of more effective clinical interventions targeting inhibitory neurons.
Animals were housed in groups in polycarbonate cages and maintained on a 12/12 hr light/dark cycle (light on at 7am) in a temperature (22°C) and humidity controlled room. Animals were given access to food and water ad libitum. All experiments performed in France complied with the European Union recommendations (2010/63/EU) and were approved by the French Ministry of Agriculture and Fisheries Epi uorescence illumination (CoolLED pE-2 excitation system) was used to visualize tdTomato-positive interneurons (Ex/Em: 555/581nm), using differential interference contrast infrared videomicroscopy.
Recordings were obtained using a Multiclamp 700B ampli er (Molecular devices, CA, USA) and signals were ltered at 2 kHz and digitized at 5 kHz via a DigiData 1440A interface (Molecular Devices, CA, USA). Data were collected and analyzed using pClamp10 software (Molecular Devices, CA, USA). Series resistance and holding current were monitored throughout the experiment. Cells with an access resistance > 25 MΩ at resting potential were excluded from analyses as well as any cell for which a >20% change in those parameters occurred during the course of the experiment. Neurons were rst clamped at −70 mV and allowed to dialyze for 5 min before any recording. All recordings were done in the presence of strychnine (10 µM) and picrotoxin (50 µM) in the bathing ACSF to block glycinergic and GABA A receptors, respectively. All drugs needed for control conditions were applied 10 minutes before recording and during the full length of the experiment. Subsequent drug effects were measured 5 minutes after a plateau was obtained and compared to baseline. In some experiments, MK801 was added to the intracellular solution to block NMDAR at the recorded cell.
Membrane and ring properties of PC and FS-PV + neurons were recorded in current-clamp mode using Kgluconate solution including or not MK801 (iMK801: 2-3 mM). Results were obtained in response of square step current injections (500 msec) into the recorded cell from -100 pA to 200 pA with 20 pA steps.
Spike frequency adaptation was measured as the ratio of the rst inter-spike interval duration over the 8th one at rheobase +40 pA. To further precise the locus of action of the drugs, the coe cient of variation (CV) analysis method was used 26 . The proportional change in the inverse square of the coe cient of variation (CV −2 ) was compared with the proportional change in the mean postsynaptic potential amplitude (M) to determine whether the quantal amplitude (q), the release probability (p) or the number of release sites (n) had changed. In a binomial distribution, CV −2 = [np/(1 -p)] and is therefore independent of q, while M = npq. When mean postsynaptic potential amplitude changes, no change in CV −2 indicates that only q has changed and a postsynaptic site is involved. A larger proportional change in CV −2 than in M indicates that p has changed, whereas a similar proportional change in the two parameters indicates that n has changed.
Both presynaptic and postsynaptic sites are affected when a smaller proportional change in CV −2 than in M is revealed.
Immunostainings and confocal microscopy. Double transgenic mice (i.e PV + -tdTOM::WT and PV + -  Statistics. Analysis was conducted using Prism 8 software (GraphPad, SanDiego, California). Results are expressed as mean ± SEM and a p-value lesser than 0.05 was considered signi cant. In the gures legends, n refers to the number of recorded cells sampled from 2-6 mice depending on conditions. Twoway ANOVA with repeated measurements were done for ring frequency, input/output curve, NMDA IV curve and short-term plasticity experiments with post-hoc multiple comparison tests using a Bonferroni-Dunn correction factor when necessary. For time-course experiment and LTP, the mean amplitude of the last 5 minutes of recording (expressed in percent of change compared to control) was tested versus 0 (and versus 100 for LTP) with one-sample Wilcoxon test. Other comparisons were done using the Wilcoxon test for paired data and the Mann-Whitney test for unpaired data.   d-serine co-agonism enables synaptic plasticity of PC excitatory output synapse onto FS-PV + interneurons. a NMDAR co-agonist and PC-to-PV synapse adaptation to increase synaptic regime.

Declarations
Normalized EPSCs amplitudes in response to stimulation trains applied at increasing frequencies (2, 20 and 50Hz). As the frequency increases, control EPSCs showed graded attenuation (Upper graphs: white circles and middle black traces). At 2Hz, acute d-serine depletion (RgDAAO, orange traces and circles)