Contribution of NMDA Receptors to Synaptic Function in Rat Hippocampal Interneurons

Abstract The ability of neurons to produce behaviorally relevant activity in the absence of pathology relies on the fine balance of synaptic inhibition to excitation. In the hippocampal CA1 microcircuit, this balance is maintained by a diverse population of inhibitory interneurons that receive largely similar glutamatergic afferents as their target pyramidal cells, with EPSCs generated by both AMPA receptors (AMPARs) and NMDA receptors (NMDARs). In this study, we take advantage of a recently generated GluN2A-null rat model to assess the contribution of GluN2A subunits to glutamatergic synaptic currents in three subclasses of interneuron found in the CA1 region of the hippocampus. For both parvalbumin-positive and somatostatin-positive interneurons, the GluN2A subunit is expressed at glutamatergic synapses and contributes to the EPSC. In contrast, in cholecystokinin (CCK)-positive interneurons, the contribution of GluN2A to the EPSC is negligible. Furthermore, synaptic potentiation at glutamatergic synapses on CCK-positive interneurons does not require the activation of GluN2A-containing NMDARs but does rely on the activation of NMDARs containing GluN2B and GluN2D subunits.


Introduction
Neuronal networks require finely balanced excitatory glutamatergic and inhibitory GABAergic synaptic transmission to allow learning, memory, and typical brain function. Alterations to this balance may result in autism and intellectual disability (Antoine et al., 2019), epilepsy (Sloviter, 1987), schizophrenia (Daskalakis et al., 2002), and depression (Czéh et al., 2015). This balance has been well described in the CA1 of the hippocampus, where local inhibition arises from a heterogeneous population of inhibitory interneurons (INs) forming dense connections with themselves and excitatory pyramidal cells (PyrCs; Pelkey et al., 2017;. INs possess diverse dendritic arbors, receiving glutamatergic inputs from both local and distant sources where synaptically released glutamate binds to and activates ligand-gated ion channels: chiefly AMPAreceptors (AMPARs) and NMDA receptors (NMDARs), that give rise to short (,10 ms) and long (100-1000 ms) synaptic events respectively (Cull-Candy et al., 2001). NMDARs have been proposed as the archetypal receptor underlying synaptic plasticity, and thus memory formation, in many cell types (Morris et al., 1990;Tsien et al., 1996;Morris, 2013).
NMDARs exist as tetrameric receptor complexes comprised of two GluN1 and two GluN2 subunits, the latter existing as four separate gene products (GluN2A-D; Gielen et al., 2009;Wyllie et al., 2013). During early brain development, GluN2B and GluN2D are the dominant isoforms in the hippocampus (Monyer et al., 1994;Flint et al., 1997;Martel et al., 2009;McKay et al., 2012). In mature principal hippocampal neurons GluN2A and GluN2B subunits predominate while GluN2D subunits have been reported in some IN populations (Engelhardt et al., 2015;Perszyk et al., 2016;Garst-Orozco et al., 2020). Native NMDARs can exist as either diheteromers (i.e., GluN1/ GluN2A) or as triheteromeric assemblies (i.e., GluN1/2A/ 2B) with ligand binding, receptor kinetics and ion channel conductance being conferred by the identity of the nature of the GluN2 subunits present (Stern et al., 1992;Wyllie et al., 1996;Gielen et al., 2009;Hansen et al., 2014). Indeed, the presence of GluN2A subunits confers fast kinetics, whereas the presence of GluN2B and GluN2D subunits confer increasingly slow kinetics (Monyer et al., 1994;Vicini et al., 1998;Wyllie et al., 1998; for review, see Wyllie et al., 2013). Synaptically released glutamate, together with glycine (or D-serine), binds to synaptic and extrasynaptic NMDARs, and at depolarized membrane potentials the relief of voltage-dependent Mg 21 block allows ion conduction (reviewed in Hansen et al., (2018)). This requirement of postsynaptic depolarization and presynaptic neurotransmitter release for NMDAR opening defines them as a coincidence detector, critical for the establishment of Hebbian plasticity (Seeburg et al., 1995). Given these properties, the stoichiometry of NMDARs subunits confer distinct functional synaptic properties (Stern et al., 1992;Cull-Candy et al., 2001;Wyllie et al., 2013). Several studies have aimed to identify the contribution of NMDAR subunits to the different excitatory and inhibitory neuronal populations. The current understanding is that mature CA1 PyrCs exclusively express GluN2A/B-containing NMDARs (Flint et al., 1997;Gray et al., 2011). Meanwhile, hippocampal INs possess varied AMPARand NMDAR-mediated responses dependent on cell type (Akgül and McBain, 2016), and all major IN classes appear to express RNA for GluN2A/B/D subunits (Perszyk et al., 2016). Functional NMDARs have been shown on a variety of IN subtypes, including those expressing parvalbumin (PV; Korotkova et al., 2010;Billingslea et al., 2014) and cholecystokinin (CCK; Kotzadimitriou et al., 2018), morphologically defined basket cells (BCs) and dendritic inhibitory cells (Matta et al., 2013), oriens/alveus INs (Hájos et al., 2002), and neurogliaform neurons . Despite this, many studies have been performed in juvenile rodent models and have not been able to determine the relative role of GluN2A in IN subtypes, because of the paucity of pharmacological modulators of this receptor subunit. Moreover, GluN2A-containing NMDARs are proposed to contribute to the induction of long-term potentiation (LTP) and GluN2B-containing NMDARs contribute to both long-term depression and LTP (Liu et al., 2004;Volianskis et al., 2013), while GluN2D NMDARs may contribute to both short-and long-term plasticity (Volianskis et al., 2013;Tozzi et al., 2016). Thus, determining the functional properties of NMDARs in INs will provide insight into their synaptic recruitment and plasticity (Booker and Wyllie, 2021).
This study assesses the relative NMDAR-mediated synaptic currents and the contribution of GluN2 subunits in morphologically and immunohistochemically identified hippocampal INs from wild-type and GluN2A-null outbred rats. We achieve this by performing whole-cell patchclamp recording from INs and PyrCs to examine the NMDAR-mediated synaptic current in the presence of pharmacological manipulation and the role of GluN2 subunits in the generation of synaptic plasticity.

Animals
All procedures were performed according to UK and University of Edinburgh guidelines and under the authority of Home Office Licence (P1351480E). GluN2A-null rats were generated through CRISPR/Cas9 deletion of the gene, with GluN2A protein loss shown previously (Strehlow et al., 2019). All rats were maintained on an outbred Long-Evans Hooded background, and were housed on a 12 h light/dark cycle with ad libitum access to food and water. Male rats were taken for recordings at 4-6 weeks of age to avoid potential confounds because of onset of the estrus cycle. All recordings were performed at 4-6 weeks of age. The average age of wild-type rats was 31.1 6 0.5 d (N = 38 rats total), and 32.4 6 0.8 d (N = 27 rats total) for GluN2A-null rats.
Whole-cell recordings of pharmacologically isolated EPSCs were performed following wash-in of Cs-gluconate solution (;2-5 min) to provide optimal voltage clamp, in the presence of picrotoxin (50 mM). EPSCs were generated by a twisted Ni:Chrome bipolar wire placed either in stratum radiatum (CA1 PyrC, PV-INs, CCK-INs) or in the alveus [somatostatin (SSt) INs], reflecting major synaptic inputs to different cell types (stimulus duration, 0.1 ms; stimulus, 20.3 6 18.3 V). Stimuli were delivered via constant voltage stimulator (Digitimer) sufficient to produce a monosynaptic EPSC in the range of ;200 pA. The amplitude of these monosynaptic AMPA-EPSCs varied between cell types (p = 0.0016, one-way ANOVA) but not within a cell type [CA1 PyrCs: p = 0.84; PV INs: p = 0.83; p = 0.84; p = 0.69; Holm-Sidak tests]. For AMPAR-EPSCs, 10 traces were collected at 20 s intervals at À70 mV. For NMDAR-EPSCs, either CNQX was bath applied first and AMPAR-EPSC blockade observed, or the cell was held at 140 mV to identify the mixed AMPAR/NMDAR-EPSCs, and then CNQX was applied. All NMDAR-EPSCs were recorded following full wash-in of CNQX at 140 mV using the same stimulation intensity as for AMPAR-EPSCs. Pharmacology for the specific NMDAR subunits was as follows: GluN2Bifenprodil tartrate (10 mM) or NAB-14 (10 mM), and both were bath applied. As NAB-14 is highly lipophilic (Yi et al., 2019), all tubing was rinsed with 100% ethanol and liberal amounts of distilled water between recordings to ensure full removal of the drug from the surfaces of perfusion tubes. All EPSC amplitudes were recorded at the peak of the EPSC, as measured over a 2 ms peak average. All rise times are reported as the 20-80% of the peak EPSC amplitude. Decay time constants were calculated from a monoexponential (AMPAR-EPSC) or biexponential (NMDAR-EPSC) curve fit to the decaying phase of the EPSC. For biexponential fits, the weighted tau was taken as the decay time constant. For all kinetic properties, only NMDAR EPSCs with amplitude .20 pA were included for analysis, to exclude measurement artifacts. NMDAR/AMPAR EPSC amplitude ratios were calculated from the average EPSC recorded for the respective epoch.
For LTP recordings, cells were recorded in current clamp, using a K-gluconate-based internal solution. Putative CCK INs were selected for recording in stratum radiatum and monosynaptic EPSPs were generated via bipolar electrode stimulation of the stratum radiatum, placed ;500 mm distal to the recorded somata. Following breakthrough into whole-cell configuration, the membrane potential was set to À70 mV with the application of a bias current, and the bridge was balanced. Then EPSPs with amplitude of ;5 mV (mean, 5.94 6 2.85mV; range, 1.64-12.35 mV) were recorded for 5 min of stable baseline (,10% change over 5 min, calculated from a linear fit of baseline period); if stability was not achieved within 7.5 min of breakthrough, the recording was abandoned. Following 5 min of stable recording, LTP was induced, with 2 Â 100 Hz of tetanus stimuli, at the same strength as the baseline EPSP. Following induction, 25 min of EPSPs were recorded to assess the potentiation of the whole-cell EPSP, which was reported as the peak amplitude measured as the peak 2 ms response. All recordings were filtered online at 10 kHz with the built-in four-pole Bessel filter and digitized at 20 kHz (Digidata1440, Molecular Devices). Traces were recorded in pCLAMP 9 (Molecular Devices) and stored on a personal computer. Analysis of electrophysiological data were performed offline using the open source software package Stimfit (Guzman et al., 2014).
Following recording, all cells were resealed by carefully withdrawing the patch-pipette in a manner akin to forming an outside-out patch. The slices were then fixed in 4% paraformaldehyde dissolved in 0.1 M phosphate buffer (PB) pH 7.35, overnight at 4°C. Slices were then transferred to 0.1 M PBS until processing for immunohistochemistry.
Confocal image stacks were collected on an invert scanning-confocal microscope (LSM 510 META Confocal with Axiovert 200, Zeiss) equipped with a 20Â (0.45 NA; Zeiss) air-immersion or a 63Â (1.4 NA; Zeiss) oil-immersion objective lens. For the identification of neurons, zstacks (1 mm steps; 1024 Â 1024 pixels) containing the somatodendritic axis and axon distribution were carried out. For neurochemical identification, the somata were imaged under high magnification, and immunolabeling was assessed over the somatic focal plane. All image analysis was performed with the FIJI package of ImageJ. For reconstruction, the neurons were imaged fully then stitched, segmented, and rendered offline in FIJI, using the Simple Neurite Tracer (SNT) plug-in (Longair et al., 2011).

Statistical analysis
All experiments and analysis were performed blind to genotype. Throughout this study, data are shown as the mean 6 SD, and the number of cells (n) and animals (N) are indicated. The required sample size was estimated based on an assumed effect size of 20% with 15% SD at 80% power, giving a required N = 7-10/group. Some data showed a .20% effect size, thus requiring fewer biological replicates (N). All data are reported as animal averages, unless stated otherwise, and are shown alongside estimation statistics of difference between means and confidence intervals (CIs) in the form (r value, effect size, 95% confidence interval, p value, test performed), the usefulness of which has been demonstrated previously (Manouze et al., 2019). We performed statistical comparisons of effect size in a paired or unpaired manner, using Student's t test, Mann-Whitney test, or Wilcoxon signed-rank test, depending on whether data were normally distributed, which was confirmed with the Anderson-Darling test. For group analysis, one-way ANOVA was performed. Statistical significance was assumed if p , 0.05.

Data availability
The data that support the findings of this study are available from the corresponding author on reasonable request.

NMDARs in hippocampal INs express variable levels of GluN2A
We next assessed the contribution of GluN2A subunits to synaptic NMDARs in identified hippocampal INs. We first sought to identify EPSCs in PV INs, which were located in and around stratum pyramidale, with large somata with multipolar dendrites. These were initially selected based on low membrane resistance and fast membrane decay times on breakthrough into the whole-cell configuration. For this study we identified 20 PV INs from 15 wildtype rats, and 14 PV INs from 13 GluN2A-null rats, which all displayed clear immunolabelling for PV. In wild-type rats, 50% (n = 10 cells) of PV INs were identified as BCs with axons localized to stratum pyramidale ( Fig. 2A) and 35% (n = 7 cells) had axons localized to stratum radiatum and stratum oriens, thus representing likely bistratified cells; the remaining three cells had axons cut close to the soma, thus preventing further identification. In GluN2Anull rats, 79% of recovered PV INs were BCs (n = 11 cells), 14% were bistratified (n = 2 cells), and one cell could not be identified because of a proximally cut axon.
Compared with wild-type rats, NMDAR-EPSCs recorded in PV INs from GluN2A-null rats had similar amplitudes of 51 6 46 pA (r = 0.54, À25.1; CI, À68.5, 18.3; p = 0.25, unpaired Student's t test) and rise times of 3.1 6 1.2 ms (r = 0.001, À0.06; CI, À0.91, 0.79; p = 0.89, unpaired Student's t test). Consistent with the contribution of GluN2A subunits to NMDAR-EPSCs evoked in PV INs, we measured a 75% increase in the decay time . Somatodendritic axis (black) and axonal arborization (red) are indicated. B, Representative EPSCs recorded from CA1 PyrCs from wild-type (black) and GluN2A-null (red) rats. AMPAR-EPSCs (inward currents) and NMDAR-EPSCs (outward currents) are shown. The peak scaled NMDAR traces are shown below to indicate the slowing of the NMDAR-EPSC. C, NMDAR-EPSC rise times, quantified for all recorded CA1 PyrCs. Data from individual cells are shown as filled circles from wild-type rats (black; n = 20 cells; N = 13 rats) and GluN2A-null rats (2A-null, red; n = 21 cells; N = 11 rats). D, NMDAR-EPSC decay time constants (tau) from the weighted tau of a biexponential curve fit. E, NMDAR/AMPAR ratio of EPSCs elicited by Schaffer collateral stimulation. Fewer rats are shown for kinetic values because of the 20 pA cutoff imposed on kinetic data. Data are shown as the mean 6 SD, alongside the difference between the means 6 CI. *p , 0.05, Student's t test.
A major subtype of hippocampal INs is composed of those INs expressing the neuropeptide CCK , which likely reflect caudal ganglionic eminence INs previously identified as containing high levels of NMDARs (Matta et al., 2013) and were proposed to express GluN2A mRNA (Perszyk et al., 2016). To determine the contribution of GluN2A to synaptic currents in these neurons, we performed recordings of AMPAR-EPSCs and NMDAR-EPSCs from identified CCK INs in wild-type rats (n = 56 cells, N = 27 rats) and GluN2A-null rats (n = 23 cells, N = 16 rats), which were all confirmed with pre-pro-CCK immunolabeling . The cells recovered comprised BCs (n = 15; Fig. 4A), Schaffer collateral-associated/apical dendrite-associated cells (SCA/ ADA, n = 21), and perforant path-associated (PPA; n = 6 cells) cells in wild-type rats; the remaining cells (n = 14) had their axon cut close to the cell body, preventing further classification. In GluN2A-null rats, we positively identified BCs (n = 7), SCA/ADA cells (n = 7), and PPA cells (n = 2); the remaining cells had a cut axon (n = 7) and, thus, could not be classified.
NMDARs are known to mediate Hebbian LTP, in both PyrCs (Malenka, 1991) and INs (Lamsa et al., 2005. Indeed, GluN2A/B-containing receptors have been shown to induce LTP (Berberich et al., 2005), with the GluN2A subunit presumed to be requisite for LTP induction (Liu et al., 2004). Given the absence of GluN2A from CCK INs, we asked whether the presence of GluN2D-containing NMDARs was sufficient to induce the Hebbian form of LTP present in these INs . In wholecell recordings from CCK INs, performed in current clamp, monosynaptic EPSPs were evoked from putative Schaffer collateral afferents in stratum radiatum in the presence of picrotoxin. The average amplitude of these EPSPs was 5.4 6 1.9 mV in wild-type CCK INs (N = 9 rats). Following 5 min of baseline recording, LTP was induced by delivering 2Â 100 Hz trains of stimuli (duration, 1 s) to the stimulated pathway, the EPSP amplitude was then measured 25 min after induction (Fig. 6C). In control recordings from wild-type CCK INs, this protocol resulted in potentiation of the EPSP to 59 6 30% increase from control EPSP at 25 min (r = 0.76, 3.4; CI, 1.9, 5.0; p = 0.001, paired Student's t test; Fig. 6D,E). In recordings from wild-type rats (N = 6 rats), NAB-14 (10 mM) was preapplied (10 min before recording), then CCK INs were recorded. The baseline EPSPs in the presence of NAB-14 had an average amplitude of 6.5 6 3.4 mV, which was not different from that of control EPSPs (r = 0.05, 1.1; CI, À1.9, 4.1; p = 0.45, unpaired Student's t test). Following the same induction as for control recordings, in the presence of NAB-14, little to no potentiation of the EPSP amplitude (6 6 37% increase) was observed at 25 min after tetanization (r = 0.004, 0.14; CI, À2.2, 2.5; p = 0.89, paired Student's t test; Fig. 6D,E), which was markedly lower than the potentiation observed in control recordings (r = 0.46, À57.7; CI, À95.0, À20.3; p = 0.005, unpaired Student's t test). To confirm that LTP in CCK INs was independent of GluN2A, we performed the same recordings in GluN2A-null rats. Following the same induction paradigm, we observed sustained synaptic potentiation in CCK INs from the GluN2Anull rat, with a 134 6 98% increase in EPSP amplitude under control conditions (N = 8 rats; r = 0.53, 7.4; CI, 1.1, 13.6; p = 0.008, Wilcoxon test; Fig. 6F,G). Contrary to LTP being reduced in the absence of GluN2A, the magnitude of EPSP potentiation was 48% greater in CCK INs of the GluN2A-null rat (r = 0.25, 76.0; CI, 3.1, 149.0; p = 0.042, unpaired Student's t test). In the presence of NAB-14, there was no potentiation of the EPSP when measured at 25 min in CCK INs from the GluN2A-null rats (N = 6 rats; 36 6 46%; r = 0.092, À0.56; CI, À2.6, 1.5; p = 0.51, paired Student's t test; Fig. 6F,G). These data reveal that LTP at Schaffer collateral synapses onto Figure 6. NMDAR-EPSCs in CCK INs are mediated by GluN2D, which is required for LTP induction. A, Representative NMDARmediated EPSCs recorded in CCK INs at 140 mV in the presence of CNQX at (black and red traces) or following 10 min wash-in of the GluN2D-negative allosteric modulator NAB-14 (10 mM; gray and pink traces). B, Plot of NMDAR-EPSC amplitudes before and after NAB-14 application in CCK INs. C, Quantification of NMDAR-EPSC amplitudes in CCK INs from GluN2A-null rats before and after NAB-14 application. D, Time course of EPSP amplitude in CCK INs from wild-type rats measured from À70 mV current clamp, following 2Â 100 Hz stimulation (double arrow) measured under control conditions (black circles) and in the presence of 10 mM NAB-14 (gray circles). Representative traces are shown above the chart, showing EPSP at baseline (black traces) and 25 min after LTP induction (gray traces). E, Quantification of LTP, reported as the percentage change in EPSP amplitude from baseline (dashed line) for wild-type CCK INs. Control LTP recordings (black circles; n = 9 cells; N = 7 rats) and those performed in the presence of 10 mM NAB-14 (gray circles; n = 5 cells; N = 5 rats) are shown. The number of tested rats is shown in parentheses. F, LTP induction in CCK INs from GluN2A-null rats according to the same scheme as in D. Data are shown for control recordings (red circles) and in the presence of NAB-14 (pink circles). Traces of each treatment are above the chart showing data at baseline (red) and after LTP induction (pink traces). G, Quantification of LTP induction in GluN2A-null rats under control (red; n = 6 cells; N = 5 rats) and in the presence of NAB-14 (pink; n = 5 cells; N = 4 rats). Data are shown as the mean 6 SD. *p , 0.05, from paired (B) and unpaired (D, F) Student's t tests.

Discussion
In the current study, we provide evidence for GluN2Aspecific modulation of synaptic NMDAR-mediated currents in hippocampal INs and PyrCs. We show that NMDAR-EPSCs in PV INs closely resemble those of CA1 PyrCs, in terms of kinetics, synaptic contribution, and GluN2A/2B composition. The closely related SSt IN cell type displays NMDAR-EPSCs that are typically faster and likely composed largely of GluN1/2A or GluN1/2A/2B receptors. Finally, we show unequivocally that CCK INs lack a prominent GluN2A component to their NMDAR-EPSCs; rather, their synaptic NMDARs are likely composed of GluN1/2B/2D triheteromeric receptors based on ifenprodil and NAB-14 sensitivity. Finally, we confirm that these NMDARs in CCK INs are required to induce a Hebbian form of LTP. Together, the data we present provide further evidence that different IN subtypes display divergent synaptic NMDAR pharmacology.

Functional identification of GluN2 subunits in hippocampal neurons
CA1 PyrCs have been postulated to express both GluN2A/B triheteromeric receptors and GluN2A or GluN2B diheteromeric receptors in mature neurons (Al-Hallaq et al., 2007;Traynelis et al., 2010;Gray et al., 2011;Tovar et al., 2013). In this study, we provide evidence that CA1 PyrCs likely express at least both GluN2B diheteromeric and GluN2A/B triheteromeric receptors at functional synapses in late juvenile rats. NMDAR-mediated EPSCs were significantly slowed in both rise and decay phase following the loss of GluN2A subunits, but did not have altered amplitudes. The observed 50-60% block produced by ifenprodil in wild-type PyrCs is consistent with the pharmacology of GluN1/2A/2B triheteromeric NMDARs observed in both heterologous cell lines (Erreger et al., 2005;Hatton and Paoletti, 2005;Marwick et al., 2019) and also in cultured dissociated neurons (Martel et al., 2009;McKay et al., 2012). Indeed, the increased block observed in the GluN2A-null rat is consistent with a switch to GluN1/2B diheteromeric receptors at synapses (Gray et al., 2011;McKay et al., 2012). Nevertheless, ifenprodil has been suggested to block GluN2A-containing receptors to some degree in cultured neurons (Kew et al., 1998;Williams, 2001); however, the IC 50 in NMDARs comprising only the Glu2A receptor is an order of magnitude higher than we used in this study (Avenet et al., 1996). Indeed, the ifenprodil block we observe is greater in the absence of GluN2A, consistent with a selective effect at GluN2Bcontaining NMDARs. Furthermore, a range of ifenprodil concentrations up to and including 10 mM have been used in ex vivo slice preparations, with overall findings consistent with GluN2B selective effects (Lei and McBain, 2002;Matta et al., 2013). Together, these data show in late juvenile rats that synaptic NMDA receptors are likely composed of a combination of GluN2A/B triheteromeric and GluN2B diheteromeric receptors, which accounts for the slower kinetics and tendency toward reduced NMDAR/AMPAR ratios at Schaffer collateral synapses in these neurons.
The role of NMDARs in the signaling, plasticity, and excitability of hippocampal INs is poorly understood, and that of the contribution of different NMDAR subunits to these functional properties even less so (Moreau and Kullmann, 2013;Akgül and McBain, 2016;Booker and Wyllie, 2021). Many studies have suggested that PV INs likely possess NMDAR-mediated EPSCs with approximately four-fold lower amplitudes compared with AMPAR-mediated EPSCs (Koh et al., 1995;Korotkova et al., 2010;Matta et al., 2013). In the neocortex, such NMDAR-EPSCs possess a GluN2A component in the neocortex (Picard et al., 2019) and display a low GluN2B content in the hippocampus of juvenile mice (Matta et al., 2013). We show that CA1 PV INs possess NMDARs of similar composition to those in CA1 PyrCs based on recordings in the GluN2A-null rat and in the presence of ifenprodil. These NMDAR-EPSCs have faster kinetic properties than neighboring CA1 PyrCs (Gray et al., 2011; see also current data), which in the absence of reduced synaptic amplitudes in the GluN2A-null rat may relate to the altered dendritic properties of PV INs compared with CA1 PyrCs (Nörenberg et al., 2010) or altered NMDA effects on such (Camiré and Topolnik, 2014). Our finding that the amplitude of NMDARs-EPSCs in PV INs is comparable to that of CA1 PyrCs, with similar effects of GluN2A loss and GluN2B blockade, offers an explanation for the presence of readily inducible Hebbian or anti-Hebbian LTP in the PV and SSt IN classes, reliant on the presence of NMDARs or calcium-permeable AMPARs, respectively (Laezza et al., 1999;Perez et al., 2001;Lamsa et al., 2005;Kullmann and Lamsa, 2007;Szabo et al., 2012;Billingslea et al., 2014). Indeed, activating these NMDARs likely recruits signaling cascades similar to those in PyrCs (Berberich et al., 2005;Gray et al., 2011), with a defined role in plasticity induction (Camiré and Topolnik, 2014). Our data suggest that, in rats, NMDAR-EPSCs may be mediated by a greater proportion of GluN2B-containing receptors than in mice, given the differential ifenprodil sensitivity observed (Matta et al., 2013), although this has some potential caveats (see above). The presence of large NMDAR-EPSCs, which are similar to CA1 PyrC synaptic NMDARs, may indicate that NMDARs underpin similar roles in terms of dendritic integration in PV INs (Branco et al., 2010;Cornford et al., 2019). While SSt INs share a common developmental origin with PV INs (Tricoire et al., 2011;Chittajallu et al., 2013) and also coexpress PV (Jinno and Kosaka, 2000;, our data indicate that through their differentiation they express functional NMDAR-EPSCs that are distinct from those of PyrCs and other PV IN morphotypes. While the single-channel conductance of NMDARs in OLM cells is similar to that of CA1 PyrCs (Hájos et al., 2002) and their dendritic filtering is rapid (Martina et al., 2000) and similar to that in PV INs (Nörenberg et al., 2010), our data indicate there are fundamental differences in the relative contribution of NMDAR subunits to EPSCs in these cells. In particular, it is plausible that SSt INs possess a more uniform population of GluN2A/2B triheteromeric receptors, first because of the ;30% block with ifenprodil (Hansen et al., 2014) and the longer decay kinetics of NMDAR-EPSCs we observe in the GluN2A-null rats. It has been shown in mice that GluN2D is expressed in PV and SSt INs in adult mice (Standaert et al., 1996;Engelhardt et al., 2015); however, given the rapid kinetics and high sensitivity of NMDAR-EPSCs to GluN2A loss and ifenprodil, our data suggest that this NMDAR subunit may not contribute to synaptic currents in older rats. Indeed, it is plausible that GluN2D is expressed in rat neurons but may contribute to nonsynaptic mechanisms. As LTP in putative SSt INs is dependent on NMDARs (Ouardouz and Lacaille, 1995), these receptors likely fulfill the typical role of NMDARs in LTP induction (Berberich et al., 2005) in this fundamental feedback interneuron subtype. One major consideration to the function of NMDARs in INs is that GluN2A subunits undergo a developmental increase in expression in PyrCs and INs alike (Flint et al., 1997;Martel et al., 2009;McKay et al., 2012;Matta et al., 2013;Wyllie et al., 2013). How this contributes to the circuit recruitment of INs has not been fully explored other than in a few IN subtypes (Matta et al., 2013). Indeed, whether or not GluN2D subunits undergo such a developmental trajectory in hippocampal INs, or indeed whether GluN2A subunits possess a delayed maturation in CCK INs remains unexplored.
We observed NMDAR-mediated currents in identified CCK IN cell types that were larger and slower than those observed in CA1 PyrCs. This is in good agreement with the presence of GluN2B/2D NMDARs, as described in stratum radiatum INs (Perszyk et al., 2016;Yi et al., 2019) and in GluN2D-containing receptors more generally (Wyllie et al., 1996;Misra et al., 2000;Jones and Gibb, 2005;Logan et al., 2007;Brothwell et al., 2008). The sensitivity of NMDAR-EPSCs in CCK INs to NAB-14 confirms the presence of GluN2C/D receptor subunits in synaptic receptors, with a reduced sensitivity to ifenprodil (Yi et al., 2019) compared with GluN2A/2B triheteromeric receptors. However, as GluN2C is not present in the hippocampus, the predominant receptor target is likely GluN2Dcontaining NMDARs. The absence of GluN2A-mediated EPSC effects is in good agreement with that in recordings performed in non-identified stratum radiatum INs, where a glycine-sensitive GluN2A antagonist did not alter NMDAR-EPSCs (Yi et al., 2019). Given that LTP was strongly attenuated by NAB-14 and the lack of CP-AMPAR-mediated LTP (Szabo et al., 2012) in identified CCK INs, our results support previous findings suggesting a central role of NMDARs in synaptic plasticity in these INs . Similarly, as LTP was readily induced in CCK-INs, we confirm previous work showing that GluN2D-containing NMDARs aid this induction (Hrabetova et al., 2000). Together, these data support the idea that GluN2A-containing NMDARs are not required for the induction of LTP in CCK INs (Berberich et al., 2005), but rather it is GluN2D-containing NMDARs that contribute to LTP expression.
Overall, our data extend the previous findings of the study by Perszyk et al. (2016), in which the reported mRNA corresponding to GluN2A, GluN2B, and GluN2D was abundant in each of the neurochemical cell types we have assessed. For example, this study shows that GluN2A receptor subunits do not appear to contribute to synaptic NMDAR-mediated currents in CCK-INs. While our data do not preclude the possibility that other NMDAR subtypes exist in all of the neuron classes tested, they indicate that such roles would be confined to other (nonsynaptic or non-ionotropic) functions, such as extrasynaptic modulation of plasticity (Ivanov et al., 2006), metabotropic modulation of inhibitory receptors (Chalifoux and Carter, 2010;Guetg et al., 2010), or mediating tonic excitation (Hanson et al., 2019). Furthermore, as we have only examined the major synaptic inputs to different classes of CA1 PyrCs and INs, synapse-specific effects of receptor composition may exist (Arrigoni and Greene, 2004;von Engelhardt et al., 2008). Indeed, recent data suggest that cell type-specific inputs may lead to withincell differences in synaptic composition of NMDARs (Lei and McBain, 2002). Our data suggest that most synaptic NMDARs contain a mixed contribution of receptor subtypes, which provides the tantalizing prospect that synapse and extrasynaptic heterogeneity of NMDARs exists to give rise to functional divergence. The data we present suggest that further work is required to disentangle the contributions of different NMDAR subtypes to synaptic and extrasynaptic functions in hippocampal INs.

Functional ramifications
Given the large heterogeneity in GluN2 subunit-specific effects on NMDAR-EPSCs in hippocampal INs, there is a wide variety of potential ramifications for human disease. First, given that GluN2A subunits contribute significantly to PV and SSt IN synaptic NMDARs, this heterogeneity emphasizes the need to further elucidate the role of these neurons in epileptic syndromes, as the GluN2A subunit may be a critical determinant of some forms of epilepsy (Marwick et al., 2019). Furthermore, as GluN2D de novo mutations have also been linked to epileptic encephalopathies (Li et al., 2016;Camp and Yuan, 2020), and given the clear role of this subunit in CCK IN synaptic function, the heterogeneity may explain the selective loss of these neurons in some forms of experimental epilepsy (Wyeth et al., 2010;Sun et al., 2014). These data suggest potential therapeutic avenues by which NMDAR function is selectively modulated in a brain region and in a cell type-specific manner. Furthermore, NMDARs contribute to the spike coupling of INs to ongoing synaptic activity (Matta et al., 2013). How divergent NMDAR kinetic properties between different IN subtypes contribute to ongoing circuit activity remains unknown, but will likely have direct ramifications for local information processing.
In summary, we provide clear evidence for GluN2A, GluN2B, and GluN2D subunit-containing NMDARs in neurochemically and morphologically defined INs and PyrCs of the rat. These findings provide insight into the mechanisms of synaptic plasticity generation in CCK INs, which rely on GluN2B-and GluN2D-containing, but not GluN2A-containing, NMDARs. These data are important to our understanding of how different IN classes integrate synaptic information.