AMPA receptor deletion in developing MGE-derived hippocampal interneurons causes a redistribution of excitatory synapses and attenuates postnatal network oscillatory activity

Inhibitory interneurons derived from the medial ganglionic eminence represent the largest cohort of GABAergic neurons in the hippocampus. In the CA1 hippocampus excitatory synapses onto these cells comprise GluA2-lacking, calcium-permeable AMPARs. Although synaptic transmission is not established until early in their postnatal life, AMPARs are expressed early in development, however their role is enigmatic. Using the Nkx2.1-cre mouse line we genetically deleted GluA1, GluA2, GluA3 selectively from MGE derived interneurons early in development. We observed that the number of MGE-derived interneurons was preserved in mature hippocampus despite early elimination of AMPARs, which resulted in >90% decrease in spontaneous excitatory synaptic activity. Of particular interest, excitatory synaptic sites were shifted from dendritic to somatic locations while maintaining a normal NMDAR content. The developmental switch of NMDARs from GluN2B-containing early in development to GluN2A-containing on maturation was similarly unperturbed despite the loss of AMPARs. Early network giant depolarizing potential oscillatory activity was compromised in early postnatal days as was both feedforward and feedback inhibition onto pyramidal neurons underscoring the importance of glutamatergic drive onto MGE-derived interneurons for hippocampal circuit function.

A reduction in both feedforward and feedback inhibitory drive following loss of GluA1-3. The elimination of GluA1-3 made it difficult to reliably evoke Schaffer-collateral mediated synaptic events onto MGE-derived interneurons (Fig. 5A,B). When events were detected the kinetics of the residual effects did not differ from WT (Fig. 5B,C). In agreement with sEPSC data (Fig. 2), the eEPSC decay time constant was comparable across groups (WT: 3.14 ± 0.15 ms, n = 8 cells from 4 animals versus KO: 3.25 ± 0.38 ms, n = 15 cells from 6 animals, P17-21; P = 0.59, Mann-Whitney Test). As expected from the near elimination of the AMPAR-mediated component the NMDAR:AMPAR ratio was markedly increased compared to WT (WT: 0.33 ± 0.04, n = 14 cells from 10 animals versus KO: 3.81 ± 1.43, n = 14 cells from 5 animals; Fig. 5C). The EPSC paired pulse ratio (PPR) showed an unexpected increase in the KO population compared to WT (PPR WT , 1.32 ± 0.08, n = 10 cells from 4 animals versus PPR KO , 1.77 ± 0.12, n = 20 cells from 6 animals; P = 0.01, Mann-Whitney Test) (Fig. 5C), suggesting that elimination of AMPARs during synapse development may have direct influence over the establishment of the release probability of the presynaptic terminal. A result similar to our observations following elimination of NMDARs in CGE-derived interneurons 13 .

Discussion
Activity, including excitatory synaptic activity, is an important factor that regulates cell survival 12,25,26 , migration 27,28 , synapse number 12 and the nature of cellular connectivity 13 in the developing brain. Activity-dependent gene expression typically relies on transient elevations of intracellular Ca 2+ -entry arising through voltage gated Ca 2+ -channels. MGE-derived inhibitory interneurons, in addition to expressing voltage gated Ca 2+ -channels express GluA2-lacking, Ca 2+ -permeable AMPARs 3,5,[29][30][31][32] , that may provide a pivotal alternative route for Ca 2+ -entry to link environment and intracellular regulation of activity driven gene expression and development in interneurons. The expression of AMPAR subunits, at time points well in advance of migration termination and the establishment of synaptic sites, is consistent with roles other than synaptic transmission for these receptors Figure 6. The GluN2B-to-2A developmental switch occurs independently of AMPARs. Schaffer collateral evoked NMDAR-mediated synaptic responses (Vhold = + 50 mV) were recorded from WT (black) or GluA1-3 KO MGE (green) interneurons in CA1 hippocampi at P5-9 (A) and P17-21 (B) in the presence or absence of the GluN2B specific NMDAR antagonist ifenprodil (smaller traces indicated by red arrow) (all recordings done in the presence of picrotoxin (50uM) and DNQX (10uM)). The group data showed a significant decrease in decay tau (C) of NMDAR-mediated EPSCs between P5-9 to P17-21 in both WT and KO groups. NMDAR-mediated events became significantly less sensitive to ifenprodil from P5-9 to P17-21 (D,E). However, the loss of GluARs was without any major impact on the kinetic change or sensitivity of ifenprodil across the developmental window tested. *P < 0.05, **P < 0.005, ***P < 0.0005. www.nature.com/scientificreports www.nature.com/scientificreports/ early in development. The Nkx2.1-Cre mouse line allowed us to eliminate GluA1,2 &3 AMPAR subunits specifically from MGE-derived interneurons early in their development. Surprisingly, although early removal of AMPARs resulted in many functional changes, their elimination had little impact on numbers or positioning of surviving cells. This is in sharp contrast to others who have shown that pharmacological intervention of AMPARs in developing neocortex and hippocampus leads to both interneuron migration and morphological defects [9][10][11] . The reason for these discrepant results is at present unclear but likely point to limitations in the use of pharmacological intervention to unequivocally determine the roles of specific glutamate receptors in interneuron migration and development.
Electrophysiological interrogation of MGE-derived interneurons in GluA1,2 &3 loss of function mice confirm the near absence of AMPAR-mediated sEPSCs (by 94% in neonatal, and by 90% in young adult). The low frequency (~5-10% of WT) of sEPSCs remaining in both age ranges tested might suggest either that the Cre/ loxP dependent knock-out is not 100% penetrant 13 or that the remaining events represent GluA4 homomeric AMPARs. GluA4 receptor expression is developmentally regulated primarily in PV-containing interneurons and does not occur until the second postnatal week of life 4,30,33 . The slight increase in frequency of remaining sEPSCs in recordings from the older age group is consistent with the postnatal upregulation of GluA4. However, the GluA4 subunit typically endows rapid kinetic properties to AMPARs 34,35 , and the events that remained did not possess more rapid decay time constants compared to their P5-9 counterparts (P5-9: 1.45 ± 0.07 ms, n = 5, versus 1.83 ± 0.05 ms, n = 8, P = 0.23, Mann-Whitney Test; Fig. 2C).
Unlike principal neurons which receive excitatory synaptic input only onto dendritic locations, particularly spines, excitatory synapses onto interneurons are found on both soma and smooth (non-spiny) dendrites 6 . Although few studies have analyzed the total number of excitatory synapses onto particular interneuron subtypes, rat PV-containing interneurons receive 5-10x more excitatory inputs than their CGE-derived counterparts 36,37 . The density of excitatory inputs onto PV-containing interneuron dendrites is almost 4-fold greater than seen on the soma, and it is unclear whether distinct afferents target somatic versus dendritic locations. In hippocampus and cerebellum, recordings from the dendrites of principal neurons show larger EPSPs with slower kinetics compared to events recorded at somatic locations [38][39][40][41] . These reports highlight that the properties of excitatory synapses distributed through soma and dendrites of neurons vary proportionally with the distance from the action potential initiation site to coordinate the input-output relationship of the neuron. It is unclear whether such an arrangement exists in interneurons, which are comparatively electrotonically compact. Interestingly, in our study, excitatory synapses were redistributed away from dendritic sites toward somatic locations in the GluA1,2&3 KO, which may be a compensation for the lack of total excitation being received by the cell. Synapse relocation toward somatic sites coupled to the observed changes in paired pulse ratio and probability of release may act to increase the likelihood of action potential firing in KO neurons in the face of reduced afferent excitatory drive.
It is well established that MGE-derived interneurons are major participants of oscillatory activity in early postnatal hippocampus 1,14,16 . We have previously shown that optogenetic inactivation of MGE interneuron activity, but not CGE interneurons eliminates ongoing GDP activity 14 . The role played by synaptic AMPARs on interneurons in generating GDPs however has not been explored. In this study the frequency and amplitude of sEPSCs received by neonatal MGE was higher (2.15 Hz, ~40pA), than we observed onto neonatal CGE interneurons at an equivalent (0.67 Hz, ~30pA) 12 suggesting a more prominent role for excitatory drive onto MGE-derived interneurons at this age. Consistent with this observation GDC frequency was reduced by ~50% in MGE KO but unchanged in CGE KO hippocampi at P5-6 confirming that excitatory input onto MGE-but not CGE-derived interneurons is a major driver of synchronized network activity in neonatal hippocampus. Interestingly, the time course of developmental downregulation of GDC activity was similar in both WT and MGE-KO suggesting a diminished role for glutamate receptor drive onto MGE-derived interneurons as this developmental epoch progresses.
In the present study the I/E ratio of feedforward inhibitory drive was reduced by >80% following AMPAR elimination, consistent with a large reduction in excitatory drive onto MGE-derived interneurons. In our previous study, the loss of AMPARs from synapses onto CGE-derived interneurons reduced the I/E ratio by a similar amount (~76%) 12 . It is difficult to reconcile these two numbers since they should not be treated as only representing the arithmetic subtraction of MGE (e.g. PV-containing) versus CGE-(e.g. CCK-containing) derived interneuron contributions to inhibitory tone 42 for a number of reasons. In both of these studies although the primary change was a loss of excitatory drive onto the MGE-(this study) or CGE-inhibitory 12 subpopulations, the numerous other changes observed also likely contribute to the overall recruitment of inhibitory drive. Specifically across these two studies we have also shown that elimination of AMPARs changes cell number and cell subpopulation survival, alterations in cell morphology and axonal arborization, changes in presynaptic release probability, and anatomical redistribution of synaptic sites across the somatic-dendritic axis, all of which will impact the recruitment of the remaining interneuron populations, complicating a meaningful direct comparison of the magnitude of inhibition recruit in the two studies.
Recent reports of de novo mutations in glutamate receptor subunits (e.g., Gria in five individuals with learning disabilities and autism [43][44][45][46] points to a link between glutamate receptors and channelopathies that lead to neurodevelopmental disorders. Similarly, somatic mutations during embryonic development are recognized as an underpinning factor for neurodevelopmental disorders 47,48 . However, these recent insights have tended to focus on the large pyramidal neuron population of the hippocampus and cortex. As interneurons derive from progenitors in distinct embryonic brain regions 49,50 (i.e., CGE-and MGE-derived interneuron subgroups), it is reasonable to hypothesize that they may differentially accumulate numerous somatic mutations that would have multiple disparate consequences for circuit activity and brain function. Here, we establish a role for AMPARs in synaptic maturation in MGE-derived interneurons that contrasts with previous reports of AMPAR subunit deletions in pyramidal neurons 51 , and CGE-derived inhibitory interneurons 12 . Taken together these reports suggest that genetic permutations depending on what progenitor pool they occur in will result in unique morphological, www.nature.com/scientificreports www.nature.com/scientificreports/ physiological and behavioral outcomes. Since we have the opportunity to make specific base editing on the genome with CRISPR system, we can now study the relationship between identified/proposed mutations and malfunctions of glutamate receptors in animal models in a cell type specific manner.

Materials and Methods
All animal studies have been approved by, and all methods were performed in accordance with the guidelines and regulations set by the National Institutes of Health's Institutional Animal Care and Use Committee (IACUC).
Histology. Tissue collection. For cell number-, and for synapse-quantification mice at P21, and P35-40 were used respectively as described in Akgul, et al. 12 tdTomato fluorescence was used to identify MGE-derived interneurons in hippocampus of 50 μm coronal sections prepared from the dissected brains, without antibody labelling.
Recombinant virus production. An AAV vector carrying the coding sequence of PSD-95 with an enhanced green fluorescent protein (EGFP) gene at the C terminus was cloned as described in Akgul and Wollmuth 54 . A FLEX-rev-PSD-95-EGFP_AAV was produced by the University of North Carolina Gene Therapy Program Vector Core.
Stereotaxic injections into hippocampus were performed as described in Akgul, et al. 12 Two control and two KO mice (P19-22) were anesthetized with isofluorane and 0.4 μl of AAV was injected at a rate of 0.1 μl/min into the hippocampus at to the following stereotaxic coordinates: −2.0 mm anterior to lambda; 2.0 mm lateral from the midline; 1.5 mm down from the dural surface, using a stereotaxic apparatus (Stoelting, Wood Dale, IL, USA). Mice were sacrificed at least two weeks after viral injection.
Imaging and quantification were performed as described in Akgul, et al. 12 For cell number quantification, 4-6 sections of 50 μm thickness were used per animal. Sections were collected using systematic-random sampling 15 , with 5 animals per group. A Zeiss LSM Confocal microscope within the Microscopy and Imaging Core, NICHD, was used to capture fluorescent images of the hippocampi with a 10X objective. For synapse number quantification 2 mouse brains from each group (WT and KO) were dissected 14-15 days post-injection. 4-5 mounted sections without antibody labelling were imaged using a Zeiss LSM 880 with Airyscan at the Microscopy and Imaging Core, NICHD, with 63X objective. The EGFP signal on tdTomato positive dendrites and soma were quantified by using Spot Detection tool of IMARIS software version 8 (Bitplane, Zurich, Switzerland). electrophysiology. Acute slices were prepared from mice between postnatal day 5-10 (P5-10, age group for GDC recordings), P5-9 (neonatal) and P17-21 (juvenile) for whole cell patch clamp electrophysiology as described in 12 . Picrotoxin (50 μM) was also used to isolate sEPSCs and eEPSCs. NMDAR-mediated EPSCs were isolated by inclusion of the AMPAR antagonist DNQX (10 μM), in addition to picrotoxin, and confirmed using the NMDAR antagonist DL-APV (100 μM). GDC's were recorded with no blockers in the bath. They were completely blocked by picrotoxin (data not shown). Multipeak, outward GDC events were detected using built-in event detection protocol with a set threshold of 100 pA amplitude and 100 ms duration in Clampfit version 10.7. The wildtype control dataset for both the feedforward and feedback inhibition experiments is replotted from our earlier study 12 since all experiments were originally performed by interleaving recordings from wildtype, MGE-KO and CGE-KO animals.

Solutions
All data were filtered at 3 kHz and acquired at a sampling rate of 10 kHz using pClamp9.2 (Molecular Devices, Sunnyvale, CA, USA).

Data availability
The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.