d-serine availability modulates prefrontal cortex inhibitory interneuron development and circuit maturation

The proper development and function of telencephalic GABAergic interneurons is critical for maintaining the excitation and inhibition (E/I) balance in cortical circuits. Glutamate contributes to cortical interneuron (CIN) development via N-methyl-d-aspartate receptors (NMDARs). NMDAR activation requires the binding of a co-agonist, either glycine or d-serine. d-serine (co-agonist at many mature forebrain synapses) is racemized by the neuronal enzyme serine racemase (SR) from l-serine. We utilized constitutive SR knockout (SR−/−) mice to investigate the effect of d-serine availability on the development of CINs and inhibitory synapses in the prelimbic cortex (PrL). We found that most immature Lhx6 + CINs expressed SR and the obligatory NMDAR subunit NR1. At embryonic day 15, SR−/− mice had an accumulation of GABA and increased mitotic proliferation in the ganglionic eminence and fewer Gad1 + (glutamic acid decarboxylase 67 kDa; GAD67) cells in the E18 neocortex. Lhx6 + cells develop into parvalbumin (PV+) and somatostatin (Sst+) CINs. In the PrL of postnatal day (PND) 16 SR−/− mice, there was a significant decrease in GAD67+ and PV+, but not SST + CIN density, which was associated with reduced inhibitory postsynaptic potentials in layer 2/3 pyramidal neurons. These results demonstrate that D-serine availability is essential for prenatal CIN development and postnatal cortical circuit maturation.


SR is widely expressed in embryonic cortical interneurons (CIN).
To determine the effect of d-serine on CIN lineage, we quantified the expression of SR (Srr) and the obligatory NMDAR subunit GluN1 (Grin1) in cells that express Lhx6 (Fig. 1A-E). We show using multiplex fluorescent in situ hybridization that nearly all Lhx6 + INs express Srr and Grin1 during prenatal and early postnatal periods ( Fig. 1A-B). The outlined cells in the intermediate zone (IZ) at E16 (Fig. 1A) show co-expression of Lhx6, Srr, and Grin1 mRNA. Figure 1A,C shows that 89% and 73% of cells at E16 in the IZ are Lhx6+/Srr+ and Lhx6+/Srr+/Grin1+, respectively. We also observed a temporal increase in the co-expression of Lhx6+/Srr+/Grin1+ cells at PND9 (74%) and 16 (95%) in the PrL (Fig 1B,D,E). We show that d-serine and l-serine levels in the E15 forebrain (Fig. 1G) are equivalent to levels observed during the first postnatal week 20 . Furthermore, d-serine is enriched in migrating interneurons of the developing neocortex ( Fig. 1H-I).
SR −/− mice display fewer immature CINs in the embryonic cortex and have an altered density of GABAergic neurons in juvenile PrL cortex. Given the widespread expression of SR and GluN1 in Lhx6+ cells embryonically, we next sought to determine if genetically eliminating SR and thus reducing d-serine would alter IN development and the levels of GABAergic neurons in juvenile mice. Since the cortical interneuron progenitor cells are derived from the ventral ganglionic eminence, we used GABA and phospho-histone3 (p-H3; postmitotic marker) immunoreactivity to determine if there are changes in the distribution of progenitors in the GE. Compared to WT, SR −/− forebrains exhibit profound accumulation of GABA ( Fig. 2A,B) and a significant increase in p-H3 in the ganglionic eminence at E15 (t 4 = 3.63, p = 0.022; Fig. 2C-E). GABA immunoreactivity at E15 in the IZ ( Fig. 2A,B), and Gad1 mRNA at E18 (Fig. 2F) in the frontal cortex, were used to visualize immature INs in WT and SR −/− neocortex. We show markedly reduced GABA immunoreactivity in the neocortex of SR −/− mice ( Fig. 2A,B). We further found that the number of Gad1+ cells in SR −/− cortex at E18 was reduced by 20% compared to WT mice (t 6 = 2.67, p = 0.037; Fig. 2G). While functional impairment of GABAergic CINs has been proposed to be one of the major factors contributing to E/I imbalance in various psychiatric disorders, most of the attention has been directed towards PV + GABAergic INs dysfunction. At PND16, we found a ~ 30% reduction of both GAD67+ and PV+ cells in the PrL of SR −/− mice across all cortical layers (total Gad67: t 9 = 2.43, p = 0.038; PV: t 8 = 3.23, p = 0.012; Fig. 3A-C, Table 1). There was no significant reduction in the number of Sst+ cells in the PrL of SR −/− mice (total SST t 7 = 0.42, p = 0.687, Fig. 3A-C, Table 1). Given that half of all hippocampal neurons originate from the MGE, we also observed a significant reduction in Gad67+, PV+ and SST+ cells in the hippocampus ( Supplementary Fig. 1). In both the PrL and hippocampus (Supplementary Fig. 2A,B), we found no notable variation in PV + density between the SR −/− and WT at postnatal day 29.
Increased E/I ratio in layer 2/3 PrL pyramidal neurons of SR −/− mice. Since SR −/− mice had fewer PV + CINs in the PrL, we investigated the properties of excitatory synaptic transmission in layer 2/3 PrL of WT and SR −/− mice (Fig. 4A). We recorded compound PSPs using whole-cell current clamps at a holding potential of − 60 mV with current injection upon superficial layer 5 stimulation. For this experiment, the peak depolarization of the PSP was set to approximately 5 mV (WT: 5.20 ± 0.18 mV, n = 15; SR −/− : 5.11 ± 0.08, n = 19, t 32 = 0.52, p = 0.605) to draw out the inhibitory component of compound PSPs (Fig. 4A). We found a significant reduction in the IPSP component of the compound EPSP/IPSP (Fig. 4B, peak IPSP amplitude, t 32 = 4.04 p < 0.001). The decrease in IPSP amplitude resulted in an increased E/I ratio ( Fig. 4C; E/I ratio, t 32 = 3.84, p < 0.001). These results suggest that a selective GABAergic impairment in SR −/− mice leads to an increase in the E/I balance.
Increased intrinsic excitability in SR −/− PrL pyramidal neurons. Next, we analyzed the number of spikes elicited during 500 ms steps of somatically injected current and found a significant increase in the intrin- www.nature.com/scientificreports/ www.nature.com/scientificreports/ sic excitability of layer 2/3 PrL pyramidal cells in SR −/− mice compared to WT mice (Fig. 4D,E). This increased intrinsic excitability was associated with a significant decrease in the rheobase (WT: 64.3 ± 6.2 pA, n = 17; SR −/− : 37.7 ± 3.9 pA, n = 18, t 33 = 3.68, p < 0.001), but did not affect the resting membrane potential or the action potential threshold (Fig. 4F,G, Table 2). The magnitude of the increased intrinsic excitability SR −/− mice was mimicked in WT mice by adding the GABA A receptor antagonist picrotoxin (Fig. 4F), suggesting reduced tonic inhibition contributed to the increased firing of SR −/− neurons. Together, these data suggest that a reduction in inhibitory input onto layer 2/3 PrL pyramidal neurons in SR −/− mice increases the E/I balance resulting in enhanced synaptically driven neuronal excitability.

Discussion
The results presented in this study show that immature Lhx6 + CINs originating from the MGE robustly express Grin1 and Srr mRNA. We show that d-serine is critical for (1) the proper distribution of CINs during embryonic development and, (2) maintaining the proper density of PV + CINs in the juvenile PrL and hippocampus. This CIN deficiency in PND16 SR −/− mice was associated with disrupted pyramidal neuron inhibition and increased firing of layer 2/3 of the PrL pyramidal neurons. Overall, our data suggest that SR, which is highly expressed in www.nature.com/scientificreports/ CINs at the investigated time points, is important for regulating CIN development, maturation, and prefrontal (E/I) balance. Our findings in embryonic SR −/− mice are consistent with previous studies suggesting that NMDAR activation is important for maintaining MGE-derived IN production, migration, morphology, and synaptic connectivity. For example, treatment of E16 rat brain slices with NMDA, in the absence of Mg 2+ , increased Ca 2+ influx into migrating CINs in the IZ 12 . NMDARs are expressed and become active at E15 21,22 , which coincides with the second and more prominent stream of INs, mainly from MGE that rapidly migrate into the neocortex  www.nature.com/scientificreports/ through the IZ 23 . We show that d-serine is also expressed in the IZ at E15, and that d-serine elimination causes a reduction of GABA in the migratory stream ( Fig. 2A-B). We observed an increase in proliferation in the eminence of SR −/− mice which is similar to results reported in cell models with either genetic mutations of Grin2B or administration of a competitive NMDAR antagonist 24 . Reducing Grin2B mRNA by 50% or treating with (2-amino-5-phosphonovalerate (APV; NMDAR antagonist) and ifenprodil (GluN2B antagonist), increased the proliferation of human neural progenitor cells 24 . Finally, d-serine produced by primary cultured NSCs supports its proliferation and neuronal differentiation 25 . Altogether, these studies suggest that constitutive SR elimination resembles embryonic NMDAR hypofunction and disrupts prenatal IN development.
We show here that d-serine reduction causes a 30% reduction of PV + CIN density, increases E/I ratio, and firing of PrL pyramidal neurons at PND16 (Fig. 3), which supports the hypothesis that PV + CINs are essential for maintaining the E/I ratio. Our findings are also supported by studies that show that activation of PV + CINs in the mPFC rescues E/I imbalance and social deficits 2 . Furthermore, a 25% reduction of adult PV CIN density in the PFC was sufficient to reduce local GABAergic transmission onto pyramidal neurons, disrupt prefrontal E/I balance, and alter the processing of afferent information from the ventral hippocampus 26 . Although we did not observe any changes in PV+ cell density at PND29 (Supplementary Fig. 2A), we cannot rule out changes in PV function. Previous studies have shown that selective knockdown of the NMDAR subunit GluN1 in the mPFC PV+ interneuron decreases its intrinsic excitability 27 . Furthermore, adult SR knockout mice showed no changes in PV density in the prelimbic cortex 28 . However, these mice showed a decreased preference for social novelty and reduced social investigation-gamma oscillation in the frontal cortex, which PV controls 29 . Together, our data suggest that there might be a delay in the maturation of these PV+ cells in the SR −/− PrL.
NMDAR hypofunction has been postulated to be central to the pathophysiology of neurodevelopmental disorders that share cognitive and social deficits 30 . However, the timing and cell-type specificity of NMDAR hypofunction needs further investigation. Conditional genetic elimination of GluN1 on Nkx2.1 + cells (MGE-Grin1 fl/ fl ) in the MGE causes widespread downregulation of synaptic transcriptional programs in PV+ and Sst + CIN subtypes at PND20 and reduces the number of PV+ neurons in the frontal cortex at PND30 13 . Furthermore, NMDAR activation during PND7-9 is essential for maintaining inhibitory synapse density at PND7 and synaptic activity of PND21 CINs 31 . Interestingly, post-adolescent deletion of Grin1 did not affect IN density, firing rate, or synchronous firing of cortical excitatory neurons, suggesting that early postnatal inhibition of NMDAR activity on INs is essential for the development of pathology 32 . In a study involving a mouse model of autism spectrum disorder (neuroligin 3 R451C knockin mice), d-Cycloserine (DCS, a partial agonist that binds at the same site as D-serine) was found to improve PV dysfunction 27 . Additionally, direct infusion of DCS into the mPFC during adolescence resulted in rescued deficits in social novelty preference in adults 27 . Further research is needed to establish the impact of administering d-serine during prenatal and early postnatal stages on the PV + cell density. Lastly, our findings are consistent with the effect of GABAA receptor antagonists on pyramidal neurons (Fig. 4) and suggest that d-serine is vital for PV development and activity in the PrL.
We hypothesize that D-serine availability during development may impact the PV and SST neuronal groups differently, and the region-specific diversity may explain the selective vulnerability of the PV+ population we observe in the cortex 33 . Firstly, the generation of SST + neurons takes place in the dorsal MGE and reaches its peak at E14.5. On the other hand, PV + neurons originate in the ventral MGE and attain their maximum only at E15.5 33 . We predict that we might observe changes in SST + neurons at earlier time points in the prelimbic cortex. Secondly, PV + interneurons comprise around 40-50% of the GABAergic interneuron population in the neocortex, while SST + interneurons contribute 30% of GABAergic interneurons. On the other hand, in the hippocampus, PV + interneurons represent 30% of the population, while SST + interneurons represent 50% of GABAergic interneurons 33 . Finally, there might be distinct expression of NMDAR subunits in PV vs. SST neuronal populations that are differentially affected by d-serine availability at these times 34 . It would be interesting to evaluate whether there are alterations in other interneuron populations such as calbindin+, calretinin+, neuropeptide Y (NPY)+, vasoactive intestinal peptide (VIP)+ and cholecystokinin (CCK)+ interneurons that may serve as compensatory mechanisms in the SR −/− PrL.
Studies suggest that NMDAR signaling regulates the MGE-derived IN subtypes in part by innate genetic programs such as, Lhx6 transcription factor. Lhx6 gives rise to PV and SST cells that populate the PFC and hippocampus. Our results are supported by scRNAseq and Ribotag-seq data from MGE-Grin1 fl/fl mice that show a significant reduction in the expression and translation of Lhx6 mRNA and its downstream targets in hippocampal IN subtypes and not cortex 13 . In this study, there were differences in the PV and SST subtype clusters in the cortex compared to the hippocampus at PND20. Interestingly, adult knockdown of Lhx6 in the hippocampus did not affect PV density and firing, suggesting that Lhx6 activity during development is essential for IN subtype www.nature.com/scientificreports/ expression and function 9 . Although we did not observe any changes in Lhx6 mRNA in E17 SR −/− forebrain (data not shown), we cannot rule out MGE-specific changes in expression and activity or any changes in postnatal expression of Lhx6. Future work is needed to uncover how d-serine and NMDAR signaling regulates the abundance and diversity within PV/SST subclasses and other IN subtypes and to identify the signaling pathways downstream of d-serine and NMDAR activation. Overall, our study demonstrates that d-serine availability is essential for interneuron number in the neocortex during prenatal development and proper circuit maturation in the PrL. Further, we show that SR deficiency results in GABAergic dysfunction through a loss of GABAergic synapses. Future studies will also aim to determine the specific impact of SR deletion in CIN progenitor cells on interneuron density, inhibitory synapses and E/I balance. The clinical implications of these findings are that NMDAR hypofunction due to reduced  Postnatal brains. At postnatal day (PND) 9 and 16, male mice were deeply anesthetized with isoflurane administered in a small chamber, then briefly intracardially perfused with cold 1 × PBS (0.5 M PB, NaCl, pH 7.4), followed by 4% paraformaldehyde (Electron Microscopy Sciences, cat# 19202), and then cryoprotected in 30% sucrose/PBS at 4 °C. For in situ hybridization, brains were sectioned at 14 μm and directly adhered on glass slides that were stored at -80 °C. For immunofluorescent staining, sections were cut at 30 μm using a Leica SM 2010R Microtome and floating brain sections were stored in a cryoprotectant solution (ethylene glycol, glycerol, 0.5 M PB, NaCl, KCl, in dH20) at − 20 °C.
Immunofluorescence staining. Brain sections were washed three times in 1 × PBS, incubated in blocking buffer (10% goat serum, 1% BSA and 0.1% triton in PBS) for an hour, and then incubated overnight with primary antibodies in blocking buffer at 4 °C. The primary antibodies used were rabbit anti-GABA (cat# A2052; 1:100), rabbit anti-phospho-H3 (cat# 06-570; 1:150), guinea pig anti-parvalbumin (cat# 195004; 1:500), guinea pig anti-somatostatin (cat# 366004;1:300), and mouse anti-GAD67 (cat# MAB5406; 1:1000). Sections were incubated in appropriate secondary antibody in blocking buffer for an hour at room temperature. The secondary antibodies used were anti-mouse Alexa 488 conjugate, anti-rabbit Alexa 488 conjugate, anti-guinea pig Alexa 488, anti-goat Alexa 647 conjugate. Sections were washed three times with 1 × PBS, incubated in Hoechst (cat# 3570), washed three times with 1 × PBS, mounted, and cover slipped using ProLong ® Gold anti-fade medium. We utilized plate 3 for our experiments 35 .  36 . Cleared brains were immunolabeled in SmartLabel (LifeCanvas Technologies, MA, USA) for 24 h using the following primary antibodies that against goat anti-PV (ab32895). Fluorescently conjugated secondary antibodies against appropriate species were applied. 36 After immunolabeling, samples were incubated in 50% EasyIndex (RI = 1.52, LifeCanvas Technologies) overnight at 37 C followed by 1 day incubation in 100% EasyIndex for refractive index matching. After index matching the samples were imaged using a SmartSPIM axially-swept light sheet microscope using a 3.6x (0.2 NA) (LifeCanvas Technologies). Samples were registered to the Allen Brain Atlas (Allen Institute: https:// portal. brain-map. org/) using an automated process (alignment performed by LifeCanvas Technologies). A NeuN channel for each brain was registered to an average NeuN atlas (generated by LCT using previously-registered samples). Registration was performed using successive rigid, affine, and b-spline warping algorithms (SimpleElastix: https:// simpl eelas tix. github. io/). Automated cell detection was performed by LifeCanvas Technologies using a custom convolutional neural network created with the Tensorflow python package (Google). The cell detection was performed by two networks in sequence. First, a fullyconvolutional detection network (https:// arxiv. org/ abs/ 1605. 06211 v1) based on a U-Net architecture (https:// arxiv. org/ abs/ 1505. 04597 v1) was used to find possible positive locations. Second, a convolutional network using a ResNet architecture (https:// arxiv. org/ abs/ 1512. 03385 v1) was used to classify each location as positive or negative. Using the previously-calculated Atlas Registration, each cell location was projected onto the Allen Brain Atlas in order to count the number of cells for each atlas-defined region.

Quantification of CIN densities in PrL.
The number of PV, Sst, and GAD67 neurons in PND16 WT and SR −/− PrL were quantified using manual counts by a trained researcher blinded to genotype using the StereoInvestigator software (MBF Bioscience). At least four sections were analyzed from each mouse (n = 4-6; section interval = 6) matched from rostral to caudal positions and spanning approximately − 2.95 mm through 1.42 mm bregma (anterior/posterior). The cells were counted at 20 × using exposure and image settings that were consistent across all sections. Density was calculated by dividing the number of counted cells by the total area counted.