Introduction

Glutamate mediates most of the fast excitatory transmission in the brain by activating primarily two types of ionotropic glutamate receptors: α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid (AMPA) and N-methyl-d-aspartate (NMDA) receptors. Whereas AMPA receptors (AMPARs) depolarize the postsynaptic neuron to set off firing, NMDA receptors (NMDARs) cause a large influx of calcium, which induces synaptic plasticity (Cull-Candy et al. 2001; Huganir and Nicoll 2013). Several subunits have been distinguished in each subfamily, and different subunit combination generates great diversity in biophysical properties, making them highly suitable for meeting complex functional requirements (Isaac et al. 2007; Kullmann and Lamsa 2007). In the neural circuit, these receptors are differentially allocated to synapses according to various factors, including presynaptic inputs, target cell types, and synaptic configurations, thereby extending the computational properties of neurons (Toth and McBain 1998; Brunel et al. 2004; Nicholson et al. 2006). In this review, I first provide a short overview of the morphological techniques required for analyzing the cellular and subcellular distributions of ionotropic glutamate receptors, and then discuss the cell type- and input-pathway-dependent allocations of synaptic glutamate receptors, focusing mainly on our recent work.

Technical considerations

Single-cell RT-PCR and isotopic in situ hybridization (ISH) studies have provided useful information about the expression patterns of ionotropic glutamate receptor subunits. ISH has been one of the most powerful methods available to examine the spatiotemporal pattern of gene expression. Although the original methods for ISH using radiolabeled probes are highly effective in detecting very low levels of transcripts (John et al. 1969), their major limitations are relatively poor cellular resolution and long exposure time. The more recently developed fluorescence ISH (FISH) protocol, using tyramide signal amplification (TSA), has overcome these limitations and proven to be efficient and useful. It allows for the detection of virtually any combination of two or three mRNAs and for histochemical labeling in combination with immunofluorescence and tracer labeling, on the same sections. This facilitates the neurochemical identification of cellular and neuronal populations expressing the mRNA of interest.

However, simply knowing the mRNA expression level of receptors is not sufficient for making functional predictions, because neurons bear elaborated dendritic arbors to receive synaptic inputs from multiple sources, and receptors are not distributed evenly on the cell surface. Thus, the precise subcellular distribution pattern, including the density in each subcellular compartment and relationship with various synaptic inputs, must be determined. In clear contrast to intracellular and extrasynaptic receptors, synaptic AMPARs and NMDARs are concentrated at the postsynaptic density (PSD), and are not detected readily by conventional immunofluorescence and preembedding immunoelectron microscopy (Fritschy et al. 1998; Watanabe et al. 1998; Fukaya and Watanabe 2000). Therefore, detection of synaptic AMPARs and NMDARs requires antigen-unmasking procedures including protease (e.g., pepsin) pretreatment or postembedding immunogold electron microscopy (EM), both of which have been proven to yield compatible results (Ottersen and Landsend 1997; Fukaya and Watanabe 2000; Fukaya et al. 2006). Of the methods used currently, postembedding immunogold EM seems to be the most reliable and convenient for localizing and quantifying postsynaptic ionotropic neurotransmitter receptors (Ottersen and Landsend 1997; Petralia and Wenthold 1999). In postembedding immunogold EM, immunolabeling takes place on the very surface of ultrathin sections where antigens are easily accessible, allowing the high-resolution localization and quantification of receptors. Double labeling with appropriate marker proteins is quite efficient in determining the neurochemical identity of pre- and post-synaptic elements, making it possible to investigate the basic principles governing the subcellular distribution of ionotropic glutamate receptors.

Target cell type- and input pathway-dependent regulation of AMPAR expression

Qualitative difference in AMPAR expression

AMPARs are hetero- or homo-tetramers derived from subunits GluA1–4 (GluR1–4; GluRA–D) (Borges and Dingledine 1998; Collingridge et al. 2009). The subunit composition of AMPARs greatly influences the mode of synapse trafficking as well as their biophysical properties (Malinow and Malenka 2002; Bredt and Nicoll 2003; Huganir and Nicoll 2013). AMPARs that lack GluA2 are permeable to Ca2+ ions, exhibit a high single-channel conductance and display an inwardly rectifying I–V relationship as they are blocked by endogenous intracellular polyamines at positive potentials (Bowie and Mayer 1995; Koh et al. 1995a). The majority of AMPARs in the central nervous system (CNS) are GluA2-containing AMPARs (Wenthold et al. 1996; Greger et al. 2002). In the forebrain, GluA1 and GluA2 are the predominant subunits, with low levels of GluA3 and GluA4 (Monyer et al. 1991; Craig et al. 1993; Geiger et al. 1995; Tsuzuki et al. 2001; Sans et al. 2003). Thus, it seems that the preferred subunit combination in pyramidal cells is GluA1/A2 with additional GluA2/A3 (Geiger et al. 1995; Wenthold et al. 1996; Tsuzuki et al. 2001; Sans et al. 2003; Lu et al. 2009). In contrast, GluA2-lacking Ca2+-permeable AMPARs are expressed preferentially in GABAergic interneurons and play crucial roles in synaptic plasticity and neuronal development (McBain and Dingledine 1993; Geiger et al. 1995; Koh et al. 1995b). Thus, synaptic AMPAR subunits show distinct, yet often overlapping, expression patterns that allow for subunit-specific function and regulation.

Consistent with previous studies (Keinanen et al. 1990; Pellegrini-Giampietro et al. 1991; Lambolez et al. 1992; Petralia et al. 1997), our FISH analyses have shown that cerebellar Purkinje cells (PC) express high levels of GluA2 and GluA3 mRNAs, with low levels of GluA1 mRNA (Fig. 1e–g), whereas molecular layer interneurons (MLIs; i.e., basket and stellate cells) express GluA2–4 mRNAs (Fig. 1f–h). These distinct combinations are also observed in quantitative immunogold EM analyses using subunit-specific antibodies for GluA1–4 (Fig. 2c–f). Considering that AMPAR-mediated EPSCs in PCs show little rectification and low Ca2+ permeability (Tempia et al. 1996; Momiyama et al. 2003), the majority of AMPARs there should contain GluA2. Therefore, it seems that the major AMPAR configuration in PCs is GluA2/A3 channels with additional GluA1/A2 channels. By contrast, synapses on MLIs exhibit inwardly rectifying EPSCs under resting conditions, indicating a predominance of GluA2-lacking Ca2+-permeable AMPARs (Liu and Cull-Candy 2000). These AMPARs consist predominantly of GluA3 homomeric channels (Keinanen et al. 1990; Sato et al. 1993) with the additional participation of GluA4 subunits (Gardner et al. 2005). However, high frequency activity at these “Ca2+-permeable synapses” produces a rapid switching from largely GluA2-lacking to GluA2-containing AMPARs (Liu and Cull-Candy 2000; Kelly et al. 2009). Thus, each cell type expresses distinct sets of GluA subunits with differential relative abundance, presumably reflecting cell type-dependent differences in the preferred subunit composition of AMPA receptors.

Fig. 1a–h
figure 1

Fluorescent in situ hybridization (FISH) showing distinct patterns of GluA mRNA expression in cerebellar Purkinje cells (PC) and molecular layer interneurons (MLI). a–d Single FISH for GluA1–4 in the adult mouse brain. Hybridization with sense probes gives no significant signals (insets). e–h Double FISH for GluA1–4 (green) and GAD67 (red) mRNAs in the cerebellar molecular layer. GluA1 mRNA is weak in GAD67 mRNA-positive large cells in the PC layer (PL; asterisks), and almost absent in GAD67-positive small cells in the molecular layer (ML; arrows) (e). GluA2 mRNA is expressed at consistently high levels in PCs, whereas it is expressed at variable levels in MLIs (f). GluA3 mRNA is expressed in PCs and MLIs (g). GluA4 mRNA is expressed moderately in MLIs (h). Small cells expressing high levels of GluA1 and GluA4 mRNAs are identified as Bergmann glia cells, because these hybridizing signals overlap with those for the glutamate transporter GLAST, which are enriched in these glia (data not shown). OB Olfactory bulb, Ctx cerebral cortex, St striatum, Hi hippocampus, Th thalamus, Mb midbrain, Cb cerebellum, MO medulla oblongata, ML molecular layer, PL Purkinje cell layer. Bars: (in d) a–d 1 mm; (in h 2) e–h 10 μm. Reproduced with permission from Yamasaki et al. (2011)

Full size image
Fig. 2a–f
figure 2

Postsynaptic labeling for GluA1–4 at three types of cerebellar synapses in wild-type mice. Quantitative immunogold EM analyses on the labeling density of GluA1–4 at three types of excitatory synapses in the cerebellar molecular layer. a Double-labeling postembedding immunogold EM for GluA3 (ϕ = 10 nm) and VGluT1 (ϕ = 15 nm). Asymmetrical synapses between VGluT1-labeled PF terminal (PF) and PC spine (Sp), and those between PF terminal and dendritic shaft of MLIs [Dn (MLI)] are labelled only occasionally for GluA3 (arrows). b Double-labeling postembedding immunogold EM for GluA3 (ϕ = 10 nm) and VGluT2 (ϕ = 15 nm). GluA3 labeling at synapses between the VGluT2-labeled CF terminal (CF) and PC spine (Sp) is clearly higher than that in VGluT2-unlabeled PF-PC synapses. c–f Quantitative immunogold EM analyses on the labeling density of GluA1–4 at three types of excitatory synapses in the cerebellar molecular layer. Note that the preferred GluA subunit combination is GluA1–3 and GluA2–4 at PF-PC and CF-PC synapses, and PF-MLI synapses, respectively. The specificity of postsynaptic labeling for each GluA subunit is confirmed by the almost blank labeling in control GluA-KO mice (black bars). Error bars represent SEM. Numbers of analyzed synapses are indicated in parentheses. Mann–Whitney U test; ***P < 0.001; NS not significant (P > 0.05). Bars 200 nm. Edges of postsynaptic density (PSD) are indicated by arrowheads. Reproduced with permission from Yamasaki et al. (2011)

Full size image

Considering that individual neurons in the brain typically receive synaptic inputs from multiple sources, the question arises whether different GluA subunits can be associated selectively with functionally distinct inputs. For example, in hippocampal CA3 interneurons, GluA2-lacking and GluA2-containing AMPARs are segregated to synapses from mossy fibers and CA3 recurrent collaterals, respectively (Toth and McBain 1998). Differential allocation of GluA subunits is also observed in fusiform cells of the dorsal cochlear nucleus: GluA2 is expressed at both parallel and auditory fiber synapses on apical dendrites, whereas GluA4 is limited to auditory fiber synapses on basal dendrites (Rubio and Wenthold 1997). However, this is not the case for cerebellar PC synapses. PCs receive two distinct excitatory afferents: numerous parallel fibers (PFs) and a single climbing fiber (CF) (Palay and Chan-Palay 1974). To distinguish these inputs, we employed double immunogold labeling for GluA subunits and terminal markers (Fig. 2a, b): VGluT1 for PF terminals and VGluT2 for CF terminals (Miyazaki et al. 2003). Quantitative immunogold EM analysis revealed that, in each of GluA1–3, the ratio of the labeling density between CF synapses and PF synapses is similar (Fig. 2c–e, white bars; Table 1). Thus, there would be no obvious difference in the relative abundance of the four GluA subunits, indicating that subunit combinations at PC synapse is determined primarily in a cell type-dependent manner with little input-dependent difference.

Table 1 Densities of immunogold particles for GluA1–4 [particles/1 μm of postsynaptic density (PSD)] in control and GluD2-KO mice. Statistical significance between the wild-type control and GluD2-KO mice was assessed by Mann–Whitney U-test. SD Standard deviation, N number synapses examined
Full size table

Quantitative difference in AMPAR expression

Another fundamental issue regarding the distribution of synaptic glutamate receptors is whether every synapse contains the same number or density of receptors. In contrast to the density of GABAA receptors, which has been shown to be uniform among cerebellar PCs, stellate cells, and dentate granule cells (Nusser et al. 1997), that of AMPARs is highly variable among different types of synapses (Nusser et al. 1998; Fukazawa and Shigemoto 2012). For example, in hippocampal CA3 pyramidal cells, mossy fiber synapses have approximately four times more AMPARs than associational/commissural fiber (A/C) synapses, suggesting input pathway-dependent regulation (Nusser et al. 1998). However, A/C synapses on interneurons have four times more AMPARs than those on pyramidal cells in the hippocampal CA3 area, suggesting target cell type-dependent regulation (Nusser et al. 1998). Likewise, individual AMPAR subunits in cerebellar molecular layer synapses are also regulated in an input pathway- and target cell type-dependent manner. In PCs, synapses from PFs have four times more AMPARs than those from CFs (Masugi-Tokita et al. 2007; Yamasaki et al. 2011). Furthermore, PF synapses on MLIs have four times more AMPARs than those on PCs (Masugi-Tokita et al. 2007). In PCs, the density of postsynaptic labeling for GluA1–3 at PF synapses is four to six times lower than that at CF synapses (Fig. 2c–e, white bars; Table 1). This trend is also observed between PF synapses on PCs and MLIs, where the labeling density for GluA3 is four times higher at PF-MLI synapses than at PF-PC synapses (Fig. 2e, white bars; Table 1).

Input pathway-dependent regulation of synaptic AMPAR in cerebellar PCs requires GluD2

What then are the molecular mechanisms underlying this skewed distribution of synaptic AMPARs? From multiple lines of evidence, we hypothesized that GluD2 (GluRδ2) could be involved in the input pathway-dependent selective abundance of AMPARs at PF-PC synapses. In cerebellar PCs, the glutamate receptor GluD2 is expressed selectively at PF synapses but not at CF synapses (Takayama et al. 1996; Landsend et al. 1997) (Fig. 3). GluD2 plays an essential role in the formation and maintenance of PF-PC synapses (Kashiwabuchi et al. 1995; Kurihara et al. 1997; Lalouette et al. 2001; Takeuchi et al. 2005; Uemura et al. 2007; Matsuda et al. 2010). Furthermore, GluD2 regulates endocytosis of AMPARs (Hirai et al. 2003) and mediates LTD at PF-PC synapses (Kashiwabuchi et al. 1995). To explore this possibility, we applied postembedding immunogold EM to GluD2-knock out (KO) and control mice, and measured the labeling density for GluA1–4 (Fig. 4; Table 1). In GluD2-KO mice, GluA1–3 shows a marked (three- to four-fold) increase at PF-PC, but not CF-PC, synapses (Fig. 4c–e, left and middle). Consequently, input pathway-dependent distribution of synaptic AMPARs in control PCs is almost eliminated in GluD2-KO PCs (Fig. 4c–e, gray bars). Unexpectedly, a modest (two-fold) increase of GluA2 and GluA3, but not GluA4, is also found at PF synapses in MLIs (Fig. 4d–f, right), where we identified a low but significant expression of GluD2 (Fig. 3e, right). These results indicate that GluD2 is involved in a common mechanism that suppresses the synaptic expression of particular GluA subunits at PF synapses in PCs and MLIs.

Fig. 3a–e
figure 3

GluD2 mRNA and protein is expressed predominantly in PCs and additionally in MLIs. a FISH showing exclusive expression of GluD2 mRNA (green) in the cerebellum in the adult mouse brain. Note that the sense probe yields no specific labeling (inset). b Double-labeling FISH for GluD2 (green) and GAD67 (red) mRNAs. In addition to strong labeling in PCs (asterisks), GAD67 mRNA-positive MLIs in the molecular layer (arrows) are labeled consistently for one of two tiny punctate signals for GluD2 mRNA. c, d Double labeling postembedding immunogold EM for GluD2 (ϕ = 10 nm) and terminal markers (ϕ = 15 nm: VGluT1 in c, VGluT2 in d). GluD2 is localized exclusively at VGluT1-positive PF-PC synapses (c), but not in VGluT2-positive CF-PC synapses (d). PF-MLI synapses are moderately labeled for GluD2 (c). e Histograms showing labeling density for GluD2 at three excitatory synapses in control (white bars) and GluD2-KO (black bars) mice. On average, the density of immunogold labeling at PF-MLI synapses amounts to 20 % of that at PF-PC synapses. Error bars Standard error of the mean (SEM). Numbers of synapses examined are indicated in parentheses. ***P < 0.0001, Mann–Whitney U test. NS Not significant. Bars a 1 mm; b 10 μm; c, d 200 nm. Reproduced with permission from Yamasaki et al. (2011)

Full size image
Fig. 4a–f
figure 4

Comparing the labeling density of GluA1–4 at three types of cerebellar synapses in the wild-type and GluD2-KO mice. a, b Double-labeling postembedding immunogold for GluA3 (ϕ = 10 nm) and VGluT1 (ϕ = 15 nm) or VGluT2 (ϕ = 15 nm). In GluD2-KO mice, while GluA3 labeling at PF-PC synapses is increased robustly (a, b), that at CF-PC synapses remains unaffected (b). c–f Quantitative immunogold EM analyses. Note that in GluD2-KO mice, the labeling density of GluA1–3 shows a three- to four-fold increase at PF-PC synapses, but not at CF-PC synapses. In addition, the labeling density of GluA2 and GluA3, but not GluA4, at PF-MLI synapses shows a two-fold increase. Error bars represent SEM. Numbers of synapses examined are indicated in parentheses. ***P < 0.0001, Mann–Whitney U test. NS not significant. Scale bars 200 nm. Edges of PSD are indicated by arrowheads. Reproduced with permission from Yamasaki et al. (2011)

Full size image

During long-term depression (LTD) at PF-PC synapses, the number of postsynaptic AMPARs is reduced by endocytosis (Matsuda et al. 2000; Wang and Linden 2000). Interestingly, application of an antibody against the putative ligand-binding domain of GluD2 induces AMPAR endocytosis, attenuates synaptic transmission, and impairs PF-LTD in cultured PCs (Hirai et al. 2003). From this evidence, it is possible to speculate that increased AMPAR expression could result from impaired PF-LTD in GluD2-KO mice (Kashiwabuchi et al. 1995). However, there is no significant increase in the density of GluA2 labeling in two major PF-LTD-deficient models: mGluR1-KO mice (Aiba et al. 1994) and GluD2ΔT mice, which lack the membrane-proximal domain of the C-terminal region of GluD2 (Uemura et al. 2007). Thus, the increase in synaptic AMPARs is not common to PF-LTD-deficient mutants, but specific to GluD2-KO mice. The other possibility is that GluD2 at PF synapses delimits the number of synaptic slots available for AMPARs. In the hippocampus, the specific interaction between PSD-95 and transmembrane AMPAR regulatory proteins (TARPs) has been shown to determine the number of synaptic AMPARs (Schnell et al. 2002; Stein et al. 2003; Ehrlich and Malinow 2004). In GluD2-KO mice, PSD-93 and PSD-95 at PF-PC synapses display concomitant increases, suggesting an increased capacity for AMPAR-TARP complexes. Compared with AMPARs, GluD2 is transported to the cell surface much more efficiently, in part, due to a strong endoplasmic reticulum exit signal in the C-terminal domain (Matsuda and Mishina 2000; Yuzaki 2009; Matsuda et al. 2010). Therefore, GluD2 appears to play a key role in the regulation of postsynaptic molecular organization and may delimit the number of slots available for AMPARs at PF synapses.

Although our data indicate that GluD2 plays an essential role in regulating the number of synaptic AMPARs at PF synapses, other mechanisms have been proposed to influence the membrane trafficking and synaptic targeting of AMPARs, including their C-terminal tail interactions with various synaptic molecules (Ziff 2007; Jackson and Nicoll 2011). These alternative processes may have distinct or synergistic functions in the regulation of synaptic AMPARs at cerebellar synapses. Thus, the amount and density, as well as subunit composition, of AMPA receptors at a given glutamatergic synapse is governed by both pre- and post-synaptic factors, resulting in functionally distinct connections.

Target cell type- and pathway-dependent regulation of NMDAR expression

Qualitative differences in NMDAR expression

NMDARs are the most important trigger for the activity-dependent long-term modification of synaptic strength (Lisman 1989; Bliss and Collingridge 1993; Cull-Candy et al. 2001), and their distribution also seems to be regulated by pre- and post-synaptic factors. Classical NMDARs consist of the obligatory subunit GluN1 (GluRζ1 or NR1) and any of the four regulatory GluN2 subunits (GluRε or NR2) (Seeburg 1993; Nakanishi and Masu 1994; Mori and Mishina 1995). Most of the diversity in the biophysical properties of NMDARs arises from the GluN2 subunit composition (Cull-Candy et al. 2001). GluN2 subunit expression displays a unique spatiotemporal profile (Lambolez et al. 1992; Watanabe et al. 1992; Monyer et al. 1994; Cull-Candy et al. 2001). For example, GluN2B (GluRε2 or NR2B) is expressed widely during prenatal development, whereas it is restricted to the forebrain in the adult brain. In sharp contrast, GluN2A (GluRε1 or NR2A) expression is ubiquitous in the CNS, starting at very low levels around the time of birth, and increasing dramatically during the second postnatal week. GluN2C (GluRε3 or NR2C) expression is first detected during the second postnatal week, and is highly enriched in the adult cerebellum. GluN2D (GluRε4 or NR2D) is expressed widely in early development, but is restricted to the brainstem and spinal cord in adulthood (Watanabe et al. 1992; Monyer et al. 1994).

GluN2A and GluN2B have received considerable attention over the past few decades, because these subunits are expressed highly in the cortex and hippocampus, and play a central role in synaptic plasticity and metaplasticity. For example, GluN2A replaces GluN2B in an activity-dependent manner (Barria and Malinow 2002), and the ratio of GluN2A to GluN2B is considered one of the crucial factors dictating the polarity of synaptic plasticity (Collingridge et al. 2004). In hippocampal CA3 pyramidal neurons in adulthood, these subunits are segregated to distinct synapses; CA3 pyramidal cell synapses from A/C input contain GluN1 and GluN2A and GluN2B, whereas synapses from mossy fibers contain only GluN1 and GluN2A (Fritschy et al. 1998; Watanabe et al. 1998). Similarly, GluN2B subunit influence at CA1 pyramidal cell synapses has been shown to differ depending on the laterality of origin of input (ipsilateral or contralateral CA3 pyramidal cells) (Kawakami et al. 2003). Therefore, differential allocation of GluN2A- and GluN2B-containing NMDARs is likely to be regulated by several mechanisms depending on the input pathway and neuronal activity.

Segregated GluN2B and GluN2D expression in trigeminal relay stations

In fetal and neonatal brains, GluN2B and GluN2D are the major regulatory subunits (Watanabe et al. 1992; Monyer et al. 1994). There are a number of contrasting physiological properties between these subunits, including channel conductance and the duration of channel opening, that might contribute to differential postsynaptic Ca2+ influx (Monyer et al. 1994; Misra et al. 2000a, b; Momiyama et al. 2003). Furthermore, GluN2B and GluN2D are organized in a pathway-dependent manner along the trigeminal pathway in both neonatal and adult mice (Yamasaki et al. 2014).

Consistent with previous studies using isotopic ISH (Watanabe et al. 1992; Monyer et al. 1994), FISH analyses have shown that GluN2B mRNA is highly expressed throughout the cortex and thalamus (Fig. 5a), whereas GluN2D mRNA is expressed only sparsely in these regions (Fig. 5b). In the primary somatosensory cortex (S1), GluN2B mRNA is expressed intensely in GAD67 mRNA(−) neurons and weakly in GAD67 mRNA(+) neurons (Fig. 5d), whereas GluN2D mRNA expression is restricted to GAD67 mRNA(+) neurons (Fig. 5e). In the thalamic ventrobasal (VB) complex, GluN2B mRNA expression is high in neurons lacking GAD67 mRNA, whereas it is low in reticular thalamic nucleus (RTN) neurons expressing GAD67 mRNA (Fig. 5f). In contrast, GluN2D mRNA expression is selective to RTN neurons expressing GAD67 mRNA (Fig. 5g). This pattern of neuronal expression is also found in the principal sensory nucleus of the trigeminal nucleus (Pr; Fig. 5h–l). Thus, it is common to each somatosensory station that non-GABAergic, most likely glutamatergic, neurons express GluN2B mRNA exclusively, whereas GABAergic neurons express GluN2D mRNA at high levels and GluN2B mRNA at low levels.

Fig. 5a–l
figure 5

FISH showing distinct neuronal expression of GluN2B and GluN2D mRNAs at trigeminal relay stations in adult mice. a–c Overall labeling patterns for mRNAs of GluN2B (a), GluN2D (b), and GAD67 (c) in a coronal forebrain section through the primary somatosensory cortex (S1), the ventrobasal thalamic nucleus (VB), and reticular thalamic nucleus (RTN). d–g Double-labeling FISH for GAD67 (green; d–g) and GluN2B (red; d, g) or GluN2D (red; e, g) mRNAs in the S1 (d, e), VB and RTN (f, g). h–l Overall labeling patterns for mRNAs of GluN2B (h), GluN2D (j), and GAD67 (l) in a coronal brainstem section through the trigeminal principal sensory nucleus (Pr). I, k Double-labeling FISH for GAD67 (green; I, k) and GluN2B (red; I) or GluN2D (red; k) mRNAs in the Pr. Note preferential labeling for GluN2B mRNA in GAD67 mRNA(−) neurons and for GluN2D mRNA in GAD67 mRNA(+) neurons (arrows) in both forebrain and brainstem sections. Hi Hippocampus, Mo trigeminal motor nucleus. Bars: a–c (in c), h, j, l (in l) 1 mm; d–g, I, k 10 μm. Reproduced with permission from Yamasaki et al. (2014)

Full size image

GluN2B and GluN2D protein localization is almost segregated along the trigeminal pathway. Immunofluorescence using parasagittal sections has revealed predominant immunolabeling of GluN2B and GluN2D proteins in the forebrain or brainstem, respectively (Fig. 6a, d), which is consistent with their distinct mRNA expressions (Fig. 6c, f). Double immunofluorescence shows that bright GluN2B(+) puncta lack detectable signals for GluN2D, whereas bright GluN2D(+) puncta show weak signals for GluN2B (Fig. 6g). The VB is crowded with bright GluN2B(+) puncta lacking GluN2D(+) immunoreactivity, whereas bright GluN2D(+) puncta in the RTN have weak immunoreactivity for GluN2B (Fig. 6h). A similar trend is also observed in the Pr (Fig. 6i). Further postembedding immunogold EM analyses revealed that GluN2B and GluN2D are specifically localized to asymmetrical synapses in each adult trigeminal relay station (S1, Fig. 7; PrV, VB, data not shown). Because GABA is contained in dendrites as well as the axon terminals of GABAergic neurons (Bolam et al. 1983), GABA can be used as a GABAergic dendritic marker (Fig. 7). Both subunits are detected preferentially on the postsynaptic membrane of asymmetrical synapses in the S1 (Fig. 7a, b). The density of GluN2B labeling per 1 μm of the postsynaptic density is three times higher at synapses on GABA(−) dendrites (mostly on dendritic spines) than synapses on GABA(+) dendrites (mostly on dendritic shafts) (Fig. 7c, left; P < 0.001, U test). By contrast, the density of GluN2D labeling is five times higher at synapses on GABA(+) dendrites than synapses on GABA(−) dendrites, and the density of the latter is almost comparable to the background labeling as determined from the corresponding synapses in GluN2D−/− mice (Fig. 7d, left; P < 0.001, U test). Similar segregated synaptic distribution is also observed in the Pr and spinal trigeminal subnucleus interpolaris (SpVi) (Fig. 7c, d, middle and right). Therefore, GluN2B is predominantly expressed at asymmetrical synapses on glutamatergic neurons, whereas GluN2D is selective to asymmetrical synapses on GABAergic neurons at each trigeminal station.

Fig. 6a–I
figure 6

Immunofluorescence showing the distinct distribution patterns of GluN2B and GluN2D immunoreactivities at trigeminal relay stations in adult mice. a, d Overall staining patterns in wild-type mouse brains for GluN2B (a) and GluN2D (D). b, e The specificity is indicated by substantial reduction of GluN2B immunolabeling in the GluN2B+/− brain compared with the GluN2B+/+ brain (b), and by the lack of GluN2D immunolabeling in the GluN2D−/− brain (e). c, f Isotopic in situ hybridization for GluN2B (c) and GluN2D (f) mRNAs in adult mouse brains. g–I High-power views of double immunofluorescence for GluN2B (red) and GluN2D (green) in the S1 (g), VB and RTN (h), and Pr(I). Intense GluN2B-positive puncta virtually lack GluN2D signals (arrowheads), and GluN2D puncta often possess weak GluN2B signals (arrows). Cb Cerebellum, CPu caudate-putamen, Cx cortex, Hi hippocampus, HT hypothalamus, Mb midbrain, MO medulla oblongata, OB olfactory bulb, Po pons, Th thalamus. Bars: a–f 1 mm; g 2 μm; h 10 μm; I 5 μm. Reproduced with permission from Yamasaki et al. (2014)

Full size image
Fig. 7a–d
figure 7

Postembedding immunogold EM showing distinct postsynaptic expression of GluN2B and GluN2D at relay stations of the adult trigeminal pathway. a, b Double-labeling postembedding immunogold EM for GABA (ϕ = 15 nm) and GluN2B (ϕ = 10 nm; a) or GluN2D (ϕ = 10 nm; b) in the S1. GluN2B is preferentially expressed at asymmetrical synapses on GABA(−) dendritic shaft and spines, and GluN2D is at GABA(+) dendrites. c, d Summary bar graphs representing preferential expression of GluN2B (c) and GluN2D (d) at synapses on GABA(−) and GABA(+) postsynaptic compartments in the S1, Pr, and SpVi. The background level of GluN2D labeling was measured at synapses on GABA-positive dendrites in GluN2D−/− mice. Arrowhead indicates the edge of the postsynaptic density. Error bars SEM. The ratio of labeled synapses to the total number of synapses examined is indicated as a fraction in each column. **P < 0.01 (U test). ***P < 0.001 (U test). Dn Dendrite; Sp spine; T ex GABA-negative (excitatory) terminal, T in GABA-positive (inhibitory) terminal. Bars 200 nm. Reproduced with permission from Yamasaki et al. (2014)

Full size image

To investigate the functional significance of segregated GluN2B and GluN2D expression in the trigeminal somatosensory pathway, we examined the formation and maturation of whisker-related patterning in GluN2B+/− and GluN2D−/− mice. Because GluN2B−/− mice display neonatal lethality (Kutsuwada et al. 1996), we instead analyzed GluN2B+/− mice, which have half the amount of GluN2B protein compared with control mice. To visualize whisker-related patterning, cytochrome oxidase (CO) histochemistry was applied to brain samples from GluN2B+/−, GluN2D−/−, and control littermates. In control mice, barrels in the S1 are obscure in all pups at P3, with segregated barrels appearing after P4 (Fig. 8a, e for GluN2B+/+ mice; Fig. 8c, g for GluN2D+/+ mice). In comparison, the appearance of barrels was delayed in GluN2B+/− mice (Fig. 8b, f, I) and advanced in GluN2D−/− mice (Fig. 8d, h, j). The termination stage of critical period plasticity, as assessed by CO histochemistry 8 days (S1) after transection of the right ION, displays similar temporal shifts. In control mice, ION transection at P3 substantially lowers overall CO intensity and blurs patterning in the contralateral hemisphere (Fig. 8k, m). By contrast, critical period plasticity termination is delayed in GluN2B+/− mice (Fig. 8p, s) and advances in GluN2D−/− mice (Fig. 8n, t). Thus, the refinement of whisker-related maps, assessed by the appearance of cortical barrels and termination of lesion-induced critical period plasticity, is delayed by nearly a day in the somatosensory cortex of GluN2B+/− mice but advanced by nearly a day in GluN2D−/− mice. These findings indicate the possibility that pathway-dependent organization of GluN2B and GluN2D-containing NMDARs is required for positive and negative modulation of somatosensory development and maturation, respectively.

Fig. 8
figure 8

Developmental appearance and termination of lesion-induced critical period plasticity in cortical barrels is delayed in GluN2B+/− mice and advanced in Glu2D−/− mice. a–h, Cytochrome oxidase (CO) histochemistry on flattened cortical sections at P3 (a–d), and P5 (e–h). I, j Bar graphs showing the 1 day delay of barrel appearance in GluN2B+/− mice (I) and 1 day advance in GluN2D−/− mice (j). k–t The right infraorbital nerve was transected at P3 (k–n), and P5 (o–r), and CO histochemistry was applied 8 days later. s, t Bar graphs representing the 1 day delay of critical period plasticity (CPP) termination in GluN2B+/− mice (s) and 1 day advance in GluN2D−/− mice (t). Bars 100 μm. Reproduced with permission from Yamasaki et al. (2014)

Full size image

Quantitative differences in NMDAR expression

Compared with AMPARs, the synaptic content of NMDARs is much less variable between individual synapses (Racca et al. 2000; Nimchinsky et al. 2004; Noguchi et al. 2005; Sobczyk et al. 2005). For example, only ~80 % of Schaffer collateral synapses onto CA1 pyramidal cell spines have a detectable number of AMPARs (Nusser et al. 1998; Takumi et al. 1999; Racca et al. 2000) and display a highly skewed distribution. By contrast, virtually all of these synapses contain immunoreactive NMDARs with small variability (Takumi et al. 1999; Racca et al. 2000). Specifically, Racca et al. (2000) showed that in CA1 pyramidal cell A/C synapses, the variability in synaptic NMDAR content [coefficient of variation (CV), 0.64–0.70] is much lower than that of the AMPAR content (CV 1.17–1.45). Furthermore, as there is no significant difference in the labeling of the obligatory NR1 subunit among synapses receiving different input sources, the net expression level of NMDAR is less likely to undergo input-specific regulation (Nyiri et al. 2003). Instead, CA1 parvalbumin (PV)-positive interneuron synapses display much lower labeling density of NR1 subunits than pyramidal cell synapses, showing a cell-type specific difference (Nyiri et al. 2003). Thus, it seems that postsynaptic cell types affect the synaptic NMDAR content, whereas presynaptic inputs are related to the distinct composition of NMDARs.

Conclusions

In summary, the above molecular-anatomical evidence has established that the synaptic allocation of AMPARs and NMDARs is regulated in an input pathway- and target cell type-dependent manner, and that it is essential for the proper functioning of neural circuits. There are, however, important questions that arise from these findings. How are these distributions regulated by neuronal activity? What are the underlying upstream and downstream molecular mechanisms? Furthermore, compared with the density or number of synaptic glutamate receptors, much less is known about subsynaptic distribution on the two-dimensional synaptic face, which also offers a powerful level of regulation over synaptic strength. Thus, it is essential to identify correlations between the functional properties and the molecular characteristics of glutamatergic synaptic connections. Knowledge gained through such studies will greatly facilitate our understanding of how these fine-tuned glutamatergic connections are established.