Layer 1 of the neocortex harbors a unique group of neurons that play crucial roles in synaptic integration and information processing. Although extensive studies have characterized the properties of layer 1 neurons in the mature neocortex, it remains unclear how these neurons progressively acquire their distinct morphological, neurochemical, and physiological traits. In this study, we systematically examined the dynamic development of Cajal-Retzius cells and γ-aminobutyric acid (GABA)-ergic interneurons in layer 1 during the first 2 postnatal weeks. Cajal-Retzius cells underwent morphological degeneration after birth and gradually disappeared from layer 1. The majority of GABAergic interneurons showed clear expression of at least 1 of the 6 distinct neurochemical markers, including Reelin, GABA-A receptor subunit delta (GABAARδ), neuropeptide Y, vasoactive intestinal peptide (VIP), calretinin, and somatostatin from postnatal day 8. Furthermore, according to firing pattern, layer 1 interneurons can be divided into 2 groups: late-spiking (LS) and burst-spiking (BS) neurons. LS neurons preferentially expressed GABAARδ, whereas BS neurons preferentially expressed VIP. Interestingly, both LS and BS neurons exhibited a rapid electrophysiological and morphological development during the first postnatal week. Our results provide new insights into the molecular, morphological, and functional developments of the neurons in layer 1 of the neocortex.
Layer 1 is among the earliest formed layers of the neocortex and the site of early synaptogenesis (Marin-Padilla 1971, 1978; Konig et al. 1975). Neurons in layer 1 undergo a more precocious morphological and neurochemical differentiation than those residing in the deep layers (layers 2–6) (Soriano et al. 1994; Super et al. 1998) and are crucial for early neocortical development and subsequent neocortical network formation (Van Eden et al. 1989; Frotscher 1998; Marin-Padilla 1998; Schwartz et al. 1998; Soda et al. 2003). Layer 1 is a relatively cell-sparse layer with 2 basic neuronal cell types: Cajal-Retzius cells and γ-aminobutyric acid (GABA)-ergic interneurons (Winer and Larue 1989; Imamoto et al. 1994; Zecevic and Milosevic 1997; Meyer and Goffinet 1998; Meyer et al. 1998; Zecevic and Rakic 2001). Our current knowledge on layer 1 is mostly gleaned from studies in rodents and primates. In rodents, layer 1 cells are of diverse origin as Cajal-Retzius cells are generated in the cortical hem, pallial septum, and subpallium (Hevner et al. 2003; Jimenez et al. 2003; Takiguchi-Hayashi et al. 2004; Bielle et al. 2005), whereas GABAergic interneurons are produced in the subpallium (Lee et al. 2010; Miyoshi et al. 2010). Neurons in the primate cortical layer 1 are also heterogeneous, with potentially more complex origins and migratory routes than in rodents (Meyer et al. 2002; Rakic and Zecevic 2003; Bystron et al. 2008; Petanjek, Berger, et al. 2009; Petanjek, Kostovic, et al. 2009; Clowry et al. 2010).
Cajal-Retzius cells are uniquely located in layer 1 and are characterized by an oval-shaped cell body and a single main long horizontal dendrite that is restricted to layer 1 (Cajal 1891; Derer 1987; Imamoto et al. 1994; Hestrin and Armstrong 1996; Radnikow et al. 2002). Previous studies suggest that most Cajal-Retzius cells disappear during the course of cortical development in both rodents and primates, but a small fraction persists into adulthood (Del Rio et al. 1995; Soda et al. 2003; Chowdhury et al. 2010). However, the precise fraction of surviving Cajal-Retzius cells and the quantitative morphological changes accompanying the death process are still unknown.
The other major layer 1 neuronal type is the GABAergic interneuron, which constitutes >90% of layer 1 neurons of the adult brain (Gabbott and Somogyi 1986; Winer and Larue 1989; Li and Schwark 1994; Prieto et al. 1994). Interneurons have been divided into many subtypes according to their morphology, neurochemical marker expression, and physiological properties (DeFelipe 1997; Kawaguchi and Kondo 2002; Markram et al. 2004; Ascoli et al. 2008; Batista-Brito and Fishell 2009). Recent evidence indicates that layer 1 interneurons mostly originate from the caudal ganglionic eminence (CGE; Xu et al. 2004; Yozu et al. 2005; Miyoshi et al. 2010) and contain Reelin-expressing late-spiking (LS) neurogliaform cells as well as bipolar/bitufted burst-spiking (BS) neurons (Lee et al. 2010; Miyoshi et al. 2010; Rudy et al. 2010). Nevertheless, little is known about the postnatal development of neurochemical, morphological, and electrophysiological properties of layer 1 interneurons.
In this study, we systematically investigated the molecular composition of layer 1 neurons as well as their developmental profile and distribution using immunohistochemistry and single-cell reverse transcription (RT)-polymerase chain reaction (PCR). We also quantitatively analyzed the development of morphological and physiological properties of layer 1 neurons by combining whole-cell patch-clamp recording and cell reconstruction. Our data showed that layer 1 Cajal-Retzius cells and GABAergic interneurons exhibit markedly different developmental profiles and may play distinct roles in neocortical circuit assembly and maturation.
Materials and Methods
CD-1 mice and GAD67-GFP (Δneo) transgenic mice (Tamamaki et al. 2003; Young and Sun 2009) were used in this study. The date of birth was defined as postnatal day 1 (P1). In the neocortex of GAD67-GFP mice, all GFP-positive (GFP+) cells were GABAergic neurons, and about 90% of the GABAergic neurons were GFP+ after birth (Tamamaki et al. 2003). All experiments were conducted in accordance with the guideline for the animal research and the use of Fudan University.
Mice at various postnatal ages (P1–14 and P50) were perfused intracardially with cold phosphate-buffered saline (PBS; pH 7.4) and 4% paraformaldehyde (PFA) in PBS (pH 7.4). The brains were removed from the skulls, fixed overnight, washed in PBS, and sectioned coronally at 60 μm on a vibratome (Leica VT1000S). Sections were incubated with the primary antibodies in PBS containing 1% bovine serum albumin, 0.5% Triton X-100, and 0.05% sodium azide for 36–48 h at 4°C. After washing in PBST (0.1% Triton X-100 in PBS) 5 times for a total of 50 min, sections were incubated with the secondary antibodies overnight at 4°C and subsequently washed 5 times in PBS for a total of 50 min. Mounted slices were visualized under epifluorescence illumination with the BX41 microscope (Olympus, Japan). The following primary antibodies were used: mouse anti-NeuN (1:1000, Millipore #MAB377), rabbit anti-NeuN (1:1000, Millipore #ABN78), rabbit anti-GABA (1:1000, Sigma #A2052), rabbit anti-calretinin (CR, 1:1000, Millipore #AB5054), goat anti-CR (1:1000, Millipore #AB1550), mouse anti-Reelin (1:1000, Millipore #MAB5364), chicken anti-GFP (1:1000, Aves #1020), rabbit anti-vasoactive intestinal peptide (VIP, 1:400, Immunostar #22700), goat anti-somatostatin (SOM, 1:250, Santa Cruz #sc-7819), and rabbit anti-GABAARδ (1:100, Phosphosolution #868-GDN). The secondary antibodies used were: donkey anti-goat (conjugated to Alexa Fluor 488 and Alexa Fluor 555, 1:500, Invitrogen), donkey anti-rabbit (Alexa Fluor 488 and Alexa Fluor 555, 1:500, Invitrogen), donkey anti-mouse (Alexa Fluor 488, Alexa Fluor 555, and Alexa Fluor 647, 1:500, Invitrogen), and donkey anti-chicken (DyLight 488, 1:500, Jackson ImmunoResearch). Confocal images were obtained by an Olympus FV1000 confocal microscope with a 40× objective. Digital images were brightness, contrast, and color balanced with Photoshop CS5 (Adobe system). For cell counting and quantification of immunolabeled neurons, sections were viewed with an Olympus epifluorescent microscope BX41, and Neurolucida software (Microbrightfield) was used to plot the distribution of the immunopositive neurons and to count neurons. We obtained quantitative data from the motor cortex (MC), somatosensory cortex (SC), and visual cortex (VC) from at least 3 discrete sections from 3 individual animals. The 3 cortical regions were determined by the stereotaxic map. The laminar borders of layer 1 drawn by Neurolucida software were identified on the basis of cell distribution of 4′,6-diamidino-2-phenylindole (DAPI) staining.
The brains of P1–20 GAD67-GFP transgenic mice were dissected from the skulls. For mice younger than P7 (including P7), the neocortex was whole-mounted. For mice older than P7, horizontal sections of the neocortex were cut at a thickness of 250 μm on a vibratome in the cutting solution containing (mM): 120 choline chloride, 2.6 KCl, 0.5 CaCl2, 7 MgCl2, 26 NaHCO3, 1.25 NaH2PO4, 15 d-glucose, and 1.3 ascorbate acid, then incubated in artificial cerebrospinal fluid (aCSF) containing (mM): 126 NaCl, 3 KCl, 1.25 KH2PO4, 1.3 MgSO4, 3.2 CaCl2, 26 NaHCO3, and 10 glucose, and bubbled with 95% O2/5% CO2. Whole-mounted and horizontal slices ensured the preservation of both the axonal and dendritic arbors of layer 1 cells. Individual cells were identified using an Olympus BX51WI upright microscope equipped with infrared-differential interference contrast (IR-DIC) and epifluorescence illumination, water immersion objectives, and an ORCA-R2 CCD camera (Hamamatsu). Intracellular solution composition was (in mM): 126 K-gluconate, 4 KCl, 10 N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid (HEPES), 4 ATP-Mg, 0.3 GTP-Na2, 10 creatine phosphate, 0.5% Alexa Fluor 568 hydrazide (Invitrogen; pH 7.2, 290–300 mOsm), and 0.4% neurobiotin (Invitrogen). The recording was performed with a Multiclamp 700B amplifier (Molecular Devices) in a current-clamp mode. After establishing the whole-cell configuration, the accumulated fine depolarizing currents (2–5 pA) were injected into the neuron to induce steady-state action potential (AP) from resting membrane potential (RMP), which can reflect intrinsic electrophysiological properties of the interneurons. Passive membrane properties were monitored periodically during the course of the experiments to ensure that the cell was healthy. Cells that showed significant rundown were discarded. Neurons were classified according to previously established criteria (Markram et al. 2004; Miyoshi et al. 2010).
To catalog the electrophysiological properties of GABAergic neurons observed in layer 1, 8 electrophysiological parameters were determined: (1) RMP measured directly after establishing the whole-cell configuration with no holding current applied; (2) AP threshold determined from the membrane potential at the onset of AP; (3) spike width measured as the time between its half-amplitude that reflects AP duration; (4) amplitude of afterhyperpolarization (AHP) measured as the difference between the ﬁring threshold and the maximum hyperpolarization following the AP; (5) time to AHP (tAHP) measured as the time interval between the spike peak and the hyperpolarization nadir; (6) AP amplitude measured from the threshold to the peak; (7) input resistance (Rin) measured by dividing the maximal averaged voltage deflection to 40 pA hyperpolarizing current pulses; (8) delay to spike measured as the time to first spike following current injection.
Cortical coronal slices (300 μm) were prepared from P15 to P20 mice and perfused with oxygenated aCSF. Patch pipettes (3–8 MΩ) were filled with 8 μL autoclaved RT-PCR internal solution containing (in mM): 144 K-gluconate, 3 MgCl2, 0.5 ethylene glycol-bis (β-aminoethyl ether)-N,N,N′,N′-tetra-acetic acid, 10 HEPES (pH 7.2, 290–300 mOsm). After whole-cell recordings were completed, cell cytoplasm was aspirated into the recording pipette while maintaining a tight seal. The tip of the pipette was then broken into a test tube, and the contents of the pipette were expelled and subjected to RT in a final volume of 10 μL as previously described (Lambolez et al. 1992; Ruano et al. 1995). Two steps of multiplex PCR were performed. The cDNAs present in the RT reaction were first simultaneously amplified using all primer pairs described in Supplementary Table 1 in a total volume of 100 μL (primer pairs were intron spanning to prevent the amplification of genomic DNA). rTaq polymerase (5 U, Takara) and 20 pmol of each primer were added to the buffer supplied by the manufacturer, and amplified over 21 PCR cycles (94°C for 30 s, 60°C for 30 s, and 72°C for 35 s). The products of the amplification reaction were then purified with the QIAquick spin column (QIAGEN) to remove primers and primer dimers. The purified PCR products (2 µL) were further individually amplified using a second pair of primers, internal to the first primers. Thirty-ﬁve PCR cycles were performed as described previously (Cauli et al. 1997), and then 10 μL of each individual PCR product was run on a 2% agarose gel using a 100 bp ladder (Tiangen) as a molecular weight marker and stained with ethidium bromide.
Morphological Reconstruction and Quantitation
Neurons in layer 1 were randomly selected, patched, and filled with neurobiotin for at least 10 min. After 10 min additional diffusion, slices were fixed in 4% PFA in 0.1 M PBS overnight. Labeled cells were visualized by overnight incubation in 1:200 Cy3-conjugated streptavidin (Jackson Immunoresearch Laboratories). Z-stack images were obtained on the spinning disk confocal system (PerkinElmer UltraView) with mosaic acquisition. A large region was scanned to ensure that entire dendritic and axonal arbors of the labeled cell were covered. Full cell 3-dimensional reconstructions and analysis were made with the Neurolucida software. Dendrites were identified by their thick, tapering and sometimes spiny appearance, whereas axons were smooth, thinner, and sometimes beaded. We measured the following morphological parameters: (1) maximum perimeter of cell soma; (2) spine density: the number of spines divided by the dendritic length; (3) length of axons and dendrites; (4) numbers of axonal and dendritic nodes: A node was assigned as the internal branch point along the axon and dendrite.
All data are presented as mean ± SEM, unless otherwise stated. All comparisons between groups were made with the Student's paired t-test or analysis of variance, and a P-value <0.05 was considered statistically significant.
Dynamic Changes in the Population of Layer 1 Neurons
We first characterized the population changes of layer 1 neurons during postnatal development by double immunolabeling with antibodies against Reelin and GFP in GAD67-GFP mice (Fig. 1A). To test whether all layer 1 GABAergic interneurons were labeled by GFP in GAD67-GFP mice, we examined the colocalization of GFP with the pan-neuronal marker NeuN in layer 1. All layer 1 GFP+ neurons expressed NeuN, and nearly all the NeuN+ cells possessed GFP from P2 to P14 (Supplementary Fig. 1B,C). On the other hand, GFP− Cajal-Retzius cells did not express NeuN (Supplementary Fig. 1A). These data suggested that virtually all of the layer 1 interneurons are selectively labeled by GFP in GAD67 mice, which allows the distinction between Cajal-Retzius cells and GABAergic interneruons, and greatly facilitates the further study of their development.
Reelin is thought to be a representative marker of Cajal-Retzius cells, but it also labels interneurons at postnatal stages (Ogawa et al. 1995; Alcantara et al. 1998). We found that a small population of Reelin+ neurons expressed GFP (Reelin+ interneurons) at early postnatal stages (P2, 1.3 ± 0.9% in SC, n = 6; P4, 0.8 ± 1.1% in SC, n = 6), and the number of Reelin+ interneurons dramatically increased after P4 (Supplementary Fig. 2). Given that Cajal-Retzius cells in the developing neocortex persistently express Reelin and do not express GABA (Hevner et al. 2003; Chowdhury et al. 2010), Reelin-positive and GFP-negative (Reelin+/GFP−) cells were considered Cajal-Retzius cells in this study. In addition, CR was shown to be another Cajal-Retzius cell marker during early developmental stages (Soriano et al. 1994; Del Rio et al. 1995). We found that Reelin+ cells were also CR immunoreactive before P4 (Supplementary Fig. 3). Considering the total population of layer 1 neurons as the sum of Cajal-Retzius cells and GFP+ GABAergic interneurons, we systematically analyzed the changes in Cajal-Retzius cells and GABAergic interneurons in 3 distinct neocortical areas: MC, SC, and VC. The population of Cajal-Retzius cells dropped significantly after P4, with only 4.1 ± 0.6% of the cells present at P2 surviving through P14 (in SC; Fig. 1B). The persisting Cajal-Retzius cells constituted 3.6 ± 0.3% of the total layer 1 cells at P14 (n = 5; Fig. 1C). On the other hand, GABAergic interneurons were abundant in layer 1 at P2 and were about 3 times as dense as Cajal-Retzius cells (Fig. 1B,D). The density of GABAergic interneurons decreased dramatically during the first postnatal week, but stabilized after P10 (Fig. 1D). The density of GABAergic interneurons at P14 was about one-third of that observed at P2. These findings demonstrated that layer 1 neuronal population undergoes dynamic developmental changes during the early postnatal period. It is worth noting that there were no significant differences in the density of Cajal-Retzius cells or in the density of GABAergic interneurons between MC, SC, and VC (Fig. 1B, D).
Temporal Evolution of Morphometric Parameters of Cajal-Retzius Cells
We next investigated the morphogenesis of layer 1 neurons. We injected individual Cajal-Retzius cells with neurobiotin in GAD67-GFP mice from P1 to P15. Cajal-Retzius cells were identified as GFP− under epifluorescence illumination (Fig. 2A). They possessed an oval-shaped soma, and one prominent, long, tapered, horizontal stem dendrite with spines, and an axon that emerged exclusively from the side of the soma opposite the dendrite (Fig. 2B,C), as previously reported (Cajal 1891; Marin-Padilla 1998; Radnikow et al. 2002). We excluded the cells that displayed characteristic morphological features of apoptosis, such as retraction of the cell body toward the dendrite and breaking down of the stem dendrite (Chowdhury et al. 2010). We reconstructed a total of 61 healthy Cajal-Retzius cells from P1 to P15 (Fig. 2D) for morphometric analysis. Their soma size (the maximum perimeter of the soma) increased with development and peaked at P5 (60.0 ± 3.0 μm, n = 8), and then decreased to 40.4 ± 2.2 μm at P15 (n = 7) (Fig. 2E). Their spine density increased continuously from P1 to P5 (P1, 3.1 ± 0.5 per 100 μm, n = 7; P3, 4.6 ± 0.9 per 100 μm, n = 11; and P5, 7.5 ± 1.4 per 100 μm, n = 8) and then stabilized from P5 to P15 (Fig. 2F). The length of axons and the number of axonal nodes peaked at P5 and then significantly decreased with development (Fig. 2G,H). During the second postnatal week, the spine density, axonal length, and nodes of Cajal-Retzius cells did not change significantly. Based on the number of dendrites emanating from the soma, Cajal-Retzius cells can be divided into 2 subtypes: Typical and atypical Cajal-Retzius cells (Radnikow et al. 2002). Typical Cajal-Retzius cells have only one dendrite originating from the soma, whereas atypical Cajal-Retzius cells have more than 2 dendrites, but only one horizontal, thick primary dendrite. Both these subtypes of Cajal-Retzius cells were observed in our analysis (Fig. 2C); however, there was no significant difference between the 2 subtypes in the morphometric parameters of the thick stem dendrite (data not shown). The length of the stem dendrite increased from 448.9 ± 168.1 μm at P1 (n = 11) to 1045.5 ± 199.5 μm at P7 (n = 6, P < 0.05) and subsequently decreased to 336.4 ± 115.3 μm at P15 (n = 7; Fig. 2I). Dendritic nodes were rare at P1 (6.8 ± 1.8, n = 11), though they reached a maximum at P3 (22.8 ± 3.8, n = 9; Fig. 2J). Furthermore, with development, the number of dendritic nodes gradually decreased to 6.6 ± 1.8 at P15 (n = 7). In summary, we found that, during the first 5 days after birth, the morphology of Cajal-Retzius cells became progressively more complex, but their axons and dendrites degenerated after P7. These results demonstrated that Cajal-Retzius cells undergo 2 phases of developmental morphological changes— first, a morphological maturation that occurs shortly after birth, followed by a morphological degeneration.
Neurochemical Characteristics of GABAergic Interneurons in Layer 1
Previous studies have shown that various neurochemical markers are expressed in neocortical GABAergic interneurons (Kawaguchi and Kondo 2002; Xu et al. 2010). To determine the subtypes of interneurons in developing layer 1, we immunostained cortical sections prepared between P1 and P14, and at P50 with the antibodies against neuropeptide Y (NPY), GABAARδ, CR, VIP, SOM, Reelin, and parvalbumin (PV). In accordance with previous studies (Ferezou et al. 2002; Lee et al. 2010), PV-expressing cells were not detected at any time points in layer 1 (data not shown). The vast majority of GFP-expressing neurons were positive for Reelin, NPY, GABAARδ, CR, VIP, and SOM (Fig. 3 and Supplementary Fig. 4), confirming that these markers effectively identify layer 1 interneurons. Most of the markers were strongly expressed in layer 1 interneurons as early as P8. We found that Reelin+ interneurons were the most abundant population in layer 1 interneurons and increased significantly from P2 to P8 (Fig. 3A). GABAARδ+ neurons were present at a lower density than Reelin+ interneurons (Fig. 3B). GABAARδ was previously reported as a marker for neurogliaform cells (Olah et al. 2009). We found that it began to appear in layer 1 at P8 and remained constant during the second postnatal week. The densities of interneurons that expressed NPY, VIP, CR, and SOM were far lower than that of Reelin+ and GABAARδ+ interneurons (Fig. 3C–F). The Reelin+, GABAARδ+, NPY+, VIP+, CR+, and SOM+ cells as a proportion of the total layer 1 interneurons were 67.5 ± 3.0%, 42.5 ± 4.3%, 20.0 ± 1.9%, 7.6 ± 0.6%, 7.0 ± 0.1%, and 2.3 ± 0.2% in SC at P14 (n = 5), respectively (Supplementary Fig. 5). It is interesting to note that the densities of GABAARδ+, VIP+ and SOM+ cells did not significantly change during the second postnatal week, since these chemical markers showed clear expression in layer 1. In contrast, the densities of Reelin+, NPY+, and CR+ interneurons exhibited a marked increase from P6 to P8.
Spatial Distribution of Neurons in Layers 1a and 1b
A recent study suggested that interneuron subtypes with distinct spatial distribution in layer 1 take part in different cortical circuits (Cruikshank et al. 2012). To ascertain the preferential and final positions of different neuronal subtypes in layer 1, we divided layer 1 into 2 equal sublayers: 1a and 1b (Fig. 4A). We found that the distribution patterns of distinct subtypes of neurons between layers 1a and 1b were dramatically different during development (Fig. 4B). Cajal-Retzius cells (Reelin+/GFP− cells) and Reelin+ interneurons strictly preferred the superficial region of layer 1a prior to P6 (approximately 75.0%), but were evenly distributed in layer 1 after P8 (Fig. 4B1,B2). CR+ interneurons accumulated in layer 1a prior to P8 (Fig. 4B3). The percentage of CR+ interneurons in layer 1a was 8 times larger than in layer 1b at P2. However, after P10, CR+ interneurons preferentially localized to layer 1b. Furthermore, VIP+ and SOM+ neurons were preferentially localized to layer 1b (Fig. 4B4,B5), and the percentage of VIP+ neurons in layer 1a decreased from 31.0 ± 4.8% at P8 (n = 6) to 3.0 ± 3.0% at P14 in SC (n = 5, P < 0.05). The developmental changes in the percentages of SOM+, GABAARδ+, and NPY+ interneurons between layers 1a and 1b were not significant (Fig. 4B5–7). The percentage of GABAARδ+ neurons in layer 1a was slightly higher than in layer 1b, which was in contrast to what was observed for NPY+ neurons. To summarize, at the end of the second postnatal week, Cajal-Retzius cells, Reelin+, GABAARδ+, and NPY+ interneurons were almost evenly distributed in layer 1, and the majority of CR+, VIP+, and SOM+ interneurons were located in layer 1b.
Our data showed that the percentages of VIP+, Reelin+, and CR+ interneuron subtypes increase in layer 1b with development. These developmental changes in neuron distribution between layers 1a and 1b may be due to the migration of cells from layer 1a to 1b and/or the selective increase of specific marker expression in layer 1b. The density of VIP+ interneurons remained relatively stable from P8 to P14 across layer 1 (Fig. 3D), suggesting that the increase of VIP+ interneurons in layer 1b is a result of cell migration from layer 1a. Interestingly, the significant increase in the percentage of Reelin+ and CR+ interneurons in layer 1b coincided with the substantial increase in their density across the entire layer 1, suggesting that their expansion in layer 1b may mainly due to an increased expression of neurochemical markers.
Coexpression of Neurochemical Markers in Layer 1 GABAergic Interneurons
Given that neurochemical markers can be expressed in interneurons in either mutually exclusive or overlapping patterns, we immunostained cortical sections with various combinations of the antibodies against the 6 neurochemical markers to study their coexpression in layer 1 interneurons (Supplementary Fig. 6). We found that the percentage of neurons with coexpressed neurochemical markers increased with development (Fig. 5). Reelin+ interneurons were the largest population in layer 1, all NPY+ and GABAARδ+ interneurons were positive for Reelin (Supplementary Fig. 6A,E), and 68.2 ± 3.5% of CR+ interneurons expressed Reelin in SC at P14 (Fig. 5 and Supplementary Fig. 6D, n = 5). In addition, 58.6 ± 2.9% of SOM+ neurons and 14.7 ± 2.6% of VIP+ neurons expressed Reelin at P14 (Fig. 5 and Supplementary Fig. 6B,C, n = 5). CR+ interneurons did not express NPY (Fig. 5 and Supplementary Fig. 6G). A small but significant population of CR+ interneurons expressed SOM in layer 1 after P12 (P14, 8.0 ± 0.7%, n = 5; Fig. 5 and Supplementary Fig. 6L). Consistent with previous studies (Kawaguchi and Kondo 2002; Miyoshi et al. 2007), SOM+ neurons lacked VIP expression (Fig. 5 and Supplementary Fig. 6K). In addition, the percentages of SOM+ neurons that expressed NPY, Reelin, and CR at P14 were 17.3 ± 3.1%, 58.0 ± 4.2%, and 69.0 ± 7.3% (n = 5), respectively (Fig. 5 and Supplementary Fig. 6J,B,L). VIP was coexpressed only with CR and Reelin (Fig. 5 and Supplementary Fig. 6C,H). The percentage of VIP-expressing CR+ neurons was 17.8 ± 3.5% at P14 (Fig. 5, n = 5).
Having established the coexpression patterns of various neurochemical markers, we quantified the percentage of interneurons expressing these markers with development (Fig. 5). Approximately 78% and 93% of GFP-expressing interneurons were labeled by these markers at P14 and P50, respectively. This difference between P14 and P50 was mainly caused by the increased expression of Reelin in interneurons with development (P14, 67.5 ± 3.0% and P50, 87.8 ± 1.0%). These results suggest that the neurochemical markers that we used to classify the subtypes of interneurons are sufficient for analyzing almost all mature layer 1 interneurons.
Electrophysiological and Molecular Features of Mature Layer 1 GABAergic Interneurons
We next investigated the electrophysiological properties of layer 1 interneurons and linked these results with their molecular characteristics. Individual layer 1 interneurons were subjected to whole-cell patch-clamp recording and single-cell RT-PCR in brain slices prepared from P15 to P20 mice. Over this postnatal period, the electrophysiological and molecular properties of interneurons reach adult levels (Zhou and Hablitz 1996; Sauer and Bartos 2011; Yang et al. 2012). In line with a previously established electrophysiology-based nomenclature system for interneurons (Miyoshi et al. 2007, 2010; Ascoli et al. 2008; Lee et al. 2010), we examined a total of 54 interneurons and classified them into 3 subtypes: late spiking 1 (LS1), late spiking 2 (LS2), and burst spiking (BS) neurons (Fig. 6A–C). The key differences among them were the delay to the initial AP spike, AP firing pattern, and afterdepolarization (ADP). LS cells were characterized by a delay with a steady ramp depolarization leading up to the initial spike at threshold current injections, as previously reported (Chu et al. 2003; Lee et al. 2010). According to the time and number of the delayed spikes, LS neurons were grouped into 2 subtypes: LS1 and LS2 neurons. While LS1 neurons had >2 delayed spikes, LS2 neurons had only 1 delayed spike. BS neurons fired a burst of APs above but not at threshold. In addition, ADPs were only observed in BS neurons (89%, 16 in 18 BS neurons; Fig. 6C, inset). We also found that some BS neurons showed adaptations in firing in response to prolonged current injections. The electrophysiological properties of layer 1 interneurons during P15–20 are summarized in Supplementary Table 2.
After the recording was completed, we analyzed the molecular properties of interneurons by performing single-cell RT-PCR experiments to simultaneously detect the transcripts of multiple interneuron markers, including GAD67, CR, nitric oxide synthase (nNOS), Reelin, PV, SOM, cholecystokinin (CCK), VIP, NPY, calbindin (CB), and GABAARδ (Fig. 6D,E). Only cells expressing GAD67 and at least one neuropeptide or calcium-binding protein were analyzed. The prominent molecular features of LS1 neurons were high mRNA levels of CR (71%), Reelin (57%), GABAARδ (57%), and CB (43%) and low mRNA levels of nNOS, SOM, CCK, and VIP. We did not detect nNOS, SOM, or VIP in LS2 neurons. The most abundant mRNA transcript in LS2 neurons was GABAARδ (86%), followed by NPY (71%), CR (57%), and CCK (43%). BS neurons had a low occurrence of GABAARδ (13%), but high occurrence of NPY (75%) and Reelin (50%). Previous studies indicate that BS neurons are VIP-expressing bipolar/bitufted neurons, and LS neurons are Reelin-expressing neurogliaform cells (Lee et al. 2010; Miyoshi et al. 2010; Rudy et al. 2010). Indeed, our data corroborated that BS neurons had higher levels of VIP than the LS1 and LS2 neurons, and both LS1 and LS2 neurons had much higher levels of CR and GABAARδ than the BS neurons, suggesting that BS neurons may be VIP+ neurons and LS (contained LS1 and LS2) neurons may be GABAARδ+ neurogliaform neurons in layer 1.
Quantitative Evaluation of Electrophysiological and Morphological Properties of Layer 1 Interneurons
To comprehensively and simultaneously examine the electrophysiological and morphological development profiles of layer 1 interneurons, we performed whole-cell recording on a large collection of GFP-expressing neurons (n = 212) from P1 to P15 GAD67-GFP mice. The recording solution contained neurobiotin for post hoc analysis of the morphological properties (Fig. 7). Eight electrophysiological features that described the passive membrane and firing properties of neurons were analyzed (Fig. 8). We found no significant differences in intrinsic firing properties between LS1 and LS2 neurons, therefore, collectively studied them as LS neurons in further analyses. In addition, the firing patterns of these neurons during the early postnatal stages were variable. Some neurons during P1 to P5 fired only one AP in response to current injection; however, only those neurons that fired sustained APs were quantitatively analyzed. Of a total of 212 layer 1 interneurons that were examined, 162 (76.4%) were LS neurons. The recorded values of passive and active membrane properties are summarized in Supplementary Table 3.
The passive and active membrane characteristics of both LS and BS neurons became more adult-like over the course of postnatal development (Figs 7 and 8). Notably, LS and BS interneurons could be distinguished by their firing properties as early as P3. Several key differences in electrophysiological properties were found between LS and BS neurons with development. Compared with BS neurons, LS neurons exhibited a remarkably longer delay time to the first AP spike during development and maturation (P < 0.001, Fig. 8A). In addition, BS neurons exhibited a shorter time to afterhyperpolarizing potentials (tAHP) than LS neurons after P7 (P < 0.01, Fig. 8H). Interestingly, the AHP amplitude of BS neurons was larger than that of LS neurons and only observed at P7 (Fig. 8G). Despite these differences, LS and BS neurons exhibited similar developmental changes in many other membrane properties. RMPs for both LS and BS neurons gradually hyperpolarized with development (Fig. 8B). The input resistance (Rin) decreased significantly during the first postnatal week (Fig. 8C). After P9, Rin remained relatively constant and was similar to that observed in mature neurons. The threshold of APs rapidly hyperpolarized between P7 and P9 and remained constant after P9 (Fig. 8D). The amplitude of APs of both LS and BS neurons gradually increased in the first postnatal week, but no significant differences were found after P7 (Fig. 8E). The change in the amplitude of APs paralleled a similar significant increase in the amplitude of AHPs (Fig. 8G). In addition, the spike width and tAHP decreased sharply between P3 and P5, and subsequently became adult-like (Fig. 8F,H). To summarize, although LS and BS neurons exhibited differences in their membrane properties, they underwent similar changes in developing their membrane properties. In addition, whereas the passive characteristics (RMP and Rin) of both BS and LS neurons change continuously over the 2 weeks after birth, most of their active properties rapidly matured in the first postnatal week.
Following recording, we analyzed the morphology of interneurons (Fig. 7B,D). All cells were obtained from whole-mount or horizontal slices for a better maintenance of the cell integrity. In total, 62 interneurons were reconstructed and analyzed from a range of postnatal developmental time points. The values of these morphological properties are summarized in Supplementary Table 4. Most of the morphological properties did not significantly differ between BS and LS neurons (Fig. 9). The maximum soma perimeter of BS neurons was significantly lower than that of LS at P3 (Fig. 9A), but no significant differences between LS and BS neurons were observed after P5. The spine density of BS and LS neurons increased significantly with development, especially from P5 to P9 (Fig. 9B). The length of the axons of both LS and BS neurons also increased drastically with development (Fig. 9C). Surprisingly, only the dendrites of BS neurons became significantly longer (Fig. 9E). The change in the length of processes (axons and dendrites) was paralleled by significant increases in the number of the nodes (Fig. 9D,F).
Validity of the GAD67-GFP Mouse Model
In this study, we investigated the developmental changes of layer 1 neurons in their molecular, morphological, and electrophysiological properties. We took advantage of the GAD67-GFP transgenic mouse to identify GABAergic interneurons. Our data demonstrated that, in these mice, while GFP labels all GABAergic neurons in layer 1, Cajal-Retzius cells is GFP negative. Different lines of transgenic mice have been created for studying interneuron function, including GAD65-GFP (Lopez-Bendito et al. 2004). While it is generally believed that most GABAergic neurons express GAD65 and GAD67 (Esclapez et al. 1993, 1994; Fukuda et al. 1997), the distribution patterns of GFP in these 2 transgenic mice are different (Tamamaki et al. 2003; Lopez-Bendito et al. 2004; Wierenga et al. 2010). The majority of GFP+ neurons in GAD65-GFP mice originates from CGE and locates in supragranular layers (Lopez-Bendito et al. 2004). However, it has also been reported that a subset of Reelin+ and CR+ interneurons in layer 1 are not GFP+ in the mature GAD65-GFP mice (Lopez-Bendito et al. 2004), indicating that GFP does not label all layer 1 interneurons in these mice. A recent study reports that the CGE-originated interneurons can be labeled in 5HT3aR-GFP transgenic mice (Lee et al. 2010). While GFP in 5HT3aR-GFP transgenic mice exclusively labels neurons from CGE, the medial ganglionic eminence (MGE)-originated Reelin+ and SOM+ interneurons in layer 1 remain unlabeled (Lee et al. 2010; Miyoshi et al. 2010). There were some differences in the percentages of layer 1 interneurons expressing specific chemical markers between our study using GAD67-GFP mice and that using 5HT3aR-GFP mouse (Lee et al. 2010). Overall, the proportion of Reelin+, NPY+, and VIP+ interneurons in layer 1 was higher among GAD67-GFP cells (Reelin: ∼85% compared with ∼73% in the 5HT3aR-GFP; NPY: ∼20% compared with ∼7%, and VIP: ∼7% compared with ∼5%). We also determined the population of SOM+ and GABAARδ+ neurons in layer 1 in GAD67-GFP mice. With the 6 interneuron chemical markers, we were able to account for about 93% of layer 1 interneurons in adult GAD67-GFP mice, versus roughly 80% in the 5HT3aR-GFP mice (Lee et al. 2010). Therefore, our use of GAD67-GFP transgenic mice minimizes the omission of any significant subsets of layer 1 interneurons.
Postnatal Temporal Evolution of Cajal-Retzius Cells
Cajal-Retzius cells are among the earliest born neurons in the neocortex (Marin-Padilla 1971, 1998; Del Rio et al. 1995) and play an important role in the structural and functional development of the neocortex (Caviness and Sidman 1973; Frotscher 1998; Rice and Curran 2001). Here, we analyzed the density as well as the morphology of Cajal-Retzius cells during the first 2 weeks after birth. Our data showed that a large number of Cajal-Retzius cells died starting at P6 and only 4.1% of Cajal-Retzius cells observed at P2 survived after P14, consistent with previous studies (Derer and Derer 1990; Del Rio et al. 1995; Soda et al. 2003; Chowdhury et al. 2010). It has been shown that Cajal-Retzius cells remain in an immature state with growth cones in the axonal and dendritic processes during the first week after birth, and the growth cones disappear when degeneration occurs in the second postnatal week (Derer and Derer 1990). In line with these observations, we found that the morphology of Cajal-Retzius cells became more complex prior to P5, with a larger soma, as well as longer and more numerous neurites. Moreover, we also observed that the morphological degeneration of Cajal-Retzius cells started after P5, as evidenced by the reduction in soma size and the complexity of dendrites and axons. Our findings suggest that the surviving Cajal-Retzius cells undergo morphological transformation during the second postnatal week. The apoptosis of Cajal-Retzius cells occurs quickly and is completed in <6 h (Chowdhury et al. 2010). Interestingly, we observed that both healthy and dying Cajal-Retzius cells at each developmental stage examined. We attempted to reconstruct healthy Cajal-Retzius cells with typical morphological properties; however, we could not completely exclude dying Cajal-Retzius cells from being reconstructed and analyzed, which may contribute to the reduction in morphological parameters observed during the second postnatal week.
Previous studies have indicated that the first postnatal week is a critical period for the maturation of interneurons and pyramidal cells in the neocortex (Yu et al. 2009; Allene et al. 2012; Inamura et al. 2012; Yu et al. 2012). Cajal-Retzius cells form gap junctions with pyramidal cells in deep layers, which could contribute to synchronous responses in the network at a very early postnatal stage in the developing neocortex (Soda et al. 2003). Furthermore, we found that the period of rapid growth in the morphology of Cajal-Retzius cells coincided with the dramatic electrophysiological and morphological development of layer 1 interneurons. In addition, the density of Cajal-Retzius cells did not decrease during this period. These findings raise the possibility that Cajal-Retzius cells may play an important role in the maturation of layer 1 interneurons in the first postnatal week. The long, outstretched, bifurcated axons of Cajal-Retzius cells could be involved in establishing synaptic contacts the terminal tuft dendrites of pyramidal cells in deep layers over a wide area of the neocortex (Radnikow et al. 2002), which may facilitate the stabilization of pyramidal neuron dendritic tufts in the most superficial part of the developing neocortex. Together, these observations suggest that Cajal-Retzius cells integrate information in layer 1 and send projections to target neurons to facilitate the formation of the neocortical network during very early stages of the development. After the early neocortical network is established, most of the Cajal-Retzius cells degenerate and die.
Developmental Process of Layer 1 Interneurons
The early postnatal development of layer 1 interneurons was systematically analyzed in this study. We showed that the densities of total layer 1 interneurons decrease in the first postnatal week. CGE- and late-born MGE-derived interneurons migrate out of layer 1 into the cortical plate during the first postnatal week (Hevner et al. 2004; Tanaka et al. 2009; Miyoshi and Fishell 2011), and very few interneurons in layer 1 are eliminated by apoptosis during the first postnatal week (Hevner et al. 2004). Furthermore, the neocortex shows marked expansion during development (Finlay and Darlington 1995). Therefore, the decrease in the density of layer 1 GABAergic interneurons in the first postnatal week is mainly due to a dilution effect and the migration to deeper layers rather than cell death. However, apoptosis may contribute to the significant decrease in the density of layer 1 interneurons during the second postnatal week (Verney et al. 2000; Southwell et al. 2012). Interestingly, we found that the majority of layer 1 interneurons abruptly expressed their distinct neurochemical markers around P8, when the radial migration from layer 1 is largely completed, suggesting that the majority of layer 1 interneurons quickly start to differentiate once they acquire their final layer-specific position. In addition, we found that VIP+ neurons migrated internally in layer 1 from P8 to P14, suggesting that a small number of interneurons still have the capacity to migrate within layer 1 during the second postnatal week.
We simultaneously and quantitatively examined the maturation of the electrophysiological and morphological properties of layer 1 interneurons. Previous studies showed that, during embryogenesis, the morphology of interneurons changes rapidly from a simple cell with few unbranched processes to a highly branched cell (Perreault et al. 2003; Wu and Cline 2003; Le Magueresse et al. 2011). As development proceeds, layer 1 interneurons become complex with more and longer processes. Extensively branching processes result in a very large surface area over which the cells can sample the input from environment, allowing them to play an important role in regulating the development of the cortical network. We found that the active membrane properties of layer 1 interneurons matured rapidly in the first postnatal week. The maturation of the passive membrane properties continued into the second postnatal week and lagged behind that of the active membrane properties by a few days. The development of the passive membrane properties was paralleled by the development of the morphology. We found that the anatomical–electrophysiological–molecular properties of layer 1 interneurons exhibited adult-like properties at the end of the second postnatal week, indicating that by then layer 1 interneurons are almost mature enough to play important roles in cortical network assembly and regulation.
Interneuron Diversity in Layer 1
Neocortical GABAergic interneurons are an extremely heterogeneous neuronal population based on their morphological, electrophysiological, and neurochemical features (Kawaguchi and Kondo 2002; Gonchar et al. 2007). The diversity of GABAergic interneurons in the deep layers of mouse neocortex is well established (Rudy et al. 2010); however, little is known about the diversity of layer 1 GABAergic interneurons. In this study, we attempted a quantitative classification of the properties of layer 1 interneurons. Consistent with previous studies (Ferezou et al. 2002; Nery et al. 2002; Lee et al. 2010; Miyoshi et al. 2010), we found that the neurochemical markers, including Reelin, GABAARδ, CR, SOM, VIP, and NPY, were expressed in layer 1 and Reelin+ interneurons that constituted the majority of layer 1 interneurons (about 85% at P50). Importantly, we identified a previously unknown subtype of layer 1 interneurons, GABAARδ+ neurogliaform cells. A previous study demonstrated that GABAARδ+ neurogliaform cells exist in the rat hippocampus (Olah et al. 2009). Similar to those in the primate neocortex (Zaitsev et al. 2009), GABAARδ+ neurogliaform cells in layer 1 of the mouse neocortex also coexpressed Reelin. VIP+ interneurons represent a special subtype of GABAergic inhibitory neurons in the mouse cortex. It was previously shown that Reelin+ interneurons do not express VIP (Miyoshi et al. 2010), but we observed that 14.2% of VIP+ interneurons in layer 1 coexpressed Reelin at P14. There was no coexpression of GABAARδ and VIP, indicating that GABAARδ+ and VIP+ interneurons are 2 distinct subtypes of layer 1 interneurons. Furthermore, the population of GABAARδ+ neurogliaform cells was 6-fold larger than that of VIP+ neurons in layer 1.
We also performed a comprehensive electrophysiological and morphological analysis to assess the diversity of layer 1 interneurons. Based on their repetitive AP firing properties, we classified layer 1 interneurons into 2 subtypes: LS and BS neurons. In agreement with previous findings (Lee et al. 2010; Miyoshi et al. 2010) and consistent with our immunohistochemistry results, the majority of layer 1 interneurons were LS neurons, which expressed GABAARδ. Previous studies showed that the morphological characteristics of LS cells are similar to those of neurogliaform cells (Schwartz et al. 1998; Chu et al. 2003; Miyoshi et al. 2010), which have dense axon collaterals and short, aspiny dendrites restricted primarily to layer 1. On the other hand, BS neurons exhibit bipolar or bitufted morphology and have long axonal branches descending downward to deeper layers, as described previously for VIP+ neurons (Kawaguchi and Kubota 1996; Rozov et al. 2001; Butt et al. 2005). Indeed, we observed that the majority of LS neurons had a neurogliaform-like morphology, but a subset of LS neurons had atypical morphology. Moreover, a subset of BS neurons had dense axon collaterals and exhibited morphological similarities to neurogliaform cells. In addition, we found that there was no significant difference in the quantitative analysis of the axons between LS and BS interneurons as they mature (Fig. 9). Meanwhile, although molecular analysis showed that LS and BS interneurons had higher expression levels of GABAARδ and VIP, respectively, we found that 7.1% of LS neurons also expressed VIP and 12.5% of BS neurons expressed GABAARδ, suggesting that the firing properties of LS and BS neurons were not strictly correlated to the expression of GABAARδ and VIP. Taken together, these results demonstrate that although electrophysiological properties of layer 1 interneurons are highly correlated with their morphological properties and expression of specific markers, the electrophysiological properties of an interneuron is not simply defined by its morphology and marker expression.
The function of GABAergic interneurons in the cortical network is largely determined by their electrical and anatomical features. The neurogliaform-like morphological features of neurons indicated that the synaptic inputs and outputs of LS neurons were confined to layer 1. The profusely ramified horizontal axons of LS/neurogliaform cells allow them to make synaptic connections with numerous targets in layer 1. LS neurons can form local circuits with LS neurons as well as with BS neurons in layer 1 (Chu et al. 2003). Meanwhile, electrical coupling is abundant among LS neurons, but infrequent between LS cells and other neurons (Chu et al. 2003). It has been shown that LS/neurogliaform cells can transfer multiple whisker inputs into GABAergic outputs to the distal dendrite of pyramidal neurons in different barrels, while the neurons with long descending processes can convert precise whisker signals into a small number of neurons in the same barrel (Zhu and Zhu 2004). On the other hand, the chemical markers are most probably necessary for the function of interneurons in cortical circuits. Interneuron-derived Reelin plays a role in modulating synaptic function (Weeber et al. 2002; Herz and Chen 2006; Forster et al. 2010). By modulating the pre- and postsynaptic neurons through the intracellular calcium signals, interneuron-derived CR is important for precise timing and plasticity of synaptic events in neuronal networks (Schurmans et al. 1997; Faas et al. 2007; Barinka and Druga 2010). NPY, VIP, and SOM also modulate neural activity and play roles in specific circuits in the nervous system (Hansel et al. 2001; Itri and Colwell 2003; Sun et al. 2003; Viollet et al. 2008; Lepousez et al. 2010; Mickey et al. 2011). Furthermore, the role of layer1 interneurons in the neuronal network is also determined by the site of the cell body in layer 1, where projections arising from different neurons are located. The axons of a substantial number of thalamic neurons project into layer 1a (Vogt et al. 1981; Rubio-Garrido et al. 2009; Cruikshank et al. 2012). They preferentially drive the LS neurons, and weakly trigger the non-LS neurons, the cell bodies of which tend to cluster in layer 1b (Cruikshank et al. 2012). On the other hand, corticocortical terminals are often found in layer 1b (Vogt et al. 1981; Miro-Bernie et al. 2006; Smith et al. 2010) and strongly evoke activation of layer 1b non-LS neurons (Cruikshank et al. 2012). We found that LS/neurogliaform neurons were distributed evenly in layer 1, and BS/VIP+ neurons accumulated in layer 1b, suggesting that they play different roles in layer 1 circuits. Along this line, LS/neurogliaform neurons have been shown to strongly inhibit L2/3 pyramidal neurons (Chu et al. 2003; Wozny and Williams 2011; Cruikshank et al. 2012). However, little is known about the function of BS neurons in the cortical network. Further studies are needed to better understand the precise function of layer 1 interneurons in both the developing and mature neocortex.
This work was supported by grants from the Natural Science Foundation of China (31070947, 31121061, and 31271157), the Ministry of Science and Technology of China (2012CB966300), the Pujiang Talent Project of the Shanghai Science and Technology Committee (10PJ1400700), Innovation Program of Shanghai Municipal Education Commission (12ZZ007), and the Foundation of Ministry of Education of China (20100071120061) to Y.-C.Y.
We thank Dr Song-Hai Shi, Dr Mu-Ming Poo, Dr Xiong-Li Yang, Yvette Chin, Kirsten M. Hively, Kate Peng Gao, and Khadeejah T. Sultan for their comments on the manuscript, Dr Lan Ma for providing GAD67-GFP (Δneo) mice, and the members of the Yu laboratory for their valuable input. Conflict of Interest: None declared.