|Home | About | Journals | Submit | Contact Us | Français|
Though vastly outnumbered inhibitory interneurons critically pace and synchronize excitatory principal cell populations to coordinate cortical information processing. Precision in this control relies upon a remarkable diversity of interneurons primarily determined during embryogenesis by genetic restriction of neuronal potential at the progenitor stage. Like their neocortical counterparts, hippocampal interneurons arise from medial and caudal ganglionic eminence (MGE and CGE) precursors. However, while studies of the early specification of neocortical interneurons are rapidly advancing, similar lineage analyses of hippocampal interneurons have lagged. A “hippocampocentric” investigation is necessary as several hippocampal interneuron subtypes remain poorly represented in the neocortical literature. Thus, we investigated the spatiotemporal origins of hippocampal interneurons using transgenic mice that specifically report MGE- and CGE-derived interneurons either constitutively or inducibly. We found that hippocampal interneurons are produced in two neurogenic waves between E9–12 and E12–16 from MGE and CGE respectively and invade the hippocampus by E14. In the mature hippocampus CGE-derived interneurons primarily localize to superficial layers in strata lacunosum moleculare and deep radiatum while MGE-derived interneurons readily populate all layers with preference for strata pyramidale and oriens. Combined molecular, anatomical, and electrophysiological interrogation of MGE-/CGE-derived interneurons revealed that MGE produces parvalbumin, somatostatin, and nitric oxide synthase expressing interneurons including fast-spiking basket, bistratified, axo-axonic, oriens-lacunosum moleculare, neurogliaform and ivy cells. In contrast CGE-derived interneurons contain cholecystokinin, calretinin, vasoactive intestinal peptide, and reelin including non-fast-spiking basket, Schaffer collateral associated, mossy fiber associated, trilaminar and additional neurogliaform cells. Our findings provide a basic blueprint of the developmental origins of hippocampal interneuron diversity.
Information processing within cortical circuits requires precision in the timing and extent of action potential generation amongst excitatory principal neuron ensembles. Such coordination is largely orchestrated by relatively few highly divergent GABAergic inhibitory interneurons with distinct molecular, anatomical, and electrophysiological properties (Freund and Buzsaki, 1996, Somogyi and Klausberger, 2005). This remarkable diversity allows interneuron networks to provide exquisite spatiotemporal control over information transfer within cortical circuits (McBain and Fisahn, 2001, Klausberger and Somogyi, 2008). Moreover, specific interneuron subtypes are implicated in regulating neuronal proliferation and migration during corticogenesis (Owens and Kreigstein, 2002) as well as postnatal maturation of cortical circuitry (Hensch, 2005; Bonifazi et al., 2009). Indeed several developmentally regulated neurological disorders such as schizophrenia are associated with deficits in the numbers and function of distinct interneuron cohorts (Di Cristo, 2007; Lewis et al., 2005). Thus, a thorough appreciation of interneuron diversity is critical to understanding cortical network development, function, and disease.
Although distinct interneurons require weeks of postnatal maturation to fully attain their subtype-defining characteristics, genetic restriction of neuronal potential at the progenitor stage is a major determinant of interneuron diversity (Wonders and Anderson, 2006, Batista-Brito and Fishell, 2009). Therefore, a complete mapping of the origins of distinct interneuron classes is fundamental to comprehending interneuron diversity. Investigations employing cell transplantation, lineage analysis, and fate mapping strategies have provided unprecedented insight into neocortical interneuron diversity delineating the precise origins, birth dates, and genetic programs governing early specification and migration of distinct neocortical interneuron subpopulations (Wichterle et al., 2001, Butt et al., 2005, 2008, Miyoshi et al., 2007, 2010, Sousa et al., 2009, Gelman et al., 2009, Fogarty et al., 2007, Xu et al., 2004, 2005, 2010; Du et al., 2008; Batista-Brito et al., 2009; Wang et al., 2010). Like their neocortical counterparts, hippocampal interneurons arise from precursors in the medial and caudal ganglionic eminences (MGE/CGE) in the basal telencephalon (Butt et al., 2005; Pleasure et al., 2000; Wonders and Anderson, 2006). However, despite an extensive literature concerning hippocampal interneuron diversity and function, fate mapping of hippocampal interneuron precursors has significantly lagged that of neocortical interneurons. A “hippocampocentric” investigation is necessitated by the fact that several cohorts of neurochemically distinct hippocampal interneurons are poorly represented in the existing neocortical literature. Furthermore, many hippocampal interneurons are classified according to laminar position and axonal projection making it difficult to identify neocortical homologs to infer developmental origins. Such stratification of a given interneuron critically dictates it's afferent/efferent connectivity, hence, governing participation during discrete patterns of hippocampal network activity (Klausberger and Somogyi, 2008). Finally, recent work revealing that neocortical neurogliaform interneurons arise from CGE progenitors (Miyoshi et al., 2010) while most hippocampal neurogliaform cells originate from MGE precursors (Tricoire et al., 2010) points to potential differences between the neocortex and hippocampus.
Here, we performed a longitudinal developmental analysis of the precursors of different mouse hippocampal interneuron subpopulations. We used a multiparametric approach combining genetically inducible fate mapping strategies, including birth dating, with immunohistochemical, electrophysiological, anatomical, and molecular analyses to directly compare and contrast MGE- and CGE-derived hippocampal interneurons.
All experiments were conducted in accordance with animal protocols approved by the National Institutes of Health. Breeding, tamoxifen-treatment and genotyping were performed as described previously (Tricoire et al, 2010). For staging of embryos, noon on the day of the appearance of a vaginal plug was treated as embryonic day 0.5 (E0.5), and the day of birth was considered P0. For migration studies using inducible lines (Mash1CreER:RCE and Olig2CreER:ZEG) pregnant females were administered 4 mg of tamoxifen (Sigma, St. Louis, MO) in corn oil (20 mg/ml) by oral gavage on E10.5, E12.5, E14.5 or E16.5.
All primary antibodies used in the study are described in Table 1. For embryonic characterization pups were dissected from anesthetized dams by cesarean, rinsed in cold phosphate buffered saline (PBS) and incubated in 4% paraformaldehyde in PBS for 1–3 hrs at 4°C. Tissue was then washed thoroughly in cold PBS 3×30 min and cryoprotected in a 25% sucrose / PBS solution at 4°C overnight. Brains were embedded in Tissue Freezing Medium (Triangle Biomedical Sciences, Durham, NC), frozen in a dry ice / ethanol bath and stored at −80°C. Slices were cut on a cryostat (Leica Microsystems, Bannockburn, IL) to a thickness of 20–30 μm directly onto microslides and processed immediately for immunohistochemistry. Tissue was first washed in PBS for 30 minutes. Slices were then transferred to a ProHisto Amplifier Tray (ProHisto, Columbia, SC) to be incubated in Amplifying IHC Wash Buffer (ProHisto) + 10% goat serum to block for 2 hrs. Subsequently, slices were incubated in chicken anti-GFP antibody (1:2000; Aves Labs, Tigard, OR) for 24–48 hours at 4°C. Samples were then rinsed in wash buffer for 3 × 30 min and then transferred to secondary antibody goat anti-chicken alexafluor 488 (1:1000) (Invitrogen, Carlsbad, CA). Tissue was again rinsed 3×30 minutes and slides were mounted and counterstained with Vectashield + DAPI (Vector Laboratories, Inc, Burlingame, CA). Immunofluorescence was imaged using an Olympus Provis AX70 microscope (Olympus America, Center Valley, PA) and images were captured using QCapture Suite (QImaging, Surrey, BC). In order to ensure that our measurements are not biased by the rapidly increasing size of the mouse brain during this period all measurements are presented as cells/mm3. To achieve this, the surface area of the region of interest was measured using ImageJ software (National Institutes of Health, Bethesda, Maryland, USA) and multiplied by the thickness of the slice. GFP+ interneurons were then manually counted using the cell counter plugin of ImageJ.
For postnatal immunohistochemical characterization of all markers except vasoactive intestinal peptide (VIP) and somatostatin (SOM) mice were perfused transcardially using a 0.1 M PBS solution containing 4% paraformaldehyde followed by overnight postfixation at 4C. For VIP and SOM staining, we used 4% paraformadehyde in PB 0.1M as fixative and postfixed for 2 hours at room temperature. Brains were then cryoprotected using 20–30% sucrose/PBS solution, sliced to 40 μm thickness using a freezing microtome, and kept at 4°C in PBS. After washing in PBS, free-floating sections were blocked for 2h at room temperature in a PBS/0.5% Triton X-100/ 1% BSA/10% normal goat serum (NGS) solution before being incubated overnight at 4°C with primary antibodies diluted in a PBS/0.5% Triton X-100/1% BSA/1% NGS solution (BGTPBS). Slices were washed with BGT-PBS before being incubated for 2 h at room temperature with secondary antibodies diluted in BGT-PBS. Nuclear counterstaining was performed with 100 ng/ml 4', 6-diamidino-2-phenylindole (DAPI, Invitrogen, Carlsbad, CA) solution in PBS for 20 min. After extensive washing in PBS, slices were mounted on gelatin-coated slides in Prolong Gold (Invitrogen, Carlsbad, CA). Secondary antibodies were used in the following concentrations: goat anti-chicken alexafluor488, F(ab)2 fragment of goat anti-rabbit alexafluor555, Goat anti guinea pig alexafluor555, Goat anti-rat alexafluor555 and Goat anti mouse alexafluor555 (1:500; Invitrogen). Fluorescent images were captured using a Retiga 4000R cooled CCD camera (Qimaging, Surrey, Canada) or using a Live duo scan confocal system (Zeiss, Germany).
M2R cellular staining differs between interneurons subtypes (Hajos et al, 1998). In this study, we focused only on those interneurons exhibiting somatodendritic M2R immunostaining which have been shown to target either septum or subiculum (Hajos et al, 1998, Jinno et al, 2007). Cholecystokinin (CCK) staining was performed using mouse ABC elite kit (Vector Laboratories) with tetramethylrhodamine conjugated tyramide (Perkin Elmer) as fluorogen. After washing in PBS sections were treated with 3% H2O2 for 30min in PBS then washed 3×15min in PBS. Blocking and washing steps were the same as above. After incubation for 2 nights at 4°C with mouse anti-CCK (1/1000), sections were incubated with biotinylated goat anti mouse (1/200, Vector Laboratories). After 3×15min wash in PBS, sections were incubated for 1h at room temperature with freshly prepared ABC complex in PBS 0.1% triton X-100 then washed again 3×15min in PBS. Staining was developed for 5 min by applying tetramethylrhodamine conjugated tyramide diluted at 1/75 in amplification plus reagent then stopped by several extensive washes in PBS.
For all embryonic and postnatal stages counting was performed on a minimum of 6 hippocampal sections from at least 2 animals. Adjascent serial sections were never counted for the same marker to avoid any potential double counting of hemisected neurons. For analysis of the fraction of GFP+ cells that are also positive for VIP+ in the GAD65-GFP line, we included cells previously shown in the supplemental material of Cea-del Rio et al, (2010) (360 of 556 GFP+ cells counted in Figure 4C). Similarly, quantitative data in Figures 3C–D and and5C5C describing the colocalization of nNOS with GPF in Nkx2-1Cre:RCE and Olig2CreER:ZEG mice were previously reported in Tricoire et al., 2010.
P14-P21 mice (of various genotypes, as indicated throughout the text) were anesthetized with isoflurane, and the brain dissected out in ice-cold saline solution containing (in mM): 130 NaCl, 25 NaHCO3, 1.25 NaH2PO4, 3.5 KCl, 4.5 MgCl2, 0.5 CaCl2, 10 glucose, saturated with 95% O2 and 5% CO2 (pH 7.4). Transverse hippocampal slices (300 μm) were cut using a VT-1000S vibratome (Leica Microsystems, Bannockburn, IL) and incubated in the above solution at 35°C for recovery (1 h), after which they were kept at room temperature until use. Individual slices were transferred to an upright microscope and visualized with infrared differential interference contrast microscopy (Axioscope FS2; Zeiss, Oberkochen, Germany). Slices were perfused (2 ml/min) with extracellular solution composed of (in mM) 130 NaCl, 24 NaHCO3, 3.5 KCl, 1.25 NaH2PO4, 2.5 CaCl2, 1.5 MgCl2, and 10 glucose, saturated with 95% O2 and 5% CO2 (pH 7.4). Recordings were performed at 32–34°C with electrodes (3–5 MΣ) pulled from borosilicate glass (World precision instruments) filled with 150mM K-gluconate, 3 MgCl2, 0.5 EGTA, 2 MgATP, 0.3 Na2GTP and 10 HEPES plus 2 mg/ml biocytin (Sigma). The pH was adjusted to 7.4 with KOH (~300 mOsm, free K+ ~ 155mM). Whole-cell patch-clamp recordings were made using a Multiclamp 700A or 700B amplifier (Molecular Devices, Sunnyvale, CA). Signals were filtered at 3 kHz (Bessel filter; Frequency Devices, Haverhill, MA) and digitized at 20 kHz (Digidata 1322A or 1440A and pClamp 9.2 or 10.2 Software; Molecular Devices). Recordings were not corrected for a liquid junction potential. To accurately determine neuronal resting membrane potential without disrupting the cell's intracellular environment, we monitored potassium-channel activation during depolarizing voltage ramps (from −100 to +200 mV) applied to cell-attached patches prior to breakthrough into the whole-cell configuration (Verheugen et al., 1999; Banke and McBain, 2006). After breaking into whole-cell configuration, membrane potential was biased to −60mV by constant current injection. Input resistance (Rm) was measured using a linear regression of voltage deflections (± 15 mV from resting potential, ~60 mV) in response to 2 s current steps of six to ten different amplitudes (increment 5 pA). Membrane time constant (m) was calculated from the mean responses to 20 successive hyperpolarizing current pulses (−20pA; 400 ms) and was determined by fitting voltage responses with a single exponential function. Action potential threshold was defined as the voltage at which the slope trajectory reaches 10 mV/ms (Stuart and Hausser, 1994). Action potential (AP) amplitude was defined as the difference in membrane potential between threshold and the peak. Afterhyperpolarization (AHP) amplitude was defined as the difference between action potential threshold and the most negative membrane potential attained during the AHP. These properties were measured for the first two action potentials elicited by a depolarizing 800 ms-long current pulse of amplitude just sufficient to bring the cell to threshold for AP generation. The adaptation ratio was defined as the ratio of the average of the last 2–3 interspike intervals relative to the first interspike interval during a 800-ms-long spike train elicited using twice the current injection necessary to obtain a just suprathreshold response. Firing frequency was calculated from the number of spikes observed during the same spike train. In some interneurons injection of hyperpolarizing current pulses induces pronounced “sag” indicative of a hyperpolarization-activated cationic current (Ih) that activates following the initial peak hyperpolarization (Maccaferri and McBain, 1996). To determine the sag index of each cell we used a series of 800ms negative current steps to create V-I plots of the peak negative voltage deflection (Vhyp) and the the steady state voltage deflection (average voltage over the last 200 ms of the current step; Vsag) and used the ratio of Vrest-Vsag/Vrest-Vhyp for current injections corresponding to Vsag=−80mV determined from polynomial fits of the V-I plots. All electrophysiological parameters were measured in pClamp or using procedures written in Igor 6 (Wavemetrics, Portland, OR).
The standard interneuron characterization scPCR protocol was designed to detect simultaneously the transcripts of parvalbumin (PV), calretinin (CR), GABA synthesizing enzymes GAD65 and 67, neuropeptide Y (NPY), somatostatin (SOM), vasoactive interstinal peptide (VIP), cholecystokinin (CCK), neuronal isoform of nitric oxide synthase (nNOS), vesicular glutamate transporter type 3 (VGluT3), preproenkephalin, neuronal PAS domain protein 1 and 3 (Npas1 and3), LIM/homeobox protein 6 (Lhx6), chicken ovalbumin upstream promoter transcription factor 2 (CoupTFII), and serotonin receptor type 3A (5HT3). For Lhx6 and CoupTFII, primers were designed to amplify all splice variants. At the end of the recording, cell cytoplasm was aspirated into the recording pipette while maintaining the tight seal. The recording pipette was then carefully withdrawn from the cell and an outside-out patch was pulled to leave the cell intact for subsequent anatomical recovery, and to prevent the entry of contaminant tissue upon removal of the pipette from the slice/recording chamber. The contents of the pipette were expelled into a test tube, and subjected to reverse transcription (RT) and two steps of polymerase chain reaction (PCR) as described previously (Lambolez et al., 1992; Cauli et al., 1997; Karagiannis et al., 2009; Tricoire et al., 2010; Cea-del Rio et al., 2010). The cDNA present in the RT reaction were first amplified simultaneously using all primer pairs described in Table S1 in a total volume of 100μl (for each primer pair, the sense and antisense primers were positioned on two different exons). A second round of PCR was then performed using 2 μl of the first PCR product as a template. In the second round, each cDNA was amplified individually with a second pair of primers internal to the primer pair used in the first PCR (nested primers, see Table 2) and positioned on two different exons. As a control for our ability to detect transcript when present the RT-PCR protocol was tested on 100 pg of total RNA purified from mouse whole brain and all transcripts were detected (not shown). The sizes of the PCR-generated fragments were as predicted by the mRNA sequences (see Table 2). Routinely, a control for mRNA contamination from surrounding tissue was performed by placing a patch pipette in the slice without establishing a seal and following the removal of the pipette its contents was processed as described above. PCR product was never obtained using this protocol. Moreover, in separate controls VGluT1 mRNA, which is always detected by our standard scPCR protocol for pyramidal cell harvests, was detected in <10% (17/175) of interneuron harvests indicating the potential for false positive signals from surrounding pyramidal cells is minimal for this scPCR protocol (for eg. see figure 3 of Vullhorst et al., 2009). Finally, to determine the potential degree of contamination by pericellular glial cells attached to targeted interneurons we probed for several glial markers in separate control interneuron harvests. Again the lack of significant detection of GFAP (0/10), NG2 (0/10), MBP (1/10) and S100β (0/10) in these interneuron harvests, all of which were positive for GAD65/67 (10/10), further indicates limited surrounding tissue contaminant detection with this protocol.
cDNA were generated as described above and subjected to 14 pre-amplification cycles in presence of a mix of specific taqman gene assays prior to aliquoting for individual measurement with the same specific gene assays by real-time PCR. Each 10 μl cDNA synthesis reaction was combined with 25 μl pre-amplification master mix (Applied Biosystems), 12.5 μl of pooled assays mix (each assay at 0.2×, final concentration: 0.05×) and 2.5μl RNAse free water. Pooled assays mix was prepared from individual 20× stock gene assays (primers +probe) and diluted in 1× TE. Gene assays for PV (ID: Mm00443100_m1), CCK (ID: Mm00446170_m1), VGluT1 (ID: Mm00812886_m1), GAD65 (ID: Mm00484623_m1) and beta actin (Taqman Endogenous control) were purchased from Applied Biosystems. Each 20× gene assay contains two unlabeled PCR primers at 18 μM each, and a FAM™ dye-labeled TaqMan MGB probe at 5 μM, except for beta actin control that contains PCR primers at 3μM and VIC dye-labeled probe at 5 μM. Gene assays were selected with primer sequences on two different exons. Thermocycling was performed as follows: 95°C 10 min, 12 cycles of (95°C, 15 s; 60°C, 4 min). This procedure preamplifies cDNA without introducing amplification bias to the sample. For each real-time PCR reaction, 5 μl aliquots of the pre-amplification reaction were combined to 10 μl gene expression master mix (Applied biosystems), 1 μl 20× gene assay and 4 μl RNAse free water. Real-time PCR was performed on StepOne instrument as follows: 50°C 2 min, 95°C 10 min, 60 cycles of (95°C, 15 s; 60°C, 1 min). Each gene was tested in duplicate and two cells were analyzed in parallel and subjected to analysis using the StepOne software. Cycle threshold (Ct) values were determined for each probe and then normalized relative to the Ct value for beta actin in the same sample serving as an endogenous control. In this way ΔCt(probed transcript) = Ct(actin) -Ct(probed transcript) provides a relative abundance measure for a given mRNA species in the sample with positive ΔCt values indicating the transcript is in greater abundance than that of actin while negative ΔCt values indicate levels lower than that of actin. As the amount of transcript doubles with each cycle the relative abundance of a given mRNA species within the sample in relation to actin is then determined by 2ΔCt.
Unsupervised clustering was performed in Matlab (MathWorks, Natick, MA) using 10 electrophysiological parameters (see above), 16 molecular parameters (GAD65, GAD67, PV, CR, NPY, VIP, SOM, CCK, nNOS, VGluT3, 5HT3, CoupTFII, Lhx6, Npas1 and 3), and embryonic origin (designated as +1 or −1 for MGE and CGE origin respectively). For each parameter, experimental observations were standardized by centering to the mean and dividing by the stardard deviation. For both Ward and K-means aggregating methods (Ward, 1963; MacQueen, 1967; Hartigan and Wong, 1979), euclidean distance was first calculated between experimental standardized data sets. The K-means algorithm was run starting from 1000 different random initial positions of the K cluster centroids. The best value of K was determined by comparison with the partition obtained from the Ward's algorithm using the adjusted Rand index as a measure of agreement (Rand, 1971, Hubert and Arabie, 1985). The significance of the classification obtained with the K-means method was further validated by comparison with randomized databases as described in Karagiannis et al. (2007). Briefly, a randomized dataset was built by permuting randomly and independently over all the cells the observations of one or several parameters and leaving the observation of other features unaffected. This operation did not alter the mean values and the SDs of the randomized parameters but disrupted the structured correlations between the measurements of the different features. Clustering of the randomized dataset was performed using the same parameters as for the K-means clustering of the original dataset. The quality of different clustering was quantitatively assessed by means of silhouette analyses (Rousseeuw, 1987). Given a data point i in a cluster A, let a(i) denote the average Euclidean distance between i and other data points in the cluster A and let b(i) denote the average Euclidean distance between i and points in the second closest cluster. The silhouette value S(i) was then computed for each cell using the following formula S(i)= [b(i)-a(i)]/Max[a(i),b(i)]. The global silhouette width S' for a given partitioning P is defined as the average S(i) over all the cells of the dataset and is used as an overall measure of quality of a possible clustering. For each set of randomized parameters, the silhouette width of the clusterings of 20 independent different randomized databases were computed and compared with the silhouette width of the K-means clustering of the original, nonrandomized dataset. The effective loss of quality was assessed by calculating the quantity [average(S'(randomized))-S'(original)]/S'(original). Comparisons were performed using Student's t-tests. In testing the effect of the number of randomized parameters on the silhouette width, two distinct sequences were used. Either the parameters to be scrambled were randomly picked among the list of 37 parameters, or scrambled parameters were picked in a pre-defined manner, starting with parameters that least altered silhouette width individually and continuing with parameters with increasing influence on silhouette width.
To begin our study we examined the migratory routes and timecourse of hippocampal invasion by cells derived from MGE and CGE progenitors throughout embryonic and postnatal development. In order to delineate the MGE- and CGE-derived cohorts of interneurons we first utilized transgenic mouse lines that selectively and constitutively GFP label cells derived from either of these two progenitor regions. Nkx2-1 is a homeodomain transcription factor which is expressed in the MGE, but absent from the CGE, that is responsible for MGE patterning and interneuron specification (Xu et al, 2004, 2008; Fogarty et al., 2007; Butt et al., 2008). Accordingly, we used Nkx2-1Cre:RCE transgenic mice to examine MGE-derived interneurons (Soussa et al, 2009; Tricoire et al, 2010). For examination of CGE-derived interneurons we took advantage of GAD65-GFP transgenic mouse line that specifically labels interneurons derived from the CGE progenitor pool (Lopez-Bendito et al, 2004).
Consistent with previous reports (Lavdas et al, 1999 and Rubin et al, 2010) interneurons derived from the MGE and CGE follow similar routes of migration from their site of genesis to the hippocampus, tangentially migrating through both the marginal zone (MZ) and the intermediate/subventricular zone (IZ/SVZ) streams (Figure 1A–C; F–H). The first interneurons enter the hippocampus on or slightly after E14.5 with peak invasion occurring between E15–E18. Indeed peak densities for interneurons of both MGE and CGE lineages occurred at a late postnatal stage, E18, reaching maximal values of 23033 ± 988 and 10271 ± 552.9 cells/ mm3 for MGE- and CGE-derived GFP+ interneurons respectively (Figure 1K,L). Interestingly, it appears that CGE-derived interneurons are added at a slower rate with a minor temporal delay in comparison to MGE-derived interneurons (Figure 1K,L) consistent with reports in the neocortex (Rubin et al, 2010; Miyoshi et al., 2007, 2010). Postnatally, both interneuron cohorts exhibit a dramatic reduction in cell densities with MGE-derived interneurons falling to 12097 ± 265 cells/ mm3 and CGE-derived interneurons falling to 3889 ± 223 cells/ mm3 within 5 days after birth representing 48% and 62% decreases respectively from the levels at P0 (Figure 1K,L). The density of interneurons from both origins continues to decrease ultimately falling to 1501.4 ± 62.5 and 3585 ± 143.4 cells/mm3 for CGE and MGE interneurons repectively. Upon arrival at the hippocampus interneurons of both lineages can be observed to populate all layers with a majority of the migrating interneurons following the MZ stream into stratum lacunosom molecular (s.l.m.). The interneurons invade the CA1 region on or around E15.5 and by E17.5 can be found in all hippocampal regions (CA1, CA3 and dentrate gyrus).
In general our migration studies illustrate a rapid increase in hippocampal interneuron density from E14.5–E18, followed by a dramatic decrease between P0–P30. Interestingly, comparison of CGE-derived interneuron densitites between the hippocampus and neocortex reveals almost identical results across all embryonic and postnatal time points examined (Lopez-Bendito et al, 2004). While some of the postnatal reduction in interneuron density reflects dilution due to expanding brain volume, earlier work has demonstrated a peak in neuronal death between P4–P8, with continued loss during the first two postnatal weeks indicating another potential factor contributing to the dramatic reduction in interneuron density (Verney et al, 2000). Throughout development CGE-derived interneurons primarily populated superficial layers of the hippocampus concentrating in stratum lacunosum moleculare (s.l.m.) and deep stratum radiatum (s.r.) whereas MGE-derived interneurons distributed throughout all layers similar to MGE/CGE patterning in the neocortex (Butt et al, 20005; Miyoshi et al, 2010; Rubin et al., 2010; Lee et al., 2010).
The constitutive expression of GFP in cells of the GAD65-GFP and Nkx2-1 transgenic mice after exiting the CGE and MGE progenitor pools allows for a global investigation of large numbers of interneurons born throughout embryogenesis. However, recent work has illustrated that an inteneuron's ultimate identity within a given cortical circuit is influenced both by birth location and date (Butt et al., 2005; Miyoshi et al., 2007, 2010). Thus, to investigate any potential role for birth date on migration to and invasion of interneurons into the hippocampus we utilized temporally inducible genetic fate mapping (Joyner and Zervas, 2006) The Mash1CreER driver line contains a tamoxifen inducible form of Cre recombinase regulated by expression of the transcription factor Mash1. Although Mash1 is expressed throughout the entire ventral telencephalon (Guillemot et al., 1993), this particular driver line when crossed with the Cre-dependent EGFP reporter (RCE:loxP) labels interneurons of the LGE and CGE, but not the MGE (Miyoshi et al., 2010). Importantly, the LGE does not contribute appreciable numbers of cortical interneurons (Wichterle et al., 2001; Nery et al., 2002; Yozu et al., 2005) allowing use of the Mash1CreER:RCE mouse to investigate CGE derived hippocampal interneurons of known birth date by providing temporally limited tamoxifen exposure at different times during embryonic development. To similarly examine temporally distinct cohorts of MGE-derived interneurons the Olig2CreER:ZEG mouse line can be used. In this line tamoxifen induced activity of Cre recombinase, under the control of the Olig2 locus, drives the expression of GFP reporter Z/EG selectively in MGE-derived interneurons (Miyoshi et al., 2007; Tricoire et al., 2010). In both lines a single tamoxifen administration to pregnant females allows for approximately 24hrs of CreER activity allowing for birth dating of GFP+ interneurons in the pups to within 1 day of tamoxifen treatment (Miyoshi et al., 2007, 2010, Tricoire et al., 2010). For these studies tamoxifen was administered at either E10.5 and E12.5 or E12.5, E14.5 and E16.5 for Olig2CreER:ZEG and Mash1CreER:RCE respectively to cover the beginning and peak time points for neurogenic waves arising from MGE and CGE (Miyoshi et al., 2007, 2010, and see below). Tissue from the pups was then examined 2,3 or 4 days later to create “snapshots” of the migration from the MGE and CGE, as well as postnatally to look at the integration of labeled interneurons into the mature circuit.
For interneurons born in the CGE (Mash1CreER:RCE) labeled by tamoxifen administration on E12.5, migration to the hippocampus required approximately 72 hours, while those labeled by tamoxifen treatment on E14.5 and E16.5 arrived in the hippocampus within 48 hrs despite a longer migratory route imposed by the rapidly increasing size of the brain (Figure 2A–F, H). Embryonically, CGE-derived interneurons tamoxifen fate mapped at E14.5 were the most concentrated, reaching 6687.8 ± 343 cells / mm3 by E18.5, while cells revealed by tamoxfien administration on E12.5 and E16.5 had maximum densities of 5295 ± 243 and 4107 ± 272 cells / mm3 respectively at the same stage of development (Figure 2H). This prevalence of E14.5 cells was not evident in the juvenile hippocampus as cell densities for E12.5, E14.5 and E16.5 tamoxifen fate mapped interneurons decreased to similar values (320±175, 454 ± 123 and 279 ± 79 cells / mm3, respectively) by P30 (Figure 2H).
In the Olig2 mouse line considerably fewer GFP+ interneurons were generated for a given tamoxifen pulse due to poor recombination efficiency in this transgenic line (Miyoshi et al., 2007). Nonetheless, the same general trends as observed for temporally distinct cohorts of CGE-derived interneurons were evident. E10.5 tamoxifen fate mapped MGE-derived interneurons required four days to migrate to the hippocampus reaching a concentration of 253 ± 49 cells / mm3 by E14.5 (Figure 2G). In contrast interneurons born later in the MGE, revealed by tamoxifen treatment on E12.5, reached the hippocampus within 48hrs at a concentration of 573 ± 123 cells / mm3 (Figure 2G). For these MGE-derived interneurons we again observed a rapid reduction in interneuron densities to 42 ± 22 and 77 ± 31.6 cells / mm3 for E10.5 and E12.5 tamoxifen fate mappedinterneurons respectively by P30 (Figure 2G). Similar to results from the Nkx2-1Cre:RCE and GAD65-GFP mice, fate-mapped interneurons from both the CGE and MGE localize to all layers of the hippocampus, and invade CA1 one day prior to CA3 and DG.
Overall our interneuron migration times from both ganglionic eminences to the hippocampus (48–72 hours) are longer than reported for migration to the neocortex (24–48 hours; Wichterle et al, 20001; Miyoshi et al, 2007 and Miyoshi and Fishell, 2010) likely reflecting the greater distance that must be traveled. The increased migration rate we observed for cells born later in development also occurs for neocortically destined interneurons (Miyoshi et al, 2010), and may result from a shift in the balance between attractive and repulsive guidance cues within the MZ and IZ/SVZ migratory paths (Lopez-Bendito et al, 2008; Chedotal and Rijli, 2009;Marin et al, 2010).
Classically, hippocampal GABAergic interneuron subtypes have been distinguished to a large degree based on unique neurochemical signatures observed with immunostaining (Freund and Buzsaki, 1996; McBain and Fisahn, 2001; Matyas et al., 2004; Somogyi and Klausberger, 2005). Thus, we next explored the developmental origin of molecularly distinct interneuron cohorts by immunostaining hippocampal sections from one- to two-month old Nkx2-1Cre:RCE and GAD65-GFP mice probing for well established markers that cover much of the overall known diversity across hippocampal interneurons (Figures 3 and and4).4). For these studies we examined the overlap between a given marker and GFP, focussing on cells within the CA1 subfield where interneuron diversity has been best characterized and laminar location aids in identifying specific interneuron subtypes (reviewed in Klausberger and Somogyi, 2008). In MGE-derived interneurons of the Nkx2-1Cre:RCE line the markers PV, SOM, and nNOS strongly colocalized with GFP in largely non-overlapping populations of cells located predominantly in strata pyramidale (s.p.), oriens (s.o.), and lacunosum moleculare respectively (s.l.m.; Figure 3). Indeed, 86% of PV+, 68% of SOM+, and 73% of nNOS+ interneurons were GFP labelled in Nkx2-1Cre:RCE mice (Figure 3A,C). Moreover, PV+, SOM+ and nNOS+ interneurons combined accounted for >90% of MGE-derived interneurons representing 30% (n=395), 24% (n=417), and 37% respectively of the total GFP+ cohort of CA1 cells in Nkx2-1Cre:RCE mice (Figure 3D). In stark contrast, PV+, SOM+ and nNOS+ cells were rarely observed to colocalize with CGE-derived GFP+ cells of GAD65-GFP mice confirming that MGE and CGE give rise to neurochemically distinct interneuron cohorts (Figure 3B–C). Instead CGE-derived GFP+ interneurons commonly stained for CR, CCK, M2R, and VIP which accounted for 19%, 27%, 6%, and 12% of GFP+ interneurons in GAD65-GFP mice respectively (Figure 4B,D). Conversely, 20%, 53%, 36% and 39% of CR+, CCK+, M2R+, and VIP+ interneurons respectively also stained for GFP in the GAD65-GFP mouse line (Fig 4C, note: our quantitative data for GFP and VIP colocalization reported here includes counts previously reported in Ceadel Rio et al, 2010). Of these markers only CR exhibited any significant, albeit minor (2% of GFP+ cells), overlap with MGE-derived GFP+ cells in Nkx2-1Cre:RCE mice further confirming that MGE- and CGE give rise to neurochemically distinct hippocampal interneuron cohorts (Figure 4A,C). It is important to note that our overall counts significantly underestimate the colocalization of GFP and CR in interneurons because of the staining of large numbers of CR-containing Cajal-Retzius cells within s.l.m. (Marchionni et al., 2010) which have distinct embryonic origins from GABAergic interneurons and hence are not GFP labelled in our mouse lines (Soriano et al., 1994, Soriano et al., 2005). Thus, while in total only 15% of all CR+ cells were GFP labelled in GAD65-GFP mice this number more than doubles to 39% when considering s.r. in isolation.
In general our findings to this point are consistent with reports in the neocortex where MGE and CGE give rise to nonoverlapping populations of neurochemically distinct interneurons. Indeed similar to our current findings MGE-derived neocortical interneurons comprise PV+ and SOM+ subgroups with a very minor contribution to CR+ cells that are likely a subset of the SOM+ cohort (Cobos et al., 2005; Flames et al., 2007; Miyoshi et al., 2007; Xu et al., 2004,2006; Fogarty et al., 2007; Butt et al., 2005). In contrast the CGE contributes CCK+, VIP+, and the majority of CR+ neocortical interneurons, also consistent with our current observations (Lopez-Bendito et al., 2004; Xu et al., 2004; Miyoshi et al., 2010; Rubin et al., 2010). In addition we confirmed that hippocampal nNOS+ interneurons primarily derive from MGE consistent with the MGE origin of NPY+ hippocampal cells as NPY and nNOS commonly colocalise in hippocampal interneurons (Fogarty et al., 2007; Pleasure, 2000; Price et al., 2005; Fuentealba et al., 2008a). Indeed we have previously shown that NPY+/nNOS+ neurogliaform and Ivy cells of the hippocampus have an MGE lineage (Tricoire et al., 2010). Our results also demonstrate a CGE origin for all M2R+ hippocampal interneurons. As M2R immunoreactivity is associated with a subset of projection interneurons in the hippocampus, such as the stratum oriens hippocampo-subiculum projecting trilaminar cell (Sik et al., 1995; Ferragutti et al., 2005; Jinno et al., 2007; Hajos et al., 1998), our findings indicate a CGE-origin for M2R+ projection cells.
Recently, the CGE has also been demonstrated to give rise to a relatively large population of reelin+ neocortical interneurons that do not overlap with the VIP/CR expressing cohort (Miyoshi et al., 2010; Rubin et al., 2010). Although many SOM+ interneurons coexpress reelin (Pesold et al., 1999; Fuentealba et al., 2010; Ramos-Moreno, 2006; Miyoshi et al., 2010) the CGE gives rise to a distinct population of reelin+/SOM− neocortical interneurons. In the hippocampus we similarly found reelin immunoreactivity in GFP+ cells for both MGE- and CGE- derived interneurons but with distinct laminar localizations (Figure 3A–D). MGE-derived GFP+/reelin+ cells in Nkx2-1Cre:RCE mice strongly localized to s.o. (82% of the total reelin+/GFP+ group) likely representing SOM/reelin coexpressing cells (Miyoshi et al., 2010; Fuentealba et al., 2010). In addition a smaller population (18%) of MGE-derived reelin+/GFP+ cells were present in s.l.m. which are likely to represent MGE-derived neurogliaform cells (Tricoire et al., 2010; Fuentealba et al., 2010). GFP+/reelin+ cells in GAD65-GFP mice were found almost exclusively in s.l.m. (79% of GFP+/reelin+) likely also representing neurogliaform cells but of CGE origin (Tricoire et al., 2010; Fuentealba et al., 2010). In addition reelin, like CR, labelled many GFP negative Cajal-Retzius cells in s.l.m. leading to an underestimate of the contribution of GFP+ interneurons to the overall reelin labelled cohort of cells (Figure 3B–C).
Finally, we examined the expression pattern of CoupTFII which has recently been associated with CGE-derived cortical interneurons (Kanatani et al., 2008, Fuentealba et al., 2010, Tricoire et al., 2010; Miyoshi et al., 2010). Consistent with these reports we also found CoupTFII immunoreactivity to commonly label GFP+ interneurons from GAD65-GFP mice (43% of GFP+ cells) with only minimal colocalization in Nkx2-1Cre:RCE mice (Figure 4A–D).
As mentioned above birth date has been shown to influence fate determination of cortical interneurons (Butt et al, 2005, Miyoshi et al, 2007, Miyoshi et al, 2010). Thus, to examine any potential temporal order to the generation of neurochemically distinct MGE- and CGE-derived interneurons we combined inducible genetic fate mapping with immunolabeling for our panel of interneuron markers in the Olig2CreER:ZEG and Mash1CreER:RCE double transgenic mice (Figures 5 and and6).6). Once again pregnant females were administered a single dose of tamoxifen by gavaging at different time points during gestation and then tissue from 3–4-week-old offspring was probed for a given marker to examine the colocalization of the marker with GFP in the CA1 area of the hippocampus. For the MGE-derived cohort tamoxifen was administrated to Olig2CreER:ZEG mothers between E9.5 and E15.5 (Miyoshi et al, 2007). However, we did not observe any GFP labelled interneurons in the hippocampus after treatment later than E13.5 despite the presence of GFP+ cells in the neocortex which confirmed that tamoxifen was successfully administrated. This indicates that hippocampal MGE derived interneurons are generated earlier and within a narrower temporal window than their neocortical homologs. For the CGE-derived cohort, tamoxifen was administrated to Mash1CreER:RCE mothers between E10.5 and E16.5 (Miyoshi et al, 2010). However, as very few GFP labelled cells were observed in the hippocampus for E10.5 and E11.5 treatments we limited our characterisation to interneurons fate-mapped from E12.5–E16.5. It is important to note that although the MGE generates significantly more cortical interneurons than the CGE (Miyoshi et al., 2010), the Olig2CreER:ZEG mouse line yields considerably lower numbers of GFP positive cells than with the Mash1CreER:RCE line for the same dose of tamoxifen due to the weak recombination efficiency within the Olig2 line (Miyoshi et al., 2007).
Consistent with our findings in the Nkx2-1Cre:RCE line PV+, SOM+ and nNOS+ interneurons represented the major populations of GFP+ fate-mapped interneurons found in the Olig2CreER:ZEG line. Although all three of these interneuron cohorts were observed for each tamoxifen time point a subtle temporal order to their generation was evident (Figure 5A–C). Whereas nNOS interneurons were generated in a bimodal manner with early (E9.5–E10.5; note, data shown in Figure 5C are replotted from Tricoire et al, 2010) and late (E13.5) peaks, PV+ and SOM+ cells emerged from the MGE primarily at early tamoxifen injection time points (E9.5 and E10.5) with their production dramatically falling off to by E13.5.
In general the Mash1CreER:RCE line revealed a delayed production of CGE-derived hippocampal interneurons relative to the MGE consistent with findings in the neocortex (Miyoshi et al., 2007, 2010; Rubin et al., 2010; Lee et al., 2010). Examination of specific subpopulations arising from the CGE indicated that CCK+, VIP+, and M2R+ interneurons are primarily generated at early time points (tamoxifen fate mapped at E12.5 and E13.5) with production falling through the E16.5 tamoxifen time point when CR+ cells become the dominant population emerging from the CGE (Figure 6A–C and F). In contrast reelin expressing interneurons are produced consistently throughout this developmental period representing approximately 40% of CGE-derived interneurons at all time points examined (Figure 6E). Similarly, CoupTFII expressing cells were consistently observed to emerge from CGE progenitors throughout development (Figure 6D) as expected for a marker that labels multiple cohorts of CGE-derived interneurons (Kanatani et al., 2008).
Overall the temporal profiles of hippocampal PV+, SOM+, VIP+ and reelin+ interneuron generation are similar to those of the neocortex but with a slight shift towards earlier embryonic stages (Miyoshi et al, 2007; Batista-Brito et al, 2009; Miyoshi et al, 2010). In contrast, CR+ internerneurons, which often coexpress VIP in the neocortex (Cauli et al, 1997), are generated at much later stages if they are destined for the hippocampus.
While immunostaining patterns are useful for broadly parsing major interneuron subgroups the ultimate identification of a given interneuron requires combined knowledge of its molecular, electrophysiological, and anatomical properties. Thus, we next investigated MGE- and CGE-derived interneurons using patch-clamp recording techniques in combination with post hoc single cell RT-PCR (scPCR) and morphological inspection of GFP+ interneurons in acute brain slices from Nkx2-1Cre:RCE and GAD65-GFP mice (Figures 7 and and8).8). Electrophysiological profiles included 20 features of intrinsic excitability and firing properties that were assayed while the scPCR protocol was designed to detect simultaneously the transcripts of 16 common interneuron markers, including most of those assayed by immunocytochemistry above (see Materials and Methods).
In the Nkx2-1Cre:RCE mice we obtained complete scPCR and electrophysiological profiles for 61 recorded interneurons throughout the different layers of CA1 and CA3; anatomical reconstructions, scPCR results, and basic electrophysiological properties of cells that represent the major subtypes observed are illustrated in Figure 7A–K. Consistent with our staining data we frequently encountered putative fast spiking PV+ perisomatic targeting basket cells (Figure 7A) as expected from the MGE origin of this cell's neocortical counterpart (Butt et al., 2005; Miyoshi et al., 2007). Similar to fast spiking basket cells from an electrophysiological perspective, we also regularly recorded another group of MGE-derived fast spiking cells displaying anatomy consistent with previously described SOM+/PV+ bistratified cells (BiCs) with axons that target the apical and basolateral dendritic domains of pyramidal cells while mostly avoiding the somatic region targeted by basket cells (Buhl., et al., 1994, 1996; Halasy et al., 1996; Maccaferri et al., 2000; Klausberger et al., 2004). The dendritic arbors of these BiCs could be oriented primarily horizontally around cell bodies located in stratum oriens (s.o.; O-BiCs; Figure 7B) or more vertically along the s.o. to s.l.m. axis from cell bodies located in s.p. (Figure 7C). On rare occasion we also encountered fast spiking cells with axonal arbors containing terminal boutons concentrated in the half of s.p bordering s.o. and in the half of s.o. adjoining s.p. suggestive of PV+ axo-axonic cells (Figure 7D; Li et al., 1992; Buhl et al., 1994, Maccaferri et al., 2000; Klausberger et al., 2003; Ganter et al., 2004). As in our previous investigation (Tricoire et al., 2010) we often recorded late-spiking, weakly accommodating, NPY/nNOS transcript containing Ivy and NGF cells throughout all CA layers (Figure 7I–K; Vida et al., 1998; Funetealba et al., 2008, 2010; Price et al., 2005). Another easily recognizable population arising from the MGE progenitor pool was the oriens-lacunosum moleculare (O-LM) interneuron (Figure 7E), a prominent SOM+ interneuron subgroup of the hippocampus (McBain et al., 1994; Sik et al., 1995; Maccaferri et al., 2000; Losonczy et al., 2002). Compared with the cells described above O-LM interneurons had slower accommodating firing properties, and more pronounced membrane sag upon hyperpolarization. Similar electrophysiological properties were observed in a group of morphologically diverse MGE-derived cells found throughout s.o., s.p., and s.r. that could not be easily matched to any previously reported hippocampal subtypes (eg. Figure 7F,G,H). One of these cells, depicted in Figure 7F, strongly resembles an s.r-s.l.m. border (R-LM) cell previously described as a perforant path-associated (PPA) interneuron with axon concentrated in s.l.m. that extends into the molecular layer of the dentate and subiculum (Vida et al., 1998, and see Hajos and Mody 1997). This categorization is uncertain as PPA interneurons are suggested to immunostain for CCK (Klausberger et al., 2005) which we never observed in GFP+ cells of Nkx2-1Cre:RCE mice. We note however that the CCK+ cells described as PPA interneurons appear distinct from those originally described having considerable axon concentration in s.r. as well as s.l.m. (compare figure 5A in Vida et al., 1998 with figures 3 and and44 in Klausberger et al., 2005). The axonal projection of the cell displayed in Figure 7G might lead to the mistaken classification as a “trilaminar” interneuron.(Sik et al., 1995). Indeed this s.o. interneuron did project to the subiculum, however, as already noted, trilaminar interneurons are M2R immunoreactive (Ferragutti et al., 2005; Hajos et al., 1998; Jinno et al., 2007) which we only observed in CGE-derived cells (see Figures 4 and and6)6) discounting this MGE-derived cell as a candidate trilaminar cell. Similarly, the cell depicted in Figure 7H is reminiscent of a Schaffer collateral associated cell (SCA), however SCA cells immunostain positive for CCK (Vida et al., 1998; Cope et al., 2002; Klausberger et al., 2005) which we only observed in CGE derived interneurons (see Figures 4 and and55).
Surprisingly, we frequently detected CCK transcript within MGE-derived interneurons in our scPCR analysis despite never observing CCK immunoreactivity in MGE derived interneurons. It is possible that the molecular profiles determined by scPCR analysis were confounded by inadvertent collection of contaminant tissue while advancing the recording electrode through slices to target GFP+ interneurons for recording. However, separate control recordings revealed that the potential for false positive signals from contamination by neighboring pyramidal or glial cells in our standard scPCR protocol was <10% (see Materials and Methods) which is far too low to account for the frequency of CCK transcript detection in MGE-derived interneurons (Figure 7K). Thus, while CCK protein detection with immunocytochemistry nicely segregates between MGE and CGE derived interneurons, CCK transcript detection with standard scPCR appears much less informative of interneuron identity. Indeed a similar disconnect between CCK protein and transcript detection has previously been observed when comparing immunocytochemical and microarray analyses of cortical PV+ fast spiking basket cells (Sugino et al., 2005). Other frequently detected transcripts from MGE derived cells included GAD65, GAD67, PV, SOM, NPY, nNOS, and Lhx6 (Figure 7K). Of these only Lhx6 was exclusively detected in MGE-derived interneurons occurring in >80% of the cells recorded in Nkx2-1Cre:RCE mice consistent with the strong expression of this transcription factor in postmitotic neurons generated from MGE precursors (Grigoriou et al., 1998; Alifragis et al., 2004; Liodis et al., 2007).
A total of 81 GFP+ interneurons were recorded in GAD65-GFP mice with representative examples illustrated in Figure 8A–K. In this CGE-derived cohort of cells we commonly observed another population of perisomatic targeting interneurons (eg. Figure 8A,B) with properties typical of CCK basket cells that exhibit higher input resistance, prominent sag upon hyperpolarization, and slow accommodating firing properties compared to the fast spiking MGE-derived basket cells described above (Vida et al., 1998; Cope et al., 2002; Klausberger et al., 2005; Daw et al., 2009, 2010; Lee et al., 2010; Cea-del Rio et al., 2010; Somogyi et al., 2004). Another easily discerned cell from the CGE was the mossy fiber associated (MFA) cell whose axon targets the proximal apical dendrites of CA3 pyramidal cells within stratum lucidum where the mossy fiber afferents of dentate granule cells reside (Figure 8C; Pelkey et al., 2005; Vida and Frotscher, 2000; Losonczy et al., 2004). As for CCK basket cells, MFA interneurons immunostain for CCK (Losonczy et al., 2004) consistent with the coexpression of GFP and CCK in GAD65-GFP and Mash1CreER:RCE mice (see Figures 4 and and6).6). An additional prominent cell group frequently encountered among CGE-derived cells resembled SCA interneurons (Figure 8D) and their CA3 counterparts (here termed associational commissural associated (ACA) interneurons, Figure 8E). SCA cells are also CCK immunoreactive and typically reside in s.r. with a major axonal projection that aligns with the CA3 pyramidal cell afferent outputs within s.r. of CA1 (Cope et al., 2002, Cea-del Rio et al., 2011; Vida et al., 1998; Lee et al., 2010). Within s.l.m we readily encountered GFP+ late spiking NGF cells in the GAD65-GFP mouse (Figure 8F,G) consistent with the partial CGE origin of hippocampal NGF interneurons (Tricoire et al., 2010; Funetealba et al., 2010). Apart from origin, the MGE and CGE derived NGF differed by a more variable expression of nNOS transcript and a complete lack of CCK transcript in CGE derived NGF cells compared to MGE-derived NGF cells all of which expressed both nNOS and CCK transcripts. Interestingly, neocortical NGF cells are known to be CGE-derived (Miyoshi et al., 2010) and also have a more variable nNOS transcript expression profile than we observed for MGE-derived hippocampal NGF cells (Karagiannis et al., 2009). Moreover, Price et al. (2005) observed a strong correlation between the presence of nNOS and CCK transcripts within a sample of hippocampal NGF cells as well as a population that lacked both transcripts similar to our MGE and CGE subgroups respectively. Amongst all the transcripts examined GAD65, GAD67, NPY, CCK, Npas1, Npas3, CoupTFII and 5HT3 were frequently observed in the CGE cohort of interneurons (Figure 8L). Of these 5HT3 most faithfully reported CGE-origin as it was never observed in MGE-derived interneurons and could be found in >75% of the CGE-derived cohort consistent with a recently reported 5HT3A-GFP mouse strain that selectively labels CGE-derived cortical interneurons (Vucurovic et al., 2010). Indeed although several of the CGE-derived interneurons we recorded were not readily identifiable with previously described hippocampal interneurons they did typically express 5HT3 transcript (eg. Figure 8H–K).
While inspection of individual recorded interneurons from basic anatomical, electrophysiological, and molecular perspectives yielded successful classification within previously established subgroups for some of our recorded cells this approach proved futile for many of the cells obtained within both the MGE and CGE datasets. Moreover, this tactic clearly revealed that cells which group together based on some similarities, such as common spiking behaviour, do not necessarily share other features, such as expression of a given transcript. In practical terms it would be useful to group together cells that share a large number of, but not necessarily all, features with minimal investigator bias. Thus, to independently probe the diversity of MGE- and CGE-derived hippocampal interneurons we attempted a polythetic classification scheme in which individual groups are not defined on the basis of any single character, but rather on a combination of characters, by subjecting our data set to unsupervised cluster analyses using the electrophysiological, molecular, and lineage parameters of the recorded cells (Cauli et al, 1997; Karagiannis et al, 2009). Practically, cell origin was set at 1 for GFP+ interneurons recorded from Nkx2-1Cre:RCE mice and -1 for cells from the GAD65-GFP line. To include all 142 recorded cells we did not consider anatomy in the clustering as axonal recovery was limited for some cells in the dataset. The hierarchical tree shown in figure 9A reveals the presence of six distinct clusters that emerged when our dataset was first subjected to a Ward's clustering algorithm (1 through 6 with n=19, 19, 36, 29, 18, and 21 cells in each cluster respectively). Interestingly, clusters 1 and 2 were comprised entirely of MGE-derived interneurons, while clusters 4, 5, and 6 contained only CGE-derived cells indicating strong segregation of interneuron subtype according to cell origin despite equal weighting of this parameter in the cluster analysis. To validate the Ward's classification we additionally performed a K-means cluster analysis of the dataset (MacQueen, 1967; Hartigan and Wong, 1979; Karagiannis et al., 2009). Unlike the Ward's method where aggregate members of a given cluster cannot exit the cluster after being linked to their nearest neighbors, suboptimal attribution of a given cell to a particular cluster is dynamically corrected with each iteration of the the K-means clustering algorithm such that a given cell may visit multiple clusters over successive iterations until an optimal affiliation is reached (see methods). While this self correcting aspect of the K-means algorithm potentially yields better discrimination between interneuron classes it does require that the number of clusters be predetermined. Based on the clusters evidenced by the Ward's method, we first ran the K-means algorithm with K=6 clusters. In comparing the clusters obtained with the two methods when K=6 we found that 90% of the cells (128/142) grouped together by Ward's clustering were similarly grouped when passed through the K-means algorithm (Figure 9B) resulting in a similarity index of 0.8 as assessed by the adjusted Rand index (Rand, 1971 and see methods). Importantly this similarity index represented a peak as the same measure decreased when the dataset was evaluated with K=5 or 7 clusters, validating the choice of 6 clusters. As for the Ward's clustering groups, 2 of the K-means determined clusters contained only MGE-derived interneurons while 3 of the clusters comprised only CGE-derived interneurons further establishing the discriminative power of embryonic origin in parsing interneuron subtypes.
The distribution of mRNAs detected in and electrophysiological features of the 6 clusters determined by K-means clustering are illustrated in Figure 9C and Table 3 respectively. Members of cluster 1 arise from the MGE, frequently express PV and SOM transcripts, and have the fastest membrane time constants and firing frequencies compared to cells of other clusters. These features are consistent with the MGE-derived putative PV+ fast spiking basket, bistratified and axo-axonic cells shown in Figures 7A–D, all of which were grouped within cluster 1. SOM transcript was detected in all cluster 2 cells which exhibited slower membrane time constants and firing properties compared to all other MGE-derived cells in clusters 1 and 3. Cluster 2 cells, exemplified by the cells depicted in Figures 7E–H, also responded with more prominent membrane sag upon membrane hyperpolarization compared to other MGE-derived cells. Cluster 3 was the only group to contain cells from both MGE and CGE. Regardless of origin the cells were characterized by NPY transcript expression, and a late spiking phenotype with just above threshold current injections as illustrated by the NGF and Ivy cells depicted in Figures 7I–K and 8F–G all of which were grouped in cluster 3. The inability of cluster analyses to dissociate MGE- and CGE-derived NGF and Ivy cells despite providing cell lineage as a parameter is consistent with the strongly overlapping properties of these cells (Tricoire et al., 2010). CGE-derived cells of clusters 4–6 all exhibited significantly greater spike adaptation than cells of clusters 1–3. Cells of clusters 4 and 5 had similar electophysiological signatures with high input resistance and slow membrane time constants, with cells of cluster 5 more likely to express CR transcript. Anatomical inspection of cells grouped into cluster 4 (eg. Figure 8D,E) typically revealed dendrite targeting cells that resemble SCA/ACA interneurons. In contrast cluster 5 cells often had little local axon (eg. Figure 8K) suggestive of GABAergic projection cells or interneuron targeting interneurons (Ascady et al., 1996; Gulyas et al., 1996). Cluster 6 cells had the highest incidence of CCK transcript typically in combination with VGluT3 and VIP, suggestive of putative CCK basket cells (Somogyi et al., 2004; Klausberger et al., 2005). Indeed anatomically, cells belonging to cluster 6 (eg. Figures 8A,B) typically exhibited perisomatic targeting local axon as expected for CCK+ basket cells. In addition MFA cells (eg. Figure 8C) were grouped into cluster 6.
To further evaluate the significance of the groupings obtained from the unsupervised cluster analyses and examine which measured parameters most heavily influenced cluster identification we examined the effect of randomization on the cluster analysis. Importantly randomization was performed in such a way that disrupted the structural correlations between the different measurements describing a given cell without changing the overall mean or standard deviation of each parameter across the dataset. For example the expression of a given transcript was randomized across all cells without changing the total number of cells that the transcript was detected in. To quantify the effect of randomization we used silhouette analysis to describe the quality of the clusters obtained with the K-means algorithm (K=6) before and after randomization (see Methods and Rousseeuw, 1987; Karragianis et al., 2009). For each cell the silhouette value (S, between −1 and +1) gives a quantitative estimation of how well a cell is sorted in its cluster with higher S indicating stronger affiliation with its cluster, negative S suggesting a potential misclassification, and S around 0 indicating the cell lies equally far away from more than one cluster. For the original non-randomized dataset almost all cells exhibit positive S values within their assigned cluster (Figure 9D) leading to a global S value (S') describing the entire dataset of 0.14. The significant reduction in S' to 0.046 following complete randomization of all parameters reflects a 67% decrease in the overall quality of the clusterings. To determine which of the measurements we obtained for each recording exhibited the greatest influence on the quality of the clusterings obtained using the K-means algorithm we investigated the effect of randomizing each parameter in isolation and in specific combinations on S' (Figure 10A–D). Amongst the commonly used interneuron marker transcripts examined SOM and nNOS randomization yielded the largest decrease in S' while for electrophysiological properties randomizing time constants and spike kinetic/frequency parameters produced the strongest decrease in S' (Figure 10A,B). As expected given the lack of overlap between CGE and MGE cells within 5 of the 6 clusters, cell origin also had high discriminative power, particularly when randomized in combination with the transcription factors and 5-HT3 that essentially serve as proxies for cell origin (Figure 10A,C). In contrast, parameters that were very common to all cells (eg. GAD65/67 expression) or rarely detected at all (eg. Enk) had little effect on S' revealing poor discriminative power. In total the silhouette analyses of clustering quality further validate the original clusters obtained from our dataset as the findings make it unlikely that the clusters resulted from accidental random correlations between the various parameters measured to describe MGE and CGE-derived interneurons.
Upon comparing the parsing of interneuron populations based on immunostaining (Figures 3–6) versus transcript detection (Figure 9C) for frequently used interneuron markers, obvious incongruities emerge. For example while CCK immunostaining was restricted to CGE-derived interneurons CCK mRNA transcript was readily observed in both MGE- and CGE-derived cells. Such incongruity is reflected in the poor discriminative power of transcripts for markers that have proved highly specific for immunolabeling nonoverlapping interneuron populations like PV and CCK (Figure 10A). While we cannot entirely rule out the possibility that scPCR molecular profiles are confounded by signals from contaminant tissue, our control recordings (see Materials and Methods) indicate that such false positive signals are negligible in our data set. Alternatively, the discrepant molecular profiles could reflect differences in the threshold limits of immunodetection versus transcript detection by scPCR. Indeed it has been shown that the detection threshold of our scPCR approach is approximately 25 mRNA molecules that is in the range of low abundance mRNA species perhaps leading to very low levels of protein that fall below immunodetection limits (Tsuzuki et al., 2001). If this is true then MGE- and CGE-derived basket cells should contain different amounts of PV and CCK transcripts. Therefore, in a final series of experiments we performed quantitative scPCR to compare the PV and CCK mRNA content of electrophysiologically and morphologically indentified MGE-derived PV+ fast spiking basket cells (FS-BCs) with CGE-derived CCK+ basket (non-FS-BCs) and SCA cells (Figure 11A–E). As controls we also examined CA1 pyramidal cells and additionally probed GAD65 and VGluT1 mRNA levels. In total four FS-BCs recorded in Nkx2-1Cre:RCE mice were compared with 5 SCA/non-FS-BCs recorded in GAD65-GFP mice as well as 5 pyramidal cells from both lines. For all steps, cells were processed in pairs (FS-BC versus SCA/non-FS-BC, pyramidal versus FS-BC, and pyramidal versus non-FS-BC) and normalization was performed using actin mRNA in each harvest. As expected CCK transcript was detected in all CGE derived non-FS-BC and SCA cells while PV transcript was successfully amplified in all four tested MGE-derived FS-BCs. As in our previous dataset we regularly detected CCK in FS-BCs (4/4 cells), however, the level of CCK mRNA was markedly less than that detected in both non-FS-BCs run in parallel (Figure 11A–E). Similarly, the level of CCK transcript detected in FS-BCs was typically less than that found in pyramidal cells which exhibited levels comparable to CGE-derived non-FS-BCs/SCAs (Figure 11A and D middle and right panels) consistent with CCK immunodetection in many pyramids (Morino et al., 1994). PV transcript was found in only 1/5 non-FS-BC/SCAs and 1/5 pyramidal cells at levels considerably lower than that detected in FS-BCs relative to actin (Figure 11D right panel). Importantly, GAD65 mRNA was not detected in any of the pyramids but was found in all interneurons at roughly the same levels confirming that the differences observed for other transcripts (CCK and PV) are specific to those mRNA species (Figure 11D,E). In contrast VGluT1 mRNA was found in all pyramids and most interneurons (8/9, see also Gallopin et al, 2006; Andjelic et al, 2009), but with an obvious enrichment in the pyramidal cell cohort further confirming the validity of our quantitative approach. Whether this intereuron VGluT1 signal revealed with the more sensitive PCR assay reflects low level contamination or true expression remains debatable. However, the enrichement of VGluT1 relative to CCK in pyramidal cell harvests further argues against pyramidal cell contamination as the source of CCK message in FS-BC harvests as CCK was enriched relative to VGluT1 in FS-BC harvests (Figure 11D,E). Together these results suggest that discrepancies between the molecular signatures generated by scPCR versus immunostaining for characterizing interneuron diversity reflect differences in the limits of threshold detection of the two methods. Thus, caution is warranted when directly comparing and attempting to extrapolate interneuron identities based on the two strategies, particularly for standard “digital” scPCR.
Coordination of glutamatergic principal cell activity within cortical networks relies upon a remarkable diversity of inhibitory interneurons that is largely predetermined during embryogenesis by genetic restriction of neuronal potential. Here we investigated the influence of embryonic origin on hippocampal interneuron diversity using longitudinal developmental analyses of the precursors of different hippocampal interneuron subpopulations. In general, based on immunohistochemistry the MGE gives rise to PV+, SOM+ and nNOS+ hippocampal interneurons. Indeed our combined anatomical, electrophysiological, and molecular characterization indicated 3 major subpopulations of MGE-derived interneurons generally conforming to these subgroups: 1) PV+ fast-spiking interneurons including basket, bistratified, and axo-axonic cells; 2) a morphologically diverse population of adapting SOM+ interneurons typified by O-LM cells; and 3) an nNOS+/NPY+ subset of late-spiking interneurons comprised of NGF/Ivy cells. In contrast CGE-derived hippocampal interneurons were immunohistochemically defined by CCK, CR, VIP and reelin expression. Interestingly, single cell profiling also indicated 4 major cohorts of CGE-derived interneurons: 1) CCK+ perisomatic targeting basket cells; 2) CCK+ dendrite targeting interneurons typified by SCA cells; 3) a late-spiking NGF cell cohort virtually indistinguishable from those with MGE lineage; and 4) an anatomically heterogenous group of high input resistance cells with limited local recovered axon likely representing M2R+, VIP+, or CR+ projection cells or interneuron-targeting-interneurons. These findings, summarized in Figure 12, reveal the developmental origins of most hippocampal interneuron subtypes.
The validity of our findings entirely depends upon the fidelity of the mouse lines used for accurately reporting MGE- and CGE-derived interneurons. Specificity of the Nkx2-1Cre:RCE line for MGE is consistent with ubiquitous expression of the Nkx2-1 homeobox gene throughout the MGE with exclusion from LGE/CGE progenitors (Kimura et al., 1996; Sussel et al., 1999; Butt et al, 2005). Indeed Nkx2-1 may be a molecular switch favoring MGE over CGE fate (Butt et al., 2008) and driver lines based on Nkx2-1, or its downstream effector Lhx6, have been employed to characterize MGE-derived neocortical interneurons (Xu et al., 2008; Fogarty et al., 2007). Moreover, loss of function studies indicate a requirement for Nkx2-1 and Lhx6 for proper specification and migration of MGE-derived interneurons (Liodis et al., 2007; Pleasure et al., 2000; Du et al., 2008; Zhao et al., 2008; Tricoire et al., 2010; Butt et al., 2008). Nkx2-1 is also expressed in preoptic area (POA) progenitors which contribute a minor population of cortical interneurons (Gelman et al., 2009). However, POA-derived interneurons fail to immunolabel for PV, SOM, or nNOS (Gelman et al., 2009) indicating that any potential POA contribution to Nkx2-1Cre:RCE GFP+ cells is minor as >90% of the GFP+ cells in this line are accounted for by these markers. Moreover, the regular detection of Lhx6 in Nkx2-1Cre:RCE GFP+ cells is inconsistent with a significant contribution of POA-derived interneurons to this population (Gelman et al., 2009). Specificity of the Olig2CreER:ZEG line for labelling MGE-derived cells has been extensively characterized elsewhere (Miyoshi et al., 2007) and is further confirmed by the shared immunolabeling profiles of GFP+ cells in Olig2CreER:ZEG and Nkx2-1Cre:RCE mice. Thus, we are confident that Nkx2-1Cre:RCE and Olig2CreER:ZEG lines accurately report MGE-derived hippocampal interneurons. However, this is not meant to imply that the lines faithfully report the entire MGE-derived interneuron population. Estimates indicate that Nkx2-1 driver lines fail to report approximately 20% of MGE-derived neocortical interneurons (Xu et al., 2008; Fogarty et al., 2007) primarily arising from progenitors in the most dorsal MGE which typically yield SOM+ interneurons (Fogarty et al., 2007; Sousa et al., 2009; Wonders et al., 2008). Indeed amongst the major populations arising from MGE the SOM+ population exhibited the least GFP coexpression in the Nkx2-1Cre:RCE line. Whether this unreported population of cells represents distinct interneuron subtypes requires further investigation.
Despite ubiquitous expression of GAD65 among interneurons the GAD65-GFP line reportedly preferentially labels CGE-derived neocortical interneurons (Lopez-Bendito et al, 2004). Our hippocampal immunohistochemical results in GAD65-GFP mice completely paralleled those obtained in Mash1CreER:RCE mice which have been characterized as CGE specific reporters (Miyoshi et al., 2010). Moreover, results obtained in both lines yielded entirely complementary findings to those from our MGE reporters. Indeed the common MGE-derived markers PV, SOM, and nNOS were essentially absent from GFP+ cells probed by immunocytochemistry in GAD65-GFP and Mash1CreER:RCE mice. Moreover, we never detected mRNA for the MGE marker Lhx6 in GFP+ cells of GAD65-GFP mice in contrast with ubiquitous expression of this transcript in Nkx2-1Cre:RCE GFP+ cells. Instead the majority of GFP+ cells in GAD65-GFP mice contained 5HT3 mRNA providing further evidence of CGE specificity as 5HT3 is an early and protracted marker of CGE-derived interneurons (Vucorovic et al., 2010; Lee et al., 2010). Indeed 5HT3-GFP reporter mice exhibit colocalization of GFP with CCK, CR, VIP, and reelin in neocortical interneurons (Vucorovic et al., 2010; Lee et al., 2010) paralleling our findings in hippocampal neurons of GAD65-GFP mice (see also Lopez-Bendito et al., 2004; Wierenga et al., 2010). Considered together the findings support use of GAD65-GFP and Mash1CreER:RCE lines to investigate the contribution of CGE progenitors to hippocampal interneuron diversity. However, once again this assertion of specificity is not meant to imply that these transgenic lines capture the entire CGE-derived interneuron population. If hippocampal CGE-derived interneurons account for ~40% of the entire hippocampal interneuron population, as suggested for neocortex (Myoshi et al., 2010, Lee etal., 2010), calculations from our density counts reveal that GAD65-GFP mice report only 50% of all CGE-derived interneurons. Assuming the Nkx2-1Cre:RCE line captures 80% of all MGE-derived interneurons as outlined above and that overall the MGE contributes 60% of cortical interneurons, our cell density counts at P30 indicate that total interneuron density should be ~7500cells/mm3, comparable to average interneuron density across rat neocortex (Beaulieu, 1993). Thus, the 1500cells/mm3 observed at P30 in the GAD65-GFP mouse represents only half the expected CGE contribution of 3000cells/mm3. This is consistent with our finding that GFP+ cells in GAD65-GFP mice only accounted for ~50% of VIP+ and CCK+ interneurons which appear to be entirely CGE-derived. Whether the GAD65-GFP line reports a fully representative cross section of CGE-derived interneuron diversity or instead labels specific subpopulations of CGE-derived interneurons remains unknown.
In addition to 5HT3, mRNA for Npas1 and CoupTFII strongly segregated to CGE-derived interneurons. For CoupTFII this preference for CGE over MGE-derived cells was also detected immunoctyochemically and is consistent with CoupTFII enrichment within the CGE (Kanatani et al., 2008). However, CoupTFII was not as exclusive as 5HT3 being detected both at message and protein levels in a small population of Nkx2-1Cre:RCE GFP+ cells. This CoupTFII expression in a minor MGE-derived interneuron cohort likely reflects interneurons generated from more dorsal MGE as these progenitors also express CoupTFII (Kanatani et al., 2008). Indeed among MGE-derived cells CoupTFII transcript most frequently occurred in cells defined by SOM mRNA expression (cluster 2 Figure 9) likely representing SOM+ interneurons arising from more dorsal MGE progenitors (Fogarty et al., 2007; Flames et al., 2007; Wonders et al., 2008; Sousa et al., 2009). Npas1 is expressed in migrating interneuron precursors (Cobos et al, 2006) and in mature hippocampal interneurons, partially colocalizing with CR and reelin (Erbiel-Sieler et al, 2004). These observation are consistent with our frequent detection of Npas1 mRNA in CGE-derived interneurons, and the fact that hippocampal Npas1 expression is unaffected in Lhx6 mutants (Zhao et al, 2008).
Overall our findings reveal that hippocampal and neocortical interneurons have mostly similar origins for homologous interneuron subtypes. In both cases MGE progenitors give rise to interneurons characterized by PV and SOM expression while CGE-derived interneurons are defined by CCK, VIP, CR, and reelin expression. Our extensive profiling allowed us to characterize the origins of hippocampal interneuron subtypes not well described in the neocortical literature such as various CCK+ interneurons. CCK immunopositive hippocampal cells only arose from the CGE consistent with recent descriptions of CGE-derived CCK interneurons (Morozov et al, 2009; Wierenga et al., 2010; Miyoshi et al., 2010; Lee et al., 2010; Lopez-Bendito et al., 2004). Indeed non-fast spiking basket, MFA, and, SCA subtypes of interneurons, all known CCK+ cells, were found amongst GFP+ cells of our CGE reporter. Similarly, we determined the origins of several other hippocampal interneurons for which no obvious neocortical homolog is yet reported such as M2R-immunoreactive projection cells (eg. trilaminar, CGE), and bistratified interneurons (MGE). The most obvious difference between neocortical and hippocampal interneurons encountered relates to the origins of late-spiking NGF cells. Consistent with our prior investigation (Tricoire et al., 2010) we observed hippocampal NGF cells to arise from both MGE and CGE progenitors while their neocortical counterpartsdisplay only CGE lineage (Miyoshi et al., 2010; Lee et al., 2010). Previously, we found that MGE- and CGE-derived NGF cells parse by nNOS immunoreactivity which identifies only MGE-derived NGFs (Tricoire et al., 2010). Here we demonstrate that MGE-derived NGF cells express Lhx6 but not 5HT3 mRNA whereas CGE-derived NGFs exhibit the opposite molecular profile further confirming dual origins for these cells.. Aside from origin the properties of these two NGF cell cohorts are so similar that our clustering algorithms failed to dissociate them even when factoring their distinct origins. Indeed this was the only population for which the cluster analyses failed to segregate MGE- and CGE-derived interneurons. It will be interesting to determine if MGE- and CGE-derived NGF cells subserve distinct network functions within the hippocampus.
In conclusion, by comparing and contrasting MGE- and CGE-derived interneurons in parallel we have established a blueprint of the developmental origins of diverse hippocampal interneurons similar to that which has emerged for the neocortex (reviewed in Wonders and Anderson, 2006; Batista-Brito and Fishell, 2009; Gelman and Marin, 2010). Importantly, we determined the lineage of several classes of interneurons uniquely described in the hippocampus. Our findings fill an important gap in the understanding of hippocampal interneuron diversity and are directly relevant to several neurological disorders precipitated by disruptions of excitation-inhibition balance due to loss of specific interneuron cohorts. Indeed a thorough appreciation of the spatiotemporal origins of distinct interneuron subtypes is critical to advance stem cell based therapies for disorders associated with dysfunction of specific interneuron subsets (Baraban et al., 2009, Maroof et al, 2010, Waldau et al., 2010).
Work supported by a National Institute of Child Health and Human Development (NICHD) intramural award to C.J.M. L.T. is an NIH Visiting Fellow and a NIH/CNRS ECTA Fellow. Daniel Abebe provided expert technical assistance. Microscopy imaging was performed at the Microscopy and Imaging Core (NICHD, NIH) with the assistance of Dr. Vincent Schram and at the imaging facilty of the IFR83 (Paris France) with the assistance of Susanne Bolt and Richard Schwartzmann. We thank Dr. Gordin V Ohning for supplying us with the CCK antibody, Dr. Gabor Szabo for providing the GAD65-GFP mouse line, Dr. Stuart Anderson for providing the Nkx2-1Cre driver line, and Dr. Gord Fishell for providing the Olig2CreER and Mash1CreER driver lines as well as the RCE reporter line.