|Home | About | Journals | Submit | Contact Us | Français|
Connectivity in the developing hippocampus displays a functional organization particularly effective in supporting network synchronization, as it includes superconnected hub neurons. We have previously shown that hub network function is supported by a subpopulation of GABA neurons. However it is unclear whether hub cells are only transiently present or later develop into distinctive subclasses of interneurons. These questions are difficult to assess given the heterogeneity of the GABA neurons and the poor early expression of markers. To circumvent this conundrum we used “genetic fate mapping” that allows for the selective labelling of GABA neurons based on their place and time of origin. We show that early generated GABA cells form a subpopulation of hub neurons, characterized by an exceptionally widespread axonal arborisation and the ability to single-handedly impact network dynamics when stimulated. Pioneer hub neurons remain into adulthood where they acquire the classical markers of long-range projecting GABA neurons.
One approach to unravelling the complexity of neuronal circuits is to understand how their connectivity emerges during brain maturation. Neuronal connectivity is very often reflected in the activity dynamics that a given network of neurons can produce. Interestingly, most developing neuronal networks spontaneously produce a variety of correlated activity dynamics that are thought to be essential for proper circuit maturation (Ben Ari, 2001; Blankenship and Feller, 2010). At early postnatal stages, the hippocampus displays spontaneous, synapse-driven network synchronizations in the form of Giant Depolarizing Potentials (GDPs) (Ben Ari et al., 1989; Garaschuk et al., 1998). We have recently shown that, during this developmental period, the CA3 region displayed a “scale-free” functional topology (Bonifazi et al., 2009) characterized by the presence of rare, superconnected hub neurons. Hub neurons are characterized by the following features (Bonifazi et al., 2009): (1) they possess high output functional connectivity; (2) their stimulation significantly affects network dynamics (whereas stimulating other neurons does not); (3) they are GABA neurons with a widespread axonal arborisation crossing subfield boundaries; (4) they receive more Excitatory PostSynaptic Potentials and have a lower threshold for action potential generation than other interneurons. Thus, the study (Bonifazi et al., 2009) confirmed the leading role of GABA neurons in shaping network oscillations (Ellender et al., 2010; Klausberger and Somogyi, 2008), that emerges as soon as the first functional synapses start to develop.
However, it is at present unknown whether hub neurons are only present transiently during development or if they persist into adulthood. If the latter, it is also unclear whether this population represents one or many morpho-physiological subtypes of interneurons. Determining the subtypes of hub neurons is experimentally challenging for several reasons. First, hub neurons are a sparse cell population (Bonifazi et al., 2009). Second, cortical GABA neurons are characterized by a bewildering heterogeneity that results in several classification challenges. This complexity is further confounded during brain development by the fact that most GABA neurons have not yet developed the characteristics that enable investigators to identify and classify them in adulthood (Hennou et al., 2002). To address the above issues, it is essential to permanently label hub neurons in a manner such that they can be examined both for their dynamics during GDPs, as well as in the adult.
Based on the following arguments, we hypothesized that hub neurons could be Early Generated interneurons (EGins) that pioneer hippocampal circuits: first, hub neurons were characterized by their advanced morpho-physiological features compared to other developing GABA neurons (Bonifazi et al., 2009). Second, theoretical models predict that scale-free networks grow according to “preferential attachment rules”, meaning that early connected neurons would turn into hub cells (Barabasi and Albert, 1999). In rodents, cortical GABA neurons are generated in the subpallium, mainly from two transient structures, the medial and caudal ganglionic eminences (Anderson et al., 1997; Batista-Brito and Fishell, 2009; Marin and Rubenstein, 2001). Peak neurogenesis of hippocampal GABA interneurons in the mouse occurs between E12 and E15 (Danglot et al., 2006). However, some GABA neurons are postmitotic and start migrating as early as embryonic day 10 (Danglot et al., 2006; Miyoshi et al., 2007). From the above, we inferred that hub neurons may be generated during the earliest phases of neurogenesis. Interestingly, recent advances in molecular genetics have produced novel tools to control the permanent expression of GFP in the interneurons of transgenic mice in accordance with their spatio-temporal embryonic origin (Miyoshi and Fishell, 2006).
To test this hypothesis, we have performed a compound analysis of EGins, using a combination of genetic fate mapping (Miyoshi and Fishell, 2006) and immunohistochemistry coupled with imaging of network dynamics and single-cell electrophysiological recordings. We find that at early postnatal stages, EGins turn into a distinct functional subclass of hub neurons (Bonifazi et al., 2009). Furthermore, we show that EGins persist in adult hippocampal networks and express markers identifying them as putative long-range projecting GABA neurons (Jinno, 2009). This indicates that these cells may retain, at least anatomically, the capacity to coordinate the timing of neuronal activity across structures. Moreover, this finding provides the means to study the involvement of hub cells in other synchronization processes such as epilepsy (Morgan and Soltesz, 2008), independently from calcium data analysis.
Despite their varied sites of origin, most, if not all, hippocampal GABA interneurons require the expression of Dlx1 and/or Dlx2 for their generation, as evidenced by the near absence of GABA interneurons in Dlx1/Dlx2 null compound mutants (Anderson et al., 1997; Bulfone et al., 1998; Long et al., 2009). Thus, in order to label as many EGins as possible we have fate mapped hippocampal interneuron precursors expressing Dlx1/2, by transiently activating a Dlx1/2CreERTM driver line (Batista-Brito et al., 2008) crossed with a Cre-dependent EGFP reporter line RCE:LoxP (Sousa et al., 2009). Recombination of the reporter allele is achieved within 24 hrs upon administration of tamoxifen, therefore providing temporal precision in the labelling of cells expressing Dlx1/2 (see Experimental Procedures). Temporal control also requires Dlx1/2 expression to be confined to post-mitotic cells, as any labelling of progenitors would overtime produce labelled cells at later ages. This condition is satisfied by using the driver Dlx1/2CreERTM since in this transgenic line Dlx1/2 is only expressed shortly after interneurons become post-mitotic (Batista-Brito et al., 2008). In order to further confirm the temporal resolution of our fate mapping approach at such unusually early force-feeding time period: (1) we performed a short term fate mapping of Dlx1/2 progenitors at E12.5 (induction at E7.5 or E9.5) and observed that GFP positive cells could be detected along the lateral border of the ganglionic eminences, but excluded from the progenitor cell region lying in the embryonic ventricular zone (Fig. S1). GFP positive cells presented relatively developed processes (Fig. S1C) and could even be found heading towards the hippocampal neuroepithelium, indicating an already advanced stage of migration (Fig. S1C); (2) we performed BrdU injections within a time window of 20 hrs following tamoxifen force-feeding (at E9.5) and found significant GFP/BrdU co-labelling in E12.5 embryos (28%, n= 1067 cells, Fig. S1A & B) whereas a BrdU injection performed later than E10.5 did not result in any labelling of GFP cells (not shown). This indicates that the EGins labelled here are likely to be born before E10.5.
We next analyzed the population of EGins using immunohistochemical approaches at two separate time points: early postnatal (P7) and adult stages (>P30). We first examined the distribution of GFP labelled cells within the adult hippocampus. Although a precise quantification of the density of EGins is impossible due to the variability in our labelling method, it is striking how few cells are labelled at these stages. Indeed, on average 4.8 ± 1.6 GFP neurons are present per hippocampal section in adult mice (n=263 sections, 7 mice, Fig. 1A, B). On average, a similar number of GFP cells could be found at P7 (4.4 ± 1.2; n=147 sections, 7 mice; Figs. 1C&2A), indicating that this subpopulation of cells is unlikely to experience massive developmental cell loss. The GFP reporter line used here provides the considerable benefit that after enhancement by immunohistochemistry, it produces a strong enough signal for the fine neuronal processes including axons to be examined. Although EGins represented only a few cells per hippocampal section, GFP axonal labelling displayed a remarkable web-like coverage of the entire hippocampal region in both age groups (Figs. 1A–B&2A). Low-density but extensive axonal arbors were distributed in all hippocampal layers. However, no prominent axonal labelling could be found in the pyramidal cell layers (Fig. 1B&2A), indicating that EGins were not principally targeting perisomatic regions. Axonal labelling was also frequently observed in the fimbria (Fig. 2C). This is in marked contrast with the spatially patterned and confined labelling obtained when interneurons are fate mapped at later developmental timepoints (Fig. 2B) or with a preferred perisomatic innervation like PV-expressing cells (Miyoshi et al., 2007). EGins could be multipolar or bipolar with horizontally or vertically oriented dendrites that, to a varying extent, were sparsely spiny (Figs. 1, ,2&2& 3). Their somata were evenly found in all hippocampal areas (Fig. 1 and and2),2), with a slightly higher proportion of them being located in the hilar region of the dentate gyrus (Fig. 1A&C). We also examined within each region, the laminar distribution of EGins. These were found in all layers but with a preference for the CA1 stratum oriens, CA3 stratum radiatum and hilar region of the Dentate Gyrus (Fig. 1C). Similar layer distributions were found in P7 and P30 hippocampal sections (Fig. 1C, p<0.05 Kolmogorov-Smirnov test). Interestingly, both the regional distribution and axonal pattern of EGins were reminiscent of those reported for interneurons with an extrahippocampal projection (Jinno et al., 2007) and Fig. 1). This subclass of interneurons was studied in great detail using immunohistochemical approaches (Gulyas et al., 2003; Jinno, 2009; Takacs et al., 2008) and was proposed to serve a hub function through an axon targeting distant regions (Buzsaki et al., 2004; Sik et al., 1994; Sik et al., 1995).
We next immunostained hippocampal sections containing EGins with a variety of classic interneuron markers. Given the late maturation of interneurons’ neurochemical content, only sections from adult mice were included here. Although parvalbumin (PV), calbindin (CB), vasoactive intestinal peptide (VIP), calretinin (CR), or nitric oxide synthase (NOS) are prominently expressed by most hippocampal interneuron classes, almost none of the EGins were positive for these markers (Fig. 3B&G–K). In contrast, a significant fraction of them were immunopositive for somatostatin (SOM) (45 ± 6 %, n = 9 animals, Fig. 3A&L). SOM-expressing hippocampal interneurons constitute a heterogenous population that includes O-LM and HIPP cells, hippocampo-subicular and hippocampo-septal projection neurons (Jinno et al., 2007) as well as bistratified interneurons. In addition to SOM, O-LM cells also express PV (Ferraguti et al., 2004) and receive strong VIP positive inputs (Acsady et al., 1996). None of the EGins was positive for both SOM and PV (Fig. 3C–E, and Fig. S2B). Moreover, EGins did not receive strong VIP positive inputs (Fig. S2A). Therefore, we can exclude that a large number of EGins become O-LM cells. Given this last result and the fact that the distribution and axonal arborisation pattern of EGins resembled that of long-range projecting neurons, we next tested for the expression of mGluR1α and M2 receptor, both being additional characteristic markers of interneurons with extra hippocampal projections (Jinno et al., 2007). We found that a large majority of EGins was positive for mGluR1α (72.2 ± 7.7%, n = 5 mice, Fig. 3A and L) and that a significant fraction of them expressed the M2 receptor (18.4 ± 2.5 %, n=4 mice, Fig. 3F&L). In addition we tested for the co-expression of SOM and mGluR1α and found that 53.4 ± 7.5% (n=4 mice) of EGins co-expressed both markers, further indicating a long-range projecting phenotype. Since neurochemical marker expression is developmentally regulated, systematic testing and quantification of their presence within EGins was difficult to assess at P7. Nevertheless, SOM, mGluR1α and M2 receptor immunoreactivities were found in EGins at early postnatal stages (Fig. 2D–F). In order to further exclude that EGins develop into basket-like or O-LM interneurons, we have patch-clamped and filled with Neurobiotin EGins focusing on the CA3 region of slices prepared from adult mice (P25, n=65 neurons). Out of 65 filled cells 38 were sufficiently recovered and 12 reconstructed. None of these cells showed any axonal or dendritic characteristics of O-LM or perisomatic interneurons (Fig. S3). In contrast, despite some heterogeneity, they shared a remarkable common property with previously filled GABA projection neurons in the CA1 region of hippocampal slices as their axon tended to run out of the slice before giving off any terminal arbor (Gulyas et al., 2003). In addition, their evoked firing pattern could be classified either as irregular/stuttering (34%, n=26, not shown) or burst adapting (46%, n=26, Fig. S3). None of them was fast spiking. The latter further excludes that EGins develop into basket-like interneurons.
To conclude, morphological analysis of EGins indicated that these cells provide wide axonal coverage to the hippocampus at early postnatal stages (P7) and that a majority of them acquire morphological characteristics of GABA projection neurons in adulthood. Since long-range and widespread axonal arborization constitute a characteristic feature of previously described hub neurons (Bonifazi et al., 2009), we next tested whether these cells developed into functional hubs at early postnatal stages (P5–7).
We first compared the morpho-physiological features of EGins to those previously observed in functional hub neurons (Bonifazi et al., 2009). As in our previous study (Bonifazi et al., 2009), we decided to focus on the CA3c hippocampal region as it is a preferred initiation site for GDPs (Menendez de la Prida et al., 1998). Previously described functional hubs could be distinguished by four times longer axonal lengths than low connectivity interneurons, a lower threshold for action potential generation and more frequent spontaneous Excitatory Post-Synaptic Potentials (sEPSPs) (Bonifazi et al., 2009). While being recorded at P5–7, genetically-labeled GFP positive cells from tamoxifen-treated Dlx1/2CreERTM;RCE:LoxP mice were filled with Neurobiotin (n=56 cells). Different types of morphologies could be recovered (Fig. 4A). Twenty cells were reconstructed for morphometric analysis (Figs. 4&5). Out of all the parameters included in the analysis (see Experimental Procedures and Figs. 4&5), EGins were most identifiable by their axonal length (average: 8241 ± 1947 μm, n=20 cells). A measure of their extended axonal coverage is the distribution of the number of intersections their axon makes with concentric circles of increasing radius centered at the soma (Sholl analysis, see Experimental Procedures, Fig. 5B). The pooled distribution of the number of intersections as a function of distance from the soma, for all reconstructed EGins, is long-tailed since it is best fitted by a log-normal function (median: 250 ±4 μm, R2:0.99, Fig. 5B), whereas the same plot for interneurons with a different embryonic origin is best fitted by a sum of two Gaussian distributions (see below and Fig. 5B). Pooling data from previously sampled hub neurons (Bonifazi et al., 2009) presented a similar log-normal distribution of axonal intersections (median: 227 ± 5 μm, R2:0.98, Fig. 5B). In fact, none of the morphometric measurements obtained for EGins was significantly different from previously sampled functional hub neurons. Indeed, we performed a multivariate cluster analysis (see Experimental Procedures) of the morphometric data of 20 EGins, 10 hubs (HC for High Connectivity cf. (Bonifazi et al., 2009)), 10 Low Connectivity neurons (LC cf. (Bonifazi et al., 2009)) and 11 GABA neurons with a protracted origin (Late Generated interneurons, LGins, see below) and found that EGins and hub neurons significantly clustered into the same group (Fig. 5C). Moreover, like functional hubs, the axonal coverage of EGins often crossed subfield boundaries. Axonal branches from 20% neurons could be seen running in the fimbria (Fig. 4A), possibly indicating an extrahippocampal projection at early postnatal stages. Within the hippocampus, axons arborized uniformly in all hippocampal layers with the exception of stratum pyramidale, which showed little axonal innervation (Fig. 4A). This is in agreement with the immunolabelling results that rules out the possibility that this population is predominantly comprised by PV-containing perisomatic cells. Regarding the basic electrophysiological features analyzed here (see Fig. 8), we found that EGins received a high frequency of sEPSPs and had a low threshold for action potential generation. These properties were significantly different from those recorded in LGins (p<0.05) but not from those measured in functional hub cells (Fig. 8, Student’s t test, p>0.1). Taken together, these results indicate that EGins display morpholo-physiological characteristics that are exceptional similar to previously described functional hub neurons.
Since this study focuses on the embryonic origin of neurons, we decided to compare EGins with LGins labelled with a similar inducible genetic approach. Due to their relatively late birthdates (Miyoshi and Fishell, 2010), we decided to focus on CGE-derived interneurons by combining the Mash1BAC-CreER driver with the RCE:loxP reporter. By chance, this line broadly labels Mash1-expressing cells in CGE and Lateral Ganglionic Eminence, but not within the MGE (Miyoshi and Fishell, 2010). In order to restrict GFP labelling to LGins, we force-fed tamoxifen to pregnant transgenic mice at late embryonic stages. GFP labelling in hippocampal sections from P7 Mash1 mice given tamoxifen at E18, indicated that LGins displayed variable morphologies, including a neurogliaform dendritic anatomy (Fig. 2B), in agreement with the previously described CGE origin of NOS-negative neurogliaform cells (Tricoire et al., 2010). Their different morphology aside, the hippocampal distribution and density of LG- and EGins were quite distinct, as LGins were numerous with a high density in the CA1 and CA3c stratum lacunosum-moleculare and absence from CA3b (Fig. 2A&B). In order to further describe their morphometric features, LGins were filled with Neurobiotin and processed post hoc. Variable morphologies could be recovered and reconstructed (n=11, Fig. 4B). Axonal length was highly variable (range: 77 – 3826 μm; Fig. 5A) but on average significantly shorter than that of previously recorded low connectivity interneurons (Bonifazi et al., 2009). Moreover, both interneuron samples clustered into the same group, which was distinct from hub and EGins (Fig. 5C). Regarding basic electrophysiological properties, LGins were comparable to low connectivity interneurons (see Fig. 8). We conclude that in contrast to EGins, LGins originating from the CGE present much less developed morpho-physiological features. We next compared the network functional connectivity of EGins and LGins in order to determine whether EGins could become functional hubs at early postnatal stages.
To test whether EGins become functional hubs, we performed multineuron calcium imaging of hippocampal slices from tamoxifen-treated Dlx1/2CreERTM;RCE:LoxP mice, focusing on CA3c, the region where functional hubs tended to concentrate. Interestingly, GFP labelling was rather frequent in that area (cf. above and Fig. 1). Unfortunately, for unknown reasons, GFP positive cells could not be efficiently labelled with the calcium-permeable indicator used here (Fura2-AM) which precluded calculating their functional connectivity index based on the analysis of their spontaneous calcium events. Nevertheless, a distinctive feature of functional hub neurons (even more striking than their high functional connectivity index) was their higher “effective connectivity” as compared to any other neuron, including high functional connectivity pyramidal cells (Bonifazi et al., 2009). Effective connectivity maps can be determined by calculating the average calcium fluorescence change across trials in every imaged cell following the stimulation of a single one. To build the effective connectivity maps of EGins and LGins, we targeted and recorded in current-clamp conditions, GFP positive neurons (n=56 cells) and stimulated them by intracellular current injections while imaging single-cell calcium responses in other imaged neurons (see Experimental Procedures). We observed that EGins displayed effective connections with 43 ± 10 % of active cells (n=8 cells, Fig. 6) while LGins displayed a significantly lower effective connectivity index (10 ± 5%, n= 6 cells, ,p<0.05, Mann-Whitney, Fig. 6).
To further test the contribution to network dynamics of EGins, we tested their influence on spontaneous network dynamics in the form of GDPs. Of those examined, only 32 experiments are considered here (see Experimental Procedures).
A phasic stimulation protocol was applied i.e. short suprathreshold current pulses repeated at 0.1 to 0.2 Hz (the frequency range of GDP occurrence). As previously described (Bonifazi et al., 2009), cell/network interaction was estimated using three metrics: (i) the frequency of occurrence of spontaneous network synchronizations (GDPs) during the stimulation relative to the resting condition; (ii) the peristimulus time histogram plotting the average fraction of cells activated by the phasic stimulation; and (iii) forward/backward GDP phase shifts relative to a harmonic oscillator mimicking GDPs rhythm in resting conditions; in this way the number of observed versus expected GDPs was estimated over time. A cell was considered as affecting network dynamics significantly if it satisfied any of the above criteria. Remarkably, when stimulated, a very large majority of EGins (72%, n=23 cells, Figs. 7A–B&8A) significantly affected network dynamics as follows: (i) in 48 % cases the average GDP frequency was altered during stimulation since inter-GDP interval distributions were statistically different in control conditions and during stimulation (p<0.05, Student’s t-test or Mann-Whitney test, see Experimental Procedures). Twenty six percent EGins decreased GDP frequency to 78 ± 8 % of control while 22% increased GDP frequency to 176 ± 26% of control; (ii) in 48% of the cases, phasic stimulation triggered network synchrony in the form of GDPs in 32 ± 4 % of the trials within 1s after the stimulus (probability p that these events occurred by chance or due to the intrinsic periodicity of GDPs was <0.05, see Experimental Procedures). (iii) Phasic stimulations induced a forward or backward GDP phase shift as compared to resting conditions, in 52 and 26% of the cases respectively (p<0.01, see Methods). EGins often displayed more than one form of cell/network interaction since 56% of them significantly affected network dynamics according to at least two of the above criteria (p<0.05). Previous description of functional hub cells (Bonifazi et al., 2009) indicated that only stimulating GABA hub neurons could similarly affect network dynamics. In addition, whatever their axonal morphology (see precedent chapter), EGins have impact on the network activity (see Fig. 7). Accordingly, in contrast to EGins, none of the LGins showed any significant effect according to the three metrics described just above (Fig. 7C, n=7, p>0.05).
This evidence therefore shows that EGins act as functional hubs for the generation of GDPs.
The developing hippocampal network comprises a functional family of GABA hub interneurons essential for synchronization (Bonifazi et al., 2009). Here we find that a subpopulation of hub cells includes the GABA neurons that are generated earliest from the embryonic ganglionic eminences. This subpopulation of hub neurons is maintained throughout adulthood when a significant fraction of them expresses characteristic markers for GABA projection neurons (Gulyas et al., 2003; Jinno et al., 2007; Jinno, 2009). Thus our findings suggest at least anatomically, that hub cells are retained in adulthood, raising the possibility that hub function may be similarly preserved. Previous studies have identified pioneer neurons, including Cajal Retzius and subplate neurons, that contribute to the establishment of cortical networks (Del Rio et al., 1997; Ghosh and Shatz, 1992; Kanold and Luhmann, 2010; Kanold and Shatz, 2006; Super et al., 1998). We identify a new population of GABA pioneer neurons with intriguing developmental functionality, and which, unlike most pioneer neurons (Kanold and Luhmann, 2010; Meyer et al., 1998; Price et al., 1997; Super et al., 1998) persist into adulthood.
The conclusion that GABA EGins are largely synonymous with previously defined hub neurons (Bonifazi et al., 2009) is supported by several lines of evidence. First, as theoretically defined (Boccaletti et al., 2006) and experimentally verified (Bonifazi et al., 2009), hub neurons should comprise a small proportion of cells: precisely as EGins do. Second, unbiased multiparametric analysis of morphometric data indicate that EGins are physically similar to hub neurons, the cardinal feature being a widespread axonal morphology. Third, EGins’ fundamental electrophysiological features are comparable to hub neurons. Fourth, EGins possess a high effective connectivity index since their stimulation leads to the activation of many neurons. Notably, high “effective connectivity” is more predictive of a hub cell identity than a high “functional connectivity”: for example pyramidal cells could display a high functional connectivity without being functional hub neurons as their effective connectivity was low and stimulating them did not affect network dynamics (Bonifazi et al., 2009). Given the intrinsically limited temporal resolution of calcium imaging approaches, the latter criterion is undoubtedly the most stringent experimental test for a hub neuron.
Early generated GABA hub neurons resemble previously termed “connector” hubs rather than the basket-like subtype (Bonifazi et al., 2009; Bullmore and Sporns, 2009) and are therefore more likely to be classified as dendritic- rather than somatic-targeting interneurons as indicated by their preferential expression of SOM. This is consistent with the paucity of PV expression, the most common marker for somatic-projecting interneurons. Indeed, their axons preferentially arborized in the dendritic layers rather than in the pyramidal cell layer. The fact that EGins largely innervate dendritic layers is also in agreement with the earlier development of GABA synapses in these layers (Gozlan and Ben Ari, 2003). In addition to SOM, the majority of EGins express mGluR1α and to a lesser extent M2R. By contrast, PV, VIP and NOS expression was virtually absent. This constellation of neurochemical expression, axonal labelling of the fimbria, in combination with the regional distribution of their somata (Jinno et al., 2007), strongly suggest that early born hub neurons most likely develop into GABA projection neurons (Ferraguti et al., 2004; Gulyas et al., 2003; Jinno et al., 2007; Jinno, 2009). Although it would be expected that CA1 septum-projecting interneurons express CB or CR and NPY, these markers were virtually absent in the EGin population. Perhaps this can be accounted for by species differences, i.e. rat versus mouse.
The finding that GABA projection neurons are the earliest generated is in agreement with the early development of a pioneer GABA pathway from the rat hippocampus, reaching the septum as early as E16 prior to glutamatergic afferents and septohippocampal connections (Linke et al., 1995). It is also compatible with the observation that GDPs, recorded in the septum in the intact septohippocampal complex in vitro, originate in the hippocampus and propagate to the septum via hippocamposeptal-projecting neurons (Leinekugel et al., 1998). Still, the exact extrahippocampal projection pattern of early born hub neurons may not be restricted to the septum as GABA projection neurons have been reported to target a variety of structures (Ceranik et al., 1997; Fuentealba et al., 2008; Higo et al., 2009; Jinno et al., 2007; Jinno, 2009; Miyashita and Rockland, 2007). Future retrograde labelling studies of the cells targeted in this study will be required to precisely address this issue. Interestingly, due to their long distance anatomic connectivity and sparseness, GABA neurons with an extrahippocampal projection were already speculated to carry a hub function and provide a wiring economy supporting the emergence of network oscillations in the adult hippocampus at a reasonable cost (Buzsaki et al., 2004).
If the intrahippocampal postsynaptic targets of EGins are not the somata of glutamatergic pyramidal neurons, their exact nature still remains to be elucidated. Interestingly, GABA projection neurons have been previously analyzed in detail at the electron microscopic level at two developmental time points (Gulyas et al., 2003; Jinno, 2009). In CA1 hippocampal slices from juvenile rats, interneurons are their major targets (Gulyas et al., 2003) whereas in adult rats in vivo, these cells were reported to selectively innervate the dendritic shafts of pyramidal cells (Jinno et al., 2007). An initial selective targeting of interneurons by EGins hub neurons would match a previous report suggesting that interneurons are the targets of the first GABA synapses formed in the CA1 hippocampal region (Gozlan and Ben Ari, 2003). Future studies are needed to test whether GABA neurons are, at least transiently, the main targets of early born hub neurons, an architecture that would provide ideal conditions for the generation of GDPs.
From the above, it is tempting to conclude that early generated hub neurons constitute a specific interneuron family. Moreover, it implies a strong genetic predetermination in the development of GABA projection neurons and suggests that in addition to their morpho-physiological features (Butt et al., 2005), specialized interneuron function may also be strongly predetermined by embryonic origin. Furthermore, the precocious maturation of hub neurons in principle makes them less susceptible to activity-dependent maturation processes as these cells likely develop in a poorly active environment.
Still, hub cells are not a homogenous interneuron family. Indeed, we had previously described GABA hub neurons with a basket-like axonal pattern (Bonifazi et al., 2009), a population that was not observed in the Neurobiotin-filled EGins during our in vitro experiments. Nevertheless, rare EGins were found to be immunopositive for PV in stratum pyramidale or granular layer at P7 and P30, suggesting the presence of occasional PV-positive perisomatic interneurons. The embryonic origin and adult fate of basket-like hub neurons therefore still remains to be determined. This subpopulation of hub neurons may similarly be maintained into adulthood, as perisomatic interneurons with the ability to time the incidence of sharp waves have recently been described in adult hippocampal slices (Ellender et al., 2010). In addition, the population of early generated hub neurons itself displays some diversity at P7, which persists in adult animals as revealed by the cell reconstructions. Heterogeneity also prevails in the population of GABA projection hippocampal neurons since at least seven different types of them have been previously described (Fuentealba et al., 2008; Jinno et al., 2007). A common embryonic origin may link these various cell types within a family of GABA projecting neurons. Alternatively, different classes of GABA neurons may progressively and transiently function as hub cells at different postnatal stages of development. If true, what we previously grouped as “hub cells” may comprise distinct populations that differentially contribute in the generation of GDPs. Such variety in hub cells could provide functional redundancy that could conceivably protect against developmental insults that impaired particular populations.
When considering early born hubs from a functional viewpoint, it is important to stress that they are solely defined by their high connectivity index, which at least theoretically allow them to act as important nodes in the flow of information between populations of neurons. Importantly, they do not create rhythms but merely convey activity to many neurons: hub neurons are not necessarily “pacemakers”. In fact, basic electrophysiological characterization of EGins did not reveal any major intrinsic oscillatory mechanism within these cells. These cells are very likely to be synaptically-driven since they display a higher sEPSPs frequency than other GABA neurons. Accordingly, electron microscopy analysis of the synaptic innervation impinging onto GABA projection neurons showed that these cells almost exclusively received glutamatergic synapses (Takacs et al., 2008). Understanding whether hub neurons are critical for the production of GDPs in physiological conditions requires approaches enabling their selective and complete elimination. The present results provide a first step towards achieving this technically challenging task.
The finding that EGins are hub cells may imply that, as proposed for other pioneer hippocampal neurons, these cells may provide as molecular cues for the development of layer specific hippocampal and extra-hippocampal afferents (Super et al., 1998). Comparison with neocortical development, a region where pioneer neurons have been extensively described, may be particularly instructive. Indeed, besides their early generation, hippocampal hub cells share several remarkable properties with subsets of subplate neurons including: (1) long distance projections (Chun et al., 1987; Kanold and Luhmann, 2010; Luhmann et al., 2009; Tamamaki and Tomioka, 2010; Voigt et al., 2001); (2) mature electrophysiological properties (Hirsch and Luhmann, 2008); (3) the expression of SOM (Chun et al., 1987; Tamamaki and Tomioka, 2010), and GAD67 (Arias et al., 2002); (4) a role in driving synchronous activity in immature cortical networks (Dupont et al., 2006; Kanold and Luhmann, 2010; Voigt et al., 2001) that identifies the subplate as a “hub station” (Kanold and Luhmann, 2010). Whether hub neurons indeed exist in the developing neocortex and persist into adulthood remains an open question.
The present finding is also interesting from the perspective of pathology. As alluded to above, these cells may provide robustness against pathological insults, in particular those resulting from environmental factors influencing brain development. Interestingly, it was previously shown that the septum-projecting subclass of CA1 SOM-containing neurons is selectively spared in a chronic rat model of Temporal Lobe Epilepsy, indicating that early born hub neurons may be resistant to epileptogenesis (Cossart et al., 2001). Whether EGins are central to synchronization processes in epileptic networks therefore remains a viable hypothesis, supported by computational simulations (Morgan and Soltesz, 2008). Now that a subpopulation of hub neurons is accessible to the conditional expression of genes of interest, including optogenetic vectors (Katzel et al., 2010), the involvement of superconnected neurons in different forms of physiological or pathological oscillations can be explored.
All animal use protocols were performed under the guidelines of the French National Ethic Comity for Sciences and Health report on “Ethic Principles for Animal Experimentations” in agreement with the European Community Directive 86/609/EEC. Double-homozygous Mash1BACCreER/CreER/RCE:LoxP+/+ and Dlx1/2CreER/CreER/RCE:LoxP+/+ (Batista-Brito et al., 2008; Miyoshi et al., 2010) male mice were crossed with 7- to 8-week-old wild-type Swiss females (C.E Janvier, France) for offspring production. To induce CreER activity, we administered a tamoxifen solution (Sigma, St. Louis, MO) by gavaging (force-feeding) pregnant mice with a silicon-protected needle (Fine Science Tools, Foster City, CA). We used 2mg of tamoxifen solution per 30g of body weight prepared at 10 mg/ml in corn oil (Sigma). Pregnant females crossed with Dlx1/2CreER/CreER/RCE:LoxP+/+ males were force-fed at embryonic days 7.5 or 9.5 days post vaginal plug (corresponding to the embryonic ages E7.5–E9.5) in order to label early expressing Dlx1/2 precursors with GFP in the embryos. For simplification purposes GFP expressing mice pups originating from these gavaged females are named tamoxifen-treated Dlx1/2CreERTM;RCE:LoxP. Similarly, pregnant females crossed with Mash1BACCreER/CreER/ RCE:LoxP+/+ males were gavaged at E18.5, in order to label late expressing Mash1 precursors in the embryos named Mash1CreERTM;RCE:LoxP mice.
To assess the temporal precision of EGins labelling, we performed six injections of 5-Bromo-2′-deoxyuridine (BrdU, 10mg/ml in PBS) every 4 hrs, starting 6 hrs after pregnant mice were force-fed with tamoxifen at E9.5 (50 μg/g i.p.). In another set of experiments BrdU was injected 31 hrs after tamoxifen force-feeding to check that tamoxifen action does not extend over 24 hrs. Sections from E12.5 embryos were immunoreacted for both GFP and BrdU as detailed in Fig. S1. Similar results were obtained when tamoxifen was force-fed at E7.5 or E9.5 (Fig. S1).
Horizontal hippocampal slices (380 μm thick) were prepared from 5 to 7-day-old (P5–7) tamoxifen-treated Dlx1/2CreERTM;RCE:LoxP (n= 56 slices) or Mash1CreERTM;RCE:LoxP (n= 39 slices) mouse pups with a Leica VT1200 S vibratome using the Vibrocheck module in ice-cold oxygenated modified artificial cerebrospinal fluid (0.5 mM CaCl2 and 7 mM MgSO4; NaCl replaced by an equimolar concentration of choline). Slices were then transferred for rest (1 hr) in oxygenated normal ACSF containing (in mM): 126 NaCl, 3.5 KCl, 1.2 NaH2PO4, 26 NaHCO3, 1.3 MgCl2, 2.0 CaCl2, and 10 D-glucose, pH 7.4. For AM-loading, slices were incubated in a small vial containing 2.5 ml of oxygenated ACSF with 25 μl of a 1 mM Fura2-AM solution (in 100% DMSO) for 20–30 min. Slices were incubated in the dark, and the incubation solution was maintained at 35°–37°C. Slices were perfused with continuously aerated (3 ml/min; O2/CO2-95/5%) normal ACSF at 35–37°C. Imaging was performed with a multibeam multiphoton pulsed laser scanning system (LaVision Biotech) coupled to a microscope as previously described (Crepel et al., 2007). Images were acquired through a CCD camera, which typically resulted in a time resolution of 50 to 150 ms per frame. Slices were imaged using a 20X, NA-0.95 objective (Olympus). Imaging depth was on average 80 μm below the surface (range: 50–100 μm).
A total of 121 neurons were recorded: 65 were recorded only for morpho-physiological characterization in adult slices, whereas 56 were recorded while imaging. Out of the latter, only 32 were considered. The other experiments (n = 24) were discarded because they did not comply either one of the following criteria: (1) stable electrophysiological recordings at resting membrane potential i.e. the holding current did not change by more than 15 pA; (2) stable network dynamics measured with calcium imaging, i.e. the coefficient of variation of the inter-GDP interval did not exceed 1; (3) complete labelling of the recorded cell; (4) good quality calcium imaging while recording. Neurons were held in current-clamp using a patch-clamp amplifier (HEKA, EPC10) in the whole-cell configuration. Intracellular solution composition was (in mM): 130 K-methylSO4, 5 KCl, 5 NaCl, 10 HEPES, 2.5 Mg-ATP, 0.3 GTP and 0.5 % Neurobiotin. No correction for liquid junction potential was applied. The osmolarity was 265–275 mOsm, pH 7.3. Microelectrodes resistance was 4–8 MOhms. Uncompensated access resistance was monitored throughout the recordings. Values below 20MOhms were considered acceptable and the results were discarded if it changed by more than 20%. Whole cell measurements were filtered at 3 kHz using a patch-clamp amplifier. Recordings were digitized online (10 kHz) with an interface card to a personal computer and acquired using Axoscope 7.0 software. Spontaneous EPSPs were detected and analyzed using the MiniAnalysis software. For most stimulation experiments, the movie acquisition time was separated evenly between (1) a ~ 3 min resting period during which the cell was held close to Vrest (i.e. zero current injection); (2) a ~3 min stimulation period during which phasic stimulation protocols were applied; (3) a ~3 min recovery period where the cell was brought back to resting membrane potential. Stimulation protocol: suprathreshold current pulses (amplitude: 100–200 pA, duration: 200 ms) repeated at 0.1–0.2 Hz. Vrest was measured as the membrane potential baseline value obtained in current-clamp mode in the absence of current injection. The action potential threshold (Vthreshold) and amplitude, membrane capacitance (Cm) and resistance (Rm) were measured offline using Clampfit. Vrest and Vthreshold were not corrected for errors due to high input resistance values in developing neurons.
Slices were fixed overnight at 4 °C in Antigenfix, rinsed in PBS containing 0.3% Triton X-100 (PBST) and incubated overnight at room temperature in cy3-streptavidin (1/1000 in PBST). Post-hoc analysis was performed using a confocal microscope. Stacks of optical sections were collected for computer-assisted neuron reconstructions.
Animals were deeply anesthetized with a ketamine (50 mg/ml) and xylazine (7.5 mg/ml) solution at a dose of 2 ml/kg (i.p.) and perfused transcardially with 4% paraformaldehyde in PB (1ml/g) at a constant flow (2 ml/min). Brains were post-fixed overnight in fixative and washed. Horizontal brain sections (40 μm) were routinely processed for multiple immunofluorescence. Briefly, after pre-incubation in 5% normal donkey serum, sections were incubated in a mixture of primary antibodies from different species, then incubated in appropriate secondary antibodies conjugated with either Al488 (1:500, Molecular Probes; Invitrogen), Cy3, Cy5, DL488, DL549, DL649 (Jackson ImmunoResearch Laboratories Inc.) for 2 hrs. The dilutions, characteristics, specificity and sources of primary antibodies are presented in Table S1. Some sections were also processed for GFP immunoperoxydase. No staining corresponding to specific labelling was observed when primary antisera were omitted. In multiple labelling experiments using primary antibodies from different species, the lack of cross-reactivity of the secondary antibodies was consistently checked.
Images were obtained with a Zeiss AxioImager Z2 microscope coupled to a camera (Zeiss AxioCam MR3). .Immunofluorescence images were acquired using a halogene HBO lamp associated with (470/40, 525/50), (545/25, 605/70) filter cubes for detection of Al488, Cy3 or DL549, Cy5 or DL649. Counts were performed manually. All results are given as percentages of total number of cells (Fig. 1C) or as means of percentages ± SEM (Fig. 3L and M), n being the number of mice per group.
Fourty three neurons were reconstructed with a computer assisted system attached to a microscope (Neurolucida, MicroBrightfield). Out of those, 31 were included in the morphometric analysis. Morphological variables included: dendritic and axonal lengths, dendritic and axonal surfaces, and number of dendritic and axonal terminals. We also performed a Sholl analysis in order to determine the distribution of the number of axonal intersections with circles of increasing radius (20μm steps) centred at the cell’s soma. Cluster analysis for morphological data was performed using Statistica software. Our analysis was performed with Euclidean distances using Ward’s method. According to Ward’s method, cases are assigned to clusters so that the variance (sum of squared deviations from the mean) within each cluster is minimized..
We used custom designed Matlab software (Bonifazi et al., 2009) which allowed: 1) automatic identification of loaded cells; 2) measuring the average fluorescence transients from each cell as a function of time; 3) detecting the onsets and offsets of calcium signals; 4) reconstructing the functional connectivity of the imaged network.
Network synchronizations (GDPs) were detected as synchronous onsets peaks including more neurons than expected by chance, as previously described (Bonifazi et al., 2009).
In order to identify cells in the network responding to phasic stimulations, for each cell we first calculated the average fluorescence change across trials in a time window between −1 and +1 seconds centred on the time of the stimulus. Cross-correlation between the average calcium signal of the cell and the calcium signal of the stimulated cell was calculated at time lags varying between −1 and +1 s. If the maximum of the cross-correlation exceeded 0.5 and occurred at positive times, indicating that the activation of the cell followed the stimulation, the cell was considered as responding to the stimulation. In order to colour-code the effective connectivity map, we built a matrix from the calcium image of the slice and we assigned to each cell its maximal cross-correlation value. The image was then convolved with a Gaussian of unitary amplitude and 8 μm radius.
To establish whether the stimulation of a single neuron was able to influence the occurrence of GDPs we used the same procedure as previously described (Bonifazi et al., 2009).
We thank Drs. R. Machold for generating the Dlx1/2-creER allele and J. Johnston for providing the Mash1CreERTM mouse. We thank Drs. Y. Ben-Ari and S. Feldt for critical comments. We thank Dr. M. Esclapez for providing occasional access to her Neurolucida system. Research in the Cossart group was supported by grants from the European Research Council (ERC FP7 Young Investigators #242852), the Fondation pour la Recherche Medicale (Equipe FRM 2008), the Fondation Bettencourt Schueller, I.N.S.E.R.M., the Ville de Marseille and Region P.A.C.A. and the F.R.C. Drs. R. Cossart and A. Baude are funded by the C.N.R.S. Research in the Fishell laboratory is supported by the National Institutes of Health (NIH RO1 grants R01MH071679 and R01NS039007).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.