THE JOURNAL OF COMPARATIVE NEUROLOGY 293:1-25 (1990)

Axo-AxonicChandelier Cells in the Rat Fascia Dentata: Golgi-ElectronMicroscopy and Immunocytochemical Studies EDUARDO SORLANO, ROBERT NJTSCH, ANI) MICHAEL FROTSCHER Institute of Anatomy, Johann Wolfgang Goethe University, D-6000 Frankfurt am Main 70, Federal Repuhlic of Germany (E.S.. R.N., M.F.); Cnit of Cell Biology, Faculty of Biology, University of Barcelona, Barcelona 08028. Spain (E.S.)

ABSTRACT Synaptic transmission can be blocked very efficiently by inhibitory synapses on axon initial segments. Inhibitory chandelier cells rorming synapses on the axon initial segment, ol‘pyramidal neurons have heen foiind in the neocortex and hippocampus proper. Here we describe a n axo-axonic local circuit neiiron in the rat fascia dentata t,hat establishes synaptic contacts with axon initial segments of numerous d e n h t e granule cells. Examination of a large number of Golgi-impregnated nongranule cells in the fascia dentata of rats revealed a group of neurons with characteristics of chandelier cells. Thus these cells exhihited a n extensive axonal plexus within t h e granular layer that characteristically formed vertical aggregations of axonal varicosities. T h e cell bodies of these neurons were located in the inner molecular layer or in t h e outer part of t h e granular layer. Their dendrites invaded t h e molecular layer, suggesting a n afferent innervation similar t,o that of the granule cells. Well impregnated putative axo-axonic cells were gold-toned for a n electron microscopic analysis. T h e cell bodies and dendrites of t,hese neurons exhibited characteristic ultrastructural features of nongranule cells, i.e.. large amounts of perinuclear cytoplasm, infoldings of t h e nuclear membrane, and a large number of synaptic contacts on t h e perikaryon and on t h e smooth dendritic shaft,s. T h e axon originating from the cell body or from a proximal dendrite gave rise to nii~nerousvesicle-filled varicosities t h a t almost. exclusively formed syinmetric synaptic contacts with axon initial segments. A semiquantitative study of five axonal complexes demonstrated t h a t 92.3 cc, of ident.ified postsynspt,ic. elements were initial segments of granule cell axons. Immunostaining with antih o c k s against glutamate decarboxylase (GAD) and parvalbumin (PAR\.:) revealed a subpopulatiun of neurons t h a t very much resembled the Golpiimpregnated axo-axonic cells with regard to cell body location, dendritic arborization, and fine structural characteristics of perikarya and dendrites. GAD and PARV were found to be coexistent. in these cells. Moreover, we found GAD- arid PARV-immunoreactive terminals in symmetric synaptic contact, with axon initial segments of granule cells. T h e present study has show11 a hitherto unknown axo-axanic cell in t h e r a t fascia dentata. On t h e basis of our immunocytochemical findings, we hypothesize that this cell exerts a slrong inhibitory effect an dentate granule cells. This way, signal transmission from t h e fascia dentata to the hippocampus proper within t h e “trisynaptic pat,hway” can efticiently be controlled by a group of highly specialized neurons. Key words: hippocampal inhibitory neurons; glutamic acid, decarboxylase,

parvalbumin

Accepted September 7,1989. Address reprint requests to Dr. Eduardo Soriano, Unit of Cell Biology, Fac ulty of Biology, University of Barcelona, Barcelona 08028, Spain.

0 1990 WILEY-LISS, INC.

E. SORIANO ET AL.

2

A

ML

H

Fig. 1. A: Golgi-impregnated chandelier cell axonal complex in the granular layer of the fascia dentata. Axonal branches form vertically oriented rows or clusters of boutons mainly in the lower half of the granular layer. Collaterals illustrated in B and C are indicated (b,c). B,C: Photomontages of the collaterals labeled in A. Note variations in packing density of collaterals (arrows) and terminal-like boutons in the granular layer. D: Several chandelier cell axonal complexes are impregnated

in the same section. It appears that they interdigitate to form very complex, regularly spaced vertical aggregations of boutons (small arrows) in the granular layer. Large arrow pointing to one of these rows seen a t higher magnification in E. Different collaterals converge to form this vertical aggregation of axonal varicosities. Bar = 100 pm; B,C, x330; D. x270; E, x750.

DENTATE AXO-AXONIC CELLS

3

Figure 1D-E

The granule cells are the dominating cell type of the fascia dentata. They are densely innervated by afferents from the entorhinal cortex, which in turn integrates information from various neocortical areas. Via the mossy fibers, the granule cells project onto pyramidal neurons in the CA3 region of the hippocampus. The CA3 pyramidal neurons give rise to the Schaffer collaterals innervating CAI pyramidal cells. Physiological studies have shown that the entorhinal afferents, the mossy fibers, and the Schaffer collaterals are excitatory (Andersen et al., '71). This trisynaptic excitatory pathway from the entorhinal cortex to the CA1 region of the hippocampus is generally regarded as the basic circuitry of the hippocampal formation. All members of this excitatory pathway are modulated by a variety of extrinsic and intrinsic fiber systems. For instance, commissural fibers from the contralateral hippocampus as well as cholinergic fibers from the septum and ascending catecholaminergic brainstem afferents have been described to terminate in both the fascia dentata and the hippocampus proper (e.g., Frotscher et al., '88). The dominating intrinsic elements that control the trisynaptic pathway are the y-aminobutyric acid (GABA)-ergic local circuit neurons. Among them are the basket cells that form a dense pericellular plexus around the cell bodies of pyramidal neurons and granule cells (Cajal, '11; Lorente de N6. '34; Ribak et al., '78; Amaral, '78; Lubbers and Frotscher, '87). Electron microscopic (EM) studies have, in fact, demonstrated that basket cell axons establish symmetric synaptic contacts on the cell bodies and proximal dendrites of these neurons (Ribak and Seress, '83; Frotscher et al., '84; Kosaka et al., '84; Liibbers and Frotscher, '87). The basket cells have accord-

b

c d g GAD GI. is I1 ML PARV S

Terminal bouton Capillary

Dendrite Granule cell Glutamate dcrarboxylase-immunoreartive Granular layer of fascia dcntala Axon initial segment Hilus Molecular layer of fascia dentata Parvalbumin-immnnoreactive SDine

ingly been regarded as the major inhibitory component in the hippocampus and fascia dentata (Andersen. '75). Inhibitory synapses on the axon initial segment where the action potential is generated can very efficiently block transmission. In the neocortex and hippocampus proper, axo-axonic chandelier cells have been found that establish multiple contacts with the axon initial segments of pyramidal neurons (Somogyi, '77, '79; Somogyi et al., '82, '83a,b, '85; Fairen and Valverde, '80; Peters et al., '82; Freund et al., '83; DeFelipe et al., '85; for review, see Peters, '84). In the fascia dentata, GABA-ergic symmetric synapses on the initial segments of granule cell axons have been described (KOsaka, '83; Kosaka et al., '84; Ribak and Seress, '83). However, the parent neuron could not yet be identified. By means of the Golgi-EM procedure ( F a i r h et al., '77), we found an axo-axonic cell in the fascia dentata, which establishes a large number of synaptic contacts with initial segments of granule cell axons. Immunocytochemical studies employing anti bodies against glutamate decarboxylase (GAD) and the Ca-binding protein parvalbumin (PARV), suggest that these neurons are PARV-containing GABAergic inhibitory cells. These neurons may thus efficiently control the hippocampal trisynaptic pathway at the level of the granule cells. Together with the above-mentioned studies on neocortical and hippocampal chandelier cells, our data provide evidence that GABA-ergic neurons with a target selectivity for axon initial segments are a constant component of inhibitory mechanisms in bhe cerebral cortex.

MAlERIALS AND METHODS Golgi impregnation Light microscopic observations were made on a collection of 120 Golgi impregnated rat hippocampi partially used in a previous study (Marti-Subirana et al., '86). Sprague-Dawley rats (11-12 weeks old) were anesthetized with ether and transcardially perfused with a mixture of I *O paraformaldehyde and 1'( glutaraldehyde in 0.1 M phosphate buffer (pH 7.4). The brains were removed from the skull and fixed in the same solution at 4 O C overnight. On the following day, the hippocampi were dissected and processed with different variants of the Golgi method as proposed b y Colonnier ('64), Braitenberg et al. ('67), and Valverde ('70). After suuerficial embedding in paraffin, sections were cut transversiy to the

E. SORIANO ET AL.

4

C

h

t Figure 2

5

DENTATE AXO-AXONIC CELLS longitudinal axis of the hippocampus on a sliding microtome. Serial sections, 150-200 pm thick, were dehydrated in ethanol and mounted in Araldite. The cells were photographed and drawn with the aid of a camera lucida and a x 63 oil-immersion objective (Zeiss N.A.: 1.4).

Gold-EM Thirty-seven Sprague-Dawley rats, 7-12 weeks of age, were used for the Golgi-EM analysis. Animals were fixed in the same solution that was used for the light microscopic studies. Since we observed the best staining of axons with Colonnier's ('64) variant of the Golgi method, most hippocampi destined for an EM analysis were processed according to this protocol. To standardize the method, tissue blocks were immersed in the dichromate solution a t constant temperature (10-12°C). This required a prolongation of the chromation period (up to 6-7 days). Thereafter, the tissue blocks were put into 0.75% silver nitrate for 24 hours at room temperature. The blocks were then processed according to a slightly modified variation of the Golgi-EM procedure described by Fair& e t al. ('77, '81). Prior to gold-toning, the sections were put into petri dishes containing absolute glycerol and illuminated with cold white light for 2 hours at room temperature (Fairkn et al., '81; Somogyi et al., '81). Gold-toning was performed by incubating the sections in 0.05% gold tetrachloride in distilled water containing 20% glycerol. Addition of glycerol made it necessary to prolong the gold-toning up to 1-2 hours. This was followed by gently washing the sections in distilled water. Reduction to metallic gold was accomplished by incubating the material in 0.0'2% oxalic acid for 4 minutes. After deimpregnation in 1%sodium thiosulphate for 1 hour, the secphosphate-buffered osmium tions were postfixed in 2 tetroxide. Thereafter, the sections were dehydrated (blockstained with uranyl acetate in 70% ethanol) and flatembedded in Araldite between aluminium foil and transparent plastic wrap. The sections were examined in the light microscope, and selected cells were processed for EM. The cells were selected with regard to the completeness of axonal staining. After t,he cells were photographed and drawn with the aid of a microscope equipped with a drawing tube, the cells were reembedded in plastic capsules and serially sectioned on a Reichert Om 173 or LKB III ultratome. To facilitate the identification of single processes in the EM, the blocks were repeatedly taken out of the ultratome, and drawings of the remaining parts of the cells were made. Thin sections were mounted on single-slot grids coated with formvar film, stained with lead citrate and uranyl acetate, and photographed in a Zeiss 109 EM.

Fig. 2. Photomontage (A) and camera-lucida drawing (C) of a chandelier cell in the fascia dentata. The cell body is located just above the granular layer and gives rise to several dendrites arhorizing in the molecular layer. The main axonal branch (arrows in A) runs parallel to the granular layer and gives off numerous collaterals forming dense vertical aggregations of axonal varicosities in the inner portion of the granular layer and in the adjacent infragranular zone. Two such vertical rows are shown in €3 (arrows). Note that only part of the numerous axonal branches is illustrated in C. Arrows in C point to the origin of three collaterals not illustrated here, which also formed dense arborizations. A, x 370; B, x620. Bar = 100 pm.

GAD immunocytochemistry: Zinc salicylate

enhancement for light microscopy Three adult Sprague-Dawley rats were processed for GAD immunohistochemistry following the zinc salicylate procedure of Mugnaini and Dahl ('83),which gives a more complete staining of dendritic processes of immunoreactive cells. Under ether anesthesia, rats were perfused with saline (0.9"; NaCl) until the blood was completely washed out. Then the animals were perfused with 500 ml of a fixative cont,aining 4 Y;, paraformaldehyde and 0.5% zinc salicylate in 0.45S0 NaCl (pH 6.5). The animals were left in situ for 2 hours and were then perfused with 10% sucrose in 0.9% NaC1. Brains were removed from the skull and stored in saline containing 30Tr sucrose and 0.1% sodium azide at 4OC for 2 days. After freezing on dry ice, the brains were cut a t 40 gm on a sliding microtome. The sections were washed in 0.5 M Tris HC1 buffer (pH 7.6) and processed "free-floating" for immunohistochemistry according to the peroxidaseantiperoxidase (PAP) technique of Sternberger et al. ('70). Tris buffer was used for rinsing as well as for the antibody dilutions. Following preincubation in 107" normal rabbit serum containing 0.1 M D/L-lysine, the sections were incubated a t room temperature with the primary GAD antibody diluted L:2,000 (a generous gift of Drs. M.L. Tappaz and W.H. Oertel). The characteristics of this antiserum have been reported previously (Oertel et al., '82). After careful washing in Tris buffer, the sections were incubated in rabbit antigoat IgG (Cappel, diluted 1:100) for 90 minutes and, thereafter, in goat PAP complex (Sternberger-Meyer Immunocyt,ochemicals, diluted 1:200, 2 hours). The tissuebound peroxidase was visualized by processing the sections in 0.03% diaminobenzidine-tetrahydrochloride (DAB; Sigma type IV) and 0.01 9;' hydrogen peroxide in 0.1 M Tris buffer (pH 7.6) for 15-20 minutes. The sections were postfixed in 0.05% osmium tetroxide and mounted on chromalum-coated slides. After air-drying, the slides were dehydrated in ethanol and coversliped in Eukitt (Merck).

EM immunocytochemistry for GAD and PARV Twelve young adult Sprague-Dawley rats were used for this part of the study. The animals were perfused under ether anesthesia with 70 ml of saline followed by 500 ml of a fixative containing 4 7;) paraformaldehyde, 0.08 % glutaraldehgde, and 15!DI. I n the cell bodies located in the inner molecular layer, asymmetric synapses clearly outnumbered symmetric junctions. Frequently, boutons were observed in synaptic contact with the gold-toned cell body or dendrite and a n adjacent nonimpregnated profile (Figs. 5I), 6A). Marly synapses. hoth symmetric and asymmetric ones, impinged on the drndritic shafts (Fig. 5R). With increasing distance from the cell hody, the port.ion of asymmetric synaptic contacts increased. Identified dendritic profiles in the outer two-thirds of t h e molecular layer were contacted aliiio3t exclusively b y houlons lorming asymmetric junct ions (Fig. 6A.R). In contrast, basal dendrites traversing the granular layer estalilished very few synaptic cont,acts. However, the number of contacts increased when they entered the hilar region (Fig. 6C). Some o f the boutons surrounding

16

E. SORIANO ET AL.

Fig. 12. Colocalization of GAD and PARV in semithin serial cryostat sections. A. Open arrow points to large GAD-immunoreactive perikaryon in the innermost portion of the molecular layer. This cell as well as several other nenrnnS (solid arrows) are also immunoreactive f o r PARV (B).h'ote smaller number of PARV-immunoreactive cells. Small

PARV. Arrows as in A. C,D: Higher magnification of the large irnmunoreactive cell in the molecular layer (C, section stained for GAD; D, sertion stained for PARV). The position of capillaries (c) is indicated. A,B, x 100; C,D, x275.

the dendritic profile shown in Figure 6 were likely to be mossy fiher collaterals: they contained numerous clear synaptic vesicles intermingled with a few dense core vesicles. T h e overwhelming majority of input synapses on apical and basal dendrites contacted dendritic shafts. Only rarely did a n axon terminal impinge on a gold-toned dendritic

spine (Fig. 6B). I n Figure 6B it can be seen that both the head and t h e stalk of t h e spine were contacted by terminals. -4xonal rornplex. A s e c t i o n i n g p r o t o c o l ( s e e Frotscher, '85)allowed us to ident.ify t h e gold-toned axons in serial thin sections. Only those parts of t h e axonal plexus

(;AD-immunoreactive cells in the molecular layer did not stain for

DENTATE AXO-AXONIC CELLS

17

Fig. 13. (:orrelated light (A) and electron (B-E) micrographs of a large GAbimmunoreartive neuron in the inner molecular layer. A Arrowhead points to origin of dendrite shown in E. Another proximal dendrite (d) is demonstrated in B. Arrow points to small GAD-immiinoreactive neuron. B: Electron micrograph of the cell body region. Arrow points to indentation of nuclear membrane. Arrowhead labels origin of dendrite shown in E. Another proximal dendrite (d) is also marked in A.

C,D: Boutons eslablishing asymmetric (C) and symmetric (D) synaptic contacts (arrows) with the immunostained perikaryon. Terminals also contact unlabeled spines (s in C) or dendrites (d in D). E: Several GADpositive boutons (arrows) impinge on cell body and proximal dendrite. The origin ofthis dendrite ib labeled by arrowheads in A and B. A, x310; B, ~ 5 , 5 0 0C,D. ; ~ 4 0 , 0 0 0E; , x 19,000.

t h a t were conipletely free of other impregnated axon structures were analyzed. T h e labeled boutons (1-2 pm in diameter) were easily identifiable by the presence of the gold precipitate, which in some cases obscured cytoplasmic

structures. 'l'he boutons contained round synaptic vesicles

arid exclusively formed symmetric synaptic junctions. A s seen in semithin sections (Fig. 7.4), rciws of boutoris wcre often located uiideriieath granule cell somata and fol-

E. SORIANO ET AL.

18

Figure 14

DENTATE AXO-AXONIC CELLS lowed a vertical course similar to that of the mossy fibers. This strongly suggests contacts on axon initial segments since granule cells are known to lack basal dendrites extending into the hilus. The postsynaptic elements of gold-toned boutons were, in fact, identified as axon initial segments on the basis of unique fine structural features such as the characteristic membrane undercoating and fascicles of microtubules (Fig. 7B-D). By analyzing serial thin sections, postsynaptic axon initial segments could occasionally be traced back to a granule cell perikaryon (see Figs. 9A,E, 10). The axon initial segment emerges from the axon hillock a t the basal pole of the granule cell and follows a straight course towards the hilus. In addition to gold-toned terminals, postsynaptic initial segments are often covered by other, nonimpregnated boutons, which also established symmetric synaptic junctions (see, e.g., Figs. 7E, 9D, I0B.C). This configuration indicates convergence of terminals from different neurons on the same postsynaptic element. Conversely, rows of impregnated boutons were observed that contacted only one axonal element (Fig. 9A,E). Various examples of contacts on initial segments formed by t,he goldtoned identified chandelier cells are illustrated in Figures 710. Most boutons were found within (Figs. 7, 9), and less often directly underneath, the granular layer, i.e., at the site of origin of the mossy fibers (Fig. 8). Synaptic contacts of gold-toned boutons were seen on proximal and distal portions of the axon initial segment, where they established symmetric junctions with the axonal shafts (see, e.g., Fig. 8C,D) as well as with small spines arising from the axon (Fig. 7D,E, 8A,R). Labeled and unlabeled boutons were often seen contacting the same axonal spine (Fig. 7E, 8B). We never observed a gold-toned terminal simultaneously contacting two or more profiles, although the boutons were thoroughly studied in serial sections. The same holds true for nonimpregnated terminals seen in apposition with initial segments. To assess the degree of synaptic specificity of the present dentate axo-axonic cells, a large number of boutons was classified according to the nature of the postsynaptic element. Table 1 summarizes the results of this semiquantitative study on Golgi-impregnat,ed boutons of the five chandelier axonal complexes. From a total of 144 terminals forming synaptic junctions, 133 (92.3 5 ) contacted profiles that were unequivocally identified as axon initial segments. Most of these synapses occurred on the shafts of initial segments, whereas 13 synapses (9 9 ; ) were observed on axonal spines or excrescences. Two boutons contacted spines that could not be traced back to the parent profile, although the ctiaracteristic membrane undercoating of the spine suggested an axonal origin. An additional 4.9% of the boutons were observed in contact with profiles lacking unequivocal ultrastructural features of axon initial segments. However, we

Fig. 14. Correlated light and electron micrographs of a PARVimmunoreactive neuron in the outer granular layer (arrow in A). B: The cell body contains an indented nucleus (open arrow) and a nuclear rod (small arrows). Note large amounts of cytoplasm rich in organelles. Several PARV-immunoreactive terminals impinge on the imniirnolabeled cell body. One of them (arrowhead) is illustrated at higher magnification in C. D: A PARV-positive bouton (asterisk) and an unlabeled terminal estahlish synaptic contacts (arrows) on the cell body. A. ~ 2 7 . 5 ;B, ~5,500;C,D, ~27,500.

19 were unable to find an identified terminal contacting a thick primary dendrite of a granule cell. In one of the five axonal complexes, two synapses on one granule cell body were ohserved. However, even in this case, the contacts were near t.he axon hillock (not shown). Thus our data suggest that chandelier cells in the fascia dentata have a high degree of target specificity. This is particularly evident in the case of gold-toned boutons in the inner molccular layer (Fig. 10). Also here, where the ascending granule cell dendrites are the predominating structures, boutons of chandelier cells established synapses on axon initial segments. It is well known that axon initial segments are very rarc in this zone and are most likely derived from ectopic granule cells.

Immunocytochemical studies Light microscopy of GAD- and PARV-immunoreactitle elements in the fascia dentata. Sections reacted for GAD using zinc salicylate to enhance the intensity of immunostaining displayed the characteristic pattern of GARA-ergic neurons and terminals in the fascia dentata as reported in the literature (Ribak et al., '78; Lubbers and Frotscher, '87). When compared to conventional GAD immunocytochemistry, dendritic profiles and their branches could be followed for considerably longer distances with the present enhancement procedure (Fig. 1lA-C). This allowed us to compare GAD-immunoreactive neurons in the inner molecular layer and in the outer granular layer with the Golgi-impregnated chandelier cells described above. In the molecular layer and in the outer granular layer, two types of GAD-immunoreactive cells were differentiated. The most frequently ohserved type was a small neuron, which could be found throughout the depth oft.he molecular layer (Fig. llA,C). Additional large GAD-immunoreactive neurons exhibited well stained dendrites, and were observed only in the inner (commissural) zone of the molecular layer arid i n t,he adjacent part of the granular layer (Fig. 1lA-C; see also Fig. 13A). In the shape and location of the cell bodies as well as in the orientation of dendrites, these cells very much resembled the Golgi-impregnated chandelier cells and were therefore subjected to a comparative fine structural analysis (see below). As in the GAD-immunostained material, large cells immunoreactive for PARV were very similar to Golgi-impregnated chandelier cells (Figs. IlD,E, 14A). Moreover, the number of large PARV-immunoreactive neurons observed in a single section (three to six) was similar to that of large GAD-positive cells in the inner molecular layer and outer granular layer.

Colocalization studies The identical distribution and the morphological similarities between PARV-positive cells and the large GADimmunoreactive neurons in the molecular and outer granular layers suggest that both substances coexist in these cells. T o ascertain this, semithin cryostat serial sections were alternatively stained with GAD or PARV antibodies. Figure 12A illustrates a relatively large number of GAD-immunoreactive cells in all layers of the fascia dentata. An adjacent section stained for PARV (Fig. 12B) displayed labeled neurons preferentially located in the infragranular zone. In the inner molecular layer, a large neuron was found to be immunoreactive for both PARV and GAD (Fig. 12C,D). Colocalization of both antigens was observed to occur in all large immunoreactive cells of the molecular layer and outer gran-

E. SORIANO ET AL.

20 ular layer. In contrast, the small GAD-positive cells of the molecular layer were never stained with PARV antibodies.

EM immunocytochemicalstudies The similarity of large GAD- and PARV-immunoreactive cells in the molecular and outer granular layer with the Golgi-impregnated chandelier cells prompted us to study the fine structure of immunoreactive neurons and terminals. A total of six large immunoreactive neurons were analyzed. Figure 13 shows a large GAD-positive cell in the inner molecular layer at both light and electron microscopic levels. Correlated micrographs of a large PARV-positive cell in the outer granular layer are demonstrated in Figure 14. Both GAD- and PARV-immunopositive cells are described together since their ultrastructural characteristics were very similar, and the coexistence of both antigens was proved. Comparable to gold-toned cell bodies of chandelier cells, large immunoreactive perikarya displayed a nucleus with infoldings and a perinuclear cytoplasm rich in organelles (Figs. 13B, 14B). Numerous terminals covered the surface of the cell bodies establishing both symmetric and asymmetric synaptic contacts (Figs. 13C-E, 14C,D). Perikarya located in the deep molecular layer exhibited a predominance of asymmetric synapses (Fig. 13c).whereas an increased number of symmetric contacts was observed when the cell bodies were situated in the outer granular layer. Often a terminal contacted both the immunoreactive perikaryon and another, unidentified profile (Fig. 13C,D). Figure 15A demonstrates the dense innervation of a dendrite arising from an ident ilied large PARV-immuaoreactive cell. A similar abundance of shaft synapses was observed on gold-toned chandelier celI dendrites. Irnmunoreactive terminals formed symmetric synaptic contacts on similarly immunostained as well as on unlabeled cell bodies and dendrites (Figs. 13E, 14C,D).We regard it as an important result of our EM immunocytochemical studies that numerous GAD- and PARV-immunoreactive boutons were found in symmetric synaptic contact with axon initial segments of granule cells (Fig. 15B.C).

DISCUSSION Axo-axonic chandelier cells are a type of cortical interneuron that is characterized by a distinctive axonal morphology and the formation of synaptic contacts exclusively with the initial segments of pyramidal cell axons (see Peters, '84). Since their discovery (Szentagothai and Arbib, '74; Szentagothai, '75; Jones, '75), they have received considerable attention and have been described in a variety of neocortical areas (see, e.g., Somogyi, '77, '79; Somogyi et al., '82; Fairen and Valverde, '80; Peters et al., '82; Valverde, '83; DeFelipe et al., '85) as well as in the hippocampus proper (Somogyi et al., '83a,b, '85; Soriano and Farinas, '87). In the present study, we provide a correlated light and electron microscopic description of an axo-axonic cell type in the fascia dentata of the rat, which established synaptic contacts almost exclusively with axon initial segments of granule cells. Our Golgi-EM observations also indicate that the location and morphology of this type of axo-axonic chandelier neuron are remarkably constant and clearly different from those reported for the basket cells (Cajal, '11; Lorente de N6, '34; Amaral, '78; Ribak and Seress, '83; Idibbers and Frotscher, '87). In a second set of experiments, we correlated the morphological features of dentate chandelier cells with those of

immunostained neurons in the same location. Our data strongly suggest that large GAD- and PARV-immunoreactive cells of the molecular and outer granular layer and the Golgi-impregnated chandelier neurons belong to the same class of cells. There are several issues that can be discussed on the basis of the present results. Following a critical evaluation of the methodological approach and a comparison of the cells described here with axo-axonic chandelier neurons in other cortical areas, we will discuss general principles of neuronal interconnections in the hippocampus and functional implications of this highly specialized type of dentate interneuron.

Target selectivity of chandelier cells in the fascia dentata Our study has confirmed, once more, that nongranule cells in the fascia dentata form a quite heterogeneous population of neurons. The presently described dentate chandelier cells were first recognized in Golgi material by some peculiar features of their axons. The axonal plexuses of axoaxonic chandelier cells in the fascia dentata, like those of the basket cells, ramify within the granular layer. However, as confirmed by the present Golgi-EM study, there are differences with regard to the postsynaptic element. Basket cells are known to establish synaptic contacts on cell bodies and dendrites (Ribak and Seress, '83; Lubbers and Frotscher, '87). Our semiquantitative analysis of chandelier cells has revealed that 92.3C0 of the postsynaptic targets were unequivocally axon initial segments of granule cells. The percentage of' contacts on initial segments may even be greater if we take into account that 6.3% of the gold-toned boutons made contacts with profiles that were cautiously classified as "unidentified." In most cases, this was due to the lack of unequivocal ultrastructural characteristics. Two gold-toned boutons were observed in contact with a granule cell body. The most likely explanation is that the chandelier cells in the fascia dentata form a few contacts on cell bodies. Alternatively, the two boutons do not belong to the chandelier axon plexus and are from a similarly impregnated axon of another cell. In fact, we did not find an axosomatic contact in any of the other four Golgi-impregnated axonal complexes studied (see Table 1).A high degree of target selectivity of the dentate axo-axonic cells is demonstrated by the axo-axonic contacts in the molecular layer (Fig. LO). Here, the neuropil is mainly composed of profiles other than initial segments, i.e., apical dendrites and dendritic spines of granule cells. The gold-toned boutons made contact with axon initial segments of superficial granule cells instead of contacting the by far more numerous dendritic profiles. Taken together, our results indicate that dentate chandelier neurons have a high degree of target specificity, similar to that reported for chandelier cells in the neocortex (Somogyi, '77, '79; Somogyi et al., '82; Fairen and Valverde, '80; Peters et al., '82; DeFelipe et al., '85) and hippocampus proper (Somogyi et al., '83a,h, '85). Moreover, the axoaxonic neurons of the fascia dentata, like neocortical and hippocampal chandelier cells. establish symmetric synaptic contacts. A few synapses are formed on spines of the axon initial segment. These obvious similarities suggest that neurons with a target selectivity for initial segments are a general, probably essential component of many regions in the cerebral cortex.

DENTATE AXO-AXONIC CELLS

Fig. 15. A: PARV-immunoreactive dendrite in the outer molecular layer arising from the neuron shown in Figure I1D. Several boutons establish asymmetric synaptic contacts (arrows) with this profile. One of them also contacts a characteristic spine (s) of the molecular layer. B,C:

21

Axon terminals immunoreactive for PARV (R) and GAD (C) in the inner granular layer establish symmetric synaptic contacts (arrows) w i t h axon initial segments of granule cells. A , x 26,400; R,C, ~34,000.

E. SORIANO ET AL.

22 When comparing t h e axon plexus of dentate axo-axonic cells with that. of chandelier cells in t h e neocortex and hippocampus proper, a few differences are noted. First, the axonal complexes of t h e dentate chandelier cells form vertically oriented groups of houtons that, are less complex t,han those observed in the neocortex and hippocampus proper of monkeys and cats (see, e.g., Peters, '84; Somogyi e t al., '86). Similarly, t h e synaptic configuration most commonly observed in the fascia dentata was a single gold-toned bouton in contact with a n axon initial segment, whereas neocortical and hippocampal chandelier cells gave rise to long strings of impregnated boutons forming multiple contacts with the same postsynaptic element. These diff'erences may have to do with t h e generally simpler organization of the fascia dentata, where the principal neurons form a single, densely packed cell layer. Our electron microscopic analysis often revealed axon initial segments that received synapses from gold-t.oned boutons and from other nonimpregnated terminals. T h e origin of these unimpregnated boutons could not b e determined. However, since no other source of axo-axonic contacts has heen reported in the fascia dentata, it is likely that these terminals belong t o other nonimpregnated dentate chandelier cells. This suggests convergence of several axo-axonic interneurons on t h e same granule cell axon. A similar convergence has been proposed for t h e chandelier cells of t h e neocortex and hippocampus proper (see Peters, '84). It cannot be ruled out t h a t other types of nongranule cells contribute 10 the synapses on axon initial segments. In the neocortex, for example, various t,ypes of interneurons that, are known to terminate preferentially on other postsynaptic elements also establish occasional contacts with axon initial segments of pyramidal cells (see, e.g., Peters and Fairen, '78; Kisvarday e t al., '85). T h e dentate chandelier cell forms a dense local axon plexus in the granular layer. Thus, a large number of granule cells may he contacted by a single axo-axonic chandelier cell. Moreover, we cannot be sure t h a t t h e complete axon plexus was impregnated in our Golgi p r e p a r a h n s . I t might he even niorr complex. However, there is no reason to assume that, these cells project outside t h e hippocampus. Molecular layer neurons were never found retrogradely labeled in tracer studies of hippocampal connections. It has been stated, on t h e other hand, that nonpyramidal cells in the hilar region give rise t o long projections (Bakst e t al., '86; Ribak e t al.. '86; Schwerdtfeger and Buhl, '86; Leranth and Frotscher, '87).

Chandelier cells in the fascia dentata are inhibitory GABA-ergicneurons Studies by F'reund et al. ('83) and Somogyi e t al. ('85) have provided evidence that. chandelier cells in the neocortex and hippocampus proper use GABA as a transmitter. By applying postembedding immunocytochemistry to semithin sect,ions of t h e cell body region of gold-toned-identified dentate chandelier cells, we recently demonst.rat.ed t h a t these neurons contain GARA (Soriano and Frotscher, '89). T h a t dentate axo-axonic cells were GABA-ergic could already b e anticipated from studies b y Kosaka e t al. ('84), who demonstrated that synapses on axon initial segments of granule cells react.ed for GAD, the GABA-synthesizing enzyme. How,ever, these authors could not identify the parent cell and concluded t h a t t h e synapses on axon initial segments derived from t h e various types of basket cells in t h e fascia dent,ata. T h e exist,ence of a specific cell type innervating the

initial segments of granule cell axons was proposed by Ribak and Seress ('83),who never found a basket cell terminal in conlact with this particular postsynaptic element. T h e present study and our parallel combined Golgi and GABAirnmunocyt.oc)iemical analysis have clearly demonstrated that t,here is a specific GABA-ergic cell type, different from t h e basket, cells, t h a t forms inhibitory synapses on initial segments of granule cell axoris. It is regularly located in t h e innermost portion of the dentate molecular layer o r in the out,er zone of t.he granular layer. In t h e present irnmunocytochemical studies, we found GAD-immunoreactive neurons i n t h e inner zone of the molecular layer t h a t were of similar size and exhibited structural characteristics similar t o t h e gold-toned chandelier cells. We have reason to believe, in fact, t.hat these large GAD-positive neurons in t h e inner molecular layer are t h e chandelier cells. Other GABA-ergic neurons in the rnolecular layer are considerably smaller. T h e remaining cell types described so far for the molecular layer are not GABA-ergic. i.e., ectopic granule cells (Cajal, '11) and small, choline acetylt,ransferase (ChAT)-positive cells (Frotscher e t al., '86). T h e large GAD-positive, supposedly axo-axonic, neurons in the inner molecular layer are also immunoreactive for PARV as demonstrated in onr colocalization e x p e r i m e n k Coexistence of GAD and PARV in the same neurons has similarly been found in the hippocampus by Kosaka e t al. ('87). T h a t PARV is contained in axo-axonic cells in t h e neocortex and hippocampus has recent,ly been demonstrated (DeFelipe e t al., '89: Katsuniaru e t a].?'88). Our E M immunocytochemical studies have accordingly shown t h a t both GAD- and PAIZV-immunoreactive boutons formed symmetric synapt,ic contacts with axon initial segments of granule cells.

Functional integration of dentate chandelier cells in hippocampal circuits In this paragraph we will try to define t h e significance of t h e presently described dentate chandelier cell wit,h regard to its relation to known hippocampal circuitries. As a starting point. we will focus on the available information about afferent systems capable of driving this particular neuron. T h e dentate chandelier cell extends long dendrites far out into the molecular layer of the fascia dentata. Its cell body is located close to or within the granular layer. Hence, all major afferent systems known to terminate in t h e molecular layer in a laminated fashion could reach either the dendrites or the cell body of this neuron. In fact, numerous terminals were l'ound o n the dendritic shafts as well as on t h e cell bodies, but also on the few dendritic spines of chandelier cells. Most afferents in the molecular layer are likely to activate this neuron because they are known t o be excitatory. This holds true. for instance, for the major input to the molecular layer, i.e., the aff'erents from the entorhinal cortex (Andersen e t al., '71). Our observations of a large number of asymtnet,ric synapses on chandelier cell dendrites extending far out into the molecular layer may be regarded as a morphological correlate. Symmetric, probably inhibitory, synaptic contacts were also observed and predominated on t h e cell body and on proximal dendrites. In fact,, in our immunocytochemical study, we found labeled boutons impinging on the large GAD- and PARV-immunoreactive cells. These t.erminals may inhibit the inhibitory axo-axonic cells resulting in disinhibition of granule cells. T h a t t h e same afTerent.s that innervate the granule cells synapse on dentate axo-axonic cells is supported by the observation of

23

DENTATE AXO-AXONIC CELLS terminals simultaneously conhcting the identified chandelier cell dendrites as well as nonimpregnated spines, most likely arising from granule cells (see Figs. 6A, 15A). AS was mentioned in Results, dentate chandelier cells do not exhibit a well developed basal dendritic tree. However, the extension of basal dendrites into the hilar region suggests a n innervation by recurrent, mossy fiber collaterals. Mossy fiber synapses directly tinderneath the granular layer are difficult to identify because they are smaller than their counterparts in the CA3 region (Blackstad, '63, '67). Only a few dendritic tips reach the zone where the first mossy fiber collaterals appear suggesting t,hat the chandelier cells located above the granular layer are less densely innervated by mossy fiber collaterals than the large basket cells mainly located underneath the granular layer with dendrites extending far into the hilar region. I t is thus tempting to speculate t,hat the GABA-ergic inhibitory chandelier cells are mainly involved in feed-forward inhibition of the granule cells. Inhibition of the granule cells appears to be very efficient because the chandelier cells contact the axon initial segments where the action potential is generated. The presently described axo-axonic chandelier cell may thus interfere with the impulse flow within the main excitatory pathway of the hippocampal lormation. This trisynaptic pathway originates in the entmhinal cortex and connects, via t.he mossy fibers, the fascia dentata with the hippocampus proper. In addition to driving the granule cells, the ent,nrhinal input may also activate the chandelier neurons, which then block transmission to the hippocampus by inhihiting a large number of granule cells. Together with t.he basket cells, chandelier neurons may thus protect the hippocampus from an overexcitation within the trisynaptic pathway. Overexcitation caused by st.imulation of entorhinal fibers result,s in structural changes in hippocampal neurons similar to those observed in epileptic brain damage (Sloviter, '83). U'e hypothesize that pathological changes in chandelier axo-axonic cells are among the factors that may contribute to the genesis o f epileptic processes in the hippocampus. This view is support,ed hy EM studies on epileptic foci in the neocortex, where a significant loss of inhibitory synapses on axon initial segments was found (Ribak, '85). A description of highly specialized axo-axonic cells raises general questions of neuronal specificity. I t appears that there are various degrees of neuronal specificity within the hippocampal formation. It is well known, for instance, that the various extrinsic alrerents of the hippocampus proper and fascia dentata not only terminate on the principal neurons, but contact other cells as well. So it has been shown that, cornmissural fibers from the contralateral hippocampus, in addition t o contacting pyramidal neurons and granule cells, also establish synapses on smooth dendrites of basket cells in CAl and in the fascia dentata (Frotscher and Zimmer, '83; Seress and Rihak, '84). Similarly, cholinergic fibers from the medial septum were found to synapse on pyramidal neurons, granule cells, and GABA-ergic inhibitory neurons (Frotscher and Leranth, '85, '86; Leranth and Frotscher, '87; Frotscher e t al., '89). We were recent,ly able to show that also the afferents from the entorhinal cortex synapse both on spines of granule cells and on smooth dendrites of PARV~immuriorcact.iveneurons in the fascia dentata (Zipp et al., '89). Ascending catecholaminergic brainstem afferents, as do the other afferent systems, terminate on pyramidal cells and GABA-ergic nonpyramidal neurons (Frotscher and Leranth, '88).

A somewhat higher degree of neuronal specificity is observed in the case of the intrinsic mossy fibers. I t is well known that granule cell axons estimate the border between CA:I arid CAI. Large mo fiber synapses are observed exclusively on '2.43 pyramidal cells. However, we were recently able t,o demonst,rate that, mossy fibers established synapses on nonpyramidal cell dendrites traversing stratum lucidum of t,he CA3 region (Frotxher, '85, '89). A certain neuronal specificity is also noted in the case of the basket cells. As was mentioned earlier, they form synaptic contacts with cell bodies and proximal dendrites and seem to avoid the a x o n initial segment.. Another example are GABA-ergic npiirons in the septum projecting to the hippocampal formation. Frcund and Antal ('88) have recently found that the terminals of the septa1 GABA-ergic neurons almost exclusively contacted cell bodies and dendrites of GABA-ergic neurons in the hippocampus, thereby serving disinhibition of the hippocampal principal cells. A high degree of specificity is seen in the case of the dentate axo-axonic cells. This specificity applies not only to a specific cell type (the granule cells) but also to a specific part o f the neuronal membrane (the axon initial segment). Even in the molecular layer, where dendrites and spines hy far outnumber other elements of the neuropil, terminals of chandelier cells contact the few axon initial segments of ectopic granule cells and avoid the other available structures. I t is a challenge to neurohiologists to unveil the factors that det,ermiiie the different degrees of neuronal specificity briefly mentioned here.

ACKNOWLEDGMENTS The authors thank E. Thielen and B. Krebs for technical assistance and 1. Szasz for the drawings. This work was supported by grants from the Deutsche Forschungsgemeinschaft (SFR 45, Fr 620/1-4) and the Ministerio de Educa( i o n y (:iencia (Spain) (CIRIT EE88-1 and FISSS 88-1111).

LITERATURE CITED Amaral. D.G. (1978) A Golgi study of cell types in the hilar region of the hippocampus in the rat. d. Comp. Neural. 182:851-914. Andersen, P. (1975) Organization of hippocampal neurons and their interconncctions. In K.L. Isaacson and K.H. Pribram (eds): The Hippocampus, V d . I. Structure and Development. New York, London: Plenum, pp. 155175.

Andersen, P., T.V.P. Bliss, and K.K. Skrede (1971) Lamellar organization of hippocampal excitatory pathways. Exp. Brain Res. 18222-238. Raksl, I., C. Avendano, .J.H. Morrison. and D.G. Amaral (1986) An experimental analysis of the origins of somatostatin-like immunoreactivity in the dentate gyrus of the rat. J . Neurosci. 6:14:12-1462. Blackstad, T.W. (1963) Ultrastructural studies on the hippocampal region. Rlackstad, T.W. (1967) Cortical grey matter. A correlation of light and electron microscopic data. In H. Hyden (ed!: The Neuron. Amsterdam: Elsevier, pp. 49-118. Rmitenberg, V.. V. Guglielmotti, and E. Sada (1967) Correlation of crystal growth with the staining of axons by the Golgi procedure. Stain Technol. -12.277-283. C ; I J ~S.K. , y (1911) Histologie du s y s t h e nerveux de l'hornrnc et des vertehres, vol 11. Paris: Maloine. ( k l i i i . M.R., W. Baier, C. Scharer, P.A. De Viragh, and C. Gerday (1988) Monoclonal antibodies directed against the calcium-binding protein parvalbumin. Cell Calcium 9t81-86. Colonnicr, M. (1964)The tangential organization of the visual cortex. J. Anat. 9x::127-344. DeFelipe, J., S.H.C. Hendry, and E.G. Jones (1989) Visualization of chandelier cell axons by parvalhumin immunoreactivity in monkey cerebral cor-

E. SORIANO ET AL.

24 tex. Proc. Nal. Sci. USA 862093-2097. DeFelipe, J.; S.H.C. Heridry, E.G. Jones, and D. Schmechel(1985) Variability in the terminations of GABAergic chandelier cell axons in the monkey sensory-motor cortex. d. Comp. Neurol. 231t3fi4-384. Fairen, A., J. DeFelipe, and R. Martinez-Ruiz (1981) The Golgi-EM procedure: A tool to study neocortical interneurons. In S. Yedoroff (ed): “ E l ~ v enth International Congress of Anatomy, Part A Glial and Neuronal Cell Biology.” New York: Alan R. Liss, Inc., pp. 291-301. Fairen, A., A. Peters. and .I. Saldanha (1977) A new procedure for examining Golgi impregnated neurons by light and electron microscopy. d. Neurocytol. 631 1 337. Fairen, A., and F. Valverde (1980) A specialized type of neuron in the visual cnrt.ex of cat. A Gulgi and electron microscopic study of chandelier reils. d. Comp. Neurol. I94)4:761-779. Freund, T.F., and M. Antal (1988) GABA-containing neurons in the septum control inhibitory interneurons in the hippocampus. Nature 336t170173. Freund, T.F., K.A.C. Martin, A.D. Smith, and I? Sonrogyi (1983) Glulamate decarboxvlase-immunoreactive terminals of Golgi-impregnated axoaxonic cells and of presumed basket cells in synaptic contact with pyramidal neurons of the cat’s visual cortex. .I. Comp. Neurol. 221 -263-278. Frotscher, M. (1985) Mossy fibres form synapses with identified pyramidal basket cells in the CA3 region of the guinea-pig hippocampus: A rombined Golgi-electron microscope study. d. Neurorytol. l4.245-259. Frotscher, M. (1989) Mossy fiher synapses on glntamate decarhoxylaseimmunoreactive neurons: Evidence for feed-forward inhibition in the CA3 region of the hippocampus. Exp. Brain Res. 7,5:441-445. Frotscher, M., P. Kugler, U. Misgeld, and K. Zilles (1988) Neurotransmissirin in the Hippocampus. Adv. Anat. Embryol. Cell Biol., Vol. 111, Berlin: Springer-Verlag. Frotscher. M., and C. Leranth (1985) Cholinergic innervation of the rat h i p pocampus as revealed by choline acetyltransferase immunocytochemistry: A combined light and electron microscopic study. J. Comp. Neurol. 299:237-246. Frotscher: M., and C. Leranth (1986) The cholinergic innervation of the rat fascia dentata: Identification of target structures on granule cells by combining choline acetyltransferase immunocytochemistry and (Mgi impregnation. J. Comp. Neurol. 24.7.58-70. Frotscher. M.. and C. Leranth (1988) Cat,echolaminergic innervation of pyramidal and GARAergic nonpyramidal neurons in the rat hippocampu>. Double label immunostaining with antibodies against tyrosine hydroxylase and glutamate decarboxylase. Histochemistry 88:313-319. Frotscher. M., C. Leranth. K. Lubbers. and W.H. Oertel (1984) Commissural afferents innervate glutamate decarboxylase immunoreactive nonpyramidal neurons in the guinea pig hippocampus. Neurosci. Lett. 46.137-143.

Frolscher, M., R. Nitsrh, and C . Leranth (1989) Cholinergic innervation of identified neurons in the hippocampus: Electron microscopic double laheling studies. In V. Chan-Palay and C. Kohler (eds): The Hippocampus-New Vistas. New York: Alan R. Liss, Inc., pp. 85-96. M. Schlander, and C. Leranth (1986) Cholinergic neurons in Frolscher, M., the hippocampus: A combined light and electron microscopic immunocytochemical study in the rat. Cell Tissue Res. 246:293-301. Frotscher, h4.. and J. Zimmer (1983) Commissural fibers terminate on nonpyramidal neurons in the guinea pig hippocampus-A combined Golgi/ EM degeneration study. Brain Kes. 265:289-293. Hsu, S.M., L. Raine, and H. Fanger (1981) The use of avidin~biotin-peroxi~ dase complex (ABC) in immunoperoxidase techniqncs: A comparison between ABC and unlabeled antibody (peroxidase) procedurcs. J. Histochem. Cytochem. 29r577-590. Jones, E.G. (1975) Varieties and distribution of non-pyramidal cells in the somatic sensory cortex of the squirrel monkey. J. Comp. Neurol. 160.205268. Katsumaru, H., T. Kosaka, C.W. Heizmann, and K. Hama (1988) Immunocytochemical study of GABAergic neurons containing the calcium-binding protein parvalbumin in the rat hipporampus. Exp. Brain Kes. 72,347362. Kisvarday, Z.F., K.A.C. Martin, D. Whitteridge, and P. Somogyi (1985) Synaptic connections of intracellularly filled clukh cells: A type of sniall basket cell in the visual cortex of the rat. d. Comp. Neurol. 241111- 137. Kosaka, T. (1983) Axon initial segments of the granule cells in the r a t dentate gyrus: Synaptic contacts on bundles of axon initial segments. Brain Res. 274.129-134. Kosaka, T., K. Hama, and J - Y . Wu (1984) GABAergic synaptic boutons in t.he granule cell layer of the rat dentate gyrus. Brain Res. 293353-359.

Kosaka; T., H. Katsumaru. K. Hama, J:Y. Wu, and C.W. Heizmann (1987) CABAergic neurons containing the Ca’ ’ -binding protein parvalbumin in the rat hippocampus and dentate gyrns. Brain Res. 419:119-130. Leranth, C., and M. Frotscher (1987) Cholinergic innervation of hipporampal GAD and somatostatin-immunoreactive commissural neurons. d. Comp. Neurul. 267:3:1-47. Lurente de N6, R. 11934) Studies on the structure of the cerebral cortex. 11. Continuation of the study of the ammonic system. J. Psychol. Neurol. 46:113-177. Lubbers, K.. and M. Frntscher (1987) Fine structure and synaptic connectiuns of identified neurons in the rat fascia dentata. Anat. Embryol. 177.1 -14. Marti-Suhirana, A., E. Soriano, and J.M. Garcia-Verdugo (1986) Morphoiogical aspects of the ectopic granule-like cellular populations in the albino rat hippocampal formation: A Golgi study. J. Anat. 144:31-47. Mugnaini, E., and A.L. Dahl (1983) Zinc-aldehyde fixation for light microsropir imrnunorytuchemistry for nervous tissues. J. Histochem. Cytochem. 31:1433-1438. Oerkl. W.H., D.E. Schmechel, E. Mugnaini, M.L. Tappaz, and I.J. Kopin i19R2) Immunocytochemical localization of glutamate decarboxylase in the rat cerebellum with a new antiserum. Neuroscience 6t2715-2735. Peters, A. 1 1 984) Chandelier cells. In A. Peters and E.G. Jones (eds): Cerebral Cortex, Vnl. 1. New Y o r k Plenum Press, pp. 361-380. Peters. A., and A. Fairen (1978) Smooth and sparsely-spined stellate cells in the visual cortex o f the rat: A study using a combined Golgi-electron microscope terhnique. J. Comp. Neurol. IRI:129-172. Peters, A,, C.C. Proskauer, and C.E. Rihak (198“) Chandelier cells in rat visual cortex. J. Comp. Neurol. 206397-416. Ribak. C.E. (1985) Axon terminals of GABAergic chandelier cells are lost at epileptic foci. Brain Kes. 326:251-260. R i h k , C.E.. and I,. Anderson (1980) Lltrastructure of the pyramidal basket rells in the dentate gyrus of the rat. J. Comp. Neurol. 192:903-916. Ribak, C.E., and L. Seress (1983) Five types of basket cell in the hippocampal dentate gyrus: A combined Golgi and electron microscopic study. J. Neurocytol. 12.577 59:. Ribak. C E. , L. Seress, G.M. Peterson, K.B. Seroogy, d.H. Fallon, and L.C. Srhrniied (1986) A GARAergic inhibitory romponent within the hippor pathway. J. Neurosci. 6-3492-3498. i’ampal ~ : o m rissural Ribak. C.E.. d.E. Vaughn, and K. Saito (19781 Irnmunocytochemical localization of glutamic acid decarboxylase in neuronal somata following colchicine inhibition of axonal transport. Brain Res. 140:315-332. Schlander. M., and M. Frotscher (19861 Non-pyramidal neurons in the guinea pig hippocampus. A combined Golgi-electron microscope study. Anat. Embryol. 17435-47. Schwerdtfcger, W.K., and E. Buhl (1986) Various types of non-pyramidal hipporampal neurons project to the septum and contralateral hippocampus. Brain Res. 386:146-154. Seress. L., and C.E. Ribak (1984) Direct commissural connections to the basket cells of thc hippocampal dentate gyrus: Anatomical evidence for feedforward inhibition. .J. Neurocytol. 13216-225. Seress, I.., and C.E. Rihak (1 985) A combined Golgi-electron microscopic study of nor1 pyramidal neurons in CAI area of the hippocampus. J. Neurocytol. 14.717-730.

Sloviter. R.S. (1983) “Epiieptic” brain damage in rats induced by sustained electrical stimulation of the perforant path. I. Acute electrophysiological and light microscopic study. Brain Res. Bull. 10:675-697. SomoEyi, P. (1977) A specific “axo-axonal” interncuron in the visual cortex of the rat. Brain Res. 136.345-350. Somogyi, P. (1379) An interneurone making synapses specifically on the axon initial segment of pyramidal cells in the cerebral cortex of the cat. J. Physiol. 2.%:18P-l9P. Sornogyi, P., T.F. Freund, and A. Cowey (1982) The axo-axonic interneuron in the cerebral cort.ex of the rat, cat and monkey. Neuroscience 735772607. Sornogyi, F’., T.F. Frcund, N. Halasz, and Z.F. Kisvarday (1981) Selectivity of neuronal (‘H)-GABA accumulation in the visual cortex as revealed by Golgi staining of (.he labeled neurons. Brain Res. 225:431-436. Somogyi, P., T.F. Freund, A.J. Hodgson, J. Somogyi, D. Beroukas, and I.W. Chubb (1YX3) Identified axo-axonic cells are immunoreactive for GABA in the hippocampus and visual cortex of the cat. Brain Res. 332143-149. Soniugyi, P., M l ; . Nunzi, A. Gorio, and A.D. Smith (1983a) A new type of specific int,erneurnn on the monkey hippocampus forming synapses exclrisively with the axon initial segments of pyramidal cells. Brain Res. 259.1:37-142.

DENTATE AXO-AXOKIC CELLS Somogyi, P., A.D. Smith, M.G. Nunzi, A. Gorio, H. Takagi, and