Previous immunocytochemical studies in the cerebral cortex of various species have shown that the calcium-binding protein calretinin (CR) labels specific subpopulations of nonspiny nonpyramidal cells (interneurons). The present study attempts to characterize morphologically and chemically the microcircuitry of CR-immunoreactive (CR-ir) neurons in the human temporal neocortex. Postembedding immunocytochemistry for CR and GABA and combination immunocytochemistry for CR and nonphosphorylated neurofilament protein (NPNFP) or for CR and the calcium-binding proteins parvalbumin (PV) and calbindin (CB) showed CR multiterminal endings frequently innervating the distal apical dendrite or the cell body and proximal dendrites of NPNFP-ir or CB-ir pyramidal cells, respectively. Cell bodies of interneurons immunoreactive for CB or PV were innervated only occasionally by CR multiterminal endings, whereas certain GABA neurons were surrounded by them. Furthermore, CR-ir axon terminals formed either symmetrical (the majority) or asymmetrical synapses with a variety of postsynaptic elements. These results indicate that different subpopulations of CR interneurons exist that are specialized for selective innervation of somatic or dendritic regions of certain pyramidal and nonpyramidal neurons.
- cortical circuitry
- pyramidal cells
- nonpyramidal cells
- calcium-binding proteins
- neurofilament protein
Neurons with local axons (interneurons) in the neocortex can be subdivided into two major groups: spiny and smooth nonpyramidal cells. Spiny nonpyramidal cells constitute a relatively small group of neurons located in the middle layers and are thought to be excitatory. Smooth nonpyramidal cells represent the largest group of interneurons and are thought to be inhibitory GABAergic cells and the main components of inhibitory neocortical circuits, which control the activity of projecting cells (pyramidal cells) (Houser et al., 1984;Hendry, 1987; Peters, 1987; Somogyi, 1989; White, 1989; Lund, 1990;Jones, 1993).
Immunocytochemical studies using antibodies against a variety of chemical compounds have revealed that certain subpopulations of smooth nonpyramidal cells display particular neurochemical characteristics (DeFelipe, 1993). Immunocytochemistry for the calcium-binding proteins parvalbumin (PV), calbindin D-28k (CB), and calretinin (CR) has been especially useful for examining the synaptic connections or chemical characteristics of certain types of neurons (DeFelipe et al., 1989;Hendry et al., 1989; Lewis and Lund, 1990; Williams et al., 1992;Kawaguchi and Kubota, 1993; del Rı́o and DeFelipe, 1994, 1995,1997; Kawaguchi, 1995). Recent colocalization experiments have shown that in the human neocortex a small, but significant, population of CR-immunoreactive (CR-ir) neurons is non-GABAergic (del Rı́o and DeFelipe, 1996a). Although the function of CR is still unknown (Andressen et al., 1993), this finding is of interest because this protein is localized in smooth nonpyramidal cells (Jacobowitz and Winsky, 1991; Glezer et al., 1992; Résibois and Rogers, 1992; Hof et al., 1993; Condé et al., 1994; Fonseca and Soriano, 1995; del Rı́o and DeFelipe, 1996a). These colocalization experiments suggest that CR immunocytochemistry labels two subpopulations of smooth nonpyramidal cells, one that consists of GABAergic neurons (the largest subpopulation) and the other of non-GABAergic neurons. This is in line with the observation that some CR-ir neurons are bipolar cells (Condé et al., 1994; del Rı́o and DeFelipe, 1996a), whereas in other studies certain bipolar cells have been found to make asymmetrical synapses (Peters and Kimerer, 1981; Fairén et al., 1984; Peters and Harriman, 1988). Because GABA neurons establish exclusively symmetrical synapses (Ribak, 1978; Houser et al., 1984), the latter bipolar cells must be non-GABAergic. However, there are no studies on synaptic connections of CR-ir neurons in the human neocortex. The aim of the present work was to characterize morphologically and chemically the microcircuitry of CR-ir neurons in the human temporal neocortex by using electron microscopy and combination immunocytochemistry for CR and nonphosphorylated neurofilament protein (NPNFP), for CR and PV, or for CR and CB. Furthermore, postembedding immunocytochemistry for CR and GABA also was performed. Immunocytochemistry for NPNFP was used because of the excellent staining of the dendritic tree of subpopulations of pyramidal cells (Campbell and Morrison, 1989), which permitted the study of possible synaptic relationships of CR-ir axons with these pyramidal cells.
Preliminary results of this investigation have been published previously (del Rı́o and DeFelipe, 1996b).
MATERIALS AND METHODS
Human cerebral cortical tissue considered to be normal (see Results) was examined from the anterolateral superior, middle, and inferior temporal gyri (Brodmann’s area 38, 21, and 20, respectively). This tissue was removed inevitably during surgical treatment of eight patients (range 19–44 years old, mean 32.25 years old) with pharmaco-resistant temporal lobe epilepsy. For all patients, informed consent was obtained before surgery. Some of this material has been used in a previous study (Marco et al., 1996). All tissue samples were immersed immediately in cold 4% paraformaldehyde in 0.1 mphosphate buffer (PB), pH 7.4, for 2–3 hr and then cut into small blocks and post-fixed in the same fixative, except for some blocks that were fixed in a mixture of 2% paraformaldehyde and 0.2% glutaraldehyde for 24 hr at 4°C. The blocks were cut at 100 μm on a Vibratome, and the sections were pretreated with a solution of ethanol and hydrogen peroxide in PB to remove endogenous peroxidase activity. Sections from tissue fixed in 4% paraformaldehyde were processed for preembedding light and electron microscope immunocytochemistry and double immunocytochemical staining, while those sections from tissue fixed in 2% paraformaldehyde and 0.2% glutaraldehyde were processed for postembedding immunocytochemistry. Adjacent sections were stained with thionine.
Preembedding light and electron microscope immunocytochemistry.Sections were washed in PB, preincubated in 3% normal goat serum in PB either with or without Triton X-100 (0.05%) for 3 hr at room temperature, and incubated for 24 hr at 4°C in rabbit anti-CR (Swant, Bellinzona, Switzerland) diluted 1:2000 in PB containing 3% normal goat serum either with or without Triton X-100 (0.05%). Then the sections were washed in PB and processed by the avidin–biotin–peroxidase method with the Vectastain ABC immunoperoxidase kit (Vector Laboratories, Burlingame, CA). Afterward, the sections were washed in PB and reacted histochemically with 0.05% 3,3′-diaminobenzidine tetrahydrochloride (DAB) and 0.01% hydrogen peroxide, washed again, and osmicated either in 0.02% osmium tetroxide for 1 min or in 1% osmium tetroxide in PB for 40 min. Sections treated with Triton X-100 were osmicated in the first osmium solution, mounted onto glass slides, dehydrated, cleared with xylene, and coverslipped. These sections were used for light microcopy alone. The sections not treated with Triton X-100 were post-fixed in 1% glutaraldehyde in PB for 1 hr, osmicated in 1% osmium tetroxide, dehydrated, and flat-embedded in Araldite resin. These plastic-embedded sections were examined by using a correlative light and electron microscopic method described in detail elsewhere (DeFelipe and Fairén, 1993). Most of the ultrathin sections were stained with uranyl acetate and lead citrate, whereas the others were examined unstained in a Jeol-1200 EX electron microscope. Neuronal size was measured directly from light microscope slides by using an eyepiece reticle.
Double immunocytochemical staining. Two methods were used for double immunocytochemical staining. The sections processed with the first method were examined by conventional light microscopy, whereas those with the second method were analyzed by confocal laser icroscopy.
For conventional light microscopy, dual immunostaining for two substances was achieved by using sequential immunocytochemistry for the first substance and then for the second substance, first with DAB–nickel and then with DAB as substrates to visualize the bound immunoglobulin–peroxidase complexes in black and brown, respectively (Hancock, 1982). Briefly, sections first were stained immunocytochemically for CR, as above, except that Triton X-100 (0.2%) was added in the preincubation and first incubation steps and that a 0.05% solution of DAB containing 2.5% nickel ammonium sulfate and 0.01% hydrogen peroxide was used to visualize immunoreactive elements. Afterward, the sections were washed and reprocessed immunocytochemically as follows: for NPNFP, using the mouse monoclonal antibody SMI 32 (Sternberger Monoclonals, Baltimore, MD) diluted 1: 5000; for CB, using rabbit anti-CB (Swant) diluted 1:2000; or for PV, using mouse monoclonal anti-PV (Swant) diluted 1:2000. The immunocytochemical procedure was the same as that described for CR, using the appropriate species-specific Vectastain ABC immunoperoxidase kits (Vector Laboratories), but the sections were reacted histochemically with 0.05% DAB and 0.01% hydrogen peroxide. Finally, the sections were dehydrated, cleared with xylene, and coverslipped.
For confocal laser microscopy, sections were double-stained only for CR and CB by the following technique. The sections were incubated for 24 hr at 4°C in a solution containing the same rabbit polyclonal anti-CR antibody and mouse monoclonal anti-CB antibody, as indicated above, diluted 1:2000. Then the sections were incubated for 1 hr at room temperature in a solution containing biotinylated goat anti-rabbit IgG (Vector Laboratories) diluted 1:200. Thereafter, the sections were washed in PB and incubated for 1 hr at room temperature in a mixture of Cy5-conjugated goat anti-mouse IgG and Cy2-conjugated streptoavidin (Amersham, Arlington Heights, IL) diluted 1:1000. Finally, the sections were mounted, coverslipped in a 1:3 solution of PB and glycerol, and examined in a Leica TCS 4D confocal laser scanning microscope equipped with an argon/krypton-mixed gas laser with excitation peaks at 489 nm (for Cy2-labeled profiles) and 649 nm (for Cy5-labeled profiles). The confocal microscope was associated with a Leitz DMIRB fluorescence microscope. Fluorescence of Cy2-labeled profiles and Cy5-labeled profiles was recorded through separate channels.
Postembedding immunocytochemistry. Vibratome sections from tissue fixed in 2% paraformaldehyde and 0.2% glutaraldehyde were post-fixed in 1% osmium tetroxide for 30 min, dehydrated, and flat-embedded in Araldite resin. Serial 1-μm-thick plastic (semithin) sections were cut with a Reichert ultramicrotome and processed by the postembedding immunocytochemical staining protocol for CR and GABA described in del Rı́o and DeFelipe (1996a). The anti-CR was the same as above, and the anti-GABA was a rabbit polyclonal antiserum (Sigma, St. Louis, MO).
Control sections for immunocytochemistry were processed as above, but with the primary antibody replaced with normal serum and, additionally in the case of CR, CB, and PV immunocytochemistry, with primary antibody adsorbed with an excess of CR, CB, and PV proteins (Swant), as indicated by the manufacturer or, in the case of GABA immunocytochemistry, with primary antiserum adsorbed with an excess of GABA albumin conjugate. No significant staining was observed under these control conditions.
The human temporal neocortex used in the present study was considered to be normal, a conclusion based on intraoperative electrocorticography and routine histopathological examinations and other immunocytochemical studies done on the resected tissue (see Marco et al., 1996). Furthermore, the normal appearance of the cytoarchitecture and the patterns of immunostaining observed in the present study were very similar to those observed in tissue sections from normal temporal neocortex that had been removed from a 34-year-old nonepileptic patient to gain access to a tumor located near the hippocampus (del Rı́o and DeFelipe, 1997). Thus, the general observations and conclusions on the connectivity of CR-ir neurons described below are likely to be found also in strictly normal human temporal cortex. Unless otherwise specified, the following description refers to observations made on material examined with conventional light microscopy.
CR-ir neurons comprised a population of nonpyramidal cells that were distributed from layers I to VI and in the subjacent white matter but that were most numerous in layers II and III. They had small somata (diameter 8–12 μm) and showed a variety of shapes (round, fusiform, bipolar, multipolar) (Fig. 1 A). As described in our earlier study (del Rı́o and DeFelipe, 1996a), the most typical morphology was the bipolar form (Feldman and Peters, 1978), which was distinguished by a round or fusiform soma, which furnished primary dendrites from the upper and lower poles, and by a long, narrow, and vertically orientated dendritic tree (Fig.1 B). Three kinds of immunoreactive elements were found in the neuropil: terminal-like puncta and axonal and dendritic processes. Axonal processes were very thin and very finely beaded with varicosities of 0.5–1 μm in diameter and were distributed nonuniformly. Dendritic processes were either thick and nonbeaded or were thin and coarsely beaded with round or ovoid varicosities; they were distributed rather uniformly (Fig. 1 B). Terminal-like puncta also were found frequently around unstained neurons, mainly in layers II and III. However, few or no CR-ir axons were observed contacting other CR-ir neurons, but close appositions between dendrites and between dendrites and somata of pairs of CR-ir neurons were found relatively often (Fig. 1 D).
Double immunocytochemical staining
The double immunocytochemical staining method used for conventional light microscopy allowed the visualization of two sets of immunoreactive elements in the same section by using DAB–nickel and DAB as substrates for the differential staining of immunoreactive elements. With DAB–nickel the immunoreactive neurons and processes were stained an intense black, whereas with DAB the immunoreactive elements were stained brown (Figs. 1 C,2 A,B). After double staining there was a decrease in intensity of immunostaining, particularly of the immunoreactive elements visualized with DAB. Because the primary objective of dual immunostaining was to study the relationship between CR-ir axons and neurons immunoreactive for NPNFP, CB, and PV, in all combinations CR immunocytochemistry was used first (with DAB–nickel), and afterward immunocytochemistry for the other substances (with DAB) was used.
The patterns of immunostaining for CR, NPNFP, CB, and PV in the human temporal neocortex have been described previously in other studies (Campbell and Morrison, 1989; Ferrer et al., 1991, 1992; Hof et al., 1991, 1993; Hayes and Lewis, 1992; del Rı́o and DeFelipe, 1994;Fonseca and Soriano, 1995). Therefore, only a brief description of immunoreactive neurons will follow.
Immunocytochemistry for NPNFP, CB, and PV labeled the cell body, axon (except for NPNFP, which labeled only the axon initial segment, but not preterminal or terminal axons), and proximal dendrites of a variety of neurons. NPNFP immunocytochemistry labeled numerous pyramidal cells (Figs. 1 C, 2 A) located in layers II through VI, but layers III, V, and VI contained the greatest number of stained cells. The size of the cell body and intensity of staining were variable, the largest and strongest-labeled pyramidal cells being located in the deep half of layer III and the superficial part of layer V. The apical dendrites of many labeled pyramidal cells could be followed for several tens or even hundreds of microns (Figs.1 C, 2 A, 3 A–D). CB-ir neurons consisted of both pyramidal (Fig. 3 E) and nonpyramidal cells (Fig. 3 F). In general, pyramidal cells were lightly labeled and located mainly in layer III, whereas nonpyramidal cells were darkly stained and found in all layers, although the majority was found in layers II and III. The shape and size of nonpyramidal cells were variable, but the most common morphology was multipolar with a rounded soma of ∼10 μm in diameter. PV immunocytochemistry labeled nonpyramidal cells located in layers II through VI. Most immunoreactive cells were darkly stained and showed a variety of forms, the multipolar morphology showing an ovoid soma of 10–15 μm in diameter being one of the most common. Some multipolar cells were similar to large basket cells (Marin-Padilla, 1969; Jones, 1975); that is, they had a large cell body (20 μm or more in diameter) and long dendrites.
Combination of CR and NPNFP immunocytochemistry
The relationship between CR-ir axons and NPNFP-ir pyramidal cells appears to be highly selective, because many NPNFP-ir pyramidal cells in layers III and V were seen innervated by CR-ir terminal-like puncta (Figs. 2 A, 3 A–D), whereas many others showed a lack of these contacts (Fig. 1 C, Table1). Furthermore, multiterminal endings frequently were observed innervating the distal apical dendrites of NPNFP-ir pyramidal cells, whereas the cell body and proximal dendrites of these pyramidal cells were innervated by few or no CR-ir terminal-like puncta (Figs.2 A, 3 A). Therefore, CR-ir axons innervate preferentially the distal apical dendrites of certain NPNFP-ir pyramidal cells. In addition, some long, vertically oriented CR-ir dendritic processes also were found in close apposition to apical dendrites of NPNFP-ir pyramidal cells (Fig.3 D).
Combination of CR and CB immunocytochemistry
Numerous CB-ir pyramidal cells also were seen innervated by CR-ir axon terminals (Table 1). However, the relationship between CR-ir axons and CB-ir pyramidal cells was inverse with respect to the relationship between CR-ir axons and NPNFP-ir pyramidal cells; that is, CR-multiterminal endings frequently were observed around the cell body and proximal dendrites of CB-ir pyramidal cells (Fig. 3 E), whereas only occasional CR-ir terminals were observed contacting the apical dendrites of these pyramidal cells. No dendro–dendritic close appositions were observed.
Because in sections from the human temporal neocortex stained for CB there were found pyramidal cells surrounded by CB-ir terminals (del Rı́o and DeFelipe, 1996a), double-labeling immunofluorescence experiments were performed to study whether CR-ir and CB-ir pericellular axons converged or not on the same pyramidal cell somata. In this material (Fig. 2 C–E) we found that numerous pyramidal cells were innervated by both CR-ir (Fig. 2 C) and CB-ir (Fig. 2 D) terminals, but double-labeled pericellular axon terminals were scarce (Fig. 2 E), which indicated that most of these axons had different origins.
Cell bodies and proximal dendrites of CB-ir nonpyramidal cells were not observed innervated by dense CR-ir multiterminal endings (Fig.3 F, Table 1): only one or two CR-ir axon terminals occasionally were observed innervating CB-ir nonpyramidal cells. No dendro–dendritic close appositions were seen.
Combination of CR and PV immunocytochemistry
The relationship between CR-ir axons and PV-ir nonpyramidal cells was similar to that found between CR-ir axons and CB-ir nonpyramidal cells. That is, few pericellular CR-ir terminals were seen around PV-ir nonpyramidal cells (Table 1). However, long dendritic segments of CR-ir neurons with bitufted or multipolar morphology frequently were observed in close apposition to dendrites of PV-ir neurons (Figs.1 E, 2 B). The dendrites in contact did not have a preferred orientation, and they often also contacted the soma (Fig. 2 B).
Postembedding immunocytochemistry for CR and GABA
To study the relationship between CR-ir neurons and GABA-ir and non-GABA-ir neurons, we used postembedding immunocytochemistry in 2-μm-thick semithin sections (Fig. 4), because in our material the quality of immunostaining (labeling of cells and puncta) was superior to that achieved with preembedding GABA immunocytochemistry. However, the dendrites of GABA-ir neurons were not stained; therefore, the examination of the relationship between CR-ir neurons and GABA-ir and non-GABA-ir neurons was limited to the innervation of their somata. In serial semithin sections it was observed that neurons that were GABA-ir, but not CR-ir, commonly were innervated by CR-ir terminals, whereas very few CR-ir terminals were seen around the soma of neurons that were immunoreactive for both CR and GABA (cells labeled with arrows in Fig. 4). Furthermore, we distinguished two types of neurons that were not immunoreactive for either CR or GABA: type 1 cells (the majority) were characterized by their somata being surrounded by numerous GABA-ir terminals, but not by CR-ir terminals (cells labeled with asterisks in Fig. 4), and type 2 cells, the somata of which were surrounded by both GABA-ir and CR-ir terminals (cells labeled with p in Fig. 4). Many cells that were not immunoreactive for GABA and CR were identified as pyramidal cells because of their triangular or conic shape from which arose an apical dendrite.
Electron microscope observations
Six immunoreactive cell bodies of small size (∼10 μm in diameter) from the middle temporal gyrus were examined at the electron microscope level. They displayed similar ultrastructural features and were characterized by a large nucleus containing clumped chromatin and by the small volume occupied by a cytoplasm poor in organelles (Fig.5 A). Furthermore, there were few axosomatic synapses. Typically, it was necessary to search three or more serial ultrathin sections to find a terminal forming an axosomatic synapse. Immunoreactive dendrites presented typical organelles and showed no particular characteristics that distinguished them from other dendrites (Fig. 5 B). However, they received numerous synapses, as compared to the cell body; in single sections of dendritic segments from a few microns long, one or more axodendritic synapses were found (arrows in Fig. 5 B).
The distribution of CR-ir puncta was studied first in semithin sections. In these sections numerous CR-ir puncta were seen around unstained neuronal somata and in the neuropil intervening among the cell bodies and blood vessels. Selected semithin sections from layers II and III were resectioned for electron microscopy. At the electron microscope level most CR-ir puncta were identified as axon terminals, but dendrites (Fig. 5 C) and myelinated (Fig. 5 D) and unmyelinated axons were also among the CR-ir elements. A majority of perisomatic CR-ir axon terminals were seen around unstained pyramidal cells (Fig. 6 A,C), and they formed symmetrical synapses (Fig. 6 B). However, some CR-ir axon terminals were observed around nonpyramidal cell somata, and they formed either asymmetrical or symmetrical synapses, but we did not intend a quantitative analysis to study further the somatic innervation of pyramidal and nonpyramidal cells.
In the neuropil, CR-ir axon terminals formed either symmetrical (Fig. 7 A,B) or asymmetrical (Fig.7 C–F) synapses with dendritic profiles. The vast majority of these dendritic profiles was nonimmunoreactive for CR, and only occasionally a CR-ir axon terminal forming a synapse with a CR-ir dendritic profile was found (Fig. 5 C). In the latter cases the postsynaptic densities were more prominent than those found on unstained postsynaptic elements. To study quantitatively the synaptic distribution of CR-ir axon terminals in the neuropil, we analyzed a random sample of 143 synapses made by CR-ir axon terminals forming clearly identifiable morphological types of synapses (asymmetrical or symmetrical) with unstained postsynaptic elements (Table2). Approximately 31% were of the asymmetrical type and 69% of the symmetrical type. The vast majority of postsynaptic elements (88%) could be identified as large proximal dendrites or thinner (presumably more distal) dendrites and dendritic spines (Fig.7). Unidentified postsynaptic elements displayed a small caliber (<1 μm), and they could not be identified as spines or dendritic shafts. Approximately 50 and 26% of CR-ir asymmetrical synapses were found on spines and dendritic shafts, respectively, whereas for CR-ir symmetrical synapses these proportions were 3 and 89%. Thus, in the neuropil, CR-ir axon terminals forming asymmetrical synapses preferentially innervated spines, whereas those forming symmetrical synapses preferentially contacted dendritic shafts.
The present work makes these main observations: (1) CR-ir multiterminal endings frequently innervated the distal apical dendrites of NPNFP-ir pyramidal cells and the cell body and proximal dendrites of CB-ir pyramidal cells. Nonpyramidal cell bodies immunoreactive for CR, CB, or PV were innervated only occasionally by CR-ir axons, whereas certain GABA-ir neurons, which were not CR-ir, were surrounded by CR-ir multiterminal endings; (2) some CR-ir dendrites were found in close apposition to the apical dendrites of NPNFP-ir pyramidal cells or dendrites and somata of other CR-ir or PV-ir nonpyramidal neurons, but no such appositions were observed with CB-ir neurons; and (3) CR-ir axon terminals formed either asymmetrical or symmetrical synapses with various postsynaptic elements.
Relationship between CR-ir multiterminal endings and pyramidal cells
Immunocytochemistry for CR revealed numerous unstained pyramidal cell bodies surrounded by CR-ir multiterminal endings. Combination immunocytochemistry for CR/NPNFP or for CR/CB showed that many pyramidal cells immunoreactive for NPNFP or CB were innervated by CR-ir multiterminal endings. However, the number of NPNFP-ir or CB-ir pyramidal cells was relatively small, as compared with the high number of pyramidal cells observed that were innervated by CR-ir axons not immunoreactive for either NPNFP or CB. Thus, NPNFP-ir and CB-ir pyramidal cells represented a subpopulation of the pyramidal cells innervated by CR-ir axons. Preferential innervation of the distal apical dendrites in the case of NPNFP-ir pyramidal cells, and of the cell body and proximal dendrites in the case of CB-ir pyramidal cells, indicates that the origins of the two kinds of CR-ir axons innervating these cells were from different types of neurons. Colocalization experiments in the human temporal neocortex showed that 30% of NPNFP-ir neurons and 42% of CB-ir neurons were double-labeled (Hayes and Lewis, 1992). Furthermore, combination immunocytochemistry for NPNFP and PV in the human temporal neocortex (del Rı́o and DeFelipe, 1994) has shown that between ∼54 and 70% of NPNFP pyramidal cells in layer III were innervated by PV-ir chandelier cell axons (which innervate axon initial segments). Therefore, the two kinds of CR-ir axons and the PV-ir chandelier cell axons may or may not converge on the same pyramidal cells; it is also possible that different combinations of these axons innervate different populations of pyramidal neurons. Additionally, double-labeling immunofluorescence experiments showed that CR-ir and CB-ir pericellular axons often converged on the same pyramidal cell somata, but the low degree of colocalization of these substances in these axons suggests that they were derived from different types of neurons. The present results further emphasize the complexity of the innervation of pyramidal cells (DeFelipe and Fariñas, 1992).
Relationship among CR-ir multiterminal endings and CR-ir, CB-ir, PV-ir, and GABA-ir nonpyramidal neurons
Cell bodies and proximal dendrites of CR-ir, CB-ir, and PV-ir nonpyramidal neurons were innervated only occasionally by CR-ir terminals. Because in the primate neocortex many of these neurons are likely also to be GABAergic and to represent a large population of GABAergic neurons (Hendry et al., 1989; van Brederode et al., 1990;Hendry and Carder, 1993; del Rı́o and DeFelipe, 1996a), our results suggest that, in general, GABAergic cell bodies are not postsynaptic targets of CR-ir terminals. This is supported by the findings for pairs of adjacent serial semithin sections processed for CR or GABA postembedding immunocytochemistry, which revealed that the majority of GABA-ir somata was devoid of CR-ir terminals. However, some GABA-ir cell somata that were not CR-ir were innervated by CR-ir multiterminal endings. Furthermore, these cells were surrounded by fewer GABA-ir terminals than CR-ir terminals; therefore, it is possible that some of these pericellular terminals originated from CR-ir/non-GABA-ir cells (see below). Noteworthy in the monkey prefrontal cortex, Gabbott and Bacon (1996), using similar methods to those in the present study, found, however, that CR-ir axons frequently were seen contacting cell bodies and proximal dendrites of CB-ir and PV-ir neurons. Whether the different findings in these studies are related to cytoarchitectonic differences in neuronal circuits or to microanatomical distinctions between human and monkey cerebral cortices remains to be determined.
Close dendro–dendritic appositions
Dendrites in close apposition with other dendrites and somata were observed in all combination pairs except between CR-ir and CB-ir neurons. Dendro–dendritic appositions appeared to be highly selective because, although the zones of contact were of considerable length, relatively few neurons displayed appositions. Dendro–dendritic synapses, dendro–dendritic or dendro–somatic gap junctions, and other junctional specializations have been described in a number of regions of the CNS (Peters et al., 1991), including the monkey neocortex (Sloper, 1971, 1972; Sloper and Powell, 1978; DeFelipe and Jones, 1985). Furthermore, Gulyás et al. (1996) found in the rat hippocampus that CR-ir neurons often form dendro–dendritic contacts; also in these contacts numerous zonula adherentia were seen at the electron microscope level. Whether in the human neocortex these contacts simply represent adhesive points or have a functional significance remains to be elucidated. However, the selectivity of the dendro–dendritic appositions and the existence of dendro–dendritic synapses and gap junctions in the monkey neocortex incline us to believe that they have an important functional significance (see alsoGulyás et al., 1996).
Synaptic connections of CR-ir neurons
Cell bodies of CR-ir neurons receive very few synapses; thus, their major synaptic input must be on their dendrites. The sources of these synaptic inputs are not known, and, although the cell bodies of the majority of CR-ir neurons are located in layers II and III, it seems clear that those neurons that have long, vertically oriented dendrites (which may transverse several layers) are involved in synaptic circuits different from those having a more local dendritic arborization.
CR-ir axon terminals formed synapses with various postsynaptic elements, including the somata of both pyramidal and nonpyramidal cells, dendritic shafts of large and small caliber, and dendritic spines. A vast majority of axo–somatic synapses were on pyramidal cells and were symmetrical, whereas few axo–somatic synapses were found on nonpyramidal cells being both symmetrical and asymmetrical. As discussed in our previous study (del Rı́o and DeFelipe, 1996a), because large basket cells are not CR-ir, the major source of CR-ir perisomatic terminals on pyramidal cells must be small basket cells. Furthermore, in double-labeled immunofluorescence material we found pyramidal cell somata innervated by both CB and CR-ir axon terminals, but they were not colocalized, which indicates that different subpopulations of small basket cells converged on the same pyramidal cells.
As reported for the rat visual cortex and hippocampus (Lüth et al., 1993; Gulyás et al., 1996), the types of synapses formed by CR-ir axon terminals in the human temporal neocortex were both symmetrical and asymmetrical. The majority of these synapses (69%) was symmetrical, and their source is very likely to be from the population of CR-ir neurons that is also GABA-ir, which represents the major population of CR-ir neurons in the human neocortex (del Rı́o and DeFelipe, 1996a). The sites of termination shown by CR-ir terminals forming asymmetrical synapses were similar to those of the subpopulation of bipolar cells that form this type of synapse in other species (Peters and Kimerer, 1981; Peters and Harriman, 1988) (see alsoFairén et al., 1984). In addition, CR-ir neurons with bipolar morphology were very frequent, and colocalization studies in the human temporal neocortex have shown that ∼26% of CR-ir neurons are not GABA-ir (del Rı́o and DeFelipe, 1996a), which is similar to that for axon terminals forming asymmetrical synapses (31%). Thus, it is likely that the source of these axon terminals is a subclass of non-GABAergic bipolar neurons. Similarly, some bipolar cells have been shown to be myelinated (Peters and Kimerer, 1981); therefore, it is possible that the CR-ir myelinated axons found in the present study originated from bipolar cells.
The vast majority of postsynaptic dendritic profiles was nonimmunoreactive for CR, which was in line with light microscope observations (see above). Although there were various postsynaptic dendritic profiles, CR-ir axon terminals forming asymmetrical synapses preferentially contacted spines (probably from pyramidal cells), whereas those forming symmetrical synapses were preferentially on dendritic shafts. Some of these dendritic shafts were from pyramidal cells, but the origin of the majority of postsynaptic dendrites could not be identified. We do not know whether the more prominent postsynaptic densities of the synaptic contacts made by CR-ir terminals on labeled postsynaptic elements, as compared with those unlabeled, was attributable simply to an enhancement produced by the deposit of electron-dense peroxidase reaction or to the different natures of these contacts.
In conclusion, these results indicate that CR-ir cells are involved in highly selective and complex synaptic microcircuits in the human temporal neocortex and that different subpopulations of excitatory and inhibitory CR-ir nonpyramidal cells exist, which are specialized for selective innervation of the somatic or dendritic regions of certain pyramidal and nonpyramidal neurons.
This work was supported by Fondo de Investigaciones Sanitarias de la Seguridad Social Grant 96/2134. We thank A. Ortiz, J. R. Rodrı́guez, and J. A. Maldonado for technical assistance.
Correspondence should be addressed to Dr. J. DeFelipe, Instituto Cajal (Consejo Superior de Investigaciones Cientı́ficas), Avenida Dr. Arce 37, 28002 Madrid, Spain.