Three-dimensional organization of dendrites and local axon collaterals of shell and core medium-sized spiny projection neurons of the rat nucleus accumbens

It has been shown previously, both in the dorsal (Gerfen et al. 1985; Penny et al. 1988; Bolam et al. 1988; Kawaguchi et al. 1989; Yelnik et al. 1994; Onn et al. 1994) and ventral striatum (Arts and Groenewegen 1992), that the geometry of the dendritic arborizations of MSN is influenced or limited by the borders between striatal compartments, i.e., the boundaries between patch and matrix. Our present observations in the core of the Acb confirm these findings, indicating that the dendrites of MSN remain within the compartment in which their parent cell body is located. This neuronal architecture supports the generally held view, primarily based on immunohistochemical and connectional data, that the patch and matrix form two largely segregated information processing compartments within the striatum (for reviews see Graybiel 1990; Gerfen 2004). However, the present study extends these observations by showing that MSN in either of these two larger subregions of the Acb, i.e., the shell and core, likewise retain their dendrites within the subregion in which the cell body resides. Like the dendrites of neuronal cell bodies that lie close to the border between patch and matrix, the dendrites of MSN in close proximity of the shell-core boundary tend to run in parallel to or are oriented away from this boundary. For neurons situated close to the boundary, this results in a very asymmetric and in some cases strongly flattened shape of the dendritic arborizations. The fact that the dendrites of MSN in shell or core do not cross the border between these two subregions of the Acb supports the notion that the shell and core constitute two morphologically and functionally largely segregated ‘compartments’ of the Acb. The distinction between shell and core on the basis of (immuno)histochemical and connectional criteria has been described extensively since its recognition by Zaborszky and colleagues (1985; for reviews, see Groenewegen et al. 1996, 1999; Zahm 1999). Likewise, the functional differentiation between shell and core has received extensive coverage in the literature (for reviews, see Cardinal et al. 2002; Di Chiara 2002; Kelley 2004). Our present findings suggest that it is unlikely that MSN in either Acb subregion are able to pick up information from incoming fibers in the adjacent subregion. This supports the idea of parallel and independent processing of information in both the shell and the core to be transmitted to their respective segregated target areas in the basal forebrain, diencephalon and mesencephalon (Groenewegen et al. 1996, 1999). As will be discussed below, recurrent axon collaterals of MSN in either core or shell might cross the border between the two subregions or the compartments therein and yet in this way providing a way of mutual influences at the striatal level (see also Van Dongen et al. 2005).

Apart from the direct influence of compartmental or subregional boundaries on the orientation and geometry of the dendritic arborizations of MSN that are situated in close proximity to such boundaries, the 3D reconstructions of most of the injected MSN in the present study show that the dendritic fields of MSN are flattened and that they show a preferred orientation. In the shell subregion, more than half of the neurons have a dendritic field that is preferentially oriented from rostral–dorsal–medial to caudal–ventral–lateral. Another, less numerous population of MSN shows a preferred orientation from dorsolateral to ventromedial. In the core the preferred orientations appear to be more heterogeneous with about one-third having an orientation from dorsomedial to ventrolateral, one-third from dorsolateral to ventromedial and the remaining neurons with a variety of orientations. It must be emphasized that the majority of neurons we injected in the shell are located in the medial part. In this part of the shell the afferent fibers from the ventral hippocampus enter, through the lateral septum, from a dorsal and medial position to run in ventral, lateral and rostral directions (Groenewegen et al. 1987). The amygdaloid afferents from the magnocellar basal amygdaloid nucleus enter the medial shell, through the bed nucleus of the stria terminalis, from caudal, medial and dorsal and run in rostral, lateral and ventral directions (Wright et al. 1996). Afferents from the ventral prelimbic and infralimbic areas enter the medial shell from rostral and run in caudal direction oriented more or less in a dorsal–medial to ventral–lateral direction (Berendse et al. 1992). In other words, the preferred orientation of the dendritic arborizations is more or less in parallel with the main afferent inputs of the medial shell of the Acb. This might be interpreted such that the dendrites of these neurons pick up information from a relatively restricted set of afferent fibers that make a maximal traverse through the dendritic arbors of individual neurons. Such an arrangement may lead to more contacts from a restricted afferent source, i.e., a form of convergence, as compared to the situation for the smaller population of MSN with a preferred dendritic arborization from dorsolateral to ventromedial. These dendritic fields tend to be oriented more or less perpendicular to the main stream of incoming fibers, and such an arrangement could result in fewer contacts with individual neurons but also with a larger number of neurons, i.e., a form of divergence. Convergence of inputs from different afferent sources onto single MSN in the shell of the Acb has been shown with both anatomical and electrophysiological methods for hippocampal and amygdaloid inputs (Mulder et al. 1998; French and Totterdell 2003) and anatomically for hippocampal and prefrontal fibers (French and Totterdell 2002). Whether the earlier mentioned differences in arrangement of incoming fibers versus orientations of dendritic arborizations make a real difference with respect to the number of contacts being made by single axons from hippocampal or amygdaloid origin with individual MSN in the medial shell remains to be established.

Within the core of the Acb the preferred orientations of dendritic arborizations of MSN show more variations. This might well be related to the less ordered input into this part of the ventral striatum and the complex 3D curvature of the shell-core boundary. Thus, although more dorsal parts of the core merge with the caudate–putamen complex and contains the typical patch-matrix structure and orientation of the dorsal striatum, the more ventral, lateral and rostral parts of the core are closely associated with the ventral and lateral shell, and the rostral shell (rostral pole) that curve around the core subregion. Although the orientation of afferent fiber systems in the caudate–putamen complex and the dorsal core is from dorsomedial to ventrolateral (cf. Berendse and Groenewegen 1990; Berendse et al. 1992; Wright et al. 1996; Voorn et al. 2004), the orientation of afferents to the ventral and rostral parts of the core is less well ordered.

In the head of the primate caudate nucleus, Walker et al. (1993) have shown that there exists a predominant dorsomedial to ventrolateral orientation of the dendritic arborizations of MSN, an orientation that parallels the main orientation of the striosomes in this part of the striatum. Desban et al. (1993) have demonstrated that the architecture of the patch network in the rat caudate–putamen likewise exhibits an overall dorsomedial to ventrolateral orientation. However, our results reveal that the preferred dendritic orientation of only a minority of the MSN in the Acb core conforms to this orientation. This might well be related to a somewhat more complex spatial orientation of the compartments and, as discussed above, a less ordered organization of afferents in this part of the striatum.

While there are quite some studies describing the morphology of the dendrites and the 3D-geometry of the dendritic arborizations of MSN in the striatum, it has clearly been more difficult to fill the very delicate recurrent collateral, intrastriatal axonal ramifications of these neurons. In the present study, in slightly more than half of the neurons that were juxtacellularly injected, both dendrites and axons were sufficiently filled to allow adequate visualization of these fine intrastriatal axons. For the majority of these neurons the main axon exiting the striatum and terminating in the ventral pallidum or ventral mesencephalon could be identified (see also Van Dongen et al. 2005), indicating the adequacy of the labeling procedure. The present study reveals two interesting aspects concerning the recurrent axon collateral network of MSN in the ventral striatum. First, in contrast to the dendrites of MSN, recurrent axon collaterals of these neurons may cross subregional (shell-core) or compartmental (patch-matrix) boundaries in the ventral striatum and in this way interconnect these functionally different subregions or compartments. Indications for patch-matrix communication on the basis of axon collaterals of the MSN projection neurons have been very sparse until now. Kawaguchi et al. (1989), on the basis of intracellular injections of biocytin in in vitro slices of the dorsal striatum, concluded that both dendrites and axons of MSN respect the boundaries between patch and matrix. While Onn et al. (1994), using in vivo intracellular labeling, describe main projection axons of MSN located in the matrix of the dorsal striatum that cross patches on their way to the pallidum, these authors explicitly state that these axons do not show terminations in these patches. In a previous study from this laboratory, using intracellular injections of Lucifer Yellow in lightly fixed slices of the ventral striatum, a single MSN in the matrix could be identified that had extensive axon collaterals in the adjacent patch compartment (Arts and Groenewegen 1992). The results of the present study confirm these observations in showing that in the core of the Acb axons of labeled MSN may enter and terminate in a nearby patch. It must be emphasized, however, that intercompartmental connections on the basis of MSN projections are scarce and intercompartmental connections might be more prominent for interneurons with long axon collaterals, such as cholinergic (Meredith et al. 1989; Kawaguchi 1993) and somatostatin interneurons (Chesselet and Graybiel 1986). The axons of the large aspiny cholinergic interneurons, however, tend to respect compartmental boundaries (Meredith et al. 1989; Kawaguchi 1993). This has been established most directly with intracellular labeling of individual neurons in the dorsal striatum (Kawaguchi 1993). With immunohistochemical studies, Chesselet and Graybiel (1986) have demonstrated that somatostatin interneurons might cross compartmental boundaries. The parvalbumin and calretinin interneurons may also be involved in patch-matrix intercommunication, as demonstrated by immunohistochemical studies (Cowan et al. 1990; Fortin and Parent 1994). As discussed previously (Van Dongen et al. 2005), MSN recurrent axon collaterals might also be the substrate for communication between shell and core of the Acb. Although these projections are bidirectional, as also shown in the present study, there appears to be dominance for core-to-shell projections when the results of larger tracer injections are taken into account.

The second interesting aspect revealed in the present study concerns the distribution of the recurrent axonal network in relationship to the dendritic arborizations of the same striatal projection neuron. Principally, three patterns were distinguished: (1) MSN with a distribution of axon collaterals largely falling within the space of the dendritic distribution (‘full overlap’); (2) MSN with axonal collaterals extending beyond the reaches of the dendritic arborizations, but with a considerable degree of overlap (‘partial overlap’); (3) MSN with recurrent axon collaterals occupying a 3D-space that showed no or only minimal overlap with the dendrites of the same neuron (no overlap). In the shell subregion, all injected neurons belonged to the second category, indicating that the axon collaterals of MSN in this part of the Acb in principle reach other shell neurons primarily within but also beyond the ‘receptive’ sphere of their dendrites. In the core subregion, neurons of all three categories were identified. One neuron with no overlap between dendrites and axon collaterals exhibited a widely distributed axonal network extending up to 1 mm away from the parent cell body. In particular, the findings in the core are largely in agreement with those of earlier studies in the dorsal striatum. Both Bishop et al. (1982) and Kawaguchi et al. (1990) distinguished two categories of striatal projection neurons. One category consists of neurons with axons that are largely overlapping the dendritic field of the same neuron. Within this category several subtypes were described on the basis of their extrinsic projections to pallidal and mesencephalic targets. A second category consists of neurons with axons that have a wide distribution of their axon collaterals, up to 1–2 mm within the striatum, and that do not project beyond the pallidum (Kawaguchi et al. 1990). Like in our sample, neurons in the latter category were far less numerous than MSN with short-range recurrent axonal fields. Whether MSN with longer ranging axon collaterals exist within the Acb shell remains uncertain; the present sample of juxtacellulary filled neurons does not include neurons of that category. Nevertheless, it has been suggested on the basis of ‘classical’ small tracer injections that the shell has more extensive and wide-ranging associative intrastriatal projections than the core and the rest of the striatum (Heimer et al. 1991; Van Dongen et al. 2005). However, whether such long-range intrastriatal projections are established by recurrent collaterals from MSN or axons of interneurons cannot be concluded on the basis of the present results.

Medium-sized spiny projection neurons are the most prevalent neuronal cell type in both the dorsal and ventral striatum, by far outnumbering the smaller and heterogeneous population of striatal interneurons. MSN with their extensive ‘receptive’ surface in the form of dendrites and spines, represent the neuronal substrate par excellence for the integration of information from functionally different cortical areas, limbic structures, and midline and intralaminar thalamic nuclei in the striatum. These inputs arrive mainly on the heads of the dendritic spines of MSN while the transfer of this information, that is excitatory in nature, can be modulated by dopaminergic inputs and contacts from striatal interneurons. Dopaminergic fibers make contact on the neck of the spines or on the dendritic shafts while the terminals of cholinergic and GABAergic interneurons are positioned more proximally on dendritic shafts and the perikaryon of the MSN (reviews, e.g., Smith and Bolam 1990; Gerfen 2004; Tepper et al. 2004). As discussed above, it has long been known that MSN have an extensive local recurrent collateral axonal network (Ramón y Cajal 1911; Preston et al. 1980; Bishop et al. 1982; Chang and Kitai 1985) and on the basis of ultrastructural studies, it has been demonstrated that these axons make inhibitory synaptic contacts on the dendritic shafts of neighboring MSN (e.g., Bolam and Izzo 1988; Meredith et al. 1993; Tunstall et al. 2002). Thus, the inhibitory inputs from spatially closely related MSN appear to be in a strategic position on the dendrites of an MSN to influence the transfer of information from the dendrites to the cell body. While it has been notoriously difficult to show functional synaptic contacts between MSN with electrophysiological methods (e.g., Jaeger et al. 1994; see also Tepper et al. 2004), several recent studies have confirmed their functional existence in both the dorsal (Tunstall et al. 2002; Czubayko and Plenz 2002; Venance et al. 2004) and ventral striatum (Taverna et al. 2004). Taverna et al. (2004), using dual whole cell patch-clamp recordings in acute slices of the Acb, showed that no less than 34% of the patched pairs of MSN had synaptic connections, most of them uni-directionally. It could further be shown that the connections between MSN exhibit various forms of short-term synaptic plasticity and that the functional connectivity can be modulated by dopamine (Taverna et al. 2004, 2005; also Czubayko and Plenz 2002).

Striatal interneurons, including the fast-spiking GABAergic interneurons, are known to contact MSN and play an important role in modulating their output (for review see Tepper et al. 2004). There are no electrophysiological data about possible connections from MSN to these interneurons. Theoretically such connections would have an influence over a wider striatal area than the interconnections between MSN because interneurons have a much more extensive axonal network than MSN. However, because little is known about MSN to interneuron projections, we concentrate our further discussion on the possible functional aspects of interactions between MSN.

In theories of striatal functioning, synaptic connections between MSN have been interpreted as being important for striatal learning and selection mechanisms through lateral inhibitory processes. Lateral inhibition between MSN is thought to play a role in the interactions between collections or ‘ensembles’ of neurons that are influenced by different sets of cortical inputs and code for competing outputs of the striatum (e.g., Pennartz et al. 1994; Plenz 2003; Tepper et al. 2004). In this way, lateral inhibition between MSN would support a ‘winner take all’ mechanism in the selection of striatal outputs. Recent insights into the role of an important class of inhibitory striatal interneurons, i.e., the fast-spiking GABAergic interneurons that are driven by excitatory cortical inputs and have a much more powerful inhibitory influence on MSN than recurrent collaterals of MSN themselves, have again questioned the significance of MSN–MSN interactions (Tepper et al. 2004). However, as argued by Plenz (2003), the inhibitory mechanisms of fast-spiking interneurons and collateral connections between MSN could both play a distinctive role in the striatal network dynamics that underlie timing and selection of cortical information that is processed through the basal ganglia circuitry. Thus, even though the direct electrophysiological effect of MSN inputs on adjacent neurons might not be as strong, the consistent findings that these MSN–MSN connections exhibit several forms of short-term plasticity and are influenced by dopamine suggests a role for such interactions at the striatal microcircuit level (Taverna et al. 2004, 2005; also Czubayko and Plenz 2002), a role that has to be still further explored.

In further considering the functional role of connections between MSN, the results of the present study with respect to the distribution of recurrent axon collaterals of MSN in the ventral striatum should be taken into account. First, most theoretical models depart from the assumption that the geometry of the axonal arborizations of MSN largely concurs with the dendritic geometry of the parent cell body. Our study, in agreement with Kawaguchi et al. (1990), shows that such congruence is not always the case and that there exist MSN with long-range axonal terminations and several forms of nonoverlapping axonal and dendritic arborizations. This could imply that such recurrent axon collaterals connect functionally different sets of MSN and provide a way of interaction or competition between such populations of neurons. Important in this context is the demonstration of directionality in the intrastriatal connections as described in our previous study (Van Dongen et al. 2005) and confirmed in the present study at the single cell level. Second, our findings of recurrent axon collaterals of MSN that cross the patch-matrix and the shell-core borders emphasize and strengthen the idea that such collaterals play a role in the interaction between functionally different compartments or ‘ensembles’ of neurons (cf also Pennartz et al. 1994; Van Dongen et al. 2005).