Long-Range GABAergic Projection in a Circuit Essential for Vocal Learning

Transcription

1 THE JOURNAL OF COMPARATIVE NEUROLOGY 403:68 84 (1999) Long-Range GABAergic Projection in a Circuit Essential for Vocal Learning MINMIN LUO AND DAVID J. PERKEL* Department of Neuroscience, University of Pennsylvania, Philadelphia, Pennsylvania ABSTRACT The anterior forebrain pathway (AFP) in the passerine song system is essential for song learning but not for song production. Several lines of evidence suggest that area X, a major nucleus in the AFP, forms part of the avian striatum. A key feature of striatal projection neurons is that they use the inhibitory neurotransmitter -aminobutyric acid (GABA). Some area X neurons express GABA-like immunoreactivity, but the neurotransmitter phenotype of the projection neurons is largely unknown. To determine whether area X projection neurons are GABAergic, we used immunocytochemistry and confocal microscopy to examine whether these neurons in adult male zebra finches express the GABA synthetic enzyme glutamic acid decarboxylase (GAD). We observed numerous large and small GAD somata in area X, and dense GAD terminals, but no GAD somata in the target of area X, the medial nucleus of the dorsolateral thalamus (DLM). The density of GAD terminals in DLM was strongly reduced by ibotenic acid lesions of area X. After tracer injection into the DLM, all of the retrogradely labeled neurons in area X were GAD. After tracer injection into area X, the vast majority of anterogradely labeled terminals in DLM were GAD. We conclude that area X neurons projecting to DLM express GAD and are thus likely GABAergic. If this projection is indeed inhibitory, information processing in the AFP is substantially more complicated than previously realized. Moreover, because a GABAergic projection to a thalamic target is reminiscent of pallidal rather than of striatal circuitry, area X may contain both striatal and pallidal components. J. Comp. Neurol. 403:68 84, Wiley-Liss, Inc. Indexing terms: birdsong; area X; DLM; glutamic acid decarboxylase; zebra finch Birdsong is a highly complex behavior acquired through two learning phases. During an initial sensory phase, a young bird forms a song template by memorizing the song of a tutor. During a later sensorimotor integration phase, which can begin before the end of the sensory phase, the bird begins to sing a highly variable subsong and gradually learns to produce a song closely resembling the tutor song (Konishi, 1965; for reviews, see Konishi, 1985, 1990; Nottebohm, 1996; Bottjer and Arnold, 1997). A group of forebrain nuclei, collectively known as the song system, is specialized for song learning and production. The song system can be divided into a primary motor pathway in the posterior forebrain and an anterior forebrain pathway (AFP; Fig. 1; for review, see Doupe, 1993). The motor pathway includes nucleus HVc (used here as the proper name, after Brenowitz et al., 1997) in the neostriatum, which projects to the robust nucleus of the archistriatum (RA; Nottebohm et al., 1976). RA in turn projects to the hypoglossal nucleus (nxiits; Nottebohm et al., 1976; Wild, 1993; Vicario, 1993), which innervates the muscles of the syrinx, the vocal organ of songbirds (Vicario and Nottebohm, 1988). This pathway is essential for song production (Nottebohm et al., 1976). The AFP provides a second, indirect pathway linking HVc to RA (Fig. 1). HVc projects to area X (Nottebohm et al., 1976). The only known efferent target for area X is the medial nucleus of the dorsolateral thalamus (DLM; Okuhata and Saito, 1987; Bottjer et al., 1989; Sohrabji et al., 1993), which in turn projects to the lateral magnocellular nucleus of the anterior neostriatum (lman; Nottebohm et al., 1982; Bottjer et al., 1989). Projection neurons within lman send their output to RA (Nottebohm et al., 1976) within the motor pathway and collateral output to area X (Nixdorf-Bergweiler et al., 1995; Vates and Nottebohm, *Correspondence to: David J. Perkel, Department of Neuroscience, 215 Stemmler Hall, University of Pennsylvania, Philadelphia, PA perkeld@mail.med.upenn.edu Received 20 May 1998; Revised 17 August 1998; Accepted 20 August WILEY-LISS, INC.

2 GABAERGIC PROJECTION IN THE SONG SYSTEM 69 Fig. 1. A highly simplified schematic view of selected major nuclei and connections in the song system. Nucleus HVc, the robust nucleus of the archistriatum (RA) and the hypoglossal nucleus (nxiits), form a descending motor pathway. The anterior forebrain pathway (AFP), formed by area X, the medial nucleus of the dorsolateral thalamus (DLM) and the lateral portion of the magnocellular nucleus of the anterior neostriatum (lman), receives input signals from HVc through area X and sends its output signals to RA through lman. In this and subsequent figures, dorsal is up and anterior is to the right. 1995). In contrast to the motor pathway, the AFP is necessary for song learning but not for song production in adult birds. Lesions in this pathway in adult birds do not have an obvious effect on song quality; however, lesions prior to song crystallization prevent the bird from matching its song to that of the tutor (Bottjer et al., 1984; Sohrabji et al., 1990; Scharff and Nottebohm, 1991). Several roles have been hypothesized for the AFP in song learning, including forming and storing the song template, providing auditory feedback to the motor system, and participating in song production (Mooney, 1992; Doupe, 1993; Doya and Sejnowski, 1995; Troyer et al., 1996; Jarvis and Nottebohm, 1997). Although there are data supporting each of these ideas, it remains difficult to assign specific functions to this pathway. Regardless of its contribution to song learning, the AFP has almost always been explicitly (Doya and Sejnowski, 1995; Troyer et al., 1996) or implicitly assumed to be an excitatory pathway. Several studies have suggested that area X, the largest nucleus within the AFP, forms a portion of the avian basal ganglia. Area X is located within the lobus parolfactorius (LPO), which, along with the paleostriatum augmentatum (PA), is thought to be homologous to mammalian striatum in nonsinging birds (Karten and Dubbeldam, 1973; Reiner et al., 1984; Medina and Reiner, 1995; Veenman et al., 1995). In addition, area X receives heavy dopaminergic input from two midbrain regions, the area ventralis of Tsai (AVT) and nucleus tegmenti pedunculopontinus, pars compacta (TPc; Lewis et al., 1981). These nuclei, sometimes referred to as AVT/substantia nigra (Bottjer, 1993), form the avian homologue of mammalian midbrain dopaminergic areas projecting to striatum (Lewis et al., 1981; Bottjer et al., 1989; Bottjer, 1993; Soha et al., 1995). Furthermore, dopamine receptors are densely expressed throughout the LPO, with somewhat higher density in area X (Casto and Ball, 1994). Finally, area X contains neurons immunoreactive for enkephalin (Bottjer and Alexander, 1995; Carillo and Doupe, 1995), as does the mammalian striatum (Parent and Hazrati, 1995). These anatomical and neurochemical data suggest that area X is a striatal portion of basal ganglia within the song system. The pallial areas HVc and lman (Striedter et al., 1998), both of which project to area X, may correspond to mammalian cerebral cortex (Bottjer, 1993). The songbird AFP connections (HVc = area X = DLM = lman) thus resemble the mammalian cerebral cortex = basal ganglia = thalamus = cortex connections (Bottjer and Johnson, 1997). Although it could be argued that the projection from area X to the thalamus distinguishes it from the mammalian striatum (but see Discussion), the neurochemical properties of area X and its inputs from pallial structures suggest functions and/or underlying processing mechanisms similar to those of the mammalian striatum. The last two stations of the AFP (DLM = lman = RA) are glutamatergic and excitatory (Kubota and Saito, 1991; Mooney, 1992; Livingston and Mooney, 1997; Boettiger and Doupe, 1998; Bottjer et al., personal communication); however, no direct evidence is available regarding the neurotransmitter phenotype of the first two connections (HVc = X = DLM). One key feature of mammalian basal ganglia is that the majority of neurons, including projection neurons in both striatum and globus pallidus, use the neurotransmitter -aminobutyric acid (GABA; for review, see Parent and Hazrati, 1995) to inhibit their targets. This is true for striatal and pallidal homologues in a variety of vertebrate species including pigeons (Medina and Reiner, 1995), which are not songbirds. Thus, a GABAergic projection from area X to DLM is a key criterion by which to examine the hypothesis that this projection participates in a circuit similar to the mammalian corticobasal ganglia thalamocortical loop. One study has suggested that DLM receives GABAergic input that could arise from area X (Grisham and Arnold, 1994). Numerous small cells in area X show GABA-like immunoreactivity (GABA-LIR). Because many terminals, but no somata, in DLM also show GABA-LIR, these terminals may arise from neurons located outside of DLM. Whereas RA provides sparse input to DLM (Wild, 1993; Vates et al., 1997), DLM receives a major afferent connection only from area X (Bottjer et al., 1989). Thus, area X is the primary candidate source for the GABA-LIR terminals in DLM. However, the projection neurons from area X to DLM are among the largest neurons within area X (Bottjer et al., 1989; Sohrabji et al., 1993), much larger than the area X neurons reported to have GABA-LIR. Three major possibilities may reconcile this discrepancy between the GABA-LIR in small cells and the large projection neurons in area X: 1) some small, GABAergic area X neurons projecting to DLM were not previously detected by tracing studies; 2) some large area X projection neurons are GABAergic but were not detected to have GABA-LIR; and 3) DLM receives GABAergic input from as yet unidentified areas other than area X. This study had two goals: to determine the connection pattern of the GABAergic neurons in the AFP and to test whether the X = DLM projection is GABAergic, a key element in the hypothesis that area X has connections similar to those of mammalian basal ganglia. We used immunostaining for glutamic acid decarboxylase (GAD),

3 70 M. LUO AND D.J. PERKEL the enzyme that synthesizes GABA from glutamic acid (Kaufman et al., 1986), to identify the neurons within the AFP that express GAD. Focal excitotoxic lesions and tract tracing were coupled with GAD immunostaining and confocal microscopy to determine whether the area X neurons projecting to DLM contain GAD immunoreactivity. We report that area X contains strikingly large, sparsely distributed, strongly GAD neurons. In addition, many small neurons are more weakly labeled with anti-gad antiserum. DLM contains no GAD somata but is filled with intensely GAD terminals. This terminal labeling is substantially reduced or eliminated following ibotenic acid lesion of area X. Double-labeling experiments show that, after tracer injection into DLM, all retrogradely labeled neurons in area X are GAD. In addition, after tracer injection into area X, the vast majority of anterogradely labeled terminals in DLM are also GAD. We conclude that the projection from area X to DLM is GABAergic and provides the major source of GABAergic input to DLM. MATERIALS AND METHODS Animals This study was carried out according to a protocol approved by the University of Pennsylvania Animal Care & Use Committee. A total of 27 adult male zebra finches and one adult mouse obtained from our own breeding facility or from a local breeder were used in this study. Of the 27 birds, one was used for Western blot analysis, one was used for GAD staining only, six were used for both GAD and Nissl staining, four were used for ibotenic acid lesion of area X and GAD staining, four were used for tracer injection into area X and GAD staining, and 11 were used for tracer injection into DLM and GAD staining. The mouse was used for the Western blot analysis. Western blot Tissue preparation and immunoblotting were performed by following methods described by Kaufman et al. (1986), with some minor modifications. Briefly, animals were given an overdose with pentobarbital (250 mg/kg) and decapitated. The brain from one mouse and the brain and leg muscle from one adult male zebra finch were homogenized separately in buffer containing 0.2 mm phenylmethylsulfonyl fluoride, 150 mm NaCl, and 50 mm tris-hcl (ph 8.0). Brain homogenate was sonicated and centrifuged to remove cell debris. Proteins in the supernatant were separated by sodium dodecylsulfate polyacrylamide gel electrophoresis (12%) and were then transferred to polyvinylidene difluoride membrane. After blocking in 5% nonfat milk at room temperature, the membrane was incubated with 1:1,000 AB108 (Chemicon, Temecula, CA), a rabbit polyclonal antiserum against feline GAD 67, in 5% nonfat milk at room temperature for 2 hours. After rinsing, the membrane was incubated in 1:2,000 horseradish peroxidase conjugated goat antiserum against rabbit IgG for 1 hour at room temperature. After rinsing, the membrane was incubated in chemiluminescence reagents (Amersham, Princeton, NJ) for 1 minute, and bound antibodies were detected by a short exposure to autoradiography film (Kodak BioMax, Rochester, NY). Injection of tracers or ibotenic acid Animals were anesthetized with sodium pentobarbital (40 mg/kg) and mounted in a stereotaxic apparatus. The skin over the skull was opened, and a small craniotomy was performed over the desired targets. A stereotaxic injection was made into the target according to stereotaxic coordinates kindly provided by M.M. Solis and A.J. Doupe. All targets were injected with a glass micropipette with a tip diameter of µm. Fluorescein dextran amine (4%; molecular weight of 10 kda; Molecular Probes, Eugene, OR) was used as the anterograde tracer; tetramethylrhodamine (TMR) dextran amine (10%; molecular weight of 3 kda, Molecular Probes) was used as the retrograde tracer. All tracers were iontophoretically injected with positive current (3 µa for anterograde tracing and 12 µa for retrograde tracing) that was pulsed 7 seconds on/7 seconds off, for a total of 30 minutes. To induce lesions of area X, ibotenic acid ( µl, 10 µg/µl) was unilaterally pressure injected with a Hamilton syringe. In two cases, a sham lesion was made by injecting 0.1 M phosphate buffer (PB) into area X of the contralateral hemisphere. After the injection, the pipette was withdrawn, the wound was closed with surgical adhesive (Nexaband, Veterinary Products Laboratories, Phoenix, AZ), and the animal was allowed to recover. After a survival time of 3 4 days, the animal was deeply anesthetized with sodium pentobarbital (250 mg/kg) and perfused transcardially with saline followed by 4% formaldehyde in PB. Histology After perfusion, each brain was postfixed in 4% formaldehyde for 2 hours and then cryoprotected in 30% sucrose in 0.1 M PB. Parasagittal sections were prepared (40 µm thickness) by using a freezing microtome. For GAD immunostaining, sections were incubated in 1:1,500 AB108 and 10% normal goat serum (NGS) in 0.1 M PB for 48 hours at 4 C. After rinsing in PB, sections were incubated with goat anti-rabbit biotinylated secondary antibody (1:200) and 10% NGS in 0.1 M PB overnight at 4 C. After rinsing in PB, sections were incubated with either Cy3- or Cy5- conjugated streptavidin, resulting in indirect immunofluorescence. After rinsing in PB, sections were mounted with Vectashield (Vector Laboratories, Burlingame, CA), coverslipped, and sealed with nail polish. For six birds, Nissl staining was performed on alternate sections. Data collection and analysis Nuclei of interest were identified by their location and cytoarchitecture in alternate Nissl-stained sections. Data were collected with confocal microscopy (Leica TCS). For double-labeled materials, laser power for each color channel was controlled to prevent cross talk between channels. To collect fluorescence information on somata, the whole section thickness was examined by optically sectioning the tissue at 1-µm intervals. To collect detailed images of nerve terminals, the tissue was sectioned optically at the minimal depth interval, 0.54 µm, by using a 100 objective and 2 electronic zoom. Data were analyzed with Scion Image (PC version of NIH Image) and Adobe Photoshop. To measure the size of both GAD neurons and retrogradely labeled neurons in area X after DLM injection, each section within a stack of confocal images was subjected to a threshold algorithm to highlight cell bodies, in which the threshold was set as the

4 GABAERGIC PROJECTION IN THE SONG SYSTEM 71 average pixel value for the section. The number of pixels within each soma and along the major and minor axes were then measured by using macro programs provided with NIH Image. Area and length were then calculated by using pixel size. The maximum value for each parameter (soma area, major axis, and minor axis) of the whole image stack of the cell of interest was taken as the value for that cell. These values for a group of cells are reported as mean SD. For double-labeled materials, images of different channels were assigned different pseudocolors and overlapped with Adobe Photoshop. Arrows and scale bars were added into figures with Microsoft PowerPoint. To present the data, multiple confocal images optically sectioned at different depths from the same area were projected into a single plane. The thickness of the plane of each section is given in the figure captions. RESULTS Antibody verification GAD has two isoforms, GAD 65 (65 kda) and GAD 67 (67 kda), arising from distinct but similar genes (Erlander et al., 1991; Erlander and Tobin, 1991). Many vertebrates, including mammals and birds, express these two isoforms (Erlander and Tobin, 1991; Ahman et al., 1996). These two GAD isoforms also exist in zebra finches, and their conserved regions are highly similar to those of human GAD isoforms (Bosma et al., 1997). In the present study, we used AB108, a rabbit antiserum against feline GAD 67,to detect GAD. Western blot and immunohistochemical studies have shown that this antibody specifically recognizes GAD 67 (Kaufman et al. 1986) in rats and labels GABAergic somata in rat, mouse, and primate (for recent examples, see Betarbet et al., 1997; Gritti et al., 1997). To verify that the labeling observed was specific for zebra finch GAD rather than due to cross reactivity, we carried out Western blot analysis. Figure 2 shows one typical immunoblotting result. A major band with apparent mobility of 63 kda could be observed in mouse brain, whereas two bands at 63 and 59 kda could be observed for zebra finch brains. With some lots of the same antibody, we also observed two bands (63 and 59 kda) for mouse brains. No major band was observed in zebra finch muscle. Previous Western blot analysis indicated that the two GAD isoforms have molecular masses of 67 and 65 kda (Kaufman et al., 1986, 1991) or 63 and 59 kda (Legay et al., 1987; Chang and Gottlieb, 1988). Such discrepancies could result from abnormal migration of GAD or molecular weight standards or from posttranslational modifications of GAD (Legay et al., 1987; Chang and Gottlieb, 1988). Our results suggest that, whereas this antibody mainly recognizes GAD 67 in mouse, it may recognize both isoforms of GAD in zebra finch. However, it appears that there is no major cross reactivity to brain proteins of molecular weight other than that of GAD, and it is thus likely that immunoreactivity in the present study represents the presence of GAD. GAD immunoreactivity in area X and DLM We observed elevated GAD immunoreactivity in HVc, RA, area X, DLM, and lman as compared with the tissue surrounding each of these areas. No neuronal labeling was observed if the primary antiserum was omitted from the histological reaction. The boundaries defined by GAD Fig. 2. The specificity of antibody AB108 against glutamic acid decarboxylase in adult male zebra finch brain was verified by immunoblotting. Arrowheads indicate molecular weight standards of 94, 67, 43, and 30 kda. Lane 1: Mouse brain. Lane 2: Zebra finch brain. Lane 3: Zebra finch muscle. The antiserum recognized one major polypeptide (63 kda) in mouse brain and two major polypeptides (63 and 59 kda; arrows) in zebra finch brain. There were no major bands in zebra finch muscle. immunostaining for each of these nuclei appeared to match the boundaries defined by Nissl staining. In HVc, RA, area X, and lman, we observed GAD immunoreactivity in both terminals and somata. In DLM, however, we observed GAD immunoreactivity only in terminals. Area X had higher GAD immunoreactivity than did the surrounding dorsal and anterior areas. Within LPO posterior and ventral to area X, GAD immunoreactivity was rather strong but was still slightly weaker than that in area X. In low power views, large, strongly GAD neurons could easily be observed; these were sparsely distributed in area X (Fig. 3A). Lightly stained axons could also be seen posteroventral to area X, in the direction of DLM. When tissue was examined at higher magnification, numerous small cells were also found to be GAD in addition to the intensely labeled large neurons (Fig. 3B). A few of the small somata were strongly stained, but overall the level of staining for small cells was apparently much weaker than that for large cells. When area X was examined at even higher magnification, many GAD processes were found. Many, but not all, large GAD somata were tightly enclosed by a ring of strongly GAD puncta (arrows in Fig. 3C) µm in diameter. These GAD puncta, presumably terminals, tightly surrounded some strongly GAD somata (Fig. 3C 1 ) or formed lines emanating from these somata, presumably following primary dendrites (Fig. 3C 2 ). Area X neurons projecting to DLM have large somata (12 15 µm in diameter as reported by Bottjer et al. 1989; µm 2 in soma area as reported by Sohrabji et al., 1993), whereas the vast majority of neurons in area X appear to have much smaller somata ( 35 µm 2 in soma

5 72 M. LUO AND D.J. PERKEL Fig. 3. The strongly glutamic acid decarboxylase positive (GAD ) neurons in area X were sparsely distributed and were large compared with the weakly stained neurons. Many large GAD neurons were also surrounded by GAD terminals. A: Low power view of GAD staining in area X. The most strongly stained neurons were large and sparsely distributed. B: High power view of GAD staining in area X. The strongly stained neurons were relatively large. Many smaller neurons were also more weakly stained. C: GAD puncta tightly surrounded the somata of many large, strongly GAD neurons. In many cases, these GAD terminals aligned to form lines close to the soma. Shown are two images of optical sections at the center of one strongly GAD soma (C 1 ) and 1 µm from the top of the soma (C 2 ). GAD puncta could be seen surrounding the soma (C 1, arrows) or forming lines (C 2, arrows) emerging from the soma. D: Histogram of soma areas of the strongly stained neurons showing that many GAD neurons have a large soma size. Scale bars 100 µm in A, 25 µm in B, 5µinC 1,C 2 ; thickness 40 µm in A,B, 2 µm in C 1,C 2. area as reported by Grisham and Arnold [1994] when using GABA-LIR). We measured the soma area and the lengths of major and minor axes of 502 strongly GAD neurons in area X from three animals to examine whether these large GAD neurons could possibly have the same size as the neurons projecting to DLM. The soma area for these neurons was µm 2, the soma major axis was µm, and the soma minor axis was µm. Figure 3D shows the soma area distribution for these strongly GAD neurons within area X. The soma size of

6 GABAERGIC PROJECTION IN THE SONG SYSTEM 73 Fig. 4. In the medial nucleus of the dorsolateral thalamus (DLM), the target of area X projection neurons, no glutamic acid decarboxylase positive (GAD ) somata were observed. However, many terminals throughout the entire extent of DLM were heavily stained for GAD. A: Low power view of GAD staining in DLM. Weakly stained axons can be seen entering DLM. B: High power view of GAD staining in DLM. There were no somata showing GAD immunoreactivity. These GAD structures are terminals forming baskets about µm in diameter. See also Figures 9 and 10. Scale bars 100 µm in A, 25 µm in B; thickness 40 µm in A, 16 µm in B. many of these strongly GAD neurons was similar to that of area X projection neurons reported by other groups. Although strong GAD immunoreactivity was observed throughout DLM, it was found only in terminals and never in somata (Fig. 4A). This observation is consistent with the GABA-LIR reported by Grisham and Arnold (1994). Weakly stained axons were seen entering the anterior aspect of DLM. Viewed at high power, GAD terminals in DLM formed connected networks resembling baskets about µm in diameter (Fig. 4B). These baskets appear to be GABAergic terminals contacting the somata of DLM neurons. GAD staining patterns in area X and DLM were consistent with the hypothesis that GAD terminals in DLM arise from area X (Grisham and Arnold; 1994), which provides the major input to DLM (Bottjer et al., 1989; Sohrabji et al., 1993; Vates et al., 1997). Lesion-induced depletion of GAD immunoreactivity in DLM To determine whether the GAD terminals in DLM arise from area X projection neurons, partial or complete unilateral lesions were made by injecting ibotenic acid stereotaxically into area X. Because the projection from area X to DLM is unilateral (Bottjer et al., 1989), the GAD staining pattern in the contralateral DLM was used as a control. Of four animals, one animal had a lesion that included all of area X and some areas anterior and ventral to area X (Fig. 5A). In this animal, little if any GAD immunoreactivity was found in DLM in the hemisphere ipsilateral to the lesion site (Fig. 5B), whereas the contralateral hemisphere had dense GAD terminals throughout the extent of DLM (Fig. 5C). After partial lesions of area X, only a portion of DLM had reduced GAD immunoreactivity (n 3; data not shown). A large fraction of the area of reduced GAD immunoreactivity had no detectable GAD terminals. Surrounding areas appeared to have reduced GAD terminal density, and the rest of the nucleus appeared to have normal GAD immunoreactivity. In one animal, with a lesion in central area X, the central area in the ipsilateral DLM did not show any GAD terminals, whereas the rest of DLM appeared to have normal GAD immunoreactivity. In another animal, with a large lesion in the central and anterior portion of area X, an anterior and central region of the ipsilateral DLM showed reduced GAD immunoreactivity. In the third animal, with a lesion in posterior area X, only the posterior portion of DLM showed reduced GAD immunoreactivity. The data from lesion-induced depletion of GAD immunoreactivity suggest that the projection from area X to DLM may follow a topographic organization, a finding mentioned also by Vates and Nottebohm (1995), but we have not yet addressed this issue directly. The major conclusion from these findings is that area X appears to supply the vast majority, if not all, of the GABAergic input to DLM. Tracer injection into DLM and GAD staining To study the projection neurons from area X to DLM more carefully, we injected the tracer TMR dextran amine iontophoretically into DLM. Because of the depth and small size of DLM, it is difficult to make injections restricted to this nucleus. Because our goal was to label as many area X projection neurons as possible and because the only known recipient of area X projections is DLM, we deliberately made large injections into and near DLM. In three animals, four hemispheres had tracer injected into DLM and its surroundings. Because we made large injections that clearly exceeded the borders of DLM, we could not address the possibility of some as yet unidentified projection to DLM. We did observe dense neuronal labeling in RA, but because of the large injection, it is unclear

7 74 M. LUO AND D.J. PERKEL Fig. 5. A unilateral lesion of area X greatly reduced glutamic acid decarboxylase (GAD) immunoreactivity in the ipsilateral part of the medial nucleus of the dorsolateral thalamus (DLM). A: A camera lucida drawing indicates the site of the lesion (filled area) that encompassed all of area X and some areas anterior and ventral to area X. B: GAD staining in DLM of the hemisphere ipsilateral to the lesioned area X was very weak or absent. C: GAD staining in DLM of the contralateral hemisphere was strong and comparable to staining in unlesioned animals. Scale bars 100 µm; thickness 40 µm in B,C. whether these neurons projected to DLM (Wild, 1993; Vates et al., 1997) or were simply labeled because their axons passed through or near the injection site. We also observed expected fiber and terminal labeling in lman (Bottjer et al., 1989) and retrogradely labeled somata in areas posterior, ventral, and medial to area X. In addition, we observed unusually elongated retrogradely labeled somata (30 10 µm) just dorsal and anterior to area X. These cells may project to DLM or the surrounding region. We concentrated on retrograde labeling in area X. Figure 6A shows one injection site that covered the dorsal half of DLM and areas anterior and dorsal to DLM. Large retrogradely labeled cells were sparsely distributed in area X (Fig. 6B). Some of these neurons were well labeled by tracer, and the details of their morphology were examined. The axons of these neurons either directly left the soma (Fig. 6C) or in some cases emerged from a process that appeared to be a primary dendrite at a point very close to the soma (Fig. 6D). The axon of each of these tracer-labeled neurons had an unusual morphology in that its diameter usually decreased after traveling µm from the starting point and then restored its size after an additional µm. By examining adjacent sections, we found that the local axonal collaterals of these retrogradely labeled neurons branched only within the vicinity of the dendritic areas of the cells (Fig. 6D). In many cases, these local axonal branches were located to one side of the dendritic tree rather than being centered on the dendritic area. Some of the well-labeled processes had varicosities (arrows in Fig. 6C) of diameter µm. We never observed these varicosities coexisting with spines on the same process. Furthermore, in 11 cases, we could clearly determine that branches arising from the axons had varicosities. Thus, the varicose processes are likely the local axonal collaterals of area X projection neurons (Bottjer et al., 1989). The dendritic trees formed a concentric region surrounding the soma about µm in diameter, with some processes reaching, in rare cases, as far as 600 µm from the soma. These tracer-labeled neurons had three to six primary dendrites. Figure 6D shows one example of the dendritic tree of an X = DLM projection neuron. When the dendrites were very well labeled, spines (arrows in Fig. 6E) could be found on distal dendrites, whereas no spines were found on proximal dendrites. To examine directly whether the area X neurons projecting to DLM are GAD, GAD staining was performed after tracer was injected into DLM. Of the hundreds of retrogradely labeled neurons examined in area X from four hemispheres in three animals, all were found to be GAD (Fig. 7). For the neurons that were well filled after tracer injection into DLM, we found that the varicosities of local axonal collaterals were also GAD (data not shown). The tracer-labeled somata in area X were not only GAD but also tightly enclosed by a basket of GAD processes (Fig. 7B). GAD puncta could clearly be seen around the soma (arrowheads in Fig. 7B), along the primary dendrites (arrows), and directly above the soma (arrowheads). Of 236 neurons from two animals, 215 neurons clearly had such a layer of GAD puncta around the GAD somata. In the remaining 21 neurons, there were GAD structures around the somata, but their morphology was less punctate and more elongated. Close contact between the distal dendrites and GAD terminals was not observed, but we did not examine the dendrites systematically.

8 GABAERGIC PROJECTION IN THE SONG SYSTEM 75 Fig. 6. Retrogradely labeled cells were sparsely distributed in area X after tracer injection into the medial nucleus of the dorsolateral thalamus (DLM). A: A camera lucida drawing indicates the injection site in DLM. For this animal, tracer was injected into dorsal DLM and areas outside of DLM (inset). The border of DLM as determined by glutamic acid decarboxylase (GAD) immunoreactivity is indicated in the inset. B: Retrogradely labeled neurons in area X were sparsely distributed. C: The axon (arrowhead) of such tracer-labeled neurons in area X branched only locally within the dendritic area of that neuron. Varicosities (arrows) could be observed on well-labeled axonal branches. D: The well-labeled neurons in area X typically show three to five primary dendrites. Shown is a neuron with four primary dendrites. The axon (arrowhead) of this neuron shared a short process with one primary dendrite. E: Many of the distal dendrites of the tracer-labeled neurons had spines, whereas no spines were found in the proximal dendrites of these neurons. Shown is a distal dendrite of the neuron shown in D with spines on it (arrows). Scale bars 200 µm in inset, 100 µm in B, 10 µm in C, 25 µm in D, 5 µm in E; thickness 40 µm in B D, 25 µm in E.

9 76 M. LUO AND D.J. PERKEL Fig. 7. In area X, the retrogradely labeled neurons after tracer injection into the medial nucleus of the dorsolateral thalamus (DLM) exhibited strong glutamic acid decarboxylase (GAD) immunoreactivity. A: Left panel shows five tracer-labeled neurons in area X (arrows); right panel shows GAD immunoreactivity in the same area. Arrows in the right panel point to the same locations as those in the left panel. B: Top row shows optical sections through a single retrogradely labeled neuron. Bottom row shows GAD staining in the same planes as those shown in the top row. The retrogradely labeled neuron showed GAD immunoreactivity. In addition, GAD terminals tightly surrounded the soma (arrowheads) and the dendrites (arrows). Scale bars 25 µm in A, 20 µm in B; thickness 40 µm in A, 2 µm in B. To examine whether the retrogradely labeled neurons have the same size distribution as the strongly GAD neurons, we measured the soma size of 333 strongly GAD neurons within the same sampling areas in tissue from three animals. Of these neurons, 110 were also retrogradely labeled. We then compared the soma size of the retrogradely labeled neurons with that of the strongly GAD, nonretrogradely labeled neurons. Figure 8B illus-

10 GABAERGIC PROJECTION IN THE SONG SYSTEM 77 Fig. 8. The retrogradely labeled neurons in area X after tracer injection into the medial nucleus of the dorsolateral thalamus (DLM) have large somata. Although the glutamic acid decarboxylase positive (GAD ) neurons are large, they are smaller overall than the tracerlabeled neurons following tracer injection into DLM. A: Histogram of soma areas shows the large size of retrogradely labeled neurons in area X. B: Cumulative histograms show the soma areas for the projection neurons (triangles) and all other GAD neurons within the same optical sections (circles). trates a cumulative histogram of the soma areas of 110 retrogradely labeled neurons (triangles) and 223 GAD, nonretrogradely labeled neurons (circles). The soma area distribution for the GAD, nonretrogradely labeled neurons ( µm 2 ) was significantly different from that for GAD, retrogradely labeled neurons ( µm 2 ; Kolmogorov-Smirnov test, d 0.40, P 0.01). Because some of the nonretrogradely labeled cells could have been projection neurons whose terminals did not encounter or pick up the tracer, this difference is probably an underestimate. For double-labeled neurons, soma size measured by tracer labeling ( µm 2 ;n 110) did not differ from soma size measured by GAD staining ( µm 2 ; paired t-test, P 0.05), showing that these two methods of measuring soma size were consistent. These results indicate that strongly GAD, retrogradely labeled neurons are larger than strongly GAD, nonretrogradely labeled neurons. Thus, in addition to the large, GAD neurons projecting to DLM, there may exist large GAD interneurons within area X that are slightly smaller than projection neurons but are still much larger than the majority of area X neurons. Tracer injection into area X and GAD staining To determine whether the GAD terminals in DLM indeed arise from area X projection neurons, tracer was injected into area X and GAD staining was then performed. After tracer injection into area X, many somata in AVT, DLM, lman, and HVc were labeled with tracer, which is consistent with previous studies (Nottebohm et al., 1976; Lewis et al., 1981; Bottjer et al., 1989; Sohrabji et al., 1993; Nixdorf-Bergweiler et al., 1995; Vates and Nottebohm, 1995). Some sparsely distributed neurons in the areas dorsal and posterior to area X were also labeled by tracer. It is not clear whether these neurons indeed project to area X or whether their axons passed through the injection sites or along the track of the injection pipettes. Tracer-labeled axons entered DLM from the anterior ventrolateral border (Fig. 9B). These axons branched very close to their targets, where they usually enlarged to about 3 µm in diameter and broke into short terminals. These terminal fields covered an area of µm in diameter. In many cases each terminal formed a tight circle about µm in diameter (Fig. 9C, top row; Okuhata and Saito, 1987), similar to the baskets observed in DLM after GAD staining. The number of baskets formed by an individual anterogradely labeled terminal was usually one or two and never more than four. These tracer-labeled baskets thus appeared to surround a very small number of DLM somata. Even though some area X terminals contacting DLM neurons did not form baskets, a single area X projection neuron could still contact only a very limited number of neurons because its axon branched within a volume comparable to that occupied by five to eight DLM somata. Of four animals that had injection sites in area X, two showed GAD staining in all anterogradely labeled terminals in DLM, and two showed GAD staining in 80% of labeled terminals. Figure 9C shows one example of a GAD, anterogradely labeled terminal in DLM. Although a small number of terminals of area X projection neurons may not express GAD, the tracer-labeled, non-gad terminals in DLM in two birds more likely reflect damage to the system by current or tracer injection. In these two animals, the density of GAD terminals in DLM ipsilateral to the injected area X appeared to be somewhat lower than that of the contralateral DLM. In some regions of DLM, the density was so low that single GAD terminals could clearly be distinguished from the background, exhibiting similar morphology to that of anterogradely labeled terminals after tracer injection into area X. This was in contrast with the control hemisphere, in which many GAD terminals appeared entangled within DLM. The underlying explanation for this unusual GAD staining pattern is not clear. Neurons in area X may have been

11 78 M. LUO AND D.J. PERKEL Fig. 9. In the medial nucleus of the dorsolateral thalamus (DLM), the anterogradely labeled terminals after tracer injection into area X showed strong glutamic acid decarboxylase (GAD) immunoreactivity. A: Camera lucida drawing of a relatively small injection site in area X. B: Low power view of GAD staining (red) and tracer labeling (green) in DLM. In this section, only two terminals are visible, presumably because of the small injection site in area X. C: The top row shows a series of three optical sections through one tracer-labeled terminal in DLM (5 µm interval). The middle row shows GAD immunoreactivity in the same area as that shown in the top row. The bottom row is a combined pseudocolor view of the top and middle rows. Yellow pixels indicate areas of colocalized tracer and GAD immunoreactivity. The tracer-labeled terminals had strong GAD immunoreactivity. Scale bars 200 µm in B, 10 µm in C; thickness 40 µm in B, 2 µm in C.

12 GABAERGIC PROJECTION IN THE SONG SYSTEM 79 damaged by current or tracer injection, and this damage may have induced depletion of GAD in their terminals in DLM. Direct evidence that the baskets formed by the GAD terminals in DLM enclose DLM somata came from some experiments in which tracer was injected into posterior area X. Because the axons of DLM neurons projecting to lman pass through but do not terminate in the posterior portion of area X (Bottjer et al., 1989; data not shown), some DLM somata were labeled after tracer injection into area X. Figure 10 shows two retrogradely labeled somata in DLM, one of which was tightly enclosed by a basket of GAD terminals. The other soma had GAD terminals surrounding it, but these terminals did not form a closed basket. In two animals, we observed several examples of such retrogradely labeled DLM somata tightly enclosed by GAD terminals. Because HVc and lman provide the major inputs to area X (Nottebohm et al., 1976; Bottjer et al., 1989; Nixdorf- Bergweiler et al., 1995; Vates and Nottebohm, 1995), we examined the retrogradely labeled somata in these two regions after tracer injection into area X (n 4) to determine whether they are GAD. Of more than 100 tracer-labeled neurons in HVc (Fig. 11) or lman (data not shown) after area X injection, none were found to be GAD. However, many GAD somata and processes were observed in these two nuclei. Thus, the HVc and lman inputs to area X appear not to be GAD. DISCUSSION Area X = DLM projection is GABAergic The main finding of the present study is that the projection neurons from area X to DLM express GAD in adult male zebra finches. Results from four approaches taken together provide strong support for this conclusion. First, a sparsely distributed population of large neurons in area X were intensely GAD. In DLM, a dense meshwork of terminals, but no somata, were GAD. Second, ibotenic acid injections into area X abolished or dramatically reduced the number of GAD terminals in DLM. The loss of GAD terminals after an ipsilateral lesion, coupled with the fact that DLM receives a major input only from ipsilateral area X, suggests that all the GAD terminals in DLM arise from the area X projection neurons. Third, after tracer injection into DLM, all the retrogradely labeled neurons in area X were GAD. Fourth, after tracer injection into area X, almost all of the anterogradely labeled terminals in DLM were GAD. These data combined strongly suggest that the area X = DLM projection is GABAergic. The GABAergic input from area X to DLM is very strong and could have a powerful impact on the firing of DLM neurons. In DLM, axon terminals of area X projection neurons form GAD baskets surrounding DLM somata (Okuhata and Saito, 1987; present results). In other connections with calyceal (Jackson and Parks, 1982) or basket-like morphology, presynaptic firing has a profound effect on the firing of the postsynaptic neuron (Hackett et al., 1982; Borst et al., 1995; Isaacson and Walmsley, 1996). Thus, area X axon terminals occupy a strategic location on DLM neurons and may exert a strong physiological influence. In an overwhelming number of different classes of vertebrate neurons, GABA has an inhibitory action (reviewed in Nicoll et al., 1990; but see Wagner et al., Fig. 10. The glutamic acid decarboxylase positive (GAD ) baskets in the medial nucleus of the dorsolateral thalamus (DLM) surround the somata of DLM neurons. A: In some cases, following tracer injection into area X, some somata in DLM were retrogradely labeled, presumably because their axons passed through the posterior portion of area X en route to the lateral portion of the magnocellular nucleus of the anterior neostriatum. Shown are two retrogradely labeled somata (arrowheads) and one anterogradely labeled terminal (arrow). B: GAD staining of the same area as that shown in A. Arrowheads and arrow in A and B point to the same locations as those in A. Scale bar 15 µm; thickness 25 µm. 1997). Thus, although the physiological effect of the X = DLM synaptic connection remains to be elucidated, it seems highly likely that it will be inhibitory, perhaps providing a veto on the firing of DLM neurons. One or more

13 80 M. LUO AND D.J. PERKEL cotransmitters may also contribute to the actions of area X neurons on these in DLM. One limitation of the present study is the focus on adult finches. The best-studied role of the AFP in song learning has been limited to juveniles. Thus, it will be important to determine whether the X = DLM connection, which develops early (Johnson and Bottjer, 1992; Nordeen and Nordeen, 1997), is GABAergic from the time of its formation. Another limitation is that we could not completely rule out the possibility that a portion of the immunoreactivity observed represented cross reactivity to some related antigen. In addition, because two isoforms of GAD are found in a variety of vertebrate species (Erlander and Tobin, 1991; Ahman et al., 1996), including the zebra finch (Bosma et al., 1997), the single antiserum may not have labeled all GAD neuronal elements. However, Western blot analysis indicated that this antiserum recognizes two bands in zebra finch brain. Thus, we could not determine which cells express which GAD isoform(s). The different staining intensity in different cell populations could reflect different degrees of GAD expression or different affinities of the antiserum for each GAD isoform. A more detailed analysis of zebra finch GAD isoforms, with appropriately selective reagents, should clarify this issue. Fig. 11. Retrogradely labeled neurons in HVc after tracer injection into area X did not show glutamic acid decarboxylase (GAD) immunoreactivity. A: Retrogradely labeled neurons in HVc (arrowheads) after tracer injection into area X. B: GAD immunoreactivity in the same area as that shown in A. Arrowheads in A and B point to the same locations. Scale bar 25 µm; thickness 40 µm in A,B. GABAergic connections within area X In addition to the neurons projecting to DLM, the majority of the remaining neurons in area X could also be GAD. A number of large GAD neurons were not retrogradely labeled, possibly because of incomplete filling of DLM with tracer. However, on average, the cells that were retrogradely labeled were larger than the GAD neurons that were not retrogradely labeled. These nonretrogradely labeled, intensely GAD neurons were somewhat smaller than the projection neurons but were still larger than most of the area X neurons. In addition, the retrogradely labeled neurons were surrounded by GAD terminals, whereas many of the large GAD cells that were not retrogradely labeled were not surrounded by GAD terminals. This observation suggests that there may be one or more types of large, sparsely distributed, GAD neurons in area X that do not project to DLM and that they may receive different degrees of somatic GABAergic input. The vast majority of neurons in area X have small somata (Nottebohm et al., 1976; Bottjer et al., 1989; Sohrabji et al., 1993). Most of these cells were weakly GAD but some were intensely stained. Grisham and Arnold (1994) observed that the vast majority of neurons in area X exhibit GABA-LIR. They reported the soma area of labeled neurons to be about 35 µm 2 and did not mention any neurons larger than 150 µm 2. Thus, our data are consistent with those of their report and extend the finding to larger neurons in area X. Therefore, it is very likely that the vast majority of neurons in area X are GABAergic. Very complex GABAergic connections may exist within area X. The projection neurons receive GABAergic contacts surrounding their somata and proximal dendrites, whose source is not entirely clear. It is unlikely that they arise from external sources, such as HVc, lman, or AVT (Bottjer, 1993), because we did not observe GAD retrogradely labeled somata in these nuclei following tracer injection into area X. Furthermore, the input from lman arises from the collateral output of lman neurons projecting to RA (Vates and Nottebohm, 1995), which use glutamate as transmitter (Kubota and Saito, 1991; Mooney and Konishi, 1991; Mooney, 1992). The GAD terminals surrounding the area X projection neurons thus likely arise from other GAD neurons in area X, either projection neurons or interneurons. The projection neurons in area X have axonal branches bearing varicosities near their dendritic field. Thus, they could form GAD terminals on other neurons within the vicinity. However, we never

14 GABAERGIC PROJECTION IN THE SONG SYSTEM 81 Fig. 12. Summary of present results. Filled elements indicate glutamic acid decarboxylase positive (GAD ) somata or terminals. Question marks indicate possible connections about which we have little evidence. For example, it is not known which cell type or types in area X receive input from HVc or the lateral portion of the magnocellular nucleus of the anterior neostriatum (lman). The projection neurons of area X are large and GAD and ramify locally, possibly contacting small cells. Some or maybe all of the small interneurons in area X are also GAD but more weakly and may contact the large projection neurons. In addition, there may exist relatively large, intensely GAD neurons that do not project to the medial nucleus of the dorsolateral thalamus (DLM). The area X projection neurons have GAD terminals around their somata and proximal dendrites. The origin of these terminals is unknown. After reaching DLM, each projection neuron contacts a small number of neurons (possibly fewer than four) and in many cases forms a tight GAD basket around the somata of the DLM neurons. Therefore, it is very likely that the area X projection neurons receive strong inhibition and have a powerful inhibitory effect on the DLM neurons. observed a single varicose process of a projection neuron forming a basket. Although terminals from several projection neurons may be able to combine to form GAD baskets, we did not observe close contacts between the axonal branches of projection neurons and other projection neurons. It seems more likely that the GAD terminals surrounding the area X projection neurons arise from GAD interneurons within area X. Based on the present results, the circuits in area X and DLM are summarized schematically in Figure 12. The area X projection neurons receive strong GABAergic input, possibly from numerous small, weakly GAD interneurons within area X, and also provide a powerful GABAergic output to DLM neurons. There also appears to be a population of large, strongly GAD area X neurons that are slightly smaller than the projection neurons and not tightly enclosed by GAD terminals. In addition, the area X projection neurons also ramify within the region delimited by their dendritic tree and thus may form local inhibitory contacts, possibly onto area X interneurons. It is not known whether the inputs from HVc and lman contact area X projection neurons, interneurons, or both. Area X and basal ganglia In mammals, a loop (the so-called direct pathway) arises from cerebral cortex, passes through synapses in the striatum, globus pallidus, ventral anterior/ventral lateral thalamus, and returns to cortex (reviewed in Parent and Hazrati, 1995). In birds, a corresponding circuit arises in the neostriatum, which Striedter et al. (1998) showed is pallial in origin. From these areas arises a loop that passes through via PA/LPO, paleostriatum primitivum (PP), and thalamus, and returns to pallium (Medina and Reiner, 1995). Area X lies within the LPO, which has many features of mammalian striatum (Karten and Dubbeldam, 1973; Reiner et al., 1984; Medina and Reiner, 1995; Veenman et al., 1995). Furthermore, area X shows heavy dopaminergic innervation from AVT (Lewis et al., 1981; Bottjer et al., 1989; Bottjer, 1993; Soha et al., 1995), dopamine receptor binding (Casto and Ball, 1994), and expression of certain neuropeptides (Bottjer and Alexander, 1995). Area X could thus participate in this loop as a portion of striatum. However, if this is the case, area X has an unexpected projection to thalamus instead of to PP, which corresponds to globus pallidus in mammals (Karten and Dubbeldam, 1973; Reiner et al., 1984; Medina and Reiner, 1995). Moreover, mammalian striatal projection neurons, the so-called medium spiny neurons (MSN), project to globus pallidus but not to thalamus, form the vast majority of striatal neurons, are densely packed, and are heavily spiny (reviewed in Parent and Hazrati, 1995). In contrast, the area X projection neurons are large, sparsely distributed, and appear to have a relatively low density of spines. Bottjer (1993) pointed out this apparent paradox and suggested that it could be resolved at least in part if area X had properties resembling both striatum and pallidum. Our results support this idea and provide further detail. For example, the small neurons in area X may correspond to the MSN and the large projection neurons correspond to pallidal neurons. Such a scenario predicts specific internal connectivity within area X, e.g., a direct connection from the small neurons to the large projection neurons. No