The constitutive differential transcriptome of a brain circuit for vocal learning

Analysis of cDNA and oligo microarray datasets

Our study consisted of 4 differential microarray screening experiments aimed at identifying molecular markers of HVC versus shelf (the area immediately ventral to HVC), RA versus ventrolateral arcopallium (VLA), striatal Area X versus ventral striato-pallidum (VSP), and nXIIts versus the supra-spinal medullary nucleus (SSP). In each case, samples were obtained from quiet, non-singing birds, which allowed us to assess the constitutive transcriptome. For song nucleus HVC, we analyzed a set of 12 spotted glass cDNA microarrays (ESTIMA:Song collection “20 k” cDNA microarrays, as detailed in [37]) previously used to screen for differentially expressed mRNAs in laser capture microdissected samples from HVC and the adjacent nidopallial shelf (n = 6 per sample; for details on sample preparation, mRNA isolation, probe amplification, labeling and hybridization, see Lovell et al. [24], Fig. 1). In the present study, we reanalyzed the datasets deposited for the HVC/Shelf study that are publically available from the Gene Expression Omnibus (GEO; GSE36270). For our reanalysis, we used a more robust data normalization and statistical analysis based on the R statistical package from BioConductor (http://www.bioconductor.org/). Briefly, the limma (Linear Models for Microarray analysis) R package [38] was used to read in the median foreground and median background fluorescence intensities. Background correction was performed using the “half” method, which subtracted the local background estimates from the foreground values, and then set any zero or negative values to 0.5 to avoid losing data. Within-array normalization was done using the “printtiploess” method to remove the dye intensity biases within each array. The resulting intensity values are in the form of M-values, which are log2(Cy5 / Cy3). However, all the experiments were done as a “common reference” design, where each sample was hybridized against the same common reference sample to facilitate comparisons across experiments. Dye-balancing was also performed, so that in half the arrays the reference was in the Cy5 channel in half the reference was in the Cy3 channel. Therefore, instead of M-values, we transformed the intensities so they were all log2(sample/reference). Between-group comparisons were calculated using the limma package in Bioconductor which uses a modified t-test to calculate p-values using an empirical Bayesian method to moderate the standard errors of the estimated log-fold changes [39]. Limma uses variance information from all the spots on the array to arrive at an estimate of per spot variance used in the t-tests. p-values were adjusted for multiple comparisons with the program’s q-value [40], which allows for selecting statistically significant genes while controlling the estimated “false discovery rate”.

For song nuclei RA, nXIIts, and Area X we performed statistical analyses of pre-normalized data sets that are publically available from NCBI Gene Expression Omnibus (GEO). Specifically, we analyzed sample sets that were previously hybridized to the Agilent Technologies Duke Taeniopygia guttata 45 K oligo array (for analysis of RA and nXIIts) or the 20 K Agilent-019785 Custom Zebra Finch Microarray (for the analysis of Area X). Complete descriptions of brain tissue sampling, mRNA isolation, oligo array hybridizations, gene expression normalization, and partial outcomes are available for nXIIts [35], RA [23], and Area X [32]. For our present analysis of these oligo arrays we used the GEOquery and limma R packages (NCBI Gene Expression Omnibus; [41] with the Benjamini & Hochberg false discovery rate method to identify differentially expressed oligos from the user-supplied normalized datasets. For RA, we analyzed 3 pairs of zebra finch samples from project GSE28395 (Avian arcopallium; [35] and performed ANOVA group comparisons between normalized expression values from RA (GSM701893, GSM701894, and GSM701895) and the ventrolateral arcopallium (VLA; GSM701896, and GSM701897, GSM7011898). For nXIIts, we analyzed 3 pairs of zebra finch samples from project GSE33667 (Avian 12th motor neuron; [23]) and performed ANOVA group comparisons between normalized expression values from nXIIts (GSM832502, GSM832503, and GSM832505) to those from the non-vocal supra-spinal medullary motor nucleus (SSP; GSM832501, and GSM832504, GSM832506). For Area X, we analyzed a subset of 9 quiet, non-singing control zebra finch sample pairs from project GSE34819 [32] and performed ANOVA group comparisons between samples from Area X (GSM855822, GSM855826, GSM855830, GSM855838, GSM855842, GSM855846, GSM855850, and GSM855854) and the ventral striato-pallidum (VSP; GSM855823, GSM855827, GSM855831, GSM855839, GSM855843, GSM855847, GSM855851, and GSM855855). Examples of the brain regions laser captured for RA vs. VLA, and Area X vs. VSP are presented in Pfenning, [23] (Fig. S17) and Hilliard, [32] (Fig. 1), respectively.

Establishment of Significance Cut-offs based on ZEBrA expression data

Efforts to identify significantly differential genes based on microarrays typically rely upon ‘blind’ analyses using ANOVA statistics with false-discovery rate correction. While informative, we have been able to improve on this approach by analyzing high resolution in situ hybridization images available on the Zebra finch Expression Brain Atlas (ZEBrA; www.zebrafinchatlas.org) for differential and non-differential genes, to establish validated p-value cut-offs for each anatomical comparison. We first examined expression for the entire set of genes with in situ hybridization data available at ZEBrA (n = ~ 490 genes) and identified 320 genes where the data allowed us to assess each nucleus of interest and its adjacent tissue. For each nucleus (with the exception of nXIIts) we then defined subsets of genes that were either differential (high or low) or expressed but non-differential, when compared to the respective adjacent regions of contrast (photomicrographs in Fig. 2a-c). A summary of the subsets of differential vs. non-differential genes that were analyzed for each nucleus is presented in Additional file 1: Table S2. In cases where no signal was detected in either brain region, labeling was inconsistent within or between sections, folds or artifacts were present that obscured evaluation, or the presence of the nucleus or its anatomical comparison could not be established, the in situ data for that gene was removed from further analysis. For each nucleus, we next retrieved the p-values for every oligo on the array that corresponded to a gene from the differential and non-differential gene sets. We then constructed p-value frequency histograms (Fig. 2a-c) for each of these gene sets, and plotted both differentials and non-differentials on the same graphs. We then established a high-confidence cut-off corresponding to a p-value that captured the highest proportion of true vs. false positives (vertical dotted lines in graphs in Fig. 2a-c). These p-values were used as validated cut-offs to define the universe of oligos that were analyzed in the subsequent scoring analysis.

Bioinformatics

Microarray Analysis and Probe Curation

We conducted a differential screening analysis to identify molecular specializations of 4 major song nuclei (HVC, RA, nXIIts, and Area X) using data from cDNA and oligo microarrays hybridized with laser-capture samples from these nuclei and adjacent areas. Although we made use of datasets that are publicly available in GEO, an extensive curatorial effort was necessary to remove a large number of systematic errors present in these datasets. First, to remove unreliable probes, we aligned the sequences for the complete sets of ESTs and oligos on the microarrays to the current zebra finch genome assembly (taegut1; WUGSC 3.2.4; [49]). This step revealed that 5518 (~ 13%) oligos and 1507 (~ 8%) ESTs were unreliable due to: (a) multiple high-scoring alignments, indicating a lack of specificity, (b) non-informative alignments (i.e. low-scoring or no alignment, or alignments exclusive to chromosome Unknown) that precluded gene identification, or (c) evidence of artifacts of mis-priming or cloning in the construction of the cDNA databases (Additional file 1: Table S1). All such problematic oligos and ESTs were removed from subsequent analyses.

At several stages of our analysis we noticed that a large number of oligos were either unannotated or mis-annotated. We therefore performed an extensive manual curation effort and directly examined the annotations for 9305 oligos (~ 22% of oligos on the oligo array; Additional file 1: Table S1). Of these, we corrected 2735 annotations that were incorrect (~ 29% of the oligos examined), updated the names of 2440 genes to current HGNC gene terms, and identified 1466 differential oligos with unknown orthology, including several that possibly represent novel genes within the songbird lineage. A complete analysis of the unknown subset will require further curatorial efforts, and will be presented in a separate study.

Gene Over-representation Analysis

Other broad pathway annotations were shared by all nuclei except for RA, reflecting the much smaller number of specific pathways present in the latter. These included G-protein coupled receptor (GPCR) signaling, calcium and non-calcium intracellular signaling, and axon guidance (Table 2C-F). However, while these nuclei all had pathways within these general categories, no specific pathways were shared across all nuclei. For instance, GPCRs for ‘metabotropic glutamate receptors’ were exclusively featured in HVC, ‘rhodopsin-like receptors’ were shared features of HVC and Area X, and nXIIts features no fewer than 6 different G-protein related receptor pathways. Lastly, other specific pathways were uniquely over-represented in individual nuclei (Table 2K). Noteworthy are the glycosaminoglycan pathway for HVC, and a broad diversity of signaling and metabolic pathways for nXIIts.

To gain further insights into features unique to individual nuclei or shared across nuclei, we constructed a Venn diagram with lists of differential genes for each song nucleus and determined all possible intersections (Fig. 3; gene sets from the main intersections are presented in Additional file 1: Table S15). This analysis revealed that collectively the nuclei shared 16 (~ 1%) markers, which were enriched for ‘cardiac conduction’ and ‘potassium channels’, as well as pathways related to molecular interactions at synapses (Table 3A). These results differed significantly from analysis of the marker set shared by telencephalic nuclei, which showed enrichments in signaling and structural pathways (Table 3B), as well as from analysis of markers shared by all DMP nuclei, which showed enrichments in reelin-related pathways and calcium- and IP3-related signaling pathways (Table 3C). In contrast, the majority (70%) of markers were unique to individual nuclei (Fig. 3), reflecting a high degree of molecular specialization across nuclei. Close examination of related pathways (Table 4) showed enrichments related to sugar and amino acid metabolism in HVC (Table 4A), inwardly-rectifying potassium channels but otherwise a paucity of enriched pathways in RA (Table 4B), and a diversity of signaling and structural pathways in nXIIts (Table 4C) and Area X (Table 4D).

A. Pathways shared by all nuclei

Cardiac conduction

Muscle contraction

Potassium Channels

Parkin-Ubiquitin Proteasomal System pathway

Class A/1 (Rhodopsin-like receptors)

Interactions of neurexins and neuroligins at synapses

B. Pathways shared by telencephalic nuclei

CDC42 signaling events

Regulation of Actin Cytoskeleton

PDGFR-beta signaling pathway

RHO GTPases Activate WASPs and WAVEs

RAC1 signaling pathway

C. Pathways shared by nuclei of the DMP

Reelin signalling pathway

Signaling by GPCR

VLDL interactions

Calmodulin induced events

CaM pathway

Stathmin and breast cancer resistance to antimicrotubule agents

Ca-dependent events

Endochondral Ossification

ECM-receptor interaction - Homo sapiens (human)

DAG and IP3 signaling

EGFR interacts with phospholipase C-gamma

PLC-gamma1 signalling

Hemostasis

Lissencephaly gene (LIS1) in neuronal migration and development

A. Pathways unique to HVC

Glycine degradation

Toxicity of botulinum toxin type A and E

Pentose Phosphate Pathway

Glucose-6-phosphate dehydrogenase deficiency

Ribose-5-phosphate isomerase deficiency

Transaldolase deficiency

Amino Acid metabolism

Ammonia Recycling

B. Pathways unique to RA

Inwardly rectifying K+ channels

Oxidative Stress

C. Pathways Unique to nXIIts

Fatty acid elongation

Endocytotic role of ndk phosphins and dynamin

Signaling by NOTCH2

agrin in postsynaptic differentiation

MET activates PTK2 signaling

GPCR signaling-G alpha s PKA and ERK

Internalization of ErbB1

Signaling by GPCR

D. Pathways Unique to Area X

role of mef2d in t-cell apoptosis

Response to elevated platelet cytosolic Ca2+

CDO in myogenesis

mechanism of gene regulation by peroxisome proliferators via ppara

Role of Calcineurin-dependent NFAT signaling in lymphocytes

Platelet degranulation

Netrin-1 signaling

Cell adhesion molecules (CAMs)

Gene family analysis

We next examined the differential expression of gene families involved in basic neuronal functions that likely influence song system organization and physiology (Figs. 4, 5, 6, 7, 8 and 9). Besides genes found by microarrays to be differentially expressed in song nuclei, we made an effort to also identify non-differential family members for a comprehensive assessment. In situ hybridization of available probes helped confirm the microarray data, visualize the brain-wide distribution of markers, and examine song nuclei not assessed by microarray analysis (representative in situ images in Figs. 10, 11, 12, 13 and 14). For some specific genes, and for LMAN, DLM, and DM, in situs were the only source of data. The majority of genes examined were differentially expressed in at least one nucleus, and all nuclei showed differential expression of several genes in each family. In situs and microarrays were largely consistent, indicating the data were robust across experiments and methods, but some divergences were also noted and are indicated in Figs. 4, 5, 6, 7, 8 and 9 by a ‘$’ symbol.

Molecular Specializations of the Song System

Ion channel physiology, which is critical for many aspects of neuronal function, emerged from both the bioinformatics and Venn diagram analyses as a major target of regulation in the song system (Fig. 3; Table 3). This outcome is consistent with our previous findings of widespread regulation of potassium channel genes in the song system [25], and likely reflects a need for the song system to tightly regulate features of cellular excitability and spiking fidelity. The present study also reveals that members of the sodium-, calcium-, chloride-channel gene families are important targets of regulation in the song system. A comprehensive analysis of these gene families and their brain expression is being prepared as part of a separate study (Friedrich et al., in preparation). Of note, the specific complements of differential ion channel genes differ across nuclei, suggesting that each nucleus uses a distinct set of mechanisms to regulate features related to excitability and spike timing. Overall, the lists of differential ion channel genes in this study represent fundamental reference sets for future pharmacological and genetic manipulations.

By far the most prominent shared enrichments were for GPCR-related pathways (Table 2B). Despite the shared pathway annotation, however, closer inspection of the complements of genes underlying these GPCR-related specializations revealed they were quite different. For example, while striatal Area X had several dopaminergic receptors, pallial HVC was enriched for cholinergic receptors. These findings are consistent with the known dopaminergic and cholinergic inputs and modulations of these respective nuclei [51–53]. Thus, the specific genes identified represent well-supported candidate targets for gene and pharmacological manipulation to better understand the contribution of these neuromodulatory pathways to vocal learning.

Other more general trends included shared enrichments in pathways related to neuromodulation, neurotransmission, and peptidergic modulation, and within the nuclei of the DMP, a shared enrichment of genes in reelin-related pathways. The fact that both RELN and its VLDLR and LRP8 receptors are differential markers of DMP nuclei suggests the possible involvement of the reelin pathway in the formation of DMP connections (discussed in more detail below). The low pathway diversity in nucleus RA is also intriguing, given the large number of differential markers of this nucleus. Possible explanations include a lower diversity of cell types compared to other nuclei or a higher representation of genes per pathway, either of which would suggest that RA is highly specialized compared to other song nuclei.

Axonal guidance and synaptic plasticity genes

The observed patterns of axonal guidance cues genes are potentially of relevance for understanding the regulation of connections across nuclei, but also raise questions about ligand-receptor pairing rules in this system. Among semaphorins and plexins, there is precedence for SEMA3E expression in presynaptic terminals having repulsive interactions with post-synaptic terminals expressing the PLXND1 receptor, but co-expression of PLXND1 with NRP1 can flip this interaction to be attractive, thus making a co-expressing nucleus an attractive target for SEMA3E-expressing terminals [54, 55]. Indeed, interactions between SEMA3E and PLXND1 play an important role in establishing specificity of cortico-thalamo-striatal connections [56]. Intriguingly, SEMA3E shows remarkably high expression in HVC and LMAN, and thus could be involved in influencing the projections of these nuclei to their common targets RA and/or striatal Area X. Consistent with this possibility, NRP1 expression is high in Area X. However, PLXND1 expression in Area X and in the striatum, as a whole was quite weak, and expression of the PLXND1/NRP1 co-receptors was not particularly marked in RA either. On the other hand, PLXND1/NRP1 are both highly expressed in HVC, which might make HVC attractive to inputs that express SEMA3E. Unfortunately, SEMA3E expression in nuclei that project to HVC (NIf, Uva, MMAN) has yet to be determined. Alternatively, there is precedence in developing sensory-motor circuits in the mammalian spinal cord for post-synaptic SEMA3E interacting with pre-synaptic PLXND1 and/or NRP [55, 56]. Under this scenario, the high expression of SEMA3E in HVC and LMAN might help to regulate inputs to these nuclei. However, this scenario seems unlikely because prominent expression of PLXND1/NRP1 appears largely lacking in nuclei that provide inputs to LMAN or HVC. In short, more work is needed to determine whether and how this particular axon guidance system operates in birds.

The high expression of PLXNA4s and NRPs in HVC and RA suggests these nuclei might be attractive targets for terminals expressing their corresponding SEMA3-type (A/C/D/F) ligands. Notably, the patterns in zebra finch are consistent with those in another vocal-learner, the Bengalese finch [57], suggesting a conserved program of axon guidance cues in the songbird lineage. While data to test this idea are not yet available, one would predict high expression of these ligands in nuclei (e.g., L/MMAN, Uva, NIf) that provide inputs to PLXNA4-expressing nuclei. SEMA7A was found to be a positive marker of all song nuclei examined, consistent with its prominent role in the maturation of mammalian cortical circuits [58]. Intriguingly, the down-regulation of the PLXNC1 receptor would make projection targets receptive to SEMA7A-positive terminals if this ligand-receptor interaction were repulsive, as in other systems [59]. Under this scenario, SEMA7A-expressing axonal terminals from a given input (this could apply to several song nuclei) would interact with PLXNC1-expressing cells outside the target nucleus and be inhibited from making contact, and/or compelled to continue towards the target by other attractants. For example, SEMA7A interacts prominently with integrin receptors, which would make the high ITGA1 expression in RA and nXIIts relevant for the HVC-to-RA and RA-to-nXIIts projections. Similarly, the high expression of PLXNC1 in DLM shell compared to its core (Fig. 14d) could play a role in the specificity of the X-to-DLM projection. The patterns observed for SEMA5A and 6D and their PLXNA4 receptors suggest similar roles in the projections of the DMP as for SEMA7A. Interestingly, the pattern of SEMA6A in zebra finch is consistent with the prominent role of this SEMA in guiding descending fibers from PLXNA4-expressing corticospinal neurons in the brainstem in mammals [60]. In this context, the high expression of PLXNA4 in RA and the prominent expression of its ligand SEMA6A in several brainstem nuclei, but down-regulation in nXIIts could help to regulate the specificity of the RA-to-nXIIts projection. In contrast to all above, the SEMA4s/PLXNBs patterns suggest a lack of prominent roles in song system projections.

Cadherins and protocadherins are a large family of genes that are involved in cell-cell adhesion and play important roles in neuronal connection formation. Our data suggests some specific song system projections where CDHs might play a role through homophilic interactions. This includes CDH6 (HVC-to-X projection), CDH13 (RA-to-nXIIts projection), and CDH9 (all projections of the DMP). The overall trend, however, was down-regulation, suggesting several CDH-related pathways are shut down in the adult zebra finch song system. Previous studies have shown differential expression of some cadherins in song nuclei of Bengalese finches [61], or explored a potential role of cadherins in vocal control [62]. While we report on a much larger set of genes in this family, consistent observations across studies include: a) the up-regulation of CDH6 (a.k.a. Cad6b) in HVC, Area X, and DM, b) up-regulation of CDH9 in HVC, c) down-regulation of CDH4 (a.k.a. Rcad) in all major forebrain song nuclei and in nXIIts, and d) down-regulation of CDH7 in RA. However, we did not observe marked up-regulation of CDH6 in RA and LMAN, nor the up-regulation of CDH10 in RA, as reported in Bengalese finches. While these could be real species differences, they could also derive from different behavioral conditions, or uncharacterized transcript variants. CDH1 was previously reported as a positive marker of HVC (Lovell et al. [24]), however a close examination of alignments to the now available zebra finch genome reveals that the reported clone is GAS8 and that CDH1 was not evaluated.

The high expression of RELN in DM and of the RELN receptors VLDLR and LRP8 in nXIIts suggests a possible role in the DM-to-nXIIts projection- an important pathway for vocal-motor patterning. Because both RELN and VLDLR are known targets of FOXP2, this brainstem vocal projection could represent a still unexplored pathway (outside of the basal ganglia) for FOXP2 modulation of vocal behaviors. Alternatively, the differential expression of RELN in DM versus nXIIts could play a role in the specificity of the projections from RA to these two brainstem targets. Our data are largely consistent with previous observations in canaries [63], although we did not detect up-regulation in Area X. Overall, the in situ data support the bioinformatics indication of the reelin pathway (Table 3C) as an important target of regulation, and the marked RELN down-regulation suggests suppression may be important for the maintenance of the song system circuitry. As for SLIT/ROBO genes, our data support previous observations of convergent specializations between RA and the human primary laryngeal motor cortex [23], and of differential regulation of SLITs and ROBOs in nuclei of the vocal-motor circuitry [64]. ROBOs are generally thought to be axonally-expressed receptors that have repulsive interactions with target-secreted SLIT ligands. Thus, the down-regulation of SLIT1 in RA and nXIIts could facilitate the targeting of ROBO-expressing terminals from HVC-to-RA and from RA-to-nXIIts, respectively. ROBO1 is largely down-regulated in RA but is highly expressed in a sparse cell population, which could correspond to a subset of projection neurons. Alternatively, the SLIT down-regulation in HVC and RA, coupled with the up-regulation of ROBOs in RA and nXIIts, might play a role in the targeting specificity of the HVC-to-RA and the RA-to-nXIIts projections. This notion has been previously suggested as helping to establish the specificity of the RA-to-nXIIts projection [64], a connection critical for learned vocalizations, but it would require the pre- and post-synaptic localizations of SLITs and ROBOs, respectively. The lack of prominent expression of SLITs or ROBOs in DM further supports a differential role of these genes in the RA-to-nXIIts projection. Lastly, the prominent differential regulation within these gene families in AFP nuclei (Area X, DLM, LMAN) suggests further roles in other parts of the song system.

The marked differential regulation of UNC5s in all song nuclei, including RA and its targets DM and nXIIts (Fig. 6c), suggests a prominent role for this pathway in song circuit connectivity. Interestingly, axonal terminals that express UNC5 receptors are repulsed by their main known ligands (netrins; [65, 66]). Our limited data suggest a lack of prominent regulation of netrins in the forebrain, but expression analysis of other netrins and examination of brainstem nuclei would be important next steps to establish the role of UNC5s. We also note that FLRTs, which provide alternative non-netrin repulsive cues to UNC5 receptors [67], and NEO1 (a.k.a. DCC), which has been implicated in chemo-attraction of netrins [68], are respectively up- and down-regulated in HVC and its targets, further suggesting UNC5-related effects at multiple sites in the song circuitry. Other classes of guidance cues (NTNGs-, NLGNs- and RTN4-related) showed down-regulation trends, with no apparent coordinated expression of known ligand-receptor pairs between connected song nuclei. We did not observe an enrichment of NTNGs in thalamo-cortical projection neurons (e.g. DLM) as reported in mammals [50, 69]. Lastly, the general down-regulation of No-Go-related cues and receptors (LINGOS, RTN4s) suggests that suppression of this axonal guidance pathway is a prominent feature of the mature song circuitry.

The marked differential expression of axonal guidance and connectivity cue genes in adult song nuclei suggests that the maintenance and/or plasticity of song system projections is under active regulation in adults. In several cases we did not see coordinated regulation of known ligand-receptor pairs in connected nuclei, suggesting that not all ligand-receptor interactions in songbirds conform to known rules of axon guidance. We note that most such rules derive from work in non-avian organisms that do not possess brain pathways for learned vocalizations. Thus, the patterns observed in finches might reflect axon guidance rules that are unique to birds, or related to the emergence of the song system in songbirds. A deeper understanding of how axonal guidance cues modulate song system connections requires developmental studies, a definition of the cellular distribution of proteins in pre-synaptic vs. post-synaptic structures, and gene manipulations to test specific hypotheses.

Excitatory and inhibitory neurotransmitter receptors

Significant insights were also gained from the analysis of inhibitory and excitatory neurotransmitter receptors. Consistent with their general roles, GABAergic and glutamatergic receptor families were broadly expressed in all song nuclei, but several genes showed evidence of regulated expression. For GABAergic receptors, the non-differential expression of the more abundant and fundamental alpha subunits of type-A receptors contrasted markedly with the differential regulation of multiple beta, delta, and epsilon subunits, suggesting these less prevalent subunits are more prominent targets of differential regulation in the song system. In particular, the up-regulation of delta subunits in all major telencephalic song nuclei, consistent with a previous report [70], suggests a distinctive feature of inhibitory neurotransmission, but its functional significance is unknown. The data on glycine receptor subunits is also intriguing, as the patterns of alpha and beta subunits seem largely complementary rather than overlapping. These receptors play important roles during development, and in adults tend to be more constrained to brainstem regions, including auditory nuclei [71–74]. Since functional receptors depend on the presence of both alpha and beta subunits, the patterns in adult finches suggest that they may play other roles, rather than reflecting functional glycine receptors.

The glutamate receptor patterns were largely consistent with previous data [26], both studies pointing to the differential regulation of multiple subunits at multiple sites as a distinctive feature of the song system. The possible implications of the observed patterns for regulating firing, deactivation, and plasticity within song nuclei (discussed more extensively in Wada et al. [26]), are not clear in the absence of pharmacological or genetic manipulations. While the present study presents the first expression data on delta-subunits in the song system, other differences between studies include the differential expression of GRIA3 in HVC and GRM4 in LMAN in the present study only, and of GRIK2/GRM8 in LMAN, GRM1/GRM5 in Area X, and GRM8 in HVC, seen by Wada et al. only. We also note that gene down- or up-regulation in distinct cell populations is more readily apparent with the non-radioactive approach used here (e.g. Fig. 12h, GRID1 in Area X; Figs. 10k-m, GRM1, GRID1, and GRID3 in HVC). Moreover, there is evidence of alternative splicing among zebra finch glutamate receptors subunits (e.g. Wada et al. [26], Fig. 5), which could further contribute to observed differences between the studies.

The expression of neuromodulatory receptors was consistent with and extends previous studies, and may help explain the differential ligand binding in autoradiographic assays. For example, the high expression of ADRA2A in HVC, LMAN, and to a lesser extent Area X, provides a likely basis for the high binding of alpha-2 adrenergic ligands to these nuclei [75]. However, the high alpha-2 ligand binding to RA and the modulation of RA via alpha-2 agonists/antagonists [76] remain unexplained, as ADRA2A was non-differential and ADRA2C was down-regulated in RA. The data on dopaminergic receptors are consistent with the more extensive study by [77], but the down-regulation of DRD2 in RA and up-regulation of DRD5 in HVC were not previously seen. With regards to DRD7, a gene markedly differential in several song nuclei, we adopted a term that better reflects its presence throughout the vertebrate phylogeny (DRD7 in Danio rerio, previously D1DR in finch [77] and DD1CR in chicken) and absence in mammals. As for serotonergic receptors, the marked up-regulation of multiple subunits in forebrain nuclei (except for HTR1B down-regulation HVC) points to serotonergic modulation as a major feature of the song system, consistent with pharmacological and electrophysiological studies (e.g. [78]). For cholinergic receptors, the data confirm the high expression of CHRM4 in Area X previously seen by Lovell et al. [24], and provide a basis for the higher binding of a muscarinic cholinergic antagonist to Area X compared to the surrounding striatum [79]. In contrast, the low expression of CHRM5 in RA might contribute to the low ligand binding described for this nucleus. We have also confirmed the high expression of CHRNA5 and CHRNA7 in HVC (previously seen by Lovell et al. [24]). This is consistent with the high alpha bungarotoxin binding for these nicotinic receptors in HVC [80]. In situ hybridization shows that CHRN4 is highly expressed in a sparse cell population within Area X. Of note, genomic alignments show that the cDNA clone annotated as CHRNA2 in Lovell et al. [24] actually corresponds to CHRNA4, thus the CHRNA2 expression pattern remains unknown in zebra finch. Overall, the data point to HVC and Area X as major targets of modulation by cholinergic inputs, particularly via CHRNA7 receptors, consistent with previous pharmacological and physiological data [81].

Our data are also consistent with and expand upon previous reports of peptide-related regulation in the song system based on gene expression and immunostaining data [24, 82–84]. In fact, peptide-related patterns provide some striking examples of differential regulation, suggesting important contributions to the physiology of the song system. They also often occur in sparse cell populations, suggesting roles in neuronal cell type diversification. We note that differential SST expression was not detected in HVC or RA, which contrasts with the described immunostained somata in these nuclei in comparison with the adjacent tissues [83]. Further, VIP was down in RA and Area X, consistent with [85], but PENK (MET) was not up in RA as expected based on the peptide distribution, suggesting that differential translation plays a role in this differential regulation. TAC1 is a remarkable marker of Area X (also reported in [86]), but is also highly expressed in nucleus accumbens (www.zebrafinchatlas.org). This finding likely explains why TAC1 was not detected as differential in Area X on the array. We also note that expression of opioid receptors in song nuclei was not detectable with our in situ protocol. This differs from reports suggesting that mu-opioid mRNA expression is not only detected, but marginally higher in HVC and RA than in surrounds [87]. We note, however that this previous report contradicts previous ligand binding autoradiographic studies in another songbird (Dark-eyed Junco) that reported no apparent differences in binding of mu-opioid ligands in the song system [88].