We have applied a comprehensive genome analysis strategy to identify the full complement of K-Channels and related genes in the zebra finch genome, and in situ hybridization to determine their brain expression, focusing on nuclei involved in birdsong production and learning. We chose humans as a starting point for K-Channel identification because they possess the largest and most completely described set of K-Channel genes. Furthermore, songbirds and humans share vocal learning, a rare trait that requires specialized vocal control brain structures that have been identified in both species. Our analysis revealed that a very large cohort of human K-Channel genes is also present in zebra finches, supporting a high degree of conservation across the two species, but we also identified several human genes that are missing in finches and some novel K-Channel genes that are present in finches but absent in humans, most of the latter representing conserved genes in non-mammalian vertebrates. Strikingly, nearly every major K-Channel sub-family (e.g. KCNA, KCNC, KCNQ, etc.) had at least one gene member that showed some degree of differential expression (higher or lower) within a song nucleus compared to its surround, supporting the notion that targeted gene regulation within this family may help shape the differential biophysical and excitable properties of vocal control areas in songbirds. Here we discuss the implications of our findings in the context of songbird biology and the evolution of circuitry for learned vocalizations.
An improved approach for identifying orthologous gene sets
In spite of a completed draft genome assembly, well-curated sets of orthologous genes are not yet available in zebra finches. Our approach to identify the full complement of K-Channel genes in the current finch assembly consisted of first compiling a list of all protein coding genes recognized as K-Channels in humans, then BLAT-aligning the predicted protein sequences to the genome assemblies of zebra finches and other vertebrates of comparative interest, followed by orthology confirmation through synteny analysis. This approach allowed us to determine which subset of K-Channel orthologs in zebra finches are correctly annotated by Ensembl, but also to identify and correct several erroneously annotated models, and some relevant loci with no predicted models. Most of the latter are in regions of low quality sequence that also contain large or several sequence gaps, often interrupting open reading frames. Our analysis also helped to confirm and/or establish several gene gains or losses within the vertebrate phylogeny. We are confident to have identified the full complement of K-Channel genes in the current finch assembly, since we systematically re-BLAT-aligned every identified finch locus to the finch genome, to ensure no loci were being missed due to low cross-species sequence conservation. This was important as some unannotated loci could only be detected through alignments to other members of the same sub-family. We believe our approach provides a highly sensitive strategy for retrieving full ortholog sets, especially from low-coverage genomes.
In search of finch-specific genomic features of K-Channel genes
The dramatic radiation of K-Channel genes in vertebrates, likely the result of ancient whole genome and chromosome duplications, as well as wholesale rearrangements , may have played a role in the rapid evolution of brain structures specialized for complex behaviors, including vocal learning. At the level of individual genes, changes affecting gene function may have included mutations, insertions and/or deletions in protein-coding sequences, changes in cis- and trans-regulatory domains, and even wholesale gene duplications/deletions. One of our goals was to determine whether the K-Channel gene family possesses features that are conserved in zebra finches (and by extension songbirds) and humans, thus pointing to possible physiological requirements associated with vocal production and learning. Specifically, we reasoned that we might find evidence for conserved K-Channel gene duplications/expansions, and/or protein coding insertions/deletions in zebra finches and humans, but absent in vocal non-learning mammals and birds (e.g. rodents, chicken).
Consistent with our hypothesis, we found that finches and humans share the vast majority of protein coding K-Channel genes, including representatives from each of the 21 major K-Channel sub-families (e.g., the 6TM and calcium-gated, the 2TM, the 4TM, the brain cyclic-nucleotide gated K-Channels, and the K-Channel tetramerization proteins). Furthermore, of the 23 human genes that appear to be missing in finches, 3 represent pseudogenes that likely do not influence cell physiology, and 7 may have been undetected in finches due to gaps in the current assembly. Thus out of 123 human K-Channel genes, only 13 (~10%) appear to be actually missing in songbirds, indicating a high degree of conservation in this gene family between these two greatly divergent species of vocal learners. Moreover, based upon consultation to the Allen Mouse Atlas of brain gene expression , 4 of these (KCNA7, KCND1, KCNK6, and KCTD13) do not appear to be expressed in the mammalian brain, 2 (KCNAB3 and KCNC3) are widely expressed throughout the brain, suggesting a general functional role, and 3 (KCNN4, KNCJ14, and KCNH3) exhibit regionalized expression that largely overlaps with that of other members of these sub-families, suggesting functional redundancy in their brain distributions. Thus, it appears that the majority of genes that are missing in songbirds (or that are novel in mammals), are either not expressed in the brain, or have redundant functions with other family members that are present in birds.
Despite the apparent conservation among vocal learners, we also discovered that the complement of K-Channel genes is highly conserved between zebra finch and chicken, the latter species being incapable of vocal learning. For instance, we found just a single finch gene, KCNE1P that is absent in chicken, and since it is a pseudogene, it is not expected to influence neuronal excitability. In fact, nearly every novel K-Channel gene in finch that is absent in humans is present in chicken and/or other vertebrate taxa, suggesting that many of these genes represent relatively ancient gene duplications that were likely lost in some or in all mammals (Additional file 1: Table S1). Thus, it seems unlikely that K-Channel gene duplication events have played a prominent role in the evolution of neural circuitry for learned vocalizations. We next looked for evidence of protein coding insertions/deletions (indels) that might change the function of channels in zebra finches as compared to chickens. Intriguingly, we identified 16 K-Channel genes that contain finch-specific indels within their predicted protein coding domains. However, only two of these K-channels (KCNA4 and KCNA6) had indels within a region containing a predicted functional domain, and in both cases this region is quite variable across taxa (e.g. mouse, lizard, human). Therefore, it seems unlikely that these amino acid substitutions are likely to affect channel function. Finally, our analysis was able to confirm and extend the observation that some K-Channel genes are under high selective pressure in finch compared to chicken [6, 12]. We note that our analysis also revealed that nearly half of all K-Channel genes that could be inferred to have high dN/dS ratios based on Ensembl Compara are incorrect, and have artificially high dN/dS ratios that stem from improperly aligned protein coding models, incomplete model predictions, and/or models derived from low quality sequence. Interestingly, our analysis revealed that of the 10 genes with relatively high dN/dS ratios, 6 are members of the KCNK family of 2-TM channels, corresponding to half of all KCNK channels under high selective pressure in zebra finch. We note that this finding is not due to the disproportionate number of KCNK channels in the zebra finch genome, as the KCTD family, the largest family of K-Channel related genes, contains 20 members and just a single gene (KCTD18) under high selective pressure compared to chicken. While the implications of this finding are unclear, they do suggest that specific families of K-Channels may play a more prominent role in the evolution of zebra finches, and/or possibly the songbird lineage. It will be interesting to see if this remains the case for other members of the songbird clade as additional genome sequences become available (e.g. ). In any case, further analysis revealed just 5 cases where non-conservative amino acid substitutions might be expected to affect a predicted functional domain, and in just two cases (KCNMB1 and KCNK16), we found that the dN/dS ratio comparing lizard vs. finch was higher than that of lizard vs. chicken, suggesting greater divergence in songbirds. Thus, overall, it appears that K-Channel genes are highly conserved among higher vertebrate lineages. It thus seems unlikely that the genomic features we have described for the K-Channel gene family in zebra finches, including novel genes or protein coding changes, are prominent mechanisms for tailoring the functional properties of circuitry for vocal production and maintenance.
As discussed next, we found strong evidence for the differential expression of K-Channel genes in song control nuclei. A comparable characterization of K-Channel expression in brain areas associated with speech and language (e.g. Broca’s, orofacial and laryngeal motor cortex, etc.) is lacking in humans, thus we cannot at present assess the degree of brain gene expression conservation across vocal learners. However, none of the detected song nuclei marker genes are specific to finches. These observations suggest that the specific properties conferred by K-Channel genes on vocal circuitry are most likely determined at the level of transcriptional regulation, pointing to promoter regulatory regions as critical elements to be examined in future studies.
Functional tuning of brain circuits for the production and maintenance of learned vocalizations
Consistent with our expectations based on the function of K-Channels and their widespread expression in the mammalian brain [42–45], we found evidence for the expression of K-Channel genes encoding pore-forming (i.e., alpha) and accessory (i.e., beta) subunits in every major song nucleus in the adult zebra finch brain (Table 1). In fact, only 3 genes for which we had probes, and were able to detect signal, appear not to be expressed in the song system. Thus, it is possible that combinatorial patterns define the molecular constituents of K-Channels within song nuclei, forming a gene network that defines the homeostatic (e.g. membrane and resting potential) as well as dynamic states (e.g. firing rate, spike waveforms, and neurotransmission) of neurons that influence the maintenance and production of adult birdsong. However, we also found that more than half of all K-Channel genes (35/56) are differentially expressed (either up or down) in at least one vocal nucleus relative to its surrounding brain region. These genes constitute molecular markers of the song system, and they may underlie the unique electrophysiological features that define the properties of neurons in vocal nuclei. A deeper understanding of the roles that these K-Channel markers play in other systems may provide important insights into the cellular properties, firing behavior, and state dynamics of vocal control pathways.
K-Channel specializations in an anterior forebrain pathway (AFP) for vocal plasticity
While lesions of AFP nuclei do not affect adult song per se, they do suppress song learning and reduce vocal plasticity, implicating the AFP as a key player in the sensorimotor learning and maintenance of birdsong. A major finding of the present study is the identification of a large number of K-Channel genes (21) that are specifically differentially expressed in the AFP (Table 1; Figure 4). The selective expression of these genes in the AFP suggests they may confer specialized electrophysiological features that are central for song learning and vocal plasticity.
Area X is located within the avian equivalent of the basal ganglia, but intriguingly it contains cells that present both striatal- and pallidal-like properties, based on anatomical, molecular, and neurophysiological criteria [46–51]. Area X receives primary input from LMAN, and in turn projects to DLM in the thalamus. While lesions of LMAN result in impoverished, but highly stereotyped adult song [16, 17], partial lesions in Area X lead to abnormal, un-crystallized, and structurally variable song . Thus, these nuclei serve functionally distinct roles during song learning. It seems reasonable to assume that these areas possess electrophysiological distinct properties, possibly related to K-Channel expression. Indeed, we found that 9 K-Channel genes are strong markers of Area X, 5 of LMAN, and 7 of both. Notably, KCNAB2 (Figure 4A), KCNA1 (Figure 4B), and KCNT2 (Figure 4E) are among the strongest and most selective markers that we have found to date for LMAN and Area X, from a database of several hundred genes, hinting that these channels likely confer specific properties that are important for a functioning adult song system.
Some identified K-Channels markers of Area X have previously been linked to sensory and motor functions in mammalian systems. For example, KCNT2, which encodes a unique class of Na+-sensitive K-Channels (K(Na+)), is widely expressed throughout brainstem auditory nuclei in mammals, where it is thought to regulate the accuracy of neuronal spike timing . Intriguingly, both singing- and auditory-related activity have been reported in Area X [49, 51], suggesting that KCNT2 might play a role in stabilizing, or locking-in, the fidelity of song-related firing patterns in Area X. Notably, Area X was also the only song nucleus showing higher differential levels of mRNA for a member of the KCNK sub-family of two-pore leak channels (i.e. KCNK2). This is intriguing as KCNK2 is the only member of this sub-family that is capable of reversibly interconverting from an open leak channel to a voltage-dependent phenotype depending on the state of phosphorylation [53, 54]. Its high expression in Area X suggests an underlying mechanism for regulating resting membrane potential, which would have important consequences for shaping cellular excitability [54, 55]. Pharmacological and double-labeling studies will be needed to address what cell types these channels are expressed in, whether/how they may interact, and how the properties of individual neurons work together to establish the underlying properties of Area X.
Our in situ data also reveal that multiple members of the same K-Channel sub-family are co-expressed within the same nucleus, and possibly even cell type, suggesting that the combined properties conferred by these channels may be important for normal function. In Area X, two members of the G-protein coupled inward rectifying K-Channel (GIRK) sub-family (i.e. KCNJ5 and KCNJ6) are highly expressed (Table 1). This is noteworthy, as GIRK channels are known to be activated by a variety of G-protein coupled receptors, including several shown to be expressed in Area X. For instance, D2 dopamine receptors are expressed in Area X  and are thought to be key players in the dopaminergic modulation of social-context dependent song learning [57–59]. Furthermore, D2 receptor activation dampens singing-related increases of egr1 (ZENK) expression [59, 60], and depresses evoked firing [58, 61], responses that are consistent with the activation of IRK-type channels in other G-protein coupled pathways. We thus suggest that GIRK activation by D2 may represent a mechanism for modulating the properties of Area X neurons, a testable hypothesis that illustrates how molecular findings can help guide targeted pharmacological and electrophysiological studies.
LMAN’s core region contains magnocellular neurons that send dual projections to Area X and RA, as well as GABAergic neurons that influence the projection neurons through inhibition [62, 63]. The magnocellular neurons are of particular interest since they provide direct input to the AFP via Area X, and the DMP via RA. Our analysis revealed that some K-Channel specializations distinguish LMAN from other song nuclei, including 1 of 32 of the KCNC sub-family members (KCNC1) for which we had probe. Interestingly, we found that KCNC3, which is prominently expressed in the brain of mammals, has been lost in birds, possibly amplifying the relative contributions of the remaining members of this channel sub-family. In mammals, KCNC-expressing neurons can be readily distinguished from other cells by their high-frequency firing, unique narrow spike waveform [64, 65], and in some cases, expression of the calcium-binding protein parvalbumin. Intracellular recordings suggest that the fast gating kinetics conferred by KCNC channels are critical for maintaining high-frequency repetitive firing , leading some to argue that virtually every neuron that is capable of sustained firing likely expresses at least one KCNC type. While few intracellular studies have been conducted on the intrinsic properties of neurons in LMAN, the presence of parvalbumin expression in both parvocellular and magnocellular neurons (Lovell and Mello, unpublished observation) predicts that KCNC channels may define the fast-spiking properties of both cell types in LMAN.
K-Channel specializations in a direct motor pathway (DMP) for singing
Lesions of HVC or RA abolish singing, indicating that the DMP provides the neural instructions or motor code for the production of learned song (Figure 1). While the exact nature of this code is unclear, the available evidence suggests a hierarchical organization with: (a) pre-motor HVC-to-RA neurons that burst sparsely during each song motif, thus apparently linked to the tempo of song, projecting to (b) RA neurons that fire repeatedly throughout the song, but in synchrony with temporal and spectral features of song syllables . Given that these processes likely require very different cellular firing properties, we were not surprised to find major differences in K-Channel complements in HVC and RA. Indeed, out of the 27 K-Channel genes identified as markers of the DMP, 7 were exclusive to HVC, and 10 to RA.
We found that genes related to assembly of delayed rectifier voltage-gated channels are highly differential in HVC (Figure 6B), but generally weakly expressed or negatively differential in RA (Figure 6; KCNA6, KCNA1). This is intriguing as when KCNA subunits are paired with KCNAB1 or KCNAB2, both of which are also elevated in HVC, the resulting inactivation gating properties become particularly suited for enhancing spike accommodation, broadening action potentials, and delaying spike after-hyperpolarizations [68, 69]. HVC contains at least three distinct classes of neurons, namely neurons that project to Area X and RA (HVCX and HVCRA, respectively), and local GABAergic interneurons (HVCINT), Importantly, HVCRA neurons, the specific cell type that participates in pre-motor projection to vocal-motor nucleus RA, exhibit strong spike accommodation [30, 31], a property that may underlie the sparse coding characteristic of this projection and that may be at least partially explained by the molecular profile we have found in HVC. We note the differential expression of KCNA channel subunits has been associated with the production of specific motor patterns in other systems. For example, in the lobster stomatogastric system, variation in K-current properties of pyloric neurons, which result from the differential expression of KCNA and KCNB subunits, has been shown to determine the firing order and phase relationships between circuit components, thus leading to the production of different motor patterns ([70, 71]). In the song system, future double-labeling studies to determine the cell type specific complement of K-channel genes, and the ability to block specific channel genes through pharmacology or molecular methods may provide novel insights into how specific K-Channels help to determine the physiology of HVC.
In contrast with HVC, virtually every K-Channel gene that is exclusively differential in RA showed relatively low expression, including both K(Na+) channel subunits (KCNJ5 and KCNJ6), both KCNT channels, and a variety of KCNB and KCND accessory subunits. These results suggest an emphasis on stabilizing neuronal networks within RA, possibly related to the preservation the low song-to-song syllable variability in firing properties that has been observed in most RA neurons [72, 73]. Interestingly, recordings from RA neurons in both anesthetized in vivo and in vitro slice preparations indicate that these neurons act like intrinsic pacemaker cells, often displaying extremely high rates of spontaneous activity . This is in stark contrast with HVC neurons (all classes), which are relatively quiescent unless stimulated by current injection . Thus, one explanation for the apparent decreased level of K-Channel mRNA expression in RA is the suppression of specific K-Channel variants (or their accessory subunits), depressing rapid changes in cellular excitability and firing adaptation.