An interneuronal CRH and CRHBP circuit stabilizes birdsong performance

Reviewer #1 (Public review):

Summary:

This study provides evidence that neuropeptide signaling, particularly via the CRH-CRHBP pathway, plays a key role in regulating the precision of vocal motor output in songbirds. By integrating gene expression profiling with targeted manipulations in the song vocal motor nucleus RA, the authors demonstrate that altering CRH and CRHBP levels bidirectionally modulate song variability. These findings reveal a previously unrecognized neuropeptidergic mechanism underlying motor performance control, supported by molecular and functional evidence.

Strengths:

Neural circuit mechanisms underlying motor variability have been intensively studied, yet the molecular bases of such variability remain poorly understood. The authors address this important gap using the songbird (Bengalese finch) as a model system for motor learning, providing experimental evidence that neuropeptide signaling contributes to vocal motor variability. They comprehensively characterize the expression patterns of neuropeptide-related genes in brain regions involved in song vocal learning and production, revealing distinct regulatory profiles compared to non-vocal related regions, as well as developmental, revealing distinct regulatory profiles compared to non-vocal regions, as well as developmental and behavioral dependencies, including altered expression following deafening and correlations with singing activity over the two days preceding sampling. Through these multi-level analyses spanning anatomy, development, and behavior, the authors identify the CRH-CRHBP pathway in the vocal motor nucleus RA as a candidate regulator of song variability. Functional manipulations further demonstrate that modulation of this pathway bidirectionally alters song variability.

Overall, this work represents an effective use of songbirds, though a well-established neuroethological framework uncovers how previously uncharacterized molecular pathways shape behavioral output at the individual level.

Weaknesses:

(1) This study uses Bengalese finches (BFs) for all experiments-bulk RNA-seq, in situ hybridization across developmental stages, deafening, gene manipulation, and CRH microinfusion-except for the sc/snRNA-seq analysis. BFs differ from zebra finches (ZFs) in several important ways, including faster song degradation after deafening and greater syllable sequence complexity. This study makes effective use of these unique BF characteristics and should be commended for doing so.

However, the major concern lies in the use of the single-cell/single-nucleus RNA-seq dataset from Colquitt et al. (2021), which combines data from both ZFs and BFs for cell-type classification. Based on our reanalysis of the publicly available dataset used in both Colquitt et al. (2021) and the present study, my lab identified two major issues:

(a) The first concern is that the quality of the single-cell RNA-seq data from BFs is extremely poor, and the number of BF-derived cells is very limited. In other words, most of the gene expression information at the single-cell (or "subcellular type") level in this study likely reflects ZF rather than BF profiles. In our verification of the authors' publicly annotated data, we found that in the song nucleus RA, only about 18 glutamatergic cells (2.3%) of a total of 787 RA_Glut (RA_Glut1+2+3) cells were derived from BFs. Similarly, in HVC, only 53 cells (4.1%) out of 1,278 Glut1+Glut4 cells were BF-derived. This clearly indicates that the cell-subtype-level expression data discussed in this study are predominantly based on ZF, not BF, expression profiles.

Recent studies have begun to report interspecies differences in the expression of many genes in the song control nuclei. It is therefore highly plausible that the expression patterns of CRHBP and other neuropeptide-signaling-related genes differ between ZFs and BFs. Yet, the current study does not appear to take this potential species difference into account. As a result, analyses such as the CellChat results (Fig. 2F and G) and the model proposed in Fig. 6G are based on ZF-derived transcriptomic information, even though the rest of the experimental data are derived from BF, which raises a critical methodological inconsistency.

(b) The second major concern involves the definition of "subcellular types" in the sc/snRNA-seq dataset. Specifically, the RA_Glut1, 2, and 3 and HVC_Glu1 and 4 clusters-classified as glutamatergic projection neuron subtypes-may in fact represent inter-individual variation within the same cell type rather than true subtypes. Following Colquitt et al. (2021), Toji et al. (PNAS, 2024) demonstrated clear individual differences in the gene expression profiles of glutamatergic projection neurons in RA.

In our reanalysis of the same dataset, we also observed multiple clusters representing the same glutamatergic projection neurons in UMAP space. This likely occurs because Seurat integration (anchor-based mutual nearest neighbor integration) was not applied, and because cells were not classified based on individual SNP information using tools such as Souporcell. When classified by individual SNPs, we confirmed that the RA_Glut1-3 and HVC_Glu1 and 4 clusters correspond simply to cells from different individuals rather than distinct subcellular types. (Although images cannot be attached in this review system, we can provide our analysis results if necessary.)

This distinction is crucial, as subsequent analyses and interpretations throughout the manuscript depend on this classification. In particular, Figure 6G presents a model based on this questionable subcellular classification. Similarly, the ligand-receptor relationships shown in Figure 2G - such as the absence of SST-SSTR1 signaling in RA_Glut3 but its presence in RA_Glut1 and 2-are more plausibly explained by inter-individual variation rather than subcellular-type specificity.

Whether these differences are interpreted as individual variation within a single cell type or as differences in projection targets among glutamatergic neurons has major implications for understanding the biological meaning of neuropeptide-related gene expression in this system.

(2) Based on the important finding that "CRHBP expression in the song motor pathway is correlated with singing," it is necessary to provide data showing that the observed changes in CRHBP and other neuropeptide-related gene expression during the song learning period or after deafening are not merely due to differences in singing amount over the two days preceding brain sampling.

Without such data, the following statement cannot be justified: "Regarding CRHBP expression in the song motor pathway increases during song acquisition and decreases following deafening."

(3) In Figure 5B, the authors should clearly distinguish between intact and deafened birds and show the singing amount for each group. In practice, deafening often leads to a reduction in both the number of song bouts and the total singing time. If, in this experiment, deafened birds also exhibited reduced singing compared to intact birds, then the decreased CRHBP expression observed in HVC and RA (Figures 3 and 4) may not reflect song deterioration, but rather a simple reduction in singing activity.

As a similar viewpoint, the authors report that CRHBP expression levels in RA and HVC increase with age during the song learning period. However, this change may not be directly related to age or the decline in vocal plasticity. Instead, it could correlate with the singing amount during the one to two days preceding brain sampling. The authors should provide data on the singing activity of the birds used for in situ hybridization during the two days prior to sampling.

Reviewer #2 (Public review):

Summary:

The results presented here are a useful extension of two of their previous papers (Colquitt et al 2021, Colquitt et al 2023), where they used single-cell transcriptomics to characterize the inhibitory and excitatory cell types and gene expression patterns of the song circuit, comparing them to mammalian and reptilian brains, and characterized the effect of deafening on these gene expression patterns. In this paper, they focus on the differential expression of various neuropeptidergic systems in the songbird brain. They discover a role for the CRHBP gene in song performance and causally show its influence on song variability.

Strengths:

The authors leverage the advantages of the 'nucleated' structure of the songbird neural circuitry and use a robust approach to compare neuropeptidergic gene expression patterns in these circuits. Their analysis of the expression patterns of the CRHBP gene in different cell types supports their conclusion that interneurons are particularly amenable to this modulation. Their use of a knockdown strategy along with pharmacological manipulation provides strong support for a causal role of neuropeptidergic modulation on song behaviour. These results have important implications as they bring into focus neuropeptide modulation of the song-motor circuit and pave the way for future studies focussing on how this signalling pathway regulates plasticity during song learning and maintenance.

Weaknesses:

While the results demonstrating the bidirectional modulation of CRH and CRHBP on song performance shed light on their role in song plasticity, it would be important to show this in juvenile finches during sensorimotor learning. We also don't get a clear picture of the 'causal' role of this signalling pathway on the song pre-motor area, HVC, as the knockdown and pharmacological manipulation studies were done in RA, whereas we see a modulation of CRHBP expression during deafening and song learning in both RA and HVC. Given the role of interneurons in the HVC in song acquisition (e.g., Vallentin et al. 2016, Science), it would have been interesting to see the results of HVC-specific manipulation of this neuropeptidergic pathway and/or how it affects the song learning process. Perhaps a short discussion of this would help to give the readers some perspective. Finally, a more direct demonstration of the neurophysiological effect of the signalling pathway would also strengthen our understanding of precisely how these modulate the song circuit plasticity, which I understand might be beyond the scope of this study.

Technical/minor:

In the Methods section, several clarifications would be beneficial. For instance, the description of the design matrices would benefit from being presented in a more general statistical form (e.g., linear model equations) rather than using R syntax. This would make the modeling approach more accessible to readers unfamiliar with software-specific syntax. In addition, while some variables (e.g., cdr_scale, frac_mito_scale) are briefly defined, others (e.g., tags, cut3,nsongs_last_two_days_cut3) could be more clearly described. This applies to the descriptions of both the gene set enrichment analysis and the neuropeptide-receptor analysis, which rely heavily on package-specific terminology (e.g., fgseaMultilevel, computeCommunProb), making it difficult for readers to understand the conceptual or statistical basis of the analyses. It would improve clarity if the authors provided a complete list of variable definitions, types (categorical or continuous), and any scaling/transformations applied would enhance clarity and reproducibility.

Reviewer #3 (Public review):

Summary:

The stable production of learned vocalizations like human language and birdsong requires auditory feedback. What happens in the brain areas that generate stable vocalizations as performance deteriorates is not well understood. Using a species of songbird, the current study investigates individual cells within the evolutionarily-conserved brain regions that generate learned vocalizations to describe that the complement of neuropeptide (short proteins) signals may be a key feature of behavioral change. Because neuropeptides are important across species, these findings may help explain diminishing stability in learned behaviors even in humans.

Strengths:

The experiments are solid and follow a strong progression from description through manipulation. The songbird model is appropriate and powerful to inform on generalizable biological mechanisms of precisely learned behaviors, including human speech.

Weaknesses:

While it is always possible to perform more experiments, most of the weaknesses are in the presentation of the project, not in the evidence or analysis, which are leading-edge and appropriate. Generally, the ability to follow the findings and to independently assess rigor would be enhanced with increased explicit mention of the statistical thresholds and subjective descriptions. In addition, two prior pieces of relevant work seem to be omitted, including one performing deafening, gene expression measures, and behavioral assessment in zebra finches, and another describing neuropeptide complements in zebra finch singing nuclei based largely on mass spectrometry. The former in particular should be related to the current findings.

Author response:

We thank the reviewers for their time and their constructive comments.

Reviewer 1 makes several incisive comments about the single-cell RNA-sequencing dataset used in this version of the manuscript, which was previously published in Colquitt, 2021. The Reviewer correctly notes that this dataset consists primarily of nuclei from zebra finches, with a relatively small proportion of the data coming from Bengalese finches. However, all other data presented here comes from assays and experiments in Bengalese finches. This discrepancy could lead to two issues of interpretation. First, there could be substantive expression differences in the CRH signaling pathway between these two species, making it difficult to interpret its cellular expression profile. Second, the Reviewer describes that in their reanalysis of this dataset they determined that what had been described as distinct cell types – namely HVC-Glut-1 vs. HVC-Glut-4 (corresponding to the HVC RA projection neurons) and the three RA-Glut types – are likely to be single cell types. The Reviewer notes that inter-individual differences in gene expression, which were not analyzed in the original publication, could have generated this apparent cell type diversity.

To the first point, we agree that the use of the published dataset that consists primarily of zebra finch data is not ideal when making claims of cell type-specific expression in Bengalese finches. To rectify this issue, we have generated additional sets of snRNA-seq from Bengalese finches that encompass multiple areas of the song system as well as adjacent comparator regions outside of the principal song areas. Our initial analysis of these datasets indicates that the cellular patterns of expression of the CRH system is consistent with what has been presented here. In our revision, we will include a reanalysis of neuropeptide expression using these more extensive datasets.

To the second point, we also agree that some of the instances of glutamatergic neuron diversity could have been generated either by issues stemming from the integration of two species or through interindividual differences. In our analysis of our newer snRNA-seq data, we also identify a single HVC RA projection neuron type (not two) and that RA projection neuron types fall into one or two classes (not three), similar to what Reviewer 1 described. We have deconvolved these datasets by genotype, as suggested by the Reviewer, and do not see substantial interindividual variation across the CRH system. However, our revision will explicitly address these issues.

Reviewer 1 also brings up several important questions concerning the relationships between CRHBP and singing and the challenge of interpreting the influences of song acquisition and deafening on CRHBP expression, given the variation in singing that generally accompanies these changes to song. To address in part this issue, our regression analysis of deafening-associated gene expression differences includes a term for the number of songs sung on the day of euthanasia as well as an interaction term between song destabilization and singing amount. This design controls for the amount that a bird sang in the period before brain collection. This analysis was included in (Colquitt et al., 2023) , and will be further elaborated and discussed in the revised version of this manuscript. Notably, CRHBP expression shows a significant interaction between song destabilization and singing amount, suggesting that reduction of CRHBP following deafening is greater than what would be expected from any reductions in singing alone. This specific analysis will be included in the revised manuscript as well.

However, despite these statistical controls, we cannot fully rule out that singing is playing a fundamental role in driving the CRHBP expression differences we see across conditions. Indeed, a number of studies have described an association between the amount a bird sings and the variability of its song (Chen et al., 2013; Hayase et al., 2018; Hilliard et al., 2012; Miller et al., 2010; Ohgushi et al., 2015) , with a general trend of higher amounts of singing correlated with a reduction in variability. This relationship is consistent with what we see for CRHBP expression in RA and HVC: high in unmanipulated adult males and decreased during states of high variability and plasticity (post-deafening and juveniles). A model that combines these observations, and that we will include in the Discussion of the revised manuscript, is one in which singing induces the expression of CRHBP in RA and HVC, limiting CRH binding to its receptors, thereby limiting this pathway’s proposed effects on the excitability and synaptic plasticity of projection neurons.

Reviewer 2 suggests multiple interesting avenues to more fully characterize the role of the CRH pathway in song performance and learning. First, we agree that HVC is a compelling target to investigate CRH’s role in song, given the similarity of CRHBP expression in HVC and RA across deafening, song acquisition, and singing. As the Reviewer notes, a number of studies have demonstrated key roles for interneurons in shaping neuronal dynamics in HVC and regulating song structure. Here, we focused on RA due to the direct influence of RA projection neurons have on syringeal and respiration motoneurons controlling song production, and the following expectation that manipulations of CRH signaling in this region would have particularly measurable effects on song. However, we agree with the reviewer that it would be of additional interest to investigate manipulations of CRH signalling in HVC. We are considering whether it will be feasible given the usual constraints of time, personnel, and other competing demands to carry such experiments as an addition to the current manuscript. Depending on how that goes, we will either add new experimental data to the manuscript, or simply acknowledge the interest of such experiments in Discussion and defer their pursuit to future study.

Likewise, Reviewer 2 suggests other ways in which an understanding of the role of CRH signalling could be further enriched with additional experiments, including investigating the influence of CRH signaling on song acquisition, when song transitions from a variable and plastic state to a precise and stereotyping state, and pursuing direct evidence that CRH influences the neurophysiology of glutamatergic neurons in HVC or RA. These are both excellent suggestions for ways in neuropeptide signalling could be further linked to alterations in behavior; As we proceed with revisions we will consider whether we can address some of these suggestions within the scope of the current manuscript, versus note them in discussion as directions for future research.

Chen Q, Heston JB, Burkett ZD, White SA. 2013. Expression analysis of the speech-related genes FoxP1 and FoxP2 and their relation to singing behavior in two songbird species. J Exp Biol 216 :3682–3692. doi:10.1242/jeb.085886

Colquitt BM, Li K, Green F, Veline R, Brainard MS. 2023. Neural circuit-wide analysis of changes to gene expression during deafening-induced birdsong destabilization. Elife 12 :e85970. doi:10.7554/eLife.85970

Hayase S, Wang H, Ohgushi E, Kobayashi M, Mori C, Horita H, Mineta K, Liu W-C, Wada K. 2018. Vocal practice regulates singing activity-dependent genes underlying age-independent vocal learning in songbirds. PLoS Biol 16 :e2006537. doi:10.1371/journal.pbio.2006537

Hilliard AT, Miller JE, Fraley ER, Horvath S, White SA. 2012. Molecular microcircuitry underlies functional specification in a basal ganglia circuit dedicated to vocal learning. Neuron 73 :537–552. doi:10.1016/j.neuron.2012.01.005

Miller JE, Hilliard AT, White SA. 2010. Song practice promotes acute vocal variability at a key stage of sensorimotor learning. PLoS One 5 :e8592. doi:10.1371/journal.pone.0008592

Ohgushi E, Mori C, Wada K. 2015. Diurnal oscillation of vocal development associated with clustered singing by juvenile songbirds. J Exp Biol 218 :2260–2268. doi:10.1242/jeb.115105