Bird song comparison using deep learning trained from avian perceptual judgments.

Our understanding of bird song, a model system for animal communication and the neurobiology of learning, depends critically on making reliable, validated comparisons between the complex multidimensional syllables that are used in songs. However, most assessments of song similarity are based on human inspection of spectrograms, or computational methods developed from human intuitions. Using a novel automated operant conditioning system, we collected a large corpus of zebra finches' (Taeniopygia guttata) decisions about song syllable similarity. We use this dataset to compare and externally validate similarity algorithms in widely-used publicly available software (Raven, Sound Analysis Pro, Luscinia). Although these methods all perform better than chance, they do not closely emulate the avian assessments. We then introduce a novel deep learning method that can produce perceptual similarity judgements trained on such avian decisions. We find that this new method outperforms the established methods in accuracy and more closely approaches the avian assessments. Inconsistent (hence ambiguous) decisions are a common occurrence in animal behavioural data; we show that a modification of the deep learning training that accommodates these leads to the strongest performance. We argue this approach is the best way to validate methods to compare song similarity, that our dataset can be used to validate novel methods, and that the general approach can easily be extended to other species.

Concurrent invasions of European starlings in Australia and North America reveal population-specific differentiation in shared genomic regions.

A species' success during the invasion of new areas hinges on an interplay between the demographic processes common to invasions and the specific ecological context of the novel environment. Evolutionary genetic studies of invasive species can investigate how genetic bottlenecks and ecological conditions shape genetic variation in invasions, and our study pairs two invasive populations that are hypothesized to be from the same source population to compare how each population evolved during and after introduction. Invasive European starlings (Sturnus vulgaris) established populations in both Australia and North America in the 19th century. Here, we compare whole-genome sequences among native and independently introduced European starling populations to determine how demographic processes interact with rapid evolution to generate similar genetic patterns in these recent and replicated invasions. Demographic models indicate that both invasive populations experienced genetic bottlenecks as expected based on invasion history, and we find that specific genomic regions have differentiated even on this short evolutionary timescale. Despite genetic bottlenecks, we suggest that genetic drift alone cannot explain differentiation in at least two of these regions. The demographic boom intrinsic to many invasions as well as potential inversions may have led to high population-specific differentiation, although the patterns of genetic variation are also consistent with the hypothesis that this infamous and highly mobile invader adapted to novel selection (e.g., extrinsic factors). We use targeted sampling of replicated invasions to identify and evaluate support for multiple, interacting evolutionary mechanisms that lead to differentiation during the invasion process.

The role of the genome in experience-dependent plasticity: Extending the analogy of the genomic action potential.

Our past experiences shape our current and future behavior. These experiences must leave some enduring imprint on our brains, altering neural circuits that mediate behavior and contributing to our individual differences. As a framework for understanding how experiences might produce lasting changes in neural circuits, Clayton [D. F. Clayton, Neurobiol. Learn. Mem. 74, 185-216 (2000)] introduced the concept of the genomic action potential (gAP)-a structured genomic response in the brain to acute experience. Similar to the familiar electrophysiological action potential (eAP), the gAP also provides a means for integrating afferent patterns of activity but on a slower timescale and with longer-lasting effects. We revisit this concept in light of contemporary work on experience-dependent modification of neural circuits. We review the "Immediate Early Gene" (IEG) response, the starting point for understanding the gAP. We discuss evidence for its involvement in the encoding of experience to long-term memory across time and biological levels of organization ranging from individual cells to cell ensembles and whole organisms. We explore distinctions between memory encoding and homeostatic functions and consider the potential for perpetuation of the imprint of experience through epigenetic mechanisms. We describe a specific example of a gAP in humans linked to individual differences in the response to stress. Finally, we identify key objectives and new tools for continuing research in this area.

Acute social isolation alters neurogenomic state in songbird forebrain.

Prolonged social isolation has negative effects on brain and behavior in humans and other social organisms, but neural mechanisms leading to these effects are not understood. Here we tested the hypothesis that even brief periods of social isolation can alter gene expression and DNA methylation in higher cognitive centers of the brain, focusing on the auditory/associative forebrain of the highly social zebra finch. Using RNA sequencing, we first identified genes that individually increase or decrease expression after isolation and observed general repression of gene sets annotated for neurotrophin pathways and axonal guidance functions. We then pursued 4 genes of large effect size: EGR1 and BDNF (decreased by isolation) and FKBP5 and UTS2B (increased). By in situ hybridization, each gene responded in different cell subsets, arguing against a single cellular mechanism. To test whether effects were specific to the social component of the isolation experience, we compared gene expression in birds isolated either alone or with a single familiar partner. Partner inclusion ameliorated the effect of solo isolation on EGR1 and BDNF, but not on FKBP5 and UTS2B nor on circulating corticosterone. By bisulfite sequencing analysis of auditory forebrain DNA, isolation caused changes in methylation of a subset of differentially expressed genes, including BDNF. Thus, social isolation has rapid consequences on gene activity in a higher integrative center of the brain, triggering epigenetic mechanisms that may influence processing of ongoing experience.

The genomics of memory and learning in songbirds.

Songbirds have unique value as a model for memory and learning. In their natural social life, they communicate through vocalizations that they must learn to produce and recognize. Song communication elicits abrupt changes in gene expression in regions of the forebrain responsible for song perception and production--what is the functional significance of this genomic response? For 20 years, the focus of research was on just a few genes [primarily ZENK, now known as egr1 (early gene response 1)]. Recently, however, DNA microarrays have been developed and applied to songbird behavioral research, and in 2010 the initial draft assembly of the zebra finch genome was published. Together, these new data reveal that the genomic involvement in song processing is far more complex than anticipated. The concepts of neurogenomic computation and biological embedding are introduced as frameworks for future research.

The genome of a songbird.

The zebra finch is an important model organism in several fields with unique relevance to human neuroscience. Like other songbirds, the zebra finch communicates through learned vocalizations, an ability otherwise documented only in humans and a few other animals and lacking in the chicken-the only bird with a sequenced genome until now. Here we present a structural, functional and comparative analysis of the genome sequence of the zebra finch (Taeniopygia guttata), which is a songbird belonging to the large avian order Passeriformes. We find that the overall structures of the genomes are similar in zebra finch and chicken, but they differ in many intrachromosomal rearrangements, lineage-specific gene family expansions, the number of long-terminal-repeat-based retrotransposons, and mechanisms of sex chromosome dosage compensation. We show that song behaviour engages gene regulatory networks in the zebra finch brain, altering the expression of long non-coding RNAs, microRNAs, transcription factors and their targets. We also show evidence for rapid molecular evolution in the songbird lineage of genes that are regulated during song experience. These results indicate an active involvement of the genome in neural processes underlying vocal communication and identify potential genetic substrates for the evolution and regulation of this behaviour.

Advancing avian behavioral neuroendocrinology through genomics.

Genome technologies are transforming all areas of biology, including the study of hormones, brain and behavior. Annotated reference genome assemblies are rapidly being produced for many avian species. Here we briefly review the basic concepts and tools used in genomics. We then consider how these are informing the study of avian behavioral neuroendocrinology, focusing in particular on lessons from the study of songbirds. We discuss the impact of having a complete "parts list" for an organism; the transformational potential of studying large sets of genes at once instead one gene at a time; the growing recognition that environmental and behavioral signals trigger massive shifts in gene expression in the brain; and the prospects for using comparative genomics to uncover the genetic roots of behavioral variation. Throughout, we identify promising new directions for bolstering the application of genomic information to further advance the study of avian brain and behavior.

Impact of experience-dependent and -independent factors on gene expression in songbird brain.

Songbirds provide rich natural models for studying the relationships between brain anatomy, behavior, environmental signals, and gene expression. Under the Songbird Neurogenomics Initiative, investigators from 11 laboratories collected brain samples from six species of songbird under a range of experimental conditions, and 488 of these samples were analyzed systematically for gene expression by microarray. ANOVA was used to test 32 planned contrasts in the data, revealing the relative impact of different factors. The brain region from which tissue was taken had the greatest influence on gene expression profile, affecting the majority of signals measured by 18,848 cDNA spots on the microarray. Social and environmental manipulations had a highly variable impact, interpreted here as a manifestation of paradoxical "constitutive plasticity" (fewer inducible genes) during periods of enhanced behavioral responsiveness. Several specific genes were identified that may be important in the evolution of linkages between environmental signals and behavior. The data were also analyzed using weighted gene coexpression network analysis, followed by gene ontology analysis. This revealed modules of coexpressed genes that are also enriched for specific functional annotations, such as "ribosome" (expressed more highly in juvenile brain) and "dopamine metabolic process" (expressed more highly in striatal song control nucleus area X). These results underscore the complexity of influences on neural gene expression and provide a resource for studying how these influences are integrated during natural experience.

Parental developmental experience affects vocal learning in offspring.

Cultural and genetic inheritance combine to enable rapid changes in trait expression, but their relative importance in determining trait expression across generations is not clear. Birdsong is a socially learned cognitive trait that is subject to both cultural and genetic inheritance, as well as being affected by early developmental conditions. We sought to test whether early-life conditions in one generation can affect song acquisition in the next generation. We exposed one generation (F1) of nestlings to elevated corticosterone (CORT) levels, allowed them to breed freely as adults, and quantified their son's (F2) ability to copy the song of their social father. We also quantified the neurogenetic response to song playback through immediate early gene (IEG) expression in the auditory forebrain. F2 males with only one corticosterone-treated parent copied their social father's song less accurately than males with two control parents. Expression of ARC in caudomedial nidopallium (NCM) correlated with father-son song similarity, and patterns of expression levels of several IEGs in caudomedial mesopallium (CMM) in response to father song playback differed between control F2 sons and those with a CORT-treated father only. This is the first study to demonstrate that developmental conditions can affect social learning and neurogenetic responses in a subsequent generation.

Sex bias and dosage compensation in the zebra finch versus chicken genomes: general and specialized patterns among birds.

We compared global patterns of gene expression between two bird species, the chicken and zebra finch, with regard to sex bias of autosomal versus Z chromosome genes, dosage compensation, and evolution of sex bias. Both species appear to lack a Z chromosome-wide mechanism of dosage compensation, because both have a similar pattern of significantly higher expression of Z genes in males relative to females. Unlike the chicken Z chromosome, which has female-specific expression of the noncoding RNA MHM (male hypermethylated) and acetylation of histone 4 lysine 16 (H4K16) near MHM, the zebra finch Z chromosome appears to lack the MHM sequence and acetylation of H4K16. The zebra finch also does not show the reduced male-to-female (M:F) ratio of gene expression near MHM similar to that found in the chicken. Although the M:F ratios of Z chromosome gene expression are similar across tissues and ages within each species, they differ between the two species. Z genes showing the greatest species difference in M:F ratio were concentrated near the MHM region of the chicken Z chromosome. This study shows that the zebra finch differs from the chicken because it lacks a specialized region of greater dosage compensation along the Z chromosome, and shows other differences in sex bias. These patterns suggest that different avian taxa may have evolved specific compensatory mechanisms.

Noninvasive diffusive optical imaging of the auditory response to birdsong in the zebra finch.

Songbirds communicate by learned vocalizations with concomitant changes in neurophysiological and genomic activities in discrete parts of the brain. Here, we tested a novel implementation of diffusive optical imaging (also known as diffuse optical imaging, DOI) for monitoring brain physiology associated with vocal signal perception. DOI noninvasively measures brain activity using red and near-infrared light delivered through optic fibers (optodes) resting on the scalp. DOI does not harm subjects, so it raises the possibility of repeatedly measuring brain activity and the effects of accumulated experience in the same subject over an entire life span, all while leaving tissue intact for further study. We developed a custom-made apparatus for interfacing optodes to the zebra finch (Taeniopygia guttata) head using 3D modeling software and rapid prototyping technology, and applied it to record responses to presentations of birdsong in isoflurane-anesthetized zebra finches. We discovered a subtle but significant difference between the hemoglobin spectra of zebra finches and mammals which has a major impact in how hemodynamic responses are interpreted in the zebra finch. Our measured responses to birdsong playback were robust, highly repeatable, and readily observed in single trials. Responses were complex in shape and closely paralleled responses described in mammals. They were localized to the caudal medial portion of the brain, consistent with response localization from prior gene expression, electrophysiological, and functional magnetic resonance imaging studies. These results define an approach for collecting neurophysiological data from songbirds that should be applicable to diverse species and adaptable for studies in awake behaving animals.

RNA-seq transcriptome analysis of male and female zebra finch cell lines.

The derivation of stably cultured cell lines has been critical to the advance of molecular biology. We profiled gene expression in the first two generally available cell lines derived from the zebra finch. Using Illumina RNA-seq, we generated ~93 million reads and mapped the majority to the recently assembled zebra finch genome. Expression of most Ensembl-annotated genes was detected, but over half of the mapped reads aligned outside annotated genes. The male-derived G266 line expressed Z-linked genes at a higher level than did the female-derived ZFTMA line, indicating persistence in culture of the distinctive lack of avian sex chromosome dosage compensation. Although these cell lines were not derived from neural tissue, many neurobiologically relevant genes were expressed, although typically at lower levels than in a reference sample from auditory forebrain. These cell lines recapitulate fundamental songbird biology and will be useful for future studies of songbird gene regulation and function.

High throughput analysis reveals dissociable gene expression profiles in two independent neural systems involved in the regulation of social behavior.

BACKGROUND: Production of contextually appropriate social behaviors involves integrated activity across many brain regions. Many songbird species produce complex vocalizations called 'songs' that serve to attract potential mates, defend territories, and/or maintain flock cohesion. There are a series of discrete interconnect brain regions that are essential for the successful production of song. The probability and intensity of singing behavior is influenced by the reproductive state. The objectives of this study were to examine the broad changes in gene expression in brain regions that control song production with a brain region that governs the reproductive state. RESULTS: We show using microarray cDNA analysis that two discrete brain systems that are both involved in governing singing behavior show markedly different gene expression profiles. We found that cortical and basal ganglia-like brain regions that control the socio-motor production of song in birds exhibit a categorical switch in gene expression that was dependent on their reproductive state. This pattern is in stark contrast to the pattern of expression observed in a hypothalamic brain region that governs the neuroendocrine control of reproduction. Subsequent gene ontology analysis revealed marked variation in the functional categories of active genes dependent on reproductive state and anatomical localization. HVC, one cortical-like structure, displayed significant gene expression changes associated with microtubule and neurofilament cytoskeleton organization, MAP kinase activity, and steroid hormone receptor complex activity. The transitions observed in the preoptic area, a nucleus that governs the motivation to engage in singing, exhibited variation in functional categories that included thyroid hormone receptor activity, epigenetic and angiogenetic processes. CONCLUSIONS: These findings highlight the importance of considering the temporal patterns of gene expression across several brain regions when engaging in social behaviors.

Discrete molecular states in the brain accompany changing responses to a vocal signal.

New experiences can trigger changes in gene expression in the brain. To understand this phenomenon better, we studied zebra finches hearing playbacks of birdsong. Earlier research had shown that initial playbacks of a novel song transiently increase the ZENK (ZIF-268, EGR1, NGFIA, KROX-24) mRNA in the auditory forebrain, but the response selectively habituates after repetition of the stimulus. Here, using DNA microarray analysis, we show that novel song exposure induces rapid changes in thousands of RNAs, with even more RNAs decreasing than increasing. Habituation training leads to the emergence of a different gene expression profile a day later, accompanied by loss of essentially all of the rapid "novel" molecular responses. The novel molecular profile is characterized by increases in genes involved in transcription and RNA processing and decreases in ion channels and putative noncoding RNAs. The "habituated" profile is dominated by changes in genes for mitochondrial proteins. A parallel proteomic analysis [2-dimensional difference gel electrophoresis (2D-DIGE) and sequencing by mass spectrometry] also detected changes in mitochondrial proteins, and direct enzyme assay demonstrated changes in both complexes I and IV in the habituated state. Thus a natural experience, in this case hearing the sound of birdsong, can lead to major shifts in energetics and macromolecular metabolism in higher centers in the brain.

Functional identification of sensory mechanisms required for developmental song learning.

A young male zebra finch (Taeniopygia guttata) learns to sing by copying the vocalizations of an older tutor in a process that parallels human speech acquisition. Brain pathways that control song production are well defined, but little is known about the sites and mechanisms of tutor song memorization. Here we test the hypothesis that molecular signaling in a sensory brain area outside of the song system is required for developmental song learning. Using controlled tutoring and a pharmacological inhibitor, we transiently suppressed the extracellular signal-regulated kinase signaling pathway in a portion of the auditory forebrain specifically during tutor song exposure. On maturation, treated birds produced poor copies of tutor song, whereas controls copied the tutor song effectively. Thus the foundation of normal song learning, the formation of a sensory memory of tutor song, requires a conserved molecular pathway in a brain area that is distinct from the circuit for song motor control.

Transcript- and annotation-guided genome assembly of the European starling.

The European starling, Sturnus vulgaris, is an ecologically significant, globally invasive avian species that is also suffering from a major decline in its native range. Here, we present the genome assembly and long-read transcriptome of an Australian-sourced European starling (S. vulgaris vAU), and a second, North American, short-read genome assembly (S. vulgaris vNA), as complementary reference genomes for population genetic and evolutionary characterization. S. vulgaris vAU combined 10× genomics linked-reads, low-coverage Nanopore sequencing, and PacBio Iso-Seq full-length transcript scaffolding to generate a 1050 Mb assembly on 6222 scaffolds (7.6 Mb scaffold N50, 94.6% busco completeness). Further scaffolding against the high-quality zebra finch (Taeniopygia guttata) genome assigned 98.6% of the assembly to 32 putative nuclear chromosome scaffolds. Species-specific transcript mapping and gene annotation revealed good gene-level assembly and high functional completeness. Using S. vulgaris vAU, we demonstrate how the multifunctional use of PacBio Iso-Seq transcript data and complementary homology-based annotation of sequential assembly steps (assessed using a new tool, saaga) can be used to assess, inform, and validate assembly workflow decisions. We also highlight some counterintuitive behaviour in traditional busco metrics, and present buscomp, a complementary tool for assembly comparison designed to be robust to differences in assembly size and base-calling quality. This work expands our knowledge of avian genomes and the available toolkit for assessing and improving genome quality. The new genomic resources presented will facilitate further global genomic and transcriptomic analysis on this ecologically important species.

Song exposure regulates known and novel microRNAs in the zebra finch auditory forebrain.

BACKGROUND: In an important model for neuroscience, songbirds learn to discriminate songs they hear during tape-recorded playbacks, as demonstrated by song-specific habituation of both behavioral and neurogenomic responses in the auditory forebrain. We hypothesized that microRNAs (miRNAs or miRs) may participate in the changing pattern of gene expression induced by song exposure. To test this, we used massively parallel Illumina sequencing to analyse small RNAs from auditory forebrain of adult zebra finches exposed to tape-recorded birdsong or silence. RESULTS: In the auditory forebrain, we identified 121 known miRNAs conserved in other vertebrates. We also identified 34 novel miRNAs that do not align to human or chicken genomes. Five conserved miRNAs showed significant and consistent changes in copy number after song exposure across three biological replications of the song-silence comparison, with two increasing (tgu-miR-25, tgu-miR-192) and three decreasing (tgu-miR-92, tgu-miR-124, tgu-miR-129-5p). We also detected a locus on the Z sex chromosome that produces three different novel miRNAs, with supporting evidence from Northern blot and TaqMan qPCR assays for differential expression in males and females and in response to song playbacks. One of these, tgu-miR-2954-3p, is predicted (by TargetScan) to regulate eight song-responsive mRNAs that all have functions in cellular proliferation and neuronal differentiation. CONCLUSIONS: The experience of hearing another bird singing alters the profile of miRNAs in the auditory forebrain of zebra finches. The response involves both known conserved miRNAs and novel miRNAs described so far only in the zebra finch, including a novel sex-linked, song-responsive miRNA. These results indicate that miRNAs are likely to contribute to the unique behavioural biology of learned song communication in songbirds.

Small molecule analysis and imaging of fatty acids in the zebra finch song system using time-of-flight-secondary ion mass spectrometry.

Fatty acids are central to brain metabolism and signaling, but their distributions within complex brain circuits have been difficult to study. Here we applied an emerging technique, time-of-flight secondary ion mass spectrometry (ToF-SIMS), to image specific fatty acids in a favorable model system for chemical analyses of brain circuits, the zebra finch (Taeniopygia guttata). The zebra finch, a songbird, produces complex learned vocalizations under the control of an interconnected set of discrete, dedicated brain nuclei 'song nuclei'. Using ToF-SIMS, the major song nuclei were visualized by virtue of differences in their content of essential and non-essential fatty acids. Essential fatty acids (arachidonic acid and docosahexaenoic acid) showed distinctive distributions across the song nuclei, and the 18-carbon fatty acids stearate and oleate discriminated the different core and shell subregions of the lateral magnocellular nucleus of the anterior nidopallium. Principal component analysis of the spectral data set provided further evidence of chemical distinctions between the song nuclei. By analyzing the robust nucleus of the arcopallium at three different ages during juvenile song learning, we obtain the first direct evidence of changes in lipid content that correlate with progression of song learning. The results demonstrate the value of ToF-SIMS to study lipids in a favorable model system for probing the function of lipids in brain organization, development and function.

The zebra finch neuropeptidome: prediction, detection and expression.

BACKGROUND: Among songbirds, the zebra finch (Taeniopygia guttata) is an excellent model system for investigating the neural mechanisms underlying complex behaviours such as vocal communication, learning and social interactions. Neuropeptides and peptide hormones are cell-to-cell signalling molecules known to mediate similar behaviours in other animals. However, in the zebra finch, this information is limited. With the newly-released zebra finch genome as a foundation, we combined bioinformatics, mass-spectrometry (MS)-enabled peptidomics and molecular techniques to identify the complete suite of neuropeptide prohormones and final peptide products and their distributions. RESULTS: Complementary bioinformatic resources were integrated to survey the zebra finch genome, identifying 70 putative prohormones. Ninety peptides derived from 24 predicted prohormones were characterized using several MS platforms; tandem MS confirmed a majority of the sequences. Most of the peptides described here were not known in the zebra finch or other avian species, although homologous prohormones exist in the chicken genome. Among the zebra finch peptides discovered were several unique vasoactive intestinal and adenylate cyclase activating polypeptide 1 peptides created by cleavage at sites previously unreported in mammalian prohormones. MS-based profiling of brain areas required for singing detected 13 peptides within one brain nucleus, HVC; in situ hybridization detected 13 of the 15 prohormone genes examined within at least one major song control nucleus. Expression mapping also identified prohormone messenger RNAs in areas associated with spatial learning and social behaviours. Based on the whole-genome analysis, 40 prohormone probes were found on a commonly used zebra finch brain microarray. Analysis of these newly annotated transcripts revealed that six prohormone probes showed altered expression after birds heard song playbacks in a paradigm of song recognition learning; we partially verify this result experimentally. CONCLUSIONS: The zebra finch peptidome and prohormone complement is now characterized. Based on previous microarray results on zebra finch vocal learning and synaptic plasticity, a number of these prohormones show significant changes during learning. Interestingly, most mammalian prohormones have counterparts in the zebra finch, demonstrating that this songbird uses similar biochemical pathways for neurotransmission and hormonal regulation. These findings enhance investigation into neuropeptide-mediated mechanisms of brain function, learning and behaviour in this model.

Acoustic developmental programming: a mechanistic and evolutionary framework.

Conditions experienced prenatally, by modulating developmental processes, have lifelong effects on individual phenotypes and fitness, ultimately influencing population dynamics. In addition to maternal biochemical cues, prenatal sound is emerging as a potent alternative source of information to direct embryonic development. Recent evidence suggests that prenatal acoustic signals can program individual phenotypes for predicted postnatal environmental conditions, which improves fitness. Across taxonomic groups, embryos have now been shown to have immediate adaptive responses to external sounds and vibrations, and direct developmental effects of sound and noise are increasingly found. Establishing the full developmental, ecological, and evolutionary impact of early soundscapes will reveal how embryos interact with the external world, and potentially transform our understanding of developmental plasticity and adaptation to changing environments.