Self-organization of songbird neural sequences during social isolation

  1. Emily L Mackevicius
  2. Shijie Gu
  3. Natalia I Denisenko
  4. Michale S Fee

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    Adult zebra finch song is highly stereotyped, and it is driven by correspondingly stereotyped neural sequences in premotor cortical nucleus HVC. By imaging HVC activity in juvenile birds isolated from social contact with tutors, the authors discover that stereotyped HVC sequences can exist even without exposure to tutor song. Interestingly, after tutoring, existing sequences in the HVC of isolate birds transitioned from being uncoupled to vocal output to highly coupled to newly copied tutor syllables. Together, these data provide a fascinating glimpse into mechanistic foundations of how nature and nurture work together to a learned motor sequence.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Behaviors emerge via a combination of experience and innate predispositions. As the brain matures, it undergoes major changes in cellular, network, and functional properties that can be due to sensory experience as well as developmental processes. In normal birdsong learning, neural sequences emerge to control song syllables learned from a tutor. Here, we disambiguate the role of tutor experience and development in neural sequence formation by delaying exposure to a tutor. Using functional calcium imaging, we observe neural sequences in the absence of tutoring, demonstrating that tutor experience is not necessary for the formation of sequences. However, after exposure to a tutor, pre-existing sequences can become tightly associated with new song syllables. Since we delayed tutoring, only half our birds learned new syllables following tutor exposure. The birds that failed to learn were the birds in which pre-tutoring neural sequences were most ‘crystallized,’ that is, already tightly associated with their (untutored) song.

Article activity feed

  1. Version published to 10.7554/elife.77262 on eLife
  2. eLife

    Evaluation Summary:

    Adult zebra finch song is highly stereotyped, and it is driven by correspondingly stereotyped neural sequences in premotor cortical nucleus HVC. By imaging HVC activity in juvenile birds isolated from social contact with tutors, the authors discover that stereotyped HVC sequences can exist even without exposure to tutor song. Interestingly, after tutoring, existing sequences in the HVC of isolate birds transitioned from being uncoupled to vocal output to highly coupled to newly copied tutor syllables. Together, these data provide a fascinating glimpse into mechanistic foundations of how nature and nurture work together to a learned motor sequence.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

  3. eLife

    Reviewer #1 (Public Review):

    Mackevicius et al image CA activity in nucleus HVC of isolated singing zebra finches before and after tutor exposure. HVC is well known for its sequential activity during singing - and isolate song is known for its abnormal variability, raising two possibilities. Tutor exposure and subsequent practice may or may not be necessary for chain foundation. Because birdsong is a learned behavior but also subject to innate predispositions, the current manuscript provides a really important test of how nature vs nurture affects the development of song - at the mechanistic level. The authors discover HVC chains do exist, but they are unusually uncoupled from vocal output. More, the more immature chain formation is at the time of tutor exposure, the more copying there is. This finding that the existing HVC chain could become time-locked to new acoustic elements is an important verification of the long assumed, but never explicitly tested, idea that plasticity in the HVC-RA pathway drives phonological change during natural development. These results are really important for the songbird field - as they mechanistically link the timing of tutor exposure to HVC chain maturity to imitation quality. These results also will be useful for the general community of biologists interested in how innate predispositions for animal behavior can express at the level of signals and circuits.

  4. eLife

    Reviewer #2 (Public Review):

    This work develops a powerful new in-vivo model for studying the interplay between innate predispositions and learned behavior. In adult zebra finches, projection neurons (PNs) in the nucleus HVC are active in robust sequences that tile the duration of these songbirds' stereotyped song motif. Juvenile birds learn this motif, a sequence of 4-7 syllables lasting 1-2 seconds (typically), by imitation of a tutor's song. Without exposure to a tutor, zebra finches develop an 'isolate' song whose syllables, and their ordering, are far more variable than that of a normally-tutored bird. The authors used calcium imaging of HVC PN ensembles in isolated zebra finches to test the interplay between developmentally specified and learned structure. Using their recently-developed unsupervised algorithm for identifying sequences in ensemble neural recordings, the authors expose sequential activation of HVC PNs underlying 'isolate' songs - showing both typical and atypical properties of such sequences in normally-tutored birds. Then, the authors make clever use of the fact that brief exposure to a tutor allows isolated juveniles to adopt new syllables within a day or two and eventually develop a normal song. They compare sequences of HVC PNs activity before and after the exposure to the tutor and discover that successful adoption of a new syllable occurs preferentially in birds whose HVC activity was less aligned to vocalization before tutoring. Further, the authors demonstrate that a new learned syllable harnesses a pre-existing sequence - an ensemble activity in HVC that, before tutoring, was loosely (or not at all) aligned with behavior and became associated with a new syllable. These findings suggest a model in which sequences emerge in HVC activity without tutoring. These activity patterns then serve as substrates in the learning of new behavior by association with new motor programs in brain structures downstream from HVC - the premotor nucleus RA innervating respiratory and syrinx-driving motor neurons in the brain stem.

    This unique model lays the grounds for a trove of new inquiries into its biophysical and network mechanisms as well as into its social implications and is of broad interest to neuroscientists, psychologists, and researchers studying artificial neural networks. Most conclusions of this paper are well supported by data, but some key conceptual and analytic aspects require framing with respect to existing literature and better detail and clarity of presentation.

    Strengths:

    Comparing activity of HVC PNs in isolated adults, normally-tutored adults, and isolated juveniles before and after a brief tutor exposure is a uniquely-powerful approach to addressing the main line of inquiry in this work. The rapid adoption of a new syllable by isolated zebra finch juveniles allowed the authors to compare sequences in HVC ensemble activity existing prior to tutoring to those time-aligned to the new syllable after tutoring. Comparisons of these sequences to those found in adult isolated as well as in normally-tutored finches are in line with the authors' conclusions that these sequences are an emerging motif of neural activity - harnessed by a learning process.

    This work also capitalizes on two novel methodological and conceptual advances made by the authors - a factorization algorithm that allows the discovery of activity sequences in data from neuronal ensembles and a method for tracking neurons across days of imaging with head-mounted miniaturized microscopes in songbirds. The manuscript provides a use case for these methods that is of great interest to both researchers of songbirds and of a large community studying neural circuits and sequential activity in the hippocampus.

    Weaknesses and points needing clearer presentation:

    1. The effect of experience, tutor exposure, and HVC maturation is very hard to separate in the current manuscript. Therefore, the ultimate goal to separate intrinsic development from experience may be too ambitious. The authors may choose, as they do in the discussion, to limit their claims to the effect of tutor exposure, a very powerful model in itself. Alternatively, the manuscript needs further citation and analyses treating several experiences, other than tutor exposure, that can contribute to the formation of sequences in HVC:

    1.1. Auditory inputs reach HVC via the nucleus Nif. These inputs can be significant even if the audio is noise patterned to mimic zebra finch song prosody[1]. The patterning of sequences may, therefore, stem from auditory inputs because:
    1.1.1. The juvenile isolates were housed with females that do vocalize.
    1.1.2. The methods describe that juveniles were only with females sometimes before hatching and sometimes starting 15 dph. While not a major concern, the authors may cite evidence that exposure to song earlier than 15dph is not causing the creation of HVC sequences. Alternatively, the authors may show that this difference in experience of juveniles does not account for their performance as learners.
    1.1.3. Juveniles were isolated starting 40-50 dph. Were they isolated from each other between 15-40dph? Do they vocalize in this time period?

    1.2. Additionally, past work showed that, on occasion, juvenile zebra finches will sing directed to females with a very strong effect on their song maturation[2]. Such an experience, although extremely unlikely before 15dph, can also support the emergence of HVC sequences.

    In sum, in setting the conceptual stage for their results, as they already do in the discussion, the authors may make clear that they study the effect of tutoring experience alone. The effect of tutoring on the development of song-aligned HVC sequences offers a strong and significant advance relevant to a broad community of system neuroscientists. Alternatively, the authors need to provide evidence supporting that juveniles had no experience that could promote the growth of HVC sequences.

    2. The adequacy of the sequence detection method (seqNMF[3]) and analyses of its outcomes need further explanation and support. This is especially needed when describing results where sequences are truncated, jittery, or otherwise variable (as some of the results indicate). The presentation of results will be strengthened by:

    2.1. A clear presentation of seqNMF's outcomes and fit to data:
    2.1.1. Explaining in the main text and methods what is meant by 'sequences' that the algorithm extracts. It is not clear if these are cells activating one after the other or any robust spatiotemporal pattern. seqNMF allows seeking 'event based' or 'part based' factorization. Please describe which was used in this manuscript.
    2.1.2. How much of the data variability is explained by sequences?
    2.1.3. How specific are neurons' activity to sequences (compared to its activity not in sequences)

    2.2. Control analyses (or citation if shown elsewhere) can show that the atypical properties of sequences are not confounded by seqNMF.
    2.2.1. For example, measures in Figure 1E-K may be compared to sequences extracted from time-shuffled data. (Similar to the 'sequenciness' approach defined by previous work of the authors[3]).
    2.2.2. Alternatively, if at all possible (because data is limited), results could be compared to analyses carried out on held-out data. For example, sequences can be discovered in training set data and used to calculate results as in Figure 1E-K on test set data.

    2.3. Is it possible to compare sequences (the W's) found before and after training? The claim that they persist needs quantitative support.

    3. The tutoring process and its effects need a clearer presentation.
    3.1. The methods are vague about the process of tutoring (specifically, how many days of tutoring each bird received).
    3.2. When describing (in text and in figure panels) the effect of tutoring it is most helpful to show:
    ○ The tutors' template.
    ○ Parts of the template that were copied by the tutee. Currently, the manuscript shows newly adopted syllables but doesn't demonstrate that these syllables were copied from the tutor.
    ○ The imitation score.

    Minor points:
    1. Activity in the nucleus LMAN is thought to drive premotor activity in the nucleus RA to produce juvenile vocalizations - without needing HVC[4]. It is not clear how the 'switch' between LMAN drive and HVC drive occurs but it may be that patterned inputs to HVC, starting from the LMAN→RA activity and reaching HVC through the thalamic nucleus Uva, also drive the chains in the juvenile's HVC (or participate in its formation). The manuscript will be strengthened by referring to such literature.
    2. The introduction could potentially relate to the 'sensitive period' existing in songbirds and humans.
    3. A clearer demonstration of song imitation and isolate song may be helpful.
    4. In Figure 2 it seems that learners' songs have less harmonic structure prior to tutoring. The figure resolution does not allow making sure but it would be good to show if there are behavioral differences prior to tutoring that predict learning.

    References:
    [1] Araki, M., Bandi, M. M. & Yazaki-Sugiyama, Y. Mind the gap: Neural coding of species identity in birdsong prosody. Science 354, 1282-1287 (2016).
    [2] Kojima, S., Doupe, A. J. & Knudsen, E. I. Social performance reveals unexpected vocal competency in young songbirds. Proc. Natl. Acad. Sci. U. S. A. 108, 1687-1692 (2011).
    [3] Mackevicius, E. L. et al. Unsupervised discovery of temporal sequences in high-dimensional datasets, with applications to neuroscience. eLife 8, e38471 (2019).
    [4] Aronov, D., Andalman, A. S. & Fee, M. S. A Specialized Forebrain Circuit for Vocal Babbling in the Juvenile Songbird. Science 320, 630-634 (2008).

  5. eLife

    Reviewer #3 (Public Review):

    This paper addresses whether the sequences of neural activity that are believed to underlie song production in songbirds emerge as a result of experience-dependent tutoring or rather preceded tutored song production. The primary approach relies on calcium imaging in HVC in untutored zebra finches. The key results include the detection of neural sequences in untutored birds, and that after late tutoring the sequences associated with the tutored song can be partially attributed to pre-existing sequences. This is a short paper that addresses an important question and seems to provide significant support for the notion that neural sequences in HVC emerge independent of tutored song, and that rather than being created by tutoring, learning exploits the presence of pre-existing sequences for song generation. The results of the paper rely in large part on the extraction of neural sequences in an unsupervised fashion, while the method used does require some assumptions (such as sequence length) the conclusions seem well supported by the data.

  6. Version published to 10.1101/2022.02.18.480996v1 on bioRxiv