A reafferent and feed-forward model of song syntax generation in the Bengalese finch

Manipulations of inhibition in cortical circuitry differentially affect spectral and temporal features of Bengalese finch song

The interplay between inhibition and excitation can regulate behavioral expression and control, including the expression of communicative behaviors like birdsong. Computational models postulate varying degrees to which inhibition within vocal motor circuitry influences birdsong, but few studies have tested these models by manipulating inhibition. Here we enhanced and attenuated inhibition in the cortical nucleus HVC (used as proper name) of Bengalese finches (Lonchura striata var. domestica). Enhancement of inhibition (with muscimol) in HVC dose-dependently reduced the amount of song produced. Infusions of higher concentrations of muscimol caused some birds to produce spectrally degraded songs, whereas infusions of lower doses of muscimol led to the production of relatively normal (nondegraded) songs. However, the spectral and temporal structures of these nondegraded songs were significantly different from songs produced under control conditions. In particular, muscimol infusions decreased the frequency and amplitude of syllables, increased various measures of acoustic entropy, and increased the variability of syllable structure. Muscimol also increased sequence durations and the variability of syllable timing and syllable sequencing. Attenuation of inhibition (with bicuculline) in HVC led to changes to song distinct from and often opposite to enhancing inhibition. For example, in contrast to muscimol, bicuculline infusions increased syllable amplitude, frequency, and duration and decreased the variability of acoustic features. However, like muscimol, bicuculline increased the variability of syllable sequencing. These data highlight the importance of inhibition to the production of stereotyped vocalizations and demonstrate that changes to neural dynamics within cortical circuitry can differentially affect spectral and temporal features of song.NEW & NOTEWORTHY We reveal that manipulations of inhibition in the cortical nucleus HVC affect the structure, timing, and sequencing of syllables in Bengalese finch song. Enhancing and blocking inhibition led to opposite changes to the acoustic structure and timing of vocalizations, but both caused similar changes to vocal sequencing. These data provide support for computational models of song control but also motivate refinement of existing models to account for differential effects on syllable structure, timing, and sequencing.

Excitable neuronal assemblies with adaptation as a building block of brain circuits for velocity-controlled signal propagation

The time scale of neuronal network dynamics is determined by synaptic interactions and neuronal signal integration, both of which occur on the time scale of milliseconds. Yet many behaviors like the generation of movements or vocalizations of sounds occur on the much slower time scale of seconds. Here we ask the question of how neuronal networks of the brain can support reliable behavior on this time scale. We argue that excitable neuronal assemblies with spike-frequency adaptation may serve as building blocks that can flexibly adjust the speed of execution of neural circuit function. We show in simulations that a chain of neuronal assemblies can propagate signals reliably, similar to the well-known synfire chain, but with the crucial difference that the propagation speed is slower and tunable to the behaviorally relevant range. Moreover we study a grid of excitable neuronal assemblies as a simplified model of the somatosensory barrel cortex of the mouse and demonstrate that various patterns of experimentally observed spatial activity propagation can be explained.

Models of activity propagation in cortical networks have often been based on feedforward structures. Here we propose a model of activity propagation, called excitation chain, which does not need such a feedforward structure. The model is composed of excitable neural assemblies with spike-frequency adaptation, connected bidirectionally in a row or a grid. This prototypical neural circuit can propagate activity forwards, backwards or in both directions. Furthermore, the propagation speed is slow enough to trigger the generation of behaviors on the time scale of hundreds of milliseconds. A two-dimensional variant of the model is able to generate different activity propagation patterns, similar to spontaneous activity and stimulus-evoked responses in anesthetized mouse barrel cortex. We propose the excitation chain model as a basic component that can be employed in various ways to create spiking neural circuit models that generate signals on behavioral time scales. In contrast to abstract models of excitable media, our model makes an explicit link to the time scale of neuronal spikes.

Morphological characterization of HVC projection neurons in the zebra finch (Taeniopygia guttata)

Singing behavior in the adult male zebra finch is dependent upon the activity of a cortical region known as HVC (proper name). The vast majority of HVC projection neurons send primary axons to either the downstream premotor nucleus RA (robust nucleus of the arcopallium, or primary motor cortex) or Area X (basal ganglia), which play important roles in song production or song learning, respectively. In addition to these long-range outputs, HVC neurons also send local axon collaterals throughout that nucleus. Despite their implications for a range of circuit models, these local processes have never been completely reconstructed. Here, we use in vivo single-neuron Neurobiotin fills to examine 40 projection neurons across 31 birds with somatic positions distributed across HVC. We show that HVC_(RA) and HVC_(X) neurons have categorically distinct dendritic fields. Additionally, these cell classes send axon collaterals that are either restricted to a small portion of HVC ("local neurons") or broadly distributed throughout the entire nucleus ("broadcast neurons"). Overall, these processes within HVC offer a structural basis for significant local processing underlying behaviorally relevant population activity.

Rhythmic Continuous-Time Coding in the Songbird Analog of Vocal Motor Cortex

Songbirds learn and produce complex sequences of vocal gestures. Adult birdsong requires premotor nucleus HVC, in which projection neurons (PNs) burst sparsely at stereotyped times in the song. It has been hypothesized that PN bursts, as a population, form a continuous sequence, while a different model of HVC function proposes that both HVC PN and interneuron activity is tightly organized around motor gestures. Using a large dataset of PNs and interneurons recorded in singing birds, we test several predictions of these models. We find that PN bursts in adult birds are continuously and nearly uniformly distributed throughout song. However, we also find that PN and interneuron firing rates exhibit significant 10-Hz rhythmicity locked to song syllables, peaking prior to syllable onsets and suppressed prior to offsets-a pattern that predominates PN and interneuron activity in HVC during early stages of vocal learning.

Rhythmic syllable-related activity in a songbird motor thalamic nucleus necessary for learned vocalizations

Birdsong is a complex behavior that exhibits hierarchical organization. While the representation of singing behavior and its hierarchical organization has been studied in some detail in avian cortical premotor circuits, our understanding of the role of the thalamus in adult birdsong is incomplete. Using a combination of behavioral and electrophysiological studies, we seek to expand on earlier work showing that the thalamic nucleus Uvaeformis (Uva) is necessary for the production of stereotyped, adult song in zebra finch (Taeniopygia guttata). We confirm that complete bilateral lesions of Uva abolish singing in the ‘directed’ social context, but find that in the ‘undirected’ social context, such lesions result in highly variable vocalizations similar to early babbling song in juvenile birds. Recordings of neural activity in Uva reveal strong syllable-related modulation, maximally active prior to syllable onsets and minimally active prior to syllable offsets. Furthermore, both song and Uva activity exhibit a pronounced coherent modulation at 10Hz—a pattern observed in downstream premotor areas in adult and, even more prominently, in juvenile birds. These findings are broadly consistent with the idea that Uva is critical in the sequential activation of behavioral modules in HVC.

Temperature Manipulation in Songbird Brain Implicates the Premotor Nucleus HVC in Birdsong Syntax

Variable motor sequences of animals are often structured and can be described by probabilistic transition rules between action elements. Examples include the songs of many songbird species such as the Bengalese finch, which consist of stereotypical syllables sequenced according to probabilistic rules (song syntax). The neural mechanisms behind such rules are poorly understood. Here, we investigate where the song syntax is encoded in the brain of the Bengalese finch by rapidly and reversibly manipulating the temperature in the song production pathway. Cooling the premotor nucleus HVC (proper name) slows down the song tempo, consistent with the idea that HVC controls moment-to-moment timings of acoustic features in the syllables. More importantly, cooling HVC alters the transition probabilities between syllables. Cooling HVC reduces the number of repetitions of long-repeated syllables and increases the randomness of syllable sequences. In contrast, cooling the downstream motor area RA (robust nucleus of the acropallium), which is critical for singing, does not affect the song syntax. Unilateral cooling of HVC shows that control of syllables is mostly lateralized to the left HVC, whereas transition probabilities between the syllables can be affected by cooling HVC in either hemisphere to varying degrees. These results show that HVC is a key site for encoding song syntax in the Bengalese finch. HVC is thus involved both in encoding timings within syllables and in sequencing probabilistic transitions between syllables. Our finding suggests that probabilistic selections and fine-grained timings of action elements can be integrated within the same neural circuits.SIGNIFICANCE STATEMENT Many animal behaviors such as birdsong consist of variable sequences of discrete actions. Where and how the probabilistic rules of such sequences are encoded in the brain is poorly understood. We locally and reversibly cooled brain areas in songbirds during singing. Mild cooling of area HVC in the Bengalese finch brain-a premotor area homologous to the mammalian premotor cortex-alters the statistics of the syllable sequences, suggesting that HVC is critical for birdsong sequences. HVC is also known for controlling moment-to-moment timings within syllables. Our results show that timing and probabilistic sequencing of actions can share the same neural circuits in local brain areas.

Dynamic effective connectivity in cortically embedded systems of recurrently coupled synfire chains

As a candidate mechanism of neural representation, large numbers of synfire chains can efficiently be embedded in a balanced recurrent cortical network model. Here we study a model in which multiple synfire chains of variable strength are randomly coupled together to form a recurrent system. The system can be implemented both as a large-scale network of integrate-and-fire neurons and as a reduced model. The latter has binary-state pools as basic units but is otherwise isomorphic to the large-scale model, and provides an efficient tool for studying its behavior. Both the large-scale system and its reduced counterpart are able to sustain ongoing endogenous activity in the form of synfire waves, the proliferation of which is regulated by negative feedback caused by collateral noise. Within this equilibrium, diverse repertoires of ongoing activity are observed, including meta-stability and multiple steady states. These states arise in concert with an effective connectivity structure (ECS). The ECS admits a family of effective connectivity graphs (ECGs), parametrized by the mean global activity level. Of these graphs, the strongly connected components and their associated out-components account to a large extent for the observed steady states of the system. These results imply a notion of dynamic effective connectivity as governing neural computation with synfire chains, and related forms of cortical circuitry with complex topologies.

Neural Sequence Generation Using Spatiotemporal Patterns of Inhibition

Stereotyped sequences of neural activity are thought to underlie reproducible behaviors and cognitive processes ranging from memory recall to arm movement. One of the most prominent theoretical models of neural sequence generation is the synfire chain, in which pulses of synchronized spiking activity propagate robustly along a chain of cells connected by highly redundant feedforward excitation. But recent experimental observations in the avian song production pathway during song generation have shown excitatory activity interacting strongly with the firing patterns of inhibitory neurons, suggesting a process of sequence generation more complex than feedforward excitation. Here we propose a model of sequence generation inspired by these observations in which a pulse travels along a spatially recurrent excitatory chain, passing repeatedly through zones of local feedback inhibition. In this model, synchrony and robust timing are maintained not through redundant excitatory connections, but rather through the interaction between the pulse and the spatiotemporal pattern of inhibition that it creates as it circulates the network. These results suggest that spatially and temporally structured inhibition may play a key role in sequence generation.

Modelling Odor Decoding in the Antennal Lobe by Combining Sequential Firing Rate Models with Bayesian Inference

The olfactory information that is received by the insect brain is encoded in the form of spatiotemporal patterns in the projection neurons of the antennal lobe. These dense and overlapping patterns are transformed into a sparse code in Kenyon cells in the mushroom body. Although it is clear that this sparse code is the basis for rapid categorization of odors, it is yet unclear how the sparse code in Kenyon cells is computed and what information it represents. Here we show that this computation can be modeled by sequential firing rate patterns using Lotka-Volterra equations and Bayesian online inference. This new model can be understood as an ‘intelligent coincidence detector’, which robustly and dynamically encodes the presence of specific odor features. We found that the model is able to qualitatively reproduce experimentally observed activity in both the projection neurons and the Kenyon cells. In particular, the model explains mechanistically how sparse activity in the Kenyon cells arises from the dense code in the projection neurons. The odor classification performance of the model proved to be robust against noise and time jitter in the observed input sequences. As in recent experimental results, we found that recognition of an odor happened very early during stimulus presentation in the model. Critically, by using the model, we found surprising but simple computational explanations for several experimental phenomena.

Odor recognition in the insect brain is amazingly fast but still not fully understood. It is known that recognition is performed in three stages. In the first stage, the sensors respond to an odor by displaying a reproducible neuronal pattern. This code is turned, in the second and third stages, into a sparse code, that is, only relatively few neurons activate over hundreds of milliseconds. It is generally assumed that the insect brain uses this temporal code to recognize an odor. We propose a new model of how this temporal code emerges using sequential activation of groups of neurons. We show that these sequential activations underlie a fast and accurate recognition which is highly robust against neuronal or sensory noise. This model replicates several key experimental findings and explains how the insect brain achieves both speed and robustness of odor recognition as observed in experiments.

An Adapting Auditory-motor Feedback Loop Can Contribute to Generating Vocal Repetition

Consecutive repetition of actions is common in behavioral sequences. Although integration of sensory feedback with internal motor programs is important for sequence generation, if and how feedback contributes to repetitive actions is poorly understood. Here we study how auditory feedback contributes to generating repetitive syllable sequences in songbirds. We propose that auditory signals provide positive feedback to ongoing motor commands, but this influence decays as feedback weakens from response adaptation during syllable repetitions. Computational models show that this mechanism explains repeat distributions observed in Bengalese finch song. We experimentally confirmed two predictions of this mechanism in Bengalese finches: removal of auditory feedback by deafening reduces syllable repetitions; and neural responses to auditory playback of repeated syllable sequences gradually adapt in sensory-motor nucleus HVC. Together, our results implicate a positive auditory-feedback loop with adaptation in generating repetitive vocalizations, and suggest sensory adaptation is important for feedback control of motor sequences.

Repetitions are common in animal vocalizations. Songs of many songbirds contain syllables that repeat a variable number of times, with non-Markovian distributions of repeat counts. The neural mechanism underlying such syllable repetitions is unknown. In this work, we show that auditory feedback plays an important role in sustaining syllable repetitions in the Bengalese finch. Deafening reduces syllable repetitions and skews the repeat number distribution towards short repeats. These effects are explained with our computational model, which suggests that syllable repeats are initially sustained by auditory feedback to the neural networks that drive the syllable production. The feedback strength weakens as the syllable repeats, increasing the likelihood that the syllable repetition stops. Neural recordings confirm such adaptation of auditory feedback to the auditory-motor circuit in the Bengalese finch. Our results suggests that sensory feedback can directly impact repetitions in motor sequences, and may provide insights into neural mechanisms of speech disorders such as stuttering.

From birdsong to human speech recognition: bayesian inference on a hierarchy of nonlinear dynamical systems

Our knowledge about the computational mechanisms underlying human learning and recognition of sound sequences, especially speech, is still very limited. One difficulty in deciphering the exact means by which humans recognize speech is that there are scarce experimental findings at a neuronal, microscopic level. Here, we show that our neuronal-computational understanding of speech learning and recognition may be vastly improved by looking at an animal model, i.e., the songbird, which faces the same challenge as humans: to learn and decode complex auditory input, in an online fashion. Motivated by striking similarities between the human and songbird neural recognition systems at the macroscopic level, we assumed that the human brain uses the same computational principles at a microscopic level and translated a birdsong model into a novel human sound learning and recognition model with an emphasis on speech. We show that the resulting Bayesian model with a hierarchy of nonlinear dynamical systems can learn speech samples such as words rapidly and recognize them robustly, even in adverse conditions. In addition, we show that recognition can be performed even when words are spoken by different speakers and with different accents—an everyday situation in which current state-of-the-art speech recognition models often fail. The model can also be used to qualitatively explain behavioral data on human speech learning and derive predictions for future experiments.

Neuroscience still lacks a concrete explanation of how humans recognize speech. Even though neuroimaging techniques are helpful in determining the brain areas involved in speech recognition, there are rarely mechanistic explanations at a neuronal level. Here, we assume that songbirds and humans solve a very similar task: extracting information from sound wave modulations produced by a singing bird or a speaking human. Given strong evidence that both humans and songbirds, although genetically very distant, converged to a similar solution, we combined the vast amount of neurobiological findings for songbirds with nonlinear dynamical systems theory to develop a hierarchical, Bayesian model which explains fundamental functions in recognition of sound sequences. We found that the resulting model is good at learning and recognizing human speech. We suggest that this translated model can be used to qualitatively explain or predict experimental data, and the underlying mechanism can be used to construct improved automatic speech recognition algorithms.

A Hebbian learning rule gives rise to mirror neurons and links them to control theoretic inverse models

Mirror neurons are neurons whose responses to the observation of a motor act resemble responses measured during production of that act. Computationally, mirror neurons have been viewed as evidence for the existence of internal inverse models. Such models, rooted within control theory, map-desired sensory targets onto the motor commands required to generate those targets. To jointly explore both the formation of mirrored responses and their functional contribution to inverse models, we develop a correlation-based theory of interactions between a sensory and a motor area. We show that a simple eligibility-weighted Hebbian learning rule, operating within a sensorimotor loop during motor explorations and stabilized by heterosynaptic competition, naturally gives rise to mirror neurons as well as control theoretic inverse models encoded in the synaptic weights from sensory to motor neurons. Crucially, we find that the correlational structure or stereotypy of the neural code underlying motor explorations determines the nature of the learned inverse model: random motor codes lead to causal inverses that map sensory activity patterns to their motor causes; such inverses are maximally useful, by allowing the imitation of arbitrary sensory target sequences. By contrast, stereotyped motor codes lead to less useful predictive inverses that map sensory activity to future motor actions. Our theory generalizes previous work on inverse models by showing that such models can be learned in a simple Hebbian framework without the need for error signals or backpropagation, and it makes new conceptual connections between the causal nature of inverse models, the statistical structure of motor variability, and the time-lag between sensory and motor responses of mirror neurons. Applied to bird song learning, our theory can account for puzzling aspects of the song system, including necessity of sensorimotor gating and selectivity of auditory responses to bird's own song (BOS) stimuli.

Stepwise acquisition of vocal combinatorial capacity in songbirds and human infants

Human language, as well as birdsong, relies on the ability to arrange vocal elements in novel sequences. However, little is known about the ontogenetic origin of this capacity. We tracked the development of vocal combinatorial capacity in three species of vocal learners, combining an experimental approach in zebra finches with an analysis of natural development of vocal transitions in Bengalese finches and pre-lingual human infants and found a common, stepwise pattern of acquiring vocal transitions across species. In our first study, juvenile zebra finches were trained to perform one song and then the training target was altered, prompting the birds to swap syllable order, or insert a new syllable into a string. All birds solved these permutation tasks in a series of steps, gradually approximating the target sequence by acquiring novel pair-wise syllable transitions, sometimes too slowly to fully accomplish the task. Similarly, in the more complex songs of Bengalese finches, branching points and bidirectional transitions in song-syntax were acquired in a stepwise manner, starting from a more restrictive set of vocal transitions. The babbling of pre-lingual human infants revealed a similar developmental pattern: instead of a single developmental shift from reduplicated to variegated babbling (i.e., from repetitive to diverse sequences), we observed multiple shifts, where each novel syllable type slowly acquired a diversity of pair-wise transitions, asynchronously over development. Collectively, these results point to a common generative process that is conserved across species, suggesting that the long-noted gap between perceptual versus motor combinatorial capabilities in human infants¹ may arise from the challenges in constructing new pair-wise transitions.

Long-range order in canary song

Bird songs range in form from the simple notes of a Chipping Sparrow to the rich performance of the nightingale. Non-adjacent correlations can be found in the syntax of some birdsongs, indicating that the choice of what to sing next is determined not only by the current syllable, but also by previous syllables sung. Here we examine the song of the domesticated canary, a complex singer whose song consists of syllables, grouped into phrases that are arranged in flexible sequences. Phrases are defined by a fundamental time-scale that is independent of the underlying syllable duration. We show that the ordering of phrases is governed by long-range rules: the choice of what phrase to sing next in a given context depends on the history of the song, and for some syllables, highly specific rules produce correlations in song over timescales of up to ten seconds. The neural basis of these long-range correlations may provide insight into how complex behaviors are assembled from more elementary, stereotyped modules.

Bird songs range in form from the simple notes of a Chipping Sparrow to the complex repertoire of the nightingale. Recent studies suggest that bird songs may contain non-adjacent dependencies where the choice of what to sing next depends on the history of what has already been produced. However, the complexity of these rules has not been examined statistically for the most elaborate avian singers. Here we show that one complex singer—the domesticated canary—produces a song that is strongly influenced by long-range rules. The choice of how long to repeat a given note or which note to choose next depends on the history of the song, and these dependencies span intervals of time much longer than previously assumed for birdsong. Like most forms of human music, the songs of canaries contain patterns expressed over long timescales, governed by rules that apply to multiple levels of a temporal hierarchy. This vocal complexity provides a valuable model to examine how ordered behaviors are assembled from more elementary neural components in a relatively simple neural circuit.

Variable sequencing is actively maintained in a well learned motor skill

Variation in sequencing of actions occurs in many natural behaviors, yet how such variation is maintained is poorly understood. We investigated maintenance of sequence variation in adult Bengalese finch song, a learned skill with rendition-to-rendition variation in the sequencing of discrete syllables (i.e., syllable "b" might transition to "c" with 70% probability and to "d" with 30% probability). We found that probabilities of transitions ordinarily remain stable but could be modified by delivering aversive noise bursts following one transition (e.g., "b→c") but not the alternative (e.g., "b→d"). Such differential reinforcement induced gradual, adaptive decreases in probabilities of targeted transitions and compensatory increases in alternative transitions. Thus, the normal stability of transition probabilities does not reflect hardwired premotor circuitry. While all variable transitions could be modified by differential reinforcement, some were less readily modified than others; these were cases that exhibited more alternation between possible transitions than predicted by chance (i.e., "b→d " would tend to follow "b→c " and vice versa). These history-dependent transitions were less modifiable than more stochastic transitions. Similarly, highly stereotyped transitions (which are completely predictable) were not modifiable. This suggests that stochastically generated variability is crucial for sequence modification. Finally, we found that, when reinforcement ceased, birds gradually restored transition probabilities to their baseline values. Hence, the nervous system retains a representation of baseline probabilities and has the impetus to restore them. Together, our results indicate that variable sequencing in a motor skill can reflect an end point of learning that is stably maintained via continual self-monitoring.

Triphasic spike-timing-dependent plasticity organizes networks to produce robust sequences of neural activity

Synfire chains have long been proposed to generate precisely timed sequences of neural activity. Such activity has been linked to numerous neural functions including sensory encoding, cognitive and motor responses. In particular, it has been argued that synfire chains underlie the precise spatiotemporal firing patterns that control song production in a variety of songbirds. Previous studies have suggested that the development of synfire chains requires either initial sparse connectivity or strong topological constraints, in addition to any synaptic learning rules. Here, we show that this necessity can be removed by using a previously reported but hitherto unconsidered spike-timing-dependent plasticity (STDP) rule and activity-dependent excitability. Under this rule the network develops stable synfire chains that possess a non-trivial, scalable multi-layer structure, in which relative layer sizes appear to follow a universal function. Using computational modeling and a coarse grained random walk model, we demonstrate the role of the STDP rule in growing, molding and stabilizing the chain, and link model parameters to the resulting structure.