Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Motor cortical representations of sensorimotor information during skill learning
(USC Thesis Other)
Motor cortical representations of sensorimotor information during skill learning
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
MOTOR CORTICAL REPRESENTATIONS OF SENSORIMOTOR INFORMATION
DURING SKILL LEARNING
by
Rachel Yuan
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(NEUROSCIENCE)
May 2020
ii
Acknowledgements
I would first and foremost like to thank my advisor Dr. Sarah Bottjer. This research project
yielded surprising and unexpected results at every turn, and I could not have navigated the
challenges without her guidance and encouragement. Her confidence in my abilities often
helped carry me through times of uncertainty, and I am grateful for her fearless enthusiasm for
tackling scientific challenges head on. I owe an immeasurable amount of my personal and
professional development over these past several years to her dedicated and caring mentorship.
Thank you to past and present committee members Drs. Bosco Tjan, Andrew Hires, Radha
Kalluri, Toby Mintz, and Jason Zevin for their time. They have provided me with valuable
insight and have always shown thoughtful engagement in my research despite each being
responsible for countless other commitments.
Thank you to past and present Bottjer lab members, especially Jennifer Achiro, Amy Paterson,
Lauren Eisenman, Christopher Meyer, and Alex Chung. Our discussions have often been both
intellectually stimulating and a much appreciated source of levity and joy. Jenny and Amy
helped me solve countless experimental challenges, and Chris’ detailed technical assistance has
been invaluable.
Thank you to my beloved friends and family, who have never failed to cheer my
accomplishments and encourage me through my struggles; I am deeply grateful for their
unwavering support throughout this journey.
Table of Contents
Acknowledgements ....................................................................................................................... ii
List of Figures ............................................................................................................................... iv
List of Tables ................................................................................................................................. v
Chapter 1: Introduction ............................................................................................................... 1
Introduction to vocal learning in songbirds ............................................................................ 1
Neural circuitry for song learning and production ................................................................ 4
Introduction to avian motor cortex ......................................................................................... 6
Chapter 2: Differential developmental changes in cortical representations of auditory-vocal
stimuli in songbirds ..................................................................................................................... 11
INTRODUCTION ................................................................................................................... 11
MATERIALS AND METHODS ........................................................................................... 14
RESULTS ................................................................................................................................. 22
DISCUSSION .......................................................................................................................... 42
Chapter 3: Motor cortical representations of active behavior in juvenile zebra finches
during sensorimotor learning .................................................................................................... 49
INTRODUCTION ................................................................................................................... 49
MATERIALS AND METHODS ........................................................................................... 50
RESULTS ................................................................................................................................. 55
DISCUSSION .......................................................................................................................... 78
Chapter 4: Concluding Remarks ............................................................................................... 84
References .................................................................................................................................... 86
iv
List of Figures
Figure 1.1 Song spectrograms of zebra finch song throughout development ......................
Figure 1.2 Zebra finch male courtship dance .......................................................................
Figure 1.3 In both mammals and songbirds, parallel loops through cortex, basal ganglia,
and thalamus mediate different aspects of motor skill learning .........................
Figure 1.4 Several streams of multi-modal information converge in avian motor cortical
regions RA-cup, RA, and AId ..............................................................................
Figure 2.1 Cortical regions RA, AId, and RA-cup are situated in close proximity within
the arcopallium .....................................................................................................
Figure 2.2 Locations of multi-unit recordings in arcopallium of anesthetized juvenile
zebra finches (42-48 dph) ...................................................................................
Figure 2.3 Locations of multi-unit recordings in arcopallium of anesthetized adult zebra
finches (> 90 dph) ...............................................................................................
Figure 2.4 RA neurons of juvenile birds respond broadly to auditory vocal stimuli but
prefer juvenile over adult vocal sounds ..............................................................
Figure 2.5 RA neurons develop selectivity for bird’s own song over other vocal stimuli
by adulthood ........................................................................................................
Figure 2.6 RA neurons develop temporally patterned responses by adulthood ....................
Figure 2.7 Single-neuron song responses in RA increase in bursting activity and
suppressed events by adulthood .................................................,.........................
Figure 2.8 Neither juvenile nor adult AId sites were responsive to song playback .............
Figure 2.9 Blocking inhibition in AId does not unmask song-evoked responses ................
Figure 2.10 RA-cup neurons respond to song playback with a variety of response
patterns ................................................................................................................
Figure 3.1 AId neurons are well situated to integrate multi-modal inputs and distribute
information across various cortical-subcortical circuits .....................................
Figure 3.2 AId neurons respond during different scored movements with excitation
and/or suppression ..............................................................................................
Figure 3.3 AId neurons show a variety of temporal response patterns during different
scored movements ...............................................................................................
3
3
4
10
13
20
21
25
28
31
34
36
37
40
57
60
64
v
Figure 3.4 AId neurons exhibit context-sensitive peck responses .......................................
Figure 3.5 A substantial population of AId neurons are responsive during singing ............
Figure 3.6 AId neurons are differentially modulated during different behavioral states .....
Figure 3.7 Movement-excited AId neurons are also modulated by other unscored factors
during different state periods ..............................................................................
Figure 3.8 Activity of non-responsive AId neurons is nonetheless modulated during
different states .....................................................................................................
67
70
73
75
77
List of Tables
Table 2.1 Proportions of song-evoked sites (multi-unit) and neurons (single-unit) in RA,
AId, and RA-cup of juveniles and adults ............................................................
Table 2.2 Average number of excited and suppressed events during song playback in
juvenile and adult RA .........................................................................................
Table 3.1 Onset and offset responses across all neurons for different movement types .....
19
34
65
1
Chapter 1: Introduction
Goal-directed motor skill learning is a fundamental process that underlies acquisition of a host of
important behaviors. Songbirds serve as a powerful model for studying goal-directed skill
learning during a sensitive period of development. Similar to infants learning speech, juvenile
songbirds learn vocal sounds used for communication by first memorizing vocalizations from an
adult tutor based on auditory experience and social interactions. During a subsequent
sensorimotor stage, juveniles gradually refine their own vocalizations until they achieve an
accurate imitation of the tutor song. This latter phase of learning requires iterative comparisons
between feedback of self-generated vocal attempts and the goal tutor song; reinforcement
learning models predict that good matches between immature song attempts and the memorized
goal song will strengthen subsequent production of those sounds, whereas poor matches between
incipient song attempts and the goal song will weaken those incorrect sounds.
Avian cortex contains highly localized, interconnected regions that are dedicated to processing
song-related information. My research focuses on uniquely tractable circuits that traverse motor
cortex and mediate essential contributions to learning and performance of complex motor skills.
Avian motor cortex – particularly in the context of sensorimotor learning during development –
has been understudied. The experiments presented herein serve as initial steps to investigating
how information important for vocal learning is represented in motor cortical regions during the
sensorimotor learning period. Below, I introduce core components of vocal learning in
songbirds and the dedicated neural circuitry that mediates this process. I then review current
literature concerning the key regions in avian motor cortex that are related to song learning and
production, which are the focus of my dissertation. Chapters 2 and 3 present results from
recordings made in motor cortical areas of anesthetized and awake juvenile birds, respectively.
Introduction to vocal learning in songbirds
Adult zebra finch song is a stereotyped sequence of complex syllables that males learn and
perform as part of courtship behavior (Thorpe, 1958, 1961; Immelmann, 1969; Marler, 1970;
Price, 1979; Konishi, 1985). To achieve an accurate imitation of their tutor’s song, juvenile male
zebra finches first require auditory experience with an adult male – either their biological father
or a foster father – from ~20-35 days post hatch (Immelmann, 1969; Böhner, 1983, 1990; Eales,
2
1985, 1989; Catchpole and Slater, 2008). At approximately 35 days post hatch, a period of
sensorimotor integration begins as the juvenile starts producing variable babbling sounds,
evaluating feedback of its vocalizations against an internal representation of the memorized goal
tutor song, and refining its motor output accordingly. Through iterative comparisons of self-
generated behavior to the goal song over a period of six-seven weeks, juveniles gradually
achieve a close imitation of the tutor song and continue to produce this stereotyped vocal pattern
throughout adulthood (Fig. 1.1).
Song learning is a highly multi-modal process. Auditory experience with an external model is
clearly involved, as developing an accurate imitation of the tutor song is known to require
exposure to an adult tutor (Price, 1979; Eales, 1989; Brainard and Doupe, 2002; Derégnaucourt
et al., 2013; Chen et al., 2016; Ljubičić et al., 2016). Song learning also requires auditory
feedback: deafened juveniles develop highly abnormal songs even following otherwise normal
experience with a tutor (Konishi, 1965, 2004). However, visual processing is also a crucial
component of vocal learning: birds raised with only passive auditory exposure to an adult song
via playback from a speaker develop abnormal songs, whereas pairing playback with a visual
stimulus such as a plastic model of an adult zebra finch enhances learning (Eales, 1989;
Derégnaucourt et al., 2013; Ljubičić et al., 2016). Moreover, visual cues such as wing strokes
from adult females in response to specific song elements may serve as performance feedback that
can help direct sensorimotor refinement of the juvenile's song (West and King, 1988; King et al.,
2005). Song production is also a complex skill that involves precise control of respiratory and
vocal muscles, and proprioceptive feedback is thus likely important for vocal learning as well
(Reiner et al., 2004; Ashmore et al., 2008; Bottjer and To, 2012).
In addition, zebra finch song is part of a complex behavioral repertoire that includes not only
vocal production but also an accompanying "dance" as part of courtship performance; these non-
vocal elements are likely also learned (Morris, 1954; Williams, 2001; Cooper and Goller, 2004;
Dalziell et al., 2013; Ota et al., 2015; Ullrich et al., 2016) (Fig. 1.2). Dance movements can
include hops, head tilts, pivots back and forth along the perch in a zig-zag pattern, and beak
wiping motions against the perch as the male advances towards the female. The dance begins
and ends in alignment with the start and end of singing bouts, and specific movements can be
temporally coordinated with specific notes in the song (Williams and Suthers, 2000; Ullrich et
3
al., 2016). Successful song learning thus requires integrating multimodal streams of information
with motor programming to produce a coordinated performance that involves both vocal and
non-vocal elements.
Figure 1.1. Song spectrograms of zebra finch song throughout development. Top to bottom:
adult tutor song; juvenile’s subsong at 40 dph; plastic song at 60 dph; stereotyped adult song at
90 dph, in which the juvenile has achieved an accurate match of the adult song at top. (Adapted
from Fee and Goldberg, 2011)
Figure 1.2. Zebra finch male courtship dance. In a typical courtship performance, the male
advances towards the female, hopping and pivoting back and forth along the perch or branch.
Timing of dance movements is coordinated with vocal production. (Morris 1954)
4
Neural circuitry for song learning and production
Vocal learning in zebra finches is mediated by a discrete system of dedicated brain nuclei,
thereby allowing for targeted study of the neural substrate that underlies this developmentally
regulated learning process. In mammals, parallel cortico-basal ganglia loops allow for
processing of different aspects of skill learning, ranging from limbic to sensorimotor aspects.
Similarly, vocal learning in songbirds is mediated by parallel cortico-basal ganglia pathways that
emanate from CORE and SHELL subregions of a cortical region called LMAN (Johnson and
Bottjer, 1992; Iyengar et al., 1999; Person et al., 2008) (Fig. 1.3). The CORE circuit carries out a
motor role in juvenile birds: lesions of LMAN-CORE in juveniles severely disrupt the bird's
ability to produce normal song (Bottjer et al., 1984; Aronov et al., 2008). In contrast, SHELL
circuitry is thought to carry out an associative or evaluative function: lesions of the SHELL
pathway in juvenile birds following tutor memorization do not induce general motor deficits, but
do impair the juvenile's ability to achieve an accurate imitation of the adult tutor song (Bottjer
and Altenau, 2010). SHELL circuitry thus plays an important role in accurate refinement of the
bird's own song, suggesting a key role in mechanisms of goal-directed learning. In adults,
LMAN lesions no longer induce disruptions of song; instead, the cortical area HVC (high vocal
center) emerges as the primary driver of song output (Nottebohm et al., 1976; Bottjer et al.,
1984; Simpson and Vicario, 1990; Yu and Margoliash, 1996; Hahnloser et al., 2002; Ölveczky et
al., 2011).
Figure 1.3. In both mammals (left) and songbirds (right), parallel loops through cortex, basal
ganglia, and thalamus mediate different aspects of motor skill learning. (Left image from
Redgrave 2010)
5
In addition to cortico-basal ganglia pathways, auditory circuits are necessary for processing tutor
songs and feedback of self-generated output. Increasing evidence implicates higher-level
auditory cortical regions in encoding information about the goal tutor song (Terpstra et al., 2004;
Phan et al., 2006; London and Clayton, 2008; Gobes et al., 2010; Hahnloser and Kotowicz, 2010;
Roberts et al., 2012; Yanagihara and Yazaki-sugiyama, 2016). For example, blockade of
second-messenger (ERK) signaling in the secondary auditory region NCM (caudomedial
nidopallium) during tutor exposure disrupts subsequent song imitation (London and Clayton,
2008). Moreover, microstimulation of the auditory cortical area NIf (telencephalic nucleus
interface) that is contingent on tutor song playback also impairs tutor song copying (Roberts et
al., 2012). There are two major pathways through which information from these higher-order
auditory areas could integrate with cortico-basal ganglia circuitry. One pathway is through a
projection to HVC, which contains a population of neurons that project to the portion of avian
basal ganglia that mediates song learning, Area X. The other pathway involves projections to a
region in avian motor cortex, RA-cup, which projects to dopaminergic neurons in the ventral
tegmental area (VTA) that in turn project to Area X (Fig. 1.3; former not shown for clarity).
Neurons in song-learning regions are highly sensorimotor: many song-control regions contain
neurons that are both active when birds are singing and responsive to playback of auditory-vocal
stimuli in anesthetized birds. Investigating responsivity of neurons to song playback is a useful
measure for gaining functional insight (Solis et al., 2000; Brainard and Doupe, 2002, 2013;
Theunissen et al., 2004). For instance, at the start of the sensorimotor learning period, LMAN-
CORE neurons which drive song output in juvenile birds respond broadly to playback of a variety
of song stimuli, ranging from the variable immature vocalizations of juveniles to the adult
stereotyped songs (Doupe, 1997; Solis and Doupe, 1997; Achiro and Bottjer, 2013). CORE
neurons become increasingly selective for each bird’s own song throughout development; even
juvenile birds who are induced to sing abnormal songs via tracheosyringeal nerve transections
possess a large number of CORE neurons that selectively respond to playback of their disrupted
songs over other conspecific songs (Solis and Doupe, 2000). This pattern of tuning may reflect a
neural representation of the bird’s own vocal motor output, which could serve a filter-like
function to aid in distinguishing self-generated vocalizations from those of others. Song
responsivity in LMAN-SHELL similarly provides insight to the function of this pathway: LMAN-
6
SHELL in juvenile birds is the sole known locus of neurons that respond selectively to playback of
the tutor song (Achiro and Bottjer, 2013). These neurons could thus serve as a neural
representation of the goal tutor song that the juvenile is trying to imitate. Remarkably, the
population of tutor-tuned neurons is present in SHELL only in juvenile birds, when a template of
the tutor song is crucial for goal-directed learning (Achiro and Bottjer, 2013). The
developmentally regulated presence of these tutor-selective neurons supports the idea of SHELL
circuitry playing a role in evaluative comparisons that guide vocal learning.
Akin to mammalian systems, cortical processing for goal-directed behaviors in these circuits
ultimately converge in avian motor cortex, serving as the final telencephalic station after which
information is either fed forward to direct motor output or integrated with various subcortical
circuits. Given this location within the circuitry, avian motor cortex is likely a crucial hub for
integrating diverse, multimodal streams of information to help mediate song learning and
production. This thesis highlights three areas of avian motor cortex – RA (robust nucleus of the
arcopallium), RA-cup, and AId (dorsal intermediate arcopallium), each of which is uniquely
situated to contribute to different aspects of song learning. RA-cup receives inputs from higher-
order auditory cortical regions, whereas RA and AId are the motor cortical output target regions
of LMAN-CORE and LMAN-SHELL, respectively (Fig. 1.4).
Introduction to avian motor cortex
RA
RA has been compared to layer 5b of mammalian motor cortex, based on both anatomical and
functional analogies: RA neurons project directly to brainstem motor neurons to drive vocal
motor output (Nottebohm et al., 1976; Wild, 1993, 2004). Single RA neurons receive dual input
from LMAN-CORE and HVC (Herrmann and Arnold, 1991; Mooney and Konishi, 1991;
Mooney, 1992; Stark and Perkel, 1999). LMAN-CORE lesions severely disrupt song behavior
only in juvenile birds, whereas HVC lesions disrupt song behavior in adult birds (Nottebohm et
al., 1976; Bottjer et al., 1984; Simpson and Vicario, 1990; Yu and Margoliash, 1996; Hahnloser
et al., 2002; Aronov et al., 2008; Ölveczky et al., 2011). HVC inputs onto RA neurons are
strengthened and subsequently pruned during the sensorimotor learning period, and the change in
7
relative strength of LMAN versus HVC synapses may contribute to the developmental shift in
cortical control of song behavior from the former to the latter (Garst-Orozco et al., 2014).
In adult birds, RA neurons respond more strongly to playback of each bird’s own stereotyped
song than to other songs, suggesting selectivity for the bird’s own behavior that may reflect an
experience-dependent neural representation of the bird’s learned vocalizations (Doupe and
Konishi, 1991; Vicario and Yohay, 1993; Dave et al., 1998; Dave and Margoliash, 2000).
Similarly, neurons in LMAN-CORE and HVC of adult songbirds are selective for playback of the
bird's own song over other conspecific songs (Margoliash and Konishi, 1985; Margoliash, 1986;
Doupe and Konishi, 1991). However, few studies have investigated the tuning of RA neurons in
juvenile birds during early stages of sensorimotor learning, when birds produce variable
vocalizations in attempts to develop an accurate match to the tutor song.
RA-cup
RA-cup is defined by its inputs from primary and secondary auditory cortical areas; these
projections form a “cup” surrounding RA, with particularly dense innervation in the area rostral
and rostro-ventral to the anterior border of RA (Kelley and Nottebohm, 1979; Vates et al., 1996;
Mello et al., 1998). These sources of auditory information could encode feedback of the bird’s
self-produced song or auditory experience with the adult tutor song.
RA-cup makes descending projections to several thalamic and brainstem regions that lie in close
relation to nuclei of the ascending auditory pathway, potentially creating feedback loops by
which auditory information integrated in higher level auditory cortical areas can modulate
ascending information (Vates et al., 1996; Mello et al., 1998; Guillery and Sherman, 2002;
Briggs and Usrey, 2008). As mentioned above, the rostro-ventral region of RA-cup projects to
areas of dopaminergic neurons in the substantia nigra and VTA that overlap with cells that
project to Area X, providing a pathway by which auditory information could directly interact
with vocal learning circuitry; thus, RA-cup may also serve as a means of relaying auditory
information important for vocal learning to the song system (Gale et al., 2008). In addition,
modulation of dopamine release from VTA may contribute to goal-directed vocal learning.
Studies of regions in avian motor cortex that overlap with the VTA-projecting portion of RA-cup
have suggested that this subset of neurons can drive adaptive plasticity in adult song
8
(Mandelblat-Cerf et al., 2014; Xiao et al., 2018). However, no published studies have
investigated the selectivity of RA-cup neurons for different auditory-vocal stimuli important for
vocal learning.
AId
Functional and anatomical evidence highlight AId as a region particularly important for vocal
learning. Lesions of AId in juvenile birds do not induce any motor disruption, but prevent the
juveniles from achieving an accurate copy of the tutor song by adulthood (Bottjer and Altenau,
2010). This result strongly suggests a role for AId in accurately directing the refinement of the
juvenile’s motor output during development. Moreover, one of AId’s efferent projection targets
includes the region of VTA that projects to Area X, providing a pathway by which information
in AId could inform dopaminergic signaling to guide the vocal learning process (Bottjer et al.,
2000).
AId receives input from LMAN-SHELL and dNCL (dorsal caudolateral nidopallium) (Bottjer et
al., 2000). LMAN- SHELL contains subpopulations of neurons that respond selectively to
playback of the bird’s own song or the tutor song (Achiro and Bottjer, 2013). Either of these
selective neural representations could be conveyed to AId, and both would be important for
vocal learning: for instance, selective tuning for the bird’s own song could help process auditory
feedback of the juvenile’s self-generated vocalizations, while tutor-selective neurons could serve
as a template of a goal behavior against which the juvenile’s vocalizations are compared.
Evaluative comparisons between these two representations would help guide accurate refinement
of the juvenile’s song. dNCL receives visual, auditory, and somatosensory information and has
also been implicated in complex learning behaviors such as visual imprinting in chicks and
working memory tasks in pigeons (Leutgeb et al., 1996; Gunturken, 1997; Metzger et al., 1998;
Braun et al., 1999; Bottjer et al., 2010; Paterson and Bottjer, 2017). Multimodal sensory
information may thus be integrated with song-related information from LMAN-SHELL in AId.
In juveniles only, single LMAN-CORE neurons that project to RA branch and send a collateral
projection into AId (Miller-Sims and Bottjer, 2012). Thus, auditory-vocal information from
song-responsive neurons in LMAN- CORE is also available to AId, specifically during the vocal
learning period. As CORE projection neurons carry motor commands to RA to drive vocal
9
output, these axon collaterals may also convey an efference copy of vocal output to AId during
learning. These inputs position AId as a developmentally regulated site of convergence, well-
situated to integrate a variety of information that is crucial for vocal learning.
Given the convergence of auditory, visual, sensorimotor, and cortico-basal ganglia circuitry in
avian motor cortex, RA, RA-cup, and AId are well-situated to integrate sensorimotor information
and contribute to different aspects of vocal learning and production. Chapter 2 of this
dissertation tests the neural responsivity of behaviorally relevant auditory vocal stimuli in these
regions in juvenile versus adult birds to investigate how information important for learning is
integrated and processed. Chapter 3 details recordings in AId of awake, behaving juvenile birds
as a first step to understanding how motor cortical regions process multimodal information in
juveniles who are actively engaged in sensorimotor learning.
10
Figure 1.4. Several streams of multi-modal information converge in avian motor cortical
regions RA-cup, RA, and AId. RA-cup receives information from higher level auditory cortical
regions and in turn projects to VTA and midbrain areas that surround auditory thalamic nuclei
(latter not shown for clarity). RA and AId receive input from LMAN-CORE and SHELL,
respectively, which mediate song learning through cortico-basal ganglia circuitry. RA also
receives input from HVC. RA projects to vocal motor circuits that drive song output; motor
drive of vocal output via RA is thought to shift from LMAN-CORE to HVC during development.
AId also receives input from dNCL, which processes multi-modal information. In juvenile birds,
AId receives input from LMAN-CORE → RA collaterals as well. AId makes several projections
to midbrain and thalamic regions to create recurrent feedback loops. Abbreviations: AId, dorsal
intermediate arcopallium; dNCL, dorsal caudolateral nidopallium; LMAN, lateral magnocellular
nucleus of the anterior nidopallium; RA, robust nucleus of the arcopallium; VTA, ventral
tegmental area
11
Chapter 2: Differential developmental changes in cortical representations of
auditory-vocal stimuli in songbirds
INTRODUCTION
Procedural learning underlies acquisition of diverse motor skills, ranging from the highly-trained
motor abilities of professional athletes to universal skills such as speech. Initially variable
attempts to execute a target motor pattern are refined based on evaluative comparisons between
feedback of self-generated motor output and an internal representation of the goal behavior. This
sensorimotor learning process requires neurons to encode information about both the animal’s
own output and the goal behavior that the animal is attempting to imitate; these two neural
representations provide the basis for comparisons between self-generated and target behaviors,
and hence are essential for guiding accurate refinement of motor behavior.
Experience-dependent representations of motor skills are altered over the course of goal-directed
learning as an accurate match to the target behavior is gradually achieved (Makino et al., 2016;
Peters et al., 2017b; Papale and Hooks, 2018). For instance, in layer 2/3 of rodent motor cortex,
the population of neurons that participates in initial attempts to perform the target behavior
expands during early phases of learning before being refined into a stereotyped ensemble that
becomes reliably associated with the learned behavior (Peters et al., 2014). However, how such
dynamic changes in neural representations occur is not well understood, particularly for learned
behaviors such as speech that are acquired during a sensitive period of development. Moreover,
while evidence suggests that cortical representations of motor behaviors differ between
superficial versus deep cortical layers, the role of the latter in encoding information important for
goal-directed learning remains poorly understood (Masamizu et al., 2014; Peters et al., 2017a).
Vocal learning in songbirds serves as a powerful model for studying neural representations
essential for goal-directed acquisition of a stereotyped behavior. We investigated neural
selectivity for behaviorally relevant stimuli such as self-generated song and target tutor sounds
within cortical regions RA, AId, and RA-cup of both juvenile birds in early stages of learning
and adults that had developed stereotyped motor behavior (Fig. 2.1). RA provides the sole
output of all telencephalic circuitry relating to vocal behavior and projects directly to the motor
neurons that drive vocal output (Wild, 2004). In addition, RA neurons in adult birds respond
12
selectively to playback of each bird’s own individual song, suggesting selectivity for self-
generated actions that may reflect the development of an experience-dependent neural
representation of the bird’s learned vocalizations (Doupe and Konishi, 1991; Vicario and Yohay,
1993; Kojima and Doupe, 2007). However, little is known about how self-generated and goal
behaviors are represented in this motor output region in juvenile birds. Understudied regions
that surround RA, AId and RA-cup, receive inputs from regions that mediate vocal learning and
behavior and are therefore well-situated to integrate auditory-vocal information that may help
guide sensorimotor learning (Kelley and Nottebohm, 1979; Wild, 1993; Mello and Clayton,
1994; Johnson et al., 1995; Vates et al., 1996; Mello et al., 1998; Bottjer et al., 2000; Miller-Sims
and Bottjer, 2012; Mandelblat-Cerf et al., 2014; Paterson and Bottjer, 2017). RA and AId have
been compared to layers 5 and 6 of motor cortex, respectively, while RA-cup has been compared
to deep layers of auditory cortex (Zeier and Karten, 1971; Jarvis, 2004; Wang et al., 2010;
Karten, 2013). Results presented here thus inform our understanding of how neurons encode
information about self-produced and goal behaviors in order to guide motor learning.
13
Figure 2.1. Cortical regions RA, AId, and RA-cup are situated in close proximity within the
arcopallium. Each receives input from cortical areas containing information about auditory
stimuli important for vocal learning. LMAN is a cortical nucleus with CORE and SHELL
subregions which convey the output of parallel cortico-basal ganglia circuits; both CORE and
SHELL contain song-responsive neurons and are essential for vocal learning. HVC is required for
song production in adult birds and also contains song-responsive neurons. RA receives direct
inputs from both HVC and LMAN-CORE and projects to downstream motor circuits that drive
vocal output. AId of juvenile birds receives inputs from both LMAN-CORE and LMAN-SHELL;
the CORE inputs are collaterals from LMAN-CORE → RA axons and are not present in adult birds.
RA-cup receives inputs from primary and secondary auditory cortical areas. Both AId and
regions of RA-cup project to dopaminergic neurons in VTA, which project in turn to song-
control basal ganglia circuity. Abbreviations: AId, dorsal intermediate arcopallium; DTZ, dorsal
thalamic zone; HVC, vocal premotor cortex (proper name); LMAN, lateral magnocellular
nucleus of the anterior nidopallium; RA, robust nucleus of the arcopallium; VTA, ventral
tegmental area.
14
MATERIALS AND METHODS
Subjects
All animal procedures were performed in accordance with the University of Southern California
animal care committee’s regulations. Thirty-two urethane-anesthetized juvenile (42 – 48 dph; n
= 18) and adult (> 90 dph; n = 14) male zebra finches were used. Table 2.1 indicates the number
of birds used in each data set. All birds were raised in group aviaries until at least 35 dph,
remaining with their natural parents and thereby receiving normal auditory and social experience
during the tutor memorization period. Juvenile birds ~45 dph are in early stages of sensorimotor
integration.
Song stimuli
Birds were placed in individual sound attenuation chambers and songs were recorded using
Sound Analysis Pro (44 kHz sampling rate; Tchernichovski et al. 2000). At least 15 song bouts
were inspected to pick a motif that represented the most frequently produced song for each bird.
For juvenile birds, the bird’s own song (OWN) was selected from recordings made within a day
of neural recordings. Juvenile conspecific songs (JUV-CON) were recorded from non-sibling
juveniles matched to the age of experimental birds. Adult conspecific songs (ADL-CON) were
recorded from retired breeders no longer present in the breeding aviaries. To generate stimuli for
playback, each song recording was edited to select segments of song behavior ranging from one
to two seconds; bouts of syllables were used for juveniles, who do not produce stereotyped
motifs at this age; two song motifs were used for adults. Stimuli were high-pass filtered at 400
Hz. Amplitude was equalized to 67 ± 1 dB SPL (measured with Extech 407735 sound level
meter, A-weighting) and played from a speaker placed 25 cm directly in front of the bird.
Twenty iterations of each stimulus were presented, with stimuli randomly interleaved and
presented at an inter-stimulus interval of 10 ± 2 sec.
Electrophysiology
One to two days prior to each recording session, birds were anesthetized with 1.5% isoflurane
(inhalation) and placed in a stereotaxic instrument; a stainless steel post was attached to the
rostral skull with dental cement. On the day of the experiment, birds were anesthetized with
20% urethane in dH2O, given as three separate intramuscular injections of 33-38 μl 30 min apart.
The steel post was fixed to the stereotax, thereby holding the bird’s head at a fixed angle while
15
leaving the ears unobstructed. Extracellular recordings were made using single glass electrodes
(0.5 – 3.5 MΩ), multi-barreled glass electrodes (Kation Scientific, Carbostar-3; 0.4 – 1.4 MΩ),
4x4 16-channel arrays (NeuroNexus, A4x4-3mm-100-125-177; 0.4 – 2.8 MΩ), or single-prong
16-channel arrays (NeuroNexus, A1x16-3mm-100-125-177; 0.4 – 2.8 MΩ). Data were
amplified, band passed between 300 and 5000 Hz (Neuralynx, Lynx-8 Amplifiers), and digitized
at 20 or 32 kHz using Spike 2 software (CED, Power1401 data acquisition interface). At the end
of each experiment, electrolytic lesions were made to verify recording location. Birds were
perfused (0.7% saline followed by 10% formalin), and brains were removed and post-fixed for
72 hours before being cryo-protected (30% sucrose solution) and frozen-sectioned (50 μm thick)
in the coronal plane. Sections were Nissl-stained with thionin to visualize recording sites.
Gabazine micro-iontophoresis
Neural recordings and micro-iontophoresis were performed with multi-barreled glass pipettes
consisting of a carbon fiber electrode (5 μm diameter, 0.4-0.8 MΩ impedance) and two attached
barrels for micro-iontophoresis (Kation Scientific, Carbostar-3). One barrel was filled with 0.9%
NaCl solution for capacitance compensation and the other was filled with the GABAA receptor
antagonist gabazine (SR95531, 2.7 mM; Sigma Aldrich). An iontophoresis pump (NeuroPhore
BH-2, Harvard Apparatus) was used to apply retaining currents ranging from -20 to -25 nA and
ejection currents ranging from 35 to 55 nA to control drug application. At each site, a constant
retaining current was applied during control song presentations (20 trials per song). After this
control period, ejection current was applied in order to administer gabazine; song presentations
were initiated two minutes after onset of ejection current.
Data Analysis
Recording sites within RA, AId, or RA-cup (based on histological examination) were considered
for further analysis; sites were considered to be in RA-cup if they fell within 300 μm of the area
directly anterior to RA, or within 600 μm ventral to anterior RA. Single units were sorted from
multi-unit data by first automatically clustering units with KlustaKwik (KD Harris, University
College London) based on the maximum and first derivative of energy of the waveforms, the
first principal components, and the rising and falling slope of the waveforms. After KlustaKwik
sorting, clusters were manually inspected across 18 different waveform features and further
16
refined as needed (MClust-3.5, AD Redish, University of Minnesota). Clusters were considered
for analysis only if <1% of spikes had an inter-spike interval < 2 ms.
Standardized Response Strength
Multi-unit sites and single neurons were tested for song responsivity by comparing the change in
average firing rate between stimulus presentation and preceding baseline (1.5 seconds; paired t-
test, p < .05; Benjamini-Hochberg correction for multiple comparisons across songs). Spiking
activity was considered to be song-responsive if at least one song stimulus evoked a significant
response. To compare responses across animals, standardized response strength was calculated
as:
𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑𝑖𝑧𝑒𝑑 𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒 𝑠𝑡𝑟𝑒𝑛𝑔𝑡 ℎ =
𝐹𝑅
̅ ̅ ̅ ̅
𝑠 − 𝐹𝑅
̅ ̅ ̅ ̅
𝑏 √𝑉𝑎 𝑟 (𝐹𝑅
𝑠 ) + 𝑉𝑎𝑟 (𝐹𝑅
𝑏 ) − 2 × 𝐶𝑜𝑣𝑎𝑟 (𝐹𝑅
𝑠 , 𝐹𝑅
𝑏 )
where FRs is the firing rate during stimulus presentation and FRb is the firing rate during
baseline. This measure is referred to as response strength throughout the text. Absolute values
of response strength were used in order to include suppressed responses in analyses.
To measure selectivity, difference scores between the standardized response strengths (sRS) of
stimulus pairs were calculated. For example, selectivity for “Song A” over “Song B” was
calculated as SelectivitySongA = sRSA – sRSB, where a positive value indicates a preference for
“Song A” and a negative value indicates preference for “Song B”. Signs of difference scores for
song-suppressed neurons were reversed in order to include all song-responsive neurons in these
selectivity measures, regardless of whether they showed increases or decreases in firing rates.
This measure is an alternative to the discriminability index d’, where response strengths between
two stimuli are first subtracted before being divided by the standard deviations. Subtracting
response strengths prior to normalization makes d’ sensitive to absolute differences in response
strength; the selectivity score used here resolves this issue by first normalizing response
strengths by standard deviation before subtraction.
Burst Fraction
Histograms of inter-spike intervals in both juvenile and adult RA neurons typically exhibited a
prominent peak at 2-3 ms. Based on these distributions, bursts were defined as groups of two or
17
more spikes separated by < 5 ms. The burst fraction was calculated by dividing the number of
spikes contained within bursts by the total number of spikes from each baseline or stimulus
period.
Correlation Index
A correlation index (CI) was calculated based on the method used by Joris et al. (Louage et al.,
2004; Joris et al., 2006). Spike trains across iterations of playback (20 trials for each song) were
binned into 1-ms windows and the number of coincident spikes that occurred within the same
time window across all pair-wise comparisons between spike trains was quantified. This count
(Nc) was then scaled to be independent of average firing rate (r), number of trials (M),
coincidence window size (ω), and stimulus duration (D) in order to generate a correlation index:
𝐶𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝐼𝑛𝑑𝑒𝑥 =
𝑁 𝑐 𝑀 (𝑀 − 1)𝑟 2
𝜔𝐷
A value of 1.0 indicates a lack of temporal structure whereas larger values indicate greater
degrees of correlation in spike timing.
To estimate significance of CI values, we compared the CI of each cell’s response to each song
against a null distribution of CI values generated by 1000 random shuffles of spike trains from
the observed response. We measured the proportion of CI values from shuffled responses that
were lower than that of the observed response and expressed this measure as a percentile. Song
responses with a CI value that fell above the 95
th
percentile of the respective null distributions
were considered to exhibit significant temporal structure. To compare the degree of temporal
patterning across cells, we calculated mean CI percentiles across cell responses to each song.
Normalized correlograms were generated from the spiking data by examining all possible pairs
of spike trains and measuring the forward time intervals between each spike from the first spike
train and each spike of the second spike train. These intervals were tallied in a histogram which
was then normalized to generate a correlogram, where height of peaks represents degree of
coincident spiking and the number of distinct peaks represents the number of coincident spiking
events across trials (Louage et al., 2004; Joris et al., 2006).
18
Excited and Suppressed Events
Stereotyped periods of excitation and suppression during song-evoked responses were
characterized based on the method used by Billimoria et al. (Billimoria, 2006). PSTHs with 5-
ms bins were constructed for each single-unit response to each song. A PSTH bin was
considered significantly different from baseline if the number of spikes it contained was greater
than two standard deviations (for excitation) or less than one standard deviation (suppression) of
the baseline response. Excited or suppressed “events” were defined as three or more contiguous
significant bins separated by at least three non-significant bins. Thus, events reflect consistent
periods of excitation or suppression across trials. This method was used to examine temporally
aligned song responses of single units in RA; although some temporally patterned responses
were also evoked from multi-unit sites in RA-cup, firing rate across recordings was highly
variable and thus calculation of events using fixed parameters across sites was not appropriate.
Statistics
All continuous data were tested for normality with Shapiro-Wilk tests. Pending normality of the
data set, one-way ANOVAs or Kruskal-Wallis tests were used to test significance of response
strengths, correlation indices, and burst fractions among stimuli in juveniles and adults.
Benjamini-Hochberg post-hoc tests were used to apply a correction for multiple comparisons
(Benjamini and Hochberg, 1995). Because single-unit data from adult RA was not normally
distributed, Mann-Whitney tests were used to compare the response strengths, OWN selectivity
over other stimuli, correlation indices, burst fraction, and excited and suppressed event counts
between juvenile versus adult RA recordings. Wilcoxon signed-rank tests were used to test
whether the number of excited and suppressed events during stimulus presentation differed from
baseline. For comparisons of proportions between two groups (for example, proportion of song-
responsive neurons in juvenile versus adult RA), Fisher’s exact tests were used. For
comparisons of proportions between more than two groups (for example, proportions of neurons
that responded to each song stimulus), χ
2
tests were performed; if significant, Fisher’s exact tests
were used for follow-up comparisons between two groups.
19
Table 2.1: Proportions of song-evoked sites (multi-unit) and neurons (single-unit) in RA,
AId, and RA-cup of juvenile and adults
Juveniles Adults
n Multi-unit sites Single units n Multi-unit sites Single units
RA
11 .97 (30/31)
excit 1.0 (30/30)
supp 0.0 (0/30)
.62 (31/50)
excit .87 (27/31)
supp .13 (4/31)
12 1.0 (31/31)
excit 1.0 (31/31)
supp 0.0 (0/31)
1.0 (35/35)
excit 1.0 (35/35)
supp 0.0 (0/35)
AId
6 0.0, (0/19) 4 .04, (1/23)
excit 0.0 (0/1)
supp 1.0 (1/1)
RA-cup
7 .40, (21/53)
excit .81 (17/21)
supp .19 (4/21)
5 .63, (46/73)
excit .67 (31/46)
supp .33 (15/46)
Numbers in parentheses indicate numbers of sites or cells that demonstrated a song-evoked
response out of total recorded.
n = number of birds
excit = excited response; supp = suppressed response
20
Figure 2.2. Locations of multi-unit recordings in arcopallium of anesthetized juvenile zebra
finches (42 – 48 dph). A-G, Caudal-to-rostral series of coronal sections mapping locations of
extracellular recordings made in RA, AId, and RA-cup; 150 μm between sections, ranging from
the posterior edge of RA (A) to 300 μm anterior to RA (G). Medial is left; dorsal is up. Black
circles represent RA sites that responded to at least one stimulus; gray circles represent
responsive sites within RA-cup. Black and gray crosses represent non-responsive sites in
RA/AId and RA-cup, respectively.
21
Figure 2.3. Locations of multi-unit recordings in arcopallium of anesthetized adult zebra finches
(> 90 dph). A-G, Caudal-to-rostral series of coronal sections mapping locations of extracellular
recordings made in RA, AId, and RA-cup; 150 μm between sections, ranging from the posterior
edge of RA (A) to 300 μm anterior to RA (G). Medial is left; dorsal is up. Black circles represent
RA/AId sites that responded to at least one stimulus; gray circles represent responsive sites within
RA-cup. Black and gray crosses represent non-responsive sites in AId and RA-cup, respectively.
22
RESULTS
We made extracellular recordings in RA, AId, and RA-cup of urethane-anesthetized juvenile
(40-48 dph) and adult (>190 dph) zebra finches during playback of each bird’s own song
(OWN), a reverse of each bird’s own song (REV-OWN), juvenile and adult conspecific songs
(JUV-CON and ADL-CON), and each bird’s tutor song (TUT). An evoked response was
defined as a significant change in average firing rate (excitation or suppression) during song
playback relative to baseline (see Materials and Methods). Figures 2.2 and 2.3 illustrate
locations of the recording sites made in juveniles and adults, respectively, with circles indicating
song-responsive sites and crosses indicating non-responsive sites.
Patterns of neural responsivity to song playback in juvenile and adult RA
Among single RA neurons in juvenile birds, 62% (31/50) showed a significant change in firing
rate during playback of at least one song stimulus and were therefore considered to be song-
responsive, whereas all but one multi-unit site met this criterion (Table 2.1). Firing rates in four
of the song-responsive neurons were suppressed during playback; the remaining 27 showed
excited responses. In adult RA, 100% of both multi-unit sites and single neurons were song-
excited; none showed response suppression. The proportion of song-responsive neurons was
significantly different between juveniles and adults (Fisher’s exact test, p < .001), indicating that
the incidence of individual RA neurons showing song-evoked responses increased substantially
between the sensorimotor learning period and adulthood. All subsequent analyses were
restricted to song-responsive sites and neurons except where noted.
Neurons in RA of juvenile birds are broadly responsive to vocal stimuli but respond more
strongly to juvenile than to adult vocal sounds
Neurons in juvenile RA were broadly responsive to a variety of vocal stimuli. Figure 2.4A (left)
shows that 90% (18/20) of multi-unit sites for which all five songs were played showed an
evoked response to two or more stimuli, with 55% (11/20) responding to all five songs.
Similarly, among single neurons for which all five stimuli were played, 90% (19/21) responded
to playback of two or more songs and 58% (12/21) responded to four or five songs (Fig. 2.4A,
middle). The chart on the right in Figure 2.4A depicts the songs to which each neuron
responded, demonstrating that the stimuli that evoked a response varied across individual
23
neurons. Thus, we did not observe a subpopulation of neurons that responded primarily to one
song stimulus. However, a greater proportion of neurons responded to OWN and JUV-CON
compared to TUT (Fig. 2.4B; Fisher’s exact test, OWN p = .008, JUVCON p = .03). Moreover,
when we ranked the response strength evoked by each song for each neuron, we found that
OWN and REV-OWN evoked the strongest response most often (7/21 and 6/21 neurons,
respectively), whereas TUT evoked the strongest response in only one neuron (Fig. 2.4C). The
combined proportion of neurons that responded most strongly to any juvenile stimulus was 76%
(16/21, blue in Fig. 2.4C), whereas the combined proportion of neurons that responded most
strongly to an adult stimulus was 24% (5/21, red in Fig. 2.4C; Fisher’s exact test, p = .002).
These data show that neurons in RA of birds engaged in early sensorimotor learning respond
broadly to different song stimuli, but tend to respond more strongly to juvenile song sounds
compared to stereotyped adult vocal stimuli.
Across cells, mean values of absolute response strength were similar for the three juvenile
stimuli (OWN, JUV-CON, and REV-OWN) at both multi- and single-unit levels (Fig. 2.4D;
ANOVA, Benjamini-Hochberg correction for multiple comparisons, p > .05 in all cases). The
bottom two rows of Figure 2.4E show rasters and PSTHs for a juvenile RA neuron that
responded equally well to playback of all three juvenile songs (sorted from the multi-unit
recording site in the raw traces above). This broad responsivity included a lack of selectivity for
forward versus reversed song stimuli: multi- and single-unit responses to OWN and REV-OWN
did not differ (multi-unit p = .41, single-unit p = .36). Thus, neurons in juvenile RA responded
comparably to immature vocal sounds regardless of the temporal patterning of those sounds.
Moreover, neural activity in juvenile RA did not distinguish the bird’s own immature
vocalizations from other juvenile songs (Fig. 2.4D, E), indicating that neurons were not selective
for self-generated vocalizations.
In comparison, we found significant differences when measuring responses to juvenile versus
adult song stimuli: the multi-unit response to each of the three juvenile stimuli was greater than
the response to TUT, and the responses to OWN and REV-OWN were greater than that to ADL-
CON (Fig. 2.4D, left; ANOVA p < .001, Benjamini-Hochberg corrected, p < .01 for TUT
comparisons, p < .05 for ADL-CON comparisons). Similarly, single-unit responses to OWN and
JUV-CON were greater than those to TUT (Fig. 2.4D, right; ANOVA p = .02, Benjamini-
24
Hochberg corrected, OWN p = .01, JUV-CON p = .02). Neither multi- nor single-unit responses
differed between the adult stimuli (ANOVA, Benjamini-Hochberg corrected, TUT versus ADL-
CON p = .07 for multi-unit, p = .16 for single-unit), indicating that juvenile RA neurons did not
distinguish the memorized tutor song from unfamiliar adult vocal sounds but rather responded
more strongly to juvenile sounds over both adult songs.
Overall, these results show that RA neurons of juvenile birds in early stages of sensorimotor
integration respond broadly to a variety of juvenile and adult vocal stimuli. However, neurons
respond more strongly to juvenile stimuli such as OWN and JUV-CON compared to adult
stimuli, suggesting that these neurons prefer acoustic features inherent to immature juvenile
vocalizations over adult vocal patterns.
25
Figure 2.4. RA neurons of juvenile birds respond broadly to auditory vocal stimuli but prefer
juvenile over adult vocal sounds. A, Proportion of juvenile RA sites (left) and single neurons
(middle) that responded to different numbers of stimuli. Multi-unit proportions are out of 20
sites to which all five stimuli were played; single-unit, out of 21 neurons. Each row in chart at
right indicates stimuli to which each neuron responded. B, Proportion of single neurons that
responded to each stimulus: 25/31 responded to OWN, 19/25 to JUV-CON, 16/27 to REV-
OWN, 17/30 to ADL-CON, 14/31 to TUT. Significance was tested using a χ
2
test. C, Proportion
of neurons that demonstrated the greatest response strength to each stimulus. Blue shading
depicts juvenile stimuli; red shading depicts adult stimuli. D, Mean standardized response
strengths to each stimulus across multi-unit sites (left) and single neurons (right). Plus signs
mark means; box plots depict medians and first, third quartiles; circles represent individual data
points. Significance was tested using a one-way ANOVA, Benjamini-Hochberg corrected. E,
Extracellular recording in juvenile RA during playback of OWN, REV-OWN, and JUV-CON.
Top: song spectrograms and raw traces of multi-unit activity. Middle: rasters of a single neuron
sorted from the multi-unit recording above (overlaid waveforms shown in inset at top right).
Bottom: PSTHs; RS, mean response strength. * p<.05, **p<.01, ***p<.005.
26
RA neurons develop a strong preference for self-generated song by adulthood
As in juvenile birds, sites and neurons in adult RA were broadly responsive: 100% of sites and
91% (31/34) of neurons to which all four adult stimuli were played responded to two or more
stimuli (compared to 90% of sites and neurons in juveniles; Fig. 2.5A; Fisher’s exact test, p > .05
in both cases). However, neurons in adult RA were more likely to respond to OWN than to any
other stimulus: 97% (34/35) of single units responded to playback of OWN, compared to 26%
(9/35) for REV-OWN, 56% (19/34) for ADL-CON, and 76% (26/34) for TUT (Fig. 2.5A, B;
Fisher’s exact test, p < .02 in all cases). This strong tendency to respond to OWN contrasts with
juvenile RA, in which similar proportions of neurons responded to OWN and other juvenile
songs (compare Figs. 2.4B and 2.5B). Moreover, whereas OWN and REV-OWN frequently
evoked the strongest responses in juvenile RA (Fig. 2.4C), by adulthood, OWN elicited the
strongest response from most neurons while REV-OWN was never the best stimulus (Fig. 2.5C).
This preference for OWN was further manifested in measures of response strength: both multi-
and single-unit responses in adult RA were strongest to OWN playback compared to all other
stimuli (Fig. 2.5D; Kruskal-Wallis p < .001, Benjamini-Hochberg correction for multiple
comparisons, p < .001 in all cases). This result contrasts with juvenile RA, in which response
strength to OWN versus other juvenile song stimuli was comparable (juvenile data re-plotted in
Fig. 2.5D for comparison; mean juvenile values shown as blue plus signs). Adults also differed
from juveniles in that responses in adult RA were stronger to all forward songs than to REV-
OWN, indicating selectivity for temporal order (Fig. 2.5D, E; Kruskal-Wallis, Benjamini-
Hochberg corrected, p < .01 in all cases). Thus, in addition to an increased response to OWN
song, neurons in adult RA also developed a sensitivity to the temporal ordering of auditory-vocal
sounds by the end of sensorimotor learning.
To quantify the degree to which single neurons preferred one song stimulus over another, we
calculated selectivity for Song A over Song B as the difference in response strength between
Song A and Song B (see Materials and Methods). The magnitude of selectivity for OWN over
CON and REV-OWN was greater in adults than in juveniles (Fig. 2.5F; Mann-Whitney test,
JUV/ADL-CON p = .002, REV-OWN p < .001). This heightened selectivity was due primarily
to an increase in response strength to OWN in adult birds compared to juveniles (Fig. 2.5D,
right; p < .001); the response to REV-OWN decreased in adults while the response to conspecific
27
song did not change (Fig. 2.5D, right; Mann-Whitney test, REV-OWN p = .02, JUV/ADL-CON
p = .87). The selectivity for OWN over other stimuli in adult RA concurs with previous studies,
although to our knowledge no published studies have tested responsivity to OWN versus TUT
(Doupe and Konishi, 1991; Vicario and Yohay, 1993).
Response strength to TUT was marginally higher than that to ADL-CON in adult RA neurons (p
= .06), reflecting the fact that responses to TUT increased by adulthood whereas responses to
ADL-CON did not change (Fig. 2.5D, right). Given the substantial increase in selectivity for
OWN, we wondered whether this stronger response to TUT was due in part to increased
similarity of TUT to OWN. To test this idea, we calculated the acoustic similarity of tutor song
and adult conspecific song to self-generated song for each bird (Sound Analysis Pro 2011,
asymmetric mean values similarity; Tchernichovski et al. 2000). Figure 2.5G plots OWN-
similarity of each adult stimulus (TUT and ADL-CON) against the mean multi-unit response
evoked by that stimulus for each bird. In 9/10 birds for which both TUT and ADL-CON were
played, whichever stimulus had higher acoustic similarity to OWN elicited a greater response
(Fig. 2.5G, 9 of 10 lines with positive slopes). This effect was not specific to tutor song identity:
in 3/10 birds, ADL-CON was more similar to OWN and also evoked a stronger response (Fig.
2.5G, 3 of 4 gray lines). This pattern of results suggests that OWN-similar vocal sounds elicit a
greater response from adult RA neurons, regardless of whether they occur within the tutor song
or an unfamiliar conspecific song.
28
Figure 2.5. RA neurons develop selectivity for bird’s own song over other vocal stimuli by
adulthood. A, Proportion of adult RA sites (left) and single neurons (middle) that responded to
different numbers of stimuli. Multi-unit proportions are out of 28 sites to which all five stimuli
were played; single-unit, out of 34. Each row in chart at right indicates stimuli to which each
neuron responded. B, Proportion of single neurons that responded to each stimulus: 34/35
OWN, 26/34 TUT, 19/34 ADL-CON, 9/35 REV-OWN. Significance was tested using a χ
2
test.
C, Proportion of neurons that demonstrated the greatest response strength to each stimulus. D,
Mean standardized response strengths to each stimulus across multi-unit sites (left) and single
neurons (right). Gray plus signs mark means from adult RA sites and neurons; box plots depict
medians and first, third quartiles; circles represent individual data points. Blue plus signs mark
means from juvenile RA responses (re-plotted from Figure 4D for comparison). Significance for
multi-unit data set was tested using a one-way ANOVA; for single-unit data set, a Kruskal-
Wallis test. E, Response strength of juvenile (left) and adult (right) RA neurons during playback
of OWN versus REV-OWN. Black lines indicate mean response. F, Selectivity of OWN over
CON (JUV-CON and ADL-CON in juveniles and adults, respectively), REV-OWN, and TUT
(see Materials and Methods) in juveniles (blue) versus adult (gray) RA. Plus signs mark means;
box plots depict medians and first, third quartiles; dotted lines, ranges. Significance was tested
using a Mann-Whitney
test. G, Similarity of ADL-CON and TUT to OWN versus mean multi-
29
unit response strength evoked by each. Lines connect data points from individual birds. Red
lines highlight birds for which TUT was more similar to OWN; gray lines, ADL-CON was more
OWN-similar.
RA neurons develop a temporally patterned auditory response by adulthood
In addition to changes in song selectivity, we also tested whether temporal patterning of song-
evoked responses in single RA neurons differed between juveniles and adults. In juvenile birds,
the pattern of spiking responses to OWN playback varied across iterations (Fig. 2.6A, left). In
contrast, OWN-evoked responses in adult RA neurons consisted of phasic bursting interspersed
with periods of suppression (Dave et al., 1998), and this pattern was consistent across trials,
resulting in peaks in the PSTH that were well-aligned to time points throughout the song (Fig.
2.6A, right).
To quantify the degree to which song-evoked responses were temporally patterned, we made
pairwise comparisons of all spike trains for each cell and calculated the number of spikes that
occurred at the same time points across trials (coincidence window = 1 ms). This value was then
normalized to generate a correlation index (CI), in which a value of 1 indicates a complete lack
of temporal structure while larger values indicate a greater degree of coincident spike timing
(Louage et al. 2004; Joris et al. 2006; see Materials and Methods). These coincident time points
can be visualized as peaks in a normalized correlogram: Figure 2.6B shows examples of
correlograms generated from single-unit spiking responses to OWN playback in juvenile versus
adult RA (see Materials and Methods). The correlogram from a juvenile neuron (left) exhibited
a low correlation index, reflecting relatively low temporal structure. In contrast, the correlogram
from an adult RA neuron (right) contains a large central peak, reflecting a greater degree and
number of coincident spiking events. Across responses, average CI value did not differ among
stimuli in juvenile RA, whereas average CI value was greater for OWN responses than for ADL-
CON and REV-OWN in adult RA (Fig. 2.6C, D; Kruskal-Wallis p = .64 for juveniles, p = .001
for adults, Benjamini-Hochberg corrected, ADL-CON p = .01, REV-OWN p < .001 for adults).
To investigate the degree to which temporal patterning occurred beyond that expected by chance,
we generated 1,000 random shuffles of each cell’s spiking response to each song in order to
create null distributions of correlation indices for each response. We measured the proportion of
30
CI values from shuffled responses that were lower than that of the observed response and
expressed this measure as a percentile score (see Materials and Methods). Thus, a higher score
indicates a greater degree of specificity in spike timing. Individual song responses that had a CI
value above the 95
th
percentile when compared against their null distributions were judged as
having significant temporal patterning. In addition, we averaged the CI percentile scores across
all responses for each song in order to compare degree of temporal patterning across song
stimuli.
A minority of song-evoked responses in individual juvenile RA neurons demonstrated significant
temporal patterning: 13% (4/31) of OWN responses, 4% (1/27) of REV-OWN responses, 20%
(5/25) of JUV-CON responses, and 3% (1/31) of TUT responses (Fig. 2.6E) exhibited CI values
that fell above the 95
th
percentile; these proportions did not differ (χ
2
test, p = .09). However,
across the population of responses for each stimulus, average CI percentile scores did not exceed
.95 for any song (Fig. 2.6F). Thus, although some responses to OWN and JUV-CON contained
coincident spiking, song responses across juvenile RA neurons did not entail time-aligned
temporal structure.
In contrast, more than half the responses to each forward song stimulus in adults exhibited a
significant degree of temporal patterning (75% (25/34) of OWN responses, 53% (10/19) of
ADL-CON responses, and 73% (19/26) of TUT responses; Fig. 2.6E). Proportions of temporally
patterned responses to OWN and TUT increased between juvenile and adult RA (Fig. 2.6E;
Fisher’s exact test, p < .001 for both comparisons), and mean CI percentile scores for responses
to all forward songs were greater in adults than in juveniles (Fig. 2.6F; Mann-Whitney test, p <
.02 in all cases). Among adult RA neurons, only CI values for OWN responses exceeded the
95
th
percentile when averaged across cells (Fig. 2.6F), indicating that RA neurons exhibit both
greater response strength (Fig. 2.5C) and a higher degree of time-aligned coincident spiking
during OWN playback compared to other stimuli. Thus, RA neurons develop temporally
patterned responses to forward song stimuli by adulthood, with a particular increase in the degree
of coincident spiking during OWN-evoked responses.
31
Figure 2.6. RA neurons develop temporally patterned responses by adulthood. A, Extracellular
recordings in juvenile (left) and adult (right) RA during OWN playback. Top: OWN
spectrograms and raw traces of multi-unit activity during playback. Middle: Rasters of single
neurons sorted from multi-unit recordings above (overlaid waveforms shown in inset at top
right). Bottom: PSTHs. B, Normalized correlograms (see Materials and Methods) of example
responses during OWN playback in single units from juvenile (left) and adult (right) RA. C,
Mean CI value of neuronal responses to each song stimulus in juvenile RA. D, Mean CI value of
neuronal responses to each song stimulus in adult RA. Significance in C and D was tested with
Kruskal-Wallis tests. E, Proportion of neuronal responses in juveniles (blue) and adults (gray)
with CI values that fell above the 95
th
percentile when compared against null distributions (see
Materials and Methods). Significance was tested using a Fisher’s exact test. F, Mean CI
32
percentile score for neuronal responses in juveniles (blue) versus adults (gray). Dotted line
marks 95
th
percentile. Significance was tested using Mann-Whitney tests. *p<.05, **p<.01,
*p<.005.
Temporally patterned responses in adult RA include increased bursting and periods of
suppression
To further compare the temporal structure of song responses between age groups, we examined
the degree of bursting activity and suppression during song-evoked responses in juvenile versus
adult RA neurons. In accord with previous studies, we found that the average firing rate and
burst fraction (fraction of spikes with inter-spike intervals < 5 ms) during baseline was greater in
adults than in juveniles (Fig. 2.7A, B; Mann-Whitney test, p < .001 in both cases) (Adret and
Margoliash, 2002). Whereas burst fraction during playback did not differ among juvenile song
responses, burst fraction among adult responses mirrored differences in response strength:
responses during OWN playback demonstrated greater bursting compared to all other stimuli,
and responses to both ADL-CON and TUT demonstrated greater bursting than REV-OWN (Fig.
2.7C, D; Kruskal-Wallis p < .001, Benjamini-Hochberg corrected, p < .01 in all cases). Thus,
the developmental increase in the adult response to forward over reversed stimuli – and in
particular to OWN playback – is accompanied by an increase in bursting activity.
Although we did not observe any instances of suppressed responses to song playback in adult
RA neurons when measuring average firing rate across entire song durations, qualitative
inspection of evoked responses revealed clear periods of suppression (Fig. 2.6A, right). To
characterize these periods, we defined suppressed “events” as three or more contiguous 5-ms
PSTH bins in which the firing rate was at least two standard deviations below baseline (see
Materials and Methods; Billimoria 2006). Juvenile neurons showed no increase in suppressed
events between baseline and song playback for any stimulus, suggesting that song responses in
juvenile RA do not recruit local inhibitory circuits in any consistent fashion (Table 2.2). In
contrast, playback of OWN and TUT in adult RA neurons (but not other song stimuli) elicited
significant increases in suppressed events relative to baseline (Table 2.2; Wilcoxon signed-rank
test, OWN p = .018, TUT p = .01), and the average number of suppressed events during these
songs was greater than in juveniles (Fig. 2.7E; Mann Whitney test, p < .001 for both cases).
33
We used the same approach to calculate the incidence of excitatory events during song playback
(see Materials and Methods) and found that the results corresponded with our earlier measures of
temporal patterning: in juveniles, only OWN and JUV-CON responses contained more excitatory
events compared to baseline (Wilcoxon signed-rank test, OWN p = .008, JUVCON p = . 03); in
adults, responses to all forward song stimuli contained a significant increase in excitatory events
– particularly for responses to OWN (Table 2.2; Wilcoxon signed-rank test, p < .001 for all
songs).
These data expand on our description of temporal structure in song responses, revealing that a
minority of OWN and JUV-CON responses in juvenile birds contain temporally patterned
spiking but lack consistent periods of inhibition. By adulthood, song-evoked responses
demonstrate increased bursting activity, and responses to OWN and TUT in particular
demonstrate temporally patterned periods of excitation and suppression.
34
Figure 2.7. Single-neuron song responses in RA increase in bursting activity and suppressed
events by adulthood. A, B, Mean firing rate (A) and burst fraction (B) during baseline in
juveniles (blue) versus adults (gray). Plus signs mark means; box plots depict medians and first,
third quartiles; dotted lines, ranges. Significance was tested using Mann-Whitney tests. C, Mean
burst fraction during song responses in juveniles (blue) versus adults (adults). Significance was
tested using Kruskal-Wallis tests, Benjamini-Hochberg corrected. D, Burst fraction during
baseline versus OWN playback in juveniles (left) and adults (right). Black lines indicate mean
burst fraction. E, Frequency distribution of suppressed events during OWN (left) and TUT
(right) responses in juveniles (blue) versus adults (gray). *p<.05, **p<.01, *p<.005
Table 2.2: Average number of excited and suppressed events during song playback in
juvenile and adult RA
Excited Suppressed
Juvenile RA Adult RA Juvenile RA Adult RA
Baseline 0.00 0.04 0.47 1.94
OWN 0.26** 4.51*** 0.48 3.26**
REV 0.12 0.29 0.33 2.46
JUV-CON 0.25* 0.89
ADL-CON 0.13 1.18*** 0.53 2.00
TUT 0.04 2.03*** 0.70 2.82**
Comparisons against respective baseline:
* p < .05, ** p < .02 , *** p < .001
35
AId neurons of juvenile and adult birds do not respond to song playback
AId is a cortical region that adjoins the lateral border of RA and receives input from the SHELL
subregion of the cortical nucleus LMAN (lateral magnocellular nucleus of the anterior
nidopallium) (Johnson et al., 1995; Bottjer et al., 2000; Paterson and Bottjer, 2017). In juvenile
birds only, AId also shares an input with RA: neurons from the CORE subregion of LMAN
project to RA and send collateral arborizations into AId of juvenile birds, providing a transient
projection to AId that is no longer present in adult birds (Fig. 2.1; Miller-Sims & Bottjer 2012).
Both CORE and SHELL regions of LMAN contain song-responsive neurons that are likely to play
an important role in vocal learning (Doupe, 1997; Solis and Doupe, 1997; Achiro and Bottjer,
2013; Achiro et al., 2017). We therefore predicted that auditory vocal information represented in
CORE and SHELL would be reflected in AId. To investigate this idea, we made multi-unit
recordings in AId of anesthetized juvenile and adult birds during song playback (Fig. 2.1; Table
2.1).
Surprisingly, we found no evidence of auditory responsivity in AId neurons: none of the 19 sites
recorded in juvenile birds demonstrated a song-evoked response. Figure 2.8A shows a
representative site in a juvenile bird that showed no response during song playback. Firing rate
during song presentations did not differ from baseline for any stimulus across all juvenile
recording sites (Fig. 2.8B), such that response strength was low and did not differ among stimuli
(0.21 ± 0.05 across all songs, mean ± SEM; ANOVA, p = .90).
Similarly, 22/23 sites in AId of adult birds demonstrated no change in firing rate during song
playback (Fig. 2.8C), and one site exhibited response suppression. As with juveniles, firing rate
was comparable between stimulus and baseline for all songs (Fig. 2.8D), and response strength
did not differ among songs (.22 ± .04 across all songs; ANOVA, p = .71). Response strength did
not differ between juveniles and adults for any stimulus (data not shown; independent t-test,
unequal variances, p > .05 in all cases). These results demonstrate that AId neurons in
anesthetized juvenile and adult zebra finches are unresponsive to playback of auditory vocal
stimuli, despite receiving direct input from regions that contain song-responsive neurons.
36
Figure 2.8. Neither juvenile nor adult AId sites were responsive to song playback. A, Example
recording in juvenile AId. Top to bottom: OWN spectrogram, raw trace, raster, and PSTH of
multi-unit activity during OWN playback. B, Mean standardized firing rate during baseline
(light blue) and stimulus (dark blue) periods in juvenile AId. Circles represent individual data
points. C, Example recording as in A, showing multi-unit activity in adult AId during OWN
playback. D, Mean firing rates as in B, showing firing rates during baseline versus stimulus
periods in adult AId. Significance in B and D was tested using ANOVAs.
One explanation for this surprising lack of response to song playback in AId is that song-evoked
responses are masked by local inhibition. To test this hypothesis, we applied the GABA-A
receptor blocker gabazine in AId via micro-iontophoresis and measured the response to song
playback (n = 5 juvenile birds, 1 adult). During gabazine application, mean firing rates increased
across both baseline and stimulus presentation periods (Fig. 2.9A); on average, firing rates
increased from 66.34 ± 8.57 to 100.48 ± 12.03 (Fig. 2.9B). However, even with GABA-A
inhibition blocked by gabazine, response strength was not significantly different from zero for
any song stimulus (Fig. 2.9C; one sample t-test, p > .05 for all cases). While the overall increase
in firing rate during gabazine application indicates the presence of local GABAergic inhibition in
AId, this inhibition does not mask auditory responses to song playback in anesthetized birds.
37
These results contrast starkly with the robust song responses recorded in RA neurons; even
though individual LMAN-CORE neurons send collaterals to both RA and AId in juvenile birds
(Miller-Sims and Bottjer, 2012), song-evoked responses are seen only in RA and not AId,
indicating a distinct difference in post-synaptic processing.
Figure 2.9. Blocking inhibition in AId does not unmask song-evoked responses. A, Example
extracellular recording in juvenile AId before (left) and during (right) gabazine micro-
iontophoresis. Top to bottom: OWN spectrogram, raw traces, rasters, and PSTHs of multi-unit
activity. B, Mean firing rate during control period versus gabazine application. Black dots
represent individual juvenile birds; triangles, adult. C, Mean response strength to stimuli during
control period (light gray) versus gabazine application (dark gray), showing lack of responsivity.
One sample t-tests comparing control and gabazine periods were not significant.
38
RA-cup contains neurons that respond broadly to playback of auditory vocal stimuli
RA-cup receives inputs from different areas of auditory cortex, including superficial (L1) and
deep (L3) layers of primary auditory cortex as well as higher-level auditory areas CM (caudal
mesopallium) and HVC-SHELF (Kelley and Nottebohm, 1979; Vates et al., 1996; Mello et al.,
1998); this cortical region is defined by the terminal fields of these auditory projections as an
area surrounding RA, with particularly dense innervation extending rostral and rostro-ventral to
RA. RA-cup may receive auditory-vocal information important for learning from these inputs
and thereby contribute to song learning (Mello and Clayton, 1994; Mandelblat-Cerf et al., 2014).
As a first step in testing whether such information is represented in RA-cup, we made
extracellular recordings in RA-cup of anesthetized juvenile and adult birds during song playback
(Fig. 2.1; Table 2.1).
Across all recordings in RA-cup, 40% (21/53) of sites in juveniles and 63% (46/73) of sites in
adults demonstrated a significant change in firing rate during presentation of at least one
stimulus; Figure 2.10B plots the proportions of responsive sites by song type. Among these
song-responsive sites, 81% (17/21) in juveniles and 67% (31/46) in adults demonstrated response
excitation; 19% (4/21) in juveniles and 33% (15/46) in adults demonstrated response suppression
(Table 2.1). These proportions did not differ between age groups (Fisher’s exact test, p = .38).
Song-evoked responses in RA-cup of both juveniles and adults demonstrated a broad variety of
response patterns. Figure 2.10A shows example rasters and PSTHs of the spiking activity during
playback of OWN in a juvenile bird and of ADL-CON in two different sites in an adult bird.
Some responses consisted of a robust decrease or increase in firing rate throughout song
playback (left; middle) while others were temporally patterned with periods of both excitation
and suppression (right). In both juvenile and adult birds, absolute values of response strength did
not differ among song types (Fig. 2.10C; Kruskal-Wallis, p = .67 for juveniles, p = .08 for
adults). Moreover, response strength did not differ between juvenile and adult birds for any song
(Fig. 2.10C; Mann-Whitney test, p > .05 in all cases). Thus, in contrast to the patterns of neural
responsivity seen in RA, juvenile and adult birds showed no developmental differences; many
sites in RA-cup were unresponsive, and the sites that did respond to playback evinced little or no
song selectivity at either of the ages studied here.
39
Due to the presence of both excitation and suppression that resulted in temporally patterned
responses at several recording sites (for example, Fig. 2.10A right), many of these responses did
not achieve a significant change in mean firing rate across the entire song duration and therefore
were not considered evoked responses. As an alternate measure of whether songs evoked a
response, we used the percentile score of each multi-unit response’s correlation index value
relative to a null distribution to investigate time-aligned responses (.95 percentile significance
threshold; see above, and Materials and Methods).
Figure 2.10D plots the distribution of CI percentile scores across all songs for all responses
recorded in juvenile (top) and adult (bottom) RA-cup, binned in increments of 0.05. The open
and filled portions of each bar indicate the proportions of responses that were significant and
non-significant, respectively, based on average firing rate measures. Thus, the filled portions of
the bars at the 95
th
percentile bins (starred bars at far right) represent responses that were non-
significant based on mean response strength but exhibited a significant degree of temporal
structure (11% (22/201) non-significant responses in juveniles and 24% (47/193) in adults with
CI percentile score > .95), indicating the presence of temporally patterned responses to song
playback that were not captured by average firing rate measures. Figure 2.10E plots
instantaneous firing rates of responses in adult RA-cup during OWN, REV-OWN, and ADL-
CON playback that demonstrated either a significant change in mean firing rate (net excitation or
net suppression; marked with green and red sidebars, respectively) or had a CI value that fell
above the 95
th
percentile (gray sidebar), illustrating the variety of response patterns observed in
these recordings – playback of each stimulus exhibited a mix of sustained increases or decreases
in firing rate and phasic activity across sites.
Although this analysis of temporally patterned activity revealed additional song-evoked
responses, a remaining 74% (179/242) of RA-cup responses in juveniles and 51% (146/288) of
responses in adults demonstrated neither a significant change in firing rate nor a high degree of
temporal patterning during song playback (Fig. 2.10D). The incidence of non-responsive sites is
surprisingly high given the multiple direct inputs to RA-cup from auditory cortical areas.
Among evoked responses, sites in both juvenile and adult RA-cup responded to a range of vocal
stimuli with a variety of response patterns; neither age group contained sites that were selectively
tuned to any particular stimulus, and there were no significant differences in responsivity
40
between the two age groups. This pattern contrasts with that seen in RA, where neurons become
selectively tuned to the bird’s own song during learning.
Figure 2.10. RA-cup neurons respond to song playback with a variety of response patterns.
A, Rasters (top) and PSTHs (bottom) of example multi-unit responses to OWN in juvenile RA-
cup (left) and ADL-CON in adult RA-cup (middle and right, from two sites within the same
41
bird). B, Proportion of neurons that responded to each stimulus in juveniles (blue) and adults
(gray). Significance between juveniles and adults was tested using Fisher’s exact tests. C, Mean
response strength to song stimuli in juveniles (blue) and adults (gray). Significance of responses
within age groups was tested with Kruskal-Wallis tests; between age groups, Mann-Whitney
tests. D, Histograms of CI percentile scores (see Materials and Methods) in bins of 0.05
increments from 0.50 to 1.0 in juveniles (top) and adults (bottom). For each bar, open portion
represents responses that were significant and filled portion represents responses that were not
significant, as measured by average firing rates. Starred bars on the far right represent CI
percentile scores that exceeded the 95
th
percentile of null distributions, indicating non-significant
(filled) responses in these bins that nevertheless demonstrated significant temporal patterning. E,
Heat maps of evoked responses to OWN, REV-OWN, and ADL-CON in adult RA-cup, as
measured by significant change in firing rate (excitation responses, green bars; suppression, red
bars) or CI value falling above the 95
th
percentile (gray bars). Due to variability in firing rate
across RA-cup sites (e.g., A), scale of color gradient for each row of each map was based on the
range of instantaneous firing rates for that response. All song stimuli were 1-2 seconds long.
42
DISCUSSION
RA, AId, and RA-cup are each well-positioned to integrate auditory-vocal information that is
important for song learning. We investigated the representation of behaviorally relevant auditory
vocal sounds in each of these regions in juvenile and adult zebra finches in order to further our
understanding of how neural tuning changes to support goal-directed refinement of variable
behavior into a stereotyped motor skill. We found that only RA neurons showed strong
developmental changes: neurons in juvenile birds were broadly tuned but preferred immature
vocal sounds over adult song; by adulthood, RA neurons developed temporally patterned
responses that were highly selective for each bird’s own song. In contrast, neural activity in RA-
cup did not show developmental changes – among sites showing song-evoked activity, song
selectivity was low and temporal patterns of responses were heterogeneous in both juveniles and
adults. AId neurons differed from both of these regions: surprisingly, neither juvenile nor adult
AId neurons responded to song playback, despite receiving afferent inputs from song-responsive
LMAN neurons (Doupe, 1997; Solis and Doupe, 1997; Iyengar et al., 1999; Miller-Sims and
Bottjer, 2012; Achiro and Bottjer, 2013).
Results from playback-evoked activity in RA address fundamental principles of motor
learning
Motor learning entails establishing associations between motor output and sensory feedback
Associating motor output and resultant sensory feedback is a fundamental component of motor
skill learning during development. These associations can develop within single cells that
demonstrate both motor and sensory properties, such as RA neurons: single RA neurons both
drive motor output and respond to playback of self-produced song, suggesting that RA neurons
not only serve a premotor role for vocal behavior but also are capable of encoding auditory
feedback of self-generated behavior (Nottebohm et al., 1976; Doupe and Konishi, 1991; Vicario
and Yohay, 1993; Wild, 1993; Dave et al., 1998; Kojima and Doupe, 2007) (Figs. 2.4 & 2.5).
The activity patterns during singing in adult RA are well-matched to those evoked by playback
of self-generated song in sleeping birds, reflecting a sensory-motor correspondence that has
developed by the end of vocal learning (Dave and Margoliash, 2000). The results from juvenile
RA presented here are consistent with the idea that linkages between sensory and motor
representations develop during learning: when the juvenile is producing highly variable motor
output, auditory responsivity is correspondingly broad while still remaining within limits of
43
vocal sounds that the juvenile is most likely to produce – i.e., responding best to a range of
immature vocal sounds and not to stereotyped adult syllables. Likewise, studies in humans have
shown strong links between perception and production in the neural representation of vocal
sounds as well (Liberman and Mattingly, 1985; Wilson et al., 2004; Pulvermüller et al., 2006;
D’Ausilio et al., 2009; Devlin and Aydelott, 2009).
Establishing sensory-motor associations in early learning requires accounting for the fact that
sensory feedback of a motor behavior is delayed relative to the premotor commands for its
output. In both mammals and birds, forward models of learning have suggested corollary
discharge as a solution: copies of premotor commands could be routed and delayed through
upstream circuitry in order to coincide with the arrival of activity evoked by sensory feedback of
the behavior (Rauschecker and Scott, 2009; Hickok et al., 2011; Prather, 2013). The recurrent
loops that comprise the song system provide ample opportunities for routing signals in this
manner – for instance, efference copy of premotor commands from LMAN-CORE that drive RA
activity could be routed through cortico-basal ganglia circuits, such that the timing of vocal
motor output is correctly associated with corresponding auditory feedback (Fig. 2.1).
Behavioral and neural variability facilitates motor exploration during early learning
RA neurons receive direct inputs from song-responsive cells in LMAN-CORE and HVC (Fig. 2.1;
Herrmann & Arnold 1991; Mooney & Konishi 1991; Mooney 1992; Stark & Perkel 1999).
LMAN-CORE is the primary driver of RA activity in 45-day-old juvenile birds (Bottjer et al.
1984; Aronov et al. 2008; Ölveczky et al. 2005; Ölveczky et al. 2011), and OWN-responsive
neurons in juvenile LMAN respond more strongly to playback of self-generated song than to
other song stimuli (Achiro and Bottjer, 2013). Surprisingly, although juvenile RA neurons drive
vocal output and receive these song-selective inputs, we found that RA neurons themselves lack
selectivity for self-generated song at this early stage of sensorimotor learning: playback of all
juvenile songs – including each bird’s own song – elicited comparable response strength in
juvenile RA. This lack of selectivity may reflect broadly tuned HVC inputs. Although
selectivity for each bird’s own song over tutor song has been reported in HVC of sleeping
juvenile birds (Volman, 1993; Nick and Konishi, 2005), no published studies have compared the
OWN-evoked responses of juvenile HVC neurons against playback of other conspecific juvenile
44
sounds. One possibility is that HVC neurons respond broadly to a variety of juvenile songs early
in sensorimotor learning, and this pattern of tuning is conveyed to RA neurons. If sensory and
motor representations are linked at the level of single RA neurons, then the lack of auditory
selectivity among juvenile songs may also reflect a broad motor representation of immature
vocal sounds. As posited by motor learning theory, a motor system in early development that is
capable of producing a wide range of outputs would confer behavioral variability that is
advantageous for facilitating exploration of motor space early in learning – in this case, a broad
motor representation of juvenile vocal sounds in RA would allow young birds to experiment with
a wide variety of vocalizations as they attempt to match goal vocal patterns (Sutton and Barto,
1998; Wu et al., 2014).
In addition to a lack of selectivity, song-evoked responses in juvenile RA neurons were also
characterized by temporal variability across playback iterations. Neuronal spiking activity in RA
of singing juvenile birds (50 dph) is variable across renditions of motifs that have recognizable
syllable sequences but are still spectrally variable (Ölveczky et al., 2011). While the lack of
temporal patterning in singing juvenile birds may thus reflect this spectral variability, our results
indicate that even when the exact same motif is presented via playback, the evoked pattern of
firing is variable across iterations. In singing adult birds, ensembles of RA neurons consistently
drive production of specific sub-syllabic elements across song renditions (Leonardo and Fee,
2005; Sober et al., 2008). The variable response pattern during song playback in juveniles may
indicate a degenerate encoding strategy early in sensorimotor learning, such that single RA
neurons participate in multiple ensembles; likewise, a particular time point in the song may be
produced by different RA ensembles across renditions, such that the song-evoked response
pattern differs across iterations. Similar degeneracy has been observed during early motor
learning in mammalian systems (Peters et al., 2014). This instability may not merely reflect an
immature network with developing connections but rather actively facilitate learning: as
proposed by reinforcement learning theory, increased spiking variability would allow the system
to explore a variety of activity patterns early in goal-directed learning to help determine the
optimal pattern for producing a given sound (Sutton and Barto, 1998; Kargo and Nitz, 2004;
Leonardo, 2005; Makino et al., 2016).
45
Coordinated refinement of excitatory and inhibitory networks may contribute to motor learning
In contrast to the variable spiking activity in juvenile responses, adult RA neurons exhibited
time-aligned periods of both excitation and suppression during song playback. In mammalian
sensorimotor systems, progression from variable to stereotyped behavioral and neural activity
patterns corresponds with an increased number of synapses in motor cortex that are subsequently
pruned, suggesting that synaptic-level changes are an important mechanism by which ensemble
activity is tuned during learning (Xu et al., 2009; Yang et al., 2009; Wang et al., 2011; Fu et al.,
2012). Synaptic inputs to RA are similarly dynamic during the sensorimotor learning period:
HVC innervation of RA neurons increases around 35 dph, and HVC inputs onto RA neurons
increase in both strength and number between 40-60 dph before being pruned (Herrmann and
Arnold, 1991; Foster and Bottjer, 1998; Roberts et al., 2010; Yip et al., 2012; Garst-Orozco et
al., 2014). Thus, early song development provides an expanded neural substrate that may
underlie variable representations of auditory-vocal stimuli and thereby facilitate early
sensorimotor learning. Importantly, experience-dependent changes in inhibitory circuits are
likely to play a role in shaping learned representations as well. For instance, song tutoring has
been associated with targeted pruning of local inhibitory synapses in RA of Bengalese finches
(Miller et al. 2017); similarly, evidence from rodents implicates coordinated inhibitory neuron
activity in the formation of stable representations of learned motor behaviors (Chen et al., 2015).
Thus, while developmentally regulated increases in synapse number would allow for an initially
expanded representation of variable vocal behavior, skill learning may also entail experience-
directed refinement of both excitatory and inhibitory inputs within RA in order to achieve the
temporally patterned responses evoked by the learned song in adults (Spiro et al., 1999; Vicario
and Raksin, 2000; Caroni et al., 2012).
AId and RA-cup are well-situated to serve as regions of sensorimotor integration that may
contribute to various aspects of vocal learning
AId of juvenile birds could process efference copy of active singing behavior
The lack of auditory responsivity in juvenile AId is surprising considering the fact that AId
receives direct inputs from song-responsive neurons in both CORE and SHELL regions of LMAN
(Bottjer et al., 2000; Miller-Sims and Bottjer, 2012; Achiro and Bottjer, 2013). Moreover,
blocking local GABA-A inhibition failed to unmask any song-evoked responses. One possible
46
explanation for the lack of auditory responsivity in juvenile AId is that motor-related inputs from
LMAN-CORE gate auditory responses: CORE→RA neurons drive vocal output in 45 dph
juveniles, and the collateral projection from these neurons into AId could convey efference copy
of vocal motor commands in singing birds (Fig. 2.1; Bottjer et al. 1984; Aronov et al. 2008;
Ölveczky et al. 2005; Ölveczky et al. 2011; Miller-Sims & Bottjer 2012). If AId contributes to
tutor song comparisons during vocal learning, gating of auditory-vocal activity in AId by
efference copy would ensure that evaluative comparisons are made only against self-produced
vocalizations (rather than external auditory inputs, including similar immature vocal utterances
by siblings). Recognition of sensory input as feedback of self-generated behavior is a
fundamental component of motor learning and execution across systems, and efference copy of
self-produced motor commands has been proposed as a mechanism for accomplishing this
objective (Crapse and Sommer, 2008; Houde and Chang, 2015; Roberts et al., 2017; Straka et al.,
2018). For example, studies in electric fish have shown that corollary discharge signals from
motor commands activating the electric organ help distinguish reafference of self-generated
electric discharges from external input (Bell, 1981; Sawtell, 2017). Similarly, studies in humans
have suggested “self-referent” neural activity that allows for identification of self-produced
sounds, enabling evaluative mechanisms for pitch control to be engaged (Behroozmand and
Larson, 2011; Behroozmand et al., 2016). Efference copy signaling self-generated behavior to
AId neurons would fulfill this crucial role and explain the surprising lack of song-evoked
responses reported here. Moreover, the projection from LMAN-CORE to AId is present only in
juvenile birds, consistent with an important role specifically during sensorimotor learning
(Miller-Sims and Bottjer, 2012).
Song responsivity in RA-cup indicates a heterogeneous population that may contribute to
general auditory-vocal processing in both juveniles and adults
Recordings from RA-cup demonstrate for the first time that neurons in this region respond
broadly to different song stimuli in both juveniles and adults. In contrast to RA, the tuning of
song-responsive sites in RA-cup did not differ between juveniles and adults. The lack of
developmental changes in selective tuning suggests that neurons in RA-cup may participate in
general processing of complex auditory information. Surprisingly, we also found many sites in
RA-cup that showed no response to the songs we presented, despite receiving multiple inputs
47
from auditory cortex – including primary auditory areas L1 and L3, and higher-level auditory
areas CM and the shelf region below HVC (Fig. 2.1; Kelley & Nottebohm 1979; Vates et al.
1996; Mello et al. 1998). Primary versus secondary auditory cortical inputs to RA-cup
demonstrate varying degrees of song selectivity (Theunissen et al., 2004; Keller and Hahnloser,
2009; Calabrese and Woolley, 2015), and these projections terminate in different patterns around
RA; responsivity and selectivity of RA-cup neurons may differ depending on whether they
receive input from all of these afferents, versus only a subset. This heterogeneity is reminiscent
of diverse response properties in deep layers of mammalian cortex and may reflect the existence
of neuronal subpopulations that are differentially involved in auditory vocal processing.
Alternatively, the high degree of convergence in RA-cup may result in neurons that are
extremely selective and prefer other complex acoustic stimuli that were not tested in this study.
Moreover, the portion of RA-cup that lies rostral and rostro-ventral to RA overlaps with part of
the ventral intermediate arcopallium (AIV), which contains neurons that are significantly
modulated during singing; thus, a subpopulation of sites recorded in RA-cup here may be
involved in auditory-vocal processing during song production in awake birds (Mandelblat-Cerf
et al. 2014).
AId and RA-cup are well-suited to integrate information across multiple cortico-thalamo-
cortical circuits
Akin to infragranular layers of mammalian cortex, AId and RA-cup each receive inputs from
multiple cortical regions and send projections to thalamic regions. AId receives input from both
LMAN and the dorsal caudolateral nidopallium (dNCL) (Bottjer et al., 2000); dNCL receives
visual, auditory, and somatosensory information and has been proposed as a region important for
integrating multimodal information during vocal learning across avian species (Leutgeb et al.,
1996; Gunturken, 1997; Metzger et al., 1998; Braun et al., 1999; Bottjer et al., 2010; Paterson
and Bottjer, 2017). AId neurons participate in multiple recurrent circuits via projections to two
higher-order thalamic nuclei, DLM (dorsolateral nucleus of the medial thalamus) and DMP
(dorsomedial nucleus of the posterior thalamus); DLM projects to LMAN-SHELL to form
recurrent feedback connections, and DMP projects indirectly to HVC (via MMAN, medial
magnocellular nucleus of anterior nidopallium) to form feed-forward connections. Thus, AId
projections give rise to both feedback loops (AId → DLM → LMAN-SHELL → AId) and
feedforward pathways (AId → DMP → MMAN → HVC → RA) that create several
48
opportunities for information transfer between both cortical and subcortical regions (Wild et al.,
1993; Foster et al., 1997; Vates et al., 1997; Bottjer et al., 2000). Similarly, efferent projections
from RA-cup include areas of midbrain and auditory thalamic nuclei – including regions that
provide input to L1 and L3 (Vates et al., 1996; Mello et al., 1998). Thus, much like deep layer
regions of mammalian auditory cortex that give rise to corticothalamic feedback pathways, RA-
cup is well-suited to integrate information across primary and secondary auditory cortical areas
and feed information back to modulate ascending auditory pathway inputs (Guillery and
Sherman, 2002; Briggs and Usrey, 2008). In addition, both AId and regions of RA-cup project
to the area of dopaminergic neurons in VTA that project into the song-control region of avian
basal ganglia, Area X (Fig. 2.1; Bottjer et al. 2000; Gale et al. 2008; Mandelblat-Cerf et al.
2014). These projections provide a means by which integrated auditory vocal information in AId
and RA-cup could integrate and modulate dopaminergic inputs to multiple brain regions and
thereby indirectly affect vocal learning.
49
Chapter 3: Motor cortical representations of active behavior in juvenile zebra
finches during sensorimotor learning
INTRODUCTION
Goal-directed motor skill learning requires iterative comparisons between feedback of variable
self-generated motor attempts and internal representations of a target motor pattern. While a key
function of motor cortex is the control of motor effectors for voluntary movement, increasing
evidence suggests that this region serves as a central locus of motor skill learning as well. For
instance, studies have demonstrated experience-dependent changes of motor cortical
representations in response to both skill learning and disease or trauma, including expansions of
motor maps which have been suggested to reflect functional reorganization that can support
newly learned movements (Sanes and Donoghue, 2000; Makino et al., 2016; Peters et al., 2017b;
Papale and Hooks, 2018). Similarly, behavioral progression from variable to stereotyped neural
activity patterns during learning corresponds with an increase in the number of motor cortical
neurons that are associated with the motor skill, followed by refinement of this population into a
stereotyped ensemble that becomes reliably associated with the learned behavior (Peters et al.,
2014).
Motor skill learning is highly sensorimotor: utilizing feedback to guide accurate refinement of
motor output requires sensorimotor integration of both internal and external sources of sensory
input, and feedback information would also be required to flexibly perform an acquired skill in
response to changing environmental demands post-learning. Correspondingly, motor cortical
representations during learning have been shown to demonstrate multi-dimensional tuning that
suggests integration across a variety of sensorimotor aspects: motor cortical neurons involved in
performance of a skill have been shown to encode not only motor parameters (e.g., movement
speed or direction) but also seemingly non-motor parameters, such as preparatory activity prior
to movement execution or visual information specific to a target object’s location in object-
directed movement tasks (Tanji and Evarts, 1976; Evarts and Fromm, 1977; Murata et al., 1997;
Shen and Alexander, 1997). Taken together, these findings suggest motor cortex as a dynamic
substrate that actively participates in sensorimotor learning.
AId is a region that has been compared to avian motor cortex and has been implicated in vocal
learning: lesions of AId in juvenile zebra finches impair the bird’s ability to achieve an accurate
50
imitation of its adult tutor song without inducing vocal motor deficits (Bottjer and Altenau,
2010). Akin to infragranular layers of motor cortex in mammals, AId receives input from
regions that process multi-modal sensory input as well as from cortico-basal ganglia circuitry
that is dedicated to vocal learning, and in turn makes a variety of descending projections that
include midbrain and hindbrain regions that contain premotor circuits as well as limbic thalamic
regions (Fig. 3.1A). AId thus serves as a powerful model for investigating motor cortical
contributions to goal-directed acquisition of complex motor skills.
To investigate the role of AId neurons in motor skill learning, we recorded from AId in awake,
behaving juvenile birds during the sensorimotor learning period. In addition to recording AId
neurons during singing, we analyzed activity during several discrete movements and investigated
patterns of activity across different behavioral state periods based on the bird’s behavior. Our
results thus represent an extensive assessment of avian motor cortical activity across a wide
variety of behaviors, thereby informing our understanding of how these neurons may contribute
to motor skill learning and production.
MATERIALS AND METHODS
Subjects
All animal procedures were performed in accordance with protocols approved by the University
of Southern California Animal Care and Use committee. Seven male juvenile zebra finches (43-
58 days post hatch, dph; mean age 46 dph on first day of recording) were used. All birds were
raised in group aviaries until at least 33 dph, remaining with their natural parents and thereby
receiving normal auditory and social experience during the tutor memorization period
(Immelmann, 1969; Böhner, 1983; Eales, 1985; Mann and Slater, 1995; Roper and Zann, 2006;
Catchpole and Slater, 2008). Juveniles were separated from group aviaries at 33-35 dph and
housed in single cages within the experimental rig at the start of experiments. Each bird’s tutor
was placed in a separate cage within view of the juvenile to help it acclimate to the experimental
rig for 2-5 days prior to the start of recording.
Electrophysiology
At 35-40 dph, birds were anesthetized with 1.5% isoflurane (inhalation) and placed in a
stereotaxic instrument. An electrode assembly consisting of four tetrodes affixed to a movable
51
microdrive was fixed to the skull using Metabond, such that the tetrodes were implanted 500 μm
dorsal to AId. Each tetrode consisted of four twisted polyimide-coated Nichrome wires (0.012
mm diameter Redi Ohm 800, Kanthal) routed through fused silica capillary tubing and
electroplated with non-cyanide gold plating solution (SIFCO 5355). One day after surgery, the
tetrode assembly was connected to a recording headstage (HS-16, Neuralynx) with a flexible
cable connected to a commutator (PSR, Neuralynx); 15 channels of neural data were amplified,
band passed between 300 and 5000 Hz (Lynx-8, Neuralynx), and digitized at 32 kHz using
Spike2 software (Power 1401 data acquisition interface, CED). Audio and video were recorded
coincident with neural activity – vocalizations were recorded with a lavalier microphone (Sanken
COS-11D) mounted in the cage; a USB-video camera (30 FPS, ELP) was placed at the front of
the cage to record video. Consecutive 30-minute recordings were made from 7:00 AM to 6:00
PM each day. Tetrodes were manually advanced with the microdrive when the cells being
recorded were lost or had already been recorded for at least two days, as indicated by stability
and consistency of the extracellular signal. At the end of each experiment, birds were perfused
(0.7% saline followed by 10% formalin), and brains were removed and post-fixed for 72 hours
before being cryo-protected (30% sucrose solution) and frozen-sectioned (50 μm thick). Sections
were Nissl-stained with thionin to visualize tetrode tracks and verify recording locations.
Movement artifact in neural recordings was correlated across recording channels and was
reduced using offline common average referencing: for each recording channel, the signal across
the remaining 14 recording channels was averaged and subtracted from that channel in order to
remove movement artifact (Ludwig et al., 2009). Channels were visually inspected after
referencing to ensure that spiking activity was not distorted. After common average reference
subtraction, single units were sorted from multi-unit data by first automatically clustering units
with KlustaKwik (KD Harris, University College London). KlustaKwik clusters were manually
inspected across 18 different waveform features and further refined using MClust (AD Redish,
University of Minnesota). Clusters were considered for analysis only if < 1% of spikes had an
inter-spike interval < 2ms.
Behavioral scoring
Videos from each 30-minute recording session were scored for movements and state periods
using ELAN (The Language Archive, Max Planck Institute for Psycholinguistics). We scored
52
each single occurrence of pecks, hops, preening behavior, beak interactions (beak wipes and
periods when the bird’s beak was in contact with perches, food cup edges, etc for longer than the
duration of a peck), fluff-ups, scratches, and stretches. Head movements occurred so frequently
that it was impractical to score all of them in all cells. However, we scored head movements that
occurred during singing and during 30 seconds of non-singing before and after each singing
episode in a subset of singing-responsive neurons (n = 41). These head movements were scored
in order to compare the response of neurons during head movements that occurred during singing
versus non-singing episodes. For 12 neurons across two recording sessions, all head and postural
movements across the entire session were also scored.
In addition to scoring discrete movements, we developed a novel way of measuring behavior
throughout recording sessions: each session was segmented into contiguous time periods that
were classified into one of five behavioral “state” periods based on the bird’s behavior: eating,
singing, active-movement, quiet-attentive, or quiescent; these state periods tiled the entire
duration of the recording session (Fig. 3.6A). Eating states were defined as periods during which
the bird was pecking at seed or grit, hulling or ingesting seeds, or pausing in between these
behaviors for at most one second. Although pecks occurred most often during eating, eating
states could also include other scored movements such as hops or preening, or unscored
movements such as head movements provided they occurred during the brief (< 1 sec) pauses
that occurred while birds were actively eating. Singing states were defined by song behavior;
they began whenever the bird produced song and lasted as long as song syllables continued to
occur within one second of each other (inter-syllable interval < 1 sec). Birds often made head
movements during singing and occasionally made scored movements such as hopping or pecking
in between bouts of singing. Active-movement states were defined as non-eating and non-
singing periods during which the bird was making active movements with pauses of at most one
second in-between movements; these periods could include any of the seven movements that
were scored as well as head and postural body movements that were not scored. Quiet-attentive
states were defined by times when the bird was not eating, singing, or moving around the cage
for more than one second; they continued as long as the bird made at most small head
movements and otherwise remained unmoving but alert. Quiescent periods were defined as
periods during which the bird was completely still and not obviously paying attention to any
53
stimulus, with eyes partially or fully closed. Quiescent state times were segmented into one-
second intervals that were used as baseline periods for analyses of scored movements (see
below).
Data Analysis
To test for significant responses during movements, firing rates across occurrences of each
movement type were compared against quiescence. Quiescent baseline periods were generated
by dividing quiescent state periods (as described above) into one-second segments. The firing
rate during two one-second quiescent segments that occurred closest in time to each movement
occurrence was used as a corresponding baseline. Fourteen neurons were recorded during
sessions that lacked quiescent state periods. For these 14 neurons, one-second baseline periods
were taken from times within quiet-attentive state periods when the bird was verified to be
unmoving (though clearly alert, unlike in quiescent state). To compare movement responses
across neurons, standardized response strength was calculated for each movement type as:
𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑𝑖 𝑧𝑒𝑑 𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒 𝑠𝑡𝑟𝑒𝑛𝑔𝑡 ℎ =
𝐹𝑅
̅ ̅ ̅ ̅
𝑚 − 𝐹𝑅
̅ ̅ ̅ ̅
𝑏 √𝑉𝑎𝑟 (𝐹𝑅
𝑚 ) + 𝑉𝑎𝑟 (𝐹𝑅
𝑏 ) − 2 × 𝐶𝑜𝑣𝑎𝑟 (𝐹𝑅
𝑚 , 𝐹𝑅
𝑏 )
where FRm is the firing rate during movement occurrences and FRb is the firing rate during
corresponding quiescent baseline periods. A positive value indicates an increase in firing rate
during the movement compared to quiescence whereas a negative value indicates a decrease in
firing rate during movement. This measure is referred to as response strength throughout the
text. To compare firing rates during states across neurons, we normalized the average firing rate
of each neuron during each state type by the standard deviation of that neuron’s firing rate across
the recording session.
To test for changes in activity around movement onsets, for each neuron we generated a 25 ms-
bin histogram of the spiking response across all occurrences of the movement; histogram
windows were one second long and centered on movement onsets. Spike times during each
movement repetition were shuffled to obtain a resulting histogram of shuffled spike data; this
was repeated 1000 times, resulting in 1000 histograms of shuffled data. Each bin of the real
spike data histogram was considered significantly excited if the count in that bin was greater than
54
95% of maximum values from the shuffled data set; likewise, the bin was considered
significantly suppressed if the count was lower than 95% of minimum values from the shuffled
data set. Onset responses were defined as responses that contained at least two consecutive bins
(50 ms) of significant maxima or minima within 100 ms of movement onset.
We tested for significant offset-aligned responses using the same method and parameters as
onset-aligned responses, except that the one-second windows used for histograms of real and
shuffled data were centered around movement offsets; offset responses were defined as
responses that contained at least two consecutive bins (50 ms) of significant maxima or minima
within 100 ms of movement offset. Due to the short duration of pecks and hops (mean hop
duration = .28 seconds; mean peck duration = .25 seconds), it was possible for the same maxima
or minima to be captured in both onset- and offset-aligned responses. To ensure that offset-
aligned activity could be accurately distinguished, for these movements we only counted excited
or suppressed offset responses that did not demonstrate significant changes in onset-aligned
activity. Preening movements were relatively long in duration (mean duration = 1.5 seconds), so
all preening-aligned offset responses were counted.
We defined “events” – brief periods of excitation and suppression – from histograms of spiking
activity. We segmented the instantaneous firing rate (IFR) across each recording session into 10
ms bins and smoothed the IFR with a moving average filter (span = 3 bins). Excited events were
defined as periods during which the smoothed IFR across five or more 10-ms bins (50 ms or
more) exceeded the average firing rate across quiescent state periods by > 3 standard deviations.
Suppressed events were defined as periods during which the smoothed IFR across five or more
50-ms bins fell below the average firing rate across quiescent state periods by < 1.5 standard
deviations. To compare across neurons, the number of events in each state was normalized by
dividing the number of excited or suppressed events in each state by the total duration of that
state for each neuron.
Statistics
Movement responses were tested for significance against quiescent baselines (see above) using
Wilcoxon signed rank tests; Benjamini-Hochberg post-hoc tests were used to apply a correction
for multiple comparisons. Neurons that demonstrated a significant difference between
55
movement and baseline for at least one movement were considered movement-responsive. To
test whether movement responses were context-selective, standardized response strengths during
movements from one context versus another were compared using Mann-Whitney tests for each
neuron (for example, comparing pecks during eating versus non-eating periods, and comparing
head movements during singing versus non-singing periods). χ
2
tests were run to compare
proportions between more than two groups (for example, proportions of neurons that were
responsive during each movement type). In case of significance, Fisher’s exact tests were used
as a follow-up to compare proportions between two groups, and Benjamini-Hochberg post-hoc
tests were used to apply a correction for multiple comparisons. Binomial tests were used to
judge whether the relative proportions of excited versus suppressed responses among movement-
responsive neurons were different from chance. Measures between different state periods
(average firing rate, normalized number of events, inter-spike intervals) were compared using
sets of pairwise linear contrasts based on trimmed means (20% trimming); this linear contrast
method has been shown to be robust to common assumption violations such as non-normality
and heteroscedasticity (Wilcox and Serang, 2017).
RESULTS
We made extracellular recordings of 119 neurons in AId of awake, behaving juvenile zebra
finches (43-58 dph) housed singly in a recording cage as they actively engaged in sensorimotor
vocal practice. A typical 30-minute recording period included various overt behaviors and
periods of quiescence when the bird was not moving. To investigate how neural activity in AId
corresponds to different behaviors, we scored seven different movements during each recording
that could be reliably identified: pecks, hops, preening episodes, beak interactions with objects in
the recording cage (e.g., beak wipes or non-peck interactions with cage bars or the food cup),
fluff-ups, stretches, and scratching episodes; we also marked periods of singing. We developed a
novel approach in which we examined spiking patterns of single neurons throughout each
recording period to investigate whether the activity of AId neurons is selective for different
movement types and/or singing behavior in juvenile birds.
56
Responsivity of AId neurons during movements
Patterns of spiking were highly variable across individual neurons, ranging from phasic to tonic
activity. In addition, each neuron’s activity was highly modulated throughout a typical recording
session, showing either excitation and/or suppression during different movements. Figure 3.1B
illustrates two different neurons recorded in one bird while it was quiescent (no overt
movements, left columns) and while it was hopping around the cage (right columns). The
neuron in the top panel fires intermittently in small bursts of at most three spikes during
quiescence, while the neuron in the bottom panel exhibits dense bursting activity. As the bird is
hopping around the cage, the neuron in the top panel shifts to longer periods of high firing
separated by relative inactivity while the neuron in the bottom panel shifts to a high tonic rate of
firing. To investigate whether such modulations were related to specific movements, for each
neuron we compared firing rates during different movement types against firing rates during
quiescent periods that were closest in time to iterations of each movement type (see Materials
and Methods). To compare movement-related activity across neurons, we also calculated the
response strength of each neuron during each movement type, defined as the standardized
difference in average firing rate during each movement type versus baseline periods (see
Materials and Methods).
57
Figure 3.1. AId neurons are well situated to integrate multi-modal inputs and distribute
information across various cortical-subcortical circuits. A, AId is a motor cortical region that
receives input from LMAN-SHELL and dNCL. LMAN-SHELL is part of a cortico-basal ganglia
loop that mediates vocal learning, whereas dNCL receives inputs from LMAN-SHELL as well as
multiple pathways processing somatosensory, visual, and auditory information. AId of juvenile
58
birds also receives inputs from LMAN-CORE via axon collaterals of single LMAN-CORE→RA
neurons that drive vocal output; these collaterals are present in juvenile birds up until ~45 dph
(Chung and Bottjer, unpublished data) and are not present in older juvenile or adult birds. AId
projects to striatum and several midbrain and thalamic regions that give rise to both feedback and
feed-forward pathways, creating several opportunities for information transfer between cortical
and subcortical regions. See text. Abbreviations: AId, dorsal intermediate arcopallium; dNCL,
dorsal caudolateral nidopallium; DTZ, dorsal thalamic zone; LMAN, lateral magnocellular
nucleus of the anterior nidopallium; RA, robust nucleus of the arcopallium; SpM, medial
spiriform nucleus; VTA, ventral tegmental area. B, Raw traces of extracellular activity
simultaneously recorded at two different sites within AId of a juvenile bird (44 dph) while the
bird was resting (left column; “quiescent”) versus hopping around the recording cage (right
column; “actively behaving”). Vertical lines above each raw activity trace indicate spikes from a
single neuron sorted from the extracellular activity.
The top plot in Figure 3.2A shows that 101 out of 119 neurons (85%) exhibited a significant
change in firing rate during at least one movement compared to quiescent baseline periods and
were thus classified as "movement responsive." Thirty-three out of 101 movement-responsive
neurons (33%) responded during only one scored movement whereas 68 out of 101 (67%)
responded during two or more movements (Fig. 3.2A, bottom). Very few cells responded during
five or six movements, and no cells responded during all seven movements (Fig. 3.2A, bottom).
Figure 3.2B depicts patterns of responsivity in these 101 neurons. For instance, 10 neurons were
either excited or suppressed during pecks but were otherwise not responsive during any of the
other six scored movements; likewise, two other groups of 10 neurons each selectively
modulated their firing rate only during preening or hops. Neurons that responded during two
movements could show either excitation or suppression during both movements, or excitation
during one and suppression during the other. In general, responsivity across cells was
heterogeneous – for instance, neurons responsive during four movements showed excitation or
suppression during various combinations of movements that could include any four of the seven
scored behaviors.
Higher proportions of neurons showed altered firing rates during pecks, preening, and/or hops
compared to other movements: 68 out of 107 neurons (64%) were significantly modulated during
pecks, 50 out of 93 (54%) during preening, and 60 out of 119 (50%) during hops (Fig. 3.2C).
These proportions did not differ (Fisher’s exact test, Benjamini-Hochberg corrected, peck versus
preen p = .23, peck versus hop p = .09, preen versus hop p = .71), and were each greater than the
59
proportions of neurons that responded during fluff-ups (21/79, 27%), scratches (14/55, 25%), and
stretches (7/59, 12%) (Fig. 3.2C; Fisher’s exact test, Benjamini-Hochberg corrected, p < .05 for
all comparisons between pecks, hops, and preening to each of the other three movements; p > .05
for all comparisons among these latter three movements).
As indicated above, we observed both excited and suppressed responses within single neurons:
21 out of 68 neurons (31%) that responded during multiple movements exhibited excitation
during some movements and suppression during others (Fig. 3.2B). The proportion of excited
versus suppressed responses did not differ (56%, 138/247 excited; 44%, 109/247 suppressed;
binomial test, p = .07), indicating a fairly even representation of excitation and suppression
across all movement responses. In addition, within each group of neurons that were responsive
during a given movement, the proportions of excited versus suppressed responses did not differ
(binomial test, p > .05 in all cases).
Overall, these results demonstrate that the responsivity of AId neurons is highly heterogeneous.
Neurons modulated their firing rate most often during pecks, hops, and/or preening but varied in
their responsivity profile: while a substantial population responded selectively during only one
movement, the majority of neurons were broadly responsive and showed significant excitation or
suppression during different combinations of two or more movements.
60
Figure 3.2. AId neurons respond during different scored movements with excitation and/or
suppression. A, Proportions of single AId neurons that responded during different numbers of
movements. Top: proportions are out of 119 neurons recorded in AId; 18/119 neurons were not
movement responsive, while 33/119 neurons responded during one movement, 25/119 during
61
two, 20/119 during three, 14/119 during four, 6/119 during five, and 3/119 during six
movements. Bottom: proportions are out of 101 AId neurons that responded during at least one
movement. B, Each row of each chart indicates the movements during which each neuron was
excited (green), suppressed (dark gray), or not responsive (light gray). Unfilled boxes indicate
that no data during that movement was recorded for that neuron. Charts are grouped according
to colors in A, based on the number of movements during which neurons responded. C,
Proportions of AId neurons that were significantly excited (green) or suppressed (dark gray)
during each movement type. Italicized numbers indicate the number of neurons recorded during
each movement type. **p<.005, ***p<.0001.
62
Temporal response patterns of AId neurons at movement onsets and offsets
Some AId neurons demonstrated consistent temporal changes in firing rate at movement onsets
and offsets that could be masked by measures of average firing rate. For example, Figure 3.3A
shows rasters and histograms for a single neuron during preening (left) and pecks (right). In both
cases, mean firing rate during the movement was significantly excited relative to quiescence
(preening response strength = 0.65; peck response strength = 1.53). However, the rasters and
histograms illustrate that the responses also contain periods of suppression at movement onset.
To capture these firing rate modulations, we tested for periods of significant excitation or
suppression at movement onsets and offsets. For each response, we compared histograms of
spiking activity spanning one second centered on movement onsets or offsets to histograms of
shuffled spike trains (25-ms bins) in order to identify bins in which the firing rate was
significantly excited or suppressed (see Methods). Figure 3.3 plots various examples of onset-
aligned (Fig. 3.3A, B; C, left) and offset-aligned (Fig. 3.3C, right) responses; green and gray
horizontal bars above the histograms mark excited and suppressed bins, respectively. Onset or
offset responses were defined as responses that included at least two contiguous bins (50 ms) of
significant excitation or suppression occurring within 100 ms of movement onsets or offsets.
We observed significant onset responses during pecking, hopping, and preening, but not other
movement types (Table 3.1, top). A total of 11 out of 107 neurons (10.3%) exhibited significant
excitation (7/11) or suppression (4/11) at peck onsets. Five out of 119 neurons demonstrated an
onset response during hopping (4.2%; 4/5 excitation, 1/5 suppression), as did four out of 93
neurons during preening (4.3%; 3/4 excitation, 1/4 suppression). Six out of 107 neurons
exhibited pecking offset responses (5.6%; 4/6 excitation, 2/6 suppression), and three out of 93
neurons showed preening offset responses (3.2%; all excitation) (Table 3.1, bottom).
Four onset responses (two pecking, two hopping) included consistent excitation at movement
onset even though the average firing rate during movement was not significantly different from
quiescent baseline (Table 3.1, top, “Not significant” row). Similarly, one preening response
demonstrated significant offset-excitation but lacked a change in average firing rate across the
entire movement (Table 3.1, bottom, “Not significant” row).
63
Furthermore, onset and offset responses could differ in sign (excitation or suppression) from
responses based on average firing rate: for example, while firing rate during the hop response
plotted in Figure 3.3B was suppressed relative to baseline (response strength = -.76), the raster
and histogram reveal an excitatory peak starting immediately before hop onset, indicating a
complex temporal response that included brief excitation followed by suppression. In total, we
found one suppressed response that showed excitation at movement onset (hopping; Fig. 3.3B),
and five excited responses that showed suppression at movement onset (four pecking, one
preening; e.g., Fig. 3.3A). In addition, one preening response was suppressed relative to baseline
but exhibited an excited peak in firing at preening offset (Table 3.1, bottom). These results
suggest that single AId neurons can be modulated by multiple factors during movements,
resulting in excitation at movement onsets or offsets and suppression during the movement itself,
or vice versa.
64
Figure 3.3. AId neurons show a variety of temporal response patterns during different scored
movements. A, Rasters and histograms illustrating the response of a single AId neuron during
preening (left) and pecks (right). B, Raster and histogram illustrating the response of a single
AId neuron during hops. C, Rasters and histograms illustrating the onset- (left) and offset-
aligned (right) preening response of an example AId neuron. Rows are sorted by movement
duration. Blue vertical lines mark movement onset; red lines mark movement offsets. Green
and gray horizontal bars above histograms denote periods of excitation or suppression,
respectively (see Materials and Methods). RS, standardized response strength over entire
duration for each movement type.
65
Table 3.1. Onset and offset responses across all neurons for different movement types
Peck-onset responses
(± 100 ms from peck
onset)
Hop-onset responses
(± 100 ms from hop
onset)
Preening-onset responses
(± 100 ms from preening
onset)
Response based
on average
firing rate
Excited Suppressed Excited Suppressed Excited Suppressed
Excited .05 (5/107) .04 (4/107) .01 (1/119) 0 .03 (3/93) .01 (1/93)
Not significant .02 (2/107) 0 .02 (2/119) 0 0 0
Suppressed 0 0 .01 (1/119) .01 (1/119) 0 0
Peck-offset responses
(± 100 ms from peck offset)
Preening-offset responses
(± 100 ms from preening offset)
Response based
on average
firing rate
Excited Suppressed Excited Suppressed
Excited .04 (4/107) .01 (1/107) .01 (1/93) 0
Not significant 0 0 .01 (1/93) 0
Suppressed 0 .01 (1/107) .01 (1/93) 0
Onset responses (top) and offset responses (bottom) are shown separately, categorized based on
whether average firing rate during the movement showed significant excitation (Excited),
suppression (Suppressed), or no response (Not significant).
66
Context-dependency of pecking behavior
Of the seven movements we scored, pecking behavior in particular tended to occur in different
contexts: birds always pecked while eating, but also pecked frequently at other objects such as
cage wires or perches. To investigate whether such context influenced responsivity, we
separately compared response strengths for pecks that occurred during eating versus other active
behaviors (non-eating). Absolute response strength differed for pecks that occurred during
eating versus non-eating in 47 out of 97 neurons (48%; Mann-Whitney test, p < .05). Figure
3.4A plots the context-sensitive cells that exhibited greater absolute response strength (positive
or negative) during eating (29/47, 62%; Fig. 3.4A, left) and non-eating (18/47, 38%; Fig. 3.4A,
right).
Among neurons that showed greater absolute response strength during eating-related pecks, most
exhibited relative excitation (higher firing rates) during eating-related pecks compared to non-
eating pecks (21/29, 72%; Fig. 3.4A left, gray lines). For example, Figure 3.4B plots the peck-
aligned responses for one neuron, illustrating strong excitation during pecks that occurred when
the bird was eating (left) versus a lack of response during non-eating (right). However, eight out
of 29 neurons (28%) showed relative response suppression (lower firing rates) during eating-
related pecks compared to non-eating pecks, suggesting the presence of relative excitation or
suppression in single AId neurons according to different behavioral contexts (Fig. 3.4A, left,
black lines). Similarly, 11 out of 18 neurons (61% ) showed greater relative excitation for pecks
during non-eating periods compared to eating periods (Fig. 3.4A, right, gray lines), whereas
seven out of 18 (39%) showed greater relative suppression during non-eating pecks (Fig. 3.4A,
right, black lines).
Overall, these results illustrate that pecking activity in many neurons was dependent on the
context in which the movement occurred, suggesting that changes in response strength of these
neurons is not specific to the physical movements of pecking behavior per se. One possibility is
that these neurons are involved in processing orofacial or external sensory information that is
present specifically in one context versus another.
67
Figure 3.4. AId neurons exhibit context-sensitive peck responses. A, Mean standardized
response strengths of neurons during pecks that occurred during eating versus non-eating
periods, grouped by neurons that showed greater absolute response strength during eating (left)
and non-eating (right). Left: Gray and black lines represent neurons that showed positive or
negative response strength, respectively, during pecks while eating. Right: Gray and black lines
represent neurons that showed positive or negative response strength, respectively, during non-
eating pecks. Firing rates during eating-related pecks were significantly different from non-
eating pecks for all plotted neurons. Rasters and histograms of an example neuron’s response
during pecks that occurred during eating (left) versus non-eating (right). Peck response strength
of this neuron is indicated by the starred plot point in A, left. Rows are sorted by peck duration.
Blue vertical lines mark peck onsets; red lines mark peck offsets.
68
Singing-responsive neurons in AId
One of AId's primary source of afferents is from LMAN-SHELL, which contains neurons that are
active during singing behavior and have been hypothesized to play an important role in guiding
accurate imitation of the tutor song during vocal learning (Fig. 3.1A) (Achiro and Bottjer, 2013;
Achiro et al., 2017). Moreover, lesions of AId in juvenile birds impair the bird’s ability to
achieve an accurate imitation of the adult tutor song without disrupting vocal motor output
(Bottjer and Altenau, 2010). Given this evidence for an important role for AId in vocal learning,
we hypothesized that firing rates of AId neurons would be modulated as juvenile birds were
actively engaged in singing behavior.
As for analysis of individual movements, we compared firing rates of single neurons during song
production against firing rates during quiescent baseline periods that were closest in proximity to
each singing episode. Firing rates were significantly modulated during singing in 71 out of 94
neurons (76%), including 48 excited responses and 23 suppressed responses. Figure 3.5A
illustrates the singing-aligned response of a neuron that increased its firing rate during song
renditions. Altered firing rates during vocal production in songbirds have typically been
interpreted as “singing-specific” activity. However, birds tend to make head movements and
postural body movements during singing, as well as beak and gular fluttering movements
specific to song production, raising the question of which movements are an intrinsic part of
singing behavior versus independent movements that may be performed simultaneously during
song production. Given the range of responsivity to movements across AId neurons (Fig. 3.2),
firing rate changes observed in singing-responsive neurons may reflect modulation by singing-
specific actions as well as actions that can be performed during both singing and non-singing
periods. As an initial test of this question, we measured neural activity during head movements
that occurred during singing episodes versus those that occurred during adjacent non-singing
periods in a subset of 41 singing-responsive neurons (see Materials and Methods).
Response strength during head movements that occurred within singing periods did not differ
from those within non-singing periods in most singing-responsive neurons (28/41, 68%) (Fig.
3.5B, C, left). Thus, the higher firing rate during singing periods in these neurons does not
reflect a contribution from spiking associated with head movements: despite the equal
contribution of spiking activity during head movements in singing versus non-singing periods,
69
firing rate increased significantly during singing. Thus, activity during head movements does
not account for the firing rate changes observed in these singing-responsive neurons; rather,
firing rate modulations in this subset of neurons may be related to singing-specific movements
such as beak movements, respiratory actions, or gular fluttering, or may reflect other non-
physical aspects of song production such as processing auditory-vocal feedback.
Five out of 41 neurons (12%) were both excited during singing episodes and demonstrated
greater response strength during head movements that occurred within singing compared to non-
singing periods (Fig. 3.5B, C, right; orange). Thus, excitation during head movements that
accompanied vocalizations contributed to an increased firing rate in these five neurons during
singing. Likewise, of the eight neurons that demonstrated lower response strength during head
movements that occurred during singing compared to non-singing (Fig. 3.5B, C, right; gray),
four were significantly suppressed during singing episodes. Together, these 13 neurons serve as
examples in which excitation or suppression during head movements that accompanied vocal
production contributed to overall increased or decreased activity, respectively, during singing.
An interesting interpretation is that these responses reflect integrative inputs: for instance, single
neurons may receive information about both head movements and singing behavior, such that
firing rate is enhanced specifically during the head and body movements that accompany vocal
motor output. Developing associations between head or postural movements and vocal behavior
may be a crucial component of learning to produce female-directed song and to perform
courtship dance movements while singing (Morris, 1954; Balaban, 1997; Williams, 2001;
Tomaszycki and Adkins-Regan, 2005)
70
Figure 3.5. A substantial population of AId neurons are responsive during singing. A, Raster
and histogram illustrating activity of an example AId neuron during singing episodes. Rows are
sorted by duration of each singing episode. Blue vertical line marks onset of each singing
episode; red lines mark ends of singing episodes. B, Proportions of 41 singing-responsive
neurons for which response strength during head movements that occurred within singing
periods was greater than (orange), less than (gray), or not different from (blue) head movements
that occurred during non-singing periods. C, Mean standardized response strengths during head
movements that occurred during singing versus non-singing periods. Lines connect data points
from single neurons. Left: neurons that showed comparable response strength during head
movements that occurred within singing versus non-singing periods. Right: neurons that showed
higher response strength during head movements that occurred within singing (orange) or non-
singing (gray) periods.
71
Additional sources of modulation of AId neurons: behavioral states
As indicated above, our goal was to assess the activity of AId neurons throughout an entire
session of active behaviors. As part of this approach, we devised a novel way of characterizing
each recording session by classifying contiguous time periods across each session into one of
five different “state” periods based on the bird’s behavior: eating, singing, active-movement,
quiet-attentive, or quiescence (Fig. 3.6A; see Materials and Methods). Eating states were
defined as periods when the bird was engaged in eating behavior, including pecking at, hulling,
and ingesting seeds; although eating state periods were dominated by eating-related behaviors,
other movements such as head movements or hops could also occur during brief pauses as the
bird visually scanned the food tin. Similarly, singing states included periods when birds were
engaged in song production, as well as brief pauses in-between song bouts during which birds
occasionally made brief hops, pecks, or head movements. During active-movement states, birds
could produce any of the seven movements we scored as well as head and/or postural body
movements that were not scored. The remaining two states characterized non-movement
periods: during quiet-attentive states, the bird was alert and could make small head movements
but was otherwise not moving; birds made no movements during quiescent states (quiescent
states included periods from which baseline intervals were sampled in the scored-movement
analyses above; see Materials and Methods).
Activity of AId neurons in awake, behaving juveniles is both dynamic and heterogeneous during
discrete movements as illustrated above (Figs. 3.1B, 3.2, 3.3). To capture these modulations
during state periods, we identified excited and suppressed spiking "events", defined as five or
more contiguous 10-ms bins in which the firing rate exceeded the average firing rate during
quiescent states by > 3 standard deviations (for excited events) or fell below the average
quiescence firing rate by < 1.5 standard deviations (for suppressed events). The average number
of excited events that occurred within active-movement and eating states was greater than during
quiet-attentive and quiescence states (Fig. 3.6B, left; pairwise linear contrasts, Benjamini-
Hochberg corrected, p < .001 in all cases). The frequency of excited events during singing states
was elevated relative to quiet-attentive and quiescence, but did not significantly differ from any
of the other four states Fig. 3.6B, left; pairwise linear contrasts, Benjamini-Hochberg corrected, p
> .05 in all cases). In accord with this pattern of results, the average inter-spike interval (ISI)
72
during active-movement and eating states was shorter compared to quiet-attentive and quiescent
states (Fig. 3.6C, middle; pairwise linear contrasts, Benjamini-Hochberg corrected, p < .001 in
all cases). ISI’s during singing states were not different from those during active-movement and
eating, but were shorter compared to quiescence (Fig. 3.6C, left, right; versus active-movement
and eating p > .05, versus quiescence p = .0001). The number of suppressed events did not differ
between state types (Fig. 3.6B, right; pairwise linear contrasts, Benjamini-Hochberg corrected, p
> .05 in all cases). These results indicate a high degree of excitatory modulation during active-
movement, eating and singing states, expressed as an increase in periods of higher firing rates
and/or shorter ISI’s relative to quiescence. The more modest incidence of excited events during
singing may indicate that firing rate modulation during singing states involves more uniform
increases in tonic spike rate.
Since single neurons could exhibit excited and/or suppressed response strength during scored
movements, we compared firing rates during different state types separately for neurons with
excited and suppressed responses to determine whether differences between behavioral states
could be accounted for by the activity of neurons that we identified above as movement-
responsive. Forty-eight out of 119 neurons demonstrated only excited responses during one or
more scored movements and were thus considered “movement-excited” neurons, whereas 32
showed only suppressed responses and were thus considered “movement-suppressed” (Fig.
3.2B); neurons that demonstrated both excitation and suppression during different movements
were excluded from this analysis. Among movement-excited neurons, the average firing rate
during both active-movement and eating states was greater than quiet-attentive and quiescence
(Fig. 3.6D, left), consistent with the fact that the majority of movement-responsive neurons
responded during pecks, hops, and preening – movements that occur almost exclusively during
active-movement and eating states (Fig. 3.2C). The mean firing rate of movement-excited
neurons during singing states was not significantly greater than quiescence, reflecting the fact
that movement-excited neurons were not necessarily singing-excited: nearly half of movement-
excited neurons (22/48, 46%) were either suppressed or not responsive during singing, or were
not recorded during singing. In contrast to movement-excited neurons, movement-suppressed
neurons did not show group differences in average firing rate (Fig. 3.6D, right).
73
74
Figure 3.6. AId neurons are differentially modulated during different behavioral states.
A, Schematics of example eating, singing, active-movement, quiet-attentive, and quiescent states.
Text boxes represent example scored and unscored (starred) movements that typically occurred
within that state type, though other behaviors could occur (see Materials and Methods).
B, Number of excited (left) and suppressed (right) events that occurred during each state type,
normalized by the total duration of each state type in a given recording session. Box-and-
whisker plots indicate medians and first and third quartiles; whiskers indicate data points not
considered outliers; circles represent data points from individual neurons. ***p<.001.
C, Histograms comparing distributions of inter-spike intervals (ISIs) during active-movement,
eating, and singing states (left); active-movement, eating, quiet-attentive, and quiescent states
(middle); singing, quiet-attentive, and quiescent states (right). Horizontal lines indicate
distributions that had significantly different means; dotted lines indicate means of the respective
distributions. p<.001 for all significant differences. D, Mean firing rate of movement-excited
(left) and movement-suppressed (right) neurons during each state type, normalized by the
standard deviation of each neuron’s firing rate across the recording session. Box-and whisker
plots as in B. **p<.01, ***p<.0005
We classified each movement-excited neuron according to which state elicited the highest firing
rate: firing rates of 24 out of 48 (50%) movement-excited neurons were highest during active-
movement states whereas firing rates of 14 neurons (29%) were highest during eating states and
eight neurons (17%) had maximal firing rates during singing (Fig. 3.7A). Interestingly, maximal
firing rates in a given state could occur in neurons that were responsive for scored movements
that occurred rarely or never during a given state. For example, even though preening
movements never occurred during singing, some preening-responsive neurons showed their
highest firing rates during singing states (Fig. 3.7B, C). Thus, single AId neurons may be
modulated by both singing-related factors as well as movements that occur outside of singing
behavior. Similarly, most neurons with maximal firing rates during eating states were peck-
responsive (11/14), but six were also responsive during hops and five were responsive during
preening even though these latter two movements rarely occurred during eating states (Fig. 3.7B,
C). Moreover, although pecking movements dominated eating states, three neurons with
maximum firing rates during eating states were not peck-responsive, indicating that the
heightened firing rate of these neurons during eating did not relate to pecking behavior (Fig.
3.7B, C). Rather, the increased firing rate may be related to other unscored factors such as head
movements or hulling behavior, or external sensory inputs. These results extend our data on
movement analyses by highlighting the complexity of the information that is integrated by AId
75
neurons: single neurons can demonstrate robust, differentially patterned responses during
specific movements while also exhibiting firing rate changes during state periods that are
unrelated to those movements.
Figure 3.7. Movement-excited AId neurons are also modulated by other unscored factors during
different state periods. A, Proportions of 48 neurons that showed excitation during one or more
scored movements, classified by maximal average firing rate during each state type. B, Relative
proportions of scored behaviors that occurred during each state type. Proportion totals include
all occurrences of each of the seven scored movements as well as all singing episodes. C,
Proportions of neurons scored as hop-, peck-, and/or preening-responsive among neurons that
showed maximal firing rate during active-movement (left), eating (middle), and singing (right)
76
states. Italicized numbers indicate the number of neurons that showed highest average firing rate
during each state type.
While the majority of neurons we recorded were responsive during one or more scored
movements, 18 out of 119 cells (15%) were not significantly modulated during any scored
movement. To investigate whether even "movement-unresponsive" neurons nonetheless
modulate their activity when the bird is actively behaving, we grouped these non-responsive
neurons by the state during which their firing rate was highest and tested whether activity was
differentially modulated between state groups. The firing rate of neurons within each preferred
state group tended to be higher relative to all other state types despite consisting entirely of non-
responsive neurons, though group differences did not reach significance due to small group sizes
(Fig. 3.8; pairwise linear contrasts, Benjamini-Hochberg corrected, p > .05 in all cases). This
pattern of results suggests that some of the neurons unresponsive during discrete movements
may be nonetheless modulated by other factors as the juvenile is actively behaving.
One remaining source of physical movement that could contribute to the modulation of AId
neurons are head and body movements. Birds constantly made quick, saccade-like movements
throughout recording periods, easily resulting in over a thousand head or postural movements in
a typical 30-minute session. As an initial test of whether firing rate was modulated relative to
these movements, we scored head movements across the entire recording session in a subset of
12 neurons. We also included neurons for which a portion of head movements had been scored
only during singing and adjacent non-singing periods (n = 41; see above). We found that the
response strength of most neurons (87%; 46/53) was significantly modulated during head
movements (32/53 excited, 14/53 suppressed). These responses included neurons that had
otherwise been considered “non-responsive” – nearly half of neurons that did not respond during
any previously scored movements (44%, 8/18) exhibited altered response strength selectively
during head movements.
77
Figure 3.8. Activity of non-responsive AId neurons is nonetheless modulated during different
states. Graphs plot mean firing rates of neurons that were not responsive for any scored
movement during each state type, classified by neurons that showed maximal average firing rate
during active-movement (top left), eating (top right), singing (bottom left), and quiet-attentive
(bottom right) states. Firing rates are normalized by the standard deviation of each neuron’s
firing rate across the recording session. Box-and-whisker plots indicate medians and first and
third quartiles; whiskers indicate data points not considered outliers; circles represent data points
from individual neurons.
78
DISCUSSION
We found that most AId neurons were selective for single movements or for different
combinations of movements. Neurons responsive during different movements frequently
demonstrated excitation during one movement and suppression during another. Moreover,
individual responses could include both transient excitation at movement onset or offset as well
as suppression of average firing rate during the movement itself, or vice versa. The diversity of
neural responses in AId is strikingly similar to the response profile of neurons in macaque motor
cortex, where single neurons demonstrate heterogeneous, multiphasic temporal patterns of
activity across reaching movements (Churchland and Shenoy, 2007). These complex responses
may result from multiple inputs relating to different movements or aspects of movements onto
single AId neurons, as well as local transformation of afferent inputs within AId. AId circuitry
includes a local inhibitory network, as evidence by the fact that blocking GABA-A receptors in
AId of anesthetized zebra finches elicits increased spontaneous firing rates, and parvalbumin is
expressed in higher levels in AId compared to surrounding motor cortex (Mello et al., 2019;
Yuan and Bottjer, 2019). Similarly, mammalian motor cortex contains a substantial population
of inhibitory interneurons, which have been implicated in both regulating plasticity during motor
skill learning and coordinating activity across motor cortex during behavior (Jacobs and
Donoghue, 1991; Hess and Donoghue, 1994; Donato et al., 2013; Chen et al., 2015). Of
particular interest is the hypothesis that selective disinhibition of different motor cortical
populations may serve to link spatially separated regions of motor cortex and thereby coordinate
activity across different motor effectors (Spiro et al., 1999; Schneider et al., 2002; Capaday,
2004). A similar mechanism would be advantageous for integrating information within AId:
avian motor cortex receives topographic input from parallel circuits that process auditory, visual,
and somatosensory information, so a local mechanism linking such inputs could provide
functional sensorimotor integration across modalities within AId (Zeier and Karten, 1971;
Bottjer et al., 2000).
The heterogeneous response profile of motor cortical neurons across taxa raises interesting
questions about what factors contribute to the tuning of these neurons. Studies from recordings
in macaque motor cortex during a reaching task found that the activity of single neurons
corresponded with a variety of different movement parameters – for instance, some showed
phase changes as a function of reach direction while others exhibited activity changes that
79
correlated with movement speed (Evarts, 1968; Georgopoulos et al., 1982, 1988; Schwartz et al.,
1988; Schwartz and Moran, 2000; Churchland and Shenoy, 2007). Highly integrated multi-
modal tuning may be a key feature of motor cortical neurons: parameters of an arm-reaching task
such as movement direction or end position of the limb could account for only a portion of
spiking patterns in single motor cortical neurons, indicating that individual neurons may be tuned
in a multidimensional space and that testing neural activity relative to any single parameter may
account only partially for multidimensional tuning profiles (Aflalo and Graziano, 2006, 2007).
Likewise, we found that single AId neurons could be modulated both during individual
movements and during state periods that did not include those movements, indicating that
neurons were modulated by multiple factors.
As expected given multiple parallel sensory inputs, the tuning profiles of neurons in both avian
and mammalian motor cortex typically include both motor and sensory responsivity. For
instance, neurons in RA, which lies adjacent to AId in songbird motor cortex, drive vocal motor
output in awake birds and demonstrate robust responses to playback of song stimuli in
anesthetized birds (Nottebohm et al., 1976; Doupe and Konishi, 1991; Vicario and Yohay, 1993;
Wild, 1993; Yu and Margoliash, 1996; Dave et al., 1998; Dave and Margoliash, 2000; Leonardo
and Fee, 2005; Kojima and Doupe, 2007; Yuan and Bottjer, 2019). Recordings from macaque
motor cortex have likewise demonstrated sensitivity to non-motor stimuli: for instance, in
visually-guided target-reaching paradigms, some motor cortical neurons exhibit consistent
activity related to the visual target, regardless of the limb trajectory used to reach that target
(Tanji and Evarts, 1976; Evarts and Fromm, 1977; Murata et al., 1997; Shen and Alexander,
1997). The activity of AId neurons may be similarly modulated by integration of a variety of
motor and sensory factors to produce heterogeneous responses with complex temporal patterning
during diverse movements.
We found that most AId neurons demonstrated altered response strength during movements
relative to quiescence; under the conditions of our recordings, firing rate modulations occurred
most often during pecks, preening, and/or hops. It is difficult to know whether activity in AId
neurons is pre-motor, modulatory, or reflects movement feedback or sensory inputs. AId
neurons project to several targets, including the striatum, a dorsal thalamic zone, the lateral
hypothalamus, the medial spiriform nucleus in the caudal thalamus, deep layers of the tectum,
80
broad areas of the pontine and midbrain reticular formation, and the ventral tegmental area (Fig.
3.1) (Bottjer et al., 2000). The medial pontine reticular formation contain premotor neurons that
contribute to neck and locomotive movements in other avian species (Steeves et al., 1987;
Valenzuela et al., 1990; Dubbeldam, 1998; Wild and Krützfeldt, 2012); peck, preening, and hop-
related activity may in part reflect the projections to these premotor centers. Moreover, previous
studies have found evidence of induced ZENK activity in AId specifically during hopping
behavior (Feenders et al., 2008). However, both prior evidence and data presented here suggest
that movement-related activity in AId is not necessarily driving peck, preening, and/or hopping
behavior per se. Importantly, lesions of AId in juvenile birds do not induce disruption of song
motor output or noticeable motor deficits in general movements, suggesting that AId neurons are
not driving voluntary pecking or hopping movements (Bottjer and Altenau, 2010). Moreover,
the results presented here demonstrate that single AId neurons do not respond consistently during
one particular type of movement. For instance, we found a substantial population of peck-
selective neurons that responded during pecks when the bird was eating but not when the bird
pecked at other objects around the cage (or vice versa), even though pecking movements in these
different contexts would recruit many of the same muscle groups. Behavioral and functional
experiments have also implicated AId circuitry in playing a role in highly integrative, complex
behaviors that extend beyond pure motor control, including ingestive behaviors, fear
conditioning, and vocal learning (Lowndes and Davies, 1994; Lowndes et al., 1994; Knudsen
and Knudsen, 1996; Aoki et al., 2006; Campanella et al., 2009; Saint-Dizier et al., 2009; Bottjer
and Altenau, 2010; Achiro et al., 2017). One such example is ANC (caudal arco-nidopallium),
which partially overlaps with AId and exhibited increased 2-deoxyglucose uptake after adult
birds who had been isolated from females for several weeks participated in their first courtship
experience after isolation; the amount of glucose consumption in ANC positively correlated with
isolation time but did not correlate with the amount of movement activity exhibited by the males
(Bischof and Herrmann, 1986).
Rather than generically driving motor behavior per se, an important function of cortical outputs
onto hindbrain motor circuitry is incorporating sensory information in order to appropriately
direct motor output based on an animal’s environment and/or goals; this sensorimotor integration
is necessary for voluntary movements such as goal-directed motor behaviors (e.g., object-
directed grasping) as well as adaptive movements based on environmental perturbations. Avian
81
and mammalian motor cortices receive multi-modal inputs and target hindbrain motor centers,
making them ideally situated to carry out sensorimotor integration for this purpose. In
macaques, neurons in motor cortical areas demonstrate a selective response when grasping at a
particular object and corresponding visual selectivity for the same object when the monkey
fixates on the object without grasping; correspondingly, inactivation of the same motor cortical
region resulted in grasping deficits resulting from preparatory hand shaping that was
inappropriate for the target object, suggesting a specific impairment in visuomotor
transformations for object grasping or manipulation rather than a gross motor impairment of
hand movements (Murata et al., 1997; Fogassi, 2001; Rizzolatti and Luppino, 2001). Some AId
neurons may serve a similar function in linking sensory information to motor output – for
example, “eating-peck” preferred responses in this framework could similarly represent an
integrated response when the visual stimulus of seed is present as the bird pecks; these neurons
could link visual information about seed with somatosensory information to contribute
specifically to food-directed pecking movements.
Although motor cortex is generally well-situated to carry out sensory-motor transformations, AId
is uniquely connected to various regions that suggest it may include a specific role in vocal
learning and behavior. In juvenile birds, LMAN-CORE neurons that drive vocal motor output
make a collateral projection into AId (Miller-Sims and Bottjer, 2012). This projection is no
longer present in older juvenile birds, suggesting a developmentally regulated role for AId
specifically during a restricted period of vocal learning. In addition, AId projects to higher-order
thalamic nuclei that are linked to vocal learning, DLM (dorsolateral nucleus of the medial
thalamus) and DMP (dorsomedial nucleus of the posterior thalamus) (Bottjer et al., 2000). DLM
is required for normal song behavior and projects to LMAN-SHELL, whereas DMP projects to
MMAN (medial magnocellular nucleus of anterior nidopallium); both LMAN and MMAN are
required for development of an accurate imitation of tutor song (Bottjer et al., 1984; Foster et al.,
1997; Vates et al., 1997; Foster and Bottjer, 2001; Aronov et al., 2008; Goldberg and Fee, 2011;
Chen et al., 2014). AId also projects to lateral hypothalamus, striatum, and the area of
dopaminergic neurons in the VTA (ventral tegmental area) that projects to a nucleus in avian
basal ganglia that is necessary for song learning (Bottjer et al., 2000); these limbic-related
projections further suggest that AId neurons are specially situated to contribute to vocal learning
and behavior.
82
One hypothesis to integrate these unique connections with the multimodal integrative function of
motor cortex is that some AId neurons may be involved in mediating movements in the context
of song behavior and learning. Song production in zebra finches is a courtship behavior during
which males vocalize while performing a dance-like sequence of hopping movements oriented
towards a female (Morris, 1954; Williams, 2001; Cooper and Goller, 2004; Dalziell et al., 2013;
Ota et al., 2015; Ullrich et al., 2016). Sensory cues contribute to establishing a social context for
courtship behavior. For instance, adult males can use visual cues to select between two female
birds shown in a silent video feed, but the addition of auditory cues induces stronger courtship
responses (Galoch and Bischof, 2006, 2007). AId’s brainstem-projecting neurons are well
situated to integrate environmental context cues when females are present to guide performance
of movements to accompany singing. In this framework, the relatively high proportions of peck,
preening, and hopping-related responses observed here may reflect the fact that beak movements
and hopping are important components of courtship behavior. In quail-chick chimeras, chicks
that received transplants of lower brainstem somites from quails retained chick-like call
structures but adopted quail-like patterns of head movements specifically during vocalizations;
similar head movements made outside of vocalization periods were not affected, indicating the
presence of specialized circuitry that mediates movements in the context of vocal behavior
(Balaban, 1997).
Circuitry processing non-vocal elements of singing in zebra finches may be similarly specialized.
For instance, we found a subpopulation of AId neurons that demonstrated differential response
strength during head movements that occurred during singing versus non-singing periods. These
neurons may serve as examples of integration between non-vocal and vocal elements of song
behavior, where neural activity is enhanced specifically during head movements that accompany
singing. This hypothesis raises an interesting prediction: although AId lesions in juvenile birds
do not induce any gross motor deficits, it is possible that hopping and/or head movements
performed during song production would be disrupted. Such a result would be consistent with
previous studies in which c-Fos immunoreactivity in AId was increased after adult males
performed non-singing courtship behaviors directed towards a live female (Kimpo and Doupe,
1997). Involvement of AId in courtship-related movements would draw an interesting parallel to
studies in mammals that have proposed that motor cortex may be parceled into “action zones”
that each process information for different ethologically relevant categories of movement
83
(Graziano, 2006; Graziano and Aflalo, 2007). For instance, stimulation of one region of
macaque motor cortex results in the animal closing its hand in a grip while bringing it to its
mouth and opening its mouth, as if eating an object, while stimulation of another region results in
the macaque raising its arm and turning its head sharply to one side as if in defense (Graziano et
al., 2002). In this context, RA and AId, which are directly adjacent to one another in motor
cortex, could serve as an “action zone” that mediates the vocal and non-vocal movements that
comprise song behavior.
While brainstem projections may play a role in active behavior, the thalamic and midbrain
projections of AId that give rise to recurrent feedback loops through cortico-basal ganglia
circuitry may integrate multi-modal information to be used in song learning. Although AId does
not drive song output (Bottjer et al., 2000; Bottjer and Altenau, 2010), we found a substantial
population of singing-responsive neurons. For a large proportion of these neurons, the changes
in firing rate during singing could not be attributed to any of the movements that we scored
(including orienting head movements). These firing rate modulations may instead be related to
singing-specific movements such as beak movements or gular fluttering; alternatively, singing-
related activity could reflect auditory-vocal processing or active evaluations of the juvenile’s
vocal behavior during sensorimotor practice. Accurate song learning requires multiple factors
beyond simply matching vocal output to an auditory goal – for instance, vocal learning in
juvenile zebra finches that are tutored with only passive playback of the tutor song is severely
impaired, whereas pairing auditory tutoring with a visual model of an adult zebra finch enhances
learning (Derégnaucourt et al., 2013; Ljubičić et al., 2016). Moreover, visual cues provided
during singing, such as wing strokes from adult females, provide feedback that can influence the
juvenile’s vocal learning (West and King, 1988; King et al., 2005). Multi-modal inputs from
dNCL and singing-activated inputs from LMAN may converge in AId, integrating important
non-vocal and vocal elements of courtship song behavior that must be learned during a sensitive
period of development.
84
Chapter 4: Concluding Remarks
RA, RA-cup and AId are key motor cortical regions that are each well situated to mediate
different aspects of vocal learning and production. However, prior to the studies presented here,
little was known about the characteristics of neural activity in avian motor cortex of juvenile
songbirds as they are actively engaged in sensorimotor learning. Chapter 2 investigated how
auditory-vocal information important for vocal learning, such as the juvenile's own song or the
adult tutor song, were represented in each of these regions in anesthetized birds. I found that
neurons in RA showed strong developmental changes: neurons in juvenile birds preferred
immature vocal sounds over adult songs but, unlike adult birds, did not prefer the juvenile's own
song over other age-matched conspecific songs. Moreover, song-evoked responses from
juvenile RA neurons lacked temporal patterning. These response properties may reflect a neural
substrate that facilitates vocal learning in juveniles: broad responsivity may reflect that RA
neurons in juvenile birds encode a broad motor representation, which would be ideal for
promoting exploration of vocal motor space in early learning; likewise, variable temporal
patterning may reflect increased spiking variability that allows the system to explore a variety of
activity patterns to help determine the optimal pattern for producing a given sound. In providing
insight to how tuning of motor cortical neurons may subserve experience-dependent motor
learning, these results make important contributions that address fundamental principles of goal-
directed skill learning.
The recordings in RA-cup and AId represent the first published studies investigating the song
responsivity of neurons in these regions and thus lay important groundwork for understanding
how each may participate in vocal learning and production. In contrast to RA, I found that
neurons in RA-cup did not show developmental changes in selective tuning, suggesting that they
may instead participate in general processing of complex auditory information. AId differed
from both RA and RA-cup in that neurons were not responsive to song playback – despite
receiving input from LMAN-SHELL, which contains song-responsive neurons. Data presented in
Chapter 3 follow up on this surprising result and illustrate the need to consider the wide range of
multimodal factors that may contribute to motor skill learning: AId contains highly
heterogeneous neurons that are unresponsive to playback of song stimuli in anesthetized birds
but are likely integrating song-related information across multiple modalities in awake, behaving
85
juveniles; I found not only a substantial population of singing-responsive neurons, but also
neurons that were modulated during different overt movements as well as other unscored factors
during different behavioral state periods. The diverse responsivity observed coupled with AId’s
place within cortical circuitry suggests that AId neurons serve as a key point of integration for
multimodal aspects of song learning and behavior, and highlight the great potential of AId as a
tractable region for dissecting mechanisms of multimodal integration in the context of goal-
directed motor skill learning. These results lay the groundwork for future studies to further
specify the tuning profile of these neurons. For instance, comparing activity of AId neurons as
juvenile birds sing alone versus during video presentation of female birds who are either
unattentive or providing visual feedback of the juvenile’s performance via wing strokes could
investigate whether AId neurons are involved in processing visual information that may be used
to guide vocal learning. In addition, the context-dependency of some AId neurons provides
another intriguing avenue for future study: for instance, what specific factors are present during
eating versus non-eating periods that elicit firing rate modulations during pecks that occur in the
former context but not the latter, and how might identifying these factors inform our
understanding of what the peck-evoked responses represent? Taken together, the results
presented here emphasize the highly integrative nature of motor cortical regions and encourage
consideration of the wide variety of modulating factors that may be involved in motor skill
learning and performance.
86
References
Achiro JM, Bottjer SW (2013) Neural representation of a target auditory memory in a cortico-
basal ganglia pathway. J Neurosci 33:14475–14488.
Achiro JM, Shen J, Bottjer SW (2017) Neural activity in cortico-basal ganglia circuits of juvenile
songbirds encodes performance during goal-directed learning. Elife 6.
Adret P, Margoliash D (2002) Metabolic and neural activity in the song system nucleus robustus
archistriatalis: effect of age and gender. J Comp Neurol 454:409–423.
Aflalo TN, Graziano MSA (2006) Partial tuning of motor cortex neurons to final posture in a
free-moving paradigm. Proc Natl Acad Sci U S A 103:2909–2914.
Aflalo TN, Graziano MSA (2007) Relationship between unconstrained arm movements and
single-neuron firing in the macaque motor cortex. J Neurosci 27:2760–2780.
Aoki N, Csillag A, Matsushima T (2006) Localized lesions of arcopallium intermedium of the
lateral forebrain caused a handling-cost aversion in the domestic chick performing a binary
choice task. Eur J Neurosci 24:2314–2326.
Aronov D, Andalman AS, Fee MS (2008) A specialized forebrain circuit for vocal babbling in
the juvenile songbird. Science (80- ) 320:630–634.
Ashmore RC, Renk JA, Schmidt MF (2008) Bottom-up activation of the vocal motor forebrain
by the respiratory brainstem. J Neurosci 28:2613–2623.
Balaban E (1997) Changes in multiple brain regions underlie species differences in a complex,
congenital behavior. Proc Natl Acad Sci U S A 94:2001–2006.
Behroozmand R, Larson CR (2011) Error-dependent modulation of speech-induced auditory
suppression for pitch-shifted voice feedback. BMC Neurosci 12.
Behroozmand R, Oya H, Nourski K V, Kawasaki H, Larson CR, Brugge JF, Howard III MA,
Greenlee JDW (2016) Neural Correlates of Vocal Production and Motor Control in Human
Heschl’s Gyrus. J Neurosci 36:2302–2315.
Bell CC (1981) An efference copy which is modified by reafferent input. Science (80- )
214:450–453.
Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful
approach to multiple testing. J R Stat Soc 57:289–300.
Billimoria CP (2006) Neuromodulation of spike-timing precision in sensory neurons. J Neurosci
26:5910–5919.
Bischof HJ, Herrmann K (1986) Arousal enhances [14C]2-deoxyglucose uptake in four forebrain
areas of the zebra finch. Behav Brain Res 21:215–221.
Böhner J (1983) Song learning in the zebra finch (taeniopygia guttata): Selectivity in the choice
of a tutor and accuracy of song copies. Anim Behav 31:231–237.
Böhner J (1990) Early acquisition of song in the zebra finch, Taeniopygia guttata. Anim Behav
39:369–374.
87
Bottjer SW, Alderete TL, Chang D (2010) Conjunction of vocal production and perception
regulates expression of the immediate early gene ZENK in a novel cortical region of
songbirds. J Neurophysiol 103:1833–1842.
Bottjer SW, Altenau B (2010) Parallel pathways for vocal learning in basal ganglia of songbirds.
Nat Neurosci 13:153–155.
Bottjer SW, Brady JD, Cribbs B (2000) Connections of a motor cortical region in zebra finches:
relation to pathways for vocal learning. J Comp Neurol 260:244–260.
Bottjer SW, Miesner EA, Arnold AP (1984) Forebrain lesions disrupt development but not
maintenance of song in passerine birds. Science (80- ) 224:901–903.
Bottjer SW, To M (2012) Afferents from Vocal Motor and Respiratory Effectors Are Recruited
during Vocal Production in Juvenile Songbirds. J Neurosci 32:10895–10906.
Brainard MS, Doupe AJ (2002) What songbirds teach us about learning. Nature 417:351–358.
Brainard MS, Doupe AJ (2013) Translating birdsong: songbirds as a model for basic and applied
medical research. Annu Rev Neurosci 36:489–517.
Braun K, Bock J, Metzger M, Jiang S, Schnabel R (1999) The dorsocaudal neostriatum of the
domestic chick: A structure serving higher associative functions. Behav Brain Res 98:211–
218.
Briggs F, Usrey WM (2008) Emerging views of corticothalamic function. Curr Opin Neurobiol
18:403–407.
Calabrese A, Woolley SMN (2015) Coding principles of the canonical cortical microcircuit in
the avian brain. Proc Natl Acad Sci 112:3517–3522.
Campanella LCA, Silva AA da, Gellert DS, Parreira C, Ramos MC, Paschoalini MA, Marino-
Neto J (2009) Tonic serotonergic control of ingestive behaviours in the pigeon (Columba
livia): The role of the arcopallium. Behav Brain Res 205:396–405.
Capaday C (2004) The integrated nature of motor cortical function. Neuroscientist 10:207–220.
Caroni P, Donato F, Muller D (2012) Structural plasticity upon learning: regulation and
functions. Nat Rev Neurosci 13:478–490.
Catchpole CK, Slater PJB (2008) Bird Song: Biological Themes and Variations, 2nd ed.
Cambridge, UK: Cambridge University Press.
Chen JR, Stepanek L, Doupe AJ (2014) Differential contributions of basal ganglia and thalamus
to song initiation, tempo, and structure. J Neurophysiol 111:248–257.
Chen SX, Kim AN, Peters AJ, Komiyama T (2015) Subtype-specific plasticity of inhibitory
circuits in motor cortex during motor learning. Nat Neurosci 18:1109–1115.
Chen Y, Matheson LE, Sakata JT (2016) Mechanisms underlying the social enhancement of
vocal learning in songbirds. Proc Natl Acad Sci 113:6641–6646.
Churchland MM, Shenoy K V. (2007) Temporal complexity and heterogeneity of single-neuron
activity in premotor and motor cortex. J Neurophysiol 97:4235–4257.
88
Cooper BG, Goller F (2004) Multimodal Signals: Enhancement and Constraint of Song Motor
Patterns by Visual Display. Science (80- ) 303:544–546.
Crapse TB, Sommer M a (2008) Corollary discharge circuits in the primate brain. Curr Opin
Neurobiol 18:552–557.
D’Ausilio A, Pulvermuller F, Salmas P, Bufalari I, Begliomini C, Fadiga L (2009) The motor
somatotopy of speech perception. Curr Biol 19:381–385.
Dalziell AH, Peters RA, Cockburn A, Dorland AD, Maisey AC, Magrath RD (2013) Dance
choreography is coordinated with song repertoire in a complex avian display. Curr Biol
23:1132–1135.
Dave AS, Margoliash D (2000) Song Replay During Sleep and Computational Rules for
Sensorimotor Vocal Learning. Science (80- ) 290:812–816.
Dave AS, Yu AC, Margoliash D (1998) Behavioral state modulation of auditory activity in a
vocal motor system. Science (80- ) 282:2250–2254.
Derégnaucourt S, Poirier C, Kant A Van der, Linden A Van der, Gahr M (2013) Comparisons of
different methods to train a young zebra finch (Taeniopygia guttata) to learn a song. J
Physiol 107:210–218.
Devlin JT, Aydelott J (2009) Speech perception: motoric contributions versus the motor theory.
Curr Biol 19:R198–R200.
Donato F, Rompani SB, Caroni P (2013) Parvalbumin-expressing basket-cell network plasticity
induced by experience regulates adult learning. Nature 504:272–276.
Doupe AJ (1997) Song-and order-selective neurons in the songbird anterior forebrain and their
emergence during vocal development. J Neurosci 17:1147–1167.
Doupe AJ, Konishi M (1991) Song-selective auditory circuits in the vocal control system of the
zebra finch. Proc Natl Acad Sci U S A 88:11339–11343.
Dubbeldam JL (1998) The neural substrate for “learned” and “nonlearned” activities in birds: A
discussion of the organization of bulbar reticular premotor systems with side-lights on the
mammalian situation. In: Acta Anatomica, pp 157–172.
Eales LA (1985) Song learning in zebra finches: some effects of song model availability on what
is learnt and when. Anim Behav 33:1293–1300.
Eales LA (1989) The influences of visual and vocal interaction on song learning in Zebra
finches. Anim Behav 37:507–508.
Evarts E V. (1968) Relation of pyramidal tract activity to force exerted during voluntary
movement. J Neurophysiol 31:14–27.
Evarts E V., Fromm C (1977) Sensory responses in motor cortex neurons during precise motor
control. Neurosci Lett 5:267–272.
Feenders G, Liedvogel M, Rivas M, Zapka M, Horita H, Hara E, Wada K, Mouritsen H, Jarvis
ED (2008) Molecular mapping of movement-associated areas in the avian brain: A motor
89
theory for vocal learning origin. PLoS One 3.
Fogassi L (2001) Cortical mechanism for the visual guidance of hand grasping movements in the
monkey: A reversible inactivation study. Brain 124:571–586.
Foster EF, Bottjer SW (1998) Axonal connections of the high vocal center and surrounding
cortical regions in juvenile and adult male zebra finches. J Comp Neurol 397:118–138.
Foster EF, Bottjer SW (2001) Lesions of a telencephalic nucleus in male zebra finches:
Influences on vocal behavior in juveniles and adults. J Neurobiol 46:142–165.
Foster EF, Mehta RP, Bottjer SW (1997) Axonal connections of the medial magnocellular
nucleus of the anterior neostriatum in zebra finches. J Comp Neurol 382:364–381.
Fu M, Yu X, Lu J, Zuo Y (2012) Repetitive motor learning induces coordinated formation of
clustered dendritic spines in vivo. Nature 483:92–96.
Gale SD, Person AL, Perkel ADJ (2008) A novel basal ganglia pathway forms a loop linking a
vocal learning circuit with its dopaminergic input. J Comp Neurol 508:824–839.
Galoch Z, Bischof HJ (2006) Zebra Finches actively choose between live images of conspecifics.
Ornithol Sci 5:57–64.
Galoch Z, Bischof HJ (2007) Behavioural responses to video playbacks by zebra finch males.
Behav Processes 74:21–26.
Garst-Orozco J, Babadi B, Ölveczky BP (2014) A neural circuit mechanism for regulating motor
variability during skill learning. Elife 3.
Georgopoulos AP, Kalaska JF, Caminiti R, Massey JT (1982) On the relations between the
direction of two-dimensional arm movements and cell discharge in primate motor cortex. J
Neurosci 2:1527–1537.
Georgopoulos AP, Kettner RE, Schwartz AB (1988) Primate motor cortex and free arm
movements to visual targets in three-dimensional space. II. Coding of the direction of
movement by a neuronal population. J Neurosci 8:2928–2937.
Gobes SMH, Zandbergen MA, Bolhuis JJ (2010) Memory in the making: Localized brain
activation related to song learning in young songbirds. Proc R Soc B Biol Sci 277:3343–
3351.
Goldberg JH, Fee MS (2011) Vocal babbling in songbirds requires the basal ganglia-recipient
motor thalamus but not the basal ganglia. J Neurophysiol 105:2729–2739.
Graziano M (2006) The organization of behavioral repertoire in motor cortex. Annu Rev
Neurosci 29:105–134.
Graziano MS., Taylor CS., Moore T (2002) Complex movements evoked by microstimulation of
precentral cortex. Neuron 34:841–851.
Graziano MSA, Aflalo TN (2007) Mapping behavioral repertoire onto the cortex. Neuron
56:239–251.
Guillery RW, Sherman SM (2002) Thalamic relay functions and their role in corticocortical
90
communication: generalizations from the visual system. Neuron 33:163–175.
Gunturken O (1997) Cognitive impairments after lesions of the neostriatum caudolaterale and its
thalamic afferent in pigeons: functional similarities to the mammalian prefrontal system? J
Hirnforsch 38:133–143.
Hahnloser RH, Kotowicz A (2010) Auditory representations and memory in birdsong learning.
Curr Opin Neurobiol 20:332–339.
Hahnloser RHR, Kozhevnikov A a, Fee MS (2002) An ultra-sparse code underlies the generation
of neural sequences in a songbird. Nature 419:65–70.
Herrmann K, Arnold AP (1991) The development of afferent projections to the robust
archistriatal nucleus in male zebra finches: a quantitative electron microscopic study. J
Neurosci 11:2063–2074.
Hess G, Donoghue JP (1994) Long-term potentiation of horizontal connections provides a
mechanism to reorganize cortical motor maps. J Neurophysiol 71:2543–2547.
Hickok G, Houde J, Rong F (2011) Sensorimotor Integration in Speech Processing:
Computational Basis and Neural Organization. Neuron 69:407–422.
Houde JF, Chang EF (2015) The cortical computations underlying feedback control in vocal
production. Curr Opin Neurobiol 33:174–181.
Immelmann K (1969) Song development in the zebra finch and other estrildid finches. In: Bird
Vocalizations (Hinde RA, ed), pp 61–74. Cambridge University Press.
Iyengar S, Viswanathan SS, Bottjer SW (1999) Development of topography within song control
circuitry of zebra finches during the sensitive period for song learning. J Neurosci 19:6037–
6057.
Jacobs K, Donoghue J (1991) Reshaping the cortical motor map by unmasking latent
intracortical connections. Science (80- ) 251:944–947.
Jarvis ED (2004) Learned birdsong and the neurobiology of human language. Ann N Y Acad Sci
1016:749–777.
Johnson F, Bottjer SW (1992) Growth and regression of thalamic efferents in the song-control
system of male zebra finches. J Comp Neurol 326:442–450.
Johnson F, Sablan MM, Bottjer SW (1995) Topographic organization of a forebrain pathway
involved with vocal learning in zebra finches. J Comp Neurol 358:260–278.
Joris PX, Louage DH, Cardoen L, van der Heijden M (2006) Correlation Index: A new metric to
quantify temporal coding. Hear Res 216–217:19–30.
Kargo WJ, Nitz DA (2004) Improvements in the signal-to-noise ratio of motor cortex cells
distinguish early versus late phases of motor skill learning. J Neurosci 24:5560–5569.
Karten HJ (2013) Neocortical evolution: Neuronal circuits arise independently of lamination.
Curr Biol 23:R12–R15.
Keller GB, Hahnloser RHR (2009) Neural processing of auditory feedback during vocal practice
91
in a songbird. Nature 457:187–190.
Kelley DB, Nottebohm F (1979) Projections of a telencephalic auditory nucleus-field L-in the
canary. J Comp Neurol 183:455–469.
Kimpo RR, Doupe AJ (1997) FOS is induced by singing in distinct neuronal populations in a
motor network. Neuron 18:315–325.
King AP, West MJ, Goldstein MH (2005) Non-vocal shaping of avian song development:
Parallels to human speech development. Ethology 111:101–117.
Knudsen EI, Knudsen PF (1996) Disruption of auditory spatial working memory by inactivation
of the forebrain archistriatum in barn owls. Nature 383:428–431.
Kojima S, Doupe AJ (2007) Song selectivity in the pallial-basal ganglia song circuit of zebra
finches raised without tutor song exposure. J Neurophysiol 98:2099–2109.
Konishi M (1965) The role of auditory feedback in the control of vocalization in the white-
crowned sparrow. Z Tierpsychol 22:770–783.
Konishi M (1985) Birdsong: From Behavior to Neuron. Annu Rev Neurosci 8:125–170.
Konishi M (2004) The role of auditory feedback in birdsong. Ann N Y Acad Sci.
Leonardo A (2005) Degenerate coding in neural systems. J Comp Physiol A Neuroethol Sensory,
Neural, Behav Physiol 191:995–1010.
Leonardo A, Fee MS (2005) Ensemble coding of vocal control in birdsong. J Neurosci 25:652–
661.
Leutgeb S, Husband S, Riters L V., Shimizu T, Bingman VP (1996) Telencephalic afferents to
the caudolateral neostriatum of the pigeon. Brain Res 730:173–181.
Liberman AM, Mattingly IG (1985) The motor theory of speech perception revised. Cognition
21:1–36.
Ljubičić I, Hyland Bruno J, Tchernichovski O (2016) Social influences on song learning. Curr
Opin Behav Sci 7:101–107.
London SE, Clayton DF (2008) Functional identification of sensory mechanisms required for
developmental song learning. Nat Neurosci 11:579–586.
Louage DHG, van der Heijden M, Joris PX (2004) Temporal properties of responses to
broadband noise in the auditory nerve. J Neurophysiol 91:2051–2065.
Lowndes M, Davies DC (1994) The Effects of Archistriatal Lesions on One‐trial Passive
Avoidance Learning in the Chick. Eur J Neurosci 6:525–530.
Lowndes M, Davies DC, Johnson MH (1994) Archistriatal Lesions Impair the Acquisition of
Filial Preferences During Imprinting in the Domestic Chick. Eur J Neurosci 6:1143–1148.
Ludwig KA, Miriani RM, Langhals NB, Joseph MD, Anderson DJ, Kipke DR (2009) Using a
common average reference to improve cortical neuron recordings from microelectrode
arrays. J Neurophysiol 101:1679–1689.
92
Makino H, Hwang EJ, Hedrick NG, Komiyama T (2016) Circuit Mechanisms of Sensorimotor
Learning. Neuron 92:705–721.
Mandelblat-Cerf Y, Las L, Denisenko N, Fee MS (2014) A role for descending auditory cortical
projections in songbird vocal learning. Elife 3.
Mann NI, Slater PJB (1995) Song tutor choice by zebra finches in aviaries. Anim Behav 49:811–
820.
Margoliash D (1986) Preference for autogenous song by auditory neurons in a song system
nucleus of the white-crowned sparrow. J Neurosci 6:1643–1661.
Margoliash D, Konishi M (1985) Auditory representation of autogenous song in the song system
of white-crowned sparrows. Proc Natl Acad Sci U S A 82:5997–6000.
Marler P (1970) A comparative approach to vocal learning: Song development in white-crowned
sparrows. J Comp Physiol Psychol 71:1–25.
Masamizu Y, Tanaka YR, Tanaka YH, Hira R, Ohkubo F, Kitamura K, Isomura Y, Okada T,
Matsuzaki M (2014) Two distinct layer-specific dynamics of cortical ensembles during
learning of a motor task. Nat Neurosci 17:987–994.
Mello C V, Clayton DF (1994) Song-induced ZENK gene expression in auditory pathways of
songbird brain and its relation to the song control system. J Neurosci 14:6652–6666.
Mello C V, Kaser T, Buckner AA, Wirthlin M, Lovell P V (2019) Molecular architecture of the
zebra finch arcopallium. J Comp Neurol 527:2512–2556.
Mello C V, Vates GE, Okuhata S, Nottebohm F (1998) Descending auditory pathways in the
adult male zebra finch (Taeniopygia guttata). J Comp Neurol 160:137–160.
Metzger M, Jiang S, Braun K (1998) Organization of the dorsocaudal neostriatal complex: a
retrograde and anterograde tracing study in the domestic chick with special emphasis on
pathways relevant to imprinting. J Comp Neurol 395:380–404.
Miller-Sims VC, Bottjer SW (2012) Auditory experience refines cortico-basal ganglia inputs to
motor cortex via remapping of single axons during vocal learning in zebra finches. J
Neurophysiol 107:1142–1156.
Mooney R (1992) Synaptic basis for developmental plasticity in a birdsong nucleus. J Neurosci
12:2464–2477.
Mooney R, Konishi M (1991) Two distinct inputs to an avian song nucleus activate different
glutamate receptor subtypes on individual neurons. Proc Natl Acad Sci U S A 88:4075–
4079.
Morris D (1954) The Reproductive Behaviour of the Zebra Finch (Poephila Guttata), With
Special Reference To Pseudofemale Behaviour and Displacement Activities. Behaviour
6:271–322.
Murata A, Fadiga L, Fogassi L, Gallese V, Raos V, Rizzolatti G (1997) Object Representation in
the Ventral Premotor Cortex (Area F5) of the Monkey. J Neurophysiol 78:2226–2230.
93
Nick TA, Konishi M (2005) Neural auditory selectivity develops in parallel with song. J
Neurobiol 62:469–481.
Nottebohm F, Stokes TM, Leonard CM (1976) Central control of song in the canary, Serinus
canarius. J Comp Neurol 165:457–486.
Ölveczky B, Otchy T, Goldberg JH, Aronov D, Fee MS (2011) Changes in the neural control of
a complex motor sequence during learning. J Neurophysiol 106:386–397.
Olveczky BP, Andalman AS, Fee MS (2005) Vocal experimentation in the juvenile songbird
requires a basal ganglia circuit. PLoS Biol 3:e153.
Ota N, Gahr M, Soma M (2015) Tap dancing birds: The multimodal mutual courtship display of
males and females in a socially monogamous songbird. Sci Rep 5:6–11.
Papale AE, Hooks BM (2018) Circuit changes in motor cortex during motor skill learning.
Neuroscience 368:283–297.
Paterson AK, Bottjer SW (2017) Cortical inter-hemispheric circuits for multimodal vocal
learning in songbirds. J Comp Neurol:3312–3340.
Person AL, Gale SD, Farries MA, Perkel DJ (2008) Organization of the songbird basal ganglia,
including area X. J Comp Neurol 508:840–866.
Peters AJ, Chen SX, Komiyama T (2014) Emergence of reproducible spatiotemporal activity
during motor learning. Nat Lett 510:263–267.
Peters AJ, Lee J, Hedrick NG, O’neil K, Komiyama T (2017a) Reorganization of corticospinal
output during motor learning. Nat Neurosci 20:1133–1141.
Peters AJ, Liu H, Komiyama T (2017b) Learning in the Rodent Motor Cortex. Annu Rev
Neurosci 40:77–97.
Phan ML, Pytte CL, Vicario DS (2006) Early auditory experience generates long-lasting
memories that may subserve vocal learning in songbirds. Proc Natl Acad Sci U S A
103:1088–1093.
Prather JF (2013) Auditory signal processing in communication: Perception and performance of
vocal sounds. Hear Res 305:144–155.
Price PH (1979) Developmental determinants of structure in zebra finch song. J Comp Physiol
Psychol 93:260–277.
Pulvermüller F, Huss M, Kherif F, Moscoso del Prado Martin F, Hauk O, Shtyrov Y (2006)
Motor cortex maps articulatory features of speech sounds. Proc Natl Acad Sci 103:7865–
7870.
Rauschecker JP, Scott SK (2009) Maps and streams in the auditory cortex: nonhuman primates
illuminate human speech processing. Nat Neurosci 12:718–724.
Reiner A et al. (2004) Revised nomenclature for avian telencephalon and some related brainstem
nuclei. J Comp Neurol 473:377–414.
Rizzolatti G, Luppino G (2001) The cortical motor system. Neuron 31:889–901.
94
Roberts T, Gobes S, Murugan M, Ölveczky B, Mooney R (2012) Motor circuits are required to
encode a sensory model for imitative learning. Nat Neurosci 15:1454–1459.
Roberts TF, Hisey E, Tanaka M, Kearney MG, Chattree G, Yang CF, Shah NM, Mooney R
(2017) Identification of a motor-to-auditory pathway important for vocal learning. Nat
Neurosci.
Roberts TF, Tschida KA, Klein ME, Mooney R (2010) Rapid spine stabilization and synaptic
enhancement at the onset of behavioural learning. Nature 463:948–952.
Roper A, Zann R (2006) The Onset of Song Learning and Song Tutor Selection in Fledgling
Zebra Finches. Ethology 112:458–470.
Saint-Dizier H, Constantin P, Davies DC, Leterrier C, Lévy F, Richard S (2009) Subdivisions of
the arcopallium/posterior pallial amygdala complex are differentially involved in the control
of fear behaviour in the Japanese quail. Brain Res Bull 79:288–295.
Sanes JN, Donoghue JP (2000) Plasticity and Primary Motor Cortex. Annu Rev Neurosci
23:393–415.
Sawtell NB (2017) Neural Mechanisms for Predicting the Sensory Consequences of Behavior:
Insights from Electrosensory Systems. Annu Rev Physiol 79:381–399.
Schneider C, Devanne H, Lavoie BA, Capaday C (2002) Neural mechanisms involved in the
functional linking of motor cortical points. Exp Brain Res 146:86–94.
Schwartz AB, Kettner RE, Georgopoulos AP (1988) Primate motor cortex and free arm
movements to visual targets in three-dimensional space. I. Relations between single cell
discharge and direction of movement. J Neurosci 8:2913–2927.
Schwartz AB, Moran DW (2000) Arm trajectory and representation of movement processing in
motor cortical activity. Eur J Neurosci 12:1851–1856.
Shen L, Alexander GE (1997) Neural correlates of a spatial sensory-to-motor transformation in
primary motor cortex. J Neurophysiol 77:1171–1194.
Simpson HB, Vicario DS (1990) Brain pathways for learned and unlearned vocalizations differ
in zebra finches. J Neurosci 10:1541–1556.
Sober SJ, Wohlgemuth MJ, Brainard MS (2008) Central contributions to acoustic variation in
birdsong. J Neurosci 28:10370–10379.
Solis M, Doupe AJ (1997) Anterior forebrain neurons develop selectivity by an intermediate
stage of birdsong learning. J Neurosci 17:6447–6462.
Solis M, Doupe AJ (2000) Compromised neural selectivity for song in birds with impaired
sensorimotor learning. Neuron 25:109–121.
Solis MM, Brainard MS, Hessler NA, Doupe AJ (2000) Song selectivity and sensorimotor
signals in vocal learning and production. Proc Natl Acad Sci U S A 97:11836–11842.
Spiro J, Dalva M, Mooney R (1999) Long-range inhibition within the zebra finch song nucleus
RA can coordinate the firing of multiple projection neurons. J Neurophysiol 81:3007–3020.
95
Stark LL, Perkel DJ (1999) Two-stage, input-specific synaptic maturation in a nucleus essential
for vocal production in the zebra finch. J Neurophysiol 19:9107–9116.
Steeves JD, Sholomenko GN, Webster DMS (1987) Stimulation of the pontomedullary reticular
formation initiates locomotion in decerebrate birds. Brain Res 401:205–212.
Straka H, Simmers J, Chagnaud BP (2018) A New Perspective on Predictive Motor Signaling.
Curr Biol 28:R193–R194.
Sutton RS, Barto AG (1998) Reinforcement Learning: An Introduction. Cambridge,
Massachusettes.
Tanji J, Evarts E V. (1976) Anticipatory activity of motor cortex neurons in relation to direction
of an intended movement. J Neurophysiol 39:1062–1068.
Tchernichovski O, Nottebohm F, Ho C, Pesaran B, Mitra P (2000) A procedure for an automated
measurement of song similarity. Anim Behav 59:1167–1176.
Terpstra NJ, Bolhuis JJ, Den Boer-Visser AM (2004) An analysis of the neural representation of
birdsong memory. J Neurosci 24:4971–4977.
Theunissen FE, Amin N, Shaevitz SS, Woolley SMN, Fremouw T, Hauber ME (2004) Song
selectivity in the song system and in the auditory forebrain. Ann N Y Acad Sci 1016:222–
245.
Thorpe WH (1958) The learning of song patterns by birds, with especial reference to the song of
the Chaffinch Fringilla coelebs. IBIS Int J Avian Sci 100:535–570.
Thorpe WH (1961) Bird-Song, The Biology of Vocal Communication. Cambridge University
Press.
Tomaszycki ML, Adkins-Regan E (2005) Experimental alteration of male song quality and
output affects female mate choice and pair bond formation in zebra finches. Anim Behav
70:785–794.
Ullrich R, Norton P, Scharff C (2016) Waltzing Taeniopygia: Integration of courtship song and
dance in the domesticated Australian zebra finch. Anim Behav 112:285–300.
Valenzuela JI, Hasan SJ, Steeves JD (1990) Stimulation of the brainstem reticular formation
evokes locomotor activity in embryonic chicken (in ovo). Brain Res Dev Brain Res 56:13–
18.
Vates GE, Broome BM, Mello C V, Nottebohm F (1996) Auditory pathways of caudal
telencephalon and their relation to the song system of adult male zebra finches (Taenopygia
guttata). J Comp Neurol 366.
Vates GE, Vicario DS, Nottebohm F (1997) Reafferent thalamo-’cortical’ loops in the song
system of oscine songbirds. J Comp Neurol 380:275–290.
Vicario DS, Raksin JN (2000) Possible roles for GABAergic inhibition in the vocal control
system of the zebra finch. Neuroreport 11:3631–3635.
Vicario DS, Yohay KH (1993) Song-selective auditory input to a forebrain vocal control nucleus
96
in the zebra finch. J Neurobiol 24:488–505.
Volman S (1993) Development of neural selectivity for birdsong during vocal learning. J
Neurosci 13:4737–4747.
Wang L, Conner JM, Rickert J, Tuszynski MH (2011) Structural plasticity within highly specific
neuronal populations identifies a unique parcellation of motor learning in the adult brain.
Proc Natl Acad Sci 108:2545–2550.
Wang Y, Brzozowska-Prechtl A, Karten HJ (2010) Laminar and columnar auditory cortex in
avian brain. Proc Natl Acad Sci 107:12676–12681.
West MJ, King AP (1988) Female visual displays affect the development of male song in the
cowbird. Nature 334:244–246.
Wilcox RR, Serang S (2017) Hypothesis Testing, p Values, Confidence Intervals, Measures of
Effect Size, and Bayesian Methods in Light of Modern Robust Techniques. Educ Psychol
Meas 77:673–689.
Wild JM (1993) Descending projections of the songbird nucleus robustus archistriatalis. J Comp
Neurol 338:225–241.
Wild JM (2004) Functional neuroanatomy of the sensorimotor control of singing. Ann N Y Acad
Sci:1–25.
Wild JM, Karten HJ, Frost BJ (1993) Connections of the auditory forebrain in the pigeon
(Columba livia). J Comp Neurol 337:32–62.
Wild JM, Krützfeldt NEO (2012) Trigeminal and telencephalic projections to jaw and other
upper vocal tract premotor neurons in songbirds: Sensorimotor circuitry for beak
movements during singing. J Comp Neurol 520:590–605.
Williams H (2001) Choreography of song, dance and beak movements in the zebra finch
(Taeniopygia guttata). J Exp Biol 204:3497–3506.
Williams MN, Suthers RA (2000) Neural Pathways for Bilateral Vocal. 426:413–426.
Wilson SM, Saygin AP, Sereno MI, Iacoboni M (2004) Listening to speech activates motor areas
involved in speech production. Nat Neurosci 7:701–702.
Wu HG, Miyamoto YR, Castro LNG, Olveczky BP, Smith MA (2014) Temporal structure of
motor variability is dynamically regulated and predicts motor learning ability. Nat Neurosci
17:312–321.
Xiao L, Chattree G, Oscos FG, Cao M, Wanat MJ, Roberts TF (2018) A Basal Ganglia Circuit
Sufficient To Guide Birdsong Learning. Neuron:1–33.
Xu T, Yu X, Perlik AJ, Tobin WF, Zweig JA, Tennant K, Jones T, Zuo Y (2009) Rapid
formation and selective stabilization of synapses for enduring motor memories. Nature
462:915–919.
Yanagihara S, Yazaki-sugiyama Y (2016) Auditory experience-dependent cortical circuit
shaping for memory formation in bird song learning. Nat Commun:1–11.
97
Yang G, Pan F, Gan WB (2009) Stably maintained dendritic spines are associated with lifelong
memories. Nature 462:920–924.
Yip ZC, Miller-Sims VC, Bottjer SW (2012) Morphology of axonal projections from the high
vocal center to vocal motor cortex in songbirds. J Comp Neurol 520:2742–2756.
Yu AC, Margoliash D (1996) Temporal hierarchical control of singing in birds. Science (80- )
273:1871–1875.
Yuan RC, Bottjer SW (2019) Differential developmental changes in cortical representations of
auditoryvocal stimuli in songbirds. J Neurophysiol 121:530–548.
Zeier H, Karten HJ (1971) The archistriatum of the pigeon: Organization of afferent and efferent
connections. Brain Res 31:313–326.
Abstract (if available)
Abstract
Goal-directed motor skill learning underlies our ability to acquire and flexibly perform countless complex behaviors that allow us to interact with our environment. Increasing evidence across taxa implicates motor cortex in not only execution of voluntary movements but also acquisition of motor skills. Juvenile songbirds learn to produce stereotyped vocalizations from an adult tutor during a sensitive period of development and can thus serve as a powerful model for investigating motor cortical contributions to skill learning. Extracellular recordings were made during song playback in anesthetized juvenile and adult zebra finches (Taeniopygia guttata) in motor cortical regions RA (robust nucleus of the arcopallium), AId (dorsal intermediate arcopallium), and RA-cup, each of which is well-situated to integrate different sources of information to contribute to vocal learning: RA neurons drive vocal motor output, AId is an adjoining region that receives multi-modal inputs and whose projections converge with basal ganglia circuitry in a dorsal thalamic zone, and RA-cup surrounds RA and receives inputs from regions of primary and secondary auditory cortex. Strong developmental differences in neural selectivity were found within RA, but not RA-cup. Juvenile RA neurons were broadly responsive to multiple songs, but preferred juvenile over adult vocal sounds
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Neural mechanisms of sensorimotor learning in cortico-basal ganglia pathways
PDF
Learning repetition rules and non-adjacent dependencies from human actions in nine-month-old infants
PDF
Neural circuits underlying the modulation and impact of defensive behaviors
PDF
Brain and behavior correlates of intrinsic motivation and skill learning
PDF
The behavioral and neural bases of tactile object localization
PDF
Contextual modulation of sensory processing via the pulvinar nucleus
PDF
The brain and behavior of motor learning: the what, how and where
PDF
Mapping multi-scale connectivity of the mouse posterior parietal cortex
PDF
Neural circuits control and modulate innate defensive behaviors
PDF
The neural correlates of skilled reading: an MRI investigation of phonological processing
PDF
Synaptic circuits for information processing along the central auditory pathway
PDF
Physiology of the inner ear: the role of the biophysical properties of spiral ganglion neurons in encoding sound intensity information at the auditory nerve
PDF
Value-based decision-making in complex choice: brain regions involved and implications of age
PDF
Exploring sensory responses in the different subdivisions of the visual thalamus
PDF
Exploiting novel transport properties of adeno-associated virus for circuit mapping and manipulation
PDF
Synaptic integration in dendrites: theories and applications
PDF
Spatial and temporal precision of inhibitory and excitatory neurons in the murine dorsal lateral geniculate nucleus
PDF
Subnetwork organization of the superior colliculus and visual system in the mouse brain
PDF
Characterizing the hippocampal synaptic and sleep abnormalities of a mouse model of human chromosome 16p11.2 microdeletion
PDF
The structure of kinematic variability during hopping: a window into motor control factors in overuse injury
Asset Metadata
Creator
Yuan, Rachel
(author)
Core Title
Motor cortical representations of sensorimotor information during skill learning
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Neuroscience
Publication Date
02/04/2020
Defense Date
01/13/2020
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
motor cortex,motor learning,OAI-PMH Harvest,sensorimotor integration,songbird,vocal learning
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Hires, Samuel (
committee chair
), Bottjer, Sarah (
committee member
), Kalluri, Radha (
committee member
), Mintz, Toben (
committee member
), Zevin, Jason (
committee member
)
Creator Email
rachelcyuan@gmail.com,rachelyu@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-265232
Unique identifier
UC11673830
Identifier
etd-YuanRachel-8143.pdf (filename),usctheses-c89-265232 (legacy record id)
Legacy Identifier
etd-YuanRachel-8143.pdf
Dmrecord
265232
Document Type
Dissertation
Rights
Yuan, Rachel
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
motor cortex
motor learning
sensorimotor integration
songbird
vocal learning