Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2013 From Songs to Synapses, Ion Channels and Mathematical Modeling Arij Daou

Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected]

THE FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

FROM SONGS TO SYNAPSES, ION CHANNELS AND MATHEMATICAL MODELING

By

ARIJ DAOU

A Dissertation submitted to the Department of Mathematics in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Degree Awarded: Summer Semester, 2013 Arij Daou defended this dissertation on May 28, 2013. The members of the supervisory committee were:

Richard Bertram Professor Directing Dissertation

Pamela Ryan University Representative

Frank Johnson Committee Member

Richard Hyson Committee Member

Wei Wu Committee Member

Giray Okten Committee Member

The Graduate School has verified and approved the above-named committee members, and certifies that the dissertation has been approved in accordance with university requirements.

ii

This dissertation is dedicated to my parents.

iii

ACKNOWLEDGEMENTS

First and foremost, I would like to thank my advisor, Dr. Richard Bertram; for being a model scientist and for all that he has done for me. Not only had he been an academic advisor where I was able to pursue with him two things I loved: birdsong research and computational modeling; he was also a second father who mentored me in that wonderful spirit of encouragement and generosity that made my journey at FSU rewarding and enjoyable. I will miss our meetings. His encouragement and support throughout my graduate career will never be forgotten. I also thank Dr. Frank Johnson and Dr. Richard Hyson for their guidance and support. I thank them for giving me the opportunity to work in their labs and for giving me an encouraging word when I needed one. I learned a tremendous amount of expertise in their labs from subject manipulation, to brain slicing, electrophysiological recording, pharmacological manipulations, intracellular staining, electrophysiological data acquisition and analysis, electrical brain lesioning, and retrograde and anterograde dye injections. Besides the academic guidance and support, I thank Dr. Johnson for the intellectual stimulation during our meetings and when discussing various topics in the lab. I enjoyed these moments and they will be truly missed. I also would like to acknowledge the Department of Mathematics for education and in specific I thank Dr. Bettye Anne Case for her great support and mentoring especially during the first year when I arrived to the department. I am so grateful and tremendously indebted to the department as well as to NFS (IOS-1146607) and NIH (DC002035) for the financial support they provided me. I’ve always felt that the least I could to pay them back partially is to work hard and deliver good research, and I truly hope I was able to do that. Also, I would like to express my gratitude and thankfulness to Dr. Jack Quine, Dr. Pamela Ryan, Dr. Giray Okten and Dr. Wei Wu, my dissertation committee members. I appreciate the time and efforts they made to be available, as well as the suggestions and comments they made about this research. Also during my years at FSU, I was able to form enduring friendships with fellow graduate students and lab mates. I would like to thank each and every one of them for being great friends and making this journey more enjoyable.

iv

And finally, I have been fortunate in having a family who are (and will always be) an abiding source of strength and support for me. No amount of gratitude can measure up to the debt I owe my parents. Their love and support have been always unconditional and transcending. There is no greater satisfaction for me than that derived from making them happy and proud. This thesis is dedicated to them.

v

TABLE OF CONTENTS

List of Tables ...... ix

List of Figures ...... x

Abstract ...... xv

1. INTRODUCTION ...... 1

2. BACKGROUND AND SIGNIFICANCE ...... 4 2.1 What is a Birdsong? ...... 4 2.2 Why Study Songbirds? ...... 6 2.3 Vocal Learners among Species ...... 9 2.4 Psychoacoustics Analysis of Birdsong ...... 11 2.5 Basic Units of Song ...... 12 2.6 The Song Learning Process ...... 13 2.7 The Song System: Anatomy ...... 19 2.8 The Song System: Functions and Roles ...... 22 2.8.1 Effects of Deafening ...... 23 2.8.2 Functions and Roles of the AFP ...... 23 2.8.3 Functions and Roles of the VMP ...... 25

3. BEHAVIORAL LEVEL: SONGSEQ ...... 32 3.1 Introduction ...... 32 3.2 Methods ...... 33 3.2.1 Identification of Syllables Using a Template ...... 34 3.2.2 Sequencing the Data of the Target Files ...... 42 3.2.3 Syntax Identification ...... 42 3.2.4 Kullback-Leibler (KL) Distance Analysis ...... 43 3.2.5 Transition Entropy Analysis ...... 48 3.3 Results ...... 49 3.3.1 Quantifying the Effects of HVC Microlesions on Syllable Sequence ...... 49 3.3.2 Developmental Changes in Syllable Sequence ...... 53 3.4 Robustness ...... 64 3.5 Discussion ...... 68

vi

4. CELLULAR LEVEL: CHARACTERIZATION OF HVC NEURONS ...... 70 4.1 Introduction ...... 70 4.2 Materials & Methods ...... 71 4.2.1 Brain Slice Electrophysiology ...... 71 4.2.1.1 Slice preparation ...... 71 4.2.1.2 Whole cell recording ...... 73 4.2.1.3 Pharmacological manipulations ...... 74 4.2.1.4 Electrophysiological identification of neurons ...... 74 4.2.1.5 In vivo injections of retrograde tracers into Area X and RA ...... 75 4.2.2 Computational Modeling ...... 80 4.2.2.1 Model HVC cells ...... 80 4.2.2.2 Voltage-gated ionic currents ...... 81 4.2.2.3 Low-voltage activated T-type Ca2+ current ...... 84 4.2.2.4 Ca2+-dependent K+ current ...... 85 4.2.2.5 Na+-dependent K+ current ...... 87 4.2.2.6 Hyperpolarization-activated inward current ...... 88 4.3 Results ...... 92 4.3.1 HVCX Neurons ...... 92 4.3.1.1 Response to applied current ...... 92 4.3.1.2 H current and inward rectification in HVCX neurons ...... 97 4.3.1.3 Actions of Ih and ICa-T in HVCX neurons ...... 100 4.3.1.4 ISK and IKNa in HVCX neurons ...... 102 4.3.2 HVCRA Neurons ...... 105 4.3.2.1 Response to applied current ...... 105 4.3.2.2 IA current in HVCRA neurons ...... 109 4.3.2.3 ISK current in HVCRA neurons ...... 112 4.3.3 HVCINT Neurons ...... 115 4.3.3.1 Response to applied current ...... 115 4.3.3.2 H current and inward rectification in HVCINT neurons ...... 119 4.3.4 Diversity of Action Potentials in HVC Neurons ...... 122 4.3.4.1 Action potentials of HVCINT neurons ...... 123 4.3.4.2 Action potentials of HVCRA neurons ...... 127 4.3.4.3 Action potentials of HVCX neurons ...... 130 4.4 Discussion ...... 134

5. NETWORK LEVEL: HVC MICROCIRCUITRY ...... 139 5.1 Introduction ...... 139 5.2 Methods ...... 140 5.2.1 HVC Cells ...... 141 5.2.2 Synaptic Connectivity ...... 142 5.2.3 Current Clamp Neuronal Firing Patterns and Rebound Bursting ...... 142 5.3 Results ...... 144 5.3.1 Network Architecture 1 ...... 145 5.3.2 Network Architecture 2 ...... 150 5.3.3 Network Architecture 3 ...... 152

vii

5.3.4 Network Architecture 4 ...... 154 5.3.5 Network Architecture 5 ...... 157 5.4 Effects of Temperature on HVC Networks ...... 158 5.5 Discussion ...... 163

6. CONCLUSIONS ...... 166 6.1 Summary ...... 166 6.2 Future Work ...... 167 6.3 Conclusion ...... 168

APPENDICES ...... 169

A. SOURCE CODES ...... 169 A1. SongSeq ...... 169 A2. HVC Models ...... 169 A3. Synchronization Index ...... 170

B. LIST OF ABBREVIATIONS ...... 171

REFERENCES ...... 173

BIOGRAPHICAL SKETCH ...... 195

viii

LIST OF TABLES

4.1 Parameter values used in all simulations ...... 89

4.2 Parameter values that vary among neuron types ...... 91

4.3 Experimental characterization of the three types of HVC neurons ...... 94

5.1 Parameter values used in all network models ...... 143

5.2 Parameter values that vary among networks ...... 143

ix

LIST OF FIGURES

2.1 An adult male zebra finch (left) and a female zebra finch (right) ...... 10

2.2 Sonogram (or spectrogram) showing a song bout produced by an adult male zebra finch ...... 13

2.3 Sonograms showing song bouts produced by a male zebra finch during subsong (A), plastic song (B), and crystallized song (C) (Johnson and Sellix 2000) ...... 16

2.4 Duration versus pitch scatter plots for juvenile vocal development (Thompson et al. 2011) ...... 18

2.5 The song system pathways ...... 20

2.6 Spike raster plot of ten HVCRA neurons and two HVC interneurons recorded in one bird during singing ...... 28

2.7 Representative sonograms recorded from a single bird with HVC heated and cooled, showing percentage song dilation relative to control ...... 30

3.1 Designation of template spreadsheet and selection of two acoustic features ...... 35

3.2 Sonogram of a typical song of a bird that received bilateral HVC microlesions ...... 35

3.3 A screen shot of SongSeq’s frame where syllable identification is processed ...... 36

3.4 Syllable Identification: Using mouse clicks, the instances of each syllable in Fig. 3.3 are painted onto the scatter plot ...... 37

3.5 A screenshot of SongSeq’s frame showing the selection of the second pair of acoustic features, and the selection of B and I to better discriminate them on the scatter plot of the new pair of features ...... 38

3.6 Second Step of Syllable Identification: Color-coded data points on the scatter plot of the second pair of acoustic features chosen in Fig. 3.5, where colors are coded based on the painting done in the first step of identification ...... 39

3.7 Second Step of Syllable Identification: Syllables B and I are painted in the new feature scatter plot ...... 40

x

3.8 The scatter plots of postoperative days 1 (A) and 3 (B) are superimposed with the painted clusters of the template ...... 41

3.9 Syllable transition probabilities on preoperative day 1 ...... 44

3.10 Syllable transition probabilities on postoperative day 1 ...... 44

3.11 Syllable transition probabilities on postoperative day 2 ...... 45

3.12 Syllable transition probabilities on postoperative day 3 ...... 45

3.13 Syllable transition probabilities on postoperative day 8 ...... 46

3.14 Syllable transition probabilities on postoperative day 12 ...... 46

3.15 Quantification of syllable transition distributions: KL distance measure quantifying the dissimilarity in song sequence between the Pre and Post days ...... 47

3.16 Quantification of syllable transition distributions: Entropy analysis quantifying the spread of the syllable transition distributions for each day of singing ...... 48

3.17 Linearity, consistency and stereotypy scores for postoperative day 12 ...... 52

3.18 Comparison of the average scores over all preoperative and postoperative days of singing...... 53

3.19 Syllable Identification for a developing bird: The scatter plot of two selected acoustic features (syllable duration versus mean FM) for a day of singing from the bird when adult ...... 54

3.20 Syllable Identification for a developing bird: The data points on the scatter plot form five clusters indicating five different motif syllables for this bird that form the template ...... 55

3.21 Syllable Identification for a developing bird: The scatter plot for week 1 is superimposed with the painted clusters of the template ...... 56

3.22 Syllable Identification for a developing bird: The scatter plot for week 6 is superimposed with the painted clusters of the template...... 57

3.23 Syllable transition probabilities are shown during a day when the bird is an adult ...... 58

3.24 Syllable transition probabilities are shown during a day of week 1 ...... 58

3.25 Syllable transition probabilities are shown during a day of week 2 ...... 59

3.26 Syllable transition probabilities are shown during a day of week 3 ...... 59

xi

3.27 Syllable transition probabilities are shown during a day of week 6 ...... 60

3.28 Syllable transition probabilities are shown during a day of week 9 ...... 60

3.29 During weeks 1 through 3, the K-L distance is large since the juvenile song is quite different from the adult song ...... 61

3.30 The entropy value on W1 is smaller compared with adult (Fig. 3.23 versus Fig. 3.22) ....62

3.31 Average linearity, consistency and stereotypy scores for the juvenile versus the adult songs ...... 63

3.32 Robustness of the results to variations in the number of notes ...... 64

3.33 The entropy analysis obtained after removing 45% of the song bouts of bird 1exhibits a similar time course of to that from the same bird when all song bouts were included (Fig. 3.16) ...... 65

3.34 The average scores obtained after removing 45% of the song bouts of bird 1 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig. 3.18) ...... 65

3.35 Robustness of the results to variations in the number of notes: SongSeq was used on the developing bird data sets ...... 66

3.36 The entropy analysis obtained after removing 45% of the song bouts of bird 2 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig. 3.30) ...... 67

3.37 The average scores obtained after removing 45% of the song bouts of bird 2 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig. 3.31) ...... 67

4.1 HVC in the slice ...... 72

4.2 An electrode patched onto an HVC cell in the slice ...... 73

4.3 Anatomical confirmation of physiological classification of HVCX neurons ...... 77

4.4 Anatomical confirmation of physiological classification of HVCRA neurons ...... 78

4.5 Recording from fluorescently labeled HVC neurons ...... 79

4.6 Steady-state activation functions for the fast gating variables (A), the slow gating variables (B, solid lines) and the steady-state inactivation functions (B, dotted lines) ....83

xii

2+ + 4.7 Ca dependence of the SK current gating function (A) and Na dependence of the KNa current (B) ...... 86

4.8 Firing properties of an actual X-projecting neuron ...... 93

4.9 Electrical recordings from retrogradely labeled HVC neurons ...... 95

4.10 Firing properties of a model X-projecting neuron ...... 96

4.11 Frequency-current relationship in X-projecting neurons ...... 97

4.12 Inward rectification in HVCX neurons ...... 98

4.13 Ih current in HVCX neurons ...... 99

4.14 Ih and ICa-T mechanisms in HVCX neurons ...... 101

4.15 ISK current in model HVCX neuron ...... 103

4.16 ISK current in actual HVCX neurons ...... 104

4.17 Firing properties of an actual RA-projecting neuron ...... 106

4.18 Firing properties of a model RA-projecting neuron ...... 108

4.19 Frequency-current relationship in RA-projecting neurons ...... 109

4.20 The effect of IA current in model HVCRA neuron ...... 110

4.21 The effect of IA current in actual HVCRA neurons ...... 111

4.22 ISK current in model HVCRA neuron ...... 113

4.23 ISK current in actual HVCRA neurons ...... 114

4.24 Firing properties of an actual HVC interneuron ...... 116

4.25 Most HVCINT neurons in the slice exhibited spontaneous firing (n=18) ...... 117

4.26 Firing properties of a model HVC interneuron ...... 118

4.27 Frequency-current relationship in HVC interneurons ...... 119

4.28 Inward rectification in HVCINT neurons ...... 120

xiii

4.29 Ih current in HVCINT neurons ...... 121

4.30 Action potentials of Type I and Type II HVC interneurons ...... 124

4.31 Difference in the magnitude of dV/dt between Type I and Type II HVC interneurons ..125

4.32 Phase plane plot of the model HVCINT neuron ...... 126

4.33 Action potentials of HVCRA neurons ...... 128

4.34 Phase plane plots of three different HVCRA neurons (A-C) ...... 129

4.35 Phase plane plot of the model HVCRA neuron ...... 130

4.36 Action potentials of HVCX neurons ...... 131

4.37 Phase plane plots of four different HVCX neurons (A-D) ...... 132

4.38 Phase plane plot of the model HVCX neuron ...... 133

4.39 Schematic models depicting the ionic currents present in each of three HVC model neurons after calibration to our data ...... 135

5.1 Synaptic interactions between the three classes of HVC neuronal populations ...... 140

5.2 Arrangement of prototype network architecture one (A) ...... 146

5.3 HVC spiking patterns of network architecture one ...... 148

5.4 Zoomed version of HVC spiking patterns of network one ...... 149

5.5 Arrangement of prototype network architecture two (A) ...... 151

5.6 Arrangement of prototype network architecture three (A) ...... 153

5.7 Arrangement of prototype network architecture four (A) ...... 155

5.8 Arrangement of prototype network architecture five (A) ...... 158

5.9 Effects of temperature on network architecture one activity ...... 160

5.10 Effects of temperature change on HVCRA neurons’ firing pattern ...... 162

xiv

ABSTRACT

Since the scientific study of birdsong began in the late 1950s, songbirds have emerged as impressive neurobiological models for aspects of human verbal communication because they learn to sequence their song elements, analogous, in many ways, to how humans learn to produce spoken sequences with syntactic structure. Thus, determining how spoken language evolved is more likely to become clearer with concerted efforts in researching songbirds. Some of the most fundamental questions in neuroscience are pursued through the study of songbirds. How does the brain generate complex sequential behaviors? How do we learn to speak? How do humans learn various behaviors by observing and imitating others? Where are the “prime movers” that control behavior? Which circuits in the brain control the order in which motor gestures of a learned behavior are generated? Among all these questions, of particular interest to us is the question of sequential behavior. Understanding the neural mechanisms that underlie sequential behavior and imitative learning is the holy grail of the field. The birdsong provided us with a uniquely powerful model for tackling this question in a system where the brain structures responsible for its generation are well known. We pursued the study of sequential neural activity in songbirds on three levels: behavioral, cellular and network. On the behavioral level, we developed a computational tool for automated, quantitative syllable-level analysis of bird song syntax. This tool aids songbird researchers and fanciers in comparing and quantifying the syntactic structure of songs produced by a bird prior to and after a manipulation such as ablation of brain region or infusion of pharmacological agents, in addition to several other purposes. As we will discuss later, this syntactic structure is highly stereotyped in songbirds and driven by neurons firing in sequential order in particular regions of the songbird’s brain. On the cellular level, the telencephalic nucleus HVC (proper name) within the songbird analogue of the mammalian pre-motor cortex is situated at a critical point in the pattern- generating premotor brain circuitry of oscine songbirds. This nucleus is of extreme importance to the songbird and produces stereotyped instructions through the motor pathway leading to precise,

xv learned vocalization by songbirds. HVC contains three populations of neurons that are interconnected, with specific patterns of excitatory and inhibitory connectivity. Characterizing the neurons in HVC is a very important requirement for decoding the neural code of the birdsong. We performed whole-cell current clamp recordings on HVC neurons within brain slices to examine their intrinsic firing properties and determine which ionic currents are responsible for their characteristic firing patterns. We also developed conductance-based models for the different neurons and calibrated the models using data from our brain slice work. These models were then used to generate predictions about the makeup of the ionic currents that are responsible for the different responses to stimuli. These predictions were then tested and verified in the slice using pharmacological manipulations. Our results are an improved characterization of the HVC neurons responsible for song production in the songbird which are the key ingredients in understanding the HVC network. We then developed prototype neural architectures of the HVC that can produce the patterns of sequential neural activity exhibited by the three types of HVC neurons during singing. Our networks consist of microcircuits of interconnected neurons which are active during different syllables of the song. The various networks that we consider assign different roles to each of the HVC neurons types in the production of the sequential activity pattern, and show great flexibility in the connectivity patterns among the neuron types. The model networks developed provide key insights into how the different types of HVC neurons can be used for sequence generation. The significance of the work presented in this dissertation is that it helps elucidate the neural mechanisms behind HVC activity. The in vitro studies we performed in brain slices and the models we developed provide critical pieces to the puzzle of sequential behavior.

xvi

CHAPTER 1

INTRODUCTION

Songbirds are feathered vocalists which, like their human counterparts, learn to pattern their respiration to produce an astonishing variety of vocal sounds (Mooney and Spiro 1997). These range from short, monosyllabic calls, to some of the longest and most complicated sounds known to science: their songs (Catchpole and Slater 1995). Songbirds are among the few animals that learn their communication sounds, as humans do. Because they share a large number of parallels with humans, they arose as an attractive and excellent model system for studying the behavioral and neurobiological mechanisms that underlie the learning and production of vocal communication signals. Perhaps another reason why birdsong has captivated humans for so many centuries is that we “hear the world” in very similar ways. In fact, after human language, birdsong is arguably the most acoustically complex and diverse communication signal known to date (Knudsen and Gentner 2010). Moreover, although people had been interested in bird song for a very long time, particularly in music and literature, its detailed scientific study is a comparatively recent phenomenon. Our overall aim was to understand the neural mechanisms of vocal production and learning by providing a quantitative description of the physiological variables that control vocal performance in a particular species of songbirds, the zebra finch. Attempting to study this topic is deceptively difficult and raises many questions on the way including what is a birdsong, how does a birdsong develop, how is the “message” being transmitted, when do zebra finches sing, what material do young birds learn, and most importantly, what is the neural code that deciphers this information between the brief bursts of neural activity. We used measurements of acoustic data, conducted electrophysiological experiments, and developed theoretical and computational frameworks in an attempt to elucidate and unveil the detailed dynamics of the vocal production and learning processes in the zebra finch. The research presented here relates to ongoing work in developmental biology, ethology, linguistics, cognitive psychology, mathematics and neuroscience. The body of the dissertation handles the study we performed on songbirds over three levels: behavioral level, cellular level, and network level.

1

Chapter 2 provides a brief introduction to the theory, terminology, anatomy, techniques and experimental manipulations used in the scientific study of birdsong. We first present a scientific explanation of birdsong, and then discuss the motivations behind the study of songbirds as well as the reasons why songbirds are considered as uniquely powerful models for studying imitative vocal learning. Next, we introduce the songbird we study (the zebra finch), explain the learning process that the zebra finch and other songbirds undergo during development, and discuss the psychoacoustics analysis of the birdsong including the basic units that constitute the song. Finally we present the anatomy of the song system and elaborate on the functionalities and roles of the major brain nuclei in the song system. These functions are unveiled through various experimental manipulations including lesioning studies, pharmacological inactivation, electrophysiological recordings from individual neurons, and several other experiments that will be discussed. A behavioral level study of songbirds will be presented in Chapter 3. Songbird researchers are quite often interested in the quantification, analysis and comparison of the bird’s songs produced prior to and after a certain experimental manipulation (such as ablation of a brain region or infusion of pharmacological agents), or through the course of the songbird’s development, or even to assess the degree of individual variation in the song structure across a population of birds. Since songbirds tend to sing hundreds to thousands of songs each day, the quantification and analysis process is a very tedious task without the help of automated tools. We present in Chapter 3 a software tool for automated, high throughput, quantitative syllable-level analysis of the birdsongs. We will discuss the primary advantages of our tool and the ease and effectiveness it provides in quantifying and comparison of birdsongs from one day of singing with one or more other days of singing. The work presented in this chapter is published in the Journal of Neuroscience Methods (Daou et al. 2012). In Chapter 4, we move to the cellular level, and discuss the intrinsic neuronal properties of the song region HVC of the zebra finch brain. HVC is analogous to the mammalian premotor cortex and the reasons why we are interested in it are discussed in details in Chapter 2. We basically performed whole-cell current clamp recordings on HVC neurons within brain slices to examine their intrinsic firing properties and determine which ionic currents are responsible for their characteristic firing patterns. We also developed conductance-based models for the different neurons and calibrated the models using data from our brain slice work. These models were then

2

used to generate predictions about the makeup of the ionic currents that are responsible for the different responses to stimuli. These predictions were then tested and verified in brain slices using pharmacological manipulations. The results we present in this chapter are published in the Journal of Neurophysiology (Daou et al. 2013). HVC model neurons are then connected into circuits in Chapter 5. In this network level study, we present five prototype neural architectures of the HVC that can produce the patterns of neural activity exhibited by the neuronal populations inside HVC during singing. Our networks consist of microcircuits of interconnected neurons which are active during different syllables of the song. The model networks provide insights into how the different types of HVC neurons can be used for sequence generation. Finally, Chapter 6 concludes the dissertation and present topics for future research.

3

CHAPTER 2

BACKGROUND AND SIGNIFICANCE

2.1 What is a Birdsong?

Experts in the fields of human speech and birdsong have always commented on the parallels between the two in terms of communication and their development (Doupe and Kuhl 1999; Marler 1970a). How can we define a birdsong scientifically and which aspects of birdsong can be used to understand and compare with human speech? Just like human speech, birdsong is a complex learned motor behavior driven by a specific set of premotor brain nuclei. These nuclei have a well-studied anatomy (Bolhuis et al. 2010). Birdsong, however, is not a language in the sense that human language is a language. First, human language is a highly elaborated signaling system that can be written, spoken, or signed. In all of these forms, humans developed rules of grammar so that the string of words they communicate through the language would convey meaning. Scientifically, human language is usually analyzed by studying semantics (the study of meaning), syntax (the order of words), prosody (the pitch, rhythm, and tempo), lexicon (words), or phonology (the study of phonemes, the smallest linguistic building block) (Doupe and Kuhl 1999). Human speech on the other hand refers to learned voice production that exhibits semantic content. Birdsong, however, lacks the rich structure and semantics of human vocal gestures, and for this reason it may be more analogous to speech than language. In other words, unlike humans, the songs of birds do not exhibit deep hierarchical relationships that convey sophisticated meanings, and the way the song elements of a birdsong are structured are thought not to generate new meanings (Berwick et al. 2011; Marler 1970a; 2000; Petkov and Jarvis 2012). On the other hand, humans can change the meaning of expressions by changing the syntactic organization of the units (Hurford 2012; Tallerman 2011). Therefore, only phonology, syntax, and perhaps prosody (in the sense that it involves control of frequency, timing, and amplitude) are the levels at which birdsong can be most usefully used to study human language, and particularly the spoken speech.

4

Nevertheless, a birdsong is more than a signal; it is an integral component of the songbird’s communication system (White 2010). Songbirds use their songs in mating and intrasexual contexts, most commonly by males to attract females and to repel rival males in fights over territorial displays (Beecher and Brenowitz 2005; Searcy and Andersson 1986). In ornithology, bird songs are often distinguished from shorter sounds, which are referred to as “calls”. The distinction between songs and calls is based upon length and context. Songs are longer and more complex and are associated with courtship and mating, while calls are produced by birds to serve functions as food begging, alarms or keeping members of a flock in contact. Moreover, irrespective of the underlying interactions that lead to song production (and which will be discussed later), at the end the song is nothing but a series of sound waves. The acoustic units of these sound waves are generated by motor patterns involving the coordinated activity of several muscles, most importantly those of the respiratory system and the syrinx. In both songbirds and humans, sounds are produced by the flow of air during expiration through a vocal system. In humans, the process is relatively well understood: Air from expiration generates a complex waveform at the vocal folds, and the components of this waveform are subsequently modified by the rest of the vocal tract (including the mouth, tongue, teeth, and lips) (Doupe and Kuhl 1999). The vocal tract acts as a filter, creating concentrations of energy at particular frequencies, called formant frequencies. With songbirds, the vocal skills are derived from the unusual structure of their powerful vocal equipment, the syrinx. The syrinx is the equivalent of the human sound box and it is a bony structure at the bottom of the trachea (unlike the larynx at the top of the mammalian trachea) (Immelmann 1969). The syrinx and a surrounding air sac resonate to sound waves that are made by membranes past which the bird forces air from the lungs. As air flows past the syringeal cords they are set into periodic oscillations, thus generating the sounds we hear and we call birdsong. The bird controls the pitch by changing the tension on the membranes and controls both pitch and volume by changing the force of exhalation. Moreover, it can control the two sides of the trachea independently, which is how some species can produce two notes at once (Nottebohm 1971; Nottebohm and Nottebohm 1976). Therefore, irrespective of the neuronal interactions that lead to song production (and which will be discussed later), the song is nothing but a series of sound waves, where sound waves are alternating changes in the pressure of the medium, which in the case of bird songs is always air.

5

2.2 Why Study Songbirds?

Biological research studies specific model organisms or animals that are well-suited for the in-depth study of a particular question or system. For example, Hodgkin and Huxley worked with the squid to study how nerves conduct impulses because squids have large neurons that were easy to study. As a result, the backbones of our modern knowledge of the conduction of nerve impulses in humans came from the research they did on squids. Similarly, the work done on fruit flies provided most of what we know about genes and pattern development in humans. Likewise, songbirds offer neuroscientists with lots of advantages and their brains are accessible to invasive research techniques that, for obvious ethical reasons, are not available in human research (Williams 2004). We are interested in songbirds and consider them as excellent models for our research questions for several reasons, as discussed below. First, the birdsong is a series of highly stereotyped complex signals that are ordered and executed by particular sets of neurons firing in sequence in well-known areas of the bird’s brain. These motor sequences are referred to as “temporal sequences”, and they are responsible for most of what humans and other organisms do. For instance, when we learn the alphabet, we do not learn 26 uncorrelated symbols; what we learn in fact is a temporal sequence because anyone of us can say the letters of the alphabet in just a few seconds, but only if we say them in one particular order – the order in which we learned them. This ability of the brain to learn a sequence of states and execute them rapidly underlies not only our ability to say the alphabet, but almost everything we do, like speaking, walking, playing music and even thinking. So understanding how the brain learns and produces complex sequences is one of the touchstone questions in neuroscience (Fee et al. 2004; Glaze and Troyer 2007), and despite the fundamental importance of understanding temporal sequences, very little is known about the biophysical and circuit mechanisms that underlie their generation and learning. For this reason, zebra finch songs have several characteristics that make it an ideal model system for understanding temporal motor sequence learning and production (Glaze and Troyer 2006; Long and Fee 2008). As a matter of fact, the motor sequence that controls the birdsong is one of the most stereotyped and precisely time-locked temporal sequences known to-date in nature.

6

Second, the acquisition of song in birds provides important insights regarding learning of speech in humans (Doupe and Kuhl 1999). In particular, learning is extremely essential to both birdsong and speech. Birds cannot sing normally and infants cannot speak, if they are not being “tutored” by adults of the species. This common behavior motivates the search for the neural foundations that control it, and the songbird provides the best platform and most tractable model system to pursue such a research question. Moreover, our knowledge of the neural substrates that control vocal learning and communication in humans and birds is increasing dramatically (Bolhuis et al. 2010; Doupe and Kuhl 1999; Mooney 2009a; Wild 1997b), and our recognition that the avian brain is based on the same organizational schema as that of mammals is growing (Doupe et al. 2005; Farries et al. 2005; Reiner et al. 2004; Williams 2004). An article by the Avian Brain Nomenclature Consortium (Jarvis et al. 2005) has highlighted the homologies and analogies between the avian and mammalian brain, and it provided insights to the extent which the songbird brain could help humans unveil the mysterious labyrinths of vocal learning and production. Third, critical and extremely sensitive periods for vocal learning are evidenced in both songbirds and humans. Neither songbirds nor babies appear to learn their communicative signals well except at specific periods in their lives. This raises the questions of what causes the brain to alter its ability to learn over time (generally referred to as plasticity), and whether the causes are the same in human infants and songbirds. Moreover, if the plasticity of the brain changes over the life cycle, what are the underlying neural mechanisms that control its changing ability to learn? (Doupe and Kuhl 1999). Because songbirds exhibit remarkable brain plasticity during their life that exemplifies that in humans, they provide an excellent model for such research questions where answers remain largely unknown. Fourth, the acquisition of complex motor sequences in humans, such as hitting a baseball or playing the piano, are not innately determined, but they are acquired through practice and can be thought of as reinforcement learning (Olveczky et al. 2005). This learning process requires the exploration of a range of motor actions, plus the evaluation of the resulting performance, reinforcing motor programs that lead to improved outcomes. Similarly, juvenile songbirds explore a large range of vocalizations by continuously varying their song, utilizing auditory feedback to improve their performance. Currently, little is known about the neural mechanisms of motor learning that is shaped by an evaluation system. Songbirds however emerge as an

7

excellent model system to answer how reinforcement learning could drive the development of complex motor sequences (Fee and Goldberg 2011). Fifth, the song system has contributed greatly to our modern understanding of neuron generation (neurogenesis) that plays a key role in learning and memory, stress, addiction, depression, neurodegenerative diseases and many other processes. Joseph Altman was the first person to suggest that new neurons were born in the adult mammalian brain (Altman 1962), but his suggestions were met with great resistance at the time. It wasn’t until the early 1λ80’s when Fernando Nottebohm renewed the interest in this field when he discovered the phenomenon of neurogenesis in songbirds while he was trying to understand how seasonal and hormonally driven changes in the size of certain song nuclei came about (Goldman and Nottebohm 1983). It was an exhilarating moment for science, and subsequently Nottebohm and colleagues were able to provide more compelling evidence for neurogenesis, including electron microscopy, retrograde neuronal labeling, and electrophysiological recordings from newly generated neurons (Alvarez-Buylla and Kirn 1997; Kirn et al. 1991). It is only recently that it became more generally accepted that adult neurogenesis also occurs in mammalian brains (Gould and Gross 2002). The increased interest in neurogenesis and stem cell research, which also produce new neurons, may in the near future be used to treat some diseases. This highlights the songbird as a perfect model system to answer such research questions. Finally, songbird species have been domesticated and can be raised and studied in the laboratory. Some species may produce three generations each year (Williams 2004). Moreover, the entire zebra finch genome had been sequenced. Therefore, the birdsong system is currently the best, if not the only, model system that combines neural and genetic approaches to study a complex natural behavior of great significance. It remains unclear why and how vocal production and learning has evolved and particularly how brain circuitries were modified to enable vocal learning, but such a development of vocal patterning that evolved in humans, birds and other mammals suggests the songbird as a tool to answer these questions (Liu et al. 2009). However, understanding of the complex neural circuitry within the brain is needed to understand the vocal learning and production mechanisms in any bird song (Reiner et al. 2004; Tchernichovski et al. 2004), and this is what we will address in the next few subsections.

8

2.3 Vocal Learners among Species

The scientific study of birdsong began in the late 1950s, with Thorpe (Thorpe 1961) and Marler (Marler 1970b). The key discovery of vocal learning however was made by Peter Marler as he was crossing neighboring valleys in Wales and heard a different song dialect in each. This observation led Marler to show that the dialects are very similar to human dialects in the sense that they were cultural in origin. Each songbird community within the valleys appears to have acquired a different song from its elders, in a process similar to the way human infants acquire language from their elders. This parallel between humans and songbirds was given additional weight by Fernando Nottebohm, Marler’s student, when he found that the song is laterized to the left hemisphere of the canary’s brain, just as language is laterized to the left hemisphere in the human brain (Nottebohm et al. 1976). In other songbird species however, like the zebra finches, there is a right-hemispheric dominance for song control (Williams et al. 1992). But since the 1λ70’s, the songbird has emerged as a model system in which to pursue answers to some of the most fundamental questions in neuroscience. Which animal species are capable of vocal learning and imitating the elder’s vocal signals? Imitation is the foundation for transmitting much of the human culture. Most importantly, our ability as children to vocally mimic the speech of our parents constitutes the foundation for our spoken language (Locke 1993). However, despite this fundamental importance of learning speech to humans, vocal learning in nonhuman species is a quite rare ability (Mooney 2009a). In particular, production of learned vocal expressions is found only in humans, bats, cetaceans (whales and dolphins), elephants, harbor seals and sea lions among mammals — a total of perhaps 300 species (Boughman 1998; Foote et al. 2006; Liu et al. 2009; Nottebohm and Liu 2010; Payne and McVay 1971; Poole et al. 2005; Ralls et al. 1985; Reiss and McCowan 1993; Richards et al. 1984; Tyack 1986; Yurk et al. 2002). On the other hand, three large groups of birds are known to learn their vocalizations: parrots and their allies (350+ species), hummingbirds (300+ species), and oscine songbirds (4,500+ species including zebra finches, canaries, and white-crowned sparrows) (Janik and Slater 1997; Jarvis 2004; Jurgens 2002; Olveczky and Gardner 2011; Poole et al. 2005; Simonyan et al. 2012; Williams 2004). Among the mentioned species, only humans have the most advanced capabilities to produce speech and language. The closest relatives to humans, the nonhuman primates, have very limited capabilities

9

to learn vocalizations (Egnor and Hauser 2004; Jurgens 2002). Similarly, the closest relatives of oscine passerine songbirds, the suboscine songbirds, lack the ability to learn new songs (Kroodsma and Konishi 1991). Thus, the fundamental question of how exactly the brain controls learned voice production for speech and song remains to be addressed through the studies in oscine songbirds only, which are referred to sometimes as the “true” songbirds. The passerine songbird that we study in the lab at FSU is the zebra finch (Fig. 2.1). Zebra finches tend to learn one song type as juveniles. Other passerine songbirds have the capability to learn several songs and they can undergo continuous learning throughout adulthood (Catchpole and Slater 1995; Okanoya 2004). The zebra finch song, learned early in life, often is very stereotyped throughout their lives. On the other hand, the songs of other species like the mockingbirds, nightingales and humpback whales show much greater variability (Catchpole and Slater 2008; Wohlgemuth et al. 2010). Because zebra finches are one of the few nonhuman animals that learn their vocalizations and because the neural circuitry for this songbird is well delineated over the past years, the zebra finch is an exceptional organism in which to explore how the brain acquires capabilities to learn complex sequences of vocal gestures, affording an experimentally tractable system where one can research the neurobiological underpinnings of vocal production and imitation (Mooney 2009a).

Figure 2.1: An adult male zebra finch (left) and a female zebra finch (right). Male zebra finches (Taeniopygia guttata) are distinguished from females by having bright orange cheek feathers, red beaks (as opposed to the orange beaks of females), and generally more striking black and white patterns.

10

2.4 Psychoacoustics Analysis of Birdsong

The invention of the sound sonogram (also known as spectrogram) at Bell Laboratories was a significant breakthrough for quantitative investigation and analysis of birdsong (Thorpe 1958). Most of the pioneering work on song development (Catchpole and Slater 1995) and on the functional anatomy of the song system (Hauser and Konishi 1999) relied on sonogram analysis. The sonogram transforms a transient stream of sound into a simple static visual image revealing the time-frequency structure of each song syllable and showing how the spectral density of a wave signal varies with time (Tchernichovski et al. 2004). Birdsongs are recorded and then analyzed by sound sonograms (Tchernichovski and Mitra 2002). Sonogram images can be measured, analyzed, and compared with one another. The elements of a sonogram are pretty simple and straightforward. The horizontal axis is time, and the vertical axis is frequency. A typical scale runs from 0 to 8,000 cycles per second, with “cycles per second” being the number of times per second that air compresses and releases to produce the sound we hear. A typical example of a sonogram showing the spectral derivatives of a song is shown in Fig. 2.2. This figure is taken with Sound Analysis Pro (Tchernichovski 2000) software developed by Ofer Tchernichovski. It is helpful for an accurate description of a bird’s song to start with a characterization of the fundamental dimensions along which the acoustic signals are deconstructed and extracted. As mentioned earlier, songbird vocalizations lack the combinatorial grammatical complexity of human language and perception is not well understood in the birdsong context; however birdsongs exhibit rich temporal structures (Knudsen and Gentner 2010; Tchernichovski et al. 2000). What are the acoustic features that are usually quantified and analyzed in the birdsong? When considering the motion of an oscillating membrane such as the vocal folds of the syrinx, the most obvious features are the period (pitch) of oscillation and regularity (entropy) of oscillation. Therefore, one can extract these dimensions from the acoustic signals and compute the pitch and the Weiner entropy of sound. Moreover, since acoustic signals unfold over time, processing in the temporal domain is important, and it makes sense to measure how pitch changes with time (frequency modulation). These, along with other “atomic” descriptions of acoustic features can be extracted with the Sound Analysis Pro software as well as with other

11

available software, making the analysis of sounds in their terms relatively easier to perform and the sonogram easier to read (Tchernichovski and Mitra 2002). Nevertheless, the sonogram has its limitations and is not always the best tool to study the birdsong. First, the most visible features of the sonogram image may not be necessarily the most important ones functionally. For example, if one looks at a spectrogram of human speech, the most apparent features are the harmonic structure the syllables exhibit. However, the harmonic structure does not carry the important information in human speech; it is rather the formants (the spectral peaks of the sound spectrum). Similarly, the syllable boundaries one observes on the spectrogram of human speech does not really determine the phrasing of words, it is rather the grammar rules that humans developed. In the case of birdsong however, it’s not totally clear what the acoustic features are that carry the relevant “information”. Moreover, similar to freezing a video and taking a snapshot image of it, the sonogram image presents a static representation of the song, which is a dynamic process (Tchernichovski et al. 2004). These and other reasons led us to develop SongSeq, a software tool that allows songbird researchers to study and analyze a large number of songs that are sung over one or multiple days of singing, thereby enabling quantification over the dynamic course of development or analyzing song following experimental manipulations like lesioning or pharmacological manipulations (Chapter 3).

2.5 Basic Units of Song

The song of an adult male zebra finch is a stereotyped series of acoustic signals with structure and modulation over a wide range of time scales, from milliseconds to several seconds (Fig. 2.2) (Fee and Scharff 2010). Long episodes of continuous singing are referred to as bouts. Song bouts, in turn, are composed of smaller acoustic units referred to as motifs each lasting about 0.5-1 second (Adret-Hausberger and Jenkins 1988; Fee and Scharff 2010; Immelmann 1969). In Fig. 2.2, the motif is bordered by a black rectangle, and the song bout is composed of two motifs. Motifs in their turn are composed of still smaller units of vocal gesture and shorter bursts of sound called notes or syllables (Price 1979); these may be analogous to the smallest units of speech, or phonetic units. The motif in the sonogram of Fig. 2.2 contains 3 syllables that are numbered as indicated. The individual notes last roughly 100-150 ms and can be broadly classified by the presence of an uninterrupted continuous energy in their spectrographic

12

representations (Fee and Goldberg 2011). Moreover, although several notes may occur in a given motif, their pattern is usually highly stereotyped between successive renditions of the same motif (Knudsen and Gentner 2010). Close examination of the song spectra and acoustic waveforms generally show syllables exhibiting transitions from periodic to aperiodic or chaotic dynamics, period doubling, and mode-locking transitions. In the songbird literature, the ordering of syllables and phrases in song is often called song syntax. Syntax is the same word applied to human speech, however, it implies grammar (Doupe and Kuhl 1999). Therefore, from the view point of a spectrogram, the song is a chain of discrete acoustic elements arranged in a particular temporal order (Catchpole and Slater 2008; Okanoya 2004; Sasahara and Ikegami 2007), where each motif is an acoustically complex event. Most importantly, both note structure and note order are learned features of the song (Mooney and Spiro 1997).

Figure 2.2: Sonogram (or spectrogram) showing a song bout produced by an adult male zebra finch. Song bouts, in turn, are composed of smaller acoustic units referred to as motifs each lasting about 0.5-1 second (black rectangle) that are repeated several times within the bout. Motifs are composed of still smaller units of sound called notes or syllables (numbered elements).

2.6 The Song Learning Process

Human infants learn to produce vocalizations rapidly in a process that appears to be simple. A few months after birth, children begin to babble and at 3-4 years of age they can produce meaningful sentences with the correct syntax. This developmental path is exactly the same regardless of language type or culture. It had been a mystery for linguists, psychologists and neuroscientists to explain how children do this, and particularly why the mechanism of language

13

acquisition is very regular despite the fact that it depends on learning and environmental factors which are generally highly variable. This mystery gets more tangled by the failure of computer science and artificial intelligence to design a robot that is able to acquire language. Therefore, cracking the “code” of language and solving this mystery remains an unsolved problem for scientists, although it is simple play for human infants. Songbirds pass through developmental paths in song learning that are extremely similar to the paths that humans pass through during language acquisition (Thorpe 1958). The first experimental evidence of song learning was the finding that juveniles exposed to conspecific (belonging to the same species) songs from other regions would subsequently sing these foreign dialects (Marler and Tamura 1964). Moreover, juvenile songbirds that are raised in isolation from other birds develop songs that lack many of the fine acoustic details that typify wild-type songs (Immelmann 1969; Marler 1970b; Marler et al. 1972; Price 1979; Thorpe 1954). Also, it had been shown that juveniles can even copy the song of another species (Baptista and Petrinovich 1984; 1986; Immelmann 1969; Marler 1991). All of these results highlight the fact that imitation for songbirds forms the basis of a normal song development. When songbirds hatch, they depend on their parents to feed them for the first 30–40 days of life. They fledge from the nest at around 20 days of age, which seems to be the time when they are ready to start vocal learning (Bottjer and Arnold 1997). After that age, songbirds become driven to sing whether they are with their parents, in community, or even in isolation. It had been hypothesized that songbirds exhibit innate perceptual predispositions for the vocal behavior of their own species, just like humans. However, innate predispositions are not enough to learn the song; social factors dramatically shape the learning process in songbirds. In order to learn to sing, juvenile songbirds need to listen to and memorize the song of their father or a neighboring adult male conspecific or even from taped renditions of song. This is called the sensory learning phase. Songbirds then conduct solo rehearsals of their own song using auditory feedback to match it to the memorized template. This is referred to as the sensorimotor learning phase. During this rehearsal process, the songbird learns to reproduce the target song with all its fine details ending up with a song that exhibits remarkable resemblance to the model song. The spectral and temporal features of every syllable as well as the correct sequence of syllables are exactly reproduced (Catchpole and Slater 1995; Mooney and Spiro 1997). Once the song is mastered, the bird retains it for the remainder of its life. This type of bird is referred to as a

14

closed-ended learner. Open-ended learners on the other hand refer to songbirds that learn their songs as juveniles; however, they have the capability to modify their existing song or even learn new songs when they become adults (Margoliash and Schmidt 2010). The model of song learning discussed above is based on the classic studies by Marler on the white-crowned sparrow (Marler 1970b). If songbirds were isolated during the sensory phase, or if they were deafened just before the sensorimotor phase, then the songs produced would be highly abnormal, bearing very little resemblance to the song of their conspecifics or the song template and showing no evidence of learning (Marler and Sherman 1983). In fact, “songs” that are produced by deafened birds are much more abnormal than those produced by birds raised in isolation (Margoliash and Schmidt 2010). So birds must be able to hear themselves for a normal vocal refinement. Similarly in humans, the vocalizations of children are highly abnormal if they are raised without the exposure to human speech, even though they have normal hearing (Fromkin et al. 1974). Moreover, if human infants are born deaf, they do not acquire spoken language (Doupe and Kuhl 1999), and if they were born with normal hearing abilities, but become deaf even late in childhood, speech deteriorates markedly (Brainard and Doupe 2000a; Waldstein 1990). Thus, birdsong, like human speech, depends on hearing the vocalizations of other individuals as well as hearing the sounds of oneself for a normal vocal production (Brainard and Doupe 2000a). Moreover, this shows that the production of adult birdsong (and similarly human speech) is not “hard-wired” in the brain and requires continuous auditory feedback. Once speech and birdsong are learned, they often remain remarkably stable. For example, it is very difficult to alter the accent of a human’s speech or to develop fluency in a new language in adulthood (Brainard and Doupe 2000a; Doupe and Kuhl 1999; Snow and Hohle 1978; Werker and Tees 1992). Likewise, in many songbird species, the fine details of the learned song remain largely unaltered during adulthood, despite the fact that songbirds are continuously being exposed to different acoustic models in their community (Brainard and Doupe 2000a; Immelmann 1969; Konishi 1985; Marler 1970b). In zebra finches, as in many other species of songbirds, song learning can be divided into three stages: subsong, plastic song and crystallized song. Just after the songbirds hatch, they initiate their earliest efforts to produce sounds by generating calls. This highlights the beginning of the earliest stage of singing, called subsong. Calls are usually generated by both male and female zebra finches, but as the males mature, a number of their calls start to exhibit complex

15

acoustic patterns which the songbirds then produce in a sequential manner (Liu et al. 2004). The production of these sequences of acoustic elements marks the subsong stage and is very similar to babbling in human infants (Doupe and Kuhl 1999; Margoliash and Schmidt 2010). Figure 2.3A depicts a sonogram showing a song bout produced by a male zebra finch during subsong (Johnson and Sellix 2000). Bouts of subsong show little or no structure in terms of note types and note sequence. Moreover, during this initial phase of vocal practice (~25–30 days post hatch, dph), juveniles begin to form memories of their tutor songs thereby generating a neural “template” of the song (Konishi 1965). In addition to listening to their tutors, juveniles also listen to the results of their vocal output and calibrate their vocal instrument, but their subsong takes a highly variable and unstructured form that lacks any reproducible acoustic elements (Immelmann 1969; Olveczky and Gardner 2011).

Figure 2.3: Sonograms showing song bouts produced by a male zebra finch during subsong (A), plastic song (B), and crystallized song (C) (Johnson and Sellix 2000). A. Bouts of subsong show little or no structure in terms of note types and note sequence. B. Bouts of plastic song show specific note types and a sequence of notes begin to emerge. However, note production is unstable and song notes are often produced out of sequence. C. Bouts of crystallized song are highly stereotyped that usually begin with a series of introductory notes followed by a conserved note sequence that is repeated several times.

16

Plastic song is the stage that follows subsong. In this stage, the acoustic elements start to resemble those in the tutor song, but are still produced in a sequence that is variable (Margoliash and Schmidt 2010). Figure 2.3B depicts a sonogram for the same bird in Fig. 2.3A but during the plastic song stage. Clearly, during this stage specific note types and note sequence begin to emerge. However, note production is unstable and song notes are often produced out of sequence. Plastic song often spans two months or more and the juvenile during this duration may sing thousands of times each day. For this reason, tens or even hundreds of thousands of song renditions may be necessary to achieve accurate imitation (Mooney 2009a). At the onset of sexual maturity, the variability that exists during plastic song is substantially eliminated. This process is called crystallization where the young bird begins to produce a normal adult song exhibiting striking resemblance to the tutor song (Fee and Scharff 2010). Song is defined as crystallized when the acoustic features of each syllable and sequence of the various syllables are relatively stable and stereotyped. Figure 2.3C shows the song during this stage exhibiting fully stereotyped bouts that begin with a series of introductory notes followed by a conserved note sequence that is repeated several times. In this stage too, the song is usually less dependent on auditory feedback. However, even though the song is less dependent on auditory feedback during crystallization, auditory feedback is still necessary for song maintenance (Margoliash and Schmidt 2010; Nordeen and Nordeen 1992). While plastic songs do not elicit any interest from adults of male or female songbirds, crystallized songs attract females and elicit aggressive responses from other males (Prather et al. 2009). In zebra finches, the crystallized song is the song they will sing for the remainder of their lives (Deregnaucourt et al. 2004; Immelmann 1969). In laboratory raised zebra finches plastic song begins around 50 dph and song crystallization between 90 and 120 dph (Immelmann 1969; Margoliash and Schmidt 2010; Roper and Zann 2006; Tchernichovski et al. 2004). The exact timing of transitions between the three stages of song development is somewhat variable and is influenced by various conditions in which the juvenile birds are raised. Among these conditions are the timing of birth, social factors, hormonal state, and nutrition (Baptista and Petrinovich 1986; Kroodsma and Pickert 1980; Nowicki et al. 2002). Another way to look at the development of song and the transitions between the different stages is to examine the evolution of specific acoustic features throughout development. Figure 2.4 shows multiple scatter plots where each scatter plot represents 1 day in each of the first 8

17

weeks of vocal development compared with a day of adult singing (Thompson et al. 2007). In each of the scatter plots note duration is plotted on the x-axis and on the y-axis one measured spectral features is plotted, which is pitch. Within these plots, each data point represents an individual note and thus discrete clusters of data points signify repeated production of a specific note type. The gray dots represent adult notes, and the black dots indicate developing notes. Notes produced during subsong (weeks 1, 2) exhibit substantial variation in duration, but are initially limited in pitch. Plastic song (weeks 3–8) marks the gradual trend toward stereotypy, during which specific note types emerge.

Figure 2.4: Duration versus pitch scatter plots during juvenile vocal development (Thompson et al. 2011). Each scatter plot represents 1 day in each of the first 8 weeks of vocal development compared with a day of adult singing (the gray dots represent adult notes, and the black dots indicate developing notes). Notes produced during subsong (weeks 1, 2) exhibit substantial variation in duration, but are initially limited in pitch. Plastic song (weeks 3–8) marks the gradual trend toward stereotypy, during which specific note types emerge.

18

Moreover, although male zebra finches produce a single song when they become adults, their songs are produced under two different behavioral contexts. When the male zebra finches sing directly to females, a mechanism called “directed songs”, they deliver the song very rapidly and their song exhibits a highly stereotyped form. When males sing alone or in the presence of other males, the song is called “undirected song”, and these songs are usually delivered with a slightly slower tempo and show a greater deal of acoustic variability than directed songs (Margoliash and Schmidt 2010).

2.7 The Song System: Anatomy

Songbirds have specialized, spatially distributed, compact and discrete brain nuclei that are interconnected through a series of pathways known as the “song system”. Very few of the neural circuits that are known to control behavior are anatomically as distinct as the song system. One can see some of the forebrain song nuclei in stained slides with the naked eye (Konishi 2010). This specialized neural circuitry is essential to song production and song learning distinguishing the songbird’s brain from the brain of birds that do not learn to vocalize (Kroodsma and Konishi 1991; Mooney 2009a; Nottebohm et al. 1982; Nottebohm et al. 1976; Wild 1997a; 2004; 1997b), and it is as well responsible for a series of mechanisms including memorization of the song template, evaluation of auditory feedback relative to the template, and continuous modification of vocal output (Brainard and Doupe 2000a). The brain circuitry involved in birdsong consists of three networks of interconnected structures that are functionally linked. The first pathway is the vocal motor pathway (VMP) that was initially discovered by Nottebohm et al. (1976) in canaries (Fig. 2.5, blue pathway). The VMP is a network of brain nuclei unique to oscine songbirds (Prather and Mooney 2004) and it is required throughout their life for normal production and acquisition of their songs (Brainard and Doupe 2000a; Nottebohm et al. 1976). The motor pathway originates its journey in the HVC. The HVC is the chief telencephalic nucleus occupying the dorsal part of the pallium (known as the neostriatum), a region of the avian brain analogous to mammalian layer III of the cortex (Fee and Scharff 2010;

Reiner et al. 2004). A subset of HVC neurons, commonly referred to as HVCRA neurons, project via a caudal route to the robust nucleus of the arcopallium (RA) (Mooney 2009b; Vicario 1991). Nucleus RA exhibits the shape of a spherical ball of neurons spanning the medio-caudal region

19

of the archistriatum, and is thought to be analogous to the mammalian layer V pyramidal neurons of the motor cortex that project directly to spinal motor neurons (Fee and Scharff 2010; Karten 1991). The output of RA includes the entire known forebrain component of the song system descending to the midbrain and brainstem involved in motor control of the syrinx and in respiratory control (Margoliash 2010). In particular, RA projection neurons send long axons innervating the neurons of the hypoglossal nucleus (the tracheosyringeal portion of the nucleus of the twelfth nerve, nXIIts), as well as the respiratory premotor neurons. The respiratory premotor neurons include the nucleus retroambigualis (RAm), which controls expiration, and the nucleus parambigualis (PAm), which controls inspiration (Nottebohm et al. 1976; Wild 1993). Unlike the mammalian cortical layer V neurons however, RA has no connections with the striatum (Jarvis 2004). The neurons of nXIIts in their turn innervate the muscles of the vocal organ, or the syrinx (Vicario and Nottebohm 1988; Wild and Arends 1987). Thus, the VMP provides the anatomical basis by which the telencephalon exerts control over vocalization, a hallmark of both human speech and birdsong.

Figure 2.5: The song system pathways. The vocal motor pathway (VMP) contains circuits that directly pattern song output. Incoming sensory information is processed by HVC, and HVC initiates motor sequences that project out to the peripheral vocal organ, the syrinx, via RA and the hindbrain nucleus nXIIts. The anterior forebrain loop (AFP) pathway contains circuits that are important for song learning and song variability. HVC sends projections to a basal ganglia loop (striatal-thalamic-cortical-striatal) which has an important output projection to the song motor pathway at nucleus RA. The auditory pathway contains circuits that process sounds, including song. Auditory signals enter the brain at the cochlear nucleus (CN) and are further processed as they pass through the forebrain by nuclei such as NCM and CM. Image courtesy of Heather Williams.

20

The second pathway of the song system is the anterior forebrain pathway (AFP, red pathway in Fig. 2.5). This pathway is functionally distinct from the VMP and plays a key role in juvenile song learning, early song production, and adult song maintenance, but it doesn’t play an important role in adult song production (Aronov et al. 2008; Bottjer et al. 1989; Bottjer et al. 1984; Brainard and Doupe 2000a; b; Nordeen and Nordeen 1993; Nottebohm et al. 1982; Nottebohm et al. 1976; Okuhata and Saito 1987; Scharff and Nottebohm 1991; Sohrabji et al. 1990). The AFP shares many similarities with the basal ganglia-thalamo cortical pathways in mammals (Bottjer and Johnson 1997; Luo and Perkel 1999; Reiner et al. 1998) that play an important role in sensorimotor learning (Person et al. 2008). Similar to the VMP, the AFP starts its journey in the HVC. In addition to HVCRA neurons discussed earlier, HVC contains another subset of projection neurons that send their axon collaterals to the basal ganglia striatal component of the song system known as Area X (Bolhuis et al. 2010); these neurons are generally referred to as HVCX neurons. Area X also receives dense dopaminergic innervation from the midbrain, including the ventral tegmental area (VTA). Area X neurons in their turn project to the thalamic nucleus DLM (medial portion of the dorsolateral thalamus), which then send their afferents to the magnocellular nucleus of the anterior nidopallium (MAN) that is contained within the pallium with HVC and RA. The medial part of MAN (MMAN) sends output to the HVC nucleus (Kubikova et al. 2007), while the lateral part of MAN (LMAN) contains two types of projection neurons, one type projects to area X and the other projects to RA. Therefore, HVC together with LMAN, provide the only known input to RA (Bottjer et al. 1989; Nottebohm et al. 1976). This connectivity pattern of the AFP forms a three-station (pallium–basal ganglia–thalamus) loop reminiscent of human cortico-striatal-thalamic-cortical loops (Bottjer et al. 1989; Luo et al. 2001; Vates et al. 1997). The third pathway is an ascending pathway involving a number of auditory cortex-like regions in the caudal pallium that are important to auditory processing and perception involved in sensory and motor song plasticity (Bolhuis et al. 2010; Jarvis et al. 2005; Mello 2004; Mello et al. 2004). This pathway comprises primary and secondary regions of the auditory telencephalon (gray areas in Fig. 2.5). In particular, afferents from the inner ear project to the cochlear nucleus in the medulla. Similar to the situation in mammals, there is both a direct and an indirect route connecting the cochlear nucleus and the auditory midbrain. In the midbrain, these pathways converge in the dorsal lateral nucleus of the mesencephalon (MLD), which is analogous to the

21

inferior colliculus (IC) in mammals (Theunissen and Shaevitz 2006). The auditory midbrain projects to ovoidalis (Ov), a relay nucleus in the thalamus, just as the IC projects to the medial geniculate body (MGB) in mammals. Ovoidalis, in turn, sends their afferents to the primary auditory area in the pallium, called field L (Zaretsky and Konishi 1976). The avian field L can be thought as analogous to the mammalian primary auditory cortex (A1). Field L has been further divided into subregions (L1, L2a, L2b, L3, L) on the basis of differences in cytoarchitecture and connectivity (Fortune and Margoliash 1992; Prather and Mooney 2004; Theunissen and Shaevitz 2006; Vates et al. 1996). The neurons of the auditory thalamic nucleus ovoidalis project in particular onto subdivisions L2a and L2b of field L (Karten 1968; Wild et al. 1993). These thalamorecipient subdivisions of field L in turn reciprocally project to L1 and L3 subdivisions of field L (Margoliash 2010). Moreover, L1, L2a, and L3 neurons project to secondary auditory structures in the caudal telencephalon (Margoliash and Schmidt 2010; Vates et al. 1996); these structures include the caudal medial nidopallium (NCM) as well as the medial (CMM) and lateral (CLM) divisions of the caudal mesopallium (CM) – all involved in song perception and are important for the recognition of tutor song (Bolhuis and Gahr 2006). The CM projects to the interfacial nucleus of the nidopallium (NIf) and the nucleus HVC of the song system (Vates et al. 1996), while the NIf projects only to the HVC. The forebrain regions L3, CM, and NCM all have the potential to transmit auditory information that influences song motor production through their projections to the motor nucleus HVC and/or its underlying shelf region (Bauer et al. 2008; Kelley and Nottebohm 1979; Woolley 2012). In addition to the input that HVC receives from these auditory areas, the forebrain nucleus uvaeformis (Uva) also sends ascending projections onto HVC. Uva in its turn receives input from some of the brainstem structures that RA projects to (Margoliash and Schmidt 2010; Schmidt and Ashmore 2008).

2.8 The Song System: Functions and Roles

The song system is an extremely distributed network that spans the entire songbird brain and conveys descending motor signals, as well as ascending auditory information (Mooney 2009a). What are the prime movers within this system that control behavior? A look at the wiring diagram alone cannot tell us much about how the song is produced and learned, but some experimental manipulations can. For instance, electrical or chemical lesions as well as

22

pharmacological inactivation can help us understand whether a brain nucleus is necessary for the expression of a particular behavior. Electrophysiological recordings can tell us whether the activity in a brain region is correlated with a behavior, as well as the underlying ionic current mechanisms that generate the electrical activity. Electrical stimulation can tell us whether activation of a brain region, or the axons passing into or through that region, can disrupt or elicit a behavior (Vu et al. 1994). Next, we will be exploring some of these manipulations that led to our better understanding of the song system and its functionalities.

2.8.1 Effects of Deafening

As pointed out earlier, for vocal imitation to succeed, hearing must be intact (Konishi 1965). In humans, large bilateral lesions of auditory cortex result in auditory agnosia, in which patients are unable to discriminate complex auditory stimuli of any kind (Doupe and Kuhl 1999). Similarly, the effects of deafening or disrupting auditory areas on song development are dramatic in juvenile songbirds, and in adult birds it causes gradual deterioration of song. While the deafening experiments alone tell us that hearing is crucial for normal song learning and production, it does not provide information about the exact roles of the song system pathways and the various brain nuclei along these pathways.

2.8.2 Functions and Roles of the AFP

The AFP is implicated in song learning but not song production for various reasons, some of which will be highlighted next. First, lesions of Area X or LMAN in the AFP in juvenile songbirds result in dramatic disruption of song because these lesions are made during the sensorimotor learning period (Bottjer et al. 1984; Doupe et al. 2005; Scharff and Nottebohm 1991; Sohrabji et al. 1990), that is during the period when the bird is learning his song. Lesions of Area X and LMAN in adult birds, however, do not have an effect on song production since the bird already learned his song (Bottjer et al. 1984; Nottebohm et al. 1976; Scharff and Nottebohm 1991). Moreover, targeted neuronal ablation of many of the HVC cells that project to Area X has no effect on the patterning or stability of adult song (Scharff et al. 2000). Thus, these results combined provide direct evidence that the AFP is not necessary for song production but plays a key role in song learning. In addition to that, various research has suggested that the AFP plays

23

an “error-correction” role (Brainard and Doupe 2000a) that was based on the postulated role of vertebrate basal ganglia circuits in reinforcement learning (Doya and Sejnowski 1995). This model suggests that the basal ganglia nuclei calculate the difference between the desired outcome and the actual performance, and as a result it outputs the difference as an “error” signal that the bird uses to improve his performance (Bolhuis et al. 2010). It had been shown that pharmacological disruption of normal neuronal activity in LMAN, which projects from the AFP into the motor pathway, during tutor song exposure (and not during vocal practice) prevents birds from producing an accurate copy of the tutor song (Basham et al. 1996). Also, lesions of LMAN prevent the gradual deterioration of song that follows deafening and causes premature crystallization of the song (Bottjer et al. 1984; Fee and Scharff 2010; Scharff and Nottebohm 1991; Sohrabji et al. 1990). Moreover, song losses that are induced by small lesions of HVC can be recovered if hearing remained intact, and this recovery is much faster if the HVC lesions are preceded by lesions of LMAN (Brainard and Doupe 2000b; Thompson et al. 2007). In addition to that, evidence suggests that there are similarities between the effects of lesioning LMAN in deafened adult birds and those of pallidotomy in Parkinson’s disease: in both cases, when motor behaviors are already well-learned, removal of abnormal cortical–basal ganglia activity leads to more normal expression of these behaviors (Doupe et al. 2005). Taken together, these results suggest that the LMAN participates in the process of song memorization, plays a key role in inducing song variability and can promote or undo vocal learning. On the other hand, permanent lesions of Area X, the basal ganglia analog in songbirds, also result in a poor imitation of the tutor song, but in contrast to LMAN lesions, the song does not crystallize, and the adult bird retains a large number of syllables but they exhibit abnormally high sequence variability (Fee and Scharff 2010; Scharff and Nottebohm 1991). This indicates that Area X plays an important role in song learning but does not induce song variability like LMAN does. And just as Area X is necessary for vocal learning in the songbird, the mammalian basal ganglia is involved in motor learning (Graybiel et al. 1994). Disruptions to the basal ganglia in mammals affect serial processing and sequential behaviors, which includes speech and language. These effects are seen in Parkinson’s disease, schizophrenia, obsessive-compulsive disorder, Huntington’s, Tourette’s, and the tardive syndromes (Brown et al. 2003; Fee and Scharff 2010; Muller et al. 1997). However, the detailed mechanisms and interactions between

24

the AFP nuclei and their roles on the learning and maintenance of vocalizations remain largely unknown.

2.8.3 Functions and Roles of the VMP

Unlike the AFP, the VMP plays a crucial role in both song production and song learning. This comes from the fact that ablations or disruptions to nuclei along the VMP alter the song significantly whether the bird is a juvenile or an adult. In particular, lesions (electrical or chemical) of HVC result in abnormal songs or muteness (Nottebohm et al. 1976). Similarly, since the neurons of RA provide the primary input to the brain regions controlling the respiratory and syringeal muscles, lesions to RA or respiratory brain regions would generate destructive effects on patterned breathing and vocal muscle activity necessary for normal song production (Suthers et al. 1999; Wild 1997b) because at the end the birdsong, just like human speech, requires the coordinated control of vocal and respiratory musculature. Moreover, the songbird brain has long been known to contain some of the most complex and selective auditory neurons ever identified, so-called “song-selective” neurons (Nagel et al. 2011). Neurons in the VMP display singing-related activity (Hahnloser et al. 2002; Hessler and Doupe 1999; Leonardo 2004; Leonardo and Fee 2005; McCasland 1987; McCasland and Konishi 1981; Yu and Margoliash 1996), and many of these neurons also respond to auditory stimulation (Doupe and Konishi 1991; Katz and Gurney 1981; Margoliash 1983; 1986; McCasland and Konishi 1981; Vicario and Yohay 1993), pointing to an important role by which auditory information could be shaping vocal control (Mooney 2009a). However, of particular interest to us is the telencephalic nucleus HVC. The pivotal location of this nucleus residing at the top of the VMP (and the AFP) makes it a crucial point of entry into the songbird’s sensorimotor system. It’s important to start by mentioning how the HVC nucleus responds to sound in order to have a better evaluation of its role. Neurons in HVC respond to auditory stimulation and this response - under anesthesia - is maximal to playbacks of the bird’s own song (BOS) than to songs of other individuals of the same species (conspecifics) or even the bird’s own song played in reverse order (Doupe 1997; Lewicki and Konishi 1995; Margoliash 1986; Margoliash and Fortune 1992; Margoliash and Konishi 1985; Mooney 2000; Rosen and Mooney

2003). However, there is no evidence that the HVCRA neurons that fire so selectively during song production respond to the auditory feedback that reaches the singing bird (Margoliash and

25

Schmidt 2010). Similarly, the human vocal motor system is sensitive to speech sounds. In human adults speech perception modulates the excitability of tongue muscles in a phoneme-specific way (Fadiga et al. 2002), and in human infants broca’s area is activated by speech and auditory input from the babbling stage onwards (Bolhuis et al. 2010; Dehaene-Lambertz et al. 2006; Imada et al. 2006). In addition to HVC neurons’ response to BOS, they are probable sites for encoding the tutor song. A recent study used a combination of song-triggered optogenetic and focal electrical stimulation methods to manipulate the activity of HVC neurons in juvenile zebra finches as they listened to the song of a tutor (Roberts et al. 2012). As a result of these manipulations, the quality of song imitation was impaired. Moreover, when they blocked the NMDA (N-methyl-D- aspartate) receptors in HVC during tutoring, the vocal imitation of the tutor song was impaired as well. This suggests that an NMDA receptor-dependent strengthening of synapses on HVC neurons is important to encoding the tutor song experience. Combined together, these findings support the idea that synapses in HVC are sites where the experience of the tutor song is encoded in the brain and also indicate that this encoding depends on brain regions that supply auditory input to HVC, like NIf (Cardin et al. 2005; Hahnloser and Fee 2007). Although these findings do not exclude the involvement of other regions downstream of HVC in this sensory encoding process (Basham et al. 1996), they do rule out prevailing models in which the auditory experience of the tutor song is first encoded in auditory regions upstream of the HVC and is only later used to guide changes in the vocal motor network during sensorimotor learning (Bolhuis and Gahr 2006; London and Clayton 2008; Phan et al. 2006; Roberts et al. 2012). A very helpful way to decipher the HVC code is to record the singing related electrical activity of identified neurons. Evidence showed that the neurophysiological activity in HVC and RA is highly correlated with song production (Chi and Margoliash 2001; McCasland 1987) with HVC being the maestro of the song having a pattern-generating role encoding its temporal features and the syllable sequences that make up the song, while RA plays the role of the musician handling the details of which notes get played when (Roberts et al. 2012). Therefore, both HVC and RA are involved in the motor control of song, but they do so in a hierarchal manner where singing-related activity propagates down the VMP, arising earlier in HVC than in RA (McCasland 1987; Vu et al. 1994; Yu and Margoliash 1996). Moreover, during singing, brief low-magnitude electrical stimulation of the HVC and RA perturbs singing (Ashmore 2005;

26

Margoliash 2010; Vu et al. 1994; Vu et al. 1998). In particular, microstimulation in HVC causes interruption of singing and restarting of the song, whereas the same stimulation in RA disrupts only the structure of syllables without altering song patterning (Brainard and Doupe 2002; Vu et al. 1994). Further data on neural coding come from studies in which electrophysiological recordings in both the HVC and RA of singing birds detected activity time-locked to individual syllables (McCasland and Konishi 1981; Mooney 2009a; Yu and Margoliash 1996). In particular, Michale

Fee and his co-workers recorded from identified HVCRA neurons using a miniature motorized microdrive (Hahnloser et al. 2002). What they found was quite remarkable: single HVCRA neurons fire only one very brief (6-10 msec) burst of action potentials at a single precise time during the entire ~1 sec motif during each rendition of the song (Fig. 2.6), with different neurons bursting at different times in the motif (Mooney 2009a), and that is the entire contribution of that neuron to the production of the learned pattern—referred to as an “ultra sparse code” (Hahnloser et al. 2002; Margoliash and Schmidt 2010). This extremely precise temporal-locking of HVCRA neurons to the bird’s vocalization is one of the most temporally precise neural sequences found in nature to date. Similar to HVCRA neurons, evidence showed that HVCX neurons displays phase-locked patterns during singing (Fujimoto et al. 2011) although they do not exhibit the same level of stereotypy as HVCRA neurons. Just like HVC projection neurons, RA neurons’ firing pattern is precisely reproduced each time the bird sings its song motif. Each RA neuron produces a fairly unique pattern of roughly 12 bursts, each lasting ~10 ms (Fee and Scharff 2010; Leonardo and Fee 2005). Moreover, RA neurons’ firing pattern is characterized by transitioning from tonic, regularly firing patterns of activity to complete inhibition punctuated by brief bursts of activity (Yu and Margoliash 1996). It is not yet known precisely how these complex sequences of bursts that are extremely stereotyped in HVC and RA are related to vocal output— for example, how the telencephalic orchestration of singing interfaces with brainstem motor and respiratory circuits downstream from RA (Fee and Scharff 2010). Therefore, within the song system, the details of note structure task appear to fall to the RA, as the activity of a given neuron in this nucleus, unlike that of a neuron in the HVC, is the same for a given note regardless of its syllabic context (Mooney and Spiro 1997).

27

Figure 2.6: Spike raster plot of ten HVCRA neurons and two HVC interneurons recorded in one bird during singing. Each row of tick marks shows spikes generated during one rendition of the song or call; roughly ten renditions are shown for each neuron. Neural activity is aligned by the acoustic onset of the nearest syllable. HVCRA neurons burst reliably at a single precise time in the song or call; however, HVC interneurons spike or burst densely throughout the vocalizations. Figure modified from Hahnloser et al (2002).

The extremely sparse and precise patterns of activity in HVCRA and RA neurons could suggest the entire ensemble of HVCRA neurons is functioning to specify the timing of syllables, notes, and even the intervening silent “gaps” between syllables. Indeed, some HVCRA and RA neurons burst during these silences, consistent with this idea. If the output of the HVCRA ensemble provides a timing signal for song, what and where is the mechanism that determines

28

and sets the song tempo? Based on the observations that RA-projecting HVC neurons generate a single burst of spikes during the song motif and that different neurons appear to burst at many different times in the motif, it has been hypothesized that these neurons generate a continuous sequence of activity over time (Fee et al. 2004; Kozhevnikov and Fee 2007). In fact, the axons of

HVCRA neurons extend local collaterals before exiting HVC, forming excitatory synapses with other HVCRA cells, as well as interneurons and HVCX cells, providing a potential substrate for a synfire chain (Mooney 2000; 2009a; Mooney and Prather 2005). In other words, at each moment in the song, there is a small ensemble of HVCRA neurons active at that time and only at that time, and each ensemble transiently activates a subset of RA neurons determined by the synaptic connections of HVC neurons in RA (Fee and Scharff 2010; Leonardo and Fee 2005). In this sense, bursting activity propagates through a chain of synaptically connected HVCRA neurons, much like a series of falling dominoes, creating a timing signal that spans the entire motif. In this case, song tempo is determined by the intrinsic biophysical properties of the HVC local network (Mooney 2009a). Additional evidence for the propagation of activity through a chain in HVC came from Michale Fee’s lab again where they bilaterally cooled HVC during singing by use of a thermoelectric heat pump (Long and Fee 2008). Bilateral cooling of HVC revealed that all aspects of song timing - duration of song syllables, interval between syllable onsets, and interval between motif onsets - are slowed by approximately 3% per degree Celsius of HVC cooling (Fig. 2.7). In contrast, bilateral cooling in RA had no effect on song timing. Thus, the generic sequential activation of the HVC chain is translated, by HVCRA neurons, into a specific precise sequence of vocal configurations (Fee and Scharff 2010). These results combined set the ground for mathematicians to develop computational models in an attempt to explain how the different types of neurons in the HVC are interconnected to produce the sequence of firing (Abarbanel et al. 2004b; Abarbanel et al. 2012; Abarbanel et al. 2004c; Abarbanel 2004; Drew and Abbott 2003; Gibb et al. 2009a; b; Jin 2009; Jin et al. 2007; Li and Greenside 2006; Long et al. 2010; Mooney and Prather 2005; Troyer and Doupe 2000a; b). Despite these attempts that provided insight to the topic, the ionic currents ingredients of HVC neurons that produce the recipe of the HVC neurons’ firing pattern remained largely unknown.

29

Figure 2.7: Representative sonograms recorded from a single bird with HVC heated and cooled, showing percentage song dilation relative to control. Hotter colours represent greater sound intensity. At colder temperatures, song motifs were produced more slowly than control songs. Figure modified from Long and Fee (2008).

An alternative possibility was suggested which is that the bursting in HVC is directly driven and controlled by synaptic inputs from circuitry within upstream brain regions, such as the thalamic nucleus Uva (Coleman and Vu 2005; Fee and Long 2011; Nottebohm et al. 1982; Williams and Vicario 1993). However, it was then shown that if proprioceptive feedback associated with vocalization is prevented from reaching HVC, e.g. by lesions of the thalamic nucleus Uvaeformis, this can result in stammering (Williams and Vicario 1993), but the feedback affects only the timing and sequencing of note delivery and the birds can still produce all the sounds of the learned song, reinforcing the belief that the learned patterns for these syllables reside in HVC (Margoliash and Schmidt 2010). In summary, despite the key role that HVC plays in the song system, precisely how the neurons in HVC are orchestrating the song sequence and phonology is very poorly understood. On one level, the “internal anatomy” of HVC neurons and their biophysical properties are largely

30

undetermined. Without a detailed explanation of the components of the ionic currents of HVC neurons and their exact contributions on spike generation, the aim to decode the song is extremely challenging and the story remains incomplete. This motivates the need to characterize HVC neurons electrophysiologically and unveil how the various ionic currents interact to control the generation of HVC neurons’ action potentials. On another level, the synaptic connectivity patterns of HVC neurons are largely undetermined. Only one study in the literature gave insight on the types of synaptic interactions between HVC neurons (Mooney and Prather 2005), and there had been very few attempts to characterize the types of synaptic connections between HVC neurons (Shea et al. 2010; Solis and Perkel 2005). As we will see in Chapter 4, we took a major step towards characterizing HVC neurons electrophysiologically and identifying their ionic currents in the slice as well as develop realistic mathematical models that generate the activity patterns of HVC neurons calibrated by our slice work. Later in Chapter 5, we connect our model neurons into plausible networks that can produce the patterns of neural activity exhibited by HVC neurons.

31

CHAPTER 3

BEHAVIORAL LEVEL: SONGSEQ

3.1 Introduction

As discussed in Chapter 2, songbirds such as zebra finches and Bengalese finches are often studied due to their patterned song. One important feature of their songs is the sequence of syllables that comprise it. Songbird researchers are quite often interested in the quantification, analysis and comparison of the sequence of syllables produced by a bird prior to and after a certain experimental manipulation (such as ablation of a brain region or infusion of pharmacological agents), or through the course of the bird’s development (such as monitoring the evolution of song through subsong, plastic and crystallized stages), or even to assess the degree of individual variation in the song structure across population of birds. Since songbirds tend to sing hundreds to thousands of songs each day, the quantification and analysis process is a very tedious task without the help of automated tools. Here we describe a computer software tool, SongSeq, for automated, high throughput, quantitative syllable-level analysis of bird song syntax that quantifies the transitions made between syllables in one or more bouts of singing by a bird, using the zebra finch as a model system. The primary advantage of our tool is the ease and effectiveness it provides in quantifying syllable sequence and performing a comparison of syllable sequence from one day of singing with one or more other days of singing. The software utilizes the output of the Feature Batch module in Sound Analysis Pro (SA+) software package (Tchernichovski et al. 2000). SA+ is a frequently used software package for converting continuous sound into discrete syllables, measuring the temporal and spectral features of each syllable produced during a day of singing, and extracting these features (such as duration, pitch, pitch goodness and entropy) into Excel spreadsheets (Tchernichovski et al. 2000). SongSeq uses these measurements to identify individual syllables based on their temporal and spectral properties, monitor the changes of the sound features across a large number of songs as well as identify and analyze transition probabilities among syllables to determine changes in syntax. Moreover, the software quantifies the consistency of syllable ordering by computing the

32

linearity, consistency, and stereotypy scores for every bout presented as well as descriptive statistics for each of these measures for each day of singing. In addition to that, we report other statistical measures that SongSeq utilizes (the Kullback-Leibler distance and the sequence entropy) to quantify the degree of similarity in syllable phonology that occurs between sequences of syllable transitions over different days of singing. In order to generate syllable-level comparisons, we developed a graphical user interface (GUI) to manually identify syllable clusters (i.e., repeated instances of the same syllable). Other methods to identify syllable clusters are available – for example, the Clustering Module in SA+ which identifies individual syllables based on Euclidean distance and high dimension filtering in acoustic feature spaces. The method presented here can be applied regardless of the syllable cluster strategy employed. SongSeq is thus an easy and effective tool for quantifying the development of syllable sequences and performing a multi-dimensional comparison of the syllable acoustic features from one day of singing with songs from other days. It therefore quantifies the ordering of syllables in songs and the frequency with which subsequences appear. This would be a useful tool for comparing the songs produced by a bird at different stages of development or prior to and after a manipulation such as partial brain ablation to portion(s) of the neural song system or infusion of pharmacological agents. The work presented in this chapter is published in the Journal of Neuroscience Methods (Daou et al. 2012).

3.2 Methods

The software described here is available for free download from http://www.math.fsu.edu/~bertram/software/birdsong/JNM_12/. There is also an online tutorial and a user manual at this site. Recordings of bird song are read into the SA+ software package as .wav files. Syllable units are generated by parsing every motif from song bouts using syllable segmentation tools in the Feature Batch module in SA+. Feature Batch generates an Excel spreadsheet for all syllables and their acoustic characteristics (for further details, see (Wu et al. 2008)). These spreadsheets are the input to SongSeq.

33

The entire procedure that SongSeq goes through is discussed below in detail and can be summarized by the following brief overview: 1) choose a template SAP spreadsheet, 2) choose two acoustic features to use for identifying syllable clusters, 3) name the syllables and assign them different colors, 4) use colored boxes to paint the different syllable clusters, 5) if syllables don’t cluster unambiguously, repeat steps 2 and 4 on two different acoustic features, 6) choose the test song files, and 7) compare features of test songs to those from the template..

3.2.1 Identification of Syllables Using a Template

To compare syllable features and sequences between one singing session and another, one must first designate one spreadsheet as the template (a typical SA+ spreadsheet containing the acoustic variables, which could be the data points for a day of preoperative singing, for example). Two acoustic features are then selected by the user from the template using SongSeq’s graphical user interface (Fig. 3.1). These features are extracted from each syllable in the template sonogram (Fig. 3.2) and displayed as a 2D scatter plot (Fig. 3.3). Within the scatter plot, each data point represents an instance of a syllable and discrete clusters of data points signify repeated production of a specific syllable type. This captures the syllable structure across multiple bouts, and defines the acoustic properties of individual syllables. The choice of acoustic features used to form the scatter plot could be important, since some feature pairs may be more effective than others in discriminating syllables. One strategy is to just try different combinations; SongSeq contains a module that enables the user to browse the different features and choose the best two features for syllable discrimination. An initial eye-identification of the spectrograms is helpful where the number of different syllables can be identified along with other acoustic features of the various syllables (like syllable duration, mean FM, amplitude and entropy). The number of clusters on the scatter plot should equal the number of different syllables on the spectrogram. We have found that syllable duration is typically a good feature to use (Wu et al. 2008). If two features are not enough to unambiguously discriminate features, SongSeq allows one to use a second pair of features for further discrimination.

34

Figure 3.1: Designation of the template spreadsheet and selection of two acoustic features. The user browses and selects a template file (SA+ spreadsheet). The template’s file storage path is displayed in the textbox after selection. Two acoustic features are then selected by the user. The acoustic features drop down menu lists all the features that SA+ generates.

Figure 3.2: Sonogram of a typical song of a bird that received bilateral HVC microlesions. The sonogram displays three introductory notes (labeled as I) and five syllables (labeled as A, B, C, D, and E).

35

Figure 3.3: A screen shot of SongSeq’s frame where syllable identification is processed. Left panelμ A 2D scatter plot of the two selected features from a preoperative day of singing (syllable duration versus mean pitch goodness). Right panel: GUI that associates the syllables clusters on the 2D scatter plot with syllables labels and colors.

The next step in syllable identification is using the graphical tools to associate syllable clusters with syllable labels (A, B, C, etc.) and colors. This is done using mouse clicks over the area in the 2D scatter plot that represents the syllable. Each mouse click paints onto the scatter plot a colored box covering the instances of a syllable (e.g., syllable A). This is used to define the boundary of that syllable in subsequent scatter plots (e.g., for different days of singing). The region covered by boxes of a single color defines a syllable (e.g., Fig. 3.4 shows five syllables labeled as A, B, C, D and E along with a cluster of introductory notes labeled as I). The colored boxes can be resized by the user (e.g., syllables A and E in Fig. 3.4 are painted with boxes of different sizes). The boxes are translucent so that one can see the syllable instances (dots) under the clusters. This procedure of syllable identification is done for every syllable, and thus for every cluster.

36

Figure 3.4: Syllable Identification: Using mouse clicks, the instances of each syllable in Fig. 3.3 are painted onto the scatter plot. The region covered by boxes of a single color defines a syllable. Here we have five syllables along with a cluster of introductory notes that is painted in black as I. On the right hand side of the frame, a tree shows the number of boxes used for each syllable.

It may happen that some syllables are easily identified with one pair of features, while others are better identified with a second pair of features. With SongSeq, one can identify the first set of syllables using one feature pair, then move to a second feature pair to improve the identification of the remaining syllables. In this process, and after all the syllables are identified with the first 2 acoustic features (Fig. 3.4), two new acoustic features are chosen along with a subset of the previously added syllables. Only this chosen subset of syllables (which can include all syllables) can be painted again in the new 2D feature space. Fig. 3.5 shows the selection of syllable duration versus mean entropy acoustic features. It also shows the selection of B and I since their clusters in Fig. 3.4 are not well isolated, and as we will see shortly syllable duration versus mean entropy pulls B and I apart. Syllables A, C, D and E are not chosen here because they are discriminated nicely with the first two acoustic features (Fig. 3.4).

37

Figure 3.5: A screenshot of SongSeq’s frame showing the selection of the second pair of acoustic features, and the selection of B and I to better discriminate them on the scatter plot of the new pair of features.

Once the second pair of features is chosen, a new 2D scatter plot appears. The data points are now color coded according to the first step of identification (Fig. 3.6). For instance, any data points that were within the blue boxes defining the borders of syllable A in the scatter plot of Fig. 3.4, will be painted in the same color (blue) in the scatter plot of Fig. 3.6. In Fig. 3.6 we see that the data points for B (green) and I (black) are pulled apart in the new 2D feature space, and it is clear that some data points previously misidentified as B actually cluster better with I (green points in the bottom right of the black cluster). Syllables A (blue), C (yellow) and D (cyan) remain well isolated, but syllable E (gray) has a very similar mean entropy as the introductory notes and therefore their corresponding clusters overlap.

38

Figure 3.6: Second Step of Syllable Identification: Color-coded data points on the scatter plot of the second pair of acoustic features chosen in Fig. 3.5, where colors are coded based on the painting done in the first step of identification. It is clear that some data points for B (green) and I (black) are pulled apart in the new 2D feature space, and some data points previously misidentified as B actually cluster better with I (green points in the bottom right of the black I cluster).

Next, the user paints in the same way as before onto the new scatter plot, but now only for the syllables (B and I) that were specified for this second round of discrimination (Fig. 3.7). For these specified syllables, a data point is considered as syllable X if and only if either of the following two conditions hold: 1) it belongs to one of the boxes defining syllable X in the first 2D scatter plot AND it belongs to one of the boxes defining syllable X in the second 2D scatter plot, 2) it belongs to one of the boxes defining syllable X in the second 2D scatter plot BUT is not a data point of a non-specified syllable (that is, a data point that is colored with blue (A), yellow (C), cyan (D) or gray (E)). Although the clusters for E and I overlap in the second scatter plot, the points originally labeled as E retain that identification since E was not selected for repainting.

39

Figure 3.7: Second Step of Syllable Identification: Syllables B and I are painted in the new feature scatter plot.

The next step is choosing the target files. Target files are typical spreadsheets generated by SA+. Information from these target files (which could be day(s) of post-operative singing, as in Fig. 3.8) is extracted and compared with the template spreadsheet to obtain sequencing details. Finally, typical transitions are selected by the user. A typical transition between two syllables is defined as one that is frequently encountered. As we will see, the typical transitions are used to calculate the consistency score that reflects the frequency with which a main or typical sequence appears.

40

A

B

Figure 3.8: The scatter plots of postoperative days 1 (A) and 3 (B) are superimposed with the painted clusters of the template. On the first day after a microlesion was made to the HVC, many notes were produced that did not fall within the syllable boundaries. On the third day after surgery, more notes were within the syllable boundaries.

41

3.2.2 Sequencing the Data of the Target Files

After SongSeq receives all required input, it probes the target files in the order they were uploaded. The operations described next are done automatically by SongSeq for each target data file. First, rows are associated with syllables based on the values of the user-specified features and are copied into a new spreadsheet, one for each syllable. This procedure is done for every defined syllable. Data points that do not fall into any of the named syllables (e.g., the points not lying in any of the colored regions in Fig. 3.4 and Fig. 3.6) are inserted into a spreadsheet called “NMS” (Non-Motif Syllable). Thus, each target file is parsed into a set of spreadsheets: one spreadsheet for every syllable name along with a spreadsheet for “NMS”. Next, these syllable spreadsheets are merged into a final “sequenced” spreadsheet that is created to contain all the notes in the order they were sung. This allows for the analysis of syllable transition probabilities. The different epochs of singing are identified in this spreadsheet by the .wav identifier.

3.2.3 Syntax Identification

Syllable transitions and scores are generated next. The sequenced spreadsheet is scanned row by row. For every block of consecutive rows that have the same .wav identifier, syllable transitions are determined. If the syllable name in row i of a song bout is "A" and the syllable name in row i+1 of the same song bout is "B", then the syllable transition is "A-B". This is done for each pair of consecutive rows within the block, producing a list of syllable transitions for every song bout. The syllable transitions are then merged and the number of occurrences of every transition is calculated. A transition probability is calculated by dividing the number of occurrences of the syllable transition by the total number of syllable transitions. The transition probabilities along with the syllable transition name are then written into a new spreadsheet. The song stereotypy is quantified using two measures that address related but different aspects of sequence stereotypy: sequence linearity addresses the way syllables are ordered in a song, and sequence consistency addresses how often a particular path is actually followed (Scharff and Nottebohm 1991). SongSeq calculates the linearity score of every song bout (.wav file) by dividing the number of different notes in the bout by the number of transition types in the bout. Since the number of different notes in a bout is always less than or equal to the number of transitions in the bout, a value of 1 represents the best (highest) linearity score. The consistency

42

score is calculated by dividing the sum of typical transitions in the bout by the sum of the total transitions in the bout (a value of 1 represents the best consistency score). A stereotypy score is then calculated as the average of the linearity and consistency scores. Scores are calculated for every song bout and then averaged over all bouts. The scores are then written into a new spreadsheet.

3.2.4 Kullback-Leibler (K-L) Distance Analysis

The Kullback-Leibler distance (K-L distance) is a measure of the difference between two probability distribution functions (Wu et al. 2008). We use this to compare syllable transition probability distributions. That is, we calculate the K-L distance between the transition probabilities of two different days to quantify the dissimilarity of the syllable sequences on those days. For one example included in Results, we were interested in the disruption of the song, and song recovery, after partially lesioning the vocal motor pathway, so we compared the transition probability distributions of postoperative days of singing to that of the preoperative day (day 1). Let n denote the number of syllables along with NMS (e.g., if the syllables entered by the user are A, B and C, then n = 4). There are possible combinations of two-syllable transitions.

If a transition never occurs, we set its probability to a non-significant small value ( ) to make , the computation numerically stable and accurate. Let represent the transition , probabilities on the preoperative day, and let represent , … , the transition probabilities on day k. For example, if is the probability of an ,A … to , B transition on day 1, then is the A to B transition probability on day 3 (e.g., Figures 3.9-3.14 shows the transition probabilities on one preoperative and five postoperative days). Then the K-L distance quantifying the degree of dissimilarity of the syllable sequences between days 1 and k is given by the following formula:

43

Figure 3.9: Syllable transition probabilities on preoperative day 1. In this figure, as well as subsequent Figs. 3.10-3.14, blue bars represent “typical transitions”, yellow bars represent “atypical transitions”, or transitions that were not listed by the user as typical, and the red bars represent “NMS transitions” involving notes that lie outside the syllable boundaries.

Figure 3.10: Syllable transition probabilities on postoperative day 1.

44

Figure 3.11: Syllable transition probabilities on postoperative day 2.

Figure 3.12: Syllable transition probabilities on postoperative day 3.

45

Figure 3.13: Syllable transition probabilities on postoperative day 8.

Figure 3.14: Syllable transition probabilities on postoperative day 12.

46

A K-L distance score is generated for every target file entered by comparing its transition probabilities with those of the first target file. The scores are then plotted in ascending order of the sequence of target files entered. Figure 3.15 shows a K-L distance analysis of the syllable sequence from one bird over two preoperative and 12 postoperative days of singing. The K-L distance of the first day of singing (Pre1) with itself is 0. Other data points describe the K-L distance between Pre1 and days following Pre1. This gives the time course of sequence dissimilarity over the days following Pre1. Larger values of the K-L distance reflect greater dissimilarity in the sequence.

Figure 3.15: Quantification of syllable transition distributions: K-L distance measure quantifying the dissimilarity in song sequence between the Pre and Post days. Pre1 and Pre2 are preoperative days of singing, and P1 through P12 are the days of postoperative singing. Dashed lines indicate the day of surgery.

47

3.2.5 Transition Entropy Analysis

The entropy of a probability distribution is a measure of the spread of that distribution (Wu et al. 2008). In our context, a high entropy means that there are many syllable transitions with non-negligible probabilities. A song has low entropy if only a few syllable transitions typically occur. If , represents the transition probabilities on day k, then the entropy corresponding to that, … day , is given by the formula:

The entropies corresponding to the transition probabilities of all target files are plotted in ascending order of the sequence of target files entered (Fig. 3.16).

Figure 3.16: Quantification of syllable transition distributions: Entropy analysis quantifying the spread of the syllable transition distributions for each day of singing. Dashed lines indicate the day of surgery.

48

3.3 Results

To illustrate the results obtained by SongSeq, we use two behavioral data sets from different male zebra finches. The first bird, an adult, received bilateral microlesions targeted at the song region HVC (proper name). As discussed earlier in Chapter 2, HVC contains projection neurons that contribute to the pathway that descends from the telencephalon to the brainstem vocal and respiratory network. This pathway, which is known as the vocal motor pathway (VMP), is necessary for the acquisition and the production of song. HVC neurons have a temporal role in song production via HVCRA projections ((Long et al. 2010) and reviewed in (Bolhuis et al. 2010)). The bird that we consider had a targeted microlesion that removed only a small region of the HVC (5-10%, (Thompson et al. 2007)), resulting in a disruption of his song post-surgery that gradually recovered to the preoperative state over a period of days. Next, we consider a second bird that had been recorded over the course of development. With the help of SongSeq, we monitor the bird’s vocal changes and the evolution of his song over the course of development.

3.3.1 Quantifying the Effects of HVC Microlesions on Syllable Sequence

Figure 3.2 shows a sonogram of a typical song for an adult bird displaying three introductory notes (labeled as I) and five syllables (labeled as A, B, C, D, and E) repeated over multiple motifs. Figure 3.3 shows a 2D scatter plot of two selected acoustic features from a preoperative day of singing (syllable duration versus mean pitch goodness). The six note clusters correspond to the five syllables repeated in song motifs and some introductory notes (the colored regions in Fig. 3.4). On the first day after a microlesion was made to the HVC, the song became very disorganized and many notes were produced that did not fall within the syllable boundaries (Fig. 3.8A). On the third day after surgery, the bird song started to recover and more notes were within the syllable boundaries (Fig. 3.8B). Figure 3.9 shows the syllable transition probabilities on the day before the surgery (the probability of every transition “X-Y” is calculated by dividing the number of occurrences for “X- Y” by the total number of transitions during that day of singing). The first eight blue bars are “typical” syllable transitions within a motif while the last blue bar (E-A) is the transition from

49

the last syllable of a motif to the first syllable of the next motif. The yellow bars represent transitions that were not listed by the user as typical, based on an initial examination of the sonograms. Finally, the red bars indicate transitions involving “NMS” notes that lie outside syllable boundaries. During the first day of singing following the HVC microlesion, motif transitions occur with low probabilities and non-motif transitions occur more frequently (Fig. 3.10). In fact, the highest transition probability is from one NMS note to another. During the second postoperative day of singing (Fig. 3.11), the probabilities of motif transitions increased (particularly transitions “I-A”, “A-I”, “A-B” and “E-A”) and the NMS transitions became more scattered. The frequent “N-N” transition that occurred on Post1 became less frequent and more “N-X” and “X-N” transitions began to occur as more notes fell within the syllable boundaries. The number of “NMS” transitions continue to decline during the third and eighth postoperative days of singing (Figs. 3.12 and 3.13) while the “I-A”, “A-I”, “A-B”, “B-C”, “C-D”, “D-E” and “E-A” transitions increased dramatically. By postoperative day 12 (Fig. 3.14), the overall pattern of syllable transition probabilities appears similar to the preoperative structure, although there are more “NMS” transitions on Post12 than on Pre1. SongSeq can also show the transition distributions in terms of pie charts (not shown). The dissimilarity in song sequence between the Pre and Post days is quantified using the K- L distance measure in Fig. 3.15. Here, Pre1 and Pre2 are preoperative days of singing, and P1 through P12 are the days of postoperative singing. Preoperative transition distribution functions are highly similar, so the K-L distance between Pre1 and Pre2 is near 0. On the first day of singing following surgery (P1) the K-L distance increases dramatically, since the typical transitions that occur on Pre1 are infrequent on P1. There is a big drop in the K-L distance between P1 and P2 due to recovery of the “I-A”, “A-B”, “A-I” and “E-A” transitions (Figs. 3.10, 3.11). Another big drop occurs between P2 and P3 due to the recovery of the “B-C”, “C-D” and “D-E” transitions and the further recovery of the “I-A”, “A-B”, “A-I” and “E-A” transitions (Figs. 3.11, 3.12). There is a third, but smaller, drop between P3 and P8 (Figs. 3.12, 3.13), making the typical transition probabilities in P8 more like those in Pre1. On subsequent days there is little change in the K-L distance.

50

Figure 3.16 shows a quantification of the spread of the transition probability distributions for each day of singing using the distribution entropy. The transition entropy values for Pre1 and Pre2 are similar, showing that the spread of the transition distribution function is similar on these two days. That is, the number of syllables and the order in which they are sung are similar. The entropy exhibited a dramatic drop on day P1 because the transition distribution on P1 has much less spread than that on any of the Pre days. Many notes sung on P1 were classified as NMS, so that many syllable transitions were lost. In fact, the NMS-NMS transition was most frequent on P1 (Fig. 3.10). By day P2 many syllables have returned, but they are coupled with “NMS” notes, resulting in a transition distribution with a larger spread as quantified by the larger entropy. The entropy on day P3 increases further as more typical transitions are recovered and the “NMS” transitions declined (Figs. 3.11, 3.12). The entropy on day P3 is higher than that on the Pre days since, in addition to the typical transitions, there are more “NMS” transitions on day P3 than on the Pre days. The entropy declines slightly after that and reaches a plateau that remains somewhat above the Pre days. This example illustrates the different types of information encoded in the K-L distance and entropy measures. Moreover, by examining the scatter plots for target files used to generate the entropy and the K-L distance values one can determine whether dissimilarity is due to increased variability in the phonology of motif syllables (change in size or shape of syllable clusters), the production of non-motif syllables (syllables that fall outside of the template clusters), or some combination of the two. Figure 3.17 shows linearity, consistency, and stereotypy scores (Scharff and Nottebohm 1991) generated for postoperative day 12. For every .wav file in the spreadsheet of day 12, the number of transitions is calculated as well as the number of transition types (the number of different transitions). Corresponding linearity, consistency and stereotypy scores are then generated based on these values. For any score that is larger than 0.85, the entry in the table is colored red, indicating that the song for this .wav file is highly stereotyped. Average linearity, consistency and stereotypy scores are then calculated and shown at the top. This procedure is done for every target file entered and thus for every preoperative and postoperative day entered by the user.

51

Figure 3.17: Linearity, consistency and stereotypy scores for postoperative day 12. For every .wav file in the spreadsheet of day 12, the number of transitions as well as the number of transition types is calculated. Corresponding linearity, consistency and stereotypy scores are then generated and listed in tabular form. For any score that is larger than 0.85, the entry in the table is colored in red, indicating that the song for this .wav file is highly stereotyped. Average linearity, consistency and stereotypy scores are then calculated and shown at the top.

Figure 3.18 shows a comparison of the average scores over all preoperative and postoperative days of singing (numbers removed for clarity). The microlesion clearly had a large impact on these measures of the song sequence during P1 and P2. The average consistency scores were higher than the average linearity scores on the Pre days. However, after surgery, and for the first five post operative days, the average linearity scores exhibit higher values than the average consistency scores. On day P6 the scores are almost equal, and for postoperative days six till twelve the consistency scores are again higher. As the bird song recovers the average scores increase gradually toward those on Pre 1.

52

Figure 3.18: Comparison of the average scores over all preoperative and postoperative days of singing. The microlesion clearly had a large impact on these measures of the song sequence during P1 and P2. However, as the bird song recovers the average scores increase gradually and by P12 they have returned to values similar to those on Pre 1. Numbers above data points are removed for clarity.

3.3.2 Developmental Changes in Syllable Sequence

In this next example we show the changes in the sequence of syllables sung by a bird during development. Figure 3.19 shows the 2D scatter plot of two selected acoustic features (syllable duration versus mean FM) for a day of singing from an adult male zebra finch.

53

Figure 3.19: Syllable Identification for a developing bird: The scatter plot of two selected acoustic features (syllable duration versus mean FM) for a day of singing from the bird when adult.

The data points on the scatter plot form five clusters (Fig. 3.19 – 3.20). There are five different motif syllables for this bird, and the canonical motif for this adult bird is ABCDCE. Thus, the typical transitions are “A-B”, “B-C”, “C-D”, “D-C”, “C-E” and “E-A”. The bird also often sings AABCDCE, making “A-A” an additional typical transition of his song. In Fig. 3.20, the five clusters are associated with syllable colors using the graphical tools as described earlier for Fig. 3.4.

54

Figure 3.20: Syllable Identification for a developing bird: The data points on the scatter plot form five clusters indicating five different motif syllables for this bird that form the template.

During the first week of development, the song is very disorganized and most of the notes produced do not fall within the syllable boundaries (Fig. 3.21). In fact, most of the data points in Fig. 3.21 represent acoustic features that do not resemble the acoustic features of the bird’s syllables as adult (Fig. 3.19) even though some of these data point lay within the syllable boundaries. These acoustic features however are due to the calls that the bird generate during his first week in development.

55

Figure 3.21: Syllable Identification for a developing bird: The scatter plot for week 1 is superimposed with the painted clusters of the template. During the first week of development, many notes were produced that did not fall within the syllable boundaries.

However, during the subsequent weeks of development, the bird song started to shape up gradually and merge toward the adult song, and by the sixth week most notes were within the syllable boundaries (Fig. 3.22). In particular, syllable C under the green boxes lies totally within the boundaries of that syllable, as well as syllable A under the blue boxes and syllable B under the black boxes. The remaining syllables are clustered and appear mostly within the boundaries of their corresponding boxes. During the subsequent weeks of development, these syllables appear to be totally within the box boundaries (not shown).

56

Figure 3.22: Syllable Identification for a developing bird: The scatter plot for week 6 is superimposed with the painted clusters of the template. During the sixth week, more notes were within the syllable boundaries.

The transition probabilities for the developing bird are shown in Figs. 3.23-3.28. As an adult, the bird song has high probability of motif syllable transitions and a low probability for non-motif syllable transitions (Fig. 3.23). In contrast, in the same bird as a juvenile, during a day of the first week of singing (post-hatch days 35-42) there are few motif transitions and many transitions associated with notes that fall outside the boundaries of the adult syllable clusters (Fig. 3.24).

57

Figure 3.23: Syllable transition probabilities are shown during a day when the bird is an adult.

Figure 3.24: Syllable transition probabilities are shown during a day of week 1.

The overall pattern of syllable transitions during weeks 2 and 3 was similar to week 1, exhibiting many transitions between an adult syllable and a “NMS” note (Figs. 3.25, 3.26).

58

Figure 3.25: Syllable transition probabilities are shown during a day of week 2.

Figure 3.26: Syllable transition probabilities are shown during a day of week 3.

At a day during week 6 of singing in the juvenile (Fig. 3.27), the probabilities of motif transitions increased and the “NMS” transitions became less frequent.

59

Figure 3.27: Syllable transition probabilities are shown during a day of week 6.

By week 9 (Fig. 3.28), the overall pattern of syllable transitions appears similar to the adult syntax structure: high probability for motif syllable transitions and low probability for non-motif transitions.

Figure 3.28: Syllable transition probabilities are shown during a day of week 9.

60

Figure 3.29 shows the vocal change in the syllable sequence over the course of development of the bird as quantified by K-L distance analysis. This is done by comparing transition probabilities of the juvenile bird from days of different weeks of singing with that of the adult bird. That is, the K-L distance at target compares the transition distribution from one

i day of week with that of the adult (the template). On a day during the first week of singing (W1), the K-L distance is large since few of the adult typical transitions are made by the juvenile bird. This is true also for singing on W2 and W3; almost all transitions involve “NMS” notes. Therefore, the K-L distance remains elevated, and is even larger on W3 than W2 due to a decline in the “B-C” transition from W2 to W3 (Figs. 3.25, 3.26). By W6, many of the adult transitions are present (Fig. 3.27), so the K-L distance is much lower.

Figure 3.29: During weeks 1 through 3, the K-L distance is large since the juvenile song is quite different from the adult song. By week 4, however, the sequence was much improved.

61

Figure 3.30 shows the transition entropy analysis quantifying the spread of the transition probability distributions for each week of singing. The entropy value on W1 is smaller compared with the adult, as one can observe by the smaller distribution spread in Fig. 3.24 versus Fig. 3.23. On subsequent weeks of singing, the entropy values first rise as more syllables emerge in the song, and then decline as the transition sequence becomes more like that of the adult. By W9 the entropy is approximately the same as that of the adult bird.

Figure 3.30: The entropy value on W1 is smaller compared with adult (Fig. 3.23 versus Fig. 3.22). On subsequent weeks of singing, the entropy values first rise as more syllables emerge in the song, and then decline as the transition sequence becomes more like that of the adult.

62

Average linearity, consistency and stereotypy scores for the juvenile versus the adult songs are shown in Fig. 3.31. The linearity score on W1 is almost the same as that of the adult, yet the song is very different. The high linearity on W1 occurs because the bird sings many single or double notes. The difference between the adult and W1 songs becomes evident, however, when one considers the consistency score, which is almost 0 on week 1. By W9 both linearity and consistency scores are similar to those of the adult song.

Figure 3.31: Average linearity, consistency and stereotypy scores for the juvenile versus the adult songs. Here, W1 through W9 represents a day during each of weeks 1 through 9.

63

3.4 Robustness

The results shown in our examples are robust to variations in the number of notes. To demonstrate this, we applied SongSeq on both birds’ data sets after removing 45% of the rows in each of the spreadsheets generated by Sound Analysis Pro (which removed 40-50% of the bouts on each day). The painted regions (and thus the syllable boundaries) on the template’s 2D scatter plot remained the same using a module of SongSeq that allows the user to upload a previous template (SongSeq saves the dimensions of every painted box on the template’s scatter plot each time a simulation is run). Figures 3.32, 3.33 and 3.34 show the K-L distance analysis, entropy analysis and average scores, respectively, for the first bird obtained after removing roughly half of the song bouts. The time courses of these measure are quite similar to those from the same bird when all song bouts were included (Figs. 3.15, 3.16 and 3.18 respectively).

Figure 3.32: Robustness of the results to variations in the number of notes. SongSeq was used on the bird 1 (HVC microlesion) data sets after removing 45% of the rows in each of the spreadsheets generated by Sound Analysis Pro. The K-L distance analysis obtained after removing 45% of the song bouts. The time course of this measure is quite similar to that from the same bird when all song bouts were included (Fig 3.15).

64

Figure 3.33: The entropy analysis obtained after removing 45% of the song bouts of bird 1 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig 3.16).

Figure 3.34: The average scores obtained after removing 45% of the song bouts of bird 1 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig 3.18).

65

Comparisons were also favorable when the same procedure was applied to the bird used in the development study (Figs. 3.35, 3.36 and 3.37, compare with Figs. 3.29, 3.30 and 3.31, respectively). Note that the corresponding values for the K-L distance, entropy and average scores at each week of singing is very similar to that before the knocking out 45% of the bouts. This shows first the bird’s songs are highly stereotyped so that even deleting a considerably large amount of his bouts would not abrupt the overall pattern of syllable sequencing.

Figure 3.35: Robustness of the results to variations in the number of notes: SongSeq was used on the developing bird data sets. The K-L distance analysis obtained after removing 45% of the song bouts. The time course of this measure is quite similar to that from the same bird when all song bouts were included (Fig. 3.29).

66

Figure 3.36: The entropy analysis obtained after removing 45% of the song bouts of bird 2 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig. 3.30).

Figure 3.37: The average scores obtained after removing 45% of the song bouts of bird 2 exhibits a similar time course of to that from the same bird when all song bouts were included (Fig. 3.31).

67

3.5 Disucssion

We have presented an automated tool, SongSeq, for analyzing birdsong syllable sequences. We showed how SongSeq can be used to monitor changes of sound features across a large number of songs, analyze transition distributions among syllables, quantify syllable ordering in terms of linearity, consistency and stereotypy scores, and quantify the degree of similarity in song syntax over different days of singing. The software uses two standard measurements from information theory, the Kullback-Leibler distance and entropy, to quantify the transition distributions and compare day-to-day differences. The Similarity Batch module in Sound Analysis Pro is a commonly used method to measure bird song similarity. For example, the method has been used to assess the vocal imitation of pupils (bird learning the song) from tutors (adult birds) (Tchernichovski et al. 2000) or recovery of song following brain injury (Coleman and Vu 2005; Thompson et al. 2007). Similarity Batch can be used to perform a large set of similarity measurements. It supports two batch modes: one is for comparing ordered pairs of sounds, and the other is for comparing sound matrices. This module is typically used to search for similarity between a single “target” motif and a .wav file comprised of multiple motifs or a set of uncategorized song units (e.g. destabilized singing following HVC microlesions). This motif-based comparison does not determine syllable-level contributions to similarity. In contrast, SongSeq’s algorithm is based on the individual syllable transitions; the entropy and K-L distance functions are based on the distributions of acoustic features of individual syllable transitions, and thus, dissimilarity between two transition probability distributions can be traced back to individual syllable contributions. SongSeq also provides an easy automated way to generate the linearity, consistency, and stereotypy scores. To our knowledge, the only automated tool available to compute stereotypy scores is the web-based program located at (http://bottjerlab.usc.edu/songinator.html). This tool computes linearity and consistency statistics for bird song but requires the user to manually enter the syllable order for every song file, and thus it requires user inspection of every sonogram. In contrast, SongSeq allows large scale computation of the stereotypy scores over a large number of .wav files (and thereby song motifs) and over a large number of days of singing (and thereby multiple SA+ spreadsheets). This is done without user interaction and the only requirement from

68

the user is to paint the clusters in the input stage to identify the syllables. Moreover, the average scores computed at the end of the batch process (Figures 3.18 and 3.31) provides an informative view of the behavior of the bird’s song over multiple days of singing. There is accumulating evidence that the bird song is coded at the sub-syllabic level (Day et al. 2009; Ravbar et al. 2012). During development, some subsyllables could form sooner than others, so that a syllable X’ could form early that is close to, but outside, the boundaries of adult syllable X. Rather than counting these notes in the “NMS” category, one could define the variant X’ in the template, by simply painting in the region of the template scatter plot where X’ is needed. Then in the adult song, there would be X syllables but few X’ syllables, while during development there would by many variants X’ of X recorded. The SongSeq software makes the identification of syllables much less labor intensive. However, there will always be misidentification of notes, just as there would be with manual identification. To reduce the frequency of misidentification, it is best to keep the syllable boundaries relatively tight. This is facilitated by using small painting boxes. A cloud of notes near, but not contiguous with, a note cluster may best be categorized as a variant X’ of the main syllable cluster X, as described above, rather than extending the boundaries of X. Also, by combining several days of singing one can get a clearer view of the syllable clusters, since the number of notes in the combined scatter plot is greater. This assumes, of course, that the song is similar on each day of singing that is combined (for example, combining days of singing from an adult bird prior to surgery). In summary, SongSeq automates the most labor intensive components of bird song analysis. In addition to automated syllable identification, it employs several algorithms to quantify song syntax. The user interface is through a GUI that requires no user programming or data manipulation. The software, along with online tutorial and a user manual, are available for free download at http://www.math.fsu.ed/~bertram/software/birdsong.

69

CHAPTER 4

CELLULAR LEVEL: CHARACTERIZATION OF HVC NEURONS

4.1 Introduction

Precisely how neurons in HVC encode song sequence and phonology is poorly understood. Models of how neurons might be organized and interconnected have been proposed (Abarbanel et al. 2004a; Abarbanel et al. 2004b; Abarbanel et al. 2012; Drew and Abbott 2003; Gibb et al. 2009a; Gibb et al. 2009c; Jin 2009; Jin et al. 2007; Katahira et al. 2007; Li and Greenside 2006; Long et al. 2010; Mooney and Prather 2005; Troyer and Doupe 2000b), and techniques for calibrating HVC models are being developed (Abarbanel et al. 2012), but a complete biophysical model is limited by the lack of detailed understanding of the properties of the component neurons. A goal of the present series of experiments is to detail the biophysical properties of cell types in HVC. There are three neuronal populations in the zebra finch HVC (Kubota and Taniguchi 1998; Mooney 2000; Mooney and Prather 2005): neurons that project to the RA

(HVCRA neurons), neurons that project to Area X (HVCX neurons), and interneurons (HVCINT neurons). The three types of HVC neurons have different functional and cellular properties which may be important for the different functions performed by the neural circuits within the HVC and for communication with the rest of the song system. Numerous in vivo and in vitro intracellular recording studies of HVC neurons have been carried out (Dutar et al. 1998; Katz and Gurney 1981; Kubota and Saito 1991; Kubota and Taniguchi 1998; Lewicki 1996; Lewicki and Konishi 1995; Long et al. 2010; Mooney 2000; Mooney et al. 2001; Mooney and Prather 2005; Schmidt and Perkel 1998; Shea et al. 2010; Solis and Perkel 2005; Wild et al. 2005). These studies shed light on several neuronal and circuit mechanisms and unveiled a variety of physiological properties within the HVC. For example, the brain slice studies demonstrated that HVCRA, HVCX, and HVCINT neurons have distinct, categorical electrophysiological phenotypes (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Mooney 2000; Mooney et al. 2001; Mooney and Prather 2005; Shea et al. 2010; Wild et al. 2005).

70

The studies published thus far have characterized HVC neurons according to their responses to injected current. HVCX neurons exhibit fast and time-dependent inward rectification where a sag appears in response to hyperpolarizing current pulses. In addition, HVCX neurons exhibit spike frequency adaptation (accommodation) in response to injected depolarizing currents. HVCRA neurons on the other hand have a more negative resting membrane potential than the other classes and do not exhibit a sag in response to hyperpolarizing current pulses. The RA-projecting neurons also fire with one or few action potentials in response to a relatively large depolarizing pulse, often with a delay. Finally, HVC interneurons exhibit a more prominent sag than HVCX neurons and produce rebound firing following hyperpolarizing current pulses. In response to depolarizing pulses, HVCINT neurons fire tonically with high firing frequency and with little or no adaptation. Despite the electrical identification of HVC neurons, the ionic currents underlying these behaviors in the different HVC neuron types remain largely unexamined and unknown. In this study, we obtained whole-cell current clamp recordings from neurons within the HVC to determine which ionic currents are responsible for their characteristic firing patterns. We also developed conductance-based models for the different neurons based on the identified ionic currents and their reported characteristics and then calibrated the models using data from our brain slice work. These models were then used to generate predictions about the frequency- response curve and the effects of blocking selected currents. Model predictions were then tested and verified in the slice using various pharmacological manipulations. The result is an improved characterization of the HVC neurons responsible for singing in the songbird.

4.2 Materials & Methods

4.2.1 Brain Slice Electrophysiology

4.2.1.1 Slice preparation. Adult male zebra finches (n = 34, > 120 d after hatch) raised in our breeding colony at Florida State University were used for experiments. Animal care and experiments were performed in accordance with National Science Foundation guidelines and approved by the FSU Animal Care and Use Committee. Birds were anesthetized with isoflourane and rapidly decapitated. The brain was quickly removed and placed in ice-cold artificial

71

cerebrospinal fluid (ACSF) that was pregassed with 95% O2-5% CO2. The hippocampus of each hemisphere was then resected over the general area of one side of the posterior lobe. This allowed us to clearly visualize HVC as an oval-like extrusion of the cerebrum which allowed for accurate sectioning (Fig. 4.1). Next, the brain was cut midsagittaly to separate the hemispheres. Each hemisphere was glued to a custom stage of a vibrating microtome immersed in ice-cold ACSF, and 175-300 µm parahorizontal slices were prepared. The ACSF used for slice preparation had equimolar sucrose partially replacing NaCl (Aghajanian and Rasmussen 1989) and contained (in mM): 72 sucrose, 83 NaCl, 3.3 MgCl2 , 0.5 CaCl2 , 1.0 NaH2PO4 , 26.2

NaHCO3, 22 glucose (osmolarity 285–295 mOsm). Slices were incubated for 15 minutes in the same sucrose ACSF solution at room temperature, followed by further incubation (45 min or more) in standard NaCl recording solution (with no sucrose) which contained (in mM): 119

NaCl, 2.5 KCl, 1.3 MgCl2 , 2.5 CaCl2 , 1.0 NaH2PO4 , 26.2 NaHCO3, 22 glucose (osmolarity 285–295 mOsm).

Figure 4.1: HVC in the slice. Under brightfield illumination, HVC has a darkened appearance due to high mylenation with striations of fibers.

72

4.2.1.2 Whole cell recording. Recordings were initiated after 1 hour of total incubation. One slice at a time was transferred to a submersion chamber where it was superfused with pregassed standard NaCl ACSF. The HVC was initially identified as a dark region in the slice (due to the heavy myelination of HVC) through a light microscope when transilluminated from below (Fig. 4.1). Whole cell patch-clamp recordings from HVC neurons were made with unpolished electrodes (4-λ MΩ). All electrodes were pulled with a Sutter Instruments (Novato, CA) P-80 micropipette puller. The electrodes were filled with the following intracellular solution

(in mM): 100 K-gluconate, 5 MgCl2 , 10 ethylene glycol-bis (aminoethyl ether) -N,N,N’,N’- tetraacetic acid (EGTA), 2 Na2- adenosine triphosphate (Na2-ATP), 0.3 Na3-guanosine 5’- triphosphate (Na3-GTP) , and 40 N-2-hydroxyethylpiperazine- N’-2-ethanesulfonic acid (HEPES), pH adjusted to 7.2–7.3 with KOH. A modified pipette solution was used during some experiments consisting of (in mM): 125 K-gluconate, 15 KCl, 1 MgCl2, 10 HEPES, 0.2 EGTA,

2 Mg-ATP and 0.3 Na3-GTP. No differences in recordings were observed between these solutions. Electrodes were injected with positive pressure and advanced through the slice until apposition to HVC neurons was clearly visualized (Fig. 4.1).

Figure 4.2: An electrode patched onto an HVC cell in the slice.

73

HVC cells were then slowly approached until the pipette resistance increased. Negative pressure was then applied until a gigaohm seal was obtained, and the patch membrane was ruptured by further application of negative pressure. Only cells that showed a stable membrane potential below -55 mV and that spiked readily in response to positive current injection were considered for further analysis. Voltage traces were corrected for an empirically measured liquid junction potential (+6 mV for standard ACSF and pipette solutions).Whole cell recordings of the membrane potential were made using a Multiclamp 700B (Axon Instruments, Foster City, CA), with an active bridge circuit for passing current, and digitized (Digidata 1322A, Axon Instruments) while connected to a PC running Axon pClamp 9 (Molecular Devices, Sunnyvale, California) acquisition software.

4.2.1.3 Pharmacological manipulations. Pharmacological tests used various drugs, which were delivered by bath application at the following concentrations: CNQX (5 and 10 µM, Tocris, 0190), picrotoxin (50 µM, Tocris, 1128), ZD 7288 (30 and 50 µM, Tocris, 1000), 4- aminopyridine (4-AP, 0.3 mM, Tocris, 0940), apamin (150 nM, Sigma, A1289), mibefradil (6 µM, Tocris, 2198), quinidine (100 µM, Tocris, 4108). The drugs CNQX, picrotoxin and quinidine were dissolved in DMSO (final concentration 0.1%) while ZD 7288, 4-AP and mibefradil were dissolved in water, and apamin was dissolved in 0.05M Acetic acid. All drugs were aliquoted and stored at -20 °C except for 4-AP which was kept at room temperature.

4.2.1.4 Electrophysiological identification of neurons. All HVC neurons in this study were classified based on their electrophysiological properties, as described previously (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Mooney 2000; Mooney et al. 2001; Mooney and Prather 2005; Schmidt and Perkel 1998; Shea et al. 2010; Wild et al. 2005). We applied three tests for the identification of HVC neurons on-line: The first test was the response obtained to depolarizing current pulses. In particular, HVCRA neurons fire only one to a few action potentials in response to +200 pA currents of ~ 0.5 s duration, whereas HVCX neurons fire more regularly with moderate spike-frequency adaptation, and HVCINT neurons fire at high frequency with little or no spike frequency adaptation (Dutar et al. 1998; Kubota and Taniguchi 1998; Mooney 2000; Mooney et al. 2001; Mooney and Prather 2005; Schmidt and Perkel 1998; Shea et al. 2010; Wild et al. 2005). The second test was the voltage response to hyperpolarizing current pulses. HVCINT neurons exhibit a prominent sag in response to

74

hyperpolarizing currents, while HVCX neurons exhibit a less prominent sag and HVCRA neurons exhibits no sag at all (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998).

The final test was an examination of the resting membrane potential. HVCINT neurons generally exhibit a resting membrane potential (RMP) of -60±6 mV, while HVCX neurons exhibit a RMP of -72±7 mV and HVCRA neurons exhibit a RMP of -85±6 mV (Dutar et al. 1998; Kubota and Taniguchi 1998). Spike afterhyperpolarization (AHP), time-to-peak (TTP), sag ratio (SR) and adaptation ratio (AR) physiological measurements were also computed off-line to help assess and confirm the on-line categorization. Spike threshold was computed as the peak of the second derivative of the voltage trace. Spike AHP and TTP were measured relative to the spike threshold point. The sag ratio SR during hyperpolarization was computed as where represents the

voltage at the end of the hyperpolarizing current pulse and represents the voltage at the nadir (the minimum voltage). The sag was computed for hyperpolarizing pulses of duration 500 msec and magnitude -200 pA. Spike frequency adaptation was quantified by computing the adaptation ratio AR between the last interspike interval and the first interspike interval for depolarizing current pulses of duration 500 msec and magnitude 150 pA. A ratio of 1 indicates that there is no adaptation, and larger magnitudes of AR indicate stronger adaptation. The electrophysiological recordings were first read with Clampfit 9.0 (Axon Instruments) and the traces were saved in Axon Text File (.atf) format. The .atf data was then read with a custom Matlab (Mathworks) routine and all further analysis was made using custom Matlab routines.

4.2.1.5 In vivo injections of retrograde tracers into Area X and RA. Physiological categorization of HVCX and HVCRA neurons was confirmed in two birds by recording from cells which were fluorescently labeled using a retrograde tracing dye that was injected into Area X or RA. The birds were deeply anesthetized with Equithesin (0.05 cc) and secured in the stereotaxic instrument, and the skull was exposed by making an incision down the center of the scalp and retracting the skin using curved forceps. The bifurcation at the midsagittal sinus was used as stereotaxic zero, and a small craniotomy over left and right Area X or RA was made using predetermined coordinates. Tracer (DiI, Life Technologies, ~400 nl) was then delivered into Area X (or RA) bilaterally via a glass micropipette attached to a gas pressure injection system (Applied Scientific Instrumentation MPPI-3).

75

After a 6 day period for tracer transport, the birds were used for brain slice electrophysiology studies. The results from one bird after DiI had been injected focally into Area X in vivo are shown in Fig. 4.4. Figure 4.4A shows the HVC under bright field illumination. As described earlier in Fig. 4.2, HVC has a darkened appearance dye to the high myelination (the dashed white line highlights the HVC region). Also shown in the slice are striations of fibers where afferent and efferent axons are projecting into and from HVC. Fig. 4.4B shows the same

HVC slice in Fig. 4.4A but under epifluorescence illumination. The red dots represent HVCX neurons that illuminates under epifluorescence because HVC neurons absorbed the fluorescent tracer DiI. As apparent, HVCX neurons are spanning the entire HVC and they could be readily observed in the slice preparation. Fig. 4.4C shows composite of the bright-field (Fig. 4.4A) and fluorescent (Fig. 4.4B) images showing a better image of HVCX neurons throughout the HVC. Figure 4.5 shows similar results after DiI injection into RA in vivo where the HVC slice is depicted under bright-field (Fig 4.5A), fluorescent (Fig 4.5B) and composites of the bright-field and the fluorescent (Fig 4.5C) illumination. HVC neurons that were retrogradely labeled were easily identified in the recording chamber when using epifluorescence illumination (Fig. 4.6).

Panel A shows an electrode patched onto an HVCX neuron under bright-field illumination from the same slice of Fig. 4.4. Under epifluorescence illumination (panel B), the same slice of panel

A shows six labeled HVCX neurons (including the neuron the electrode is patched onto). As a confirmation, panel C shows composite of the bright-field (A) and fluorescent images (B).

76

A

B

C

Figure 4.3: Anatomical confirmation of physiological classification of HVCX neurons. A. Under brightfield illumination, HVC has a darkened appearance with striations of fibers (the dashed white line highlights the HVC region). B. Focal infusions of DiI into Area X in vivo, retrogradely labeling HVCX cells that could be readily observed in the slice preparation under epifluorescence illumination. C. Composite of the bright-field (A) and fluorescent images show HVCX neurons throughout the HVC.

77

A

B

C

Figure 4.4: Anatomical confirmation of physiological classification of HVCRA neurons. A. Under brightfield illumination, HVC has a darkened appearance with striations of fibers. B. Focal infusions of DiI into RA in vivo, retrogradely labeling HVCRA cells that could be readily observed in the slice preparation under epifluorescence illumination. C. Composite of the bright-field (A) and fluorescent images show HVCRA neurons throughout the HVC.

78

A

B

C

Figure 4.5: Recording from fluorescently labeled HVC neurons. A. An electrode patched onto an HVCX neuron under bright-field illumination in the same slice of Fig. 4.3. B. The same slice under epifluorescence illumination showing six labeled HVCX neurons (including the neuron the electrode is patched onto). C. Composite of the bright-field (A) and fluorescent images confirming the HVCX neuron.

79

4.2.2 Computational Modeling

Single-compartment conductance-based biophysical models of cells from the HVC were developed, based on our current-clamp data. Simulations of these model neurons were performed using Matlab (Mathworks). The source codes containing the models are available at http://www.math.fsu.edu/~bertram/software/birdsong, as well as at ModelDB website located at http://senselab.med.yale.edu/modeldb/.

4.2.2.1 Model HVC cells. We used Hodgkin-Huxley-type models with additional currents added to reproduce features of the voltage traces observed in our current-pulse studies. The functional forms of activation/inactivation functions and time constants were based on published neural models (Destexhe and Babloyantz 1993; Dunmyre et al. 2011; Hodgkin and Huxley 1952; Terman et al. 2002; Wang et al. 2003), and the majority of our fitting parameters were maximum conductance, which are likely to vary among cell types. The sag seen in HVCX and HVCINT neurons (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998) suggests that a hyperpolarization-activated inward current conductance is present, and the post-inhibitory rebound firing seen in these neurons suggest that a low-threshold T-type Ca2+ current conductance may be present. Also, the delay to spiking seen in the response of HVCRA neurons to depolarizing pulses (Kubota and Taniguchi 1998; Mooney and Prather 2005) hints at expression of the A-type K+ current in these neurons. Moreover, the depolarization block and plateau potential that some HVC neurons exhibit in response to positive current pulses (Kubota and Saito 1991) is an indicator of the existence of a persistent sodium current or some other long-lasting inward current. The presence of some conductances had been shown in previous studies and in these cases they were included in our models. For example, Na+ - and Ca2+ - dependent K+ conductances have been identified before (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Schmidt and Perkel 1998), and a high-threshold Ca2+ conductance was shown to exist in HVC neurons (Kubota and Saito 1991; Long et al. 2010). With this background, the model was designed to include spike-producing currents ( and 2+ 2+ ), a high-threshold L-type Ca current ( ), a low-threshold T-type Ca current ( ), a 2+ + + + small-conductance Ca -activated K current ( ), a persistent Na current ( ), a Na - + + dependent K current ( ), an A-type K current ( ), a hyperpolarization-activated cation

80

current ( ), and a leak current ( ). The membrane potential of each HVC neuron obeys the current balance equation:

where represents the constant applied current.

4.2.2.2 Voltage-gated ionic currents. The constant-conductance leak current is

. The voltage-dependent currents have non-constant conductances with

activation/inactivation kinetics characterized as fast, instantaneous or slow as described below. The voltage-dependent currents are:

where

and where is the half activation (or inactivation) voltage for gating variable and is the

T slope factor for that variable. The term is the thermal voltage, where is the temperature of the bathing solution (25 , R is the gas constant, and F is Faraday’s constant. 2+ + is the external Ca concentration,℃ which K is 2.5 mM in the bathing solution. The persistent Na current is modeled as in Dunmyre et al (2011) with an instantaneous activation and a slow

81

inactivation variables. The steady state activation functions for the fast gating variables that are treated as instantaneous are shown in Fig. 4.7A ( (black), (blue), (cyan), and

(green)).

The gating variables , , and are slower and have first-order kinetics governed by

where for , and is given above by (7), and for is given as follows

where

and

The steady state activation for the slow variable is shown in Fig. 4.7B (solid magenta), as are the inactivation functions for the slow variables (dotted black), (dotted blue) and (dotted green). Activation functions are plotted as solid lines while inactivation functions are dotted. The time constants and are given in Table 4.1 while and are voltage- dependent and given by

82

Fast Gating Variables A

Slow Gating Variables B

Figure 4.6: Steady-state activation functions for the fast gating variables (A), the slow gating variables (B, solid lines) and the steady-state inactivation functions (B, dotted lines).

83

4.2.2.3 Low-voltage activated T-type Ca2+ current. The low-voltage activated T-type Ca2+ current is modeled similar to Terman et al (2002), except that the single-channel current is described by the Goldman-Hodgkin-Katz formula,

The activation gating for the rapidly activating channel ( ) is treated as instantaneous

(Fig. 4.7A, magenta) and is given by

The inactivation variable ( ) is also treated as instantaneous and depends on the slowly operating variable , which reflects the availability of the current. This is given by

The slowly operating gating variable is governed by

where

84

and

Fig. 4.7B shows as a function of voltage when is at equilibrium (dotted cyan). We

2+ used the GHK (Goldman –Hodgkin–Katz) form to model the L-type and T-type Ca currents. The ohmic form is another common form in the neural modeling literature, but the GHK form is often a better and more biologically realistic representation if one wishes to focus on the effects of the Ca2+ current. However, although we do not believe it is necessary to use the GHK form in this study as it is a more complex representation, using GHK form would provide more biologically accurate model neurons and give insights into the details of the Ca2+ currents in these neurons.

4.2.2.4 Ca2+-dependent K+ Current. The small-conductance Ca2+-dependent K+ current is modeled as

where is the steady state activation function of the SK current that is based on the levels of intracellular calcium (Fig. 4.8A) and given by

The constant is the dissociation constant of the calcium-dependent current and is the 2+ intracellular concentration of free Ca ions and governed by

85

The constant represents the fraction of free to total cytosolic Ca2+ while the constant combines the effects of buffers, cell volume, and the molar charge of calcium. Also, the constant

is the calcium pump rate constant, and represents the basal level of Ca2+.

A

B

2+ + Figure 4.7: Ca dependence of the SK current gating function (A) and Na dependence of the KNa current (B).

86

4.2.2.5 Na+-dependent K+ current. The Na+-dependent K+ current, modeled as in Wang et al. (2003), is given by

The activation function is given by

and plotted in Fig. 4.8B. is the intracellular concentration of Na+ ions and governed by

The influx of is controlled by . The Na+ concentration in HVC -1 2 neurons has not been measured, so we choose mM (msec µA) cm so that the increase of is ~ 100 µM per action potential, which is similar to what has been reported in hippocampal pyramidal neurons (Jaffe et al. 1992; Rose et al. 1999; Rose and Ransom 1997).

The extrusion of is assumed to be largely due to a Na+/K+ ionic pump, which extrudes + + three Na ions for every two K ions brought into the cell (Fain 1999). The extrusion by the ion pump was modeled in (Li et al. 1996) as where

The sodium concentration at the resting state is assumed to be mM. The slow + + kinetics of the Na /K ionic pump is important for the model and plays a role in adaptation.

87

4.2.2.6 Hyperpolarization-activated inward current. The hyperpolarization activated inward current’s activation is modeled as in Destexhe and Babloyantz (1993) using a fast component ( ) and a slow component ( ) as follows

The fast activation component obeys

where

is shown in Fig. 4.7A (red) and its time constant is given by

Similarly, the slow activation component obeys

where

88

is shown in Fig. 4.7B (solid red) and is given in Table 4.1.

Manual adjustment of model parameters was performed to qualitatively reproduce membrane potential trajectories. The parameters that were manually adjusted and tuned are listed in Table 4.2. HVC neurons of the same type behaved similarly in response to depolarizing and hyperpolarizing current pulses, as described earlier. A candidate neuron of each HVC cell type was selected and the model parameters were fit by eye using iterative manual tuning. For example, the adaptation seen in HVCX and HVCRA neurons was achieved by tuning the gSK and gKNa parameters. Similarly, the sag seen in HVCX and HVCINT neurons was achieved by tuning the kr and gh parameters. The features that were the target of the fit were the number of spikes, the shape of the spikes, the steepness of the sag, and the strength of the rebound firing. Fixed parameter values for HVC neurons used in the simulations are given in Table 4.1 and parameters that vary between the different model neurons are shown in Table 4.2.

Table 4.1: Parameter values used in all simulations

Parameter Value Parameter Value

-70 mV 0.4 mV

-90 mV -67 mV

50 mV 68 mV

85 mV -5 mV

-30 mV -5 mV

2 nS -0.05 mV

89

Table 4.1 - continued

Parameter Value Parameter Value

19 nS -6 mV

1 nS 6 mV

10 msec -10 mV

1000 msec 5 mV

20 msec 5 mV

1 msec 25 mV

1500 msec -7.8 mV

200 msec -0.1 mV

87.5 msec 2 mV

-35 mV 2.2 mV

-30 mV 0.1

-20 mV 0.0015 pA-1µM msec-1

-40 mV 0.3 msec-1

-48 mV 0.1 M

-20 mV 0.5 µM

-60 mV 0.0006 mM/msec

-105 mV 15 mM

-105 mV 100

-65 mV

90

Table 4.2: Parameter values that vary among neuron types

HVCX HVCRA HVCINT

450 nS 300 nS 800 nS

50 nS 400 nS 1700 nS

6 nS 27 nS 1 nS

40 nS 500 nS 1 nS

4 nS 1 nS 4 nS

5 nS 150 nS 1 nS

2.7 nS 0.6 nS 1.1 nS

100 pF 20 pF 75 pF

0.3 0.95 0.01

91

4.3 Results

Our goal was to characterize the mix of ionic currents present in HVCRA, HVCX and

HVCINT neurons. To facilitate this, we developed biophysical models of the three cell types using current clamp data, and then used these to predict and interpret the effects of pharmacological blockers. All slice recordings were done using the whole-cell patch clamp technique. These data are based on intracellular recordings from 105 neurons from 34 birds. In the figures, when experimental data was used to calibrate the model, the data are shown first. When the model was used to predict responses to manipulations, the model simulation is shown first.

4.3.1 HVCX Neurons

4.3.1.1 Response to applied current. The X-projecting neurons (n=47) have two key identifying features. The first is the spike frequency adaptation (SFA) observed in response to depolarizing current pulses (Fig. 4.9A). Upon stimulating an HVCX neuron with a relatively weak depolarizing pulse, the neuron fires with high frequency and quickly switches to a lower frequency that gradually decreases over the course of current application (n = 46). The adaptation ratio (AR, subsection 4.3.1.4) for HVCX neurons was 3.91 ± 2.1 (n = 13). A second feature of HVCX neurons is a moderate sag generated in response to hyperpolarizing current pulses (Fig. 4.9B), followed by rebound firing or rebound depolarization upon the termination of the pulse (n=45). The sag and the rebound were larger as for larger hyperpolarization current pulses. The sag ratio for this neuronal population (SR, section 4.3.1.4) was 0.10 ± 0.014 (n = 19).

In addition to these two characteristic features that readily identify an HVCX neuron, they also exhibit a variety of other electrical properties detailed in Table 4.3. In response to depolarizing current pulses, the spikes emanate from a depolarized plateau (Fig. 4.9A) and exhibit a large- amplitude AHP (14.5 ± 0.4 mV, n = 24) and a slow time-to-peak (13 ± 1.2 ms, n = 26). A subset of the HVCX neurons (n=7) exhibited spontaneous firing, which was quickly abolished after dual application of the AMPA/kainate receptor antagonist CNQX (5 µM, n=4 and 10 µM, n = 3) and the GABAA receptor antagonist PTX (50 µM, n=7) (not shown). Spontaneous activity was absent in an additional 14 HVCX neurons which were recorded in the presence of these receptor antagonists, suggesting that HVCX neurons are silent in the absence of synaptic drive.

92

Experiment A

B

Figure 4.8: Firing properties of an actual X-projecting neuron. A. An X-projecting neuron exhibits some spike frequency adaptation in response to a depolarizing current pulse (150 pA). B. A weak sag followed by postinhibitory rebound firing is generated in response to a hyperpolarizing current pulse (-200 pA).

93

Table 4.3: Experimental characterization of the three types of HVC neurons. Abbreviations: AHP (after hyperpolarization), TTP (time-to-peak), AP (action potential), RMP (resting membrane potential), AR (Adaptation Ratio), SR (Sag Ratio). (x/y) means x number of neurons out of the y number of neurons tested exhibit the corresponding property. (n=x) means x number of neurons were used to calculate the corresponding ratio.

Electrophysiological property Blocked by drug(s)

Spike frequency adaptation (46/47) Spike from depolarized plateau (42/47)

Large AHP (24/47) Apamin (4/4)

Slow TTP (26/47) AR of 3.91 ± 2.1 (n=13)

neurons Moderate sag (45/47) ZD 7288 (8/8) X SR of 0.10 ± 0.014 (n=19)

HVC Rebound firing (35/47) Mibefradil (6/7) Rebound depolarization (9/47) Mibefradil & ZD 7288 (3/3) Spontaneous activity (7/47) CNQX & PTX (7/7) Low excitability (33/33) Apamin (6/6), Quinidine (3/3)

Delayed spike (14/33) 4-AP (5/8) Spikes from depolarized plateau (32/33) Fast large AHP (17/33) 4-AP (4/8) Fast TTP (19/33)

neurons Absence of sag (33/33) RA SR of 0.009 ± 0.003 (n=11)

HVC Fast inward rectification (23/23) Absence of rebound firing (30/33) Hyperpolarized RMP (33/33) Apamin (6/6), Quinidine (3/3), 4-AP (8/8) High firing frequency (25/25) Sharp downstrokes of AP (25/25) Spike amplitude of 78±5 mV (18/25) Spike amplitude of 55±8 mV (7/25) Large AHP (16/25) neurons neurons Short TTP (17/25) INT Prominent sag (25/25) ZD 7288 (5/5)

HVC SR of 0.24 ± 0.065 (n=9) Rebound firing (25/25) Mibefradil (4/4) & ZD 7288 Spontaneous activity (18/25)

94

These features were useful in distinguishing the X-projecting neurons from the other classes of HVC neurons on-line, and these classification criteria were confirmed by recording retrogradely labeled HVCX neurons. Figure 4.10B shows the firing pattern of a retrogradely labeled HVCX neuron (same neuron in Fig. 4.10A) that exhibits similar properties to the ones identified off-line.

A B

C D

Figure 4.9: Electrical recordings from retrogradely labeled HVC neurons. A. Same neuron in Fig 4.5C. B. Recording from the labeled neuron in A showing a firing pattern that is characteristic of HVCX neurons (n=5). C. Recording from an HVCRA neuron that was labeled using similar methods shows its characteristic properties (n=2). D. Recording from a neuron that was not fluorescently labeled shows properties that resembles those of an HVC interneuron.

Next, model HVCX neuron parameters were calibrated to fit the voltage traces. Figure

4.11A shows the firing pattern of a model HVCX neuron in response to a 150 pA applied current. The model neuron exhibits spike frequency adaptation that is due to the Ca2+ - dependent K+ + + current (ISK) and the Na - dependent K current (IKNa). The model was also calibrated to fit the sag and the rebound firing seen in HVCX neurons (Fig. 4.11B). In the model, the sag is primarily due to the Ih current that is activated upon hyperpolarization and gradually depolarizes the

95

neuron. The prominence of the sag is controlled by the parameter that governs the weight put on the fast component relative to the slow component of the Ih current. For model HVCX neurons, is set to 0.3, thereby giving 70% of the weight to the slow component of the current that is responsible for the slow depolarization that characterizes the sag. The rebound firing is 2+ primarily due to the low-threshold T-type Ca current (ICa-T) in the model, working in cooperation with Ih. The hyperpolarizing current pulse activates Ih and removes the inactivation of ICa-T. Upon removal of the hyperpolarizing current, Ih and ICa-T together depolarize the cell beyond its resting potential and beyond the spike threshold. The rebound spiking terminates as Ih deactivates and ICa-T inactivates. Longer or stronger current application augments deinactivation of ICa-T, enhancing the rebound.

Model A

B

Figure 4.10: Firing properties of a model X-projecting neuron. A. An X-projecting neuron exhibits some spike frequency adaptation in response to a depolarizing current pulse (150 pA). B. A weak sag followed by postinhibitory rebound firing is generated in response to a hyperpolarizing current pulse (-200 pA).

96

As a test, we used the calibrated model to predict the average firing frequency over a 1 sec duration of applied current for a range of current magnitudes and compared this prediction to the response of actual HVCX neurons (Fig. 4.12). Both model and actual neurons had a roughly linear frequency response that ranged from 5 Hz to 40 Hz. Since the frequency response of the model was not a feature used in the calibration, the good fit to experimental data provides support for the model and the manner in which it was calibrated.

Figure 4.11: Frequency-current relationship in X-projecting neurons. The firing frequency of a model HVCX neuron under various current injections (triangles) closely matches the data from X-projecting neurons in the slice (circles, n=21, mean ± SE). For both model and actual HVCX neurons, the duration of applied current was 1 sec.

4.3.1.2 H current and inward rectification in HVCX neurons. The sag and postinhibitory rebound produced by HVCX neurons are examined further in Fig. 4.13. Panel A shows the current-voltage relationship of the model HVCX neuron, using the same parameters as in Fig. 4.11A. The solid curve with circles represents the voltage at the end of the hyperpolarizing current pulses, while the dashed curve with triangles is the voltage at the nadir (the minimum voltage). As the magnitude of the current pulse increases, the difference between

97

the voltage responses at the end of the current pulse and at its nadir increases, indicating an increase in the sag. In the slice, HVCX neurons showed a similar behavior (n=35). A representative example is shown in Fig. 4.13B.

A Model Model

B Experiment

Figure 4.12: Inward rectification in HVCX neurons. A. Current-voltage relationship of a model HVCX neuron using same parameters as in Fig. 4.10A. Circles with solid curve represent voltage responses measured at the end of the current pulses while triangles with dashed curve represent voltage at the nadir. B. Current-voltage relationship for an actual HVCX neuron.

98

The sag generated in the model HVCX neuron is due to the hyperpolarization-activated inward current (Fig. 4.14A) that also contributes to the rebound (black trace). The inset of Fig.

4.14A shows the total Ih conductance gh. It is near 0 at rest, but increases upon hyperpolarization.

The fast component causes a rapid initial increase in gh, followed by a slower rise due to the slow component. This slow rise is reflected in the voltage as a sag. When the current pulse is terminated, the rapid component of gh quickly goes to zero, while the slow component remains. It is this residual slow component that contributes to the rebound firing.

A Model Model

B Experiment

Figure 4.13: Ih current in HVCX neurons. A. The sag generated in the model HVCX neuron is due to the hyperpolarization activated inward current (Ih, with gh = 4 nS) that is contributing partially to the rebound spike (black trace). Blocking Ih by setting gh to 0 eliminates the sag but keeps a small rebound depolarization due to ICa-T (red trace). Figure inset shows the total Ih conductance. B. Blocking Ih using ZD 7288 (50 µm) eliminates the sag and the rebound firing (red trace, 20 minutes after ZD 7288 application) in the same X-projecting cell used in Fig. 4.8B, matching model predictions.

99

Blockade of Ih is simulated in the model by setting gh = 0. This results in a small decline in the resting membrane potential, and a much larger hyperpolarization during the current pulse

(Fig. 4.14A, red trace). This reflects the absence of the fast component of Ih. The sag is gone and there is no rebound, reflecting the absence of the slow component of Ih. There is, however, a small voltage rebound depolarization due to the ICa-T current, as is discussed shortly. Model predictions were tested and verified in the slice by pharmacologically blocking the

Ih current using ZD 7288 (50 µm, Fig. 4.14B). ZD application eliminated the sag and the weak rebound firing but kept a small rebound depolarization (red trace, 20 minutes after drug application) in the same X-projecting cell used in Fig. 4.9B. Moreover, the resting membrane potential was hyperpolarized by ~ 5 mV. Similar behavior was observed in all of the 7 additional

HVCX neurons examined, 2 of which were given a lower concentration of ZD 7288 (30 µm).

Therefore, the sag seen in HVCX neurons is due to the hyperpolarization-activated inward current, which also contributes to rebound firing.

4.3.1.3 Actions of Ih and ICa-T in HVCX neurons. The interplay between the Ih and ICa-T currents and their role in shaping the firing pattern of HVCX neurons was further examined.

Figure 4.15A shows a model HVCX neuron exhibiting a sag and rebound firing (black trace, gK =

1700 nS, gSK = 1 nS, Iapp = -120 pA). Setting gCa-T to 0 eliminates the rebound firing and any rebound depolarization (not shown). The rebound firing can be eliminated and a voltage depolarization retained by partial blockade of ICa-T (gCa-T = 0.1 nS, 3.3% of original conductance), which leaves the sag unaltered (red trace). The sag and rebound depolarization are eliminated by additional blockade of Ih (gh = 0 nS, green trace). The figure inset shows the total

ICa-T current conductance (gCa-T) in the model HVCX neuron under the three conditions. The conductance is near 0 during the hyperpolarization since the channel is deactivated. However, it also loses any inactivation, so when the applied hyperpolarizing current is removed, the Ca-T current quickly activates and depolarizes the cell to spike threshold (black trace). When the current is mostly blocked (red trace), its residual total conductance has sufficient strength to activate and depolarize the model cell; however, it is still too small to bring the cell to spike threshold. When Ih is blocked, the voltage does not go high enough after the pulse termination to activate the Ca-T current, so gCa-T remains low.

100

A Model Model

B Experiment

Figure 4.14: Ih and ICa-T mechanisms in HVCX neurons. A. Black trace shows a model HVCX neuron exhibiting a sag and rebound firing (gK = 1700 nS, gSK = 1 nS, Iapp = -120 pA). The rebound firing is eliminated by partial blockade of ICa-T (gCa-T = 0.1 nS, 3.3% of original conductance), leaving a small rebound depolarization and keeping the sag intact (red trace). The sag and rebound depolarization are eliminated by additional blockade of Ih (gh = 0 nS, green trace). Figure inset shows the total ICa-T current conductance (gCa-T) in the model HVCX neuron under the three conditions. B. In an HVCX neuron, model predictions are matched by blocking ICa-T (red trace) using Mibefradil (6 µm, 40 minutes after drug application, Iapp = -120 pA). A subsequent application of ZD 7288 (50 µm, 30 minutes after ZD application) to block Ih eliminates the sag and the rebound depolarization (green trace).

101

Model predictions were matched in the slice by blocking ICa-T (Fig. 4.15B, red trace) using mibefradil (6 µm, 40 minutes after drug application, Iapp = -120 pA) (Lacinova 2004; 2005). Mibefradil kept the sag intact, eliminated the rebound firing and left a rebound depolarization. A subsequent application of ZD 7288 (50 µm, 30 minutes after ZD application) to block Ih eliminated the sag and the rebound depolarization (green trace). Similar results were observed in

2 other HVCX neurons that were tested. Mibefradil eliminated rebound spiking in 3 additional neurons, reaffirming that rebound firing in these neurons requires the low-threshold T-type Ca2+ current.

4.3.1.4 ISK and IKNa in HVCX neurons. The spike frequency adaptation seen in HVCX neurons was further examined. Figure 4.16A shows a model HVCX neuron exhibiting adaptation, using the same parameters as in Figure 4.11 but with gNa=800 nS and gSK=5 nS. Blocking the 2+ + Ca -dependent K current by setting its conductance (gSK) to zero increases the model neuron’s excitability slightly, but the adaptation is retained due the presence of the Na+-dependent K+ current (Fig. 4.16B). Moreover, the resting membrane potential increased by a few millivolts (~3 mV), and the shape of the interspike voltage time course has changed.

Model predictions were tested in the slice by blocking ISK using apamin (150 nM). Figure

4.17A shows a HVCX neuron exhibiting spike frequency adaptation. Apamin application had little effect on adaptation, and changed the shape of the interspike voltage time course, as in the model (Fig. 4.17B). Moreover, the resting membrane potential was depolarized by ~ 5 mV. Unlike the model response however, apamin application reduced the AHP following each action potential significantly. Similar results were observed in 3 additional HVCX neurons that were tested.

These results suggest that the spike frequency adaptation in HVCX neurons is not mediated by SK current, but the AHP following each action potential is. This agrees well with findings in Kubota and Saito (1991) and Schmidt and Perkel (1998). The former identified a Na+-dependent K+ conductance in HVC neurons, but due to the nonselectivity of pharmacological blockers of

KNa current we did not pursue this further.

102

Model

A

B

Figure 4.15: ISK current in model HVCX neuron. A. A model HVCX neuron exhibits spike frequency adaptation (Iapp = 150 pA). B. Blocking the SK current by setting its conductance (gSK) to zero increases the model neuron’s excitability slightly, but the adaptation is preserved. Also, the resting membrane potential is increased by a few millivolts (~3 mV), and the interspike voltage time course changed its shape.

103

Experiment

A

B

Figure 4.16: ISK current in actual HVCX neurons. A. An actual HVCX neuron exhibits spike frequency adaptation for a current pulse of 100 pA. B. Blocking the SK current using apamin (150 nM) had little effect on adaptation, changed the shape of the interspike voltage time course, and reduced the AHP following each action potential. It also depolarized the resting membrane potential by ~ 5 mV.

104

4.3.2 HVCRA Neurons

4.3.2.1 Response to applied current. The RA-projecting neurons (n=33) have three key identifying features. The first is their relative lack of excitability in response to depolarizing current pulses (Fig. 4.18A). The neuron fires with one (n=17) or few spikes (n=12) in response to a relatively large depolarizing pulse. The second is the absence of sag in response to hyperpolarizing current pulses (n=33, Fig. 4.18B), along with an absence of rebound firing upon the termination of the hyperpolarizing pulses (n=30). The sag ratio (SR) in HVCRA neurons was 0.009 ± 0.003 (n = 11). In a small subset of these neurons however, a small rebound depolarization was seen (n=3). The third identifying feature is the extremely hyperpolarized resting membrane potential of -85±6 mV (n=33). In addition to these features, the firing pattern of RA-projecting neurons exhibits other important characteristics (Table 4.3). There is often a delay to spiking in response to depolarization (n=14). In a subset of these neurons, tonic firing was observed (n=4). Moreover, the spikes emanate from a depolarized plateau but have sharp downstrokes, where each spike is followed by a fast AHP that exhibits a large amplitude (16 ± 0.8 mV, n=17) and a short time-to- peak (3 ± 0.4 ms, n=19). Although no sag is present in response to hyperpolarizing current pulses, the fact that the spacing between voltage traces becomes smaller with greater hyperpolarizing currents indicates that there is fast inward rectification (n=23). That is, a depolarizing current is being activated at the lower voltages. These features were very useful in distinguishing the RA-projecting neurons from the other classes of HVC neurons on-line, and these classification criteria were confirmed by recording retrogradely labeled HVCRA neurons (Fig. 4.10C). The recording from the RA-projecting neuron in Fig. 4.10C was done in the brain slice of Fig. 4.5. Note the similarity between the firing pattern of the labeled neuron in Fig. 4.10C and the firing pattern of the neuron in Fig. 4.18A. Both neurons, exhibit an extremely hyperpolarized resting membrane potential and a long delay to spiking. Moreover, the labeled neuron showed no sag to hyperpolarizing current pulses (not shown).

105

Experiment A

B

Figure 4.17: Firing properties of an actual RA-projecting neuron. A. An RA-projecting neuron fires a single action potential, often with a long delay, in response to a relatively large depolarizing pulse (175 pA). B. No sag is present in response to hyperpolarizing current pulses (-160 pA to -20 pA, steps of 20).

106

Model HVCRA neuron parameters were calibrated to fit the voltage traces in response to a depolarizing and a hyperpolarizing current pulse. Figure 4.19A shows the firing pattern of a model HVCRA neuron in response to a 150 pA applied current. The model neuron exhibits delay + to spiking that is due to the A-type K current (IA), which contributes to the lack of excitability with the cooperation of ISK and IKNa. The inset of Fig. 4.19A shows the total IA current conductance gA. It is near 0 at rest, but increases rapidly upon depolarization due to fast activation. This rapid increase halts after a few milliseconds and switches to a slow decrease that is due to slow inactivation. The slow decrease is reflected in the voltage trace as a slow depolarization in the membrane potential, and this allows the model neuron to escape the inhibition produced by IA and fire a delayed spike. The model was also calibrated to fit the voltage response to hyperpolarizing currents (Fig. 4.19B). In the model, the absence of the sag is primarily due to the large value of , which is set to 0.95, thereby giving almost all the weight to the fast component of the Ih current and very little to the slow component. The absence of the rebound is due to the small values of the T-type Ca2+ current conductance (gCa-T = 1 nS) and Ih current conductance (gh = 1 nS), which prevent the model cell from depolarizing beyond its resting potential and spike threshold. The small Ih that is present, however, is sufficient to produce inward rectification of the membrane upon hyperpolarization; as the membrane is hyperpolarized to lower voltages, Ih activation increases, reducing the spacing between the voltage curves. If gh is set to 0, inward rectification disappears and the spacing between the voltage traces becomes equal in response to equal steps of hyperpolarizing current pulses (not shown). We used the model neuron to predict the average firing frequency over a 1 sec duration of applied current for a range of current magnitudes and compared this prediction to the response of actual HVCRA neurons (Fig. 4.20). Both model and actual HVCRA neurons had a roughly linear frequency response that ranged from 3 Hz to 25 Hz. For pA, both model and actual

HVCRA neurons did not spike during the depolarization, so no average firing frequency was computed. For pA, the model neuron showed no activity, but the actual neurons fired with very low frequency. Overall, the predicted model neuron response agreed well with the frequency responses of the actual HVCRA neurons (n=6).

107

Model A

B

Figure 4.18: Firing properties of a model RA-projecting neuron. A. HVCRA model neuron parameters are calibrated to match the experimental recording (Iapp = 150 pA). The long delay to spiking is due to the A- + type K current (IA). Figure inset shows the total IA conductance during the current pulse. B. For the same parameter values used in A, the HVCRA model neuron has no sag, but exhibits inward rectification in the spacing between voltage traces, as in Fig. 4.17B.

108

Figure 4.19: Frequency-current relationship in RA-projecting neurons. The firing frequency of a model HVCRA neuron under various current injections (triangles) closely matches the data from RA-projecting neurons in the slice (circles, n=6, mean ± SE). For both model and actual HVCRA neurons, the duration of applied current was 1 sec.

4.3.2.2 IA current in HVCRA neurons. The delay to spiking in HVCRA neurons was further examined. Figure 4.21A shows a model HVCRA neuron with delay to spiking (gKNa =

1000 nS, Iapp = 200 pA). Setting gA to 0 eliminates the delay and increases excitability (Fig.

4.21B). Moreover, the resting membrane potential is slightly depolarized when IA is eliminated.

Figure 4.22 shows an actual HVCRA neuron before (panel A) and after (panel B) application of the IA blocker 4-AP (0.3 mM, Iapp = 200 pA). Drug application eliminated the delay to spiking, increased the neuron’s firing frequency and left a strong adaptation. Moreover, the resting membrane potential was depolarized by ~ 3 mV. Unlike the model response however, 4-AP reduced the afterhyperpolarization (AHP) following each spike. Similar results were observed in 4 additional HVCRA neurons that were tested.

109

Model

A

B

Figure 4.20: The effect of IA current in model HVCRA neuron. A. In a model HVCRA neuron there is a delay to spiking following application of a depolarizing current (Iapp = 200 pA). B. Blocking IA eliminates the delay to spiking. The adaptation present at the beginning of the voltage trace after blockade is due to the Na+-dependent and Ca2+-dependent K+ currents.

110

Experiment

A

B

Figure 4.21: The effect of IA current in actual HVCRA neurons. A. An RA-projecting neuron exhibits a delay to spiking following application of a depolarizing current (200 pA). B. Model predictions are verified experimentally by the application of 4-AP (0.3 mM) that eliminated the delay to spiking. The resting membrane potential increased after drug application in both the model and the actual cell.

111

Larger dosages of 4-AP as well as long durations of application switched the neurons’ firing pattern from tonic spiking to episodes of large plateau oscillations (n=3). Similar oscillations were observed following application of quinidine (250 µm) in an attempt to block the IKNa pharmacologically. This also caused a dramatic increase in the resting membrane potential (~12±3 mV increase, data not shown). Quinidine is known to be a non-selective blocker of K+ currents, as is 4-AP at high dosages. It is therefore hard to know what mixture of blocked currents produces this behavior. Bursting patterns were produced in our model HVCRA neuron by setting gA = gKNa = 0 and reducing the magnitude of gK to 15 nS. However, this bursting pattern did not fully resemble the plateau oscillations produced by the actual neuron. Due to the ambiguity in current blockage, we did not pursue this further.

2+ 4.3.2.3 ISK current in HVCRA neurons. We next investigated what role the Ca - + dependent K current plays in the lack of excitability of HVCRA neurons. Figure 4.23A shows a model HVCRA neuron using the same parameters as in Fig 4.21A, but with gKNa = 100 nS, gA = 0 2+ + nS, gSK = 35 nS and gCa-T = 9 nS. Blocking the Ca -dependent K current (gSK = 0) caused a dramatic increase in excitability, decreased the spike amplitude slightly, and increased the resting membrane potential by ~ 8 mV (Fig. 4.23B).

Figure 4.24A shows an actual HVCRA neuron’s response to 150 pA. This neuron does not exhibit a delay to spiking (n=12 HVCRA neurons had no delay). Application of apamin (150 nM) greatly increased the neuron’s excitability, decreased the spike amplitude, and depolarized the resting membrane potential by ~ 5 mV (Fig. 4.24B). However, unlike the model response, no adaptation was seen (AR = 1.04). Similar results were observed in 3 additional HVCRA neurons

(one where adaptation was present). In 2 other HVCRA neurons, apamin application had similar effects, but there was a very high plateau potential and the amplitude of the spikes was much smaller. These results show that the SK current plays a key role in damping the excitability of

HVCRA neurons. It also contributes to maintaining the extremely negative resting membrane potential of these neurons.

112

Model

A

B

Figure 4.22: ISK current in model HVCRA neuron. A. A model HVCRA neuron with the same parameters as in Fig. 4.20A but with gKNa = 100 nS, gA = 0 nS, gSK = 35 nS and gCa-T = 9 nS. B. Blocking the SK current in the model (gSK = 0) caused a dramatic increase in excitability, decreased the spike amplitude slightly, increased the resting membrane potential by ~ 8 mV, and exhibited adaptation due to IKNa.

113

Experiment

A

B

Figure 4.23: ISK current in actual HVCRA neurons. A. An RA-projecting neuron exhibits a single action potential with no delay. B. Blocking ISK using apamin (150 nM) greatly increased the neuron’s excitability, decreased the spike amplitude, and depolarized the resting membrane potential by ~ 5 mV. Unlike the model response however (Fig. 4.22B), no adaptation was seen in this neuron.

114

4.3.3 HVCINT Neurons

4.3.3.1 Response to applied current. HVC interneurons were less frequently encountered (n=25) and have two identifying features. The first is the high firing frequency observed in response to depolarizing current pulses (Fig. 4.25A). Upon stimulating an HVCINT neuron with a very small depolarizing pulse, the neuron fires with high frequency (n=25). A second identifying feature is a prominent sag generated in response to hyperpolarizing current pulses (Fig. 4.25B), followed by rebound firing upon the termination of the pulse (n=25). The sag ratio for this neuronal population (SR) was 0.24 ± 0.065 (n = 9). Other features of HVCINT neuron firing patterns are described in Table 4.3. In response to depolarizing current pulses, the spikes were followed by a large-amplitude AHP (23 ± 0.7 mV, n = 16) with a short time-to-peak (4 ± 0.5 ms, n = 17). Most of the interneurons seen had spikes that overshot 0 mV (n=18, spike amplitude=78±5 mV), but in a subset of neurons the spikes remained below 0 mV (n=7, spike amplitude=55±8 mV). Moreover, 18 interneurons fired spontaneously with a variety of patterns.

Figure 4.26 shows a sample of four different HVCINT neurons firing spontaneously over long periods of time (~2.5 minutes). Unlike HVCX neurons, this spontaneous activity was reduced, but not abolished, after CNQX (5 µM) and PTX (50 µM) application to block synaptic input (n=4, not shown).

Model HVCINT neuron parameters were calibrated to fit the voltage traces in response to depolarizing and hyperpolarizing current pulses (but not the spike amplitudes, which fall into two classes). Figure 4.27A shows the firing pattern of a model HVCINT neuron in response to a 75 pA applied current. The model neuron exhibits high firing frequency and no adaptation, both features which are due to the very small magnitudes of the ISK and IKNa conductances, thereby increasing the excitability of the model cell. Moreover, the model HVCINT neuron undershoots the resting membrane potential, due to the relatively large magnitude of the delayed rectifier K+ current conductance. The model was also calibrated to fit the sag and the rebound firing seen in

HVCINT neurons (Fig. 4.27B). As with the HVCX model neuron, the sag in the HVCINT model neuron is due to the Ih current that is activated upon hyperpolarization and gradually depolarizes the neuron. The prominence of the sag is primarily due to the extremely small value of (0.01), thereby giving almost all the weight to the slow component of the Ih current. Also, similar to the HVCX model neuron, the rebound firing is due to ICa-T and Ih. Model HVCINT neurons however do not generate the spontaneous firing seen in HVCINT neurons (Fig. 4.26).

115

Experiment

A

B

Figure 4.24: Firing properties of an actual HVC interneuron. A. An HVC interneuron fires tonically at high frequencies in response to a depolarizing current pulse (75 pA). B. A prominent sag followed by postinhibitory rebound firing is generated in response to a hyperpolarizing current pulse (-120 pA).

116

A

B

C

D

Figure 4.25: Most HVCINT neurons in the slice exhibited spontaneous firing (n=18). A-D. Spontaneous firing in four different HVCINT neurons.

117

Model

A

B

Figure 4.26: Firing properties of a model HVC interneuron. A. HVCINT model neuron parameters were calibrated to match the voltage traces of Fig. 4.24. A model HVC interneuron fires tonically at high frequencies in response to a depolarizing current pulse (75 pA). B. A prominent sag followed by postinhibitory rebound firing is generated in response to a hyperpolarizing current pulse (-120 pA) in model HVCINT neuron.

118

The model HVCINT neuron was used to predict the average firing frequency over a 1 sec duration of applied current for a range of current magnitudes and compared to the response of actual HVCINT neurons (Fig. 4.28). The frequency was much higher than in other HVC neurons, ranging from 10 Hz to 70 Hz. The predicted model neuron response agreed well with the frequency responses of the actual HVCINT neurons (n=15).

Figure 4.27: Frequency-current relationship in HVC inter neurons. The firing frequency of a model HVCINT neuron under various current injections (triangles) closely matches the data from interneurons in the slice (circles, n=15, mean ± SE). For both model and actual HVCINT neurons, the duration of applied current was 1 sec.

4.3.3.2 H current and inward rectification in HVCINT neurons. The sag and rebound of

HVCINT neurons are examined further in Figs. 4.29 and 4.30. The current-voltage relationship of the model HVCINT neuron is shown in Fig. 4.29A. The solid curve with circles represents the voltage at the end of the current pulses, while the dashed curve with triangles is the voltage at the nadir. The difference between the voltage responses at the end of the current pulse and at its nadir, i.e. the sag, increases as the magnitude of the hyperpolarizing current pulse increases. This difference is much larger than that seen in HVCX neurons (compare with Fig. 4.13B). In the slice, HVCINT neurons showed a similar behavior (n=18). A representative example is shown in Fig. 4.29B.

119

A Model Model

B Experiment

Figure 4.28: Inward rectification in HVCINT neurons. A. Current-voltage relationship of a model HVCINT neuron using same parameters as in Fig. 4.26. Circles with solid curve represent voltage responses measured at the end of the current pulses while triangles with dashed curve represent voltage at the nadir. B. An HVC interneuron exhibits inward rectification over a range of currents that is captured by the model.

Similar to HVCX neurons, the sag generated in the model HVCINT neuron is due to the hyperpolarization-activated inward current (Fig. 4.30A). The inset of Fig. 4.30A shows the total

Ih conductance gh. Because the HVCINT resting membrane potential is relatively high (~ -65 mV), some Ih conductance is activated at rest. In contrast, the resting level of gh was almost 0 in the model HVCX neuron (see Fig. 4.14A inset) since the resting membrane potential was lower.

When the hyperpolarizing pulse is applied, gh slowly rises, resulting in a voltage sag. The increase is slow, since gh is dominated by the slow component in this model neuron. When the

120

current pulse is terminated, the slow component of gh remains at an elevated level, contributing to the rebound firing.

A Model Model

B Experiment

Figure 4.29: Ih current in HVCINT neurons. A. The sag generated in the model HVCINT neuron is due to Ih (gh = 4 nS), which also contributes to the rebound (black trace). Blocking Ih by setting gh to 0 eliminates the sag and generates a weak rebound with a long delay due to ICa-T. Figure inset shows the total Ih conductance increasing during the negative current pulse. B. Blocking Ih using ZD 7288 (50 µm) eliminates the sag and generates a delayed rebound (red trace, 25 minutes after ZD 7288 application) in the same interneuron used in Fig. 4.24B, matching model predictions.

121

Blockade of Ih is simulated in the model by setting gh = 0. This results in a small decline in the resting membrane potential, and a much larger hyperpolarization during the current pulse.

This is because the resting Ih current conductance has been eliminated, so there is less Ih current to resist the hyperpolarization. The sag is gone and there is a delayed weak rebound spike, due to

ICa-T. Model predictions were tested and verified in the slice by pharmacologically blocking the

Ih current using ZD 7288 (50 µm, Fig. 4.30B). ZD application eliminated the sag and the strong rebound firing, but a weak delayed rebound spike remained (red trace, 25 minutes after drug application) in the same interneuron used in Fig. 4.25B. Moreover, the resting membrane potential was hyperpolarized by ~ 8 mV. Similar behavior was observed in 4 additional HVCINT neurons. Therefore, the sag seen in HVCINT neurons is due to the hyperpolarization-activated inward current, which also contributes to rebound firing. In another 4 HVCINT neurons, mibefradil application eliminated rebound firing (not shown) confirming the existence and role of ICa-T in these neurons as well.

4.3.4 Diversity of Action Potentials in HVC Neurons

The shape of action potentials differed considerably among the different types of HVC neurons. In general, the most noticeable features of the firing pattern of a neuron include the frequency of firing, bursting versus non-bursting activity and adapting versus non-adapting behavior. These are probably more important physiological properties than the shape of the spikes; however, one cannot cleanly separate them because differences in ionic conductances produce differences in firing patterns that will eventually produce differences in spike shape. In this section, we illustrate the differences in the shape and pattern of firing of action potentials between the various types of HVC neurons, focusing on understanding the role of ion channels in generating these differences. A simple but remarkably informative way of examining the properties of action potentials is to plot the time derivative of the voltage ( ) versus the voltage: a so-called phase-plane plot (Jenerick 1963). Information about the time course of the spike is lost with phase plots, but some aspects of the spike are clearer than in a simple display of voltage versus time (Bean 2007). The phase plane plot for a membrane action potential gives a direct read-out of the net ionic currents as a function of voltage during the various phases of the action potential. For example,

122

the maximum sodium current flowing during the spike can be estimated from the maximum reached during the upstroke, when contributions from other channels are likely to be

small. Similarly, information on spike threshold, strength of upstrokes and downstrokes of action potentials, as well as interspike intervals can be all deduced from phase plane plots.

4.3.4.1 Action potentials of HVCINT neurons. HVC interneurons were divided into two categories based on the shape of their action potentials, the magnitudes of and the trajectories on phase plane plots. Both types of interneurons (Type I and Type II) are fast -spiking neurons exhibiting very narrow spikes (Fig. 4.31A1, B1). The action potentials of Type I are preceded by a smaller, earlier component that is usually referred to as “the kink” (Coombs et al. 1957; Shu et al. 2007). This is illustrated in Fig. 4.31A1 with the arrow indicating the abrupt rise in the action potential (kink). Type II neurons on the other hand exhibit slightly narrower spikes, but no kink precedes their action potentials (Fig. 4.31B1). The corresponding time derivatives of the voltage traces in panels A1 and B1 are shown in panels A2 and B2, respectively. Note the fast repolarization rate in both cases illustrating the fast-spiking behavior of HVC interneurons. In a wide variety of central and pyramidal neurons, the kink and the early component seen here in Type I HVC interneurons had been interpreted as reflecting the initiation of the spike in the initial segment of the axon (Bean 2007; Clark et al. 2005; Grace and Bunney 1983; Hausser et al. 1995; Palmer and Stuart 2006; Shu et al. 2007; Stuart et al. 1997). To examine that further, panel A3 shows the phase plane plot of the trajectories in A1 illustrating the distinct components of the action potentials of Type I interneurons (illustrated by arrows on A3). These components on the phase plot are the hallmark of central and pyramidal neurons, and they include the initiation in the initial segment (IS) and subsequent somatic dendritic (SD) spike. The phase plane plot of Type II neurons, however, does not exhibit these distinct components due to the absence of the kink (B3).

123

Experiment

Type I Type II

2 ms 2 ms

Figure 4.30: Action potentials of Type I and Type II HVC interneurons. A1. Action potentials at high resolution of a Type I HVC interneuron illustrating the abrupt rise (kink). A2. Time derivative of the membrane voltage ( ) of the same trace in A1. Note the fast repolarization rate. A3. Phase plane plot of the trajectories in A1 illustrating the distinct components that usually reflects the initiation in the initial segment (IS) and subsequent somatic dendritic (SD) spike as seen in central and pyramidal neurons (see text). B1-B3. Action potentials, time derivative and phase plane plot of a Type II HVC interneuron. Note lack of kink and IS component.

In addition to the difference with the kink, Type I and Type II HVC interneurons differ in the relative magnitudes of the time derivative as illustrated in Fig. 4.32. Panels A1 and A2 show two different HVC neurons of Type I, while panels B1 and B2 show another two INT different HVCINT neurons of Type II. HVC interneurons of the same type exhibit very similar

124

trajectories on the phase plane plot as illustrated in panels A3 and B3 where the corresponding trajectories in A1 and A2 (or B1 and B2, respectively) are superimposed. Most notably, the time derivative of Type I is much larger than that of Type II in all of the interneurons examined (n=8, Type I and n=5, Type II). Moreover, Type II interneurons do not overshoot zero while Type I neurons do (Fig. 4.32).

Experiment

Type I Type II

Figure 4.31: Difference in the magnitude of between Type I and Type II HVC interneurons. A1- A2. Phase plane plots of two different HVCINT neurons of Type I. B1-B2. Phase plane plots of two different HVCINT neurons of Type II. HVC interneurons of the same type exhibit very similar trajectories on the phase plane plot as illustrated in A3 (trajectories of A1 and A2 superimposed) and B3 (trajectories of B1 and B2 superimposed). Moreover, Type II interneurons do not overshoot 0 mV, while Type I interneurons do.

125

The model HVCINT neuron does not exhibit the kink seen in actual Type I HVC interneurons (Fig. 4.33) and it is thus similar to Type II interneurons in this aspect. However, the magnitude of of the model HVCINT neuron is similar to that of Type I interneurons as it reached ~200 mV/msec (compare with Fig. 4.32A1-A3). Moreover, the model HVCINT neuron is a fast-spiking neuron with narrow spikes and it overshoots 0 mV, similar to Type I. In general, cells with narrow spikes commonly display fast-spiking behavior and are capable of firing at high frequencies with little decrease in frequency during prolonged stimulation (Bean 2007; Connors and Gutnick 1990; Descalzo et al. 2005; Erisir et al. 1999; Kawaguchi 1995; McCormick et al. 1985; Nowak et al. 2003; Tateno et al. 2004; Zhou and Hablitz 1996). In the cortex, and hippocampus, GABA ( -aminobutyric acid)-releasing interneurons generally have narrower spikes than glutamatergic pyramidal neurons, as seen by various intracellular recordings (Bean 2007; Kawaguchi 1995; McCormick et al. 1985; Nowak et al. 2003). The fast- spiking phenotype has been related to expression of the Kv3 family of voltage-gated potassium channels, the rapid and steeply voltage-dependent activation and deactivation kinetics of which are well-suited for generating narrow action potentials and short refractory periods (Bean 2007; Erisir et al. 1999; Martina et al. 1998; Massengill et al. 1997; Rudy and McBain 2001). This + agrees with our model HVCINT neuron that exhibits a large magnitude of the K conductance (gK = 1700 ns). The voltage-dependent K+ current is playing an important role in shaping the model + HVCINT neuron’s firing pattern; however, measuring the K conductance in HVC interneurons remains to be further investigated.

Model

Figure 4.32: Phase plane plot of the model HVCINT neuron. Note the lack of kink (resembling Type II interneurons) and the large magnitude of (resembling Type I interneurons).

126

4.3.4.2 Action potentials of HVCRA neurons. All of the HVCRA neurons examined (n=7) belonged to one category and exhibited very similar shape and pattern of firing, magnitude of voltage time derivative and trajectories on phase plane plots. Figure 4.34A shows a representative example of the activity recorded from an RA-projecting neuron exhibiting a delay to spiking for a depolarizing current pulse of 250 pA. Despite the lack of excitability of these neurons, their action potentials exhibit sharp upstrokes and downstrokes. This is illustrated in panel B where two action potentials of the same voltage trace in A are depicted at higher resolution. There is an abrupt rise (kink) in the voltage trace (shown by the arrow) and the interspike interval is close to, but larger than, that seen in HVCINT neurons. The time derivative of the voltage trace in B is shown in panel C. As mentioned earlier, HVCRA neurons’ action potentials are sharp and this is reflected in the fast depolarization and repolarization phases of the time derivative trace (Fig. 4.34C). The phase plane plot of the trajectories in B is shown in panel D. Note the two distinct components in the trajectories on the phase plane that are also seen in

Type I HVCINT neurons and that are the hallmark of IS and SD spikes in other types of neurons.

HVCRA neurons show very strong stereotypy not only in their temporal precision while firing during singing, but also in the shapes of their action potentials as reflected by the trajectories on the phase plane plots (Fig. 4.35). Panels A-C shows three different HVCRA neurons having very similar trajectories. These three neurons (and the remaining four examined, not shown) exhibited the kink in their voltage traces which is reflected in the phase plots of panels A-C by the two distinct components representing the IS and SD spikes. The similarity between the HVCRA neurons firing pattern is more obvious in panel D by superimposing the trajectories of A-C. The magnitude of the time derivatives of voltage traces of HVCRA neurons were also pretty close to each other and ranged between -70 and 125 mV/msec (n=7). Therefore, from the perspective of the space that spans, HVCRA neurons are similar to that of Type II

HVCINT neurons. These results however do not rule out the possibility that there could be more than one type of HVCRA neuron because the number of neurons examined is small.

127

Experiment

250 pA

Figure 4.33: Action potentials of HVCRA neurons. A. Activity recorded from an RA-projecting neuron exhibiting a delay to spiking for a current pulse of 250 pA. B. Two action potentials of the same neuron in A at higher resolution illustrating the abrupt rise (kink) in the voltage trace. C. Time derivative of the voltage trace in B. Note the fast repolarization as in HVC interneurons. D. Phase plane plot of the trajectories in B illustrating the distinct components reflecting the initiation in the initial segment (IS) and subsequent somatic dendritic (SD) spike.

128

Experiment

Figure 4.34: Phase plane plots of three different HVCRA neurons (A-C). D. HVCRA neurons exhibit very similar trajectories on the phase plane plot as illustrated in D where trajectories of A-C are superimposed.

The model HVCRA neuron exhibits the kink seen in actual HVCRA neurons (Fig. 4.36) although the kink appears at more depolarized voltages and its trajectory looks dissimilar to that of the actual HVCRA neurons. However, the magnitude of the time derivative of model HVCRA neuron voltage is very similar to that of actual RA-projecting neurons. + HVCRA neurons express a variety of K currents, which as we have seen earlier in this chapter, play a key role in shaping the RA-projecting neuron’s activity pattern. This could explain the sharp downstrokes these neurons exhibit in their action potentials, which make

HVCRA neurons along with HVC interneurons unique in the HVC from this perspective. The extremely important premotor role that HVCRA neurons play during singing makes sense for these neurons to have their spikes issued rapidly and reliably enabling the temporal precision in firing that is the hallmark of HVCRA neurons. However, the contribution of the particular ionic currents to HVCRA neurons’ action potentials, that is, the “internal anatomy” of their spikes remains to be investigated with various clamp techniques.

129

Model

Figure 4.35: Phase plane plot of the model HVCRA neuron. Note the presence of the kink that occurs at larger values of voltage than actual HVCRA neurons. The magnitude of , however, is similar to actual RA-projecting neurons.

4.3.4.3 Action potentials of HVCX neurons. All of the HVCX neurons examined (n=14) belonged to one category and exhibited very similar shape and pattern of firing, magnitude of and trajectories on phase plane plots. Figure 4.37A shows a representative example of the

activity recorded from an X-projecting neuron exhibiting spike frequency adaptation for a depolarizing current pulse of 150 pA. In panel B two action potentials of the same voltage trace in A are depicted at higher resolution. Just as in Type I interneurons and RA-projecting neurons, there is an abrupt rise (kink) in the voltage trace. The interspike interval of HVCX neurons is considerably larger than that seen for HVCINT and HVCRA neurons. The time derivative of the voltage trace in B is shown in panel C. Note the slow repolarization phase that HVCX neurons exhibit (Fig. 4.37C) which is different than that seen in HVCINT and HVCRA neurons (compare with Fig. 4.31A1-A2 and Fig. 4.34C). The phase plane plot of the trajectories in B is shown in panel D. The two distinct components that are seen in Type I HVCINT neurons and HVCRA neurons are evident. The attenuation of spikes (successive decrease in the spike amplitude) seen in all of the HVCX neurons is reflected in the phase plane plot as two trajectories wrapped within each other.

130

Experiment 150 pA

Figure 4.36: Action potentials of HVCX neurons. A. Activity recorded from an X-projecting neuron exhibiting spike frequently adaptation for a current pulse of 150 pA. B. Two action potentials of the same neuron in A at higher resolution illustrating the abrupt rise (kink) in the voltage trace. Note the attenuation of spikes (successive decrease in the spike amplitude). C. Time derivative of the voltage trace in B showing the slow repolarization unlike HVCRA and HVCINT neurons. D. Phase plane plot of the trajectories in B. The two distinct components that are seen in Type I HVCINT neurons and HVCRA neurons are evident.

131

Similar to HVCRA and HVCINT neurons, HVCX neurons exhibit very similar trajectories on phase plane plots as illustrated in Fig. 4.38. Panels A-D shows four different HVCX neurons exhibiting extremely similar firing patterns with attenuation and adaptation. All of the HVCX neurons examined exhibited the kink in their voltage traces which is reflected in the phase plots of panels A-D by the two distinct components representing the IS and SD spikes. The similarity between the HVCX neurons firing pattern is more obvious in panel E by superimposing the trajectories in A-D. Most importantly, the magnitude of the time derivative of the voltage traces of HVCX neurons is the largest of the three classes of HVC neurons and ranged between -100 and 400 mV/msec (compare with Figs 4.32 and 4.35).

Figure 4.37: Phase plane plots of four different HVCX neurons (A-D). Note the attenuation and adaptation. E. HVCX neurons exhibit very similar trajectories on the phase plane plot as illustrated by superimposing the trajectories of A-D. The magnitude of in HVCX neurons is much larger than that in HVCINT and HVCRA neurons (compare with Figs 4.31 and 4.34)

132

Unlike model HVCINT and model HVCRA neurons, the model HVCX neuron is the closest to actual HVCX neurons in its firing pattern properties and phase plane plots. Figure 4.39 shows the phase plane plot for the model HVCX neuron exhibiting the two distinct components as actual HVCX neurons, although they are not pronounced as in the real neurons. Moreover, and more dramatically, the magnitude of the time derivative of the model HVCX neuron voltage is very similar to that of actual X-projecting neurons, reaching high values (~ 350 mV/msec, compare with Fig. 4.38). In addition, the model HVCX neuron exhibits attenuation of spikes and this is evident in Fig. 3.39 by the concentric trajectories.

Figure 4.38: Phase plane plot of the model HVCX neuron. The kink is evident, but less pronounced than in actual neurons. More importantly, the magnitude of is very similar to that of actual X-projecting neurons (compare with Fig. 4.37).

In summary, cataloguing all the components of the ionic currents of HVC neurons and their exact contributions on spike generation is extremely challenging, as it requires characterizing the gating kinetics of every single component. However, these results motivate the need to accurately describe channel gating on both fast and slow timescales, as well as understand how the voltage-dependent currents interact to control the generation of HVC neurons’ action potentials. Our whole-cell recordings had been conducted from the cell body of intact HVC

133

neurons, and for this reason interpretation of the phase-plane plots for action potentials is not a trivial task, as the presence of a dendritic tree and axon means that not all ionic current goes to charge or discharge the somatic membrane. The analysis done here, however, shows that there is a possibility that the spikes of HVC neurons occur first farther from the soma, in the initial segment of the axon or possibly in the nodes of Ranvier. This had been seen in layer 5 cortical pyramidal neurons and other types of interneurons (Clark et al. 2005; Colbert and Johnston 1996; Khaliq and Raman 2006; Palmer and Stuart 2006). In cerebellar Purkinje neurons, blocking sodium entry at the first node of Ranvier has no effect on the somatic spike wave-form (including the kink), whereas inhibiting sodium entry at the initial segment reduces the kink (Khaliq and Raman 2006).

4.4 Discussion

This study identified ionic currents present in zebra finch HVC neurons, which was facilitated by the development of computational models of the three types of HVC neurons. These models were based on current-clamp recordings reported in the literature (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Mooney 2000; Mooney et al. 2001; Mooney and Prather 2005; Shea et al. 2010; Wild et al. 2005) and by our own current-clamp data. The calibrated models were used to generate predictions about the roles that the different ionic currents play in shaping the characteristic firing patterns of each of the HVC neurons types. The predictions were then tested and verified in the slice by varying the applied current or pharmacologically blocking ionic currents of HVC neurons. Our data identified a 2+ hyperpolarization-activated inward current (Ih) and a low-threshold T-type Ca current in the 2+ + + HVCX and HVCINT neurons, a Ca -activated K current (ISK) and an A-type K current (IA) in + + the HVCRA neurons, and highlights a possible role for the Na -dependent K current in the

HVCX neurons. Figure 4.40 summarizes the conductance values presented in Tables 4.1 & 4.2 and shows a schematic of the three model HVC neurons depicting their corresponding ionic currents. The relative magnitudes of the various conductance parameters within a particular model neuron are reflected by the sizes of the corresponding cartoon ion channels (size

134

computed based on logarithmic scale). The variation in the balance of these various ionic currents shapes the firing properties of model HVC neurons.

Figure 4.39: Schematic models depicting the ionic currents present in each of three HVC model neurons after calibration to our data. Similar ionic conductances across the three classes of HVC neurons are coded with the same colors. The relative magnitudes of the various conductance parameters presented in Tables 4.1 & 4.2 for a particular model neuron are reflected by the sizes of the corresponding cartoon ion channels (size computed based on logarithmic scale). This shows indirectly the contributions of the various ionic currents to shaping the firing properties of the model HVC neurons. The h-current conductance has fast (ghf) and slow (ghs) components.

Previous research characterized the electrophysiological properties of the different classes of HVC neurons (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Mooney 2000; Schmidt and Perkel 1998; Shea et al. 2010). These properties include the resting membrane potential, input resistance, sag, spike duration, spike threshold, AHP amplitude and AHP time-to-peak. Moreover, morphological characterization and visualization of HVC neurons had been performed post hoc in a few of these and other studies (Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Mooney 2000; Mooney and Prather 2005; Shea et al. 2010; Wild et al. 2005). We used this electrophysiological characterization to identify our HVC neurons and performed our own morphological studies to support this means of identification. Our electrical recordings were consistent with the previous studies and our anatomically

135

identified cells confirmed the classification of neurons based on their physiological properties. A panel of intrinsic property measurements for each of the classes of HVC neurons is reported in Table 4.3. Hyperpolarization-activated ionic conductances are a prevalent feature of electrically excitable cells, including frog skeletal muscle fibers, heart, photoreceptors, neurons in the peripheral and central nervous system, and some types of axons (Pape 1996). The sag we observed in HVCX and HVCINT neurons in response to hyperpolarizing current pulses was completely abolished after the application of the drug ZD 7288 in all of the neurons tested (Figs. 4.14, 4.15 & 4.30), indicating that the sag in these HVC neurons is due to the hyperpolarization- activated inward current (Ih). In other types of neurons, two kinetically distinct components of Ih have been identified with time constants in the range of hundreds of milliseconds and seconds, respectively (Banks et al. 1993; Budde et al. 1994; Solomon et al. 1993). We found that our models fit the data best when the relative strength of these components is different in the three types of HVC neurons.

The unique nature of Ih, i.e., an inward current activated upon hyperpolarization beyond resting potential, makes it particularly useful in rhythmogenesis. Reported examples include (a) rhythmic activity of the crustacean somatogastric ganglion, where Ih in lateral pyloric neurons controls phase relationships in the pyloric network (Golowasch et al. 1992; Golowasch and Marder 1992); and (b) slow oscillatory activity of interneurons that control heartbeat in the leech, where Ih mediates the escape from inhibition that times the phase transition of two interconnected neurons (Angstadt and Calabrese 1989). Another study done by Thoby-Brisson et al (2000) on the respiratory rhythm generated by the pre-Botzinger complex (PBC) showed that inhibiting Ih can actually increase the frequency of rhythmic activity, which is counter-intuitive and demonstrates the subtle actions of this subthreshold current. In thalamocortical neurons, the properties of the Ih current and the critical role it plays in slow thalamic oscillations has been extensively investigated (Budde et al. 1997; McCormick and Pape 1990; Munsch and Pape 1999;

Pape 1996). In the HVC of songbirds, the role that Ih current plays in controlling their vocal pattern remains to be investigated. The existence of a low-voltage activated Ca2+ current in HVC neurons was first shown by Kubota and Saito (1991), and we have shown here the key role it plays in rebound spiking (Figs. 4.14, 4.15 and 4.30). This current is known to be important in other systems as an ionic current

136

for burst generation (Huguenard 1996). This has been characterized in several cell types that include thalamic reticular (Mulle et al. 1986) and relay cells (Deschenes et al. 1982; Llinas and Jahnsen 1982), inferior olive cells (Llinas and Yarom 1981a; b), hippocampal interneurons (Fraser and MacVicar 1991), lateral habenular neurons (Wilcox et al. 1988), a subpopulation of pontine reticular formation cells (Gerber et al. 1989), and neocortical neurons (Friedman and Gutnick 1987). Bursting activity in HVC neurons has been reported to occur in vivo during singing (Hahnloser et al. 2002; Kozhevnikov and Fee 2007; Long et al. 2010). Moreover, Lewicki (1996) showed that auditory HVC neurons in vivo exhibit a significant hyperpolarization before emitting their bursts, and the strength of the burst is correlated with the degree of hyperpolarization. The role the T-type Ca2+ current plays in HVC neuron bursting in vivo has yet to be investigated.

The adaptation we observed in HVCX neurons was unaltered by apamin application (n=4). This demonstrates that SK channels are not solely responsible for adaptation in these neurons. However, apamin did decrease the AHP following each action potential, consistent with prior reports (Kubota and Saito 1991; Schmidt and Perkel 1998). Apamin also depolarized the resting membrane potential by a few millivolts in these neurons. Apamin application however, had a dramatic effect on HVCRA neurons. These neurons normally fire one or few spikes upon depolarization, but when the SK current is blocked they fired throughout the duration of the depolarizing current pulse (Fig. 4.24). Thus, SK current plays a major role in inhibiting the excitability of HVCRA neurons. We also investigated the role played by A-type K+ channels, using the blocker 4-AP. This eliminated the delay to spiking in HVCRA neurons and increased their resting membrane potential. It also reduced the AHP following each spike. Since both 4-AP and apamin increased the resting membrane potential of HVCRA neurons, IA and ISK must both contribute to the extremely hyperpolarized resting membrane potential of HVCRA neurons, a distinctive feature of these neurons. The ionic currents presented here motivate speculation on the role they play from a circuit- level perspective. Evidence shows that sequential activity in the HVC propagates preferentially along its rostrocaudal axis (Day et al. 2013; Nottebohm et al. 1982; Stauffer et al. 2012). Moreover, Mooney and Prather (2005) identified several synaptic interactions among the three HVC neuronal subpopulations in zebra finch brain slices. These findings, combined with our

137

models, could serve as a starting point to develop realistic and plausible neural architectures that would reflect an accurate topography of the nucleus as well as produce the characteristic patterns of neural activity exhibited by HVCRA, HVCX and HVCINT during singing. This of course requires more investigation on the connectivity patterns among the different types of HVC neurons. The models that we developed consist of a single compartment. One could construct multi- compartment models to better represent the spatial aspects of HVC neurons, although to do this well it would be necessary to first study dendritic properties of the neuron, including ion channel distribution. Also, in our model description of ionic currents, we utilized functional forms used in prior published neural modeling studies of non-HVC neurons. We did not attempt to calibrate the shape parameters of these functions, which would best be done using a voltage-clamp protocol or using dynamical estimation methods as in Toth et al (2011). Doing this calibration would improve the fit of the models. Finally, pharmacological blockers could be used with a range of current pulses (rather than just one depolarizing pulse and one hyperpolarizing pulse, as done here) to better constrain the model parameter values. This being said, the current models provide a large step forward in describing the biophysics of HVC neurons.

138

CHAPTER 5

NETWORK LEVEL: HVC MICROCIRCUITRY

5.1 Introduction

In this chapter, we investigate plausible neural architectures within the HVC that can produce the patterns of neural activity exhibited by HVCRA neurons, HVCX neurons, and

HVCINT neurons during singing. As constraints, we use the connectivity information gathered from brain slices (Mooney and Prather 2005) and the activity patterns of the three subpopulations of neurons recorded from brain slices (Daou et al. 2013; Dutar et al. 1998; Kubota and Saito 1991; Kubota and Taniguchi 1998; Mooney et al. 2001; Mooney and Prather 2005; Schmidt and Perkel 1998; Shea et al. 2010; Wild et al. 2005) and in vivo in anesthetized birds (Lewicki 1996; Mooney 2000). Conductance-based models we developed in Chapter 4 for the individual neurons and parameters were adjusted and calibrated to match the spiking data from brain slice and anesthetized birds. Here we connect the model neurons in different ways so as to match the neuronal patterns of activity produced while the bird is singing (Hahnloser et al. 2002; Kozhevnikov and Fee 2007; Long et al. 2010). Using paired intracellular recordings and antidromic stimulation in zebra finch brain slices, several synaptic interactions among the three HVC neuronal subpopulations were determined

(Mooney and Prather 2005). These interactions are illustrated in Fig. 5.1. Both HVCRA and

HVCX neurons excite HVC interneurons. The interneurons are themselves inhibitory to both projection neurons. The HVCRA projection neurons send excitatory afferents to HVCRA neurons, and receive excitatory input from HVCX neurons (Mooney and Prather 2005). Moreover, as we saw in Chapter 4, HVC interneurons tend to be highly excitable and fire spontaneously, while

HVCX neurons tend to fire sparsely with adaptation and HVCRA neurons are famous for their lack of excitability.

139

Figure 5.1: Synaptic interactions between the three classes of HVC neuronal populations. RA-projecting neurons (HVCRA), X-projecting neurons (HVCX), and interneurons (HVCINT). Blue curves with arrow heads are excitatory and red curves with arrow heads are inhibitory. Electrodes depict a cartoon of the in vitro recordings performed in Chapter 4 from the three classes of HVC neurons.

We present five prototype network architectures that show how the sequential activity in the HVC projection neurons can propagate. Our goal is not to isolate the best network architecture but to explore ways in which the different types of HVC neurons can interact to produce a stereotyped HVCRA sequence. Each architecture is comprised of a chain of microcircuits, each of which produces an HVCRA burst during one syllable of the motif. With these simple model circuits, we are able to determine which components of the cells’ intrinsic behavior and connectivity pattern are responsible for the duration of individual bursts, the length of silent intervals between the bursts, and the resulting syllable, intersyllable, motif and inter- motif durations. In most architectures, postinhibitory rebound serves a major role that preserves precise spike timing information, enabling reliable propagation of sequential activity throughout the microcircuits.

5.2 Methods

We assume that neurons of the HVC that are involved in the timing of syllables form microcircuits. Each microcircuit contains subpopulations of neurons, neurons, and

neurons. The many neurons within a subpopulation are assumed to be electrically active

140

at the same time, so we represent this subpopulation by a single representative neuron. For example, the subpopulation in a microcircuit is represented by the model neuron .

The representative neurons , , and are then synaptically coupled together to form a microcircuit, or to project to another microcircuit as a connecting link in the microcircuit chain. Single-compartment conductance-based biophysical models of cells from the HVC were developed, based on current-clamp data (section 4.3.2). Simulations of these model neurons and of model networks composed of synaptically coupled , , and neurons were performed using the CVODE numerical integrator in XPPAUT (http://www.math.pitt.edu/~bard/xpp/xpp.html).

5.2.1 HVC cells

HVC model cells that are used to connect the networks are the same model cells developed in Chapter 4 (section 4.3.2). In particular, each model HVC neuron was designed to include spike-producing currents ( and ), a high-threshold L-type Ca2+ current ( ), a low- 2+ 2+ + threshold T-type Ca current ( ), a small-conductance Ca -activated K current ( ), a + + + + persistent Na current ( ), a Na -dependent K current ( ), an A-type K current ( ), a hyperpolarization-activated cation current ( ), and a leak current ( ). Thus, the membrane potential of each HVC neuron obeys the current balance equation:

The is an external current, where H can be X, RA or INT representing input to X- projecting, RA-projecting or interneurons, respectively. We treat this as stochastic input and model its stochasticity by adding standard Gaussian white noise as follows: where is the noise amplitude, is a random number drawn at each time step from a normal distribution, and is a basal constant applied current value.

141

5.2.2 Synaptic Connectivity

The current represents the synaptic input(s) from the presynaptic cell(s) to the particular HVC model neuron, and is modeled as where . Here the summation is taken over the presynaptic HVC neurons where X represents a presynaptic cell, Y represents a postsynaptic cell, is the reversal potential for the synapse in the postsynaptic cell with = for excitatory input and = for inhibitory input. Each synaptic variable satisfies a first-order differential equation where is the membrane potential of the presynaptic HVC neuron X, and approximates a Heaviside

function.

5.2.3 Current Clamp Neuronal Firing Patterns and Rebound Bursting

As discussed in Chapter 4, , , and neurons in brain slices exhibit different spiking responses to depolarizing DC current pulses. With the injection of a depolarizing current, the neurons fire with characteristic patterns. neurons typically spike repetitively in response to the depolarizing current pulse and exhibit spike-frequency adaptation (Fig. 4.9A). In contrast, neurons typically produce a single spike, often with a delay, in response to a depolarizing current pulse (Fig. 4.18A). Finally, neurons exhibit high firing frequency with little or no spike-frequency adaptation (Fig. 4.25A). Parameters for model HVC neurons were then calibrated to exhibit spiking patterns that are similar to those of actual HVC neurons from our brain slice studies (Figs. 4.11, 4.19 and 4.27, respectively). Model HVC neurons are also able to predict accurately the frequency-current relationships of the actual HVC neurons (Figs 4.12, 4.20 and 4.28). In addition to that, it is important to note that the rebound firing seen in HVCX and HVCINT neurons (Figs. 4.9, 4.15 and 4.25) and that is replicated by the models (Figs. 4.11, 4.15 and 4.27) serve as a key element of the mechanism for the sequential firing pattern generated by most of the networks during simulated singing, as we will see later. Fixed parameter values for HVC neurons used in the simulations are given in Tables 4.1 and 4.2 and 5.1, and parameters that vary between the different network architectures are shown in Table 5.2.

142

Table 5.1: Parameter values used in all network models. These parameters are different from those in Tables 4.1 and 4.2 and used to connect the networks in all the simulations.

Parameter Value

40 nS

5 nS

0 mV

-100 mV

10 nS

50 nS

20 nS

1 nS

Table 5.2: Parameter values that vary among networks.

Network 1 Network 2 Network 3 Network 4 Network 6

25 25 25 25 20

5 5 5 5 3

22 22 22 22 23

1 1 1 1 15

1 1 1 1 15

1 1 1 1 15

1 1 1 1 15

143

Table 5.2 - continued

313 313 313 313 270

5 3 3 3 0.1

0 0 0 0 0.1

12 15 15 15 0.1

12 15 14 22 0.1

12 15 13 18 0.1

12 15 13 17 0.1

5.3 Results

We next describe the activity generated by five prototype networks of model HVC neurons, and explain how the firing activity of neurons propagates through the network in sequential bursts of activity. In each case we describe the network components that are essential for the patterned output of the system. We also describe the internal dynamics of the network that govern the duration of individual bursts as well as the duration of silent gaps between bursts. The five prototype networks are composed of chains of five HVC microcircuits, each with its own intrinsic dynamics. Each mean field neuron in a microcircuit is representative of a neural population. The number of microcircuits in the chain determines the number of syllables in the motif. We envision the HVC to be composed of many copies of such microcircuit chains that are associated with syllables with roughly synchronized activity (Fee et al. 2004). Thus, the first neuron of each microcircuit chain bursts during the first syllable, although bursts need

not start or end at the same time. The second neuron of each chain bursts during the second syllable, etc. The five prototype network architectures differ in both the connectivity patterns within a microcircuit and the connectivity among the microcircuits.

144

HVC neurons receive afferent projections from several extrinsic forebrain and brainstem regions (Cardin and Schmidt 2004; Foster et al. 1997; Nottebohm et al. 1982). Since little is known about the firing pattern of this input, we treat it as a DC current which is on during singing and applied to the three HVC neuronal populations. Each microcircuit is identical to the others, so to produce a sequence initiator we increase the DC input to one microcircuit relative to the others. The DC input is stochastic, producing a variable inter-motif interval, as occurs in the zebra finch.

5.3.1 Network Architecture 1

The first network architecture is illustrated in Fig. 5.2A. Within each of the five microcircuits (the first of which is enclosed by a dashed rectangle), there are excitatory projections from and to , and inhibitory projections from to

and . In this network, as in all others, there are excitatory projections from

back into . Each receives excitatory drive (DC current) from outside the HVC

(downward arrows in diagram). The microcircuits interact with each other via the projections from from one microcircuit to in a following microcircuit.

The activity patterns that this network displays are illustrated in Fig 5.2B. The interneurons spike continuously, thereby inhibiting the activity of . Whenever the stochastic DC current is sufficiently large to overcome the inhibitory input, it drives above the spike threshold. The spiking in is amplified by the excitatory projection back onto .

This positive feedback results in a burst of activity, which generates a burst of activity in

due to excitatory coupling. The burst in generates a hyperpolarization in

due to inhibitory coupling. These events are illustrated in the first column of Fig. 5.2B. Each column of the figure is a time point, with t1 < t2 < … < t5. The bursts in and

during the first time point (t1) are illustrated, as well as the quiescence of . 2+ 2+ Intracellular Ca accumulates during the burst. This result in a buildup of Ca - + activated K current ( ) that terminates the burst, which in turn terminates the burst since no longer provides excitation ( doesn’t play a significant role here due to the small magnitude of its conductance). After these bursts are over, escapes from inhibition and fires a postinhibitory rebound burst. Since excites of the second microcircuit

145

in the chain, a new burst is initiated in (Fig. 5.2B, column 2). This activity propagates throughout the chain of microcircuits causing a sequence of bursts.

Figure 5.2: Arrangement of prototype network architecture one (A). In this figure and Figs. 5.6-5.8, each of the six prototype networks is composed of five microcircuits. The first microcircuit is enclosed by a dashed rectangle to represent its role as a leading microcircuit in the network. Each HVCRA population receives excitatory drive (DC current) from outside the HVC (downward arrows). Stronger DC current is represented by larger downward arrows. Blue curves with arrow heads represent excitatory connections, and red curves with circles represent inhibitory connections. Dotted lines represent random projections. B. Activity pattern for network architecture one. Bursts elicited at their correct times are represented by (

), “unwanted” bursts at wrong times by ( ) and silent cells by ( ). Blue curves with arrows represent excitatory connections, and red curves with circles represent inhibitory connections. The times ti are time points during the motif, with ti+1 > ti. Same apply to Figs. 5.6-5.8.

146

The duration of bursts is determined by two factors. The first is the rate of Ca2+ accumulation, which results in accumulation of the hyperpolarizing SK current. The second is the strength and duration of the rebound bursting generated by after release from inhibition. The duration of silent intervals between the bursts, for instance between

and , is determined by the degree of inhibition of . The stronger the

inhibition of , the longer it takes to elicit a rebound burst, and thus the longer the silent gap. The degree of inhibition is determined by the synaptic conductance from to

( ) as well as the intensity of spiking in neurons.

The inter-motif interval is determined by intrinsic and extrinsic factors. The primary intrinsic factor is the current in . During a motif, initiates the first burst of activity resulting in an increase in its intracellular calcium and in . Once the burst terminates, there is a prolonged afterhyperpolarization as Ca2+ concentration slowly declines before

can burst and initiate a new motif. This afterhyperpolarization is controlled by the

conductance, and the conductance, . The extrinsic factor is the magnitude of the applied current ( ) to . The stronger the applied current to , the shorter the time required to elicit a new episode of bursting. The motif duration is determined by the number of syllables in the song, and thus by the number of microcircuits within the network, the duration of bursts and the duration of silent gaps between the bursts. The activity patterns that this network displays are shown in Figs. 5.3 and 5.4. The sequential bursting of is illustrated in Fig. 5.3A and on a shorter time scale in Fig. 5.4A.

The five burst and inter-burst durations are the same throughout the sequence since the microcircuits are homogeneous. The firing pattern of a sample interneuron ( ) is shown in

Fig. 5.3B and 5.4B. In this network, is continuously spiking until elicits a burst of activity in . The sparse irregular spiking of is illustrated in Fig. 5.3C and 5.4C.

There is a burst of activity in following the burst in . Otherwise, the spiking is irregular.

147

Figure 5.3: HVC spiking patterns of network architecture one. A. Sequential bursting of of the five microcircuits within the network (labeled with numbers) showing four episodes of activity corresponding to four motifs within a bout of singing. B. Firing pattern of a sample interneuron in the network ( ) over the course of the four motifs showing one burst per motif as a result of the excitation from . C. Firing pattern of a sample X-projecting neuron ( ) showing the sparse irregular spiking of cells.

148

Figure 5.4: Zoomed version of HVC spiking patterns of network architecture one. Panels A, B, and C show activity from panels A, B, and C of Fig. 5.3, respectively, on a shorter time scale. The burst in generates a burst in due to excitatory coupling. The burst in then generates a hyperpolarization in due to inhibitory coupling. then escapes from inhibition and fires a

postinhibitory rebound burst which starts the activity of of the second microcircuit in the chain.

149

Although this network produces the desired behavior of HVC neurons, one possible drawback is that HVC interneurons fire only one burst in response to excitation from bursts. Recordings of identified HVC neurons in singing birds have shown that

neurons can fire up to 20 bursts per song motif (Hahnloser et al. 2002; Hahnloser 2006;

Leonardo and Fee 2005). Another possible drawback is that the sequence of activity in neurons is driven primarily by neurons. While some studies have shown an

important role for – interactions in the production of the song and the bird’s recognition of its own song (Hahnloser et al. 2002; Kubota and Taniguchi 1998; Mooney 2000; Rosen and Mooney 2003), one study showed that targeted neuronal death in neurons left song production unaltered (Scharff et al. 2000). We address this finding in subsection 5.3.5. These factors also motivate the investigation of additional networks.

5.3.2 Network Architecture 2

We next consider architecture Fig. 5.5A, which gives more complex network dynamics than the previous network. Unlike before, sends afferent excitatory projections to

in the same microcircuit and to in the next microcircuit (Fig. 5.5A). This

holds except for , which send projections to only. Also, send inhibitory afferents to in the same microcircuit and to in the next microcircuit. This holds except for , which inhibits only. The connectivity patterns between and

remains the same as in network one.

The activity pattern that this network displays is shown in Fig. 5.5B. When bursts t at time 1, two bursts of activity are generated in the interneurons due to excitatory coupling; the first in within the same microcircuit and the other in of the next microcircuit.

The burst in hyperpolarizes , while the burst in hyperpolarizes . These events are illustrated in the first column of Fig. 5.5B.

150

Figure 5.5: Arrangement of prototype network architecture two (A). Blue curves with arrow heads represent excitatory connections, and red curves with circles represent inhibitory connections. Dotted lines represent random projections. B. Activity patterns for network architecture two. For more info on symbols, refer to legend of Fig. 5.2.

151

The accumulation of intracellular Ca2+ terminates the burst, which in turn terminates the and bursts. After and bursts are over, both

and escape from inhibition and fire postinhibitory rebound bursts at time t2. Since

excites and excites , two new bursts in and

neurons are now initiated at time t2 (Fig. 5.5B ). The burst in is at the correct place in the sequence, but the burst of is not. The  connection is included in the network to prevent the burst. The burst induces a burst in

, which projects to and inhibits at time t 2. also inhibits , t terminating its burst and causing hyperpolarization during the second interval of time 2 (division in row of Fig. 5.5B). The net result is that does not burst at time t2.

As in network one, the propagation of sequential activity in is driven by the rebound firing of . However, unlike network one, and issue two bursts of activity within a song motif. This network exemplifies how additional well-placed connections can alter the number of and bursts in the motif without altering the number of

bursts.

5.3.3 Network Architecture 3

The third network includes randomness in the selectivity of some projections. It is represented by the model system in Fig. 5.6A and the activity patterns that it displays are depicted in Fig. 5.6B. Unlike before, sends excitatory projections to in the same microcircuit and to a random selection of other from the other microcircuits in the network. Since projects randomly to , sequential bursting in HVC’s projection neurons requires the development of appropriate counteracting connections, as described below (and not shown in Fig. 5.6A). Consider the burst firing of at time t1. Suppose that

is synaptically coupled with and , in addition to . This coupling produces bursts in all three interneuron populations at time t . The burst in 1 hyperpolarizes , and the bursts in and hyperpolarize

and , respectively. The first column of Fig. 5.6B illustrates these events.

152

Figure 5.6: Arrangement of prototype network architecture three (A). Blue curves with arrow heads represent excitatory connections, and red curves with circles represent inhibitory connections. Dotted lines represent random projections. B. Activity patterns for network architecture three. For more info on symbols, refer to legend of Fig. 5.2.

153

The buildup of intracellular Ca2+ terminates activity, which in turn terminates the

, and bursts. This allows , and to escape from

inhibition and fire a postinhibitory rebound burst at time t2 . Since excites ,

, and are stimulated at time t2 (Fig. 5.6B). The resulting burst is

at the correct time in the sequence, while potential and bursts are not. Suppose that synapses onto and as well as . Then the burst induces bursts in , and . These interneurons would then need to synapse onto and inhibit and so that the ill-timed bursts are suppressed. For example,

 ,  and  connections would be effective in

doing this. A similar scenario of random efferents countered by appropriate

efferents would apply at each microcircuit.

Network three requires appropriate  inhibitory connections to prohibit

“unwanted” bursts at the wrong times in the sequence. The appropriate connections are determined by the randomly chosen  connections. The number of bursts produced by each is determined by the number of excitatory projections it receives. The number of bursts produced by each is determined by the number of inhibitory projections it receives as well as by the number of bursts produced by these neurons.

Thus, this network allows for a large number of and bursts during a motif while preserving the sequence of activity in the neurons. Like networks one and two, the propagation of sequential behavior is driven by the rebound firing of cells.

5.3.4 Network Architecture 4

Network four builds on network three by adding random projections from to

. Network four is illustrated in Fig. 5.7A and the activity patterns it displays are depicted in

Fig. 5.7B. Unlike the previous three networks, each sends inhibitory projections to

in the same microcircuit and to a random selection of in the other microcircuits.

Similar to network three, sequential bursting in HVC’s projection neurons requires a careful development of appropriate counteracting connections.

154

Figure 5.7: Arrangement of prototype network architecture four (A). Blue curves with arrow heads represent excitatory connections, and red curves with circles represent inhibitory connections. Dotted lines represent random projections. B. Activity patterns for network architecture four. For more info on symbols, refer to legend of Fig. 5.2.

155

Consider the burst firing of at time t1. Suppose that these neurons are synaptically coupled with and , in addition to . This coupling produces bursts in all four interneuron populations at time t1. Suppose also that is synaptically coupled with and in addition to , and is synaptically coupled with

and in addition to . The burst in hyperpolarizes

and , the burst in hyperpolarizes and and the

bursts in and hyperpolarize and , respectively. The first column of Fig. 5.7B illustrates these events.

Intracellular Ca2+ buildup terminates activity, which in turn terminates the ,

and bursts. This allows , and to escape

from inhibition and fire postinhibitory rebound bursts at time t 2 . Since excites , bursts in , and are potentially elicited at time t 2 (Fig. 5.7B). The

burst is at the correct time in the sequence, while the and

bursts are not. Suppose that the burst induces bursts in , and

due to the random excitatory coupling. To suppress the unwanted bursts, these

interneurons could project as  ,  and  .

Therefore the inhibitory projections from , and onto

and prohibit and from issuing bursts at time t2. Similar scenarios follow in the other microcircuits.

Similar to network three, network four requires appropriate  inhibitory connections to prevent the bursts that occur at the wrong times, and these appropriate  connections are determined by the randomly chosen connections. Since

’s rebound bursting is stronger in this network due to the large number of

inhibitory projections they receive, ’s excitation is stronger and thus this network

requires stronger inhibition from to suppress the unwanted bursts. In addition to the high number of bursts produced by , network four also allows for a larger number of

bursts than in the previous three networks.

The first four networks have one key common element: the sequential bursting that occurs in the RA-projecting neurons is driven by the rebound firing of the X-projecting neurons after

156

their release from inhibition. The next network produce sequential firing of that is largely independent of and activity.

5.3.5 Network Architecture 5

The last network architecture we consider in Fig. 5.8A suggests a functional independence of the neurons from both and neurons. In this network, sends excitatory projections to in the next microcircuit. Synaptic weights for  and  are set at low values so that activity is almost independent of the activity of the other neural populations. Figure 5.8B displays the activity pattern for this network. The activity patterns for the and are not shown in Fig. 5.8B because they have little influence on the

sequential activity of . When bursts at time t1, it excites due to the excitatory coupling. The synaptic weights for the  projections are set to moderately low values so that there is some latency before produces its first spike. The burst generated by excites in its turn, which then fire its burst at time t3 due to the

 excitatory coupling. The sequential firing in the RA-projecting neurons

propagates throughout the microcircuit chain in a similar fashion. The bursts overlap, and the onset of the overlap between and depends on the strength of the synaptic conductance from to ( ). The stronger this synaptic conductance is, the greater the overlap. We also developed another network architecture where the propagation of sequential activity is dependent on both and neurons solely. In this network, the only way sequential activity can propagate is through a rebound firing mechanism in neurons.

However since our slice work studies (Chapter 4) have shown that neurons do not elicit rebound firing after release from inhibitory applied currents, we did not pursue that further.

Moreover, this shows that the only possible neural architecture that could support the Scharff et al work (2000) is network architecture 5 where activity is controlled by neurons solely.

157

Figure 5.8: Arrangement of prototype network architecture five (A). Blue curves with arrow heads represent excitatory connections, and red curves with circles represent inhibitory connections. B. Activity patterns for network architecture five. For more info on symbols, refer to legend of Fig. 5.2.

5.4 Effects of Temperature on the HVC Network

As we explained in section 2.8.3, the song tempo is hypothesized to be determined by the intrinsic biophysical properties of the HVC local network (Mooney 2009b). Supporting evidence on that came from Michale Fee’s group when they bilaterally cooled the HVC during singing by use of a thermoelectric heat pump (Long and Fee 2008). Bilateral cooling of HVC revealed that all aspects of song timing - duration of song syllable, interval between syllable onsets, and interval between motif onsets - are slowed after HVC cooling (Fig. 2.7).

158

We attempted here to simulate cooling and heating the HVC networks to monitor the effects of temperature on the propagation of sequential activity. The electrical properties in the model that were affected by temperature are the maximal conductances, the rates of channel opening and closing and synaptic connections, using the usual -formalism (Hille 2001;

Rinberg et al. 2013; Tang et al. 2010). In particular,

and ,

where is any of the maximal conductances in the network, is the value of at the reference temperature, , which is chosen to be 23 . Moreover, is the factor by which the parameter is scaled at the current temperature, ; and℃ describes the temperature sensitivity. With this alteration to the model, we hope to illuminate temperature effects on HVC network activity and how they relate to the different inward and outward conductance dynamics. We are particularly interested in determining whether the HVC network remains oscillatory and sequential activity is preserved, or whether it becomes quiescent (non-oscillatory), and the sequential propagation of activity crashes down, since these are the most salient features of the HVC network. Figure 5.9 shows the results of cooling or heating HVC network architecture one by 15 . Panel A shows the membrane potential of the five HVC neurons in the architecture under RA ℃ control conditions, that is the temperature was not altered. The HVCRA neurons fire sequentially spanning a motif of duration ~500 msec. Figure 5.9B shows network’s one activity when the system is cooled down by 15 . Cooling down the kinetics of the network slowed down its activity as the motif duration increased℃ by ~200 msec. Moreover, the individual HVCRA neurons’ burst durations are shorter and the durations of the silent gaps between bursts are longer. This agrees with Long and Fee (2008) results where they observed that at colder temperatures, song motifs were produced more slowly than control songs and there was a significant increase in motif duration during cooling (Fig. 2.7). Figure 5.9C shows the network activity when the system is heated up by 15 . Heating the network decreased the motif duration by ~ 100 msec, increased the durations of ℃individual bursts and decreased the interburst intervals. This agrees again with the Long and Fee (2008) results as warming up the HVC caused a decrease in motif duration (Fig. 2.7).

159

Figure 5.9: Effects of temperature on network architecture one activity. A. Membrane potential of

HVCRA neurons under control conditions exhibiting sequential firing. B. Cooling the network by 15 slowed down the activity. Note the time scale difference between A and B. Moreover, burst durations are shorter and the silent gaps between bursts are longer. C. Heating the network by 15 increased the℃ network activity. The interburst intervals are shorter, the intraburst intervals are longer and overall the motif is shorter by ~ 100 msec. ℃

160

Figure 5.10 shows the effects of temperature on individual HVCRA neurons’ firing pattern. Panel A shows neuron’s bursting pattern under control conditions. Cooling the network decreased the frequency of firing within the neuron’s burst (panel B). It also increased the interspike and intraspike intervals within the burst. Heating the network on the other hand had an opposite effect. The frequency of firing increased within the burst, the interspike and intraspike intervals decreased, and the burst overall duration increased as well (panel C). As described earlier, our HVC model neurons connected in the networks are representative neurons of larger populations, and therefore individual model HVCRA bursts represent an ensemble of HVCRA neurons bursting during a particular syllable, and silent intervals between model HVCRA bursts represent silent gaps between syllables. With this comparison in mind, the results here second all the results obtained by Long and Fee (2008) as shown in Figs. 5.9 and 5.10. Long and Fee (2008) also noticed that the stretch of syllable-onset intervals was significantly correlated with stretch of the syllable contained within that interval. In other words, for syllables that exhibited a larger stretch than average, the onset interval to the following syllable also exhibited a larger stretch than average. Our results here also agree with their findings as we show that the onset of each syllable in the microcircuits is linked to and triggered by the end of the previous syllable. Their finding that the silent gaps between syllables were also significantly dilated by HVC cooling was also observed in our simulations. Combined, these results suggest that the biophysical dynamics in HVC are involved in controlling song timing on a fine timescale including syllable durations, interval between syllable onset and silent gaps between syllables. These results raise interesting questions: Given the fact that HVC is a bilateral structure, how are the two HVCs coordinated during singing? The network architectures we developed here do not take interhemispheric coordination into account, yet this raises the question whether both HVCs contribute to song timing and how. Long and Fee (2008) observed that cooling either HVC alone caused song slowing intermediate to that seen for bilateral cooling thus ruling out the possibility that song timing is controlled by dynamics in a single hemisphere. They also show that there may be some switching of the control of song timing between the two HVCs on the timescale of song syllables or long subsyllabic elements. These results combined motivate the need to develop network architectures that would take interhemispheric coordination into account.

161

Figure 5.10: Effects of temperature change on HVCRA neurons’ firing pattern. A. Membrane potential of neuron under control conditions. B. Cooling the network by 15 decreased the number of spikes as well as increased the interspike intervals within the burst. C. Heating the network by 15 increased the number of spikes, reduced the interspike intervals within the℃ burst and increased the burst duration. ℃

162

5.5 Discussion

We have presented computational models that describe how sequential activity in HVC projection neurons can be produced during singing. These microcircuit chains are envisioned to have many copies throughout the HVC, and the chain ensemble is envisioned to drive singing. The different chain architectures that we described produce different activity patterns of HVC neurons, and lead to different predictions. For example, inhibitory projections are crucial for the sequential activity in the first four architectures, so antagonism of GABA receptors would terminate the activity pattern. The first four architectures predict a key role for postinhibitory rebound of neurons. The different networks also predict a different number of bursts per motif of and of neurons. At present, there are insufficient data to determine which of the proposed architectures are best representative of the HVC circuitry. However, the model networks provide insight into how the different types of HVC neurons can be used for sequence generation. The different networks vary in their robustness to parameter variation. This robustness was determined by simulating each network 100 times with random variation in a single parameter (the AHP conductances, synaptic conductances or applied current) about its default value. The network response was considered accurate if the network generated sequential bursting in at least 90% of the simulations. The range of variation of the randomly-varying parameter was increased up until the point that the network ceased to be accurate. This range of variation was then taken as the measure of robustness to variation in that parameter. Networks one and two were robust to ~40% variations in synaptic conductances, ~15% variations in AHP conductances and ~20% variations in the external applied currents, where the variation could be an increase or decrease in the parameter value. Networks three and four were less robust to variations: they were robust to ~30% variations in synaptic conductances, ~10% in AHP conductances and ~15% in external applied currents. Network five was the most robust of all network architectures as it was robust to variations of ~30% synaptic conductances, ~30% AHP conductances and ~25% applied current. Thus, all network models are more robust to random variations in synaptic conductances than to variations in the AHP current or the input to HVC. Chain models have previously been proposed for the HVC (Jin et al. 2007; Li and Greenside 2006; Long et al. 2010). Although the HVC neurons in our model are segregated into

163

a chain-like network of microcircuits, we have focused on the interactions among the three classes of HVC neurons. Unlike other models, our simulations and analysis propose that postinhibitory rebound in most networks, particularly networks one through four, plays a key role in preserving precise timing information that enables reliable propagation of sequential activity throughout the HVC microcircuits. This mechanism is a characteristic of some invertebrate central pattern generators which use inhibition and recurrent excitation and which exhibit postinhibitory rebound (Goaillard et al. 2010; Satterlie 1985). In a previous model of the HVC by Troyer and Doupe (2000b), it was proposed that sequence generation results from a reciprocal excitatory chaining of motor and sensory representations of HVC projection neurons. The motor representation resides in the HVCRA neurons and the sensory representation resides in an efference copy in the HVCX neurons. They proposed that the connections from HVCX to HVCRA are necessary for sequence generation. Although their approach was different than ours, their results match the results of our first four architectures in which sequence generation is driven by the HVCX  HVCRA excitatory coupling after HVCX neurons escape their inhibition and fire a rebound burst. Interactions between the inhibitory interneurons and the RA-projecting neurons play a key role in the Drew and Abbott (2003) model of song selectivity and sequence generation in HVC. As in our models, their model includes an AHP current that plays a crucial role in the generation of temporal sequences. The sequence in their model is driven by periodic input to HVC neurons. Katahira et al (2007) also proposed a model in which groups of inhibitory interneurons in HVC form branching chain networks, and sequential activity in HVCRA groups is generated by periodic burst inputs from Uva to the HVCRA neurons. While Uva provides input to HVC, it has not been established that this input is needed to drive sequential bursting in the HVC projection neurons (Coleman and Vu 2005). In contrast, our networks assume constant input to the HVC; the sequential bursting is produced by the excitatory and inhibitory interactions among the different neuronal populations. In addition, the syllable and intersyllable durations are determined by the intrinsic dynamics of the HVC neurons. Other models in which inhibitory neurons play an important role are (Jin 2009) and (Gibb et al. 2009b). In the former, syllables are driven by chain networks of RA-projecting neurons, and random connections between the HVCRA and HVCINT neurons provide mutual inhibition among the HVCRA neurons enabling reliable propagation of sequential activity. In the later,

164

HVCRA neurons are organized as a chain of bistable clusters. Once in the “on” state, an adaptation current builds up causing spike frequency adaptation and eventual burst termination. The various networks that we have examined raise developmental questions. What developmental mechanism could generate the pattern of HVCINT  HVCRA connectivity in networks two, three and four? Little is known about the development of the inhibitory inputs from interneurons onto the projection neurons within the HVC. However, we suggest a mechanism whereby the excitatory coupling between the projection neurons and the interneurons develops early in the bird’s life and interneuron connectivity develops later as the bird attempts to reproduce the song of its tutor. Thus, the interneurons inhibit specific projection neurons to move the brain circuitry in a direction that minimizes the error in vocal output relative to that of the tutor’s song.

165

CHAPTER 6

CONCLUSIONS

6.1 Summary

This dissertation provided new insights to the study of songbirds on different levels. In Chapter 3, we introduced SongSeq, the first automated tool designed for analyzing birdsong syllable sequences. We showed how this software can be used to monitor changes of sound features across a large number of songs, analyze transition distributions among syllables, quantify syllable ordering, and quantify the degree of similarity in song syntax over different days of singing. In Chapter 4, we provided the first characterization of the three types of HVC neurons and introduced the first biophysically accurate computational model of the HVC neurons. Using in vitro whole-cell electrical recordings, we recorded and identified HVC neurons within brain slices of the zebra finch and examined their intrinsic firing properties. We determined the ionic currents that are responsible for the HVC neurons’ characteristic firing patterns. We also developed conductance-based models for the different neurons and calibrated these models using the data from our brain slice work. The model predictions were all tested and verified in the slice using pharmacological manipulations which shows the validity of our models and their biophysically accurate representations of HVC neurons. In the last part of the dissertation (Chapter 5), we presented computational models that describe how sequential activity in HVC projection neurons can be produced during singing. Although chain models had been proposed before, we presented the first network models where the interactions are among the three classes of HVC neurons. The model networks provide insight into how the different types of HVC neurons can be used for sequence generation.

166

6.2 Future Work

The dissertation addressed the “internal anatomy” of the HVC from the biophysical and cellular level as well as from the network level. In the process, it also revealed a number of obstacles and raised several questions that have yet to be investigated. It is of great significance that we took the first step in identifying the ionic currents of HVC neurons and showed that they play a key role in controlling their characteristic firing patterns; however, the exact roles that these ionic currents exert in controlling the vocal pattern during singing remains to be investigated. The in vitro studies presented here motivate the speculation on the roles these ionic currents play from a circuit-level perspective. The hidden knights behind the symphony orchestrated by HVC during song production are the ionic currents of HVC neurons and the synaptic interactions between the HVC neurons themselves, because it is these components of the HVC that encode the song. In addition to that, it remains to be investigated what roles the ionic currents play during development. As we discussed in Chapter 2, the birdsong passes through a series of phases during development that ranges from little or no structure due to unstable notes production to crystallized songs that are highly stereotyped. Are the developmental changes seen in behavior due to intrinsic changes within the HVC neurons that could be due to changes in ionic currents gating? Or is it merely a consequence of building the synaptic interactions necessary to generate a network that is able to produce a crystallized song? The models that we developed are single compartment models. One could construct multi- compartment models to better represent the spatial aspects of HVC neurons, and to do this well it would be necessary to first study dendritic properties of the neurons, including ion channel distribution. In our model description of ionic currents we utilized functional forms used in prior published neural modeling studies of non-HVC neurons. To fit these to HVC neurons it would be necessary to obtain voltage-clamp recordings to calibrate shape parameters. Finally, pharmacological blockers could be used with a range of current pulses to better constrain the model parameter values. The various networks that we have examined raised developmental questions. What developmental mechanisms could generate the pattern of connectivity between the various HVC neurons? Little is known about the development of the inhibitory inputs from interneurons onto

167

the projection neurons within the HVC. However, our results suggest a mechanism whereby the excitatory coupling between the projection neurons and the interneurons develops early in the bird’s life and interneuron connectivity develops later as the bird attempts to reproduce the song of its tutor. Thus, the interneurons inhibit specific projection neurons to move the brain circuitry in a direction that minimizes the error in vocal output relative to that of the tutor’s song. These predictions proposed by our models motivate the need to test them experimentally. This of course requires more investigation on the connectivity patterns among the different types of HVC neurons.

6.3 Conclusion

In closing, the work presented in this dissertation has the potential to make a considerable impact in the field of songbird research. The ionic currents identified in brains slices and the computational models developed provide a large step forward in describing the biophysics of HVC neurons. This work serves as step towards discussing realistic and plausible neural architectures that would reflect an accurate topography of the nucleus as well as produce the characteristic patterns of neural activity exhibited by the various HVC neurons during singing. Being able to explain the complete story of the neural code behind HVC’s sequential activity could open a new era to science.

168

APPENDIX A

SOURCE CODES

The simulations done in this dissertation are written using various programming languages, including JAVA, Matlab and XPPAUT. The corresponding source codes are extremely large, and for this reason we only summarize their details here and provide links to the urls where they are available for free download online.

A1 SongSeq

SongSeq is written using the JAVA programming language, which is available for free download at http://www.oracle.com/technetwork/java/index.html. As graphical user interface software, SongSeq utilized the Swing and AWT packages in JAVA for their nice graphical capabilities. Uploading and parsing the Excel spreadsheets generated by SA+ is programmed by utilizing the JDBC (Java Data Base Connectivity) package and the HSSF package from POI. The software is written using 97 JAVA classes, and they are all available online for free download along with a user manual on how to use it and instructions for download on the following url: http://www.math.fsu.edu/~bertram/software/birdsong/JNM_12/. The work done with this software is published in the Journal of Neuroscience Methods (Daou et al. 2012).

A2 HVC Models

The simulations done with the HVC model (Chapter 4) are done using the CVODE numerical integrator in XPPAUT (http://www.math.pitt.edu/~bard/xpp/xpp.html) and the ode45 solver in Matlab (http://www.mathworks.com/products/matlab/). The source codes for these simulations containing the models including the source code to generate each figure are available online at http://www.math.fsu.edu/~bertram/software/birdsong, as well as at ModelDB website located at http://senselab.med.yale.edu/modeldb/. The results in this work are published in the Journal of Neurophysiology (Daou et al. 2013).

169

A3 Synchronization Index

In a separate work that is not mentioned in this dissertation, we investigated the ability of imposed glucose waveforms to synchronize a population of mouse islets of Langerhans. Very briefly, pancreatic islets secrete insulin in a pulsatile manner, and the individual islets are synchronized, producing in vivo oscillations. Michael Roper’s group at FSU designed a microfluidic system that was used to deliver glucose waveforms to ~20 islets while fura 2 fluorescence was imaged. They entrained all islets to a sinusoidal waveform of glucose producing synchronized oscillations of fura 2 fluorescence. During perfusion with constant glucose, oscillations of fura 2 fluorescence were observed in individual islets. To quantify the experimental data collected, we developed the synchronization index () that measures the phase difference between two oscillators. We used this synchronization index to measure the period of fura 2 fluorescence oscillations and evaluate the synchronization of islets. We then wrote graphical user interface software using Matlab that determines the degree of synchrony among two or more calcium oscillations. The source code is available online for free download at http://www.math.fsu.edu/~bertram/software/islet/AJP_11b/. Moreover, the results of this work are published in the American Journal of Physiology (Zhang et al. 2011).

170

APPENDIX B

LIST OF ABBREVIATIONS

SAP – Sound Analysis Pro HVC – High Vocal Center VMP – Vocal Motor Pathway RA – Robust nucleus of the Arcopallium RAm – Nucleus Retroambigualis PAm – Nucleus Parambigualis nXIIts – Tracheosyringeal portion of the nucleus of the twelfth nerve AFP – Anterior Forebrain Pathway VTA – Ventral Tegmental Area LMAN – Lateral Magnocellular nucleus of the Anterior Nidopallium MMAN – Medial Magnocellular nucleus of the Anterior Nidopallium MLD – Dorsal Lateral nucleus of the Mesencephalon IC – Inferior Colliculus Ov – Ovoidalis MGB – Medial Geniculate Body NCM – Caudal Medial Nidopallium CMM – Caudal Medial Mesopallium CLM – Caudal Lateral Mesopallium CM – Caudal Mesopallium NIf – Interfacial Nucleus of the Nidopallium Uva – Uvaeformis FM – Frequency Modulation K-L – Kullback-Leibler NMDA – N-methyl-D-Aspartate CNQX – 6-cyano-7-nitroquinoxaline-2,3-dione ACSF – Artificial Cerebro Spinal Fluid EGTA – Ethylene Glycol Tetraacetic Acid

171

HEPES – 4-(2-HydroxyEthyl)-1-PiperazineEthaneSulfonic acid AMPA – 2-Amino-3-(3-hydroxy-5-Methyl-isoxazol-4-yl) Propanoic acid GABA – Gamma-AminoButyric Acid 4-AP – 4-AminoPyridine RMP – Resting Membrane Potential AHP – After HyperPolarization TTP – Time-To-Peak SR – Sag Ratio AR – Adaptation Ratio AP – Action Potential

HVCRA – HVC neurons projecting to RA

HVCX – HVC neurons projecting to area X

HVCINT – HVC interneurons

Ih – Hyperpolarization activated inward current + IA – A-type K current 2+ + ISK or SK – Ca -dependent K current 2+ + IKNa – Na -dependent K current 2+ ICa-T – Low-threshold T-type Ca current + INap – Persistent Na current + INa – Transient Na current + IK – Delayed rectifier K current

IL – Leak current

172

REFERENCES

Abarbanel H, Talathi S, Mindlin G, Rabinovich M, and Gibb L. Dynamical model of birdsong maintenance and control. Physical Review E 70: 051911, 2004a.

Abarbanel HD, Gibb L, Mindlin GB, Rabinovich MI, and Talathi S. Spike timing and synaptic plasticity in the premotor pathway of birdsong. Biol Cybern 91: 159-167, 2004b.

Abarbanel HD, Meliza C, Kostuk M, Huang H, and Margoliash D. Dynamic parameter estimation for biophysical models of HVC neurons. In: Society for Neuroscience. New Orleans, LA: 2012.

Abarbanel HD, Talathi SS, Mindlin G, Rabinovich M, and Gibb L. Dynamical model of birdsong maintenance and control. Phys Rev E Stat Nonlin Soft Matter Phys 70: 051911, 2004c.

Abarbanel HDI. Mapping Neural Architectures Onto Acoustic Features of Birdsong. Journal of Neurophysiology 92: 96-110, 2004.

Adret-Hausberger M, and Jenkins PF. Complex organization of the warbling song in starlings. Behaviour 107: 138-155, 1988.

Aghajanian GK, and Rasmussen K. Intracellular studies in the facial nucleus illustrating a simple new method for obtaining viable motoneurons in adult rat brain slices. Synapse 3: 331-338, 1989.

Altman J. Are new neurons formed in the brains of adult mammals? Science 135: 1127-1128, 1962.

Alvarez-Buylla A, and Kirn JR. Birth, migration, incorporation, and death of vocal control neurons in adult songbirds. J Neurobiol 33: 585-601, 1997.

Angstadt JD, and Calabrese RL. A hyperpolarization-activated inward current in heart interneurons of the medicinal leech. J Neurosci 9: 2846-2857, 1989.

Aronov D, Andalman AS, and Fee MS. A specialized forebrain circuit for vocal babbling in the juvenile songbird. Science 320: 630-634, 2008.

Ashmore RC. Brainstem and Forebrain Contributions to the Generation of Learned Motor Behaviors for Song. Journal of Neuroscience 25: 8543-8554, 2005.

173

Banks MI, Pearce RA, and Smith PH. Hyperpolarization-activated cation current (Ih) in neurons of the medial nucleus of the trapezoid body: voltage-clamp analysis and enhancement by norepinephrine and cAMP suggest a modulatory mechanism in the auditory brain stem. J Neurophysiol 70: 1420-1432, 1993.

Baptista LF, and Petrinovich L. Social interaction, sensitive phases and the song template hypothesis in the white-crowned sparrow. Anim Behav 32: 172-174, 1984.

Baptista LF, and Petrinovich L. Song development in the white-crowned sparrow: social factors and sex differences. Animal Behaviour 34: 1359-1371, 1986.

Basham ME, Nordeen EJ, and Nordeen KW. Blockade of NMDA receptors in the anterior forebrain impairs sensory acquisition in the zebra finch (Poephila guttata). Neurobiol Learn Mem 66: 295-304, 1996.

Bauer EE, Coleman MJ, Roberts TF, Roy A, Prather JF, and Mooney R. A synaptic basis for auditory-vocal integration in the songbird. J Neurosci 28: 1509-1522, 2008.

Bean BP. The action potential in mammalian central neurons. Nat Rev Neurosci 8: 451-465, 2007.

Beecher MD, and Brenowitz EA. Functional aspects of song learning in songbirds. Trends Ecol Evol 20: 143-149, 2005.

Berwick RC, Okanoya K, Beckers GJ, and Bolhuis JJ. Songs to syntax: the linguistics of birdsong. Trends Cogn Sci 15: 113-121, 2011.

Bolhuis JJ, and Gahr M. Neural mechanisms of birdsong memory. Nat Rev Neurosci 7: 347-357, 2006.

Bolhuis JJ, Okanoya K, and Scharff C. Twitter evolution: converging mechanisms in birdsong and human speech. Nat Rev Neurosci 11: 747-759, 2010.

Bottjer SW, and Arnold AP. Developmental plasticity in neural circuits for a learned behavior. Annu Rev Neurosci 20: 459-481, 1997.

Bottjer SW, Halsema KA, Brown SA, and Miesner EA. Axonal connections of a forebrain nucleus involved with vocal learning in zebra finches. J Comp Neurol 279: 312-326, 1989.

174

Bottjer SW, and Johnson F. Circuits, hormones, and learning: vocal behavior in songbirds. J Neurobiol 33: 602-618, 1997.

Bottjer SW, Miesner EA, and Arnold AP. Forebrain lesions disrupt development but not maintenance of song in passerine birds. Science 224: 901-903, 1984.

Boughman JW. Vocal learning by greater spear-nosed bats. Proc Biol Sci 265: 227-233, 1998.

Brainard MS, and Doupe AJ. Auditory feedback in learning and maintenance of vocal behaviour. Nat Rev Neurosci 1: 31-40, 2000a.

Brainard MS, and Doupe AJ. Interruption of a basal ganglia-forebrain circuit prevents plasticity of learned vocalizations. Nature 404: 762-766, 2000b.

Brainard MS, and Doupe AJ. What songbirds teach us about learning. Nature 417: 351-358, 2002.

Brown RG, Jahanshahi M, Limousin-Dowsey P, Thomas D, Quinn NP, and Rothwell JC. Pallidotomy and incidental sequence learning in Parkinson's disease. Neuroreport 14: 21-24, 2003.

Budde T, Biella G, Munsch T, and Pape HC. Lack of regulation by intracellular Ca2+ of the hyperpolarization-activated cation current in rat thalamic neurones. J Physiol 503 ( Pt 1): 79-85, 1997.

+ + Budde T, White JA, and Kay AR. Hyperpolarization-activated Na -K current (Ih) in neocortical neurons is blocked by external proteolysis and internal TEA. J Neurophysiol 72: 2737-2742, 1994.

Cardin JA, Raksin JN, and Schmidt MF. Sensorimotor nucleus NIf is necessary for auditory processing but not vocal motor output in the avian song system. J Neurophysiol 93: 2157-2166, 2005.

Cardin JA, and Schmidt MF. Noradrenergic inputs mediate state dependence of auditory responses in the avian song system. J Neurosci 24: 7745-7753, 2004.

Catchpole CK, and Slater PJB. Bird Song: Biological Themes and Variations. Cambridge University Press, 2008.

175

Catchpole CK, and Slater PJB. Bird Song: Biological Themes and Variations. Cambridge: Cambridge University Press, 1995.

Chi Z, and Margoliash D. Temporal precision and temporal drift in brain and behavior of zebra finch song. Neuron 32: 899-910, 2001.

Clark BA, Monsivais P, Branco T, London M, and Hausser M. The site of action potential initiation in cerebellar Purkinje neurons. Nat Neurosci 8: 137-139, 2005.

Colbert CM, and Johnston D. Axonal action-potential initiation and Na+ channel densities in the soma and axon initial segment of subicular pyramidal neurons. J Neurosci 16: 6676-6686, 1996.

Coleman MJ, and Vu ET. Recovery of impaired songs following unilateral but not bilateral lesions of nucleus uvaeformis of adult zebra finches. J Neurobiol 63: 70-89, 2005.

Connors BW, and Gutnick MJ. Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci 13: 99-104, 1990.

Coombs JS, Curtis DR, and Eccles JC. The interpretation of spike potentials of motoneurones. J Physiol 139: 198-231, 1957.

Daou A, Johnson F, Wu W, and Bertram R. A computational tool for automated large-scale analysis and measurement of bird-song syntax. J Neurosci Methods 210: 147-160, 2012.

Daou A, Ross MT, Johnson F, Hyson RL, and Bertram R. Electrophysiological characterization and computational models of HVC neurons in the zebra finch Journal of Neurophysiology, 2013.

Day NF, Kinnischtzke AK, Adam M, and Nick TA. Daily and developmental modulation of "premotor" activity in the birdsong system. Dev Neurobiol 69: 796-810, 2009.

Day NF, Terleski KL, Nykamp DQ, and Nick TA. Directed functional connectivity matures with motor learning in a cortical pattern generator. J Neurophysiol 109: 913-923, 2013.

Dehaene-Lambertz G, Hertz-Pannier L, Dubois J, Meriaux S, Roche A, Sigman M, and Dehaene S. Functional organization of perisylvian activation during presentation of sentences in preverbal infants. Proc Natl Acad Sci U S A 103: 14240-14245, 2006.

Deregnaucourt S, Mitra PP, Feher O, Maul KK, Lints TJ, and Tchernichovski O. Song development: in search of the error-signal. Ann N Y Acad Sci 1016: 364-376, 2004.

176

Descalzo VF, Nowak LG, Brumberg JC, McCormick DA, and Sanchez-Vives MV. Slow adaptation in fast-spiking neurons of visual cortex. J Neurophysiol 93: 1111-1118, 2005.

Deschenes M, Roy JP, and Steriade M. Thalamic bursting mechanism: an inward slow current revealed by membrane hyperpolarization. Brain Res 239: 289-293, 1982.

Destexhe A, and Babloyantz A. A model of the inward current Ih and its possible role in thalamocortical oscillations. Neuroreport 4: 223-226, 1993.

Doupe A, Perkel D, Reiner A, and Stern E. Birdbrains could teach basal ganglia research a new song. Trends in Neurosciences 28: 353-363, 2005.

Doupe AJ. Song- and order-selective neurons in the songbird anterior forebrain and their emergence during vocal development. J Neurosci 17: 1147-1167, 1997.

Doupe AJ, and Konishi M. Song-selective auditory circuits in the vocal control system of the zebra finch. Proc Natl Acad Sci U S A 88: 11339-11343, 1991.

Doupe AJ, and Kuhl PK. Birdsong and human speech: common themes and mechanisms. Annu Rev Neurosci 22: 567-631, 1999.

Doya K, and Sejnowski TJ. A novel reinforcement model of birdsong vocalization learning. Adv Neural Inf Process Syst 7: 101-108, 1995.

Drew PJ, and Abbott LF. Model of song selectivity and sequence generation in area HVc of the songbird. J Neurophysiol 89: 2697-2706, 2003.

Dunmyre JR, Del Negro CA, and Rubin JE. Interactions of persistent sodium and calcium- activated nonspecific cationic currents yield dynamically distinct bursting regimes in a model of respiratory neurons. J Comput Neurosci 31: 305-328, 2011.

Dutar P, Vu HM, and Perkel DJ. Multiple cell types distinguished by physiological, pharmacological, and anatomic properties in nucleus HVC of the adult zebra finch. J Neurophysiol 80: 1828-1838, 1998.

Egnor SE, and Hauser MD. A paradox in the evolution of primate vocal learning. Trends Neurosci 27: 649-654, 2004.

Erisir A, Lau D, Rudy B, and Leonard CS. Function of specific K+ channels in sustained high- frequency firing of fast-spiking neocortical interneurons. J Neurophysiol 82: 2476-2489, 1999.

177

Fadiga L, Craighero L, Buccino G, and Rizzolatti G. Speech listening specifically modulates the excitability of tongue muscles: a TMS study. Eur J Neurosci 15: 399-402, 2002.

Fain G. Molecular and Cellular Physiology of Neurons. Cambridge, MA: Harvard Univ. Press, 1999.

Farries MA, Ding L, and Perkel DJ. Evidence for direct and indirect pathways through the song system basal ganglia. The Journal of Comparative Neurology 484: 93-104, 2005.

Fee MS, and Goldberg JH. A hypothesis for basal ganglia-dependent reinforcement learning in the songbird. Neuroscience 198: 152-170, 2011.

Fee MS, Kozhevnikov AA, and Hahnloser RH. Neural mechanisms of vocal sequence generation in the songbird. Ann N Y Acad Sci 1016: 153-170, 2004.

Fee MS, and Long MA. New methods for localizing and manipulating neuronal dynamics in behaving animals. Curr Opin Neurobiol 21: 693-700, 2011.

Fee MS, and Scharff C. The songbird as a model for the generation and learning of complex sequential behaviors. ILAR J 51: 362-377, 2010.

Foote AD, Griffin RM, Howitt D, Larsson L, Miller PJ, and Hoelzel AR. Killer whales are capable of vocal learning. Biol Lett 2: 509-512, 2006.

Fortune ES, and Margoliash D. Cytoarchitectonic organization and morphology of cells of the field L complex in male zebra finches (Taenopygia guttata). J Comp Neurol 325: 388-404, 1992.

Foster EF, Mehta RP, and Bottjer SW. Axonal connections of the medial magnocellular nucleus of the anterior neostriatum in zebra finches. J Comp Neurol 382: 364-381, 1997.

Fraser DD, and MacVicar BA. Low-threshold transient calcium current in rat hippocampal lacunosum-moleculare interneurons: kinetics and modulation by neurotransmitters. J Neurosci 11: 2812-2820, 1991.

Friedman A, and Gutnick MJ. Low-threshold calcium electrogenesis in neocortical neurons. Neurosci Lett 81: 117-122, 1987.

Fromkin V, Krashen S, Curtis S, Rigler D, and Rigler M. The development of language in Genie: a case of language acquisition beyond the 'critical period'. Brain and Language 1: 81-107, 1974.

178

Fujimoto H, Hasegawa T, and Watanabe D. Neural coding of syntactic structure in learned vocalizations in the songbird. J Neurosci 31: 10023-10033, 2011.

Gerber U, Greene RW, and McCarley RW. Repetitive firing properties of medial pontine reticular formation neurones of the rat recorded in vitro. J Physiol 410: 533-560, 1989.

Gibb L, Gentner TQ, and Abarbanel HD. Brain stem feedback in a computational model of birdsong sequencing. J Neurophysiol 102: 1763-1778, 2009a.

Gibb L, Gentner TQ, and Abarbanel HD. Inhibition and recurrent excitation in a computational model of sparse bursting in song nucleus HVC. J Neurophysiol 102: 1748-1762, 2009b.

Gibb L, Gentner TQ, and Abarbanel HDI. Inhibition and recurrent excitation in a computational model of sparse bursting in song nucleus HVC. Journal of Neurophysiology 102: 1748-1762, 2009c.

Glaze CM, and Troyer TW. Behavioral measurements of a temporally precise motor code for birdsong. J Neurosci 27: 7631-7639, 2007.

Glaze CM, and Troyer TW. Temporal structure in zebra finch song: implications for motor coding. J Neurosci 26: 991-1005, 2006.

Goaillard JM, Taylor AL, Pulver SR, and Marder E. Slow and persistent postinhibitory rebound acts as an intrinsic short-term memory mechanism. J Neurosci 30: 4687-4692, 2010.

Goldman SA, and Nottebohm F. Neuronal production, migration, and differentiation in a vocal control nucleus of the adult female canary brain. Proc Natl Acad Sci U S A 80: 2390-2394, 1983.

Golowasch J, Buchholtz F, Epstein IR, and Marder E. Contribution of individual ionic currents to activity of a model stomatogastric ganglion neuron. J Neurophysiol 67: 341-349, 1992.

Golowasch J, and Marder E. Ionic currents of the lateral pyloric neuron of the stomatogastric ganglion of the crab. J Neurophysiol 67: 318-331, 1992.

Gould E, and Gross CG. Neurogenesis in adult mammals: some progress and problems. J Neurosci 22: 619-623, 2002.

Grace AA, and Bunney BS. Intracellular and extracellular electrophysiology of nigral dopaminergic neurons. Action potential generating mechanisms and morphological correlates. Neuroscience 10: 317-331, 1983.

179

Graybiel AM, Aosaki T, Flaherty AW, and Kimura M. The basal ganglia and adaptive motor control. Science 265: 1826-1831, 1994.

Hahnloser RH, Kozhevnikov AA, and Fee MS. An ultra-sparse code underlies the generation of neural sequences in a songbird. Nature 419: 65-70, 2002.

Hahnloser RHR. Sleep-related neural activity in a premotor and a basal-ganglia pathway of the songbird. J Neurophysiol 96: 794-812, 2006.

Hahnloser RHR, and Fee MS. Sleep-related spike bursts in HVC are driven by the nucleus interface of the nidopallium. J Neurophysiol 97: 423-435, 2007.

Hauser MD, and Konishi M. The Design of Animal Communication. Cambridge, MA: MIT Press, 1999.

Hausser M, Stuart G, Racca C, and Sakmann B. Axonal initiation and active dendritic propagation of action potentials in substantia nigra neurons. Neuron 15: 637-647, 1995.

Hessler NA, and Doupe AJ. Singing-related neural activity in a dorsal forebrain-basal ganglia circuit of adult zebra finches. J Neurosci 19: 10461-10481, 1999.

Hille B. Ion Channels of Excitable Membranes. Sunderland, MA: Sinauer, 2001.

Hodgkin AL, and Huxley AF. A quantitative description of membrane current and its application to conduction and excitation in nerve. J Physiol 117: 500-544, 1952.

Huguenard JR. Low-threshold calcium currents in central nervous system neurons. Annu Rev Physiol 58: 329-348, 1996.

Hurford JR. The Origins of Grammar: Language in the Light of Evolution. Oxford: Oxford University Press, 2012.

Imada T, Zhang Y, Cheour M, Taulu S, Ahonen A, and Kuhl PK. Infant speech perception activates Broca's area: a developmental magnetoencephalography study. Neuroreport 17: 957- 962, 2006.

Immelmann K. Song development in the zebra fi nch and other estrildid finches. Inμ Bird Vocalizations, edited by Hinde RA. New York: Cambridge University Press, 1969, p. 61-74.

180

Jaffe DB, Johnston D, Lasser-Ross N, Lisman JE, Miyakawa H, and Ross WN. The spread of Na+ spikes determines the pattern of dendritic Ca2+ entry into hippocampal neurons. Nature 357: 244-246, 1992.

Janik VM, and Slater PJB. Vocal learning in mammals. In: Advances in the study of behavior1997, p. 59-99.

Jarvis ED. Learned birdsong and the neurobiology of human language. Ann N Y Acad Sci 1016: 749-777, 2004.

Jarvis ED, Gunturkun O, Bruce L, Csillag A, Karten H, Kuenzel W, Medina L, Paxinos G, Perkel DJ, Shimizu T, Striedter G, Wild JM, Ball GF, Dugas-Ford J, Durand SE, Hough GE, Husband S, Kubikova L, Lee DW, Mello CV, Powers A, Siang C, Smulders TV, Wada K, White SA, Yamamoto K, Yu J, Reiner A, and Butler AB. Avian brains and a new understanding of vertebrate brain evolution. Nat Rev Neurosci 6: 151-159, 2005.

Jenerick H. Phase plane trajectories of the muscle spike potential. Biophys J 3: 363-377, 1963.

Jin DZ. Generating variable birdsong syllable sequences with branching chain networks in avian premotor nucleus HVC. Phys Rev E 80: 051902, 2009.

Jin DZ, Ramazanoglu FM, and Seung HS. Intrinsic bursting enhances the robustness of a neural network model of sequence generation by avian brain area HVC. J Comput Neurosci 23: 283- 299, 2007.

Johnson F, and Sellix M. Reorganization of a telencephalic motor region during sexual differentiation and vocal learning in zebra finches. Brain Res Dev Brain Res 121: 253-263, 2000.

Jurgens U. Neural pathways underlying vocal control. Neurosci Biobehav Rev 26: 235-258, 2002.

Karten HJ. The ascending auditory pathway in the pigeon (Columba livia). II. Telencephalic projections of the nucleus ovoidalis thalami. Brain Res 11: 134-153, 1968.

Karten HJ. Homology and evolutionary origins of the 'neocortex'. Brain Behav Evol 38: 264-272, 1991.

Katahira K, Okanoya K, and Okada M. A neural network model for generating complex birdsong syntax. Biol Cybern 97: 441-448, 2007.

181

Katz LC, and Gurney ME. Auditory responses in the zebra finch's motor system for song. Brain Res 221: 192-197, 1981.

Kawaguchi Y. Physiological subgroups of nonpyramidal cells with specific morphological characteristics in layer II/III of rat frontal cortex. J Neurosci 15: 2638-2655, 1995.

Kelley DB, and Nottebohm F. Projections of a telencephalic auditory nucleus-field L-in the canary. J Comp Neurol 183: 455-469, 1979.

Khaliq ZM, and Raman IM. Relative contributions of axonal and somatic Na channels to action potential initiation in cerebellar Purkinje neurons. J Neurosci 26: 1935-1944, 2006.

Kirn JR, Alvarez-Buylla A, and Nottebohm F. Production and survival of projection neurons in a forebrain vocal center of adult male canaries. J Neurosci 11: 1756-1762, 1991.

Knudsen DP, and Gentner TQ. Mechanisms of song perception in oscine birds. Brain Lang 115: 59-68, 2010.

Konishi M. Birdsong: From Behavior to Neuron. Ann Rev Neurosci 8: 125-170, 1985.

Konishi M. From central pattern generator to sensory template in the evolution of birdsong. Brain Lang 115: 18-20, 2010.

Konishi M. The role of auditory feedback in the control of vocalization in the white-crowned sparrow. Z Tierpsychol 22: 770-783, 1965.

Kozhevnikov AA, and Fee MS. Singing-related activity of identified HVC neurons in the zebra finch. J Neurophysiol 97: 4271-4283, 2007.

Kroodsma D, and Konishi M. A suboscine bird (eastern phoebe, Sayornis phoebe) develops normal song without auditory feedback. Animal Behaviour 42: 477–487, 1991.

Kroodsma D, and Pickert R. Environmentally dependent sensitive periods for avian vocal learning. Nature 288: 477-479, 1980.

Kubikova L, Turner EA, and Jarvis ED. The pallial basal ganglia pathway modulates the behaviorally driven gene expression of the motor pathway. Eur J Neurosci 25: 2145-2160, 2007.

Kubota M, and Saito N. Sodium- and calcium-dependent conductances of neurones in the zebra finch hyperstriatum ventrale pars caudale in vitro. J Physiol 440: 131-142, 1991.

182

Kubota M, and Taniguchi I. Electrophysiological characteristics of classes of neuron in the HVC of the zebra finch. J Neurophysiol 80: 914-923, 1998.

Lacinova L. Pharmacology of recombinant low-voltage activated calcium channels. Curr Drug Targets CNS Neurol Disord 3: 105-111, 2004.

Lacinova L. Voltage-dependent calcium channels. Gen Physiol Biophys 24 Suppl 1: 1-78, 2005.

Leonardo A. Experimental test of the birdsong error-correction model. Proc Natl Acad Sci U S A 101: 16935-16940, 2004.

Leonardo A, and Fee MS. Ensemble coding of vocal control in birdsong. J Neurosci 25: 652- 661, 2005.

Lewicki MS. Intracellular characterization of song-specific neurons in the zebra finch auditory forebrain. J Neurosci 16: 5855-5863, 1996.

Lewicki MS, and Konishi M. Mechanisms underlying the sensitivity of songbird forebrain neurons to temporal order. Proc Natl Acad Sci U S A 92: 5582-5586, 1995.

Li M, and Greenside H. Stable propagation of a burst through a one-dimensional homogeneous excitatory chain model of songbird nucleus HVC. Phys Rev E Stat Nonlin Soft Matter Phys 74: 011918, 2006.

Li YX, Bertram R, and Rinzel J. Modeling N-methyl-D-aspartate-induced bursting in dopamine neurons. Neuroscience 71: 397-410, 1996.

Liu WC, Gardner TJ, and Nottebohm F. Juvenile zebra finches can use multiple strategies to learn the same song. Proc Natl Acad Sci U S A 101: 18177-18182, 2004.

Liu WC, Wada K, and Nottebohm F. Variable food begging calls are harbingers of vocal learning. PLoS One 4: e5929, 2009.

Llinas R, and Jahnsen H. Electrophysiology of mammalian thalamic neurones in vitro. Nature 297: 406-408, 1982.

Llinas R, and Yarom Y. Electrophysiology of mammalian inferior olivary neurones in vitro. Different types of voltage-dependent ionic conductances. J Physiol 315: 549-567, 1981a.

183

Llinas R, and Yarom Y. Properties and distribution of ionic conductances generating electroresponsiveness of mammalian inferior olivary neurones in vitro. J Physiol 315: 569-584, 1981b.

Locke JL. The child's path to spoken language. Cambridge: Harvard University Press, 1993.

London SE, and Clayton DF. Functional identification of sensory mechanisms required for developmental song learning. Nat Neurosci 11: 579-586, 2008.

Long MA, and Fee MS. Using temperature to analyse temporal dynamics in the songbird motor pathway. Nature 456: 189-194, 2008.

Long MA, Jin DZ, and Fee MS. Support for a synaptic chain model of neuronal sequence generation. Nature 468: 394-399, 2010.

Luo M, Ding L, and Perkel DJ. An avian basal ganglia pathway essential for vocal learning forms a closed topographic loop. J Neurosci 21: 6836-6845, 2001.

Luo M, and Perkel DJ. Long-range GABAergic projection in a circuit essential for vocal learning. J Comp Neurol 403: 68-84, 1999.

Margoliash D. Acoustic parameters underlying the responses of song-specific neurons in the white-crowned sparrow. J Neurosci 3: 1039-1057, 1983.

Margoliash D. Preference for autogenous song by auditory neurons in a song system nucleus of the white-crowned sparrow. J Neurosci 6: 1643-1661, 1986.

Margoliash D. Sleep, learning, and birdsong. ILAR J 51: 378-386, 2010.

Margoliash D, and Fortune ES. Temporal and harmonic combination-sensitive neurons in the zebra finch's HVc. J Neurosci 12: 4309-4326, 1992.

Margoliash D, and Konishi M. Auditory representation of autogenous song in the song system of white-crowned sparrows. Proc Natl Acad Sci U S A 82: 5997-6000, 1985.

Margoliash D, and Schmidt MF. Sleep, off-line processing, and vocal learning. Brain Lang 115: 45-58, 2010.

Marler P. Birdsong and speech development: could there be parallels? Am Sci 58: 669-673, 1970a.

184

Marler P. A comparative approach to vocal learning: Song development in white-crowned sparrows. Journal of Comparative and Physiological Psychology 71: 1-25, 1970b.

Marler P. Origins of music and speech: insights from animals. In: The Origins of Music. Cambridge: Massachusetts Institute of Technology, 2000, p. 31-50.

Marler P. Song-learning behavior: the interface with neuroethology. Trends Neurosci 14: 199- 206, 1991.

Marler P, Mundinger P, Waser MS, and Lutjen A. Effects of acoustical stimulation and deprivation on song development in red-winged blackbirds (Agelaius phoeniceus). Anim Behav 20: 586-606, 1972.

Marler P, and Sherman V. Song structure without auditory feedback: emendations of the auditory template hypothesis. J Neurosci 3: 517-531, 1983.

Marler P, and Tamura M. CULTURALLY TRANSMITTED PATTERNS OF VOCAL BEHAVIOR IN SPARROWS. Science 146: 1483-1486, 1964.

Martina M, Schultz JH, Ehmke H, Monyer H, and Jonas P. Functional and molecular differences between voltage-gated K+ channels of fast-spiking interneurons and pyramidal neurons of rat hippocampus. J Neurosci 18: 8111-8125, 1998.

Massengill JL, Smith MA, Son DI, and O'Dowd DK. Differential expression of K4-AP currents and Kv3.1 potassium channel transcripts in cortical neurons that develop distinct firing phenotypes. J Neurosci 17: 3136-3147, 1997.

McCasland JS. Neuronal control of bird song production. J Neurosci 7: 23-39, 1987.

McCasland JS, and Konishi M. Interaction between auditory and motor activities in an avian song control nucleus. Proc Natl Acad Sci U S A 78: 7815-7819, 1981.

McCormick DA, Connors BW, Lighthall JW, and Prince DA. Comparative electrophysiology of pyramidal and sparsely spiny stellate neurons of the neocortex. J Neurophysiol 54: 782-806, 1985.

McCormick DA, and Pape HC. Properties of a hyperpolarization-activated cation current and its role in rhythmic oscillation in thalamic relay neurones. J Physiol 431: 291-318, 1990.

185

Mello CV. Gene regulation by song in the auditory telencephalon of songbirds. Front Biosci 9: 63-73, 2004.

Mello CV, Velho TA, and Pinaud R. Song-induced gene expression: a window on song auditory processing and perception. Ann N Y Acad Sci 1016: 263-281, 2004.

Mooney R. Different subthreshold mechanisms underlie song selectivity in identified HVc neurons of the zebra finch. J Neurosci 20: 5420-5436, 2000.

Mooney R. Neural mechanisms for learned birdsong. Learn Mem 16: 655-669, 2009a.

Mooney R. Neurobiology of song learning. Curr Opin Neurobiol 19: 654-660, 2009b.

Mooney R, Hoese W, and Nowicki S. Auditory representation of the vocal repertoire in a songbird with multiple song types. Proc Natl Acad Sci U S A 98: 12778-12783, 2001.

Mooney R, and Prather JF. The HVC microcircuit: the synaptic basis for interactions between song motor and vocal plasticity pathways. J Neurosci 25: 1952-1964, 2005.

Mooney R, and Spiro JE. Bird song: of tone and tempo in the telencephalon. Curr Biol 7: R289- 291, 1997.

Mulle C, Madariaga A, and Deschenes M. Morphology and electrophysiological properties of reticularis thalami neurons in cat: in vivo study of a thalamic pacemaker. J Neurosci 6: 2134- 2145, 1986.

Muller N, Putz A, Kathmann N, Lehle R, Gunther W, and Straube A. Characteristics of obsessive-compulsive symptoms in Tourette's syndrome, obsessive-compulsive disorder, and Parkinson's disease. Psychiatry Res 70: 105-114, 1997.

Munsch T, and Pape HC. Modulation of the hyperpolarization-activated cation current of rat thalamic relay neurones by intracellular pH. J Physiol 519 Pt 2: 493-504, 1999.

Nagel K, Kim G, McLendon H, and Doupe A. A bird brain's view of auditory processing and perception. Hear Res 273: 123-133, 2011.

Nordeen KW, and Nordeen EJ. Auditory feedback is necessary for the maintenance of stereotyped song in adult zebra finches. Behav Neural Biol 57: 58-66, 1992.

186

Nordeen KW, and Nordeen EJ. Long-term maintenance of song in adult zebra finches is not affected by lesions of a forebrain region involved in song learning. Behav Neural Biol 59: 79-82, 1993.

Nottebohm F. Neural lateralization of vocal control in a passerine bird. I. Song. J Exp Zool 177: 229-261, 1971.

Nottebohm F, Kelley DB, and Paton JA. Connections of vocal control nuclei in the canary telencephalon. J Comp Neurol 207: 344-357, 1982.

Nottebohm F, and Liu WC. The origins of vocal learning: New sounds, new circuits, new cells. Brain Lang 115: 3-17, 2010.

Nottebohm F, and Nottebohm ME. Left hypoglossal dominance in control of canady and white- crowned sparrow song. J Comp Physiol 108: 171-192, 1976.

Nottebohm F, Stokes TM, and Leonard CM. Central control of song in the canary, Serinus canarius. J Comp Neurol 165: 457-486, 1976.

Nowak LG, Azouz R, Sanchez-Vives MV, Gray CM, and McCormick DA. Electrophysiological classes of cat primary visual cortical neurons in vivo as revealed by quantitative analyses. J Neurophysiol 89: 1541-1566, 2003.

Nowicki S, Searcy WA, and Peters S. Brain development, song learning and mate choice in birds: a review and experimental test of the "nutritional stress hypothesis". J Comp Physiol A Neuroethol Sens Neural Behav Physiol 188: 1003-1014, 2002.

Okanoya K. The Bengalese finch: a window on the behavioral neurobiology of birdsong syntax. Ann N Y Acad Sci 1016: 724-735, 2004.

Okuhata S, and Saito N. Synaptic connections of thalamo-cerebral vocal nuclei of the canary. Brain Res Bull 18: 35-44, 1987.

Olveczky BP, Andalman AS, and Fee MS. Vocal experimentation in the juvenile songbird requires a basal ganglia circuit. PLoS Biol 3: e153, 2005.

Olveczky BP, and Gardner TJ. A bird's eye view of neural circuit formation. Curr Opin Neurobiol 21: 124-131, 2011.

187

Palmer LM, and Stuart GJ. Site of action potential initiation in layer 5 pyramidal neurons. J Neurosci 26: 1854-1863, 2006.

Pape HC. Queer current and pacemaker: the hyperpolarization-activated cation current in neurons. Annu Rev Physiol 58: 299-327, 1996.

Payne RS, and McVay S. Songs of humpback whales. Science 173: 585-597, 1971.

Person AL, Gale SD, Farries MA, and Perkel DJ. Organization of the songbird basal ganglia, including area X. The Journal of Comparative Neurology 508: 840-866, 2008.

Petkov CI, and Jarvis ED. Birds, primates, and spoken language origins: behavioral phenotypes and neurobiological substrates. Front Evol Neurosci 4: 12, 2012.

Phan ML, Pytte CL, and Vicario DS. Early auditory experience generates long-lasting memories that may subserve vocal learning in songbirds. Proc Natl Acad Sci U S A 103: 1088-1093, 2006.

Poole JH, Tyack PL, Stoeger-Horwath AS, and Watwood S. Animal behaviour: elephants are capable of vocal learning. Nature 434: 455-456, 2005.

Prather J, and Mooney R. Neural correlates of learned song in the avian forebrain: simultaneous representation of self and others. Current Opinion in Neurobiology 14: 496-502, 2004.

Prather JF, Nowicki S, Anderson RC, Peters S, and Mooney R. Neural correlates of categorical perception in learned vocal communication. Nat Neurosci 12: 221-228, 2009.

Price PH. Developmental determinants of structure in zebra finch song. J Comp & Physio Psych 93: 260-277, 1979.

Ralls K, Fiorelli P, and Gish S. Vocalizations and vocal mimicry in captive harbor seals, Phoca vitulina. Canadian Journal of Zoology 63: 1050-1056, 1985.

Ravbar P, Lipkind D, Parra LC, and Tchernichovski O. Vocal exploration is locally regulated during song learning. J Neurosci 32: 3422-3432, 2012.

Reiner A, Medina L, and Veenman CL. Structural and functional evolution of the basal ganglia in vertebrates. Brain Res Brain Res Rev 28: 235-285, 1998.

188

Reiner A, Perkel DJ, Bruce LL, Butler AB, Csillag A, Kuenzel W, Medina L, Paxinos G, Shimizu T, Striedter G, Wild M, Ball GF, Durand S, Gütürkün O, Lee DW, Mello CV, Powers A, White SA, Hough G, Kubikova L, Smulders TV, Wada K, Dugas-Ford J, Husband S, Yamamoto K, Yu J, Siang C, and Jarvis ED. Revised nomenclature for avian telencephalon and some related brainstem nuclei. The Journal of Comparative Neurology 473: 377-414, 2004.

Reiss D, and McCowan B. Spontaneous vocal mimicry and production by bottlenose dolphins (Tursiops truncatus): evidence for vocal learning. J Comp Psychol 107: 301-312, 1993.

Richards DG, Wolz JP, and Herman LM. Vocal mimicry of computer-generated sounds and vocal labeling of objects by a bottlenosed dolphin, Tursiops truncatus. J Comp Psychol 98: 10- 28, 1984.

Rinberg A, Taylor AL, and Marder E. The effects of temperature on the stability of a neuronal oscillator. PLoS Comput Biol 9: e1002857, 2013.

Roberts TF, Gobes SM, Murugan M, Olveczky BP, and Mooney R. Motor circuits are required to encode a sensory model for imitative learning. Nat Neurosci 15: 1454-1459, 2012.

Roper A, and Zann R. The Onset of Song Learning and Song Tutor Selection in Fledgling Zebra Finches Ethology 112: 458-470, 2006.

Rose CR, Kovalchuk Y, Eilers J, and Konnerth A. Two-photon Na+ imaging in spines and fine dendrites of central neurons. Pflugers Arch 439: 201-207, 1999.

Rose CR, and Ransom BR. Regulation of intracellular sodium in cultured rat hippocampal neurones. J Physiol 499 ( Pt 3): 573-587, 1997.

Rosen MJ, and Mooney R. Inhibitory and excitatory mechanisms underlying auditory responses to learned vocalizations in the songbird nucleus HVC. Neuron 39: 177-194, 2003.

Rudy B, and McBain CJ. Kv3 channels: voltage-gated K+ channels designed for high-frequency repetitive firing. Trends Neurosci 24: 517-526, 2001.

Sasahara K, and Ikegami T. Evolution of birdsong syntax by interjection communication. Artif Life 13: 259-277, 2007.

Satterlie RA. Reciprocal inhibition and postinhibitory rebound produce reverberation in a locomotor pattern generator. Science 229: 402-404, 1985.

189

Scharff C, Kirn JR, Grossman M, Macklis JD, and Nottebohm F. Targeted neuronal death affects neuronal replacement and vocal behavior in adult songbirds. Neuron 25: 481-492, 2000.

Scharff C, and Nottebohm F. A comparative study of the behavioral deficits following lesions of various parts of the zebra finch song system: implications for vocal learning. J Neurosci 11: 2896-2913, 1991.

Schmidt MF, and Ashmore RC. Integrating breathing and singing: Forebrain and brainstem mechanisms. In: Neuroscience of birdsong, edited by Zeigler HP, and Marler P. Cambridge: Cambridge Univ. Press, 2008, p. 285–322.

Schmidt MF, and Perkel DJ. Slow synaptic inhibition in nucleus HVC of the adult zebra finch. J Neurosci 18: 895-904, 1998.

Searcy WA, and Andersson M. Sexual selection and the evolution of birds. Ann Rev Ecol Syst 17: 507-533: 1986.

Shea SD, Koch H, Baleckaitis D, Ramirez JM, and Margoliash D. Neuron-specific cholinergic modulation of a forebrain song control nucleus. J Neurophysiol 103: 733-745, 2010.

Shu Y, Duque A, Yu Y, Haider B, and McCormick DA. Properties of action-potential initiation in neocortical pyramidal cells: evidence from whole cell axon recordings. J Neurophysiol 97: 746-760, 2007.

Simonyan K, Horwitz B, and Jarvis ED. Dopamine regulation of human speech and bird song: a critical review. Brain Lang 122: 142-150, 2012.

Snow CE, and Hohle MH. The critical period for language acquisition: Evidence from second language learning. Child Development 49: 1978.

Sohrabji F, Nordeen EJ, and Nordeen KW. Selective impairment of song learning following lesions of a forebrain nucleus in the juvenile zebra finch. Behav Neural Biol 53: 51-63, 1990.

Solis MM, and Perkel D. Rhythmic Activity in a Forebrain Vocal Control Nucleus In Vitro. Journal of Neuroscience 25: 2811-2822, 2005.

Solomon JS, Doyle JF, Burkhalter A, and Nerbonne JM. Differential expression of hyperpolarization-activated currents reveals distinct classes of visual cortical projection neurons. J Neurosci 13: 5082-5091, 1993.

190

Stauffer TR, Elliott KC, Ross MT, Basista MJ, Hyson RL, and Johnson F. Axial organization of a brain region that sequences a learned pattern of behavior. J Neurosci 32: 9312-9322, 2012.

Stuart G, Schiller J, and Sakmann B. Action potential initiation and propagation in rat neocortical pyramidal neurons. J Physiol 505 ( Pt 3): 617-632, 1997.

Suthers RA, Goller F, and Pytte C. The neuromuscular control of birdsong. Philos Trans R Soc Lond B Biol Sci 354: 927-939, 1999.

Tallerman M. What is syntax? . In: The Oxford Handbook of Language Evolution 2011, p. 442- 455.

Tang LS, Goeritz ML, Caplan JS, Taylor AL, Fisek M, and Marder E. Precise temperature compensation of phase in a rhythmic motor pattern. PLoS Biol 8: 2010.

Tateno T, Harsch A, and Robinson HP. Threshold firing frequency-current relationships of neurons in rat somatosensory cortex: type 1 and type 2 dynamics. J Neurophysiol 92: 2283-2294, 2004.

Tchernichovski O. Sound Analysis Pro http://ofer.sci.ccny.cuny.edu/sound_analysis_pro.

Tchernichovski O, Lints TJ, Deregnaucourt S, Cimenser A, and Mitra PP. Studying the song development process: rationale and methods. Ann N Y Acad Sci 1016: 348-363, 2004.

Tchernichovski O, and Mitra PP. Towards quantification of vocal imitation in the zebra finch. J Comp Physiol A Neuroethol Sens Neural Behav Physiol 188: 867-878, 2002.

Tchernichovski O, Nottebohm F, Ho CE, Pesaran B, and Mitra PP. A procedure for an automated measurement of song similarity. Anim Behav 59: 1167-1176, 2000.

Terman D, Rubin JE, Yew AC, and Wilson CJ. Activity patterns in a model for the subthalamopallidal network of the basal ganglia. J Neurosci 22: 2963-2976, 2002.

Theunissen FE, and Shaevitz SS. Auditory processing of vocal sounds in birds. Curr Opin Neurobiol 16: 400-407, 2006.

Thoby-Brisson M, Telgkamp P, and Ramirez JM. The role of the hyperpolarization-activated current in modulating rhythmic activity in the isolated respiratory network of mice. J Neurosci 20: 2994-3005, 2000.

191

Thompson JA, Basista MJ, Wu W, Bertram R, and Johnson F. Dual pre-motor contribution to songbird syllable variation. J Neurosci 31: 322-330, 2011.

Thompson JA, Wu W, Bertram R, and Johnson F. Auditory-dependent vocal recovery in adult male zebra finches is facilitated by lesion of a forebrain pathway that includes the basal ganglia. J Neurosci 27: 12308-12320, 2007.

Thorpe WH. Bird-Song. Cambridge: Cambridge Univ. Press, 1961.

Thorpe WH. The Learning of Song Patterns by Birds, with Special Reference to the Song of the Chaffinch. Ibis 100: 535-570, 1958.

Thorpe WH. The Process of Song-Learning in the Chaffinch as Studied by Means of the Sound Spectrograph. Nature 173: 465 - 469, 1954.

Toth BA, Kostuk M, Meliza CD, Margoliash D, and Abarbanel HD. Dynamical estimation of neuron and network properties I: variational methods. Biol Cybern 105: 217-237, 2011.

Troyer TW, and Doupe AJ. An associational model of birdsong sensorimotor learning I. Efference copy and the learning of song syllables. J Neurophysiol 84: 1204-1223, 2000a.

Troyer TW, and Doupe AJ. An associational model of birdsong sensorimotor learning II. Temporal hierarchies and the learning of song sequence. J Neurophysiol 84: 1224-1239, 2000b.

Tyack P. Whistle repertoires of two bottlenose dolphins, Tursiops truncatus: Mimicry of signature whistles? Behavioral Ecology and Sociobiology 18: 251-257, 1986.

Vates GE, Broome BM, Mello CV, and Nottebohm F. Auditory pathways of caudal telencephalon and their relation to the song system of adult male zebra finches. J Comp Neurol 366: 613-642, 1996.

Vates GE, Vicario DS, and Nottebohm F. Reafferent thalamo- "cortical" loops in the song system of oscine songbirds. J Comp Neurol 380: 275-290, 1997.

Vicario DS. Organization of the zebra finch song control system: II. Functional organization of outputs from nucleus robustus archistriatalis. J Comp Neurol 309: 486-494, 1991.

Vicario DS, and Nottebohm F. Organization of the zebra finch song control system: I. Representation of syringeal muscles in the hypoglossal nucleus. J Comp Neurol 271: 346-354, 1988.

192

Vicario DS, and Yohay KH. Song-selective auditory input to a forebrain vocal control nucleus in the zebra finch. J Neurobiol 24: 488-505, 1993.

Vu ET, Mazurek ME, and Kuo YC. Identification of a forebrain motor programming network for the learned song of zebra finches. J Neurosci 14: 6924-6934, 1994.

Vu ET, Schmidt MF, and Mazurek ME. Interhemispheric coordination of premotor neural activity during singing in adult zebra finches. J Neurosci 18: 9088-9098, 1998.

Waldstein RS. Effects of postlingual deafness on speech production: implications for the role of auditory feedback. J Acoust Soc Am 88: 2099-2114, 1990.

Wang XJ, Liu Y, Sanchez-Vives MV, and McCormick DA. Adaptation and temporal decorrelation by single neurons in the primary visual cortex. J Neurophysiol 89: 3279-3293, 2003.

Werker JF, and Tees RC. The organization and reorganization of human speech perception. Annu Rev Neurosci 15: 377-402, 1992.

White DJ. A social ethological perspective applied to care of and research on songbirds. ILAR J 51: 387-393, 2010.

Wilcox KS, Gutnick MJ, and Christoph GR. Electrophysiological properties of neurons in the lateral habenula nucleus: an in vitro study. J Neurophysiol 59: 212-225, 1988.

Wild JM. Descending projections of the songbird nucleus robustus archistriatalis. J Comp Neurol 338: 225-241, 1993.

Wild JM. Functional anatomy of neural pathways contributing to the control of song production in birds. Eur J Morphol 35: 303-325, 1997a.

Wild JM. Functional neuroanatomy of the sensorimotor control of singing. Ann N Y Acad Sci 1016: 438-462, 2004.

Wild JM. Neural pathways for the control of birdsong production. J Neurobiol 33: 653-670, 1997b.

Wild JM, and Arends JJ. A respiratory-vocal pathway in the brainstem of the pigeon. Brain Res 407: 191-194, 1987.

193

Wild JM, Karten HJ, and Frost BJ. Connections of the auditory forebrain in the pigeon (Columba livia). J Comp Neurol 337: 32-62, 1993.

Wild JM, Williams MN, Howie GJ, and Mooney R. Calcium-binding proteins define interneurons in HVC of the zebra finch (Taeniopygia guttata). J Comp Neurol 483: 76-90, 2005.

Williams H. Birdsong and singing behavior. Ann N Y Acad Sci 1016: 1-30, 2004.

Williams H, Crane LA, Hale TK, Esposito MA, and Nottebohm F. Right-side dominance for song control in the zebra finch. J Neurobiol 23: 1006-1020, 1992.

Williams H, and Vicario DS. Temporal patterning of song production: participation of nucleus uvaeformis of the thalamus. J Neurobiol 24: 903-912, 1993.

Wohlgemuth MJ, Sober SJ, and Brainard MS. Linked control of syllable sequence and phonology in birdsong. J Neurosci 30: 12936-12949, 2010.

Woolley SM. Early experience shapes vocal neural coding and perception in songbirds. Dev Psychobiol 54: 612-631, 2012.

Wu W, Thompson JA, Bertram R, and Johnson F. A statistical method for quantifying songbird phonology and syntax. J Neurosci Methods 174: 147-154, 2008.

Yu AC, and Margoliash D. Temporal hierarchical control of singing in birds. Science 273: 1871- 1875, 1996.

Yurk H, Barrett-Lennard L, Ford JKB, and Matkin CO. Cultural transmission within maternal lineages: vocal clans in resident killer whales in southern Alaska. Animal Behaviour 63: 1103- 1119, 2002.

Zaretsky MD, and Konishi M. Tonotopic organization in the avian telencephalon. Brain Res 111: 167-171, 1976.

Zhang X, Daou A, Truong TM, Bertram R, and Roper MG. Synchronization of mouse islets of Langerhans by glucose waveforms. Am J Physiol Endocrinol Metab 301: E742-747, 2011.

Zhou FM, and Hablitz JJ. Layer I neurons of rat neocortex. I. Action potential and repetitive firing properties. J Neurophysiol 76: 651-667, 1996.

194

BIOGRAPHICAL SKETCH

EDUCATION Florida State University BioMathematics, P.H.D. 2013, May Advisor: Richard Bertram Florida State University BioMathematics, M.S. 2010, July Advisor: Richard Bertram American University of Beirut Computer Science and Mathematics, B.S. (with honors) 2008, July

PUBLICATIONS 1. Arij Daou, Matthew Ross, Frank Johnson, Richard Hyson, Richard Bertram (2013), Electrophysiological Characterization and Computational Models of HVC Neurons of the Zebra Finch, Journal of Neurophysiology (to appear). 2. Arij Daou, Frank Johnson, Wei Wu, Richard Bertram (2012), A Computational Tool for Automated Large-Scale Analysis and Measurement of Birdsong Syntax, Journal of Neuroscience Methods, vol. 210, pp. 147-160. 3. Xinyu Zhang, Arij Daou, Tuan Truong, Richard Bertram, Michael Roper (2011), Synchronization of Mouse Islets of Langerhans by Glucose Waveforms, American Journal of Physiology, 301: E742-E747

AWARDS 1. Distinguished Teaching Assistant (FSU Mathematics Department, 2013) 2. Honorable Student Award (American University of Beirut, 2006) 3. Achievement Award (American University of Beirut, 2005)

195

SCIENTIFIC PRESENTATIONS 1. Arij Daou, Matthew Ross, Frank Johnson, Richard Hyson, Richard Bertram (2012), Electrophysiological Characterization and Computational Models of HVC Neurons of the Zebra Finch, Society for Neuroscience Meeting, New Orleans, USA. 2. Kevin Elliott, Arij Daou, Wei Wu, Richard Hyson, Richard Bertram, Frank Johnson (2012), A Cortical Premotor Region That Generates Vocal Sequences Receives Spatially-Organized Afferent Innervation, Society for Neuroscience Meeting, New Orleans, USA. 3. Arij Daou, Frank Johnson, Wei Wu, Richard Bertram (2011), A Computational Tool for Automated Large-Scale Analysis and Measurement of Birdsong Syntax, Society for Neuroscience Meeting, Washington DC, USA. 4. Mark Basista, Arij Daou, Wei Wu, Richard Bertram, Frank Johnson (2011), Spatially Organized Neural Activity Underlies a Temporally Organized Behavior, Society for Neuroscience Meeting, Washington DC, USA. 5. Arij Daou, Frank Johnson, Wei Wu, Richard Bertram (2010), Computational Model of Microcircuit Dynamics Underlying Birdsong in the Zebra Finch, Society for Neuroscience Meeting, San Diego, USA.

LAB EXPERIENCE - Performed in vitro whole-cell current and voltage clamp recordings on neurons within the HVC. This includes subject manipulation, brain slicing, electrophysiological recordings, pharmacological manipulations, intracellular staining and imaging, and electrophysiological data acquisition and analysis, 2011-2012. - Conducted various electrical lesioning experiments of different regions of the zebra finch brain (particularly HVC, Area X, and LMAN), 2009-2011. - Conducted dye injection experiments by pressure injecting retro- and anterograde neural tracers (DiI and DiO) into different subregions of the HVC, in an attempt to study the spatial and topographic organization of the HVC, 2011.

196

TEACHING EXPERIENCE - Instructor for Calculus I Spring, 2012 - Instructor for Pre-Calculus Fall 2011 & Summer 2011 - Instructor for Bio-Calculus Spring 2008, Fall 2008 & Spring 2009 Contributed to the lecture notes of BioCalc and wrote the “Logistic Equation” lesson - Course Assistant for various classes 2008-2011 Trigonometry, Business Calculus, College Algebra, Mathematical Logic

COMPUTER PROGRAMMING SKILLS - Expert in JAVA with 8+ years of programming experience and 4+ years of IT industry experience. Developed several software with GUI (Graphical User Interface), both application-based (Swing, AWT and SWT) and web-based (GWT, Java Servlets, and Javascript) using JDBC and RMI. The latest is presented in Daou et al., 2012. - Expert in MATLAB, XPP and SQL. - Experience in C++, C, Python, OpenGL, Adobe Photoshop and Adobe Illustrator.

PERSONAL AFFILIATIONS - Member, Society for Neuroscience, 2009 – present - Member, American Mathematical Society, 2008 – present - Member, Pi Mu Epsilon, 2008 – present - Graduate Co-Sponsor, Pi Mu Epsilon – Florida Beta Chapter, 2013 – present - Organizer, Mathematics Graduate Student Seminar (at FSU), 2009 - Organizer, Songbird Journal Club (at FSU), 2009 – present

INVITED PRESENTATIONS University of Chicago: Presentation to Dr. Daniel Margoliash lab (2012).

197

SEMINAR TALKS - Florida State University, Graduate Student Seminar Fall 2012, Fall 2011, Fall 2010, Summer 2009, Fall 2009 - Florida State University, Biomathematics Seminar Fall 2012, Spring 2011, Fall 2011, Spring 2010, Fall 2010, Fall 2009 - Florida State University, Genetics Seminar Spring 2009, Spring 2008 - Florida State University, Biomathematics Journal Club Every semester, 2008-present - Florida State University, Songbird Journal Club Every semester, 2011-present

REFERENCES - Dr. Richard Bertram Department of Mathematics & Program in Neuroscience Florida State University, Tallahassee, FL 32306 Telephone: 850-644-7195 Email: [email protected]

- Dr. Frank Johnson Department of Psychology & Program in Neuroscience Florida State University, Tallahassee, FL 32306 Telephone: 850-644-8566 Email: [email protected]

- Dr. Richard Hyson Department of Psychology & Program in Neuroscience Florida State University, Tallahassee, FL 32306 Telephone: 850-644-5824 Email: [email protected]

198