brains and birdsong - duke university

Chapter 8

Brains and birdsongERICH D. JARVIS

INTRODUCTION

The special brain structures for singing, andlearning how to sing, were discovered in 1976.Since then, with still growing momentum, therehas been a wealth of fascinating discoveries onthe structure, function, and evolution of thebrains of birds, especially those that engage invocal learning. Less than half of all birds possessthe ability to learn and reproduce new sounds.These vocal learners, parrots, hummingbirds, andsongbirds belong to only 3 of the 23 major birdgroups. They have the necessary forebrainanatomy for producing learned vocalizations. Allother birds use only basal brain structures forvocal production and their vocalizations areinnate, and genetically inherited from theirparents.

Of the birds that are vocal learners, most isknown about songbirds, especially the canaryand the zebra finch. These are subjects of choicebecause they breed easily in captivity and displayopposite extremes of vocal learning behavior. Inzebra finches, only the males sing, have vocallearning, and possess the appropriate forebrainstructures for the purpose; they are closed-endedvocal learners, developing one song motif asjuveniles and singing it for life. In canaries, bothmales and females sing, and both have vocallearning brain structures; they are open-endedlearners and, like humans, continue to learn newsongs as adults. An understanding of these

contrasts and their underlying principles holdspromise of new and fundamental insights intohow brains make learning possible. The discoveryof similar sets of brain structures in parrots andhummingbirds throws new light on how brainstructures for vocal learning have evolved. Theshared features imply that there are strongepigenetic constraints placed on the brain andbehavior during evolution, and suggest thatperhaps humans evolved somewhat similar brainstructures for speech and singing. Although thecognitive side of song learning hardly bearscomparison with human speech, ‘bird-brained’need no longer be a pejorative. If a parrot imitatesits own and other species’ sounds so readily anda child can hardly be prevented from learning tospeak, how is it that a chimpanzee does neither?The answer perhaps lies in the fact that no brainstructures have been found in chimpanzeesequivalent to those present in songbirds andhumans.

Other important findings about brains werefirst made in the songbird vocal learning system.These include the startling discovery of largesexual dimorphisms in the brain, the role ofhormones in regulating brain structure andlearning, and the ability of the adult songbirdbrain to generate new neurons, paving the wayfor current brainstem cell research. This chapterwill review some of these pioneering studies,and how the brain controls singing behavior,which is the theme of this book.

Brains and birdsong 227

GENERAL BRAIN ORGANIZATION

Understanding the science of birdsong ultimatelyrequires a basic understanding of the vertebratebrain. Birds and mammals, including humans,are vertebrate animals. Regardless of brain andbody size, all vertebrates share a general brainorganization that consists of five basic regions.These are the spinal cord within the vertebralbackbone, the hindbrain in front of it, then themidbrain, the thalamus, and the cerebrum, withthe cerebellum (meaning ‘small brain’) on topof the hindbrain (Fig. 8.1A). Among thosevertebrates that are amniotes, having embryosthat develop within amniotic fluid-filled sacs,that is reptiles, birds, and mammals, all have asimilar structural organization in these five brainregions, except for one, the cerebrum (Veenmanet al. 1995; Reiner et al. 1998). In reptiles andbirds the cerebrum is organized into large cellclusters; in mammals only the bottom part isorganized into clusters, called the basal ganglia,whereas the upper part is arranged in layers thattogether form the cortex. Because mammals havelong been regarded as more intelligent than birdsor reptiles, it was assumed that one of their majoranatomical differences, the mammalian cortex,was responsible for their more intelligent behavior(Herrick 1956). Many viewed evolution asproceeding linearly with its ultimate goal thecreation of ‘man,’ ‘his’ language, civilization, largebrain size, and folded cortex. Accordingly, almostall subdivisions of the cerebrum of birds andreptiles were given names that designate themammalian basal ganglia, which were thoughtto be primitive (Edinger et al. 1903; Edinger1908; Herrick 1956; Fig. 8.1B, oldnomenclature). We now know that theorganization of the bird cerebrum is much moresimilar to the mammalian cerebrum thanpreviously thought (Reiner et al. 1998 in pressA, B; Puelles et al. 1999). In 2002, at aNomenclature Forum held at Duke University,the subdivisions of the entire bird cerebrum wererenamed to reflect more accurately the manyhomologies that exist between avian andmammalian brains (Jarvis et al. 2002; Medina

et al. in press; Reiner et al. 2004a; Fig. 8.1B,new nomenclature).

The cerebrum of the bird and the mammalconsists of three major cell zones, the pallium,the striatum, and the pallidum. The striatiumand pallidum portions combined are the basalganglia, and in most animals they sit at the baseof the cerebrum (Marin et al. 1998). The palliumzone sits above the striatum at the top of thecerebrum, and includes the cortex in mammalsand four major brain subdivisions in birds (Reineret al. 1998; Puelles et al. 1999; Swanson 2000b).Brain size and folding are more related to animalsize than to behavioral complexity (Van Essen1997). The smaller the animal, the less foldingthere is of the cerebrum. Hummingbirds withone of the smallest bird brains and no foldinghave more advanced types of behavior, such asvocal learning, than many mammals with largeand folded brains, such as horses.

Neurons, the communicating cells of the brain,typically have a cell body with axons that makelong distance connections to other neurons (Fig.8.1C). Neurons also have dendrites that receivethese axonal connections. The meeting places ofthe axons and dendrites are the synapses. In thebasic connectivity plan of both birds andmammals, sensory information from the outsideworld, and from the inner body, destined forthe cerebrum, first passes through sensoryreceptor neurons that have one axon projectingout into the body or the exterior, and anotherinto the spinal cord for the lower body or intothe hindbrain for the upper body. Here theysynapse onto their appropriate cell groups, whichin turn project up to the midbrain. The midbrainneurons send projections into the thalamus. Thethalamic neurons process and project theinformation to the cerebrum, where they synapsewith a network of cells involving the pallium,striatum, and pallidum. Neurons that controlmovement form a network that traverses in theopposite direction, from the cerebrum, to themidbrain, and spinal cord. Neurons of the spinalcord, lower motor neurons, send their axons tomuscles. The cerebrum controls complex learningand voluntary behaviors, and ‘conscious’

228 Nature’s Music

(A) Side view SongbirdThe Brain

Human

Figure 8.1 (A) Songbird and human brains compared. (B) Old and new nomenclature for the bird brain.Dashed gray lines separate regions that differ by cell density and size, such as the six layers of the cortex.The darkest gray area of the human cerebrum are all axon pathways, typically called white matter,connecting the cortex with various cerebral and non-cerebral areas. Abbreviations: E, entopallial nucleus;B, basorostral nucleus; L2, field L2 nucleus; OB, olfactory bulb; HP, hippocampus; CDL, Corticoid dorsallateral (C) A diagram of neurons and their connections.

Dendrites:receptors

(B)

Oldnomenclature

The bird brain

Newnomenclature

Cerebellum

Thalamus

MidbrainHindbrain

Cerebellum

ThalamusMidbrain

Hindbrain

cerebrum = telencephalonpalliumstriatumpallidum

Cell body

Pre-synaptic neuron

e-action potentials

Dendrite spinesDendrites

Synapse:neurotransmitters-pre-synaptic

receptors-post-synaptic

(C) How neurons communicate

Nucleus

Cytoplasm

Cell bodyAxon

Post-synaptic neuron

The human brain

Cortexlateral

ventricle StriatumClaustrum

PallidumThalamus

Midbrain

Hindbrain

HyperpalliumMesopallium

Nidopallium

Iateralventricle

HpCDL

L2

E

StriatumPallidum

Arcopallium

Amygdaloid

Complex

BOB

Am

ygdala

OB


processing, whereas the midbrain is involved ininnate behaviors and ‘non-conscious’ processing.Glial cells typically provide physical support andnourishment to neurons.

THE SEARCH FOR VOCALLEARNING BRAIN AREAS

Vocalizations are learned in some species andinnate in others, and this behavioral contrast isexpected to reflect a brain difference. As humans,however, we often have difficulty inunderstanding its true basis because, in everydaylanguage, we do not distinguish auditory learningfrom vocal learning. Vocal learning is the abilityto imitate the sounds that you hear or improvise.It is a rare trait, found so far in only six animalgroups: three birds (parrots, hummingbirds, andsongbirds) and three mammals (humans, bats,and cetaceans, the whales and dolphins [Thorpe1961; Marler 1970b; Wiley 1971; Caldwell& Caldwell 1972; Nottebohm 1972; Kroodsma& Baylis 1982; Dooling et al. 1987a; Guinee& Payne 1988; Baptista & Schuchmann 1990;Rubsamen & Schafer 1990; Esser 1994; Gauntet al. 1994; McCowan & Reiss 1995]). Auditorylearning is the ability to form memories of thesounds that you hear, and to associate soundswith objects and living things in the environment.Auditory learning is much more widespread. Itis present in nearly all land and many aquaticvertebrates. An example helps in understandingthis distinction. A dog can learn the meaning ofthe sounds sit (in English), sientese (in Spanish),or osuwali (in Japanese). Dogs are not born withthis knowledge of human words. They acquireit through auditory learning. However, a dogcannot imitate the sounds sit, sientese, or osuwali.Humans, parrots, and some songbirds can. Thisis vocal learning, and though it depends uponauditory learning (Konishi 1965a), it is distinctfrom it. Aside from mimics, most vocal learnersonly imitate sounds of their own species.

The history of the search for vocal learningbrain areas in songbirds has its roots in the searchfor language brain areas in humans, beginning

nearly 200 years ago. Vocal learning is thebehavioral substrate for spoken language. In thefirst recognized breakthrough, in 1861, a Frenchphysician, Paul Broca, published a paper aboutan autopsy of one of his patients who had astroke (Broca 1861). The patient could only speakone word, ‘tic,’ and had extensive brain damage,with a central location for the lesion in the nowfamous Broca’s Area of the cortex. From thisand several subsequent patients, Broca concludedtwo things: (i) that language within the brainwas localized, not spread out and (ii) its locationwas on the left side of the frontal lobe. Forty-four years later in 1905, Oswald Kalischer, aGerman scientist, attempted to determine ifparrots have a Broca’s-like Area (Kalischer 1905).He performed left and bilateral hemispherelesions, and stimulation experiments, on Amazonparrots. He found some vocal effects, but thiswas preliminary work and was never followedup. Studies on vocal non-learning animals,including mice and monkeys, revealed no cerebralvocal regions (Kuypers 1958b; Jürgens 1995,1998). Only midbrain vocal regions that controlinnate vocalizations were found. It was not until115 years after Broca’s discovery that FernandoNottebohm, and two of his colleagues, TegnerStokes and Christiana Leonard, in 1976 publisheda pioneering report, announcing the discoveryof cerebral vocal structures for learnedvocalizations in a non-human species, the canary,a songbird (Nottebohm et al. 1976).

Nottebohm began his neuroanatomicalinvestigations in the peripheral nervous systemoutside the spinal cord. He was studying theorganization and function of the songbird syrinxand its connection with the axon nerve bundlethat controls it, the tracheosyringeal nerve(Nottebohm 1971a, b), so-called because its axonsinnervate both the trachea and syrinx. Thesongbird syrinx is the main organ that produceslearned vocalizations (see Chapter 9). Nottebohmmade the interesting finding that after surgicaldisconnection of the left tracheosyringeal nervefrom the syrinx, canaries and other songbirdspecies had more difficulty in producing theirlearned songs than after disconnecting the right


nerve (Nottebohm 1971b; Nottebohm &Nottebohm 1976). He then reasoned that thistracheosyringeal nerve dominance might originatein the central nervous system, perhaps in thecerebrum, as is the case for human language.He and his colleagues made what turned out tobe a lucky guess and created surgical lesions inbrain regions next to the auditory pathway. Theyfigured that a vocal system might be next to anauditory system (Nottebohm et al. 1976). Theregion they focused on included what is nowcalled the auditory pallium. In their very firsttry, they found that the canary ceased to singnormally, particularly with lesions to one side ofthe brain.

This was the first demonstration of lateralizationin the cerebrum of a non-human species; likehumans, canaries were left-side dominant.Repeating their lesions in more birds, theynarrowed down the location that led to loss oflearned song to one region. The structure in thisregion was given the name HVC, for hyperstriatumventrale pars caudale, later renamed as the HighVocal Center. Lesions to the left canary HVCresulted in greater song deficits than to the rightHVC. They then labeled degenerating axons fromthe HVC lesions and determined that HVCprojected to another structure in the palliuminvolved in production of learned vocalizations,that is now called Robust nucleus of theArcopallium, RA, because of its robust appearancein stained tissue sections. HVC also projected toa structure in the striatum that they called AreaX, because at the time lesions there did not leadto noticeable song deficits (Nottebohm et al.1976). In stained brain sections, these vocal nucleican be seen under low magnification, andsometimes with the naked eye. It is remarkablethat they were not identified earlier in anatomicalinvestigations of the avian brain.

Soon thereafter, there was an explosion ofstudies on the cerebral vocal structures of thesongbird brain. Mark Konishi and his studentsperformed the first electrophysiologicalexperiments in the brain of awake songbirds (Katz& Gurney 1981; McCasland & Konishi 1981),in search of the brain regions that integrate

learned auditory information with learnedvocalizations. Nottebohm and Konishi alsointroduced into neurobiology the zebra finch, abird in which vocal learning had been describedin detail by Immelmann (1969) and his students.It was believed that study of the differencesbetween zebra finches and canaries would revealthe underlying brain mechanisms for continuedadult brain plasticity. Zebra finches pass throughthe juvenile phase of vocal learning relativelyfast, within 90 days after hatching, instead of afull year as in some other species, and can breedseveral times a year. Since then, zebra finchesand canaries have become the mainstay ofsongbird neurobiology research. In 1981, JohnPaton, Kirk Manogue, and Nottebohm usedelectrophysiological and neuronal connectivityapproaches to describe several similar brainstructures in a parrot (Paton et al. 1981). In2000, Claudio Mello and I used behavioral andmolecular approaches to reveal the entire set ofcerebral vocal structures in hummingbirds (Jarviset al. 2000). Because we had used the sameapproach with all three vocal learning groups,we were able for the first time to compare cerebralvocal structures in hummingbirds, parrots, andsongbirds. For studies in humans, the inventionof the new imaging techniques of PositronEmission Tomography (PET) and MagneticResonance Imaging (MRI) in the 1980s and1990s, and their use to study language-activatedbrain areas in humans, also propelled the brainand language field forward (Poeppel 1996; Binder1997). However, it is still not possible for scientiststo access human brains as readily as we can dowith birds.

Hundreds of papers later, a new research fieldhas emerged, focused on the neurobiology ofsongbird vocal communication, with over 98laboratories worldwide as of 2003, resulting inmany detailed analyses of the brain networkinvolved in the learned vocal communication ofsongbirds. Thus, over 140 years after Broca’sdiscoveries in humans and 26 years afterNottebohm’s discoveries in canaries, we now knowmore about brain pathways for vocal learning inbirds than we do in humans.


THE SONGBIRD VOCALCOMMUNICATION BRAINNETWORK

Networks of neuronal connectivity can bedetermined by a variety of methods. The twobest are based on neuronal tract-tracing andelectrical activity. For tract-tracing, a colored and/or fluorescent dye is surgically injected into thebrain region of interest; over several days thedye is transported through axons and/or dendritesto connecting regions; the brain is then dissectedfrom the animal, sectioned and examinedunderneath a microscope for presence of thetracer in the connecting regions. Alternatively,

electrical activity in two or more neurons can berecorded simultaneously and the circuitry isdeciphered mathematically based on the relativetiming of their firing. Tract-tracing is good atdefining the global circuitry, whereas electricalactivity is better at defining the microcircuitry.Both methods have been used in defining thebrain pathways responsible for songbird vocalcommunication.

The key network consists of three different,interconnected pathways, one posterior, oneanterior, and an auditory pathway. Of severalversions, the one I favor most is presented inFigure 8.2 (Jarvis in press). The posterior andanterior vocal pathways are often calledcollectively the ‘vocal control nuclei,’ ‘song control

Figure 8.2 (A) The anterior and posterior vocal pathways in the songbird brain. (B) There is one mainauditory pathway. Abbreviations not in the main text: CN, cochlear nucleus; LLD, lateral lemniscus, dorsalnucleus; LLI, lateral lemniscus, intermediate nucleus; LLV, lateral lemniscus, ventral nucleus; SO, superiorolivary nucleus; PAm, para-ambigualis nucleus; RAm, retroambigualis nucleus. Black arrows connect thenuclei of the posterior vocal pathway (light background). White arrows link nuclei of the anterior vocalpathway (dark background). Dashed arrows connect nuclei between the two pathways. Within the cerebrum:dark gray is pallium, medium gray is striatum, and light gray is pallidum. Connectivity was extrapolatedfrom the following studies: Nottebohm et al. 1976, 1982; Okuhata & Saito 1987; Bottjer et al. 1989, 2000;Wild 1994, 1997; Johnson et al. 1995; Nixdorf-Bergweiler et al. 1995; Vates & Nottebohm 1995; Livingston& Mooney 1997; Vates et al. 1997; Wild et al. 1997a, 2000; Iyengar et al. 1999; Luo & Perkel 1999a, b;Perkel & Farries 2000. Syrinx drawing from Suthers (1997).

Vocal and Auditory Pathways in the Songbird Brain

(A) Vocal pathways Cerebrum (B) Auditory pathway Cerebrum

Cerebellum

Uva

ShellThalamus

Midbrain

Hindbrain

ear haircells

cochlearganglion

Ram-Pam

nXma

Cerebellum

MidbrainHindbrain

nXma

Ram-Pam

respiratorymotor

neurons

muscles ofvocal

organs:trachea

&syrinx

Areax

Man

Mo

HyperpalliumMescopalliumNidopalliumNlf

CSf

Striatum

Pallidum

ArcopalliumRA

Cup

CLM/CMMAv

L1

L3L2NCMSh

elf

HVC

Ov

DLMDM

MLD

CN

SO

LLDLLI

LLV

HVC

AV

Uva

Thalamus

Areax

Man

Mo

HyperpalliumMescopalliumNidopalliumNlf

Striatum

Pallidum

ArcopalliumRA

Ov

DLMDM

MLD


nuclei,’ or the ‘song control system,’ nuclei herereferring to collections of neurons.

The posterior vocal pathway consists of fourcerebral nuclei all located within the back partof the pallium (Fig. 8.2A), HVC and RA, andtwo less commonly studied nuclei, the InterfacialNucleus, NIf, and Avalanche, Av. They areintegrated in a circuit with the final output fromRA to the midbrain vocal nucleus DM and to aset of hindbrain areas that includes respiratorynuclei RAm and PAm, and the tracheosyringealnucleus, the so-called 12th nucleus, nXIIts. Fromthere, axons of the respiratory nuclei makesynapses onto other motor neurons of the spinalcord, which in turn control muscles of the chestwall and air sacs involved in respiration. Axonsof the tracheosyringeal nucleus synapse onto themuscles of the trachea and syrinx. Thus, theposterior vocal pathway controls motor neurons,producing sounds and modulating breathingwhile doing so. In contrast to birds, mammalsuse the larynx to produce vocalizations (seeChapter 9). So far, no connections have beenfound in songbirds from the posterior vocalpathway onto motor neurons that control thetongue and beak movements although we expectto find connections there because song ismodulated by beak movements (see Chapter 10).In general, cerebral projections connectingdirectly to hindbrain and spinal cord motorneurons are a telltale sign of involvement inlearned movements.

The anterior vocal pathway consists of threecerebral nuclei located towards the front of thecerebrum, Area X, the Magnocellular nucleus ofthe Anterior Nidopallium, MAN, and the ovalnucleus of the Mesopallium, Mo, which is themost recently discovered (Jarvis et al. 1998),and one nucleus in the thalamus (Fig. 8.2A).Little is known about the function of Mo.Connections between the others form a loopfrom MAN to Area X to the dorsal part of thethalamus and back to MAN, which we canabbreviate as MAN → Area X → DorsalThalamus → MAN. This pathway is dividedinto two parallel parts, lateral (l) and medial(m) as lMAN → lArea X → DLM → lMAN

and mMAN → mArea X → ?DIP → mMAN.The posterior pathway sends input into theanterior pathway’s medial and lateral parts byway of a projection from HVC → all of Area X.The anterior pathway in turn sends output tothe posterior pathway by way of a projectionfrom lateral and medial MAN to different partsof the posterior pathway; lMAN → RA andmMAN → HVC (Fig. 8.2A; Box 28, p. 233).Thus, the anterior vocal pathway has little directinteraction with vocal motor neurons of thebrainstem (hindbrain and midbrain), but throughits interactions with the posterior pathway ispoised for other functions, including the learningof vocalizations.

The microcircuitry is important inunderstanding how these pathways operate. TheHVC has three major neuron types: (i) the RA-projecting neurons that send axons from HVC→ RA, (ii) the X-projecting neurons that sendaxons from HVC → Area X, and (iii) theinterneurons that make connections within HVC,between its RA- and X-projecting neurons. Thus,the X-projecting neurons are the input cell typeto the anterior pathway. The output cell type ofthe anterior pathway in MAN has two axonsgoing in different directions; for lMAN, oneaxon projects to lArea X, staying within theanterior pathway, and the other to RA, in theposterior pathway. Presumably this cell type inmMAN has one axon projecting to mArea Xand the other to HVC. In this manner, the lateralpart of the anterior pathway influences RA inthe posterior pathway and the medial part ofthe anterior pathway influences HVC in theposterior pathway.

The auditory pathway, processing sound as afirst step in song learning, begins at the haircells in the cochlea of the ear (Fig. 8.2B; seeChapter 7), activated by sound (Hudspeth 1997).They send axons to cochlear, CN, and leminiscal,LL, sensory nuclei in the hindbrain. These inturn project to the midbrain auditory nucleusMLd. MLd projects to Ovoidalis, Ov, in thethalamus. Ov projects to the primary auditorycells of the pallium, L2, then to secondary areas,L1 and L3, and then onto tertiary auditory areas,


BOX 28

THE SONG SYSTEM: A MINIHISTORY AND SOME CONNECTIVITY DETAILS

As is often the case in science, discoveries are made in bits and pieces. The order of discovery influencesthinking about how things work. In the song system, the pathways were named differently at first. The posteriorcircuit was considered a ‘direct pathway,’ by which HVC sends information to RA; the anterior circuit was calledan ‘indirect pathway,’ from HVC to RA (Okuhata & Saito 1987), terminology that was well established by the early1990s. The anterior pathway was also sometimes called the recursive loop (Williams 1989) or accessory loop(Doupe & Konishi 1991). The direct pathway was said to begin in HVC with neurons projecting to RA. The indirectpathway was also said to begin in HVC with neurons projecting to Area X; Area X to DLM, DLM to lMAN, andfinally lMAN back to RA. In this view, HVC belonged to both pathways. In the mid-1990s, a connection from lMANto lArea X was found, effectively forming a loop, (Nixdorf-Bergweiler et al. 1995; Vates & Nottebohm 1995), latershown to be a closed loop (Luo et al. 2001). The pathway then began to look more like the so-called cortical–basal ganglia–thalamic–cortical loops of the mammalian brain (Bottjer & Johnson 1997; Perkel & Farries 2000).In the late 1990s, the indirect pathway was renamed the anterior forebrain pathway (AFP), because of its location(Doupe 1993), and this has become common usage. In addition, connections between the medial part of thepathway were discovered and it was realized that they parallel the more commonly studied lateral part (Fosteret al. 1997; Jarvis et al. 1998). To complement the anterior circuit’s name, the direct pathway was renamed theposterior pathway (Jarvis et al. 1998). Finally, comparisons with other vocal learners and mammals (Jarvisunpublished) led to the view presented in this chapter, with the loop of the anterior pathway considered as thebasic pattern, also present in the brains of other vocal learners and in nonvocal pathways of mammals. I arguethat differences between species arise primarily in the connections between the anterior and posterior nuclei.

Some connectivity detailsThere are many complex connections within the vocal pathways of songbirds. RA projects to UVa in the thalamus,DM in the midbrain, the tracheosyringeal motor nucleus (nXIIts) in the brainstem, and to at least four premotornuclei in the brainstem that control breathing, abbreviated PBvl, IOS, RVL, rVRG, and RAm (Wild 1997). UVaprojects up to HVC, providing a direct route for feedback from RA to HVC (Striedter & Vu 1997). DM, like RA,projects to the tracheosyringeal motor nucleus and the same four breathing-related brainstem nuclei, providinga means for DM to modulate activity from RA onto these nuclei. It is this DM connectivity that coordinates thesyringeal and respiratory muscles during vocalizing in non learning species. In songbirds, it appears that RA hastaken over the innate vocalizing system by synapsing onto DM and all of its downstream targets (Vicario 1994).The respiratory nucleus RAm makes synapses onto motor neurons of the spinal cord which, in turn, controlmuscles of the chest wall and air sacs involved in breathing. In order to coordinate the two sides of the brain, UVaand DM send axons to their connecting counterparts on the other side of the brain.

Within the cerebrum, the HVC and lMAN neurons contact the dendrites of the same cells in RA (Mooney &Konishi 1991), but not always in the same location. This is how lMAN may modulate HVC activity into RA. In HVC,the interneurons contact both its X-projecting and RA-projecting neurons, coordinating their activity (Mooney2000). In the anterior vocal pathway, the connections form closed loops (Luo et al. 2001). This means that therewill be contact from a given neuron in lMAN to lArea X, and from there, to DLM, which will send its axon back tothe same lMAN neuron within that loop. Adjacent will be another neuronal closed loop parallel to it. An unusualfeature of the anterior vocal pathway, and perhaps of the avian brain generally, is that in Area X, in the striatum,there is a smaller number of pallidal-like neurons to which the striatal neurons are thought to make contact(Perkel & Farries 2000). If this is correct, HVC and lMAN project onto the striatal neurons in Area X which, in turn,converge onto the pallidal neurons in Area X and, from there, to the DLM of the thalamus. In the mammalianstriatum, the pallidal cells appear so far to be entirely separate from the striatum.

Erich Jarvis


Caudal Medial Nidopallium, NCM, CaudalMedial Mesopallium, CMM, the HVC shelf ofthe nidopallium and the RA cup of arcopallium,and a Caudal Medial part of the lateral Striatum,CMS. The auditory pathway also has adescending feedback projection with connectionssimilar to the vocal pathway, from the HVCshelf to the RA cup to the shells of thalamic andmidbrain auditory regions. One key feature ofthe auditory pathway is that, once projectionsfrom the auditory thalamus, Ov, reach thecerebrum’s primary auditory receiving cells, L2in the pallium, the L2 cells send projectionsthat spread out to many other pallial and onesubpallial auditory areas for further processing.Sensory input ascending into cerebral regions isa telltale sign of sensory learning.

It is an open question, where auditoryinformation enters the vocal pathways. It waslong thought that the main auditory input wasdirectly from L2 into HVC (Kelley & Nottebohm1979). However, this view was based on non-specific tract-tracing leakage into the adjacentNIf vocal nucleus that projects to HVC. Instead,L2 projects to an auditory region called the HVCshelf, which in turn sends only a few axons intoHVC (Fortune & Margoliash 1995; Vates et al.1996). Given that the HVC shelf projects to theRA cup, one might wonder if this is how auditoryinformation enters, but the cup also sends veryfew if any axons into RA (Mello et al. 1998).Two other less well-studied locations may bepotential sources of significant auditory input.Preliminary results show that auditory electricalactivity into HVC requires input from NIf, andnot from the adjacent field L2 region (Boco &Margoliash 2001). Input to NIf comes from Uvaof the thalamus (Box 28, p. 233). Anotherpotential source of auditory input is a regioncalled para-HVC, a thin layer of cells medial toHVC, on the surface of NCM, connected withNCM, and projecting into Area X of the anteriorvocal pathway (Foster & Bottjer 1998). Takentogether, the major source of auditory input intothe posterior vocal pathway may come directlyfrom brainstem auditory areas into Uva intovocal NIf; direct auditory information into the

anterior pathway may come from the NCM shellinto Area X. The final answers on the linkagesbetween the auditory and vocal networks awaitfurther investigation.

FUNCTIONS OF VOCAL PATHWAYSAND NUCLEI

Use of Lesions, ElectrophysiologicalRecordings, and Gene Activation

In total, the songbird vocal communicationnetwork consists of seven vocal cerebral nuclei,one thalamic nucleus, one midbrain nucleus,one hindbrain nucleus, and a comparable numberof auditory nuclei (Fig. 8.2). The layout suggeststhat different parts serve distinct behavioral andphysiological functions. To decipher thesefunctions, three basic approaches have been used:(i) the creation of lesions as performed insongbirds by Nottebohm and colleagues (1976),(ii) recording electrical activity as first performedin songbirds by Leppelsack (1978), Konishi andstudents (Katz & Gurney 1981; McCasland &Konishi 1981), and (iii) examining the molecularbiology of the circuitry as first performed byClayton, Mello, myself, and colleagues (Melloet al. 1992; Jarvis et al. 1995; Clayton 1997).To understand how the vocal pathways function,it is useful to review the underlying logic ofthese three approaches, all widely used in whatis collectively called neuroethology, a terminvented by Jerram Brown (Brown & Hunsperger1963), studying the neural basis of mammalianvocalizations.

1. With lesions, results are interpreted in termsof loss of function. If a brain region is destroyedand a particular aspect of behavior is affected,then the interpretation is that the brain regionis responsible for that aspect of the behavior.

2. With electrical activity, current changesacross neuron membranes are measured. Actionpotentials are large currents that travel downaxons to communicate with connecting neurons.There are also sub-threshold potentials; smalllocal current changes on the membranes that


communicate within and between neurons. Tomeasure activity, metal or glass electrodes areinserted into the brain regions or cells of interest.The electrical potentials picked up are sent bywire to detection devices. Easiest to measure ismultiunit activity, registering the summed activityfrom several neurons near the electrode tip. Thiscan be performed in awake animals. In contrast,single unit activity, the activity of a single neuron,is very difficult to measure, whether inside oroutside the cell. It is feasible with tissue slices oranaesthetized animals, but difficult in awakeanimals; with each movement, the electrodeattached to the head shifts slightly and losescontact with the cell. One solution has been touse movable electrodes that can be adjusted witha small micro motor attached to the bird’s head(Fee & Leonardo 2001). Sometimes it is possibleto extract single unit activity out of multiunitrecordings. Multiunit measurements are sufficientto establish general relationships between activityand behavior, whereas single unit measurementsare better for exploring mechanisms ofcommunication within a neural network.

3. Molecular studies in songbirds have hadtheir biggest impact by measuring the synthesisof gene products, messenger RNA (mRNA), andproteins. The order of events for gene productsynthesis is that gene in the DNA is used tosynthesize mRNA by a process called transcriptionand the mRNA is then used to translate thegenetic code and synthesize the protein by aprocess called translation. In situ hybridizationis used to detect mRNA synthesis andimmunocytochemistry for detecting proteinsynthesis. Most useful for songbird studies havebeen activity-dependent genes; their productsare synthesized as a result of neurons firing actionpotentials. The activity-dependent gene moststudied is ZENK. The molecular function ofthe ZENK protein is to bind to the DNAregulatory regions of select genes and eitherenhance or repress their mRNA transcription.Because of this, ZENK is called a transcriptionfactor. It is also called an immediate early gene,or IEG for short, because in the brain ZENK isnormally synthesized at very low levels and then

is briefly up regulated in cells after a short periodof increased brain electrical activity (Worley etal. 1991). Once the activity ceases or returns tobaseline levels, new synthesis stops. Because thehalf-lives of the mRNA and protein are short,about 15 min and 30 min respectively, theaccumulation and presence of the gene productis short-lived (Herdegen & Leah 1998). In thismanner, the IEGs can be used in ways somewhatsimilar to functional magnetic resonance imaging(fMRI), in the sense that fMRI activity impliesrecent electrical activity. However, like fMRI,the relationship of ZENK expression with neuronelectrical activity is not one-to-one. Many areasof the thalamus, the pallidum, and primarysensory neurons of the pallium do not synthesizeZENK, regardless of the level of activity. OtherIEGs are differently distributed in the brain,and other factors besides electrical activity canaffect IEG synthesis (Box 29, p. 236).

The Posterior Vocal Pathway

When either HVC or RA, the two main cerebralstructures of the posterior pathway, is lesionedbilaterally, on both sides of the brain, songbirdsare unable to produce learned vocalizations(Nottebohm et al. 1976). Lesioned canaries stillattempt to sing as judged by their posture andthroat movements, but they are silent or produceonly faint sounds (Fig. 8.3A). Zebra finch becomeunable to produce learned calls, or losemodifications that they have learned, revertingto the innate version (Simpson & Vicario 1990).Innate vocalizations are retained. When Uva orNIf are lesioned, most learned syllables areretained, but syntax, the ordering of learnedvocalizations, is affected and becomes morevariable (Williams & Vicario 1993; Hosino &Okanoya 2000). With large lesions incorporatingthe midbrain vocal nucleus DM or thetracheosyringeal nucleus, birds can no longerproduce either learned or innate vocalizations,becoming mute (Brown 1965; Nottebohm etal. 1976; Seller 1981). Smaller lesions of DMcan reduce the motivation to vocalize. Takentogether, these experiments show that to utter


BOX 29

ACTIVITY-REGULATED GENE EXPRESSION IN THE BRAIN: A HYPOTHESIS

The association between electrophysiological activity and the synthesis of so-called immediate early genes (IEGs) is wellestablished, but the exact relationship is not known. IEGs may be directly responsive, not to electrical potentials inneurons, but to the neurotransmitters that are released at synapses. The neurotransmitters, such as glutamate anddopamine, then bind to their respective receptors. If enough receptors are occupied, two independent events arehypothesized: (i) depolarization of the post-synaptic neuron and subsequent firing of it’s own action potential to generatebehavior or process sensory information; and (ii) activation of second messenger pathways that leads to synthesis ofZENK, c-fos, BDNF, and other IEGs also in the post-synaptic neuron. Second messengers are molecules such ascalcium ions that enter the cell or cyclic AMP molecules that are formed within the cell, in both cases after the transmitterbinds to its particular cell receptor. The second messengers then attach to and activate protein transcription factors readyto go into the cell. These then bind to the promotor regions of select IEGs, such as ZENK, to turn on their mRNA andsubsequent protein synthesis. Soon after performing their various functions, the IEGs are rapidly degraded by enzymes.Methods that reveal the presence of IEG activation provide valuable insights into which brain areas and circuits wereengaged immediately beforehand.

Post-synaptic neurons that do not express the necessary receptor or second messenger systems to activate synthesisof a particular IEG, such as ZENK, will still fire action potentials in response to pre-synaptic input on its receptors, butwill not turn on ZENK expression. One can also experimentally dissociate post-synaptic activity of neurons, preventingthem from firing, and still get neurotransmitter-induced ZENK synthesis (Keefe & Gerfen 1999). However, in most partsof the brain, in contrast with other tissues, electrophysiological activity and IEG synthesis are co-induced by synapticneurotransmitter release. I hypothesize that the level of IEG expression is controlled by the firing rate of the presynapticneurons. A literature analysis suggests that the higher the rate of action potentials, the greater the amount of IEGsynthesized, at least for ZENK (Chew et al. 1995; Mello et al. 1995; Stripling et al. 1997; Jarvis et al. 1998; Hessler &Doupe 1999b). Brain areas that do not express ZENK, such as the pallidum, primary sensory neurons of the pallium, andparts of the dorsal thalamus (Mello & Clayton 1994; Jarvis et al. 1998), presumably lack the appropriate receptors, orhave receptors that inhibit its expression. Other IEGs are synthesized in response to neuronal activity in different subsetsof brain regions. For example, the IEG c-fos, also a transcription factor, is synthesized in the same areas as ZENK, butit has a higher threshold for induction by activity, and it is synthesized at relatively higher levels in pallial song nuclei bysinging (Kimpo & Doupe 1997; Wada & Jarvis unpublished). The mRNA of BDNF, a brain growth factor, is synthesizedafter singing only in pallial regions of the cerebrum, including the song nuclei, and not in the striatal nucleus Area X (Liet al. 2001). Presumably the striatum, including Area X, does not have the necessary receptor combination to induceBDNF by neurotransmitter release. The multiplicity of receptor types for different neurotransmitters means that stimulimay trigger many different electrical and molecular responses throughout the brain.

Erich D. Jarvis

Proposed Model of Activity-regulated Gene Expression

Pre-synapticaxon:

Cortex (mammals)HVC (songbirds)

Behavior-inducedaction potentials

e–

Post-synapticdendrite:

glutamate receptors

dopamine receptors

Glutamate release

Dopaminerelease

Modulation actionpotentials

e–Pre-synaptic

axon:dopaminemidbrainneurons

Neurofilaments

Synapsins

Targetgenes

ZENK

Ca2+ Post-synaptic neuron

ZENKmRNA

ZENKprotein

Striatum:Striatum (mammals)Area X (songbirds)

Existingtranscription

factors


learned vocalizations, most cerebral nuclei ofthe posterior pathway are required; midbrainand hindbrain nuclei are needed for productionand contextual modulation of all vocalizations,innate and learned.

Electrophysiological recordings show thatimmediately before singing, neurons in NIf firefirst, followed by HVC, and then RA (McCasland1987). This type of premotor activity occursmilliseconds before sound output (Fig. 8.3B).During singing these nuclei continue to fire actionpotentials and stop milliseconds before the soundoutput ceases. This firing pattern indicates thatthe posterior pathway neurons are the direct braingenerator of the learned vocalizations. Molecularstudies show that the act of singing induces alarge increase in synthesis of ZENK and otherIEGs in the posterior vocal pathway nuclei (Jarvis& Nottebohm 1997; Kimpo & Doupe 1997).

Low levels of ZENK mRNA appear within thefirst 5–10 min of singing, peak after 30 min,and stay at a steady state as long as the birdcontinues to sing at a regular rate. The amountsynthesized is related to the number of songs abird utters (Jarvis & Nottebohm 1997). Thismeans that either the mRNA is stabilized, or itis degraded and re-synthesized at a steady rate asthe bird continues to vocalize. The interactionsbetween behavior, electrical activity, and geneexpression, are such that electrical activity inthe brain network leads to muscular activity, inturn producing vocal behavior and at the sametime inducing gene expression in that pathway(Box 29, p. 236). The gene products synthesizedthen regulate the expression of other genes.

Two open questions are: how does the electricalactivity of the posterior vocal pathway actuallygenerate singing behavior and what are the cellular

Effect of lesionsSome Functions of the Posterior Vocal Pathway

Canary

Before HVC lesion

After HVC lesion

1 2 3 4 5Time (s)

(A)

Electrical activity Zebra finch10

0

10

0

Fre

quen

cy(k

Hz)

Act

ivit

y(m

v)

300Time (ms)

HVC 400Time (ms)

RA

(B)

Fre

quen

cy (

kHz)

8

6

420

8

64

2

0

Figure 8.3 (A) Sonograms of canary song before and 9 months after bilateral lesions to HVC. Modifiedfrom Nottebohm et al. (1976). (B) Electrical activity during singing in HVC and RA of two different malezebra finches. Modifed from Yu & Margoliash (1996).


and behavioral consequences of regulated geneexpression. Insight into the first question comesfrom single unit recordings of individual HVCand RA neurons in zebra finches while singing,or in other states with neurons firing as they doduring singing. In HVC, each RA-projectingneuron produces a burst of action potentials onceper song motif, the repeated unit of zebra finchsong, at a precise time within the motif. TheseHVC neurons burst in a sequence (Hahnloseret al. 2002). In contrast, each HVC interneuron,as well as neurons in RA, fires multiple timesduring a song motif, in synchrony with eachother and with the single burst of the RA-projecting neurons (Yu & Margoliash 1996;Hahnloser et al. 2002). The X-projecting neuronsof HVC fire at rates intermediate between theRA-projecting and interneurons. These findingsindicate that each HVC interneuron and eachneuron in RA connected to the tracheosyringealnuclei receives convergent input from multipleHVC RA-projecting neurons. They also suggestthat HVC has many subpopulations of RA-projecting neurons, each active at different timesin a song motif and each activating a differentensemble of RA neurons (Hahnloser et al. 2002).The different RA ensembles then producepatterns of activity in the muscles of the syrinxand respiratory apparatus, controlling the timingof muscle contraction and relaxation appropriatefor the sounds produced. That is, RA has to firein a manner that coordinates timing of bothsyringeal muscles for vocalizing and abdominalmuscles for breathing (Wild 1997). This appearsto be achieved by two different but interconnectedsets of neurons in RA, one projecting to thetracheosyringeal motor neurons and the otherto respiratory premotor neurons (Mooney et al.2002). During breathing without singing, therespiratory neurons regulate breathingindependently of RA. During singing, RA aswell as DM take over (Vicario 1994; Wild et al.1997a; Mooney et al. 2002). Theelectrophysiological findings further suggest thatHVC interneurons make strong inhibitorycontacts on the X-projecting neurons, modulatingtheir firing during singing and thus HVC’s signals

to the anterior vocal pathway. However, moreinvestigation is required to decipher the exactnature of the relationship of electrical signalsfrom Uva to NIf to HVC to RA to thetracheosyringeal and respiratory nuclei, and ontothe syringeal and respiratory muscle, as well assignals to the anterior vocal pathway.

Insight into the cellular and behavioralconsequences of regulated gene expression comesfrom cell culture and gene blocking experimentsin other species. In cultured mouse cells, theZENK protein binds to the regulatory regionsof genes involved in modulating the structureof neurons and transport of molecules insideneurons (Box 29, p. 236). It has also beenhypothesized that IEGs like ZENK act asmolecular switches that convert short-termmemories into long-term ones (Goelet et al.1986). However, songbirds that are singing well-learned songs still produce ZENK (Jarvis &Nottebohm 1997). I hypothesize that in theposterior vocal pathway, perhaps every time thebird sings, ZENK is induced to help replaceproteins that get used up during the act of singing,and maintaining the song motor memories; whenthe bird sings 30 min later, the pathway is readyto produce song again.

The Anterior Vocal Pathway

In contrast to the posterior vocal pathway, manyaspects of the anterior vocal pathway have beenenigmatic since its discovery as, for example,the naming of ‘Area X’ (Nottebohm et al. 1976).In 1999, the neurobiology graduate students ofDuke University captured the essence of thisenigma with a play they wrote for thedepartmental retreat, called the ‘Area X-Files.’Interestingly, the brain region in which Area Xis located, the striatum, puzzled scientists studyinghumans and other mammals years before it wasofficially accepted at the Duke University 2002nomenclature forum that this area in birds ishomologous to the mammalian striatum (Wilson1914; DeLong & Georgopoulos 1981; Parent& Hazrati 1995; Brown & Marsden 1998).Nevertheless, the anterior vocal pathway of


songbirds has special functional properties thathelp us gain a general understanding of how thecerebrum works.

When nuclei of the anterior vocal pathwayare bilaterally lesioned, birds of all ages stillproduce some form of learned vocalizations. Theycan sing, but ongoing vocal learning is disruptedand they can no longer imitate new sounds. Forexample, when juvenile zebra finches are lesionedin lArea X during the sensitive period for vocallearning, the bird’s song remains plastic and thusnothing new can be imitated or crystallized.When lesioned in lMAN, the bird’s song rapidlybecomes stereotyped and again nothing new canbe imitated (Bottjer et al. 1984; Sohrabji et al.1990; Scharff & Nottebohm 1991; Nordeen &Nordeen 1993). Note that most so-called AreaX lesions, are actually in its lateral portion lAreaX. In zebra finch adults that have mastered theirsong and are no longer imitating anything new,lesions to lMAN and lArea X have no apparenteffect. However, in species that still learn as adults,either by imitation or improvisation (canaries)or re-develop old songs after a non-singingseasonal lapse (white-crowned sparrows), lesionsin lMAN do have effects (Nottebohm et al. 1990;Suter et al. 1990; Benton et al. 1998). In theplastic song of adult canaries, which occurs inthe fall, lesions to lMAN reduce vocal plasticity,whereas during stereotyped singing, in the spring,lesions have no effect. In white-crowned sparrowsbefore the yearly re-initiation of singing in thefall, lesions to lMAN result in stereotyped songfirst becoming plastic and then crystallizing to adifferent song; but lesions during the highlystereotyped singing season in the spring have noeffect. And in zebra finches, adult song can berendered plastic again either by partial cuttingof the tracheosyringeal nerve or by deafeningthe animal. These manipulations lead to gradualmodification and deterioration of the song, butprior lesions to lMAN prevent this experimentallyinduced plasticity (Williams & Mehta 1999;Brainard & Doupe 2000).

Thus, regardless of age, the anterior vocalpathway is not needed for song production, butis necessary for naturally occurring orexperimentally induced song learning or

modification. If the pallial structure lMAN isremoved when song is plastic, song becomesstereotyped; but if song is already stereotyped,then there is no change. If the striatal structurelArea X is removed when song is plastic, it remainsso; but if song is already stereotyped, then thereis also no change. This suggests that during vocallearning lMAN of the pallium adds variabilitywhereas lArea X of the striatum adds stereotypyto the vocalizations. A balance between the twoenables learning to occur. For these reasons, itappeared at first that during non-imitative stagesof life, such as the adult phase in zebra finches,the anterior vocal pathway is no longer used.

Then a challenge emerged. In both juvenilesand adults the act of vocalizing induces ZENKgene expression in nuclei of the anterior vocalpathway of zebra finches, in all of MAN and ofArea X (Jarvis & Nottebohm 1997). In fact inadults as well as young, ZENK synthesis in AreaX was the highest of all vocal nuclei. Thenelectrophysiological studies of adults revealedaction potentials milliseconds before andthroughout singing in both lMAN and lArea X(Hessler & Doupe 1999a). In lAreaX, tonicallyactive neurons were recorded and, besidesincreased activity, some decreased activity duringsinging. Neither a change in electrophysiologicalactivity nor ZENK expression occurred whenthe birds simply heard another bird’s song; smallincreases of activity occurred in half of the animalswhen the birds heard playbacks of their ownsong. In contrast, large increases of activity andZENK expression occurred when deafened birdswere actively singing, at levels not detectablydifferent from intact birds singing. These findingssuggest that electrophysiological activity and geneexpression in the anterior vocal pathway aremotor-driven, as in the posterior vocal pathway.This still begs the question of why lesions of theanterior vocal pathway, at least the lateral half,do not affect singing in adults with well-learnedsong, when the pathway is highly active duringsinging.

Social Context

One answer came when the vocalizing-driven


BOX 30

DIRECTED AND UNDIRECTED SONG: ZEBRA FINCHES IN CAPTIVITY AND THE FIELD

Zebra finches sing frequently in the field and in captivity. Throughout the year, at almost any time of the day whenbirds are stationary in trees and shrubs, one can hear the cheerful, mechanical sounds that constitute songphrases of the zebra finch. During pre-copulatory courtship, directed song is emitted by a sexually aroused malewhen he sings directly at the female a few centimetres away as he dances towards her. The visual and vocalcomponents combine to form a powerful sexual signal that is often ignored, or avoided, by most females; but ifconditions are right, she responds with a tail vibration display that is an invitation for the male to mount andcopulate. In wild flocks, males will confront and sing to any new female that lands near them, but copulation israrely invited except by their own mated female at a private location in the few days before egg-laying. Whenchoosing sexual partners laboratory females prefer males with a high rate of singing and with complex songphrases (Collins et al. 1994). Undirected song, originally called solitary song, is more frequently heard. Males,often perched alone on the tops of bushes, will stare straight ahead, and sing many phrases that usually appearto be completely ignored by other zebra finches. Undirected singing is also common in resting flocks when malesseem to find enough solitude for a few phrases. In captivity, visual isolation from conspecifics frequently increasesbouts of undirected singing. During undirected song, males remain stationary and never make any courtshipmovements. The two versions of song are equally loud, and directed song is a more intense performance: faster,more notes and longer bouts, though the differences are subtle (see below; Sossinka & Böhner 1981). Whiledirected song is clearly a sexual signal, the function of undirected song is far from clear. When a female partnerwas experimentally removed, a wild male immediately increased his undirected song rate and reduced it whenshe returned (Dunn & Zann 1996a). This suggests males are advertising their quality and unmated status viaundirected song and the presence of their mate inhibits such performance. Close proximity of male companionsalso inhibits undirected song but the functional significance of this behavior is unknown. During breeding,undirected song is most commonly performed during nest building when the female has just entered the partlybuilt enclosed nest and he is just outside on the way to collect more nesting material. If he does not sing thereis a good chance she will leave the nest shortly afterwards and this could result in extra-pair mating during herfertile period or allow other females to dump eggs in her nest. Thus undirected song in this context appears tobe a form of mate and nest guarding (Dunn & Zann 1996b).

Richard A. Zann

Zebra Finch SongUndirected Directed

1 s 1 s

Fre

quen

cy (

kHz)

8

4

0


synthesis of ZENK in lMAN and lArea X, aswell as the lower two-thirds of RA, was found tobe dependent upon social context (Jarvis et al.1998). In zebra finches, undirected singing, givenwhile not facing another bird, increases ZENKsynthesis in lMAN, lArea X, and RA. Directedsinging, facing another bird while singing, usuallya female, induces much less ZENK synthesis inthese nuclei. In contrast, ZENK expression in

the medial part of the anterior pathway, mMANand mArea X, and in both RA-projecting andX-projecting neurons of HVC is similar duringdirected and undirected singing. Electrophysio-logical recordings are consistent with these results,showing that in lMAN, lArea X, and RA, electricalactivity is different during undirected and directedsinging, whereas in HVC it does not differ(Hessler & Doupe 1999b; Dave & Margoliash

BOX 31

ANTERIOR FOREBRAIN PATHWAY LESIONS DISRUPT SONG IN ADULT BENGALESE FINCHES

Lesioning Area X or LMAN in the brains of juvenile zebra finches had profound effects on song learning andperformance, but no effect was detected in adult zebra finches (see p. 000). Nevertheless, the nuclei of theanterior forebrain pathway do not regress in adulthood. Furthermore, a part of this system is apparently activewhile singing, as shown by electrophysiological recordings and gene expression studies. Thus, the function of theanterior forebrain pathway in adulthood remains an enigma. We used adult Bengalese finches to re-examine thereal-time involvement of the anterior forebrain pathway in song production. We selected this species becauseBengalese finches are critically dependent upon real-time auditory feedback when producing the adult song(Okanoya & Yamaguchi 1997; Wooley & Rubel 1997). We reasoned that the feedback control might be mediatedby the anterior forebrain pathway. When a partial lesion of Area X was made in adult Bengalese finches, a markeddeficit was observed; the number of song note repetitions increased dramatically after the lesion. Curiously, theeffect occurred only in the portion of the song where the number of repetitions was naturally variable; the part ofthe song where the number of repetitions was fixed was not affected at all. The effect of surgery lasted up to twoweeks, after which the original song was recovered, identical with that in the preoperative recordings (Kobayasiet al. 2001). We suspect that the symptom observed here might be somewhat similar to Huntington’s disease inhumans in that this behavior becomes difficult to stop once it has started. It seems that the anterior forebrainpathway, including the basal ganglia, may be involved in the real-time control of song production, especiallyregarding the temporal precision of song duration.

Kohta I. Kobayasi & Kazuo Okanoya

Pre-Area X lesion

Post-Area X lesion

Bengalese Finch Song

stuttering*

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5Time (s)

Fre

quen

cy (

kHz)

10

6

2


2000; Jarvis et al. 2002). During undirectedsinging, both lMAN and lArea X, the two regionsbest studied electrophysiologically, show robustfiring throughout song bouts in a relatively noisymanner (Hessler & Doupe 1999b). Duringdirected singing, much less firing occurs and inlMAN the noise drops out leaving a patternmore matched to individual song syllables.Because lMAN projects to RA and mMANprojects to HVC, and because there is a socialcontext difference in RA and not in HVC, Ipropose that the lateral part of the anterior vocalpathway regulates RA and the medial partregulates HVC.

The comparison of activity during directedand undirected singing has given rise to variousalternative hypotheses about the functional roleof the anterior vocal pathway in adults. Onehypothesis is that, besides learning, the anteriorvocal pathway is also used to produce smallmoment-to-moment differences in singing outputto modulate the meaning of vocalizations forlistening birds (Jarvis et al. 1998). In zebra finches,there are small differences between directed andundirected singing (Box 30, p. 241), includingthe presence of more introductory notes, slightlyfaster motifs, and longer song bouts in directedsong (Sossinka & Bohner 1980). Anotherhypothesis is that the anterior vocal pathway issomehow connected with the attention that thebird gives to its surroundings during directedversus undirected singing (Hessler & Doupe1999b). Still another suggests that undirectedsong is simply practice, and that use of the anteriorvocal pathway and the production of ZENKmaintains the pathway’s health and sustains motormemories (Jarvis & Nottebohm 1997; Jarvis etal. 1998). During directed singing, stimulatingthe female, as the object of desire, is presumablymore important than song practice. A proposedrole for auditory feedback will be discussed later.However, if there is a maintenance function, lesionsto the pathway should result in slow deteriorationof song. This has not been found in adult zebrafinches. An effect has been seen in well-learnedsong of a close relative, the Bengalese finch(Kobayasi et al. 2001). Lesions to a large portion

of Bengalese finch Area X resulted in temporaryeffects on song syntax. The birds stuttered whenproducing syllables that were often repeated innormal song (Box 31, p. 242), suggesting a rolefor the anterior vocal pathway in the generationof some adult syntax, but the role with regard tosocial context is not clear. There were no differen-tial effects on directed versus undirected singing.

Interestingly, lesions to the medial part of theanterior vocal pathway, the part that is alwayshighly active during both directed and undirectedsinging, results in an immediate effect on syntaxin adult zebra finches (Foster & Bottjer 2001).Lesions to adult mMAN caused increased syntaxvariability, but had very little effect on songsyllable structure. Lesions to mMAN in youngbirds did not appear to affect their early plasticsong, but the birds could not crystallize astereotyped syntax, though they still had relativelystereotyped syllable structure. It should be notedthat these nuclei are very tightly packed together,and a number of the lesions encompassed mHVoabove and/or mArea X below mMAN. Generally,the medial part of the anterior pathway appearsto be involved in syntax learning and production,with mMAN possibly influencing HVC to helpform stereotyped syntax. Taken together withthe effects of lesions in the posterior pathwaysuggests that removal of any of the vocal nucleiinputs to HVC, such as NIf, Uva, and mMAN,results in the inability of HVC to produce normalstereotyped syntax (Foster & Bottjer 2001).

Despite the progress, the basic function ofthe anterior vocal pathway remains rather elusive.It surely plays a role in vocal learning, and it isvery active during singing at all stages of life.But beyond that, the situation is less clear, as isalso true of the anterior forebrain-basal gangliapathway of mammals (Wilson 1914; DeLong& Georgopoulos 1981; Parent & Hazrati 1995;Brown & Marsden 1998). Deciphering the basicfunctions of the anterior vocal pathway may yieldnew insights into cerebral functioning in general.But to understand vocal learning, we must alsoconsider how the anterior vocal pathway andthe song system in general gain access to auditoryinformation.


THE BRAIN AND AUDITION

Functions of the Auditory Pathway

For some years after discovery of the songbirdvocal control system, the search in the brain forthe locus of auditory song processing focusedon the vocal nuclei. There was a conviction thatthe place to look was where auditory and motorinformation meet. In the first electrophysiologicalstudies of HVC, action potentials were detectedin response to sound, especially to playbacks ofthe bird’s own song (Katz & Gurney 1981;McCasland & Konishi 1981), thus identifyingthe vocal nuclei as possible auditory processingstations. It came as a surprise when molecularbiologists played songs to zebra finches andcanaries and found induced ZENK synthesis,not in the vocal nuclei, but in another set ofcerebral brain areas, including NCM (Mello etal. 1992; Fig. 8.4A). This finding set NCM onthe map and eventually six other cerebral auditoryresponsive areas as a network (Fig. 8.3) importantin the processing of song as stimuli (L3, L1,HVC shelf, CMM, RA cup, and CMS; Mello& Clayton 1994). The electrophysiology of someof these auditory areas had been studied earlier

(Leppelsack 1978), mostly in the context of non-song sounds (Müller & Leppelsack 1985; Müller& Scheich 1985), but their potential relevanceto song processing and learning was notappreciated at the time.

We now know that when male or femalesongbirds hear songs of their own species, highrates of action potential firing and stronginduction of ZENK synthesis occur in these sevenauditory cerebral areas. When they hear songsof other species, less ZENK is synthesized inthese areas (Fig. 8.4A). When they hear shortduration pure tones, ZENK is not induced, andelectrical activity is actually inhibited belowbaseline (Mello & Clayton 1994; Chew et al.1995; Jarvis & Nottebohm 1997; Stripling etal. 1997; Jarvis et al. 2002). Thus,electrophysiological and molecular responses ofthe songbird auditory cerebrum are highlysensitive and species-specific. One area in thepallium, L2, and one in the thalamus, Ov, areknown to be auditory as judged by their sound-induced electrical activity and connectivity, butZENK is not expressed in them. In the midbrainauditory region, MLd, ZENK synthesis doesoccur in response to hearing song.

The species-specific molecular response in

Figure 8.4 Hearing-induced gene expression in the zebra finch brain. (A) Accumulated ZENK mRNA inNCM in the auditory pathway, relative to silence, when zebra finches hear various stimuli for 30 min; theother species was canary. From Mello et al. 1992. (B) Accumulated ZENK mRNA in NCM when birds heardfirst a familiar song, then a novel one. Modified from Mello et al. 1995.

Gene Induction in the Auditory Pathway (NCM)

Species specificity

Hearing

ZE

NK

fol

d-in

duct

ion

in N

CM

10

8

6

4

2

0Silence Tones Other- Own-

species speciessong song

(A)

0 30 60 90 120 150 180 210Song novelty

Time (min)

Hearing

Song A Song B

ZE

NK

fol

d-in

duct

ion

in N

CM 6

5

4

3

2

1

00 90 180 270 360 450 540 630

Presentation number

(B)


NCM is learned while the birds are growing up.When young zebra finches have been raised insocial isolation, playback of their own-speciessongs does not induce ZENK synthesis in NCM(Jin & Clayton 1997). When zebra finches areraised by canaries, as adults, the most potentstimulus at activating ZENK synthesis in theauditory cerebrum is not zebra finch song butcanary song (Ribeiro 2000). Thus, what a birdhears as a juvenile can influence its brainmolecular responses as an adult.

Electrical activity and ZENK synthesis in thesecerebral areas is also influenced by song familiarity.When males or females hear playbacks of novelzebra finch songs, ZENK synthesis is strong inthe cerebral auditory regions, but is reduced whenthey become familiar with songs after repetitions,as if they have a molecular memory of recentlyheard songs (Mello et al. 1995; Fig. 8.4A). Novelsongs also induce electrical activity, with an initialburst of high firing rates in NCM neurons thatalso decreased as the songs were repeated. Whenhearing a familiar song however, the firing ratenever decreased to silent baseline levels. If thenow familiar song is presented several hours or aday or two later, the firing rates start where it leftoff (Chew et al. 1995; Stripling et al. 1997; Fig.8.4B). This is a form of neuronal memory calledlong-term habituation. It appears that auditorymemories of song are being stored in NCM andother connected cerebral auditory areas.

Lesion studies also show that long-term songmemories can involve NCM. Ikebuchi &Okanoya (2000) trained Bengalese finches todiscriminate between songs and measured heartrates when the birds listened to novel and familiarsongs (Ikebuchi et al. 2003). Female heart ratesincreased when they heard novel songs. As theybecame familiar with the song, their heart ratesdecreased to a steady level. When NCM waslesioned, the females’ heart rates no longerincreased on hearing novel songs, suggesting thatNCM is required for discrimination of novelfrom familiar songs. They could learn a behavioraldiscrimination task in which birds had todiscriminate between two songs heard, but couldnot remember later what they had learned. It

appears that NCM is not required for the short-term auditory memory of song, but is neededfor long-term memory of them. These memorydeficits of NCM lesions are somewhatreminiscent of the symptoms of Alzheimer’sdisease in humans.

ZENK and other genes may be involved inthe formation of long-term memories. Whengeneral inhibitors of mRNA or protein synthesisare injected into NCM at the time of novelsong playback, they do not affect the short-termhabituation of neuronal activity that occurs withrepeated stimulation. However, when the songsare played later, after 3 hours, the electrical activityrate is as high as with a novel song (Chew et al.1995). Inhibiting mRNA and protein synthesisspecifically at 3 hours and 6 hours after hearingnovel songs also prevents further long-termmaintenance of the habituated memory of songs.Since ZENK and other IEGs are expressedquickly, within the first hour that birds hear thenovel songs, and because blocking synthesis ofgene products at this time affects long-termmemory, it appears that these genes may act asmolecular switches that convert short-termmemories into long-term memories (Goelet etal. 1986). The effect at later times suggests thatthere are several waves of gene expression involvedin the process.

The electrophysiological mechanisms by whichthe auditory pathway processes and learns species-specific sounds are only just beginning to beunderstood. As information is conveyed fromthe midbrain to the auditory regions of thecerebrum, each station shows more complexityin its response to sounds, including responsivenessto species-specific songs (Chew et al. 1995, 1996).In the cerebrum, field L2 responds first, followedby L1, L3, CMM, and NCM, in the order inwhich they are connected. L2 responses areevoked by many types of sounds, in a linearfashion, and in a tonotopic manner, with low-frequency neurons at the top of L2 and highfrequency at the bottom (Capsius & Leppelsack1999; Sen et al. 2000). At each subsequentstation, the neurons then respond less linearly,registering specific features such as frequency


modulation, syllable combinations, and down-sweeps and up-sweeps. These more sophisticatedresponses are also evident in the ZENK geneexpression response, with different species-specificsyllable types being processed in particular partsof NCM (Ribeiro et al. 1998). The perceptionof song probably involves all of these anatomicalregions, with those furthest removed from L2,in the Nidopallium (L1, L3, NCM), Mesopallium(CMM) and Lateral Striatum (CMS), possiblyserving as the prime locations of the song‘percept’.

Although we now know that NCM and otherareas are auditory centers for processing songs,this does not explain the lack of induced ZENKexpression and other IEGs in the vocal controlnuclei after hearing songs. This lack of increasedexpression is elucidated by the later discoverythat there was less hearing-induced electricalactivity in vocal nuclei when the animals wereawake, than when they were anesthetized orsleeping (Dave et al. 1998; Schmidt & Konishi1998; Hessler & Doupe 1999a; Nick & Konishi2001). Most earlier electrophysiological workwas conducted on anaesthetized subjects, whereasthe gene expression studies were done withawake animals. This new paradox, awake versussleeping-induced hearing activity, still left a majorquestion unanswered: where do the functionalinteractions between the auditory and vocalpathways take place?

Auditory Feedback and the Template

As is intuitively obvious, vocal learning, and vocalimitation in particular, requires that a bird hearthe tutor’s song that he will imitate. In addition,Konishi (1965a) found that a songbird also needsto hear himself practice that song in order toimitate it accurately. If a songbird is allowed tolisten and form an auditory memory of the tutor’ssong, but then is deafened before he physicallypractices that song, the bird will not learn toaccurately produce what he had earlier heard.From these behavioral experiments arose the ideaof the song template (Marler & Tamura 1964;Konishi 1965a; Fig. 8.5). This hypothesis

proposed that during song learning, a youngbird first forms an auditory memory, a templatesomewhere in the brain, based on the tutor’ssongs that he hears. Sometime thereafter, daysto almost a year depending upon the species,the bird begins to practice singing and by listeningto his own song, he will try to match his producedvocal output with the auditory template in hisbrain. Since the formulation of this hypothesis,the search for the auditory template and the sitefor auditory feedback in the songbird brain hasbeen a major area of research. The search haslead to findings that are both interesting andelusive.

As already mentioned, the exact source ofauditory input into the vocal pathways isunresolved, even though some auditory cellpopulations, L2, HVC shelf, and RA cup, aredirectly adjacent to vocal nuclei cell populations,NIf, HVC, and RA. However, the search forsites of auditory–vocal integration, orsensorimotor integration more generally, was notlimited to connectivity studies. It was thoughtthat HVC, showing electrical activity both whilevocalizing and when the bird hears hisown song played to him, should be a goodcandidate site for such integration and the songtemplate (McCasland & Konishi 1981;McCasland 1987).

But soon thereafter auditory responses werefound throughout all known cerebral vocal nuclei(HVC, RA, lArea X, lMAN) as well as in thetracheosyringeal motor neurons (Williams &Nottebohm 1985; Williams 1989; Doupe &Konishi 1991). Depending upon the anestheticused, auditory responses were induced not onlyby playbacks of the bird’s own song, but also byother birds’ songs and non-biological sounds,such as tones (Katz & Gurney 1981; Williams& Nottebohm 1985), leading Williams andNottebohm (1985) to propose the Motor Theoryof Song Perception, as a counterpart to the MotorTheory of Speech Perception for humans(Liberman et al. 1967; Liberman & Mattingly1985). They hypothesized that the same brainareas used to produce sounds (vocal–motor areas)were also used to perceive them (hearing–sensory


areas). This theory was part of a larger debatebegun even before Broca’s time, between the‘diffusionists’ and the ‘locationalists.’ Diffusionistsbelieved that brain functions were spread outand that the same area is used for more thanone function. Locationalists believed that specificregions of the brain served specific functions, aspropounded by Gall & Spurzheim (1810–1819;Benson & Ardila 1996). For songbirds, the firingof neurons throughout the vocal control systemin response to any sound brought the debatecloser to the diffusionist’s model. This findingalso indicated that the vocal pathways couldprocess more than just the birds’ own song,throwing some doubt on whether vocal pathwaysstore only a song template for auditory feedbackcomparison or simply process sounds for auditoryperception in general.

The next candidate was the anterior vocalpathway (Doupe & Konishi 1991), which atthe time did not seem to do anything in adultanimals after learning was complete. It wasproposed that when a juvenile zebra finch forms

an auditory memory of a tutor’s song that theauditory template is stored in the anterior vocalpathway. The pathway would then act as acoincidence detector between the song producedand the song heard. Specifically, it was proposedthat during juvenile practice, when vocal motoractivity originating in HVC generates song, thebird would hear itself sing, and this informationwould be transferred from the auditory pathwayto HVC. HVC in turn would send this auditoryfeedback signal into the anterior vocal pathwayvia its X-projecting neurons. The anteriorpathway would then compare the producedfeedback song with the expected song asrepresented by the template. With a mismatchthe anterior pathway would then correct thatdiscrepancy by changing the activity of RAneurons via lMAN (Williams 1989; Doupe &Solis 1997; Fig. 8.5).

A major feature of this hypothesis is thepresumption of responsiveness of the anteriorvocal pathway to auditory stimuli, particularlyof the bird’s own song. Three kinds of results

Song Learning, Auditory Feedback, and Template Hypotheses

Template Hypothesis 1965 Template Hypothesis 2000Tutor’s song

Tutor’s song Brain

Brain

(A)

(B)

Auditorymemory

Template

Ears

Auditoryfeedback

Syrinx

Produced song

Ears

Auditoryfeedback

Produced song

Syrinx

RA

Motorcopy

RApXp

HVCEfference copy

TemplateAreaXDLMlMAN

Sensory ormotor?

Figure 8.5 Template models of song learning. (A) Template hypothesis as proposed by Konishi 1965a.Tutor’s song is heard and stored as an auditory template. The bird tries to produce an imitated song, hearshimself sing and compares that with the template. (B) A template hypothesis proposed by Troyer & Doupe2000. First, the anterior pathway receives an efference copy of the song from the X-projecting neurons ofHVC, coming in turn from RA-projecting neurons of HVC. Second, auditory feedback is sent to the HVC’sX-projecting neurons to the anterior pathway and compared there with the efference copy.


began to change this view. First ZENK synthesisin the anterior vocal pathway, including MANand Area X, was found to be activated not byhearing playbacks of the bird’s own song, but bysinging (Jarvis & Nottebohm 1997). Moreover,when birds were deafened, singing still droveincreases in ZENK synthesis in the anterior andposterior vocal pathways to the same levels foundwhen they could hear. Second, like the posteriorvocal pathway, auditory activity in the anteriorvocal pathway was found to be shut off or highlydiminished when zebra finches were awake(Margoliash 1997; Hessler & Doupe 1999a). Inaddition, like the posterior vocal pathway, thesinging electrical activity in lMAN and lArea Xwas premotor, firing before song output, andoccurring whether the birds could hear or weredeaf (Hessler & Doupe 1999b). Thus, in theanterior vocal pathway there is motorelectrophysiological activity and gene expressionwith no requirement for auditory feedback. Third,were the discoveries of closure of the anteriorforebrain loop via the lMAN to lArea Xconnection (Nixdorf-Bergweiler et al. 1995; Vates& Nottebohm 1995; Luo et al. 2001), and medialand lateral parallel connections of the anteriorpathway, one with an output to RA and theother to HVC (Foster et al. 1997; Jarvis et al.1998). This suggested the more global view, thatthe anterior vocal pathway serves, not as anindirect route of transferring information fromHVC to RA, but as a closed-loop processingstation receiving information from HVC andsending that processed information orinstructions to HVC and RA. Debate continuesabout whether there is low auditory activity ornone at all in the anterior and posterior vocalpathway nuclei. However, there may be speciesdifferences; preliminary data reveal robusthearing-induced activity in HVC of awake songsparrows (Nealen & Schmidt 2002).

Striving to reconcile these contradictions,Troyer & Doupe (2000) proposed a theoreticalmodel, showing how the anterior vocal pathwaymay still function in auditory feedback with astored song template. They argue that whereas amotor song signal is transported from HVC to

RA, via HVC’s RA-projecting neurons, at thesame time an ‘internal sensory efference copy’of song is transported in parallel to the anteriorvocal pathway from HVC to Area X, via HVC’sX-projecting neurons (Fig. 8.5B). The term‘efference’ refers to activity directed away from acentral location in the brain, in this case HVC.The two signals, one motor, the other an internalsensory efference copy, would arrive in RA andbe compared there, but auditory feedback fromthe bird’s own song would be compared in HVC’sX-projecting neurons. Problems confronting thismodel are the requirement that the signals fromHVC to Area X should be sensory, when thereis no known sensory input for the efference copyin this model, and the fact that in anesthetizedbirds all HVC neuron types, X-projecting, RA-projecting, and interneurons, are equallyresponsive to auditory activity (Mooney 2000).In addition, in awake birds, all HVC neurontypes show motor-driven ZENK expression andpremotor electrical activity during singing (Jarvis& Nottebohm 1997; Hahnloser et al. 2002).However, it is possible that the X-projectingneurons send a ‘motor efference copy’ of song(Margoliash 1997; Mooney et al. 2002).

If the auditory responses in vocal nuclei occurmainly when the birds are anesthetized, what istheir role in real life? Interesting answers werealso found in sleeping birds (Dave & Margoliash2000). When sleeping zebra finches hearplaybacks of their own song, the hearing-inducedelectrical responses in HVC and RA suddenlyappear, as if the birds were anesthetized. Whenthe birds wake up, the hearing-induced responsescease. In HVC, these auditory responses weremost prominent when the birds heard songplaybacks during deep, slow-wave sleep (Nick& Konishi 2001). When sleeping zebra finchesare not hearing playbacks of song, HVC andRA sometimes display spontaneous complexpatterns of electrical activity that match thesinging activity which the bird produced earlierthat day (Dave et al. 1998; Hahnloser et al.2002), as though the birds’ vocal motor pathwayis replaying the singing activity produced duringthe day, but without the bird actually singing or


hearing song. Replay during sleep might stabilizemotor memories of a bird’s songs so that it cansing them again later (Dave et al. 1998), or thebird may simply dream about singing (Box 32,p. 249).

The best evidence for a role of the anteriorvocal pathway in some form of auditory feedbackis that lesioning of lMAN prevents the songdeterioration resulting from tracheosyringealnerve cutting and deafening (Williams & Mehta1999; Brainard & Doupe 2000). Given thelimited auditory responses of awake animals, thisrole might only be permissive, with actualtemplate instructions coming from elsewhere,possibly from the auditory pathway. The species-specific molecular responses of ZENK andelectrophysiological activity in NCM and relatedareas have led to a focus on them as a potentiallocation of the template. When zebra finchesraised alone with a tutor became adults, inductionof ZENK synthesis was highest when the birdsheard song of their tutors (Bolhuis et al. 2000).Moreover, the closer a tutor was matched, thehigher the levels of ZENK synthesis in NCM.This response is counterintuitive, given thatfamiliar songs of non-tutor birds induce lessZENK expression (Mello et al. 1995; Fig. 8.4).It is possible that hearing tutor songs in novelsocial contexts may induce high levels, but thishas yet to be tested. Nevertheless, responsivenessof NCM and related areas is clearly selective,especially to sounds that resemble the birds’ ownsong, so that they could be involved in auditoryfeedback.

Another finding points in this direction. Whensongbirds hear themselves sing, ZENK is inducedin the bird’s NCM and other auditory areas in amuch more restricted pattern than when theyhear other birds sing, or their own song from atape recorder (Jarvis et al. 1998). Apparently awakebirds identify playbacks of their own song as anovel song from some other bird, sounding similarto their own. Thus, tests of the auditory feedbackand template hypotheses using tape recordedplaybacks of the birds’ own song may be deficientbecause this method does not mimic the state ofthe animal while actually singing himself.

In summary, the search for mechanismsunderlying auditory feedback and thehypothesized template continues, and has thepotential to yield exciting findings. It is not yetknown if the interactions between auditory–sensory information and vocal–motorinformation occur in the vocal pathway, theanterior vocal pathway loop, the auditory pathwayor somewhere not yet studied. Wherever theytake place, the sensorimotor interactions mightinvolve sub-threshold electrical responses, or somenon-electrical mechanism. It looks as thoughthe answers will be found by studying animalswhile they hear themselves vocalize, rather thanwhile hearing song playbacks, especially perhapsduring sensitive periods for vocal learning whenbirds are actively using auditory feedback togenerate imitated songs.

SENSITIVE PERIODS

‘Critical’ or ‘sensitive’ learning periods are stagesin life when there is a limited window ofheightened ability to learn new information. Theyoccur in all animals, and are prominent in humandevelopment (Rauschecker & Marler 1987).Songbirds have become a premier model forunderstanding the neuroethology of sensitiveperiods, and similarities with those for languagelearning in humans (Doupe & Kuhl 1999) makesongbirds especially interesting. Vocal learningin both songbirds and humans has four phases.In songbirds, the first is called the auditoryacquisition phase in which they form auditorymemories of the songs that they hear; the secondis a babbling-like subsong phase in which theyengage in vocal motor practice, but do notproduce imitations; in the third, plastic songphase, they practice imitating sounds withincreasing precision; fourth is a crystallizationphase in which they go through a period that ispuberty-like, the voice becomes stabilized andwell learnt, and the birds become adults andready to breed. The timing of these phases variesin different species and they sometimes overlap(see Chapter 3).


BOX 32

DO BIRDS SING IN THEIR SLEEP?

Sleep, as defined by specific postures, elevated sensory thresholds, and brain states, appears to be ubiquitous in birdsand mammals, yet its functions are poorly understood. It has been studied extensively in mammals but only recently hasthe phenomenon of sleep in songbirds begun to be addressed. All birds show REM and non-REM periods of sleep(Rattenborg et al. 2002), but the pattern and frequency may be different than in mammals. Small birds have short periodsof REM sleep (circa 4 s.; Szymczak et al. 1993). This may reflect the trend relating the period of sleep cycles to animalsize but, in general, birds may have less REM sleep than do mammals (Siegel 1995). Recent observations have giventhe study of birdsong an unexpected focus in sleep research. We have known for some twenty years that some neuronsin the song system of the brain respond best to playback of song, especially the bird’s own song; most detailed studieswere conducted on anesthetized white-crowned sparrows and zebra finches (Margoliash 1983; Margoliash & Fortune1992; Theunissen & Doupe 1998). The situation in unanesthetized birds is more complicated. In awake zebra finches,neurons in parts of the song system, such as RA, respond more weakly than expected to song playback, or hardly at all.The same RA neurons respond vigorously to song playback in sleeping birds (Dave et al. 1998). Whereas sensorythresholds in general are elevated during sleep, there are clearly exceptions to this rule. The implications of thisphenomenon for basic and perhaps even clinical research have yet to be fully appreciated. In birds, one hypothesis isthat suppression of auditory activity in awake animals in the motor pathways, including RA, is the expression of analready established, well-controlled adult song, and that auditorily-driven activity in RA during singing represents an errorsignal. The expression of auditory activity during sleep may reflect a singing-like state the song system enters, enablingit to access the hypothesized error signals that guide adaptive changes to re-stabilize song during sleep.

The initial RA recordings during sleep were from clusters of neurons, but eventually techniques were developed torecord from single cells in RA while birds sang, and to re-record them later during undisturbed sleep. The result was aremarkable finding (see fig.). In virtually every cell encountered, there was bursting activity during undisturbed sleep thatmatched the same bursting patterns observed for that cell when the bird was singing while awake (Dave & Margoliash2000). Analysis of electrical brain activity during sleep is challenging because there is no immediate, reliable timereferent linked to behavior, calling for some sophisticated statistical analysis. Nevertheless, obvious matches were foundbetween the sleeping and waking states (see below), as for example where an RA neuron emitted a long train of 30 ormore spikes organized into multiple bursts, each of which matched the sequencing of the activity patterns of that sameneuron during singing. Are birds singing in their sleep? The occurrence of coordinated bursting in song system neuronssuggests that song is being replayed in the sleeping brain. A similar phenomenon has been reported in the hippocampalspatial memory system of sleeping rats after they have been exploring a maze (Wilson & McNaughton 1994). Replayconcepts are at the heart of some theories postulating that memory consolidation takes place during sleep, but a role forsleep in the acquisition or maintenance of birdsong is still speculative. Do birds dream of singing? Owners of petbudgerigars and other birds will not need much convincing, but the truth is that we still have no objective access to themental imagery that birds may experience during sleep. In some cases RA neurons give trains of bursts during sleeprepresenting whole syllable sequences or motifs, and apparently the population of RA neurons bursts synchronouslyduring sleep. If we could learn how to release brainstem inhibition of the bird’s vocal system, we might find that sleepingbirds would actually burst into song.

Daniel Margoliash

RA Bursting in Sleep Matches Daytime Tutor Song

TutorSongRA

Bursting


Critical periods in vocal learning behaviorare controlled by hormones and the brain.Songbirds hatch from the egg presumably withcerebral vocal nuclei already present, but theycannot yet produce learned vocalizations. Day10 after hatching is the earliest that the vocalnuclei have been sought and found (Nixdorf-Bergweiler 2001). At this age, they produce theinnate begging calls that induce feeding by theparents (see Chapter 5). As the hatchling birdgrows so does its brain. The cerebral vocal nucleigrow at different rates (Nixdorf-Bergweiler 2001):in zebra finches, lMAN size peaks before thesubsong phase and then decreases; Area X, HVC,and RA sizes peak right after the subsong phaseand are maintained to adulthood.

Not all connectivity is in place after hatching.The connectivity of the anterior vocal pathwayis present, but that for the posterior vocal pathwayis not yet completed (Mooney & Rao 1994).During the auditory acquisition phase, whichoverlaps with the subsong phase in the zebrafinch and several sparrows, neurons from HVCto RA wait outside of the RA nucleus. As thebird begins to produce subsong, these HVCneurons grow into RA and find connections there.At this time, the lMAN axons in RA are spreadthroughout RA. Later as the bird begins toproduce plastic song, for the first time practisingimitations in earnest, the axonal spread is prunedback until the appropriate connections are made(Herrmann & Arnold 1991). During this pruningprocess, electrical activity in the anterior vocalpathway, from lMAN to RA, also undergoeschanges. Before subsong, the speed of electricalsignaling between the two regions is fast. Duringand after subsong, the speed decreases (Livingston& Mooney 1997; Mooney 1999). The growthof connections from HVC into RA is thoughtto prepare the posterior vocal pathway forproducing sounds. The pruning of connectionsfrom lMAN to RA and the decrease in signalingspeed are thought to aid in providing RA withthe necessary information from the anterior vocalpathway on how to sing the imitated song.

Changes in gene expression also occur duringsensitive periods. In young animals, before the

subsong phase, the amount of ZENK synthesisin the auditory regions, in vocal nuclei, and inother brain areas is high, without the need forhearing song or for singing (Jin & Clayton 1997;Jarvis et al. 1998; Whitney et al. 2000; Striplinget al. 2001). During the subsong phase, the basallevels start to decrease, and ZENK can then beinduced by hearing song, in the auditory pathway,and by singing subsong, in the vocal pathway(Jarvis & Nottebohm 1997; Jin & Clayton 1997).By adulthood, the basal levels are dramaticallyreduced and the induction of ZENK by singingand hearing is further amplified. The higher levelsof basal ZENK in juveniles are thought to reflecttheir higher brain plasticity compared with adults.During these successive developmental stages, ahost of other genes undergo changes in expressionin the vocal nuclei. These include genes forsynaptic transmission, affecting the glutamatereceptors that are responsible for changing thespeed of communication between neurons, andgenes for generating connectivity and hormonereceptors (Gahr & Kosar 1996; Soha et al. 1996;Singh et al. 2000). It is not clear how thesechanges in gene regulation influence sensitiveperiods, but some insight has been gained byexamining species and sex differences.

SPECIES AND SEXUALDIFFERENCES

There is just one vocal learning primate, humans.The vocal learning suborder of oscine songbirdsincludes over 4000 species (Nottebohm 1972;Sibley & Ahlquist 1990), with a wide spectrumof variation in vocal learning behavior, fromclosed-ended vocal learners, such as zebra finches,to highly plastic opened-ended vocal learners,such as canaries (Nottebohm et al. 1990;Catchpole & Slater 1995; see Chapter 3). Asjuveniles, most species go through similar phasesof song learning leading up to crystallized adultsong. However some, such as the zebra finchcrystallize one song as an adult, and others, suchas canaries, go through seasonal phases of plasticsong, and learn new song themes as adults; yet


others can learn new songs any time of the year.Like humans, however, most open-ended vocallearners still learn song most easily as juveniles.Other variations include differences in the abilityto imitate other species’ sounds, the ability toimprovise, and to increase repertoire size, andsyntactical complexity (see Chapter 4). We canassume that there is something significant aboutthe mechanisms operating in songbird brainsthat make this great variation in vocal learningbehavior possible.

Species Differences

Could there be different sets of vocal nuclei fromone species to another? Nearly a hundred songbirdspecies have now been examined, and they areremarkably consistent. All have the four largecerebral vocal nuclei, HVC, RA, MAN, and AreaX (Brenowitz 1991; DeVoogd et al. 1993;Brenowitz 1997). When carefully examined,many also have the smaller cerebral vocal nucleiNIf, Av, and Mo (personal observations). Thus,the behavioral variability is not explained by thepresence or absence of certain vocal nuclei.

A second possibility is differences in anatomicalconnectivity. This is more challenging; it takesyears of study to determine connectivity in onespecies. Nevertheless, connectivity between zebrafinches and canaries has been compared in detail,and although comparisons are not complete, nomajor differences have been found (Vates &Nottebohm 1995; Vates et al. 1997).

The third source considered was variation inthe size of vocal nuclei (Nottebohm et al. 1981;DeVoogd et al. 1993; MacDougall-Shackletonet al. 1998). Size differences between speciesand between individuals of the same species dooccur, but the data are contradictory andcontroversial. Some reports show that both acrossand within species, those with large vocalrepertoires, measured as the number of songs abird has, or the number of different syllables orphrases it employs, have relatively larger HVCs(Nottebohm et al. 1981; DeVoogd et al. 1993;Airey & DeVoogd 2000; Airey et al. 2000; Fig.8.6). In some species with seasonal learningperiods, HVC size increases during learningperiods (Nottebohm et al. 1986). However, othershave failed to replicate the finding that repertoire

Song Repertoire and Brain Space (HVC)

SpeciesGeneraFamilies

–0.10 –0.06 –0.02 0.02 0.06Relative HVC volume

Rel

ativ

e so

ng r

eper

toir

e si

ze

0.25

0.15

0.05

–0.05

–0.15

–0.25

Figure 8.6 Song repertoires and brain space. X-axis values were calculated as relative HVC/cerebrumvolume of one species minus relative HVC/cerebrum volume of another species (�), average HVC ofvolume of one genus minus that of another (O), or HVC volume of one family minus HVC of another (+).The same was done for y-axis values, calculating song repertoire size differences between species,genera, or families; males of ~40 species were used for these calculations. There is a positive correlationat all three levels of analysis. Modified from DeVoogd et al. 1993.


size is related to vocal nuclei size, at least withina species (MacDougall-Shackleton et al. 1998;Ward et al. 1998). The interpretation of thesecontradictory results may be complicated by thefinding that the amount of singing a bird performsinfluences the size of the vocal nuclei and thenumber of cells in them (Li & Jarvis 2001; Sartoret al. 2002; Alvarez-Borda & Nottebohm 2002).Thus, the changes seen in seasonally singing birdsmay occur because they sing more during learningperiods. Differences between species orindividuals of a species may reflect the amountof song a particular species or individual is proneto produce. As with muscles of the body, themore the bird sings, the greater the potentialincrease in size of the cerebral vocal nuclei, asdiscussed later. Yet there is some supportiveevidence that the relative size of the vocal nucleiis important; zebra finches have a larger lArea Xrelative to lMAN than do canaries (unpublishedobservations), a difference that is consistent withthe hypothesis that lArea X endows a bird’s songwith stereotypy and lMAN endows it withplasticity.

A fourth compelling source of speciesdifferences is genetic, especially the differentialexpression of genes for synaptic transmission.Synaptic transmission involves neurotransmittersreleased at synapses, and glutamate is the mostabundant neurotransmitter in the brain. The‘pre-synaptic neuron’ releases the glutamate fromits axons and the ‘post-synaptic neuron’ receivesthe glutamate on receptors in its dendrites. Eachtype of neuron has one or more of the 24 differentkinds of glutamate receptors. The receptorstransmit information about the electrical signalsbetween pre- and post-synaptic neurons, toinfluence in different ways the expression of genessuch as ZENK, in post-synaptic neurons (Box29, p. 236). Compared to the rest of the brain,the patterns of expression of glutamate receptorsare unusual in vocal control nuclei, apparentlyeither higher or lower than the surrounding brainsubdivision to which they belong (Wada et al.2001). Furthermore, each species examined hadits own unique pattern of expression in the vocalnuclei. Specializations of mGluR2, NR2A, and

NR2B glutamate receptor distribution werestrongly correlated with the type of syntacticalcomplexity displayed in the song of each species.Lower mGluR2 and NR2A, and higher NR2Bwere related to higher syntax variability.Syntactical complexity and vocal learning styleare often related, with open-ended vocal learnershaving more variable syntax than closed-endedspecies (see Chapter 4). It may be that one sourcefor some of the variability in vocal learning is tobe found in differences in the distribution ofancient gene families.

Sex Differences

There was once a widespread belief that, exceptfor breasts, genitalia, and overall body size, menand women were basically the same, particularlytheir brains. It came as a shock to many thatsome male songbirds have large cerebral vocalnuclei and females have smaller ones, or evennone at all (Nottebohm & Arnold 1976; Arnold& Mathews 1988). Although small sex differenceshad been detected earlier in other brain areas,this first demonstration of significant genderdifferences, seen even with the naked eye, in thebrain of an evolutionarily advanced vertebrate,challenged the notion that there are no significantdifferences between males and females (Raisman& Field 1971; Ball & Macdougall-Shackleton2001). Soon thereafter, scientists began to findsignificant sexual differences in select brain areasof other birds, and of mammals, particularly inparts of the hypothalamus that controlreproductive functions (Ball & Macdougall-Shackleton 2001), and cerebral areas that controllanguage (Harasty et al. 1997; Cooke et al. 1998).

In those songbird species in which the femaleshave few or no detectable vocal nuclei, such aszebra finches, the females do not engage in vocallearning. Female zebra finches are born withcerebral vocal nuclei, but as they mature theirvocal nuclei atrophy. As adults their cerebral vocalnuclei are hardly noticeable (Nixdorf-Bergweiler2001). One of them, lMAN, shrinks lessdramatically than others and in adult females ofsome species, destroying it hampers the perceptual


processing of the male’s song (Hamilton et al.1997; Burt et al. 2000). In canaries, femaleshave vocal nuclei, but they are smaller than thosein males, and females do not learn as much songor sing as much as males (Nottebohm & Arnold1976). In other species, such as bay wrens, vocalnuclei are comparable in size in males and females,and they counter-sing with each other (Arnoldet al. 1986; Ball & Macdougall-Shackleton 2001;Balthazart & Adkins-Regan 2002).

These gender differences may be related toenvironmental constraints. Most songbird studieshave focused on birds living in temperate climates,especially in North America and Europe, wherethere are more species in which only males sing(Morton 1996a), leading some to assume thatthe natural order of vocal behavior in birds isthat only males learn vocalizations and sing.However, many more species live in the tropicsand often both males and females produce learnedsong (Kroodsma et al. 1996). Morton (1996a)proposed that the changing seasons and harsherwinters could have led to selection for moredivision of labor, with males singing and femalesspecializing more in selecting mates and tendingto the young. This temperate–tropical contrastis apparent even when comparing differentpopulations of closely related species: At oneextreme, the Carolina wren lives in a temperatezone climate, and females have no detectablecerebral vocal nuclei and do not sing; at theother extreme are its tropical cousins, the baywren and buff-breasted wren, whose females havevocal nuclei and song repertoires equal in size tothose of males (Arnold et al. 1986; Brenowitz &Arnold 1986; Morton 1996a; Nealen & Perkel2000). Thus, the environment may in a sensehave had an influence on the evolution of genderdifferences in brain and behavior. Notsurprisingly, such differences are in parthormonally controlled.

The Role of Hormones

Testosterone and estrogen are found in allvertebrates; testosterone is often assumed toregulate male sexual behavior and estrogen

assumed to regulate female sexual behavior.However, this still popular notion has becomeoutdated, partly because of surprises about theinterrelationships of hormones, brains, and sexuncovered in songbirds. The first surprise waswhen Gurney & Konishi (1980) found that ifthey injected a young female zebra finch withextra doses of estrogen she would grow male-like vocal nuclei. Testosterone did not have thiseffect, but did cause estrogen-injected femalesto sing as adults later in life (Pohl-Apel & Sossinka1984). These females did not display maleplumage, but were sufficiently masculinized thatthey attempted to breed with other females. Thus,hormonal treatment of a developing songbirdcan convert a vocal non-learner into a learner.However, it is not the case that injecting hormonesinduces cerebral vocal nuclei and vocal learningin non-vocal learning species, such as quail; nordoes injecting estrogen into vocal learning malesincrease the size of their vocal nuclei; similarly,blocking estrogen and testosterone in young vocallearners or removing their gonads does notprevent formation of cerebral vocal nuclei(Adkins-Regan & Ascenzi 1990; Balthazart etal. 1995; Wade 2001), but does lead to a decreasein size of the vocal nuclei and reduction in theamount of singing, though the birds still learnto imitate song and sing (Wade 2001; Alvarez-Borda & Nottebohm 2002). In addition, femalecanaries, which sing less than males, sing morewhen given testosterone, and their vocal nucleiincrease in size during the process (Nottebohm1980; DeVoogd & Nottebohm 1981). Debatesensued about the roles and sources of testosteroneversus estrogen in the development of cerebralvocal nuclei and song learning (Wade 2001;Balthazart & Adkins-Regan 2002). The searchfor solutions is ongoing and not withoutcontradictions (Box 33, p. 255), but the followingis a synthesis of the facts as they now stand.

Testosterone is actually a pro-hormone, ahormone precursor. It is converted by threedifferent enzymes, aromatase, 5α-reductase, and5β-reductase, into estrogen, 5α-dihyrotestosterone, and 5β-dihydrotestosterone(Ball & Balthazart 2002; Balthazart & Adkins-


Regan 2002). The synthesized estrogen crossescell membranes and binds to estrogen receptorsinside the cells. The bound receptor acts as atranscription factor causing, amongst otheractions, changes in gene expression. Likewise,the androgen 5α-dihyrotestosterone crosses cellmembranes and binds to androgen receptorsinside the cell. The bound receptor then causesa different set of changes in gene expression,amongst other actions. The synthesized 5β-dihydrotestosterone is the inactive form of thetestosterone pro-hormone, and is treated as waste.

There are two sources of these hormones: thegonads and adrenals, and the brain itself(Holloway & Clayton 2001; Schlinger et al.2001). Testosterone is produced and releasedinto the blood stream by the testes in males andby the adrenal glands in males and females;

estrogen by the ovaries. Testosterone crosses theblood–brain barrier to act in the brain, asindicated, where it is converted into estrogen,in both males and females. Brain-synthesizedestrogen is released back into the blood streamat levels as high as those released from the gonads(Schlinger & Arnold 1992). Within the brain,distribution of hormone receptors and thehormone-synthesizing enzymes differs with thebrain region, sex, species, season, and stage ofdevelopment, somewhat like the glutamatereceptors. The androgen-synthesizing enzyme,5α-reductase, is present throughout much ofthe brain of songbirds, but the estrogen-synthesizing enzyme, aromatase, is selectivelyexpressed in NCM and related auditory areas.This is true of both male and female songbirds,whether or not females have vocal nuclei

BOX 33

PATHWAYS FOR HORMONAL INFLUENCE ON BIRDSONG

Birdsong is seasonal, and one of the earliest observations made by field biologists and natural historians is thatsinging behavior is positively correlated with various measures of reproductive physiology. Experiments establishedthat testosterone from the gonads enhances seasonal changes in song production. Both estrogenic and androgenicmetabolites of testosterone appear to be involved in these effects of testosterone on song. They exert theireffects on behavior by binding to intracellular receptors. Pioneering autoradiography studies in the early 1970srevealed that receptors for testosterone are found in several of the forebrain song control nuclei including HVC,RA and lMAN (Arnold et al. 1976). This observation was somewhat unexpected in that androgen and estrogenreceptors in the brain otherwise seemed to be restricted to the limbic regions and selected areas in the diencephalonand mesencephalon. This finding, along with the observation that the syrinx itself has androgen receptors(Lieberburg & Nottebohm 1979), suggested that testosterone might activate song by binding directly to receptorsin the vocal control system as well as in the syrinx. Anatomical studies employing either immunohistochemicalmethods for the localization of androgen receptor (AR) and estrogen receptor (ER) proteins as well as in situhybridization studies of the messenger RNA for AR and ER confirmed and extended the initial autoradiographicstudies (Balthazart et al. 1992; Bernard et al. 1999). However, at least in temperate zone birds, song is part ofa still larger suite of male reproductive behaviors, extending beyond the domain of the song system in the strictsense. In both birds and mammals the preoptic region of the brain is known to coordinate the effects of testosteroneon male sexual behaviors. In European starlings, lesions to the preoptic region block song behavior as well asother reproductive behaviors (Riters & Ball 1999). Catecholamine cell groups that project to the forebrain songcontrol nuclei are also known to express both androgen and estrogen receptors (Appeltants et al. 1999; Maneyet al. 2001). It is therefore possible that the effects of testosterone on song are at least in part mediated by itsaction in the preoptic region and/or in brainstem catecholamine cell groups that, in turn, project to forebrain songcontrol regions. Although it seems likely that direct action of testosterone in song control nuclei regulates thequality of song produced, testosterone effects on the motivation to sing may involve other parts of the brainas well.

Gregory F. Ball


(Metzdorf et al. 1999; Soma et al. 1999; Saldanhaet al. 2000). In the cerebral vocal pathways, onlyHVC has estrogen receptors. In adult canaries,the receptor levels are high during each breedingseason, whereas in zebra finches, they are presentonly once in life, during early juveniledevelopment, in the HVC of both males andfemales (Gahr & Konishi 1988; Gahr & Metzdorf1997; Jacobs et al. 1999). In adults of mostspecies, Area X lacks androgen receptors, whereasall pallial vocal nuclei contain androgen receptorsbut at different levels in different species (Jacobset al. 1999; Tramontin & Brenowitz 2000; Ballet al. 2002). These hormone and brain differenceshave a genetic basis.

In mammals, the XX chromosome pair makesa female and the XY pair a male. In birds, theopposite occurs; a ZZ pair makes a male and aZW pair makes a female. Thus, in mammals,the female is the ‘genetic default;’ a Y chromosomeis needed to make a male body from a female.In birds, the male is the genetic default, and aW chromosome is needed to make a female bodyfrom a male (Balthazart & Adkins-Regan 2002).Estrogen synthesized in the egg helps mold themale testes into female ovaries, a change probablyfacilitated by the W chromosome. In birds as inmammals, circulating testosterone converted toestrogen in the brain masculinizes the brain.Strong evidence for sex linkage came from thebrain of a zebra finch that was gynandromorphic,half male, half female (Box 34, p. 249).

Bringing this knowledge together, thefollowing tentative picture emerges (Fig. 8.7).During early development, an as yet unknowngenetic program that includes genes on the Zchromosome starts the formation of the cerebralvocal nuclei in both sexes. Sometime aroundhatching, circulating testosterone that enters thebrain is converted to estrogen via aromatase; theestrogen binds to its receptor in HVC, facilitatinggrowth of HVC and connecting nuclei. Thecirculating testosterone is also converted in thebrain to the androgen 5α-dihyrotestosteroneusing 5α-reductase, which then binds to itsreceptors in pallial vocal nuclei to modulate nottissue growth, but crystallization of connections

and later enhancement of singing output.For those species in which the females do not

engage in vocal learning, it is probable that geneson their W chromosome actively prevent orreverse the growth of cerebral vocal nuclei. Thisgenetic mechanism presumably evolved after theevolution of songbird vocal learning in tropicalzones, for species living thereafter in temperatezones. However, when these females are givenextra doses of estrogen early in life, as in thezebra finch, inhibition of vocal nuclei growth isovercome. In addition, if one of the two Zchromosomes of males is not fully inactivated,they have the potential like mammals tosynthesize extra doses of gene products responsiblefor the formation of vocal nuclei.

During the juvenile-to-adulthood transition,an androgen surge helps crystallize the song toadult form and readies the bird for breeding.There is a parallel with human puberty; largehormonal increases occur, an adult-like voicecrystallizes, and the young teenager, male orfemale, is biologically ready to have children.The underlying mechanisms in the two sexesare not the same, and in birds the sex differencesare not yet fully understood. In species whereonly males imitate song, female vocal nuclei havealready shrunk in size by this time; so thesefemales go through a puberty-like phase, butwithout engaging in vocal learning. In speciesthat learn new songs seasonally, a similar processreoccurs; song becomes plastic, learning occurs,and then testosterone products recrystallize thenew songs and ready the bird’s singing for theupcoming breeding season.

Androgens are thought to crystallize the voiceby binding to pallial vocal nuclei receptors,activating genes that stabilize synapses betweenvocal nuclei and helping the survival of newneurons arriving in HVC (Rasika et al. 1994;White et al. 1999). A castrated male gives plasticsong but fails to crystallize until given a seasonallyappropriate dose of testosterone (Marler et al.1988). Artificially high doses of testosterone injuveniles or in the plastic phases of adult singingalso induce song crystallization (Korsia & Bottjer1991; Nottebohm 1993). Androgens stimulate


BOX 34

WHAT MAKES A BIRD BRAIN FEMALE OR MALE?

Even casual observations of birds reveal a profound sex difference in the ability to sing. Song is a quintessentialexample of male sexual advertisement and, in many species, females sing little or not at all. The male’s greaterability is directly caused by a marked sex difference in the structure of the neural song circuit. In zebra finches,for example, several brain regions controlling song are about five times larger in males than in females. This largesex difference in brain structure, the first to be discovered in any vertebrate (Nottebohm & Arnold 1976), invitesan obvious question; what causes brain development to differ in the two sexes?

The first answer was suggested by studies in mammals, in which hormones secreted by the testes act on thefetus to make the brain male. The important hormones are testosterone and its metabolite, estradiol. Whenfemale zebra finches were treated with estradiol on the day of hatching, they developed a much more male-likeneural song circuit and a fairly good male song (Gurney & Konishi 1980). Androgens also seem to play animportant role because the masculinizing actions of estradiol are blocked by an anti-androgen, flutamide (Grishamet al. 2002). Thus, although estrogen induces a masculine pattern of development, it appears to require anandrogen-dependent step. These studies suggest that, in birds also, male sex hormones make a male brain.Some doubts about this idea arose, however, because blocking the action of these hormones does not preventa male from singing and developing a male-like song circuit (Arnold 1997). The hormonal theory was furtherchallenged when genetically female zebra finches, induced by a drug injection to develop large testes rather thanovaries, were found to have a feminine neural song circuit (Wade et al. 1999), contradicting the earlier findings.

An alternative idea is that the genetic sex of brain cells plays a role in determining whether the brain has maleor female characteristics. The strongest support for this view comes from the analysis of an unusual bilateral

A GynandromorphicZebra Finch

right left

gynandromorphic zebra finch that arose by mutation(Agate et al. 2003). This bird was genetically male onthe right half of its body, and genetically female on theleft half (see fig.). Male plumage covered only theright half of its body, and the right gonad was a testisand the left gonad an ovary. Amazingly, the neuralsong circuit was more masculine on the right half ofthe brain than on the left, indicating that the differencesin genetic sex on the two sides of the brain influencedtheir sexual characteristics. Although the origin of sexdifferences is not fully understood, both hormonal andintrinsic genetic factors appear to play critical roles.One intriguing, as yet unsupported, idea is thatgenetically male brain cells may themselves producemore sex hormones than female brain cells, and thatsex hormones originating in the brain are importantfor sexual differentiation as well as for the control ofadult behavior (Schlinger et al. 2001).

Arthur P. Arnold


singing, and a positive feedback mechanism isthought to be involved, with androgen bindingto vocal nuclei receptors enhancing singing, whichin turn enhances testosterone release by thegonads, which then encourages more singing(Fig. 8.7). This positive feedback loop can bebroken in the non-breeding season of seasonallysinging species, when androgen receptor levelsin the vocal nuclei are low, and the birds justsing for short 10-min bouts (Ball et al. 2002).As a possible mechanism by which androgenenhances singing, binding to androgen receptorsin pallial vocal nuclei neurons may reduce thethreshold for electrical activity, as occurs inmammals (McEwen 1994; Ramirez et al. 1996),making the neurons associated with singing moreready to fire. The general picture is that hormoneshave three types of facilitating effects on thesongbird vocal communication system: they helpform the system, they stabilize synapses andneurons involved in the learning and

crystallization of song, and they induce songproduction.

However, hormone receptors are not presentin the cerebral vocal nuclei of all vocal learners.The HVC-like vocal nuclei of hummingbirdshave high levels of androgen receptors, but noestrogen receptors (Gahr 2000). In the one speciesof parrots examined, budgerigars, there are noandrogen or estrogen receptors in their vocalnuclei (Gahr 2000). Only songbirds have higharomatase in their auditory regions (Metzdorfet al. 1999; Soma et al. 1999; Gahr 2000;Saldanha et al. 2000). Thus, it looks as thoughhormones may have different roles in the vocalsystems of different vocal learners.

ADULT NEUROGENESISDISCOVERED

One of the major outgrowths of these hormone

Possible Hormonal Mechanisms for Sex Differences in the Brain and Singing

Genes on Z &other chromosomesDevelopment

Adulthood

(A)

(B)

brain estrogen

X

hatching

song imitation

begin formation of cerebral vocal nuclei

enhance vocal nuclei growth and formation

in females of some species, W chromosomegene(s) possibly inhibits vocal nuclei growth

song crystallization adulthood

testis &/or brainandrogen surge

puberty-liketransition

stimulatesmore adult androgen enhanced song production & vocal nuclei size

stimulates

X

blocked by other factors andmay be seasonally controlled

Figure 8.7 Diagrams of possible interactions between (A) sex hormones, brain development, and (B) theformation and activation of the vocal learning system in youth and adulthood.


studies in songbirds was the discovery that theadult brain is able to make new neurons.

Nearly 100 years ago, Ramon y Cajal, an earlyNobel Laureate in neuroscience, stated that,unlike skin, which is constantly renewed, in theadult brain “the nerve paths are something fixed,ended, immutable. Everything may die, nothingnew may be rejuvenated” (Ramon y Cajal 1913).This view became dogma, epitomized in thestatement “do not destroy your brain cells becauseyou will not make any more.” Over the years,various attempts were made to challenge thisdoctrine by new methods, trying to induce newneuronal growth in adult brains, but withoutsuccess. The view prevailed that warm-bloodedvertebrates do not generate new neurons as adults,perhaps because it would be too costly to holdonto long-term memories (Rakic 1985). In the1960s, Altman (1962) found new neuronproliferation in the adult rat brain. However,his work met with strong criticism, and wasdifficult to repeat, discouraging further investiga-tion.

Song Plasticity and Rejuvenation of the Brain

Goldman & Nottebohm (1983) were startledwhen, twenty years later, as they studied therelationship between seasonal re-growth in thesize of HVC and hormones in adult canary brains,they found newly labeled neurons (Fig. 8.8).Nottebohm had found earlier that injection oftestosterone in female canaries resulted in a 90percent increase in the size of HVC. To identifydividing cells, the animals were injected withradioactively labeled thymidine, one of thenucleotides of DNA. When cells replicate, theradioactive nucleotide is taken up into the DNAof the new cell and can be detected. Goldmanand Nottebohm went to great lengths to provethat the new cells were actually neurons. Perhapsnot surprisingly, their results were met withskepticism, and at first were difficult to getpublished. Later Paton, Alvarez-Buylla, and Kirndecisively proved that the newly generated cellsin the songbird brain were indeed new neuronsthat were functional in the song system (Paton

New Neurons in the Adult Brain

Discovered in songbird HVC in 1983 Discovered in human hippocampus in 1998

FluroGold labels ‘neurons’ connected to another brain region.Tritiated thymidine, 3H-Thy, labels newly divided cells.

(A)

NeuN, a neuron-specific gene, identifies cells as neurons.Bromodeoxyuridine, BrdU, also labels newly divided cells.

(B)

Figure 8.8 (A) Evidence of newly formed neurons in the adult songbird brain from Alvarez-Buylla & Kirn1997. Left: canaries were injected with 3H-thymidine; after allowing new neurons to reach HVC, RA wasinjected with FluoroGold, which was taken up by the axons of HVC’s RA-projecting neurons. Right: at highpower, Flo-labeled cells containing 3H-Thy are seen. (B) Evidence of newly formed neurons in the adulthuman brain (Eriksson et al. 1998). Left: frontal sections of a human hippocampus labeled with a markerthat identifies the neuron-specific gene NeuN (dark staining areas). Right: high power of the granule cellularlayer (GCL) double-labeled with NeuN and bromodeoxyuridine (BrdU) showing a neuron born in the brainof this patient when he/she was an adult. Pictures kindly provided by John Kirn (A) and Fred Gage (B).


& Nottebohm 1984; Alvarez-Buylla et al. 19881990, 1992). Remarkably, neuron death andreplacement were found to occur in HVC, at ahigh rate on a daily basis, with nearly all of theneurons of a given cell type replaced each year(Kirn et al. 1991; Nottebohm 2002). Inretrospect, it is hard to imagine how neurogenesisin the adult brain was missed.

But the skepticism persisted. The vocal nucleiare highly specialized brain tissues, and it wasargued that perhaps new adult neurons are foundonly in these unusual structures. Alvarez-Buyllaand collaborators then showed that new neuronswere present throughout the songbird cerebrum,and in the cerebrums of vocal non-learning birdsas well (Nottebohm & Alvarez-Buylla 1993; Linget al. 1997). Then, using the same techniques,they found new neurons in the cerebrum of amammal, the mouse (Lois & Alvarez-Buylla1993; Lois et al. 1996; Doetsch et al. 1997),finally confirming Altman’s (1962) findings. Thenew neurons they found were mostly transportedto the olfactory bulb.

Skeptics retreated to another line of defense,arguing that the olfactory bulb is a primitivestructure and of little importance in holdingonto complex memories. However, around thisperiod, studies of mice using the same approachesas those used on songbirds, demonstrated theincorporation of new neurons in the mammalianhippocampus, a region involved in learning andmemory (Cameron et al. 1993; Gould &McEwen 1993). The numbers of new neuronsentering the hippocampus each day werestaggering, in the thousands, and weresignificantly reduced by stress and enhanced bylearning (Gould et al. 1999a; Gould & Tanapat1999). Most animals kept in cages are stressedand this may be one reason why adultneurogenesis was overlooked in the past. Gouldthen also found new neurons entering the adultprimate cortex, an area certainly responsible forcomplex learning (Gould et al. 1999b). Finally,Gage, Eriksson, and colleagues in 1998, studyingthe autopsied brains of patients who had beeninjected days or years earlier with a cell divisionmarker (bromodeoxyuridine) used to diagnose

the severity of their cancer, found evidence ofnew neurons in the human hippocampus thatsurvived in the patients until their death (Erikssonet al. 1998; Fig. 8.8). So, the 100-year-olddoctrine of no new neurons in adulthoodgradually crumbled (Specter 2001; Nottebohm2002), though still not without challenges(Kornack & Rakic 2001; Rakic 2002a, b).

The discovery of new neurons in the vocalnuclei of songbirds led Nottebohm to propose arole for them in the formation of new songmemories (Nottebohm 1984). Several findingsappeared to support this view. In canaries, oldneurons die and are replaced by new neurons ata higher rate during times of the year when thebirds learn new song syllables, after the molt inthe autumn and around the breeding season inApril to May (Kirn et al. 1994; Fig. 8.9).However, other studies seemed to challenge thisnotion. Zebra finches continue to show neuronaldeath and new neuron addition in HVC wellafter song learning is complete (Ward et al. 2001).Other species that undergo seasonal changes inbreeding and singing behavior accompanied bychanges in death and incorporation of new vocalnuclei neurons do not add new songs to theirrepertoires (Tramontin & Brenowitz 1999, 2000).

If there is a link to learning, it may be indirect,as suggested by Li and colleagues (2000), whofound that just the act of singing without anyevidence of learning new songs, drives theincreased survival of new neurons in HVC withina few days (Fig. 8.9), with the number of newlyarrived neurons proportional to the amount ofsinging. This motor-driven survival of newneurons adds further to testosterone-inducedenhancement of survival (Alvarez-Borda &Nottebohm 2002). Similarly, in the rat brain,wheel running for extended periods of timeincreases proliferation and incorporation of newneurons in the hippocampus (Van Pragg et al.1999). It seems that one of the most salientfactors influencing the addition of new neuronsis exercise, the act of performing a behavior ratherthan a need to incorporate new memories. Arelationship to exercise does not necessarilyexclude learning; however, the new neurons


Behavior, Dead Cells, and New Neurons (canary)

Dead cells

New syllables

New neu

rons

Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May

Seasonal changes

(A)

Changes with and without singing

# ne

w n

euro

ns p

er m

m3

800

600

400

200

0

Not-singing Singing

# ne

w n

euro

ns i

n H

VC

325

275

225

175

0In HVC In hippocampus # song bouts per hour

(B)

10 20 30 40 50

Figure 8.9 (A) Changes in cell death, song learning, and new neuronal survival across the year in brainsof males of a seasonal breeding species, the canary. Peaks in cell death are followed by peaks in new songsyllables and new neurons. Modified from Alvarez-Buylla & Kirn 1997, Kirn et al. 1994, and Nottebohm etal. 1986. (B) Singing-driven enhancement of new neuron survival in HVC. Left: canaries were eitherallowed to sing for 7 days, or prevented from singing by interrupting them. Singing increased the survivalof neurons that entered HVC but had no affect on new neurons that entered the hippocampus, a non-vocalarea. Right: the enhancement of new neuron survival is proportional to the rate of singing over a 30-dayperiod. This enhancement was independent of the gonads; these animals had their testes removed.Modified from Li & Jarvis 2001 and Alvarez-Borda & Nottebohm 2002.

presumably inherit or acquire the memoriesembodied in the old neurons.

Not all neuronal types are replaced. Neuronswith short-distance axonal connections, such aslocal neurons, are more likely to be replacedthan those that traverse long distances, such asthose from the cortex or pallium to the spinalcord. Neurons in the adult thalamus andmidbrain are not replaced (Nottebohm &Alvarez-Buylla 1993). Interestingly, the latter are

areas involved mainly in innate behaviors. Inthe songbird vocal system, only neurons of HVCand Area X are continually replaced in adulthood,and not those of lMAN and RA (Kirn et al.1999). The arcopallium in which RA residesgenerally does not undergo neuron replacementand, like RA, is the part of the avian brain thatsends long descending axons to the brainstemand spinal cord (Zeier & Karten 1971; Kirn etal. 1999). Within HVC, its RA-projecting


neurons are replaced throughout adult life, butits X-projecting neurons are not (Alvarez-Buyllaet al. 1988; Kirn et al. 1999). Using a laser-lesion procedure, Scharff et al. (2000) selectivelylabeled the RA-projecting and X-projectingneurons of HVC with a dye, which kills thecells when zapped with the laser at a particularwavelength of light. When they killed off the X-projecting neurons in adult zebra finches, songwas temporarily affected for a day or two andthen returned to normal. This is similar to whathappens when lArea X is lesioned. The birdsshowed no induced regeneration of the X-projecting neuron type in HVC. HVC justbecame smaller, due to the neuron loss. Whenthey killed off the RA-projecting neurons,however, the birds could not sing, as happenswhen the entire HVC is lesioned. After severalmonths, the birds slowly recovered the same ora very similar song to that produced before thelesion, and newly incorporated RA-projectingneurons were found throughout HVC. Thus,when damaged, the RA-projecting neuronpopulation of HVC can be restored, andremembered behavior can be recovered. Thisshows that the song template is not stored inthe RA-projecting neurons.

The discovery of neurogenesis in adult birdbrains helped to revitalize the field of adultmammalian brainstem cell research, and the searchfor new ways to repair brain injuries (Gage 2000).Although scientists have a long way to go, thereare exciting prospects. Adult brainstem cells ofboth birds and mammals have been found withinthe walls of the cerebral ventricles, in areas calledthe ventricular zones, which line the cavity wherethe cerebral spinal fluid flows (Doetsch et al.1999). The stem cells divide and give rise todaughter cells that then migrate to their destinationin the brain and become neurons. One of theproblems in taking advantage of this mechanismto induce repair of damaged brain tissue, is thatthere will be a need to control the incorporationof cells that normally get replaced, to induce andcontrol those not normally replaced, where theymigrate to, and their eventual connectivity. Vocallearning birds can serve as a useful animal model

for the development of these kinds of brain repairtechniques.

EVOLUTION OF VOCAL LEARNING:SONGBIRDS, HUMMINGBIRDS,AND PARROTS

Can vocal learning birds serve as useful animalmodels for language? To answer this questionwe need to gain a better understanding of theevolution of vocal learning. All songbirdsexamined, including birds that are not typicalsingers, such as corvids – crows, ravens, jays andmagpies – have cerebral vocal nuclei (DeVoogdet al. 1993; Brenowitz & Kroodsma 1996). Incorvids they are presumably used for learningand producing their complex calls and lowvolume songs (see Chapter 5). Thus, vocallearning was probably present early in songbirdevolution. How did this come about? Nottebohm(1972) proposed that songbirds, hummingbirds,and parrots evolved their vocal learning abilitiesindependently. His reasoning was that most oftheir close relatives, including the sub-oscinesongbirds (but see Chapter 3), are all vocal non-learners, and presumably represent the ancestralcondition for birds. After the discovery that someparrot vocal nuclei have similarities with thoseof songbirds (Paton et al. 1981), Brenowitz (1991)proposed that just as the behavior they controlis similar, the cerebral vocal nuclei of songbirdsand parrots evolved independently but withshared properties. In his detailed study ofbudgerigar vocal nuclei connectivity, Striedter(1994) gave parrot vocal nuclei different namesthan in songbirds, assuming that if they evolvedindependently, then they should not have thesame names. My colleagues and I did the samefor hummingbirds, and used molecular mappingof vocal areas to provide new perspectives (Jarviset al. 2000).

The late Luis Baptista first proved thathummingbirds are vocal learners (Baptista &Schuchmann 1990). He was a lover ofhummingbirds, and his work was in part the


inspiration for the direction of my own researchon vocal learning. Unlike songbirds andbudgerigars, hummingbirds are difficult to workwith. They are among the smallest and fastestflying birds. They do not breed readily in captivity,and they are very territorial and often kill eachother if they are kept in close quarters. Becauseof this, we took our molecular brain mappingexperiments into the field. In the Atlantic Forestregion of Brazil, near Santa Teresa, we found asite with one of the highest populations ofhummingbirds in the world. Unlike songbirdsand parrots, hummingbirds live only in theAmericas. They consist of two main lineages,the Trochilinae and the Phaethornithinae. Thetwo species we studied were the somberhummingbird, an ancient Trochilinid, and therufous-breasted hermit, an ancient Phaethor-nithinid (Box 35, p. 263). After we observed abird, either singing, or listening without singingfor 30 min in the morning, we attracted himinto a cage with a sugar water bottle, capturedhim, and examined ZENK expression in hisbrain. This approach, similar to the one we hadused earlier on songbirds and parrots (Jarvis &Mello 2000), helped us to identify vocal nucleithat were not revealed by other methods.

Comparisons of the ZENK expression patternsin the three taxa of vocal learners revealed someremarkable similarities. All had seven cerebralvocal structures that showed gene activationduring singing. We found three of the seven innearly identical brain locations, in anterior partsof the cerebrum, though they have differentshapes. In songbirds, these structures are part ofthe anterior vocal pathway. The remaining fourof the seven are all in different brain locationsin each of the vocal learning groups, but are stillwithin the same brain subdivisions relative toeach other. Two of these, the HVC-like and theRA-like structures have similar shapes. Insongbirds, these structures are part of the posteriorvocal pathway. All three vocal learning groupshad similar auditory areas, showing geneactivation after hearing species-specificvocalizations, all in the same relative brainlocations. The similarity of these auditory areas

is not surprising, since vocal non-learning birdswere already known to have such auditory regions(Wild et al. 1993). Interestingly, the location ofthe four posterior nuclei relative to the auditoryareas differs, in accordance with the relative ageof each order. Parrots are the oldest, and theirposterior vocal nuclei are far away from theauditory areas; hummingbirds are the next oldest,and their posterior vocal nuclei are closer andadjacent to auditory regions; songbirds are themost recent, and their posterior vocal nuclei areshifted further back, embedded within theauditory areas.

Lesion studies confirmed that the cerebral vocalstructures of parrots, at least those studied, arerequired for producing and learning vocalizations(Heaton & Brauth 2000a, b). Interestingly, lesionsin the parrot nucleus parallel to HVC, NLc,revealed that it is required for the birds to speakEnglish words (Lavenex 2000). This was the firsttime anyone had established the role of a structurein a non-human brain for the production ofimitated human speech. Electrophysiologicalstudies in anaesthetized budgerigars show thattheir vocal nuclei also have auditory responses,and this may turn out to be a basic feature ofvocal learners. Neither lesioning norelectrophysiological studies have yet beenconducted in the vocal nuclei of hummingbirds,though there have been preliminaryhummingbird connectivity studies (Gahr 2000).

The comparisons show that all three vocallearner groups have similar posterior vocalpathways, connecting a region of the nidopalliumto the arcopallium, then to the tracheosyringealmotor neurons, and then to the syrinx and theabdominal respiratory muscles. Connectivity ofthe anterior vocal nuclei has only been studiedin the songbird and the parrot. Both have similaranterior vocal pathway connectivity, forming apallial–striatal–thalamic–pallial loop (Striedter1994; Durand et al. 1997). Where theconnectivity differs greatly, is in the connectionsbetween their posterior and anterior vocalpathways. In the parrot brain, the anterior vocalpathway does not receive its posterior pathwayinput from the HVC-like structure, as occurs in


BOX 35

COMPLEX SONGS WITH A SMALL BRAIN: HUMMINGBIRDS

Hummingbirds have many fascinating characteristics, including one of the highest brain-to-body size ratios inexistence, and a large species-specific diversity of vocalizations, some of which display local dialects (Wiley1971; Snow 1973; Stiles 1982; Rehkaemper et al. 1991; Gaunt et al. 1994; Ventura & Takase 1994; Vielliard1994; Rusch et al. 1996; Schuchmann 1999; Ficken et al. 2000). Given Luis Baptista’s demonstration of vocallearning in hummingbirds (Baptista & Schuchmann 1990), we wonder how such a small brain, the size of the tipof a man’s pinky could do what many animals with bigger brains cannot. With the discovery of the hummingbirdcerebral vocal system, of a kind not found in non-vocal learning animals (Jarvis et al. 2000), the answer appearsto lie in the patterning of neuron connectivity rather than absolute brain size. We have been studying two species,the sombre hummingbird (A) and the rufous-breasted hermit (B). Sombre hummingbird song has a relativelystereotyped and rhythmic syntax, like a zebra finch, but the syllables are complex and two-voiced, with rapidfrequency modulations (CD #—). Rufous-breasted hermit song has a complex syntax, consisting of rollingsequences of increasing and decreasing pitches and modulations, generating a great variety of different syllables(CD #—). Like songbirds, these species appear to produce their songs in affective contexts, either in an undirectedmanner or directed to the opposite sex during courtship (Ferreira et al. 2003). A wealth of information may begained by studying the brains of the smallest birds in existence.

Erich D. Jarvis & Adriana R. Ferreira

(A) (B)


songbirds. Instead, part of the posterior vocalpathway’s RA-like nucleus projects into the twoanterior vocal nuclei, Mo-, and MAN-like nuclei.In the parrot, the output of the anterior pathwaydoes not have well-separated medial and lateraldivisions. Instead, the same region of the MAN-like nucleus projects to both its RA-like and itsHVC-like vocal nuclei. Since connectivity of thesongbird Mo nucleus has not been yetdetermined, a complete comparison with parrotsis not possible (but see below).

In parrots and songbirds auditory pathwaysfrom the ear to the auditory pallium (NCMand other areas) are very similar to those in vocalnon-learning birds. Thus, auditory pathwayconnections appear to be an ancient inheritance.However, there are differences in the mode ofinteraction of the auditory and vocal pathways.Besides the major difference in location of theposterior vocal nuclei relative to the auditoryregions, the parrot vocal pathways receive auditoryinput from two different auditory pathways: oneis like that in songbirds, from Ov of the thalamusto field L2 to L1 and L3; the other is from amore frontal pathway, less well characterized,that involves a projection from the laterallemniscus intermediate (LLI) nucleus of thehindbrain to nucleus basorostralis. In parrots,both pathways, L1 and L3, and nucleusbasorostralis then project to the NIf-like vocalnucleus, which then projects into the parrot nucleiof the posterior (HVC-like) and anterior (Mo)pathways.

By combining the connectivity and geneexpression findings for all three vocal learninggroups, we get an indication of some basicrequirements for avian vocal learning. There areseven cerebral vocal nuclei organized into aposterior vocal pathway that projects to the lowervocal motor neurons and an anterior vocalpathway that forms a loop and is involved inother aspects of vocal learning. Anotherrequirement is that every major cerebralsubdivision except the pallidum and hyperpalliumhas at least one vocal nucleus. This in turn suggestsa basic requirement for avian cerebrally controlledbehaviors where, as in mammals, each cerebral

function involves some part of a six-layered cortexplus the basal ganglia (Swanson 2000a). Variationseems to be permissible, in the shapes of thevocal nuclei, the absolute brain location of theposterior vocal pathway relative to the auditorypathway, the connectivity between the two vocalpathways, and in the connectivity between thevocal and auditory pathways. There are alsovariations in the relative sizes of vocal nuclei.The parrot Mo-like and MAN-like vocal nucleiare relatively much larger than their songbirdand hummingbird counterparts; like canaries,with a relatively large MAN to Area X ratio,parrots are thought to display more vocal plasticitythan most songbirds or hummingbirds.

Three Evolutionary Hypotheses

Now that we have all of this information inhand, we can begin to answer questions aboutthe evolution of vocal learning brain areas inbirds. Modern birds are said to have evolvedfrom a common ancestor sometime around thecretaceous–tertiary boundary at the time of theextinction of dinosaurs (Feduccia 1995; Fig.8.10). How did seven similar brain structureswith somewhat similar connectivity evolve inthree distantly related vocal learning bird groupsin the past 65 million years? To put the questionin perspective, the phylogenetic distance betweenparrots and songbirds (Sibley & Ahlquist 1990)is as great as that between humans and dolphins(Novacek 1992; Fig. 8.11). To explain the sharedsimilarities, three hypotheses have been proposed(Jarvis et al. 2000).

Hypothesis 1: Three out of the 23 avian ordersevolved vocal learning independently (Fig 8.10,solid circles). Each time, they evolved sevensimilar brain structures to serve the purposes oflearned vocal communication. This would suggestthat the evolution of brain structures for vocallearning, and for complex behavior in general,is under strong epigenetic constraints. If true,then a similar scenario can be made for the vocallearning mammals: humans, cetaceans, and bats(Fig. 8.11, solid circles). The evolution of wings


provides an analogy. Wings evolvedindependently at least four times, in birds, bats,pterosaurs (ancient flying dinosaurs), and insects.In each case, they evolved at the sides of thebody, usually one on each side, and not one onthe head, the other on the tail, or elsewhere.One hypothesis is that wings evolved in similarways because of a strong constraint, the centerof gravity of the body, dictating the mostenergetically efficient manner for flight.According to this logic, one can predict that if,in another half a million years or so, pigeons

were to evolve vocal learning, then they too wouldhave seven similar brain regions in well-definedlocations.

Hypothesis 2: An alternative hypothesis is thatthere was a common avian ancestor with vocallearning, possessing the seven cerebral vocalnuclei. These traits were only retained in thethree current vocal learning orders and lost atleast four times independently in the interrelatedvocal non-learning orders (Fig. 8.10, opencircles). Repeated losses would suggest that

Brain Structure and the Phylogenetic Relatedness of BirdsAVIAN FAMILY TREE 23 ORDERS

~65 MYA to the PRESENT

Three alternative hypotheses3 independent gains4 independent losseseverybody has it to varying degrees

* Vocal learners

Struthioniformes (ostrich)Tinamiformes (tinamous)

Craciformes (scrubfowl)Galliformes (chickens, quail)Anseriformes (ducks, geese)

Turniciformes (buttonquails)

Piciformes (toucans, woodpeckers)

Galbuliformes (puffbirds)Bucerotiformes (hornbills)Upupiformes (hoopoes)Trogoniformes (trogons)Coraciiformes (rollers)

Coliiformes (mousebirds)Cuculiformes (cuckoos, roadrunners)Psittaciformes (parrots)*

Apodiformes (swifts)Trochiliformes (hummingbirds)*Musophagiformes (turacos)Strigiformes (owls)

Columbiformes (pigeons, doves)Gruiformes (cranes, trumpeters)Ciconiiformes (shorebirds, birds of prey)Passeriformes (songbirds/suboscines)Passeriformes (songbirds/oscines)*

Figure 8.10 A phylogenetic tree of living birds according to DNA relationships. The Passeriform order isdivided into its two suborders, suboscine and oscine songbirds. Lines at the root of the tree indicate extinctorders. Open and closed circles show the minimum ancestral nodes where vocal learning could haveevolved or been independently lost. Vocal learners are in bold. Modified from Sibley & Alhquist (1990).


maintenance of vocal learning and the underlyingcerebral vocal structures may be under anothertype of epigenetic constraint; there wouldpresumably be considerable costs to retainingvocal learning or evolving in adaptivecircumstances that did not require vocal learning.If the losses occurred independently, several times,then again a similar scenario can theoreticallybe advanced for the vocal learning mammals:humans, cetaceans, and bats (Fig. 8.11, opencircles). However, chimpanzees and otherprimates would have had to lose the trait recently,multiple independent times.

Hypothesis 3: A third hypothesis is that avianvocal non-learners have some rudimentary systemof cerebral vocal nuclei that scientists have missedpreviously, and these systems were independentlyamplified in the vocal learners. If true, this wouldpresent a challenge to the hypothesis that cerebralvocal nuclei are unique to vocal learners. It impliesthat all birds have at least the primordia for thenecessary brain structures, with the potential forvocal learning to varying degrees. A similarscenario could be argued for mammals andperhaps all advanced vertebrates; they all haveat least the primordia for vocal cerebral brainstructures and for vocal learning, includingchimpanzees, lions, tigers, and bears, to varyingdegrees, and that these were amplified in thevocal learners.

Whichever hypothesis is correct, singly or incombination, the answer will be fascinating. Theyall suggest that the evolution of brain pathwaysfor complex behaviors is constrained by factorsas yet unknown. All lead to the same prediction,that vocal learning mammals, including humans,may have evolved vocal brain pathways withfeatures held in common with vocal learningbirds.

PARALLELS BETWEEN BIRDS ANDHUMANS?

It has been appreciated for some time that thevocal learning behavior of songbirds shares

developmental characteristics with humanlanguage learning (Thorpe 1961; Marler 1970b).It may seem bizarre to even suggest that birdbrains can teach us something new about howhuman brains work. But in an era whengeneticists have shown us how profitable it canbe to move from humans to fruit flies to bacteriaand back, I am encouraged to offer somesuggestions about how human/avian comparisonsmight proceed.

The precise relationships between songbirdand human vocal brain regions remainunexplored. The omission exists in part becauseof the erroneous historical belief that the avianbrain is one large basal ganglion, that humansare much more special than they really are, andbecause of the obvious limitations on theexperimental study of humans. Now that wehave a new picture of the relationships betweenavian and mammalian brains (Jarvis et al. underreview; Reiner et al. in press B) and given ourexpanding knowledge of the brain pathwaysfor learned vocal communication amongstdistantly related birds, the time may be ripe fora reappraisal of the neurobiology of humanlanguage as viewed from a comparativeperspective.

For comparisons to be productive, sometranslations need to be made between theterminologies used in bird and human vocalcommunication research (Doupe & Kuhl 1999).We can separate human vocal communicationinto three categories: comprehension, speech,and language. Comprehension is the perceptionand understanding of language; speech is theproduction of language sounds, includingphonemes and words; and language is thesyntactical sequencing of these sounds intomeaningful phrases and sentences. The term‘language’ has also been applied to non-vocalcommunication, such as reading and signing.Most neurobiologists interested in language havestrived to project these behavioral terms ontodifferent brain structures, comprehension andlanguage to the cortex, in Wernicke’s and Broca’sareas, and speech to lower brainstem motor areas,the periaqueductal gray and 12th nucleus (Hollien


Figure 8.11 The mammalian family tree and the evolution of vocal learning. The phylogenetic tree ofliving mammals is based on relationships compiled by Novacek (1992). Except for primates, the Latinname of each order is given with examples of some common species. Vocal learners are asterisked. Openand closed circles show the minimum ancestral nodes where vocal learning could have either evolved orbeen independently lost. Independent losses would, at a minimum, have required a common vocal learningancestor node located by the arrow. Within the primates, there would have to be at least 6 independentlosses (tree shrews, prosimians, new and old world monkeys, apes, and chimps), and one loss (treeshrews), followed by a regain of vocal learning (humans). This assumes that all other primates are vocalnon-learners. Compare with Fig. 8.10.

Phylogenetic Relatedness of Mammalian Vocal Learners

Mammalian family tree

Mesozoic140 130 120 110 100 90 80 70 60 50 40 30 20 10 0

27 orders

Cenozoic

Triconodonts

Multituberculata

Monotremata

Marsupalia

Edentata

Pholidota

Lagomorpha

Rodentia

Macroscelidea

Primates/humans

Scandentia

Dermoptera

Chiroptera

Insectivora

Carnivora

Artiodactyla

Cetacea

Tubulidentata

Perissodactyla

Hyracoidea

Proboscidea

Embrithopoda

Desmostylia

Sirenia

Creodonta

Condylarthra

Commonancestor

Three alternative hypotheses3 independent gains>7 independent losses

— everybody has it to varying degrees

Pulaeoryctoids

myr


1975; Geschwind 1979; Fitch 2000a; Kent 2000),though linguistic definitions can vary widelydepending upon the subfield or the investigator.In songbird research, there is no equivalentdistinction between language-like and speech-like properties. One word, ‘song,’ represents bothproduction of learned sounds and theirsequencing. In this manner, the linguisticdefinition of ‘language’ for humans is at leastsomewhat similar to the avian neurobiologists’definition of ‘song;’ the linguistic definitions of‘speech’ and ‘singing’ for humans are similar tothe avian neurobiologists’ definitions of ‘call andsong production.’ Using the rules of avianneurobiologists, speech and singing combinedwould be called ‘language production.’ Ratherthan projecting predefined behavioral terms ontodifferent brain structures, however, students ofbirdsong have tended to project definitions inthe opposite direction, from brain structures tothe behaviors they appear to control.

The terminology for behavioral deficitsresulting from brain lesions also varies from onescientist to another. For humans, we tend tospeak of dysarthria (slurred speech), verbal aphasia(poor syntax production), and auditory aphasia(poor language comprehension; Benson & Ardila1996). For songbirds, the equivalent terms arenot well formalized, but they include songdegradation, disrupted syllable structure,disrupted sequence or syntax production, andimpaired song discrimination, the latter beingequivalent to a comprehension deficit. With thebenefit of translations like this, the newcomparative brain nomenclature, and a detailedliterature review (Jarvis 2001), I propose thefollowing speculative thesis on the neurobiologyof human language.

We can think of humans, like vocal learningbirds, as having three basic cerebral pathwaysfor vocal behavior, one posterior, one anterior,plus an auditory pathway; together they areresponsible for human vocal learning, speech,and singing. The posterior vocal pathway beginswith facial motor cortex in the cerebrum and isthe only one in which some connections havebeen determined. Using an old tract-tracing

method, staining stroke-induced degeneratingaxons from autopsied human brains, Kuypers(1958a) showed that the human face motor cortexprojects directly to the lower motor neurons thatcontrol vocalizations, the nucleus ambiguous.As the tracheosyringeal nucleus in birds projectsto the syrinx, so the nucleus ambiguous inmammals sends axons that control muscles ofthe larynx. However, a connection from the cortexto the nucleus ambiguous is lacking in our closestrelative, the chimpanzee, and in other non-humanprimates (Kuypers 1958b; Jürgens 1995) as wellas pigeons and other vocal non-learning birds(Wild et al. 1997a). Thus, these properties ofthe human face motor cortex are analogous tothose of the pallial nuclei HVC- and RA-likecombined, in vocal learning birds (Fig. 8.12).

I propose that the human anterior vocalpathway consists of the classical Broca’s area ofthe cerebral cortex, plus an entire band of cortexon either side of Broca’s area. Human patientssuffering damage in these areas can produce thesounds of language, but they have difficultylearning new speech sounds and in producingcorrect syntax (Benson & Ardila 1996). Inhumans, these two deficits are identified as poorrepetition and verbal aphasia; in songbirds theequivalent terms used are poor imitation ordisrupted song learning and syntax deterioration.Thus, the properties of this strip of cortex, forwhich I have suggested the name ‘the languagestrip’ (Jarvis 2001), most resemble those of acombination of the avian pallial MAN-like andMo-like vocal nuclei. Damage to two other majorbrain regions in humans also leads to languagedeficits in imitation and syntax, withouteliminating production entirely, thoughsometimes the effects are worse than those afterdamage in the language cortex. These are theanterior portion of the human striatum and aregion of the anterior human dorsal thalamus(Damasio et al. 1982; Naeser et al. 1982; Graff-Radford et al. 1985; Alexander et al. 1987; Fig.8.12). The worst deficits may perhaps occurbecause, as in the brains of non-human mammals,connections from the cortex converge in thestriatum, and then into the thalamus,


concentrating function into smaller and smallerareas (Parent & Hazrati 1995). These propertiesof the human anterior striatum and thalamusare analogous to those of the striatal Area X-likeand thalamic DLM-like vocal nuclei of vocallearning birds. My interpretation is that thehuman anterior pathway, consisting of a languagecortex strip, anterior striatum, and anterior dorsalthalamus, is connected in a cortical–basal ganglia–thalamic–cortical loop, much as in non-vocalareas of non-human primates (Parent & Hazrati1995) and in the anterior vocal pathways of

vocal learning birds.I suggest that the human auditory pathway

for language comprehension is a sensoryperceptual learning pathway with projectionsfrom the hair cells in the ear that reach themidbrain (the inferior colliculis), the thalamus(the medial geniculate), the primary auditorycortex, and the secondary auditory cortex(Wernicke’s Area). All mammals, birds, andreptiles examined have a similar set of connections(Carr & Code 2000). When tested, the cerebralpart of this pathway in various species has proved,

Figure 8.12 A speculative comparison of left hemisphere brain organization and connectivity of vocal andauditory brain areas of parrots, hummingbirds, humans and songbirds. White arrows indicate proposedanterior vocal pathways and black arrows indicate proposed posterior vocal pathways. Connectivity wasextrapolated from the following sources: songbird (cited in Fig. 8.2 legend); parrot (Paton et al. 1981;Brauth et al. 1987, 1994, 2001; Brauth & McHale 1988; Striedter 1994; Durand et al. 1997; Farabaugh &Wild 1997; Wild et al. 1997b); hummingbird (Gahr 2000). Connectivity in hummingbirds is not known, andhumans is conjectured, or based upon findings in non-human mammals and vocal learning birds (Kuypers1958a).


as in songbirds, to be tuned to the perception ofspecies-specific sounds (Wang et al. 1995).

There is an additional parallel in the occurrenceof cerebral dominance. As mentioned earlier, inhumans and in many songbirds, the left side ofthe brain is dominant over the right in vocalcommunication. According to my proposedmodel, the left human posterior vocal pathwaycontrols the production of speech phonemes andwords and the left human anterior vocal pathwaycontrols the sequencing of phonemes into wordsand words into sentences. The human vocalpathways on the right side control singing in asimilar manner. When the left language areasare damaged, humans have trouble producingor sequencing speech (aphasia), but when theright language areas are damaged they havetrouble producing and sequencing singing(amusia; Wertheim 1969; Benson & Ardila1996). Like vocal learning birds, there is a sexdifference in the language areas of humans, butin the reverse direction. Women, although leftdominant, are more prone than men to utilizeboth sides of their brains for speech and language;they have more connections between the twohemispheres, a higher density of language areasynapses, and larger language brain regions thanmen (Shaywitz et al. 1995; Witelson et al. 1995;Harasty et al. 1997). Such differences are thoughtto explain why, from a young age, women havebetter language skills than men (Bradshaw 1989;Mann et al. 1990; Halpern 1992). These braindifferences between the sexes, although lessextreme than in many birds, are also thought tobe under the influence of testosterone andestrogen (Kimura 1996).

From this strictly anatomical point of view,and leaving aside all of the cognitive requirementsfor language, I offer the speculation that the‘main’ difference between the vocalcommunication systems of humans and vocallearning birds lies, not in the presence or absenceof particular pathways, or in their pallialorganization, but in the relative sizes of structuresthat make up their vocal pathways. In the anteriorvocal pathway, humans appear to have a muchlarger pallial representation (MAN- and Mo-

like regions) relative to the striatum (Area X-like region), compared with vocal learning birds.Such a huge size difference is consistent withthe prodigious variability and complexity ofhuman vocal behavior that appears to be greaterthan all of the vocal learning birds combined.The pallial organizational differences betweenmammals and birds, a layered cortex in humansand a globular pallium in birds (Fig. 8.1), mustmean that there is more than one type of brainorganization that can produce vocal learningbehavior.

One wonders then how humans, members ofa relatively young, approximately 120,000-year-old species, evolved brain pathways for learnedvocal communication that appear to be somewhatsimilar to those used by the very distantly relatedvocal learning birds? As far as their auditorypathways go, birds and humans probablyinherited them from a common reptilian-likeancestor. The general ability to engage in auditorylearning can explain why dogs, chimpanzees,and other animals are able to acquire anunderstanding of human speech sounds. Dogspresumably have a Wernicke’s-like area forlearning to process complex sounds.

With regard to the vocal pathways, humansobviously must have evolved their cerebralvocal pathways independently of a commonancestor with birds and reptiles, given thatvocal learning is so rare. This conclusion callsfor some caution, given the occasional reportsof other animals, such as seals, chimpanzees,Japanese macaques, and prairie grouse said tohave imitated either human speech or calls ofwhat were thought to be a different dialect oftheir species (Hayes & Hayes 1951; Sparling1979; Ralls et al. 1985; Masataka & Fujita 1989;Perry & Terhune 1999). But we need moreexperimental study before these anecdotes canbe properly assessed; most have been difficult, ifnot impossible, to replicate (Owren et al. 1993;Marler 1999).

With independent evolution as the mostplausible hypothesis, then the similarity of thevocal pathways of humans and vocal learningbirds would suggest that they evolved under some


type of environmental, epigenetic constraint. Oneadvantage of learned vocalizations is the abilityto adapt sounds for more efficient communicationin different environments (see Chapter 6).Humans are thought to have evolved in the EastAfrican Rift Valley where within limited areasyou can find tropical forest, savannah, and desert.Perhaps vocal learning evolved in both birds andhumans to cope with diverse environments,favoring diverse vocal behavior, exploiting thepotential of brain pathways, based upon a designthat was already part of the basic vertebrate plan.To make the necessary changes in the brain, amotor system would be needed for producinglearned movements, with posterior and anteriorpathway components already present, to bedirected to the lower motor neurons that controlthe muscles of the vocal organs; once connectedto a sensory pathway for hearing, this couldbecome a recipe for vocal learning.

CONCLUSIONS

Scientists have a come a long way in their studiesof brains and birdsong. As we acquire newinformation, the ideas and concepts presentedhere will be subject to change. Some of theexciting advances we can anticipate will comefrom deciphering the role that the anterior vocalpathway plays in learned vocal communication.Another profitable area will be the role of replayedelectrical activity in the brain associated with

singing during sleep. We still have much to learnabout the mechanisms of auditory–vocalintegration and the search for a template. Thediscovery of new neurons in the adult brain hasrevolutionary implications for medical science.The molecular biology of vocal learning is helpfulin understanding genetic mechanisms of behavior,and in resolving the great mystery of how vocallearning evolved. Some areas as yet unexploredinclude the study of vocal brain areas in theother mammalian vocal learners, cetaceans, andbats. The extensive knowledge we now have aboutvocal learning in birds may provide a useful guidein how best to approach the study of thesemammalian vocal learners, though cetaceans willalways be a challenge. New techniques mayemerge for exploring brain connectivity andbehaviorally-driven gene expression in humanbrains in an ethically responsible manner, thoughit is not yet clear how best to proceed. Studentsof both human and non-human vocal behaviorneed to share mindsets and develop a sharedvocabulary, if the translation of discoveries fromone field to the other is to be facilitated. This isespecially important if research on vocal learningbirds is to be of any help in understanding theunderpinnings of language in the human brain.Last, but not least, we must not lose sight of thecontinuing need for studies on the peripheralstructures, the beak, the syrinx and its muscles,which, like our lips, tongue, and throat, play akey role in the physical production of learnedsounds.

brains and birdsong - duke university

Documents