AUDITORY-VOCAL INTEGRATION IN THE SONG SYSTEM

R. Mooney Æ M.J. Rosen Æ C.B. Sturdy

A bird’s eye view: top down intracellular analysesof auditory selectivity for learned vocalizations

Received: 5 October 2001 /Revised: 5 January 2002 /Accepted: 10 June 2002 / Published online: 24 October 2002� Springer-Verlag 2002

Abstract The ‘‘song system’’ refers to a group of inter-connected brain nuclei necessary for the utterance oflearned song and for the generation of vocal plasticityimportant to both song learning and adult song main-tenance. Although song learning and, in some species,song maintenance depend on auditory feedback, howaudition influences vocalization remains unknown. Oneattractive idea is that auditory signals propagate directlyto those telencephalic nuclei implicated in song pat-terning, providing a convenient substrate for sensori-motor integration. Consistent with this idea, auditoryneurons highly selective for the bird’s own song havebeen detected in telencephalic song nuclei, and lesions ofthese structures can impair song perception as well assong production. This review discusses evidence for anauditory-perceptual role of the song system, the ana-tomical pathways by which auditory information entersthe song system, the synaptic events underlying highlyselective action potential responses to learned song, andthe possible roles such activity could play in songlearning and maintenance.

Keywords Error correction Æ HVc Æ Inhibition ÆSong learning Æ Song selectivity

Abbreviations AFP anterior forebrain pathway Æ area Xarea X of the lobus parolfactorius Æ BOS: bird’s ownsong Æ BOSrev reverse BOS Æ BOS-RS reverse syllableorder BOS Æ CON conspecific song Æ DLM medialnucleus of the dorsolateral thalamus Æ EPSP excitatorypostsynaptic potential Æ field L field L of Rose Æ HETheterospecific song Æ HVc used here as the propername Æ IPSP inhibitory postsynaptic potential ÆLMAN lateral part of the magnocellular nucleus of the

anterior neostriatum Æ mMAN medial part of themagnocellular nucleus of the anterior neostriatum ÆnAm nucleus ambiguus Æ NIf nucleus interfacialis ÆnRAm nucleus retroambigualis Æ nXIIts tracheosyringealportion of the hypoglossal nucleus Æ Ov nucleusovoidalis Æ RA robust nucleus of the archistriatum ÆUva nucleus uvaeformis Æ WN white noise

The song system: a circuit for singing and listening

In addition to the song system’s obligatory role insinging, several studies point to its importance to theperception of conspecific vocalizations. Such a dual rolefor circuitry controlling learned vocalizations is notunique to songbirds. In humans, speech production andperception have been localized to separate brain regions(Broca’s and Wernicke’s areas, respectively), yet there isconsiderable overlap in function, such that damage toBroca’s area induces perception as well as productiondeficits (e.g., Carpenter and Rutherford 1973). In asimilar vein, a link between production and perceptionhas been demonstrated in songbirds, in that selectivelesions in circuits essential to singing also disrupt songperception.

Neural circuits important to song include two func-tionally distinct yet interconnected pathways emanatingfrom the song nucleus HVc (Fig. 1A, B) (Nottebohmet al. 1976; Fortune and Margoliash 1995; Foster andBottjer 1998). One pathway projects to vocal and re-spiratory areas used for singing while the second path-way projects through the anterior forebrain andinfluences song learning and maintenance. Several lesionstudies implicate both branches of the song circuit inconspecific song perception, and hint that this percep-tual role is most pronounced in those songbirds thatactually produce song. For example, male zebra fincheswith anterior forebrain pathway (AFP) lesions areslower in the acquisition of a discrimination between thebird’s own song (BOS) and other conspecific songs,while the capacity for heterospecific (canary) song

J Comp Physiol A (2002) 188: 879–895DOI 10.1007/s00359-002-0353-3

R. Mooney (&) Æ M.J. Rosen Æ C.B. SturdyDepartment of Neurobiology,Duke University Medical Center,Durham, NC 27710, USAE-mail: mooney@neuro.duke.eduFax: +1-919-6844431

discrimination remains intact (Scharff et al. 1998).Similarly, even partial lesions of HVc lead to conspecificsong discrimination deficits in both sexes of starlings (aspecies in which both sexes sing) (Gentner et al. 2000);female canaries, which also are capable of singing (whentreated with testosterone), lose their conspecific songpreference following HVc lesions (Brenowitz 1991), andare impaired on several auditory discrimination tasksfollowing lesions of the lateral part of the magnocellularnucleus of the anterior neostriatum (LMAN) (Burt et al.2000). These results in singing birds contrast with those

from species in which females cannot sing: female zebrafinches maintain their conspecific song preference inspite of HVc lesions (Macdougall-Shackleton et al.1998). The dichotomous nature of deficits observed inbirds that can produce learned song and those thatcannot raises the likelihood that circuits used for singingare also exploited for auditory processing important tosong perception.

Why the perception and processing of conspecificsong would differ depending on whether the song circuitis capable of producing song is not readily apparent.However, clear behavioral evidence exists that the songproduced by this neural circuit is developed and main-tained through auditory experience. At the most basiclevel, auditory experience of others’ as well as one’s ownsong is needed for normal song development (Konishi1965; Price 1979). Birds acoustically isolated from thetime of hatching develop aberrant isolate songs, whiledeafening after exposure to a tutor also derails normalsong development, pointing to the necessity of auditoryfeedback. An auditory feedback role is also important tothe maintenance of adult song, at least in some songbirdspecies: both adult deafening and artificially distortedauditory feedback imposed during singing perturb pre-viously stable adult song in zebra finches (Nordeen andNordeen 1992; Leonardo and Konishi 1999). Finally,auditory information can be seen to acutely influence theamount and type of song production in the wild, asterritorial male song sparrows alter their song output inan attempt to match the song types most recentlyemitted by neighboring males (Stoddard et al. 1992).Such an important role for audition in vocal develop-ment and performance may require the song system tofunction in a dual auditory-vocal capacity.

Indeed, the song system is likely to be a site of au-ditory-vocal integration necessary for song plasticity andmaintenance. One idea is that song is learned andmaintained by an error correction process, in which thebird uses auditory feedback to actively match its vocaloutput with a memorized model (Marr 1969; Knudsen1994; Troyer and Doupe 2000a, 2000b). Such error

Fig. 1 A A sagittal schematic of the neural circuit for songacquisition and production. Nuclei essential for the patterning andproduction of learned song include nucleus interfacialis (NIf), HVc,robust nucleus of the archistriatum (RA), nucleus ambiguus(nAm)/nucleus retroambigualis (nRAm) and tracheosyringealportion of the hypoglossal nucleus (nXIIts) (shown in white). Theanterior forebrain pathway (AFP; light gray) is critical foraudition-guided vocal plasticity in both juvenile and adult zebrafinches, and contains area X of the lobus parolfactorius (area X),the medial nucleus of the dorsolateral thalamus (DLM), and thelateral part of the magnocellular nucleus of the anterior neostri-atum (LMAN). Note that separate populations of HVc neuronsproject to RA and area X, respectively. Auditory structures (darkgray) include the thalamic nucleus ovoidalis (Ov) and the primaryauditory thalamo-recipient zone of the telencephalon, field L;ascending auditory pathways are depicted with a broken line toindicate that intermediary structures have been omitted for the sakeof simplicity. Field L of Rose (field L) may communicate with HVcvia NIf and via regions immediately ventral to HVc (the shelf, darkgray). Some structures have been omitted for clarity; D dorsal, Rrostral. B An equivalent flow chart diagram of the song system,with the AFP shown in gray. C A schematic of auditory and non-auditory afferents to HVc. In birds, auditory information istransmitted from the thalamic nucleus ovoidalis to the telence-phalic auditory region field L, which has three subdivisions, L1–L3(Wild et al. 1993). Many field L1 and L3 axons terminate in a‘‘shelf’’ region extending 500 lm from the HVc ventral border,which may provide one site of auditory input to HVc (see text); NIfis likely to provide auditory afferents to HVc. Field L mayinnervate NIf directly, and indirectly through the caudal hyper-striatum ventrale (cHV, not shown; Vates et al. 1996a). Dashedlines indicate putative connections. The auditory properties of themedial part of the magnocellular nucleus of the anterior neostri-atum (mMAN) and nucleus uvaeformis (Uva) neurons await morecomplete descriptions (see Vates et al. 1997; Williams 1989). Thesyrinx is the avian song organ

880

correction could occur online, as the bird sings, or off-line, when the bird is not vocalizing (perhaps even dur-ing sleep) (Dave and Margoliash 2000). In one onlinemodel (Troyer and Doupe 2000a, 2000b), auditoryfeedback is proposed to drive an instructive signal thatactually guides the production of subsequent songsyllables based on the prior sequences heard.

Whether online or offline, the anatomical organiza-tion of the song system supports the idea that it functionsin part for error correction: LMAN (the output of theAFP) is essential to audition dependent vocal plasticity(Bottjer et al. 1984; Brainard and Doupe 2000) andsynaptically communicates with the robust nucleus of thearchistriatum (RA), the song system’s sole telencephalicoutput to the brainstem vocal and respiratory areas usedin singing (Fig. 1A, B). RA is implicated in the tempo-rally precise, low-level patterning of song and receivessynaptic input fromHVc, an area essential for generatingor relaying higher-level song patterning (Vu et al. 1994;Yu and Margoliash 1996). Such convergent organizationmay make RA a pivotal site where LMAN and HVcterminals can interact to influence song quality.

An essential role for LMAN in audition-dependentvocal plasticity gains support from a wide variety ofbehavioral experiments performed in juvenile and adultsongbirds. In juvenile zebra finches, LMAN lesions ar-rest song development, resulting in prematurely stereo-typed and abnormally impoverished songs (Bottjer et al.1984; Scharff and Nottebohm 1991). Even in adult zebrafinches, LMAN is implicated in audition-dependentvocal stability. Although adult zebra finch song nor-mally remains highly stereotyped, deafening still leads toa slow and progressive deterioration of song’s acousticstructure (Nordeen and Nordeen 1992), analogous tospeech deterioration in deafened humans (Waldstein1990). This deafening-induced song degradation can beblocked by prior bilateral LMAN lesions (Brainard andDoupe 2000), suggesting that auditory feedback activelymaintains the stable song of the adult zebra finch. Oneinference is that LMAN provides a permissive or in-structive signal that can generate high levels of songvariability during juvenile song learning and perhapssmaller amounts of variability in adult song that can beexploited for ongoing error correction. LMAN is wellsuited to provide such signals because its activity ismodulated both by singing and hearing song (Doupeand Konishi 1991; Vates et al. 1996a; Hessler and Doupe1999), and because it trophically regulates the synapsesthat HVc neurons make with RA (Kittelberger andMooney 1999), a synapse which is likely to directlyinfluence song structure.

While the relationship between neuronal activity inthe song system and singing has grown increasinglyclear, the neuronal correlates underlying song perceptionremain obscure. Both chronic recordings and microsti-mulation experiments support the idea that HVc andRA form a hierarchical pattern generator for song(McCasland 1987; Vu et al. 1994; Yu and Margoliash1996). A perceptual function of song system neurons is

unknown but may be reflected in one of their mostfascinating properties, that of selectivity for the BOS(Fig. 2 and 6). Electrophysiological recordings from avariety of song nuclei [e.g., HVc, RA, the tracheosyrin-geal portion of the hypoglossal nucleus (nXIIts), area Xof the lobus parolfactorius (area X), the medial nucleusof the dorsolateral thalamus (DLM) and LMAN] inawake and urethane-anesthetized birds have detectedstrong neuronal responses to song playback, where theBOS functions as the most potent stimulus (McCaslandand Konishi 1981; Williams and Nottebohm 1985;Margoliash 1986; Doupe and Konishi 1991; Vicario andYohay 1993; Volman 1996; Yu and Margoliash 1996;Dave et al. 1998; Schmidt and Konishi 1998; Theunissenand Doupe 1998; Hessler and Doupe 1999; Mooney2000; Lewicki and Arthur 1996). Such selective neuronalresponses are well suited for auditory feedback func-tions, and also could underlie the perception of con-specific song. In support of this latter idea, male zebrafinches discriminate conspecific songs more quicklywhen one song is their own (Cynx and Nottebohm1992), while surgically-muted juvenile male zebra finchesdevelop into adults that lack the conspecific song pref-erences seen in finches that can sing (Pytte and Suthers1999). A useful step will be to test whether auditoryresponses in any song nucleus, and especially BOS-selective responses in HVc, are evoked during songdiscrimination tasks.

One attractive idea is that an equivalent motor andauditory code for song could enable auditory feedbackto be compared directly to the code for motor output inthose areas essential to song patterning (i.e., HVc andRA). In both vocalizing and listening states, neuronalactivity in several song nuclei, including HVc, RA andLMAN, is tightly correlated with the acoustic structureof the song (McCasland and Konishi 1981; McCasland1987; Yu and Margoliash 1996; Hessler and Doupe1999; Dave and Margoliash 2000). An intriguingequivalence in auditory and vocal motor activity actu-ally has been detected with chronic recordings made inthe RA of unrestrained zebra finches, either duringsinging or playback of the BOS. Most notably, patternsof activity observed in RA during singing are similar tothose displayed by the same neurons while the bird ex-posed to playback of its own song during sleep (Daveand Margoliash 2000). The finding that BOS playbackgenerates neuronal activity highly similar to that seenduring singing may reflect an equivalent coding of au-ditory and vocal motor song representations and raisesthe intriguing possibility that BOS playback can be usedto reveal circuit dynamics relevant to motor as well asauditory functions of the song system.

Sources of auditory input to the song system

Despite the longstanding observation that song nucleipossess both vocal motor and auditory attributes, sig-nificant gaps persist in our knowledge of the anatomical

881

pathways via which auditory information enters thesong system. In this review, we focus on auditory inputsto the song nucleus HVc, because of its potential im-portance as a site for auditory-vocal interaction, andbecause both anatomical and functional data supportthe idea that HVc serves as the sole source of auditoryinput to the AFP, including LMAN, and to the nuclei ofthe descending motor pathway, including RA andnXIIts. Indeed, all of these areas lose their auditory-evoked responses upon HVc inactivation (Williams andNottebohm 1985; Doupe and Konishi 1991; Vicario andYohay 1993), further underscoring the likelihood thatauditory-vocal integration occurs via HVc and notthrough other auditory pathways. Elucidating the exactsources and types of auditory information received byHVc, and the relationship between auditory activity inHVc and other song nuclei, will provide greater insightinto the mechanisms of auditory-vocal integration andthe auditory processing of conspecific communicationsounds.

Prior to discussing auditory selectivity in the songsystem, it is worthwhile to consider song as an auditorystimulus. Complex vocalizations, including speech and

Fig. 2 In vivo intracellular recordings can be used to studysubthreshold and suprathreshold properties of identified neuronsin the song system. These subthreshold and suprathresholdacoustically-evoked responses in identified HVc projection neuronswere recorded intracellularly in an individual adult male zebrafinch. Responses of an RA projecting cell (top) and area X-projecting cell (bottom), are shown to six different stimuli (stim;BOS, bird’s own song; BOSrev reverse BOS; BOS-RS reversesyllable order BOS; CON conspecific song; WN white noise; HETheterospecific (e.g., swamp sparrow) song; 20 iterations each).Suprathreshold (action potential) responses are shown as cumula-tive peristimulus time histograms (spikes/bin; 25-ms bin width) andsubthreshold responses are shown as median-filtered averagedmembrane potential records [Vm (mV); see Mooney 2000]. In theRA-projecting cell, all stimuli evoked large and prolongeddepolarizing subthreshold responses, although only the BOS andBOS-RS evoked suprathreshold activity. In the area X-projectingneuron, subthreshold depolarizing responses were either morephasic (BOS, BOS-RS, CON) or largely absent (BOSrev, WN,HET); unlike the RA-projecting cell, hyperpolarizing responsesalso were evoked by the BOS and BOS-RS. Similar to the RA-projecting cell, suprathreshold responses in the area X-projectingneuron were evoked by the BOS and the BOS-RS (note also anonset response occurs to the swamp sparrow song)

882

birdsong, typically vary in their pitch and loudness overtime, i.e., display frequency and amplitude modulations(FM and AM). An important characteristic of such atime-varying signal is that its total spectral energy isidentical to its time-reversed counterpart; this spectralinvariance can be exploited to test whether a neuron’sresponse to a sound is triggered by the sound’s time-varying features, or instead by spectral components in-dependent of their temporal contexts. In the songbird, auseful approach has been to assess a neuron’s temporalsensitivity by contrasting its responses to forward andreverse playback of the BOS, a behaviorally relevantsound highly effective at eliciting action potentialresponses from song system neurons. In addition,higher-order temporal sensitivity has been probed bycomparing a neuron’s responses to the BOS and reversesyllable BOS (reverse syllable BOS contains each syllableplayed in the forward direction but in an overall reversedsequence). Differential neuronal responses to forwardversus reverse syllable BOS point to temporal integra-tion across syllable boundaries, whereas different re-sponses to forward and reverse BOS may simply reflectlow level temporal sensitivity to certain AM or FMelements in a single syllable.

The use of song playback is a two-edged sword:although behaviorally relevant and capable of drivingthe highly selective auditory neurons found in the songsystem, birdsongs often are acoustically complex. Thus,an important goal is to determine which acoustical fea-tures of song are salient to song-responsive neurons;although beyond this review’s scope, the sound synthesisand reverse correlation techniques essential to this un-dertaking are now being applied in the songbird’s au-ditory forebrain (Theunissen et al. 2000; Sen et al. 2001).In addition, although syllable sequence is often learned,and thus neuronal sensitivity to such sequences likelyarises via experience, the same may not hold for neu-ronal sensitivity to forward over reverse BOS, becausebirdsongs could exhibit temporal asymmetries indepen-dent of learning. Indeed, temporal asymmetries arecommon to a wide variety of musical sounds, and thesharp attack of a plucked (guitar or harpsichord) orstruck (piano) string sound highly unnatural whenplayed in reverse. With respect to birdsong, recentspectro-temporal analyses of zebra finch songs detect aconsistent temporal asymmetry, prominent in the AMdomain, which may be a species-wide feature that dis-tinguishes conspecific songs from those of other species(Theunissen et al. 2001). Selectivity for such species-typical features may emerge quite early in the auditorysystem, perhaps not surprising given how importantdecoding communication sounds can be to the individ-ual’s reproductive success. Indeed, a response bias forconspecific vocal sounds has been detected in the frog’sauditory nerve (Rieke et al. 1995) and in the zebrafinch’s auditory midbrain (Woolley and Casseday 2001).Despite the likelihood that certain biases toward con-specific vocal sounds are established prior to the songsystem, important insights have been gleaned by using

birdsong playback to probe the song system and itspotential auditory afferents.

As previously mentioned, HVc and other song systemneurons are highly BOS-selective, raising the question asto where such selectivity arises. Indeed, two features thatdistinguish HVc and other song system neurons frommost neurons in the primary auditory telencephalon arean action potential responsiveness to the BOS but notother conspecific songs, and an enhanced temporalsensitivity, evinced as a marked increase in the propor-tion of neurons showing a preference for forward overreverse and reverse syllable BOS (Fig. 2; Lewicki andArthur 1996). An especially daunting challenge to find-ing how and where such selectivity arises is the com-plexity of HVc inputs: HVc is the site of pronouncedanatomical convergence, and more than one of its manyafferents may have an auditory function (Fig. 1C). Thetype and source of auditory information entering HVcremain unknown, despite many elegant anatomicalstudies of HVc connectivity (Nottebohm et al. 1976,1982; Fortune and Margoliash 1995; Vates et al. 1996a;Foster et al. 1997). This ignorance persists because of thehighly divergent and convergent connections betweenHVc and its pre- and postsynaptic partners: HVc’s twosets of efferent axons (RA- and area X-projecting) in-termingle with three (and possibly four) different sets ofafferents [arising from nucleus interfacialis (NIf), nu-cleus uvaeformis (Uva), medial part of the magnocellu-lar nucleus of the anterior neostriatum (mMAN) andpossibly field L of Rose (field L); Fig. 1], several ofwhich have auditory properties. The resulting fibers ofpassage problem cannot be surmounted solely withconventional extracellular tract tracing or electricalstimulation techniques (Fortune and Margoliash 1995;Vates et al. 1996a), but instead also will require intra-cellular approaches.

A significant challenge is that HVc may receive song-related auditory input from several sources (see Fig. 1,legend), including the avian primary auditory telen-cephalon (field L) and NIf, a structure anatomicallyembedded in field L and synaptically interposed betweenfield L and HVc (Fig. 1C) (Wild 1994; Fortune andMargoliash 1995). Although NIf is immediately presy-naptic to HVc, debate remains as to whether field Lneurons also innervate HVc directly. Contrary to thisnotion, the earliest study of field L axonal projections insongbirds revealed their heavy termination in a ‘‘shelf’’region extending 500 lm from the HVc ventral border(Fig. 1) (Kelley and Nottebohm 1979; Fortune andMargoliash 1995; Vates et al. 1996a), rather than withinHVc itself. Certain HVc neurons extend their dendritesinto dorsal regions of the shelf, providing a potentialmeans for auditory information to enter HVc (Fortuneand Margoliash 1995; Vates et al. 1996b; Benton et al.1998), but it is unknown whether field L axons directlysynapse onto HVc dendrites in the shelf, indirectly in-nervate HVc via shelf interneurons, both, or neither. Inany case, such an indirect pathway between field L andHVc would only provide auditory input to a minority of

883

HVc neurons; other HVc neurons necessarily wouldhave to derive their auditory input from HVc neuronscontiguous to the shelf, or from other sources altogether.

The shelf also complicates attempts to map connec-tivity between field L and HVc using conventional tract-tracing techniques: tracer injections made into HVcretrogradely label some field L neurons (Fortune andMargoliash 1995), but the HVc shallow dorso-ventralprofile makes spillover to the shelf difficult to avoid,raising the possibility of false-positive retrograde label-ing. Functional studies will help clarify field L-HVcconnectivity: a recent intriguing hint of monosynapticconnections stems from cross-correlation analyses ofsimultaneous extracellular recordings made in field Land HVc, which reveal very short latency field L-leadingpeaks (Shaevitz and Theunissen 2001). This finding isconsistent with a monosynaptic projection from field Lto HVc, but false positives could arise if common syn-aptic inputs to the two sites are filtered through differentpostsynaptic membrane time constants, thus generatingstaggered spike onset latencies, or if the common inputsexhibit contrasting transmission times (i.e., due to con-trasting diameters of different branches of a parentaxon). If some field L neurons do innervate HVc, a re-maining issue is the degree to which these field L neuronsexhibit selectivity for the BOS, and the amount of in-dividual variation in their response properties. Althoughfield L neurons as a population are not BOS-selective,responding equally vigorously to forward and reversesong playback, a small minority responds more stronglyto forward song, and an even smaller group discrimi-nates forward over reverse syllable BOS (Lewicki andArthur 1996). Despite their forward BOS bias, these fieldL neurons differ from most HVc neurons in displayingsustained action potential activity above spontaneouslevels to the temporal BOS variants. Such field L neu-rons may reflect a stage of song processing intermediatebetween that of most field L neurons and those in HVc,but it is unknown whether any of these BOS-selectivefield L neurons innervate HVc, even indirectly. An im-portant goal will be to better understand whether andwhich individual field L neurons synaptically contactHVc, which perhaps will be most readily accomplishedby intracellular staining of individual field L neuronsthat have been well characterized in terms of their songselectivity, and by probing the synaptic connectivitybetween field L and HVc in brain slice preparations.

Perhaps the leading candidate to supply HVc withauditory input is NIf, as it is immediately presynaptic toHVc and most of its neurons show a response bias to theBOS intermediate between that of field L and HVc(Janata and Margoliash 1999). Unlike field L, the ter-mination of NIf axons in HVc is well-established; similarto field L neurons, NIf neurons fire in a sustainedmanner to reverse BOS (Janata and Margoliash 1999),and thus are well-suited to drive the sustained synapticpotentials actually seen in HVc in response to thisstimulus (Fig. 2; also see Mooney 2000). As with field L-HVc recordings (and subject to the same qualifications),

cross-correlograms of NIf and HVc BOS-evoked spiketrains reveal coincident or NIf-leading positive peaks(Janata and Margoliash 1999); the latter signature isconsistent with an excitatory monosynaptic projectionfrom NIf to HVc, while coincident peaks raise the pos-sibility that NIf and HVc receive a common auditoryinput. Common sources of input to HVc and NIf includethe thalamic nucleus Uva (McCasland 1987; Wild 1994),a structure with auditory properties yet to be fullycharacterized, and perhaps field L, although its con-nections to NIf are even less well established than thoseit may make with HVc (but see Fortune and Margoliash1995).

Certainly, one way to better understand the nature ofany extrinsic contributions to HVc song-selectivity couldbe to monitor HVc auditory responses when either NIfor field L is inactivated temporarily. However, the re-sults of such experiments may be difficult to interpretbecause: (1) in field L, selective neurons are embeddedwithin a larger population of non-selective cells; (2) fieldL may supply auditory input to both NIf and HVc; (3)NIf is interposed between different laminae of field L,making selective inactivation of NIf difficult to achieve;and (4) NIf may be one of only several auditory inputsto HVc. As with mapping of field L connectivity, in vivointracellular recordings could be used to characterize theauditory responses of NIf neurons that have axonalterminals in HVc; simultaneous intracellular recordingsin NIf and HVc would also help better resolve the timingof their activity and the sign of their connections.

Ultimately, a major obstacle to understanding thesource of HVc auditory responses is the lack of an ex-plicit wiring diagram for this nucleus. For example, thehighly selective BOS responses seen in HVc could resultfrom the anatomical ‘‘pipelining’’ of a small minority ofhighly BOS-selective afferents, or instead may arise vialocal refinement (i.e., thresholding by HVc neurons) ofmore broadly responsive inputs. Distinguishing betweenthese two ideas requires knowing which of many can-didate auditory afferent neurons actually synapticallycommunicate with HVc, as well as the sign of theirconnections. Although a bottom-up approach will con-tinue to help identify candidate auditory inputs to HVcand other song nuclei, moving forward in part will re-quire that these approaches be accompanied by a par-allel assessment of cellular connectivity based onanatomical (intracellular staining, for example) andfunctional (cross-correlation and synaptic physiology,for example) methods.

A bird’s eye view of song selectivity: top down,intracellular approaches

In synaptically connected neurons, the action potentialin a presynaptic neuron can be regarded as its ‘‘output,’’and the synaptic response it evokes in the postsynapticcell can be thought of as ‘‘input.’’ Measuring the output-input relationships of neuronal circuits constitutes a

884

major challenge in understanding the genesis of highlyselective sensory responses, and more generally in as-sessing the transfer function of a given brain region,such as the song nucleus HVc. The challenge stems fromthe difficulty in measuring functioning synaptic inputsonto neurons, which are often subthreshold, and thusundetectable in extracellular records. Although extra-cellular recordings are undoubtedly useful, they can onlyassess a neuron’s output; the input a given neuron re-ceives must be inferred, which is difficult in areas such asHVc that receive highly convergent input. Therefore,given the uncertainty about the auditory nature and signof HVc’s many inputs, a useful approach is to measurethe auditory-evoked synaptic input received by HVcneurons.

We have used in vivo intracellular recordings andsong playback to probe the synaptic events underlyingsong-evoked firing in identified HVc neurons (Fig. 2)(Mooney 2000). Such an approach enables intracellularstaining, which can determine which morphological celltypes respond to auditory stimuli, and can detect audi-tory-evoked activity in the absence of action potentialdischarge, which is a great aid in the song system, wheresuprathreshold responses often are elicited exclusively bythe BOS. Another advantage of an intracellular ap-proach is that it can help assess the sign of song-evokedsynaptic activity, yielding important clues about func-tional connectivity within and between song nuclei, in-dependent of their auditory role. Furthermore, whencoupled with pharmacological inactivation of the localcircuit, extrinsic and local contributions to a neuron’sauditory responses can be unmasked, giving insight intolocal circuit functions (Ferster et al. 1996; Rosen andMooney 2000a). Indeed, the subthreshold responses ofan impaled cell can be used to indirectly measure theaction potential activity of those neurons providing itwith synaptic input. In this sense, intracellular recordsprovide not only single unit data, but also informationabout the population of cells presynaptic to the recordedcell. Ultimately, intracellular methods provide a meansof unfolding the network from the postsynaptic celloutwards to its presynaptic partners, allowing mea-surements of the response properties of those neuronssynaptically communicating with the impaled cell.

HVc’s pivotal role in singing combined with its cel-lular heterogeneity make it an essential site to probe atan intracellular level. Unlike many other song nuclei,HVc possesses not one but two anatomically and func-tionally distinct projection neuron types. One type in-nervates the song motor nucleus RA, providing apathway by which HVc influences song patterning, whilethe other cell type innervates a basal ganglia-like struc-ture (i.e., area X) within the AFP (Fig. 1A, B; Alvarez-Buylla et al. 1988; Nixdorf et al. 1989). HVc also isthe probable site for important motor and auditorytransformations related to birdsong, as suggested by itsposition in the song patterning hierarchy and by theheightened auditory selectivity HVc displays at thepopulation level versus its extrinsic afferents (Lewicki

and Arthur 1996; Yu and Margoliash 1996). Indeed,such transformations could be mediated by the laby-rinthine local HVc circuit, which consists of both dedi-cated local interneurons and the extensive axoncollaterals elaborated by both projection neuron types(Mooney 2000).

To more precisely understand the neural basis of songlearning, two important and related goals are to un-derstand the degree and type of auditory-evoked activityin each HVc cell type. First, a clearer picture of whereauditory-vocal integration can occur requires knowingwhether area X- and RA-projecting neurons displayauditory-evoked activity. Second, local circuit mecha-nisms important to HVc function can be detected bycomparing a single cell’s subthreshold (synaptic) andsuprathreshold (action potential) responses to auditorystimuli, by measuring subthreshold response patternsbefore and after local circuit inactivation, and bysearching for specific interneuron-projection neuron in-teractions. Finally, clarifying how and when signalspropagate from HVc to RA and to the AFP requires anassessment of the timing of suprathreshold activity inidentified HVc projection neurons.

An intracellular analysis of song-selectivityin identified HVc neurons

HVc’s various cell types are electrically and morpho-logically distinct, consistent with their functional spe-cializations (Dutar et al. 1998; Kubota and Taniguchi1998; Mooney 2000). Using in vivo intracellular re-cordings, we relied on both morphological and electricalproperties to identify cells, and found that all HVc celltypes show robust, song-selective action potential dis-charge (Fig. 2; Mooney 2000). Such widespread audi-tory responses contrast with prior in vivo intracellularstudies that identified auditory responses only in areaX-projecting cells (Katz and Gurney 1981; Lewicki1996), and instead extend earlier extracellular studies(Doupe and Konishi 1991; Vicario and Yohay 1993)suggesting that HVc can relay auditory and vocal motorinformation directly to RA. Beyond providing anotherpath in addition to the AFP for conveying song-relatedauditory information from HVc to RA, BOS-responsiveRA-projecting cells could serve as a cellular substrate forauditory-vocal integration, at least if these same cellsalso exhibit song motor activity. This appears to be areasonable assumption, because all RA-projecting cellswe have encountered respond to BOS playback, butdistinct populations of auditory and purely motorRA-projecting neurons could exist.

The two separate (area X- versus RA-projecting)auditory representations that emerge from HVc couldfunction in a comparator circuit, either within HVc orwhen these two paths converge in RA. A comparatorfunction might be expected to involve one path trans-mitting a prediction of the sound expected from certainvocal gestures and the other conveying the actual sound

885

produced. In this latter case, we might expect certaindifferences in the auditory activity evoked in the twoHVc projection cell types, with one cell type displayingless selective responsiveness to acoustical stimuli. Such adistinction is not apparent at the suprathreshold level inurethane-anesthetized birds, as both projection cell typesdisplay highly similar action potential responses to BOSplayback (Mooney 2000).

Despite their similar suprathreshold profiles, mark-edly different subthreshold events, suggestive of diver-gent synaptic processes, underlie the song-evoked,highly phasic firing patterns in the two projection neu-ron types in HVc, suggesting that HVc potentially cantransmit different kinds of activity to RA and the AFP(Fig. 2; Mooney 2000). Specifically, BOS playbackevokes sustained depolarizing (i.e., subthreshold excit-atory) responses in RA-projecting neurons but elicitsmore phasic depolarizations alternating with prolongedhyperpolarizations in area X-projecting cells. In contrastto the phasic action potential responses of both projec-tion cells, interneurons discharge action potentials ton-ically throughout playback, with their highest firingrates closely coinciding with maximum membranepotential negativity in area X-projecting cells. Thesefindings suggest to us that a specific interneuron-area X-projecting cell interaction generates the subthresholddifferences seen in the two projection neurons types andhints that one function of the HVc local circuit is togenerate patterns of song-evoked inhibition in areaX-projecting cells. Important issues remaining to be re-solved include the functional implications of such dif-ferences for HVc and its downstream targets, namelyRA and the AFP, and the degree to which the sub-threshold responses, particularly in RA-projecting cells,reflect the action potential activity in HVc auditoryafferents.

The relative absence of auditory evoked inhibition inRA-projecting neurons, specifically in contrast to thepronounced BOS-evoked inhibition seen in areaX-projecting cells, may permit distinct kinds of auditorysignaling through these two HVc outputs. AlthoughRA-projecting neurons are highly selective and generateaction potentials almost exclusively to BOS playback,they also show subthreshold depolarizing responses toreverse BOS, as well as to many other song and non-song stimuli (Fig. 2). The non-BOS stimuli can driveRA-projecting neurons to spike when otherwise sub-threshold tonic positive current is injected through therecording electrode (Mooney 2000), showing that thesecells can relay information about a wide variety ofacoustical signals, at least when sufficiently depolarized.Therefore, factors which positively shift the restingmembrane potential of the RA-projecting cell type, suchas the augmentation of endogenous tonic excitatorydrive onto these cells that might occur during singing orcertain other behavioral states, could enable them torelay information about sounds that the bird emits re-gardless of whether they closely match the BOS. Incontrast, inhibition onto area X-projecting cells may

ensure a more absolute response specificity to the BOS,regardless of excitatory tone. Outside of this morespeculative functional contrast, a striking feature of thetwo cell types is the remarkable similarity of their su-prathreshold output in the face of quite different syn-aptic underpinnings. Certainly, an amazing and as yetpoorly understood feature of the HVc circuit is thethresholding mechanism that enables RA-projectingneurons to fire almost exclusively (and even then quitesparsely) to the BOS, despite receiving marked sub-threshold excitatory input activated by many othersounds.

The subthreshold excitatory patterns of RA-project-ing neurons can reveal much about the source and na-ture of HVc auditory afferents. Ultimately, the diversesubthreshold responses seen in RA-projecting cells in-dicate that they receive more extensive auditory infor-mation than is revealed in their firing patterns, and thatthe presynaptic neurons that innervate these RA-pro-jecting cells fire action potentials in response to a widevariety of stimuli besides the BOS. Despite beingbroadly responsive to songs and other sounds, thispresynaptic population is nonetheless likely to be BOS-selective, because the subthreshold responses seen inRA-projecting cells are strongly biased to the BOS(Mooney 2000).

The population of auditory cells presynaptic to RA-projecting neurons could be located in one or more sitesin or outside of HVc. One way to assess the contributionfrom extrinsic sources is to record from RA-projectingneurons and reversibly inactivate the local HVc circuit,as can be achieved by focally applying concentratedGABA (Rosen and Mooney 2000a, 2000b). Synapticresponses that persist in GABA are assumed to be drivenby neurons located outside of HVc, specifically becausethe saturating GABA concentrations potently activateGABA receptors on HVc neurons, shunting their so-matic and dendritic membranes and blocking theirability to generate synaptically-driven action potentials.Our initial observation using this approach is that al-though both HVc projection neuron types and inter-neurons are inactivated by GABA application, sustainedsynaptic responses to forward and reverse BOS can stillbe detected in RA-projecting cells during inactivation,and these responses remain BOS-selective, being largerin amplitude to forward than reverse BOS (Rosen andMooney 2000b). One conclusion to draw from thesefindings is that, as a population, HVc auditory afferentsare BOS-selective and fire throughout song playback.Individual NIf neurons display exactly such qualities(Janata and Margoliash 1999), making them leadingcandidates to supply RA-projecting neurons with audi-tory input.

The contrasting song-evoked subthreshold responsesin the two HVc projection cell types also could provideimportant clues about HVc local circuit function. Oneidea is that RA- and area X-projecting neurons receivecommon excitatory drive originating outside of HVc,and that the observed subthreshold differences in the

886

two cell types arise through local inhibition restricted toarea X-projecting cells (Fig. 3, left). This idea gainssupport from the close concordance between interneu-ron firing rates and periods of maximal inhibition in areaX-projecting cells (Mooney 2000), but the absoluteconnectivity between these cell types remains to be es-tablished. Alternatively, RA- and area X-projectingneurons could receive different patterns of extrinsicallygenerated auditory-evoked excitation (Fig. 3, right), anidea bolstered by the finding that area X-projecting cellslargely lack reverse BOS responses altogether, a featurefurther distinguishing them from RA-projecting neurons(Mooney 2000). These two models predict differentoutcomes of local HVc inactivation: in the former case,forward BOS-evoked responses in area X-projectingcells should become exclusively depolarizing andbroader in their duration, while reverse BOS-evokeddepolarizations should appear, thus resembling re-sponses seen in RA-projecting neurons. In the lattermodel, area X-projecting cells should retain highlyphasic depolarizing responses evoked only by the for-ward BOS. We presently are testing the effects of GABAinactivation on the response profile of area X-projectingcells to determine whether the two projection cells re-ceive common excitatory drive from auditory afferentsfrom outside HVc, and whether the local inhibitorycircuit sculpts the response patterns of area X- project-ing cells. Despite skillful efforts to unravel the HVc localcircuit (Dutar et al. 1998; Schmidt and Perkel 1998),many important issues concerning HVc inhibitionremain to be addressed, including the number of HVc

interneuron types and the sources of their inputs, whichcertainly include the axon collaterals of RA-projectingneurons (see below), but could also include input fromHVc afferents, such as NIf.

Possible functions of auditory-evoked inhibition in HVc

An obvious question is the functional significance ofauditory-evoked inhibition in area X-projecting neu-rons. Four ideas are that this inhibition serves to: (1)create more temporally precise or exclusive responsive-ness to the BOS, by sculpting more prolonged BOS-evoked excitation and/or suppressing responses to otherstimuli; (2) facilitate syllable combination sensitivity,through postsynaptic priming; (3) regulate activitywithin the AFP, perhaps by disinhibition of the outputof the AFP, the nucleus LMAN; and (4) generate avocally-triggered cancellation signal for error correctionpurposes (Figs. 4, 5). These possibilities are not mutu-ally exclusive, but the latter two hinge on the idea thatthe HVc circuit may function to convey alternate or‘‘mirror’’ images in response to the same stimulus.

The heightened sensitivity to syllable combinationsdistinguishes HVc from its auditory afferents, suggestingthat this feature is an emergent property of the HVcnetwork (Lewicki and Arthur 1996). Although all areaX-projecting HVc neurons show a strong suprathresholdbias to forward over reverse BOS, a smaller subsetshows absolute combination sensitivity, firing actionpotentials only in response to certain syllable sequencespresent in the BOS (Margoliash 1983). Pioneering in-tracellular studies of combination sensitive HVc neuronsshowed that inhibition is a key feature of their responseto preferred syllable sequences (Lewicki and Konishi1995; Lewicki 1996). Given that BOS-evoked inhibitionis detected in all area X-projecting cells (Mooney 2000),regardless of their combination sensitivity (R. Mooney,personal observations), one idea is that inhibitionactively masks more prolonged excitatory responsesactivated by the BOS, such as actually detectedin RA-projecting cells (Fig. 4, lower left). In thismodel, inhibition would tightly regulate when all area

Fig. 3 A circuit diagram depicting two possible mechanisms forthe different subthreshold response patterns observed in RA- versusarea X-projecting neurons. One possibility (left) is that anexcitatory extrinsic auditory afferent common to both projectioncell types is further modified by local circuit inhibition, sculptingthe response to BOS and non-BOS stimuli in area X-projecting cells(see also Fig. 4). Another idea is that the two projection cellsreceive distinct types of auditory input arising from structuresoutside of HVc (right). In the model shown at left, the inhibitoryinterneuron ‘‘sees’’ input from extrinsic sources indirectly, throughan interposed layer of RA-projecting cells; other possibilitiesinclude the extrinsic input directly activating the interneuronpopulation

887

X-projecting cells fire to the BOS, and also act in com-bination-sensitive cells to completely suppress actionpotential responses to non-preferred syllable sequences.Selective intracellular blockade of BOS-evoked inhibi-

tion in area X-projecting neurons, as has been accom-plished in orientation-selective neurons in themammalian visual cortex (Nelson et al. 1994), could beuseful to further explore the degree to which inhibitionsculpts excitatory responses to the BOS and accounts forabsolute combination sensitivity.

Another idea well suited to explain some aspects ofcombination sensitivity is that inhibition evoked by thefirst syllable in the preferred sequence actively ‘‘primes’’the postsynaptic membrane, perhaps by de-inactivatinga low threshold calcium conductance in the areaX-projecting neuron (Kubota and Saito 1991). Suchpriming could enable the cell to fire action potentialbursts when subsequently excited by the second syllablein the pair (Fig. 4, lower right). This model is attractivebecause it can account for the order sensitivity of theseneurons (i.e., inhibitory priming must lead excitation),and because song- and electrically evoked inhibition inarea X-projecting cells can last many hundreds of mil-liseconds (Schmidt and Perkel 1998), similar to thelongest inter-syllable integration times displayed bycombination-sensitive cells (Margoliash 1983). Initialexperiments aimed at reconstituting this highly non-linear behavior simply by pairing postsynaptic hyper-polarization with playback of only the second syllablehave failed, either because somatic current injection isinsufficient to relieve inactivation (perhaps due to adistal dendritic location of the calcium channels and/orthe low postsynaptic input impedance encountered invivo), or because syllable combination sensitivity reflectsa network property of HVc (Lewicki and Konishi 1995).In cells with absolute syllable combination sensitivity,

Fig. 4 One possible function of song-evoked inhibition in areaX-projecting HVc neurons is to enhance auditory selectivity for theBOS. A general model (top) shows how combination sensitivityarises through an interaction between the HVc local circuit andextrinsic auditory inputs to HVc that convey BOS-biased infor-mation without a high degree of sensitivity to syllable combina-tions. In this model, interneuronal selectivity is enhanced relative toHVc excitatory inputs because the interneurons do not receivedirect excitation from HVc extrinsic auditory input, but insteadthrough interposed RA-projecting neurons, which threshold theseextrinsic inputs. This architecture can account for the observationthat inhibition in area X-projecting cells is largely activated by theBOS, but not by other stimuli that drive only subthresholdresponses in RA-projecting cells (Fig. 2; Mooney 2000); otherarrangements include direct activation of the interneuron pool byHVc afferents. Two specific mechanisms for generating enhancedselectivity include inhibitory sculpting of area X-projecting cellresponses (lower left), and inhibition-triggered postsynaptic prim-ing (i.e., through deactivation of voltage-gated channels; lowerright). In the case of sculpting inhibition, convergent andconcurrent acoustically evoked inhibition and excitation lead to amore phasic response to the BOS, and could suppress responses toother stimuli altogether (the dotted line represents the spikethreshold of the area X-projecting cell; the oscillogram belowdepicts the playback of the BOS). In the priming model, syllable 1evokes an inhibitory postsynaptic potential (IPSP) that primes thepostsynaptic membrane by deinactivating a voltage-gated channel(represented by the movement of the tethered ball from the lumenof the channel); when presented in isolation, syllable 2 generates asubthreshold excitatory PSP (EPSP), but triggers suprathresholdbursting when following syllable 1, because of prior membranepriming

888

selective intracellular inhibitory blockade could testwhether hyperpolarizing inhibition is needed to gener-ate any suprathreshold responses to the preferred sylla-ble combination, as predicted of a priming mechanism.

Implicit in studying the regulation of song-evokedfiring in HVc is the idea that these firing patterns haveimportant consequences for activity in downstream ar-eas important to song production (RA) and vocal plas-ticity (the AFP). With regards to area X-projecting cells,one possibility is that patterns of synchronous inhibitionalso could directly affect the timing and amount of ac-tivity in the AFP (Fig. 5). The emerging parallels be-tween the AFP and the mammalian basal gangliainclude the presence of large, GAD-positive outputneurons in area X (similar to neurons in the internalsegment of the globus pallidus), that spontaneously fireat high rates and form calycal inhibitory terminals onthalamic neurons in DLM (Luo and Perkel 1999b;Farries and Perkel 2000; Luo et al. 2001). Similar tomammalian thalamic relay neurons, DLM neurons firelow threshold calcium spikes when released from inhi-bition and form excitatory synapses on their telence-phalic partners (in LMAN) (Livingston and Mooney1997; Boettiger and Doupe 1998; Luo and Perkel 1999a).Despite such circuit-breaking studies, the dynamicproperties of the AFP are still poorly understood.

An important issue is the nature of the synaptictransfer function between HVc and the output neuronsof area X, which could be net inhibitory, net excitatory,or perhaps a mixture of both, an arrangement mostclosely resembling the organization of mammalian basalganglia pathways (Fig. 5). As previously noted (Luo andPerkel 1999a), concordant strength of firing to the BOSin both LMAN and HVc could be accounted for by anadditional inhibitory synapse between area X-projectingHVc neurons, which are believed to be excitatory, andarea X output neurons; this arrangement ultimatelywould result in a net excitatory relationship betweenHVc and LMAN (Fig. 5A, lower). In contrast, net ex-citatory connections between HVc and area X outputneurons (either due to a monosynaptic excitatoryconnection or even numbers of interposed inhibitoryneurons between HVc and area X output neurons),would result in a net inhibitory relationship betweenHVc and LMAN (Fig. 5A, upper). Although at firstglance this latter arrangement might seem unable toexplain concordant firing to BOS playback in HVc andLMAN, synchronous inhibition onto a population ofarea X-projecting cells could result in a disinhibitoryrelationship between HVc and LMAN, specifically if thespontaneous firing of many area X-projecting neurons isneeded to maintain high spontaneous discharge rates inindividual area X output neurons. In this case, syn-chronized inhibition onto the area X-projecting cellpopulation would transiently depress an area X neuron’saction potential discharge, thus releasing DLM neuronsfrom their ‘‘resting’’ state of tonic inhibition and drivingthem to fire bursts of action potentials, strongly excitingLMAN neurons (Fig. 5A, upper).

One prediction of a disinhibitory model is thatLMAN will be excited by those stimuli that inhibit areaX-projecting cells, and inhibited by stimuli that syn-chronously excite these HVc neurons. At first glance,this prediction does not seem to be supported by theconcordance in strength of BOS-evoked firing in HVcand LMAN, but an unresolved issue is the degree oftemporal concordance in BOS-evoked activity in thesetwo areas. We have noted a contrast between the patternof BOS-evoked firing in area X-projecting HVc cells,which is highly phasic (Mooney 2000), and the pro-longed subthreshold responses that BOS playbackevokes in LMAN neurons (Rosen and Mooney 2000a).In this sense, the time-course of inhibition onto areaX-projecting cells may better match the prolonged sub-threshold depolarizations evoked in LMAN by the BOS.Another possibility is that different area X-projectingcells fire at different times to the BOS, and convergentorganization of net inhibitory connections between HVcand area X output cells generates the more prolongedresponses seen in LMAN. Moving beyond the realm ofpure speculation here will require simultaneous record-ings from identified area X-projecting HVc cells andLMAN neurons coupled with inactivation techniques,as well as detailed in vitro experiments aimed at furtherelucidating the synaptic connectivity between HVc andarea X.

The relationship between HVc and AFP activity inresponse to auditory stimuli is likely to hold an impor-tant key to understanding audition-dependent vocalmodification. Our recent analysis of BOS-evoked pat-terns of inhibition in area X-projecting HVc cells of theswamp sparrow (Mooney et al. 2001), a bird with mul-tiple song types, suggests that knowing whether theHVc-AFP relationship is disinhibitory will be crucial tounderstanding the functional significance of auditory-evoked activity in HVc. An intriguing feature of certainarea X-projecting cells in swamp sparrows is that theymay fire action potentials in response to only a singlesong type but can display subthreshold inhibition to allsong types in the bird’s repertoire (but not to the songsof other birds; Fig. 6). Clearly, a more complete ap-preciation of HVc auditory activity rests on an under-standing of the transfer function between HVc and theAFP. If so, the swamp sparrow data suggest that inhi-bition onto area X-projecting cells may provide a con-textual filter for distinguishing self-generated signalsfrom those of other birds, even when the bird’s ownsongs vary in their acoustical structures.

In the context of error correction, a remaining chal-lenge is to relate BOS-evoked activity in LMAN with theeffects of LMAN lesions on vocal plasticity. Brainardand Doupe’s (2000) finding that LMAN lesions blockaudition-dependent vocal plasticity suggests, when takenat face value, that the complete absence of LMAN ac-tivity is functionally equivalent to error-free vocal per-formance. The obvious question that arises is why BOSplayback should elicit the most potent response fromHVc and LMAN neurons, at least with respect to

889

temporal variants of the BOS and other conspecificsongs, if the BOS arguably is closer to the desired vocaloutput than these other stimuli. Two possibilities arethat the auditory activity evoked in the passive playbackstate differs from that generated during vocal perfor-mance, and that LMAN lesions and/or deafening inducepathological states that affect vocal change throughmechanisms distinct from those normally engaged dur-ing vocalization.

An important idea that may partially reconcile theseconcerns is that inhibition in area X-projecting cellsduring vocalization could serve to cancel out anticipatedauditory feedback (Fig. 5B). Cancellation signals havebeen described in electric fish, which navigate by sensingthe weak distortions external objects create in the fish’sself-generated electrical field. In order to detect suchweak signals, the fish must render transparent thelarger electrical distortions created by its own body

890

movements, which it does by using appropriatelyweighted efference copies of the motor commands forthese movements to cancel out resultant electrosensorysignals (Bell et al. 1997). In the case of singing, a can-cellation signal could be used to screen out vocal soundsthat the system has been trained through experience toexpect (Troyer and Doupe 2000a, 2000b); signals outsideof the realm of expectations, due perhaps to errors invocal control, could filter through and adaptively mod-ify the vocal control network (through HVc, RA and theAFP, e.g.; Fig. 5B and legend). Such a mechanism couldexplain the difficulty in detecting auditory responses tothe BOS in singing birds (McCasland and Konishi1981): accurate renditions of the BOS are cancelled out;and might also account for how artificially delayed au-ditory feedback signals disrupt song quality (Leonardoand Konishi 1999), i.e., they mismatch the filter and arenot effectively cancelled out, and thus modulate AFPactivity. Answers to these questions will require chronicrecordings from identified HVc and AFP neurons insinging birds, especially in the presence of artificiallydelayed or otherwise distorted auditory feedback.

Fig. 5A,B Auditory-evoked inhibition in area X-projecting HVcneurons is likely to affect activity in the AFP, and also could pointto a mechanism for vocally-triggered cancellation of auditoryfeedback. A Upper: synchronous acoustically-evoked inhibition ofarea X-projecting cells could trigger the disinhibition of LMAN. Inthis diagram, the equivalent circuit model is shown at the top, andthe action potential activity for each neuron type is shown instylized form either during silence or during song playback(‘‘song’’). In the silent state, low rates of spontaneous firing inarea X-projecting cells are relayed through convergent netexcitatory synapses to maintain the high ‘‘resting’’ firing rates ofprojection neurons in area X, which in turn form inhibitorysynapses on DLM neurons. Tonic inhibition from area X silencesDLM neurons, preventing them from exciting LMAN neurons.During song playback (as of the BOS, shown as an oscillogram atthe bottom left), synchronized inhibition in area X-projecting HVcneurons transiently suppresses the firing of area X projectionneurons, leading to the excitation of DLM and LMAN neurons.A Lower: an additional or alternative circuit organization involvesan intervening inhibitory neuron in area X located between theHVc terminal and the area X output neuron; this arrangementresults in a net excitatory relationship between HVc and LMAN. BVocal signals in RA-projecting neurons also could provide feed-forward inhibition onto area X-projecting cells that cancels outanticipated auditory feedback. In this model, an efference copy ofthe vocal command is conveyed via an excitatory axon collateral ofan RA-projecting cell to an HVc interneuron, which in turn formsan inhibitory synapse on an area X-projecting neuron. Theinhibitory signal acts to cancel out anticipated auditory feedbackconveyed to the area X-projecting cell through an independentexcitatory channel (depicted by the dashed line and arrowhead).Through experience, the cancellation signal is scaled to exactlycancel out the auditory feedback generated by an errorlessrendition of the BOS – when vocal error is absent, the net activityof the area X projecting remains unchanged from the resting state(see Troyer and Doupe (2000a, 2000b) for a detailed modelincorporating similar features). In the case of vocal error, themismatch between the inhibitory cancellation signal and theauditory feedback signal leads to a net increase or decrease in theactivity of the area X-projecting cell, modulating the activity of theAFP and leading to error correction of the vocal command signalpropagating from HVc to RA

Fig. 6 Subthreshold and suprathreshold responses (conventions asin Fig. 2) to song playback (oscillograms at bottom) recordedintracellularly from an individual area X-projecting cell in a swampsparrow, a songbird with multiple song types in its repertoire (i.e.,BOSA, B and C). Significant action potential responses (spikes/bin)were evoked only by song type C and by the conspecific song(CON), although the suprathreshold response to song type C washighly phasic, distinct from the low intensity, sustained actionpotential discharges elicited by the other bird’s song. In contrast,hyperpolarizing responses were evoked by playback of all the bird’ssong types, but not by the conspecific song [middle row; Vm (mV)].Vertical scale bar equals 10 mV; horizontal scale bar equals 1 s

b

891

The intrinsic HVc circuitry seems well suited to sucha cancellation function, because electrical excitation ofRA-projecting neurons potently inhibits area X-projecting cells (Fig. 7). The flow of activity, from RA-projecting axon collateral to inhibitory interneuron toarea X-projecting cell, is suggested by the finding thatthe same stimulation protocols that inhibit area X-pro-jecting cells also excite HVc interneurons (Fig. 7A); theinhibitory postsynaptic potentials (IPSPs) evoked in thismanner in area X-projecting cells have long onset la-tencies and are blocked by AMPA and NMDA receptorblockers (R. Mooney, personal observations), implicat-ing an intervening excitatory synapse. The indirect in-hibitory coupling between the HVc projection cells canbe extremely potent: in preliminary studies using si-multaneous recordings from RA- and area X-projectingcells, we find that d.c.-evoked spiking in a single RA-projecting cell is sufficient to evoke an IPSP in a nearbyarea X-projecting cell (Fig. 7B). Indeed, such organiza-tion could account for the opposing subthreshold re-sponses these cell types show to the BOS playback:auditory evoked excitation of RA-projecting cells woulddrive feed-forward inhibition onto area X-projecting

cells. In the awake behaving bird, one function of thiscircuit structure could be to enable RA-projecting neu-rons conveying motor commands during singing totransmit an inverted efference copy through the localnetwork, masking area X-projecting cells from resultantauditory signals (i.e., the BOS; Fig. 5B). In this regard,the relatively slow activation of G-protein coupled in-hibition present within HVc (Dutar et al. 2000) maysufficiently delay a vocally-generated cancellation signalso that it coincides with auditory feedback. One pre-diction of this model is that slight distortions of theauditory feedback signal (induced by singing-triggeredplayback of the BOS with slight temporal and/or spec-tral shifts) should result in detectable alterations in areaX-projecting neuronal activity from their normal song-related pattern.

Conclusions

The role of audition in vocal learning and maintenanceis likely to be mediated through auditory activity inthose brain nuclei involved in song patterning, especially

892

the song nucleus HVc. To date, the source and nature ofauditory input to HVc has remained surprisingly ob-scure, despite evidence that HVc performs both auditoryand vocal functions, and the finding that its neuronsdisplay selective BOS-evoked activity in the awake bird.Functional anatomical approaches, combining in vivoand in vitro intracellular recordings with cross-correla-tion analysis and reversible inactivation techniques willhelp clarify HVc’s sources of auditory input, and lead toa fuller appreciation of any differences between the au-ditory selectivity of HVc versus its afferents. The sub-threshold differences in auditory-evoked activity in theHVc projection cells that innervate either RA or theAFP likely arise through local processing and point tofunctional specializations in the two cell populations,including the manner in which they signal their postsy-naptic targets. An explicit knowledge of the connectionsbetween different HVc neurons, as well as betweenidentified HVc projection neurons and their downstreamtargets, is still lacking. Our present view is that song-evoked inhibition arises in area X-projecting cellsthrough local circuit interactions, and could serve toenhance song-selectivity while also signaling areas im-portant to audition-guided vocal plasticity. An impor-tant and challenging goal in the future will be to relatethe detailed circuit information gleaned from intracel-lular recordings made in brain slices and anesthetizedsongbirds with the extracellular activity patterns

recorded in the singing bird. Uniting these two levels ofanalysis will provide a powerful link between synapticmechanisms and the development and production oflearned vocalizations.

References

Alvarez-Buylla A, Theelen M, Nottebohm F (1988) Birth of pro-jection neurons in the higher vocal center of the canary fore-brain before, during, and after song learning. Proc Natl AcadSci USA 85:8722–8726

Bell C, Bodznick D, Montgomery J, Bastian J (1997) The genera-tion and subtraction of sensory expectations within cerebellum-like structures. Brain Behav Evol 50:17–31

Benton S, Cardin JA, Devoogd TJ (1998) Lucifer yellow filling ofarea X-projecting neurons in the high vocal center of femalecanaries. Brain Res 799:138–147

Boettiger CA, Doupe AJ (1998) Intrinsic and thalamic excitatoryinputs onto songbird LMAN neurons differ in their pharma-cological and temporal properties. J Neurophysiol 79:2615–2628

Bottjer SW, Miesner EA, Arnold AP (1984) Forebrain lesionsdisrupt development but not maintenance of song in passerinebirds. Science 224:901–903

Brainard M, Doupe A (2000) Interruption of a forebrain-basalganglia circuit prevents plasticity of learned vocalizations.Nature 404:762–766

Brenowitz EA (1991) Altered perception of species-specific song byfemale birds after lesions of a forebrain nucleus. Science251:303–305

Burt JM, Lent KL, Beecher MD, Brenowitz EA (2000) Lesions ofthe anterior forebrain song control pathway in female canariesaffect song perception in an operant task. J Neurobiol 42:1–13

Carpenter RL, Rutherford DR (1973) Acoustic cue discriminationin adult aphasia. J Speech Hear Res 16:534–544

Cynx J, Nottebohm F (1992) Role of gender, season, and famil-iarity in discrimination of conspecific song by zebra finches(Taeniopygia guttata). Proc Natl Acad Sci USA 89:1368–1371

Dave A, Margoliash D (2000) Song replay during sleep and com-putational rules for sensorimotor vocal learning. Science290:812–816

Dave AS, Yu AC, Margoliash D (1998) Behavioral state modula-tion of auditory activity in a vocal motor system. Science282:2250–2254

Doupe AJ, Konishi M (1991) Song-selective auditory circuits in thevocal control system of the zebra finch. Proc Natl Acad SciUSA 88:11339–11343

Dutar P, Vu HM, Perkel DJ (1998) Multiple cell types distin-guished by physiological, pharmacological, and anatomicproperties in nucleus Hvc of the adult zebra finch. J Neuro-physiol 80:1828–1838

Dutar P, Petrozzino J, Vu H, Schmidt M, Perkel D (2000) Slowsynaptic inhibition mediated by metabotropic glutamate recep-tor activation of GIRK channels. J Neurophysiol 84:2284–2290

Farries MA, Perkel DJ (2000) Electrophysiological properties ofavian basal ganglia neurons recorded in vitro. J Neurophysiol84:2502–2513

Ferster D, Chung S, Wheat H (1996) Orientation selectivity ofthalamic input to simple cells of cat visual cortex. Nature380:249–252

Fortune ES, Margoliash D (1995) Parallel pathways and conver-gence onto HVc and adjacent neostriatum of adult zebra finches(Taeniopygia guttata). J Comp Neurol 360:413–441

Foster EF, Bottjer SW (1998) Axonal connections of the high vocalcenter and surrounding cortical regions in juvenile and adultmale zebra finches. J Comp Neurol 397:118–138

Foster EF, Mehta RP, Bottjer SW (1997) Axonal connections ofthe medial magnocellular nucleus of the anterior neostriatum inzebra finches. J Comp Neurol 382:364–381

Fig. 7 Local circuit organization of HVc, schematized at upperleft, includes feed-forward inhibition from RA- to area X-projecting neurons. A Intracellular recordings made from identifiedHVc neurons in zebra finch brain slices reveal that electricalstimulation of the axons of RA-projecting neurons evokes IPSPs inarea X-projecting cells and EPSPs in interneurons (subthresholdand suprathreshold EPSPs are shown for the interneuron and wereevoked with the same stimulus current applied to the RA-projecting axons). B Simultaneous, dual intracellular recordingsmade from an area X- and an RA-projecting neuron in a brain slicereveal that pairs of action potentials in the RA-projecting cell aresufficient to trigger an IPSP in the area X-projecting cell. A positivecurrent (bottom trace, +0.75 nA) injected into the RA-projectingcell was adjusted to elicit one (left) or sometimes two (middle andright) action potentials; spike doublets but not singlets consistentlyelicited a PSP in the area X-projecting cell. Artificially varying themembrane potential of the area X-projecting cell with currentinjection from its resting value (� –65 mV; not shown) to verynegative values (� –100 mV; left and middle) and to moderatelydepolarized values (� –42 mV; right) reveals that the evoked PSPbecomes hyperpolarizing at � –50 mV, consistent with a GABAA-mediated IPSP. One interpretation of this result is that spikedoublets in the RA-projecting cell are required to drive anintervening inhibitory interneuron to spike threshold; anotherpossibility is that the RA-projecting cell forms an inhibitorysynapse directly on the area X-projecting cell, and that this synapserequires paired spikes to achieve transmitter release. This latter ideais less likely, because RA-projecting HVc neurons are known toform excitatory, glutamatergic synapses on RA neurons (Mooneyand Konishi 1991). The evidence gathered from dual intracellularrecordings underscores the potency of the effective inhibitorycoupling between RA- and area X-projecting HVc neurons, anarrangement likely to account for some of the subthresholddifferences in song-evoked activity observed in these cell typesand that could serve the cancellation function outlined in Fig. 5B

b

893

Gentner TQ, Hulse SH, Bentley GE, Ball GF (2000) Individualvocal recognition and the effect of partial lesions to HVc ondiscrimination, learning, and categorization of conspecific songin adult songbirds. J Neurobiol 42:117–133

Hessler NA, Doupe AJ (1999) Singing-related neural activity in adorsal forebrain-basal ganglia circuit of adult zebra finches.J Neurosci 19:10461–10481

Janata P, Margoliash D (1999) Gradual emergence of song selec-tivity in sensorimotor structures of the male zebra finch songsystem. J Neurosci 19:5108–5118

Katz LC, Gurney ME (1981) Auditory responses in the zebrafinch’s motor system for song. Brain Res 221:192–197

Kelley DB, Nottebohm F (1979) Projections of a telencephalic au-ditory nucleus-field L in the canary. J CompNeurol 183:455–469

Kittelberger J, Mooney R (1999) Lesions of an avian forebrainnucleus that disrupt song development alter synaptic connec-tivity and transmission in the vocal premotor pathway.J Neurosci 19:9385–9398

Knudsen EI (1994) Supervised learning in the brain. J Neurosci14:3985–3997

Konishi M (1965) Effects of deafening on song development inAmerican robins and black-headed grosbeaks. Z Tierpsychol22:584–599

Kubota M, Saito N (1991) Sodium- and calcium-dependent con-ductances of neurones in the zebra finch hyperstriatum ventralepars caudale in vitro. J Physiol (Lond) 440:131–142

Kubota M, Taniguchi I (1998) Electrophysiological characteristicsof classes of neuron in the HVc of the zebra finch. J Neuro-physiol 80:914–923

Leonardo A, Konishi M (1999) Decrystallization of adult birdsongby perturbation of auditory feedback. Nature 399:466–470

Lewicki MS (1996) Intracellular characterization of song-specificneurons in the zebra finch auditory forebrain. J Neurosci16:5855–5863

Lewicki MS, Arthur BJ (1996) Hierarchical organization of audi-tory temporal context sensitivity. J Neurosci 16:6987–6998

Lewicki MS, Konishi M (1995) Mechanisms underlying the sensi-tivity of songbird forebrain neurons to temporal order. ProcNatl Acad Sci USA 92:5582–5586

Livingston FS, Mooney R (1997) Development of intrinsic andsynaptic properties in a forebrain nucleus essential to aviansong learning. J Neurosci 17:8997–9009

Luo M, Perkel DJ (1999a) A GABAergic, strongly inhibitoryprojection to a thalamic nucleus in the zebra finch song system.J Neurosci 19:6700–6711

Luo M, Perkel DJ (1999b) Long-range GABAergic projection in acircuit essential for vocal learning. J Comp Neurol 403:68–84

Luo M, Ding L, Perkel DJ (2001) An avian basal ganglia pathwayessential for vocal learning forms a closed topographic loop.J Neurosci 21:6836–6845

Macdougall-Shackleton SA, Hulse SH, Ball GF (1998) Neuralbases of song preferences in female zebra finches (Taeniopygiaguttata). Neuroreport 9:3047–3052

Margoliash D (1983) Acoustic parameters underlying the responsesof song-specific neurons in the white-crowned sparrow. J Neu-rosci 3:1039–1057

Margoliash D (1986) Preference for autogenous song by auditoryneurons in a song system nucleus of the white-crowned spar-row. J Neurosci 6:1643–1661

Marr D (1969) A theory of cerebellar cortex. J Physiol (Lond)202:437–470

McCasland JS (1987) Neuronal control of birdsong production.J Neurosci 7:23–39

McCasland JS, Konishi M (1981) Interaction between auditory andmotor activities in an avian song control nucleus. Proc NatlAcad Sci USA 78:7815–7819

Mooney R (2000) Different subthreshold mechanisms underliesong-selectivity in identified HVc neurons of the zebra fiNeurosci 20:5420–5436

Mooney R, Konishi M (1991) Two distinct inputs to an avian songnucleus activate different glutamate receptor subtypes on indi-vidual neurons. Proc Natl Acad Sci USA 88:4075–4079

Mooney R, Hoese W, Nowicki S (2001) Auditory representation ofthe vocal repertoire in a songbird with multiple song types. ProcNatl Acad Sci USA 98:12778–12783

Nelson S, Toth L, Sheth B, Sur M (1994) Orientation selectivity ofcortical neurons during intracellular blockade of inhibition.Science 265:774–777

Nixdorf BE, Davis SS, DeVoogd TJ (1989) Morphology of Golgi-impregnated neurons in hyperstriatum ventralis, pares caudalisin adult male and female canaries. J Comp Neurol 284:337–349

Nordeen KW, Nordeen EJ (1992) Auditory feedback is necessaryfor the maintenance of stereotyped song in adult zebra finches.Behav Neural Biol 57:58–66

Nottebohm F, Stokes TM, Leonard CM (1976) Central control ofsong in the canary, Serinus canarius. J Comp Neurol 165:457–486

Nottebohm F, Kelley DB, Paton JA (1982) Connections of vocalcontrol nuclei in the canary telencephalon. J Comp Neurol207:344–357

Price PH (1979) Developmental determinants of structure in zebrafinch song. J Comp Physiol Psychol 93:260–277

Pytte C, Suthers R (1999) A bird’s own song contributes to con-specific song perception. Neuroreport 10:1773–1778

Rieke F, Bodnar DA, Bialek W (1995) Naturalistic stimuli increasethe rate and efficiency of information transmission by primaryauditory afferents. Proc R Soc Lond Ser B 262:259–265

Rosen M, Mooney R (2000a) Intrinsic and extrinsic contributionsto auditory selectivity in a song nucleus critical for vocal plas-ticity. J Neurosci 20:5437–5448

Rosen MJ, Mooney R (2000b) Local and extrinsic contributions tosong-selectivity in the zebra finch song nucleus HVC. SocNeurosci Abstr 26:2030

Scharff C, Nottebohm F (1991) A comparative study of the be-havioral deficits following lesions of various parts of the zebrafinch song system: implications for vocal learning. J Neurosci11:2896–2913

Scharff C, Nottebohm F, Cynx J (1998) Conspecific and hetero-specific song discrimination in male zebra finches with lesions inthe anterior forebrain pathway. J Neurobiol 36:81–90

Schmidt MF, Konishi M (1998) Gating of auditory responses in thevocal control system of awake songbirds. Nature Neurosci1:513–518

Schmidt MF, Perkel DJ (1998) Slow synaptic inhibition in nucleusHVc of the adult zebra finch. J Neurosci 18:895–904

Sen K, Theunissen FE, Doupe AJ (2001) Feature analysis of nat-ural sounds in the songbird auditory forebrain. J Neurophysiol86:1445–1458

Shaevitz SS, Theunissen FE (2001) Functional connectivity bet-ween field L and HVC in the male zebra finch. Soc NeurosciAbstr 27:724.716

Stoddard PK, Beecher MD, Campbell SE, Horning CL (1992)Song-type matching in the song sparrow. Can J Zool 70:1440–1444

Theunissen FE, Doupe AJ (1998) Temporal and spectral sensitivityof complex auditory neurons in the nucleus HVc of male zebrafinches. J Neurosci 18:3786–3802

Theunissen FE, Sen K, Doupe AJ (2000) Spectral-temporalreceptive fields of nonlinear auditory neurons obtained usingnatural sounds. J Neurosci 20:2315–2331

Theunissen FE, David SV, Singh NC, Hsu A, Vinje WE, GallantJL (2001) Estimating spatio-temporal receptive fields of audi-tory and visual neurons from their responses to natural stimuli.Network 12:289–316

Troyer T, Doupe AJ (2000a) An associational model of birdsongsensorimotor learning. I. Efference copy and the learning ofsong syllables. J Neurophysiol 84:1204–1223

Troyer T, Doupe AJ (2000b) An associational model of birdsongsensorimotor learning. II. Temporal hierarchies and the learn-ing of song sequence. J Neurophysiol 84:1224–1239

Vates GE, Broome BM, Mello CV, Nottebohm F (1996a) Audi-tory pathways of caudal telencephalon and their relation tothe song system of adult male zebra finches. J Comp Neurol366:613–642

894

Vates GE, Vicario DS, Nottebohm F (1996b) A new source ofauditory input to the songbird anterior forebrain pathway. SocNeurosci Abstr 22:66.61

Vates GE, Vicario DS, Nottebohm F (1997) Reafferent thalamo-‘‘cortical’’ loops in the song system of oscine songbirds. J CompNeurol 380:275–290

Vicario DS, Yohay KH (1993) Song-selective auditory input to aforebrain vocal control nucleus in the zebra finch. J Neurobiol24:488–505

Volman SF (1996) Quantitative assessment of song-selectivity inthe zebra finch ‘‘high vocal center’’. J Comp Physiol A 178:849–862

Vu ET, Mazurek ME, Kuo YC (1994) Identification of a forebrainmotor programming network for the learned song of zebrafinches. J Neurosci 14:6924–6934

Waldstein RS (1990) Effects of postlingual deafness on speechproduction: implications for the role of auditory feedback.J Acoust Soc Am 88:2099–2144

Wild JM (1994) Visual and somatosensory inputs to the avian songsystem via nucleus uvaeformis (Uva) and a comparison with theprojections of a similar thalamic nucleus in a nonsongbird,Columba livia. J Comp Neurol 349:512–535

Wild JM, Karten HJ, Frost BJ (1993) Connections of the auditoryforebrain in the pigeon (Columba livia). J Comp Neurol 337:32–62

Williams H (1989) Multiple representations and auditory-motorinteractions in the avian song system. Ann N Y Acad Sci563:148–164

Williams H, Nottebohm F (1985) Auditory responses in avianvocal motor neurons: a motor theory for song perception inbirds. Science 229:279–282

Woolley SMN, Casseday JH (2001) Acoustic tuning to songs andcalls in songbird auditory midbrain neurons. In: 6th Int CongrNeuroethol. University of Bonn, p 144

Yu A, Margoliash D (1996) Temporal hierarchical control ofsinging in birds. Science 273:1871–1875

895