Mechanisms for Complex Feature Selectivity in the Songbird Auditory Forebrain

Neurons in many secondary sensory areas exhibit high selectivity for specific stimuli. These representationsare thought to underlie the ability of animals to quickly recognize complex objects under a wide range ofconditions, but it is not well understood how this selectivity is built up from the simpler response propertiesof neurons at earlier stages of processing. The songs of European starlings (Sturnus vulgaris) incorpo-rate large repertoires of motifs (each a complex vocalization circa 1 s in duration) that are unique to eachindividual bird, and contribute to individual recognition by song [1–3]. We explored the receptive fieldproperties of cells in the caudomedial mesopallium (CMM), which respond selectively to learned songs andmay contribute to vocal recognition behavior [4]. In a series of awake-restrained and urethane anesthetizedrecordings from 5 starlings, we recorded the responses of 70 well-isolated single units in CMM to noveland familiar motifs. Out of 59 neurons that gave significant responses to at least one motif, 36 (61%) weresignificantly selective, responding robustly to only a small number of motifs. Selective neurons had sig-nificantly lower spontaneous firing rates, and exhibited phasic firing patterns, with one or more preciselytimed peaks of excitation. Because these response properties are poorly predicted by linear spectrotempo-ral receptive field (STRF) models [4], we examined instead if CMM neurons were responding to multipleauditory features of the stimuli. Motifs were decomposed into spectrotemporal features that correspondedto stereotyped vocal gestures, which are employed by many individuals. Both selective and non-selectiveneurons responded robustly to these features when they were presented in isolation, but whereas nonselec-tive neurons tended to give excitatory responses to a broad range of features, selective neurons respondedto exhibited excitatory responses to a few features and suppressive responses to many other features. Theresponses of CMM neurons to motifs could be predicted from the sum of the excitatory and suppressivecontributions of the constituent features with a high degree of accuracy (median correlation coefficient =0.50; inter-quartile range 0.24–0.68; 86 motifs, 40 neurons). These data suggest that CMM neurons achievespecificity for particular stimuli by sampling from disjoint regions of an underlying feature space, and bymaking use of the temporal interplay between excitation and suppression to further narrow tuning.AcknowledgmentsThis work was supported by NIH grants DC007206 and F32DC008752 (CDM).References[1] M. Eens. Understanding the complex song of the European starling: an integrative approach. Advancesin the Study of Behavior, 26:355–434, 1997.[2] T. Q. Gentner and S. H. Hulse. Perceptual mechanisms for individual vocal recognition in Europeanstarlings, Sturnus vulgaris. Anim Behav, 56(3):579–594, Sep 1998.[3] T. Q. Gentner and S. H. Hulse. Perceptual classification based on the component structure of song inEuropean starlings. J Acoust Soc Am, 107(6):3369–3381, Jun 2000.[4] T. Q. Gentner and D. Margoliash. Neuronal populations and single cells representing learned auditoryobjects. Nature, 424(6949):669–674, Aug 2003.