Central auditory neurons have composite receptive fields

Abstract

High-level neurons processing complex, behaviorally relevant signals are sensitive to conjunctions of features. Characterizing the receptive fields of such neurons is difficult with standard statistical tools, however, and the principles governing their organization remain poorly understood. Here, we demonstrate multiple distinct receptivefield features in individual high-level auditory neurons in a songbird, European starling, in response to natural vocal signals (songs). We then show that receptive fields with similar characteristics can be reproduced by an unsupervised neural network trained to represent starling songs with a single learning rule that enforces sparseness and divisive normalization. We conclude that central auditory neurons have composite receptive fields that can arise through a combination of sparseness and normalization in neural circuits. Our results, along with descriptions of random, discontinuous receptive fields in the central olfactory neurons in mammals and insects, suggest general principles of neural computation across sensory systems and animal classes. is an important question in sensory neuroscience. Dimensionality reduction involves extracting a hierarchy of features to obtain a selective and invariant (categorical) representation useful for behavior. To understand better the principles underlying this process in the central auditory system, we characterized receptive fields of neurons in the caudo-medial nidopallium (NCM) of the European starling (Sturnus vulgaris), a songbird with an acoustically rich vocal repertoire (1). The NCM, a secondary auditory cortexlike region in songbirds, receives convergent inputs from the primary thalamorecipient region, Field L, and other secondary auditory regions (2) and contains neurons selectively tuned to birdsong, a behaviorally relevant natural stimulus (3-5).We recorded action potentials extracellularly from individual well-isolated NCM neurons during the playback of starling songs and estimated the structure of the neurons' receptive fields using the Maximum Noise Entropy (MNE) method (6). Statistical inference methods in this class (7,8) maximize the noise entropy of the conditional response distribution to produce models that are constrained by a given set of stimulus-response correlations but that are otherwise as random, and therefore as unbiased, as possible. Unlike the spike-triggered covariance (STC) method (9), MNE works well with natural stimuli; in contrast to the maximally informative dimensions (MID) method (10), MNE can identify any number of relevant receptive-field features. Results Single NCM Neurons Respond to Multiple Distinct Features of StarlingSong. We recorded neuronal responses to six different 1-minutelong songs, each repeated 30 times. These songs were recorded from three male starlings, and together they contained over 200 motifs, brief segments of starling song that are perceived as distinct auditory objects (11). An NCM neuron usually responds to a variety of motifs (4, 12) (Fig. 1), and NCM neurons display ...

Network Receptive Field Modeling Reveals Extensive Integration and Multi-feature Selectivity in Auditory Cortical Neurons

“…Another recent avian forebrain study [54] used a maximum noise entropy (MNE) approach to uncover multiple receptive fields sensitive to second-order aspects of the stimulus. Unlike the above two GNM [24,53] approaches, this model does not have hidden units with sigmoidal nonlinearities, but finds multiple quadratic features.…”

Comparative Aspects of Hearing in Vertebrates and Insects with Antennal Ears

“…Crucially, the same neuronal circuit that can implement Gaussian tuning for selectivity can also implement, by using different parameter values, a MAX-like operation (output = maximum(input)), which is important for invariance and has been observed experimentally in the mammalian visual cortex [102,103] and in the central auditory system of a songbird [104]. The predicted computational flexibility was experimentally verified in the European starling, a species of songbird, where individual central auditory neurons could switch between the MAX-like operation for invariance and a tuning operation for selectivity, depending on the stimulus and network state [104], displaying sparse responses and multidimensional selectivity [105].A Gaussian tuning function is also useful for generalization after training on a limited data set. Specifically, radial Gaussian basis functions are smooth and interpolate well from training data to new data [106].…”

Multidimensional receptive field processing by cat primary auditory cortical neurons

“…1E). Suppressive features have been observed in visual cortex (Rust et al, 2005, Chen et al, 2007), though rarely in the auditory system (though see (Harper et al, 2016, Kozlov and Gentner, 2016). For the neuron in Fig.…”

“…In the first, the MNE analysis was applied to neurons in a secondary auditory cortex-like center in the songbird (Kozlov and Gentner, 2016). In concord with our results, the authors found multiple excitatory and suppressive features.…”

“…The feature may be seen as a stimulus that drives the neuron to respond, or it may be interpreted as a descriptor of neuronal stimulus processing (Sharpee et al, 2004). The single feature STRF description has provided important insights, though in recent years it has become apparent that many auditory cortical neurons simultaneously encode information about more than one stimulus feature in their spiking activity, and thus the single feature may not reveal the true richness of auditory cortical processing (Atencio et al, 2008, 2009, 2012, Harper et al, 2016, Kozlov and Gentner, 2016). …”