TRENDS in Neurosciences Vol.24 No.12 December 2001 http://tins.trends.com 0166-2236/01/$ – see front matter © 2001 Elsevier Science Ltd. All rights reserved. PII: S0166-2236(00)01953-6 677 Research Update Research News To localize a sound source in space, the auditory system detects minute differences in the arrival time of a sound between the tw o ears. It has long been assumed that delay lines and coincidence detectors turn these time differences into a labelled line code for source position, but recent studies challenge this view. For many animals, the auditory system provides an essential ‘early warning system’, detecting the sound of approaching danger or opportunity. To serve this role effectively, the auditory system must both detect a sound, and also determine the likely direction of its source. For this purpose, sensitivity to several spatial acoustic cues has evolved 1 . One such cue lies in the interaural time differences (ITDs) that arise when a sound comes from one side, reaching one ear before the other. The magnitude of an ITD is directly related to the angular position of a sound source: the further the sound is from the midline, the larger the ITD. However, because of the high speed of sound and the relatively small distance between the ears, ITDs tend to be tiny, typically only a fraction of a millisecond. The smallest ITD humans can detect is in the order of 10–20 μs. A humble fly, Ormia ochracea, might even be able to do a staggering 200 times better than that, with an ability to resolve ITDs of only 50 ns (Ref. 2). The question of how the comparatively sluggish neural hardware can efficiently detect and encode such small delays has fascinated auditory neuroscientists for a long time. However, one of their most cherished working hypotheses, the so called ‘Jeffress Coincidence Model’ 3 , might now have to be revised, in the light of a recent study by McAlpine et al. 4 Signals from each ear are relayed through highly specialized synapses in the cochlear nuclei to converge on neurons in the medial superior olive (MSO). According to the Jeffress model, axons projecting to the MSO form ‘delay lines’, adding a small delay to the signal from one ear. Neurons in the MSO act as ‘coincidence detectors’ or ‘cross- correlators’ 5 , which respond strongly only if the inputs from each ear are precisely matched in time. Consequently, MSO neurons become tuned to the ITD value that is precisely offset by the axonal delay. According to the model, the axonal input delay varies systematically from one MSO neuron to the next. This establishes a ‘labelled line’ code for ITD, and hence the direction or the source of the sound. These ‘labelled lines’ then project to the inferior colliculus (IC) and are relayed from there to other auditory structures. Hunting in the dark – Jeffress style Early studies of the auditory system of the barn owl, Tyto alba, generated much evidence in support of the Jeffress model 6 . In the barn owl nucleus laminaris (NL, the avian homologue of the MSO), for example, the systematic delay line properties postulated by the Jeffress model have been demonstrated directly 7 . Barn owls are highly unusual animals that can strike prey in complete darkness, guided by sound localization alone. To achieve this astonishing feat, these animals have evolved several auditory specializations, including a pair of highly asymmetric outer ears, with one ear pointing up, the other down. Consequently, interaural level differences (ILDs) vary as a function of sound-source elevation for these animals 8 . This arrangement is extremely convenient, because it enables neurons in the barn owl IC to determine both the azimuth and the elevation of a sound source by combining the ILD information extracted in the NL with elevational (ILD) information extracted in another nucleus of the owl’s brainstem 9 . This ingenious organization has no direct parallel in the mammalian auditory system. Nevertheless, it has long been thought that the encoding of sound azimuth through ITD sensitivity might happen in an analogous fashion in mammals, in accordance with Jeffress’ original model. This notion is now under pressure, however, as predictions derived from the model fail to be borne out. Phantom direction labels suggest a response-gradient code in mammals Spatial hearing is particularly acute for sound sources directly in front of the animal, and one might expect that the auditory system would possess many neurons sharply tuned to ITDs near zero, which could encode frontal positions at high resolution. However, when McAlpine and colleagues investigated the ITD sensitivity of a large number of neurons in the IC of the guinea pig, they found that the great majority was tuned to ITDs that were larger than any ITD normally experienced by the animal 4 . Assuming that these neurons represent sound azimuth based on a labelled line code for ITD would lead to the perplexing conclusion that they encode non-existent source directions. McAlpine and colleagues also noted that ITD tuning was generally broader than that observed in barn owls, and that tuning for ITD and for sound frequency were not independent. There was a strong tendency for neurons tuned to lower frequencies to show broader ITD tuning and larger optimal ITDs than neurons Of delays, coincidences and efficient coding for space in the auditory pathway Jan Schnupp