Multisensory contributions to low-level, ‘unisensory’ processingIntroductionVisual and somatosensory responses in auditory cortexAnatomical mechanisms of multisensory convergence in auditory cortexCortical sourcesSubcortical sources?Functional implications of multisensory convergence in low-level cortical processingConclusions: low-level multisensory convergence and ‘the sensory hierarchy’AcknowledgementsReferences and recommended readingMultisensory contributions to low-level, ‘unisensory’ processingCharles E Schroeder and John FoxeNeurobiologists have traditionally assumed that multisensoryintegration is a higher order process that occurs after sensorysignals have undergone extensive processing through ahierarchy of unisensory subcortical and cortical regions.Recent findings, however, question this assumption. Studies inhumans, nonhuman primates and other species demonstratemultisensory convergence in low level cortical structures thatwere generally believed to be unisensory in function. In additionto enriching current models of multisensory processing andperceptual functions, these new findings require a revision inour thinking about unisensory processing in low level corticalareas.AddressesCognitive Neuroscience and Schizophrenia Program, The Nathan KlineInstitute for Psychiatric Research, 140 Old Orangeburg Rd, Orangeburg,NY 10962 and Program in Cognitive Neuroscience, Department ofPsychology, The City College of the City University of New York, NorthAcademic Complex, 138thStreet and Convent Avenue, New York, NY10031, USACorresponding author: Schroeder, Charles E ([email protected])Current Opinion in Neurobiology 2005, 15:454–458This review comes from a themed issue onSensory systemsEdited by David Julius and Andrew KingAvailable online 12th July 20050959-4388/$ – see front matter# 2005 Elsevier Ltd. All rights reserved.DOI 10.1016/j.conb.2005.06.008IntroductionRecent studies in both monkey and human subjects haveprovided evidence for multisensory convergence at low-level, putatively unisensory, stages of the sensory corticalpathways [1–10,11]. For example, somatosensoryresponses can be observed in auditory belt cortical regions(i.e. at the second level of auditory processing [4]), andeye position input modulates auditory responses even atthe primary cortical (A1) level [12,13]. Parallel findingshave emerged in carnivores [12–14]. Most dramatically,two laboratories [15,16] have shown anatomical intercon-nections between low-level visual and low-level auditoryareas, which include the primary cortices (V1 and A1), andtwo others have shown that eye position can affect thegain of auditory responses in A1 [17,18]. Recent reviews[9,19,20] have highlighted the fact that low-level (early)multisensory convergence is paradoxical from a hierarch-ical sensory perspective, and its functions are not yetclear.This review focuses on low-level multisensory conver-gence in the primate auditory system. First, we review theneurophysiological evidence in this area. Second, wediscuss the potential anatomical sources of non-auditoryinput, and the types of projections used (i.e. feedforward,feedback, lateral). Finally, we consider the functionalimplications of early multisensory integration in the con-text of the hierarchical model of auditory processing. Toavoid confusion, we will use the term ‘low-level’ to referto the anatomical stage at which a multisensory process isobserved, reserving the term ‘early’ for reference to thetime domain.Visual and somatosensory responses inauditory cortexStudies using event related potentials (ERPs) in humanshave demonstrated short latency audio–visual [3,7] andaudio–somatosensory [21,22] interactions, and haveraised the possibility that these interactions occur inauditory cortices of the superior temporal plane. Locali-zation of multisensory interactions within the superiortemporal plane is independently supported by findingsfrom other brain imaging techniques that have betteranatomical resolution, including magnetoencephalogra-phy (MEG) [1,23,24] and functional magnetic resonanceimaging (fMRI) [11,25].Intracranial recordings in several macaque species directlyconfirm multisensory convergence in auditory cortex[4,5,8,26], showing somatosensory and visual inputs inregions posterior to A1. The initial report of non-auditoryinputs [4] used multi-electrode recordings in awake mon-keys to show that somatosensory responses triggered byelectrical stimulation of the median nerve have approxi-mately the same latency as co-located auditory responses,and have a similar, although not identical, feedforwardlaminar profile. That is, the response begins in Lamina 4and is followed by responses in the supra- and infra-granular laminae [27]. The laminar profile of these inputscontrasts strongly with that of nearby visual inputs [5,27].These inputs have a feedback profile, in that, activitybegins outside of Lamina 4, typically in the supragranularlaminae, and then spreads to Lamina 4 [27]. In addition, astudy [8] used microelectrode recordings in anesthetizedmonkeys to confirm that convergence occurs at the singleneuron level [12,13], and showed that, although cutaneous,proprioceptive and vibratory inputs are present, the domi-nant specific type of input is a cutaneous representationbiased toward the skin surfaces of the head and neck.Current Opinion in Neurobiology 2005, 15:454–458 www.sciencedirect.comCorresponding studies of specific visual properties havenot yet been conducted, although other studies (e.g.[11,25]) predict strong motion sensitivity.These results fit into a complex of earlier findings onsomatosensory inputs into the region of posterior auditorycortex in macaque monkeys. Leinonen et al.[28] reportedauditory–somatosensory co-representation in Area Tpt,the parabelt region occupying the posterior most portionof the superior temporal plane in macaque monkeys. Alsoin macaques, Robinson and Burton [29] described a bodymap in a medial retro insular (RI) region of the superiortemporal plane, in a location just medial to the caudo-medial (CM) belt region of auditory cortex; our ongoingstudies (P Lakatos and CE Schroeder, unpublished) showthat RI receives auditory inputs. Subsequently, Krubitzeret al.[30] suggested the existence of one or more bodysurface maps in a region they referred to as ventralsomatosensory area (VS), which adjoins the medial edge(foot representation) of parietal operculum areas S2 andPV and extends out