Specificity and randomness in the visual cortexKenichi Ohki and R Clay ReidResearch on the functional anatomy of visual cortical circuitshas recently zoomed in from the macroscopic level to themicroscopic. High-resolution functional imaging has revealedthat the functional architecture of orientation maps in highermammals is built with single-cell precision. By contrast,orientation selectivity in rodents is dispersed on visual cortex ina salt-and-pepper fashion, despite highly tuned visualresponses. Recent studies of synaptic physiology indicate thatthere are disjoint subnetworks of interconnected cells in therodent visual cortex. These intermingled subnetworks,described in vitro, may relate to the intermingled ensembles ofcells tuned to different orientations, described in vivo. Thishypothesis may soon be tested with new anatomictechniques that promise to reveal the detailed wiring diagram ofcortical circuits.AddressesDepartment of Neurobiology, Harvard Medical School,220 Longwood Avenue, Boston, MA 02115, USACorresponding author: Reid, R Clay ([email protected])Current Opinion in Neurobiology 2007, 17:401–407This review comes from a themed issue onSensory systemsEdited by Peter Mombaerts and Tony ZadorAvailable online 27th August 20070959-4388/$ – see front matter# 2007 Elsevier Ltd. All rights reserved.DOI 10.1016/j.conb.2007.07.007IntroductionOver the past 50 years, the visual cortex has served as amodel system for the study of cerebr al cortical circuits.Several themes have dominated this literature: hierarch-ical versus recurrent processing, specific versus randomsynaptic connectivity, and functional architecture versusintermingling of response types. A classical view of visualcortical processing [1] has concentrated on one side ofeach of these dichotomies that the visual cortex is bestunderstood as a hierarchical system, whose receptivefields are created through specific connections, within aframework of crystalline functional architecture. At var-ious times, however, each of these views has been calledinto question.Orientation selectivity and hierarchyThe hierarchical model of visual processing was pro-posed at the beginning of mo dern studies of the visualcortex, when it was proposed that selectivity for stimulu sorientat ion emer ges from the specific connecti ons fromthalamus to cortical simple cells [2]. In the cat visualcortex, there is evidence that multiple thalamic afferents,each of which is substantially unoriented, add togethe r toproduce a stro ngly oriented afferent inp ut to simple cellsin layer 4 [3,4]. This evidence com es from cross-corre-latio n studies [5] as w ell as studies of ori entat ion selec-tivity of thalamo-rec ipient cells when the intr acortica lcircuit has been silenced [6–8].While an increasing body of evidence has sho wn thatthalamo-cortical connections are related to the establish-ment of orientation selectivity, the role of intracorticalconnections has been much more difficult to study. Cross-correlation techni ques have show n that simple cells inlayer 4 connect to iso-orientation complex cells [9], butthe detailed logic of these connections is not clear. Thereis new evidence that simple cells predominate in layer 4and complex cells in other cortical layers [10], but therelative impo rtance of feedforward, recurrent, and feed-back connections in the cortical circuit is still vigorouslydebated. Without newer tools to study the relationshipbetween intracortical synaptic connections and visualresponse properties, it will be, however, difficult toresolve these issues.While there is strong physio logical evidence for the rol e ofthalamo-cortical input in orientation selectivity, a moredirect anatomical proof has been elusive. In some species,however, the main thalamo-recipien t cells in layer 4 arenot orientation selective, but cells in the next stage—layer 2/3 are orientation selective. Their orientation selec-tivity must develop through intracortical connectionsfrom layer 4 to layer 2/3. In an elegant recent study ofone such species, the tree shrew, an elongated topo-graphic organization of connections from layer 4 to layer2/3 correlated well with the emergence of orientationspecificity [11].Coarse-grained and fine-grained specificity ofintracortical connectionsSince the Hubel and Wiesel’s model was first put for th,the specificity of the anatomical connections has beenstudi ed at i ncreasingly fine levels of detail: from arealmaps to func tional or laminar architecture, fine-sc alegeometry, and finally individual synaptic connections.At the coarsest l evels a re cortic al regions, w hich ofte nhave large-scale m aps , such as the retinotopy of the visualcortex. At the next level i s functional architecture, inwhich different features can be segreg ated at a scale oftens to hundreds of microns [1]. The afferent con nectionsto cortex often respect thes e functional boundaries, suchwww.sciencedirect.com Current Opinion in Neurobiology 2007, 17:401–407as i n the ocular dominance columns in the input layersof primate visual cortex [12], or in the cortico-corticalprojections between different functional comp artments[13,14]. Many studies have also demonstrated spe cificityin intracortical connections, either b etween corticallayers [15,16 ] or in long-range co nnection s within layers[17,18].The next finer scale has been termed the geometrical levelat which the close proximity of axons and dendrites areconsidered on the scale of !0.5–2 mm [19,20",21]. At thislevel and the next – that of actual synaptic connect ions –the debate about specificity has been couched in terms ofPeters’ rule [19,22,23], that is, axons make connectionsrandomly in direct proportion to the occurrence of allsynaptic targets in the adjacent neuropil, with no localspecificity.Recently, several fine-scale geometrical analyses havebeen performed for cortical circuits. For pyramidalneurons in somatosensory cortex, at least one directapposition of axons and dendrites (regardless of theexistence of a synapse) was observed for every pair ofpyramidal neurons sharing the same cortical columnwithin 300 mm [20",21]. This supports Peters’ rule forthe geometrical contact between axons and dendrites.But this level can only reveal potential connectivity:Actual synaptic boutons were found in only a fractionof these potential connectivities. The number of synapticboutons was correlated to the synapt ic responses of pairsof neurons [20"].A version of Peters’