Spatiotemporal Elements of Macaque V1 Receptive Fields

Nicole C. Rust, Odelia Schwartz, J. Anthony Movshon, Eero P. Simoncelli
2005 Neuron  
components from physiological data. For each cell, one would like to extract a set of relevant filters and the nonlinear rule by which their outputs are combined. Summary Simple cells can be analyzed using the well-known method of reverse correlation (Jones and Palmer, 1987; Neurons in primary visual cortex (V1) are commonly DeAngelis et al., 1993) . Specifically, an unbiased esticlassified as simple or complex based upon their senmate of the linear filter can be recovered by taking the
more » ... to the sign of stimulus contrast. The respatiotemporal average of the stimuli preceding spikessponses of both cell types can be described by a the spike-triggered average (STA). However, the nonlingeneral model in which the outputs of a set of linear earity of complex cells prevents analysis by reverse filters are nonlinearly combined. We estimated the correlation, since these cells are driven equally by a model for a population of V1 neurons by analyzing given stimulus and an otherwise identical stimulus of the mean and covariance of the spatiotemporal distriopposite contrast polarity. These stimuli cancel when bution of random bar stimuli that were associated averaged, and the STA is therefore flat. But an analysis with spikes. This analysis reveals an unsuspected of the spike-triggered covariance (STC) can resolve richness of neuronal computation within V1. Specifimechanisms that have this type of symmetric nonlinear cally, simple and complex cell responses are best deinfluence on response (de Ruyter van Steveninck and scribed using more linear filters than the one or two Bialek, 1988; Brenner et al., 2000; Simoncelli et al., found in standard models. Many filters revealed by 2004). STC analysis has successfully revealed some rethe model contribute suppressive signals that appear ceptive field elements of cat V1 complex cells (Touryan to have a predominantly divisive influence on neuet al., 2002) and macaque retinal ganglion cells (Schwartz ronal firing. Suppressive signals are especially potent et al., 2002). in direction-selective cells, where they reduce re- Here we use a combination of STA and STC analysis sponses to stimuli moving in the nonpreferred dito reveal unexpected structure within the receptive rection. fields of both simple and complex cells in macaque V1. In particular, the responses of both cell types are best Introduction described using more linear filters than the standard models predict. Moreover, some of the filters contribute Neurons in primary visual cortex are classically divided suppressive signals that appear to have a predomiinto two groups. Simple cells respond precisely to the nantly divisive influence on neuronal firing. These addilocation and contrast polarity of features in the visual tional filters have a significant impact on responses, scene, while complex cells measure the magnitude of even to simple stimuli. These findings extend and enlocal contrast without regard to the polarity or precise rich our understanding of the computations performed position of stimulus features (Hubel and Wiesel, 1962; by V1 cells and place additional constraints on biolog-Hubel and Wiesel, 1968). Simple receptive fields can be ically based models. A brief report of some of this work described by a single linear spatiotemporal filter whose has appeared elsewhere (Rust et al., 2004) . output is half-wave rectified and squared ( Figure 1A) (Movshon et al., 1978b; Heeger, 1992a) . Complex re-Results ceptive fields are economically described by two linear spatiotemporal filters whose outputs are squared and Recovering the Linear Filters summed (the "energy model," Figure 1B) (Movshon et We stimulated each neuron with a binary random bar al., 1978a; Adelson and Bergen, 1985; Spitzer and stimulus whose bars covered a region slightly larger Hochstein, 1985). The two cell types are characteristithan the classical receptive field and were aligned with cally found in different cortical layers and are thought the preferred orientation ( Figure 1D, left) . We assume to receive different patterns of input, suggesting that an "LNP" functional model consisting of a set of linear they form distinct groups (Hubel and Wiesel, 1962; Hu- filters (L), an instantaneous nonlinearity (N) that combel and Wiesel, 1968). Recent theory and experiments bines filter outputs to yield a rate, and a Poisson spike have shown, however, that many simple and complex generator (P) that converts the rate signal into spikes ( Figure 1C ). We assume that the filters operate within a 16 frame (160 ms) interval ( Figure 1D, middle) . The *Correspondence: rust@cns.nyu.edu 4 These authors contributed equally to this work. ensemble of these space-time stimulus intervals pre-Neuron 946 Figure 1. LNP Functional Models for V1 Neurons, and Their Characterization Using Spike-Triggered Analyses (A) A standard simple cell model, based on a single space-time oriented filter. The stimulus is convolved with the filter, and the output is passed through a half-wave rectifying and squaring nonlinearity. This signal determines the instantaneous rate of a Poisson spike generator. (B) The "energy model" of a complex cell, based on a pair of space-time oriented filters with a quadrature (90°) phase relationship (Adelson and Bergen, 1985). Each filter is convolved with the stimulus, and the responses are squared and summed. The resulting signal drives a Poisson spike generator. (C) The generalized linear-nonlinear-Poisson (LNP) response model used in this paper. The cell is described by a set of n linear filters (L), which can be excitatory (E) or suppressive (S). The model response is computed by first convolving each of the filters with the stimulus. An instantaneous nonlinearity (N) governs the combination of excitatory and suppressive signals that drives a Poisson spike generator (P). (D) Spike-triggered analysis. (Left panel) A schematic of the random binary bar stimulus used to drive V1 neurons. The bars were aligned with the neuron's preferred orientation, and the stimulus array spanned the classical receptive field. (Middle panel) A 2D representation of the stimulus sequence made by taking a slice through the stimulus ensemble in the plane orthogonal to the preferred orientation-each pixel represents the intensity of a bar at a particular location in one frame. The collection of stimulus blocks during the 16 frames (160 ms) before each spike (gray box) forms the spike-triggered stimulus distribution. (Right panel) The STA is a block of pixels, each corresponding to the average of the corresponding pixel values over the distribution. The STC is a matrix whose entries contain the average product of each pair of pixels after the mean has been projected out. See Experimental Procedures for details. Supplemental Data Geisler, W.S., and Albrecht, D.G. (1992). Cortical neurons: isolation of contrast gain control. Vision Res. 32, 1409-1410. The Supplemental Data for this article can be found at http:// www.neuron.org/cgi/content/full/46/6/945/DC1/. Heeger, D.J. (1992a). Half-squaring in responses of cat striate cells. Vis. Neurosci. 9, 427-443. Acknowledgments Heeger, D.J. (1992b). Normalization of cell responses in cat striate cortex. Vis. Neurosci. 9, 181-197. We are grateful to Neot Doron and Mian Hou for help with histology; low signals are enhanced by spatiotemporal luminance contrast in mayleh for help during the experiments; and to Adam Kohn and macaque V1. J. Neurophysiol. 93, 2263-2278. Jonathan Pillow for helpful discussions. This work was supported Hubel, D.H., and Wiesel, T.N. (1962). Receptive fields, binocular inby HHMI Investigatorships to E.P.S. and J.A.M. and by a grant to teraction and functional architecture in the cat's visual cortex. J. J.A.M. from the National Institutes of Health (EY 2017). Physiol. 160, 106-154.
doi:10.1016/j.neuron.2005.05.021 pmid:15953422 fatcat:bwqllw7lwnb3tfdp3tghtzzeim