Transient inhibition to light explains stronger V1 responses to dark stimuli

Neurons in the primary visual cortex (V1) receive excitation and inhibition from distinct parallel pathways processing lightness (ON) and darkness (OFF). V1 neurons overall respond more strongly to dark than light stimuli, consistent with a preponderance of darker regions in natural images, as well as human psychophysics. However, it has been unclear whether this “dark-dominance” is due to more excitation from the OFF pathway or more inhibition from the ON pathway. To understand the mechanisms behind dark-dominance, we record electrophysiological responses of individual simple-type V1 neurons to natural image stimuli and then train biologically inspired convolutional neural networks to predict the neurons’ responses. Analyzing a sample of 74 neurons (in anesthetized, paralyzed cats) has revealed their responses to be more driven by dark than light stimuli, consistent with previous investigations. We show that this asymmetry is predominantly due to slower inhibition to dark stimuli rather than to stronger excitation from the thalamocortical OFF pathway. Consistent with dark-dominant neurons having faster responses than light-dominant neurons, we find dark-dominance to solely occur in the early latencies of neurons’ responses. Neurons that are strongly dark-dominated also tend to be less orientation selective. This novel approach gives us new insight into the dark-dominance phenomenon and provides an avenue to address new questions about excitatory and inhibitory integration in cortical neurons. Significance Neurons in the early visual cortex respond on average more strongly to dark than to light stimuli, but the mechanisms behind this bias have been unclear. Here we address this issue by combining single-unit electrophysiology with a novel machine learning model to analyze neurons’ responses to natural image stimuli in primary visual cortex. Using these techniques, we find slower inhibition to light than to dark stimuli to be the leading mechanism behind stronger dark responses. This slower inhibition to light might help explain other empirical findings, such as why orientation selectivity is weaker at earlier response latencies. These results demonstrate how imbalances in excitation vs. inhibition can give rise to response asymmetries in cortical neuron responses.


Introduction
provided. Topical phenylephrine hydrochloride (2.5%) and atropine sulfate (1%), or 149 cyclopentolate (1.0 %) in later experiments, were administered daily.  In total, 110 single units from 37 penetrations in 8 cats (4 males, 4 females) were 166 analyzed. These recording experiments involved lab personnel working on other projects. 167 Out of these neurons, 6 were rejected because part of their receptive fields was outside 168 the screen, and 30 were rejected because the predictive performance of the fitted model   For the subsequent data analysis (described below), all images were resized from 195 480x480 to 40x40 before training (see below) to avoid overparameterization of the fitted model. Resizing was done using the Image module from the Python Image Library (PIL; 197 Umesh, 2012).   The stimulus images are convolved with a pair of parametrized 2D gaussian filters 231 (with positive or negative polarity for the ON and OFF pathways, respectively), each 232 followed with a half-wave rectification (ReLU). The 2D gaussians represent receptive 233 fields of LGN neurons in which the weaker surrounds (Croner & Kaplan, 1995) are 234 neglected, as follows:    The data is separated into training, validation, and test sets, corresponding to the 302 three sets of stimulus movies. The model parameters are fit to the training set using a 303 mini-batch size of 100 stimulus-response pairs. As an additional regularization measure, 304 training is stopped if there is no improvement on the validation set in the preceding 50 305 epochs -then we use the model at its peak performance (i.e. 50 epochs before training 306 stops) in subsequent analyses. We use a third, separate test set to obtain an unbiased 307 estimate of predictive performance. pass training procedure, with each pass improving the spatial resolution of the receptive 316 field estimate. In the first pass, we optimize the model parameters using the full 480x480 317 stimulus images downsampled to 40x40. We then manually designate a square cropping 318 window that encloses an area slightly larger than the apparent receptive field. Next, we 319 crop the stimulus images within that window, and rescale each image within it to 40x40.

320
Due to the resizing, this cropped image then has much better spatial resolution than the 321 40x40 image from the first-pass. This image is used to re-train the model in the second 322 pass, where we repeat the procedure, but with the cropped image. In the third pass, we  This three-pass training procedure allows us to characterize a neuron's receptive field 327 with high resolution and substantially increases predictive performance.   The latency t with the highest @6AB? will be referred to as T.
This index varies from -1.0 to 1.0, with positive LDB values indicating a neuron 391 is light-dominated, and negative values that it is dark-dominated.

392
Excitation/Inhibition balance 395 The excitation/inhibition balance (EIB) index is similar to LDB, but contrasts 396 excitation with inhibition instead of light with dark: conditions. The 40x40 stimuli were tailored to each neuron's spatial receptive field, 408 which we estimated by using Recon(i,j,T) (Eq. 13) at each neuron's peak latency T (see  These responses were used to compute the orientation selectivity of each neuron using a 432 vector summation method (Wörgötter & Eysel, 1987; Swindale, 1998): where N is the number of sinewave gratings, xi is the orientation angle, and R(xi) 438 represents the simulated responses. The orientation selectivity index, OS, was computed 439 separately for each latency in individual neurons.   observe at the optimal latency instead seems to be due to a strong imbalance between ON 528 and OFF inhibition: while inhibition is on average 37.95% stronger from the ON than 529 from the OFF pathway ( Figure 3C), there is no significant difference between excitation 530 from the ON and OFF pathways ( Figure 3D; t = 0.81, df = 73, p = 0.42). The difference 531 between ON and OFF inhibition ( Figure 3C) is also significantly stronger (t = 2.96, df = 532 73, p = 0.0042) than the difference between ON and OFF excitation ( Figure 3D).

533
In addition, whether a given neuron is light-or dark-dominated is strongly related 534 to whether ON inhibition exceeds OFF inhibition ( Figure 3C, red points above 1:1 line 535 vs. blue points below). However, the imbalance of ON vs. OFF excitation poorly predicts 536 whether a neuron is light or dark-dominant ( Figure 3D). Overall, these results suggest the dark-dominance effect to be more driven by an imbalance in ON/OFF inhibition than by 538 an imbalance in ON/OFF excitation.

633
These findings suggest that inhibition is slower to dark than light stimuli, which leads to 634 dark-dominance at the 13.3-26.7 ms latency.  Figure 6A). But as we hypothesized, we do obtain dark-665 dominance at the 13-27 ms latency with stimuli that both excite and inhibit the neuron's 666 response (conditions 3 and 4, Figure 6B). These results support the idea that dark-

705
We next investigate the relationship between orientation selectivity and light-dark 706 balance at each neuron's optimal latency, which can be seen in Figure 7B. Neurons  Using a novel model-fitting approach to natural image responses, we find V1 747 neurons respond more strongly to dark than to light stimuli at early but not at later 748 latencies, due to slower inhibition to dark than light stimuli. Dark-dominance occurs 749 when inhibition is differentially recruited, for example when there is a light stimulus on 750 the dark-driven region of a neuron's receptive field (or vice-versa). As can be seen in Inference of excitation and inhibition from model-fitting 763 We use a machine learning algorithm to fit a model based on   Since this study utilized recordings from polytrodes that did not extend across all 813 the cortical layers, a laminar analysis was not feasible. A useful future direction could be 814 to replicate this experiment with linear-array probes to obtain simultaneous recording 815 across all V1 layers, to investigate the laminar dependence of dark-dominance.  In conclusion, we use a novel machine learning approach to bring new insights to 863 the phenomenon of stronger dark responses in visual cortex neurons. We find the dark-864 dominance effect to only occur in the early latencies, and to be due to slower inhibition to 865 dark stimuli. We also show how weaker average inhibition to dark stimuli is related to