Local origin of field potentials in visual cortex

Abstract

The local field potential (LFP) is increasingly used to measure the combined activity of neurons within a region of tissue. Yet, available estimates of the size of this region are highly disparate, ranging from several hundred microns to a few millimeters. To measure the size of this region directly, we used a combination of multielectrode recordings and optical imaging. We determined the orientation selectivity of stimulus-evoked LFP signals in primary visual cortex and were able to predict it on the basis of the surrounding map of orientation preference. The results show that > 95% of the LFP signal originates within 250 microm of the recording electrode. This quantitative estimate indicates that LFPs are more local than often recognized and provides a guide to the interpretation of the increasing number of studies that rest on LFP recordings.

Figure 1

Orientation selectivity of LFPs evoked by orientation noise stimuli in primary visual cortex. (A) Mean across orientations of the LFP response. Responses were aligned to grating onset (dashed line), and averaged across grating orientations. The oscillation ~30 ms after stimulus onset is largest because it is aligned with the occurrence of grating stimuli (excluding blank stimuli). (B) Tuned components of the LFP response for three stimulus orientations. Responses were aligned to the onset of a single orientation (dashed line), and the mean across orientations (shown in A) was subtracted. (C) Estimate of the orientation tuning curve. We applied Singular Value Decomposition to the responses and obtained the orientation tuning curve and the response time course plotted at the margins. (D) Amplitude of the frequency spectrum for the mean across orientations of the LFP response shown in A. (E) Same, for the tuned components of the LFP response shown in B. (F) Layout of the 10×10 electrode array (dots), aligned with the map of orientation preference measured with voltage-sensitive dyes. Hue indicates preferred orientation and saturation indicates tuning strength (legend). (G) Orientation tuning of LFP responses measured in the electrode array. Half of the sites were classified as strongly tuned (black) and the other half as weakly tuned (gray) based on the standard deviation across orientations. Rectangle: recording site shown in A–E. (H) Relation between orientation preference of LFPs and of the corresponding pixels in orientation preference map. (I) Relation between orientation preference of LFPs and simultaneously recorded multi-unit spike activity. To remove outliers, only sites with strong LFP tuning (shown in G) contributed to these scatter plots.

Figure 2

Predicting the orientation selectivity of LFPs based on the orientation preference map. (A–C) LFP tuning curves predicted from the map of orientation preference, for three values of σ, the radius of integration. For each value of σ, the region of integration is shown (top) together with the LFP tuning curves (bottom) predicted by the model (red). The measured LFP tuning curve is plotted for comparison (black). Dashed lines indicate preferred orientation determined by a cosine fit. (D) Correlation between predicted and measured LFP responses, across all orientations and recording sites, as a function of σ. Error bars indicate standard error of the mean (bootstrap tests). (E) Relation between orientation preference of predicted and measured LFPs at the optimal radius of integration (σ = 100 μm). To remove outliers, only sites with strong LFP tuning (defined in Figure 1) contributed to this analysis. (F) Predicted LFP orientation tuning curves (red) superimposed on measured LFP orientation tuning curves (gray and black) in the electrode array. Predictions were restricted to recording sites with reliable signal/noise ratios in the map of orientation preference. Rectangle indicates the recording site of panels A–C and of Figure 1A–E.

Figure 3

Comparison of LFP selectivity determined with orientation noise stimuli and contrast-reversing gratings. (A) Amplitude spectra of the responses to contrast-reversing gratings for three stimulus orientations, recorded at a single electrode. Stimuli reversed in contrast at 4 Hz, leading to LFP responses at 8 Hz (second harmonic, or F2 component). (B) The orientation tuning curves of F2 responses measured across the electrode array (black). The tuning curves for orientation noise stimuli are provided for comparison (gray). Rectangle indicates the recording site of panel A. (C) Relation between orientation preferences determined with orientation noise and with contrast-reversing gratings. To remove outliers, only well-tuned sites are compared in this analysis (top 75%, based on the standard deviation across orientations). (D) Distribution of ratios of tuning widths for the same set of sites. For each site, the width of the F2 tuning curve was divided by the width of the tuning curve obtained with orientation noise stimuli.

Figure 4

Comparison of tuning curves for evoked LFP responses and induced LFP activity. We examined brief intervals in the stimulus sequence, where gratings were followed by a blank. (A) Amplitude of the frequency spectrum for the evoked response. Vertical lines indicate the gamma band (25–90 Hz) (B) Selectivity of the evoked response, determined by averaging spectral amplitudes below (open circles) and within (filled squares) the gamma range. Error bars represent 95% bootstrap confidence intervals for the mean. (C–D) Same, for induced activity, i.e. oscillations whose amplitude is modulated by the stimulus. (E) Comparison of selectivity for induced LFP activity in the gamma band (black) and for evoked LFP responses (gray, same as in Figure 1G). Rectangle indicates the recording site shown in A–D. (F) Relation between orientation preferences of evoked responses obtained with random noise stimuli and orientation preferences of induced gamma band activity. Weakly-tuned recording sites were excluded from the comparison (conventions as in Figure 1). (G) Distribution of ratios of tuning widths for the same set of sites. For each site, the width of the tuning curve for induced activity was divided by the width of the tuning curve for evoked responses.