Comparing the feature selectivity of the gamma-band of the local field potential and the underlying spiking activity in primate visual cortex

Abstract

The local field potential (LFP), comprised of low-frequency extra-cellular voltage fluctuations, has been used extensively to study the mechanisms of brain function. In particular, oscillations in the gamma-band (30-90 Hz) are ubiquitous in the cortex of many species during various cognitive processes. Surprisingly little is known about the underlying biophysical processes generating this signal. Here, we examine the relationship of the local field potential to the activity of localized populations of neurons by simultaneously recording spiking activity and LFP from the primary visual cortex (V1) of awake, behaving macaques. The spatial organization of orientation tuning and ocular dominance in this area provides an excellent opportunity to study this question, because orientation tuning is organized at a scale around one order of magnitude finer than the size of ocular dominance columns. While we find a surprisingly weak correlation between the preferred orientation of multi-unit activity and gamma-band LFP recorded on the same tetrode, there is a strong correlation between the ocular preferences of both signals. Given the spatial arrangement of orientation tuning and ocular dominance, this leads us to conclude that the gamma-band of the LFP seems to sample an area considerably larger than orientation columns. Rather, its spatial resolution lies at the scale of ocular dominance columns.

Keywords: feature selectivity; local field potential; macaque; multi unit activity; primary visual cortex; spatial resolution.

Figure 1

Example of LFP (A and B) and MU activity (C and D). Two orthogonal stimulus conditions recorded at the same site are shown. Panels (A and C) and (B and D) show identical trials of the two signals, respectively. Vertical lines indicate onset and offset of visual stimulation. During the stimulation period clear gamma-band oscillations in the LFP can be seen. However, in (B) these are much more prominent then in (A). Note that MU response strength is higher in (C) than in (D).

Figure 2

LFP spectral properties. (A) Example of the power spectrum of a single typical recording site during baseline activity (–300 to stimulus onset, dashed line, light grey error bars) and visual stimulation (200–500 ms after stimulus onset, solid lines). The two solid lines show spectra under stimulation with two gratings with different orientations (dark grey: 22°, medium grey: 90°). This site shows stronger gamma-oscillations in response to the former orientation. All shaded regions indicate one standard error of the mean. Site is from dataset A2. (B) Power difference spectrum between stimulus and fixation demonstrating a prominent increase in power in the gamma-band (30–70 Hz) at the same site as shown in (A). (C) Population average over the relative power of the LFP spectra under stimulation to baseline for all sites in dataset A2. (D) Mean relative power of all sites (n = 196) in the gamma-band for all used data groups. Error bars show standard error of the mean.

Figure 3

Orientation tuning properties of the LFP. (A) Mean tuning index d′ as a function of frequency for the dataset A2 (n = 96, circles). This tuning index is an indication of the discriminability between the preferred and the orthogonal orientation (see Materials and Methods). Also shown is a shuffle corrected tuning index (squares), where trials were randomly assigned to conditions. A tuning index that is significantly greater than chance is observed in the gamma-band. (B) Mean fraction of tuned sites in the gamma-band. The left grey bar of each dataset shows the percentage for the original data, while the right white bar was obtained from the shuffled data. The number was computed in a 10 Hz band around the frequency with maximal tuning index. (C) Mean fraction of tuned sites in frequency bands over 100 Hz. Colors as in (B). (D) Mean tuning index d′ averaged over all sites in the same bands used for (B) (gamma band). (E) Comparison of the tuning index of the gamma-band (x-axis) with tuning index at frequencies over 100 Hz. The cross indicates the mean. Note that it lies well below the main diagonal. Therefore the gamma-band is on average more tuned than the frequencies over 100 Hz. (F) Comparison of the tuning index of the gamma-band (x-axis) with tuning index of the event-related potential. The cross indicates the mean. Note that it lies well below the main diagonal. Therefore the gamma-band is on average more tuned than the event-related potential.

Figure 4

Tuning strength of LFP compared to MU activity. Scatter plot of the tuning index d′ for the MU activity and LFP obtained from the same recording site. Tuning strength for the LFP was computed in the best tuned frequency band in the gamma range. The cross indicates the population average, showing that the MU activity is more strongly tuned than the gamma band of the LFP. Marginal histograms are shown on the left and the bottom.

Figure 5

Preferred orientation of the LFP compared with MU activity. Scatter plot of the preferred orientations of the gamma-band power of the LFP and MU activity recorded from the same site obtained by fitting a circular Gaussian tuning function (see Materials and Methods). White circles indicate data from monkey A (datasets A1 and A2), while black circles indicate monkey B (dataset B1). Marginal histograms are shown on the left and the bottom. The inner histograms show data from monkey A, the outer from monkey B. While the preferred directions of the LFP signal cluster around a mean value, MU preferred directions are distributed across the entire space of orientations (for a detailed discussion, see text).

Figure 6

Examples of MU and LFP tuning functions and relationship of tuning strength to preference difference. (A and B) Two typical examples of tuning functions of MU (top row) and LFP (bottom row). Site (A) shows similar tuning curves for MU and LFP, whereas site (B) has nearly orthogonal preferred orientations for the two signals. Note also that the relative increase from minimal to maximal activity is larger in both MU tuning curves. (C) Population histogram of the difference in preferred orientations between MU and LFP. While many sites have close tuning, there is a large fraction with nearly orthogonal tuning as well. Small arrows labeled “close” and “far” indicate the groups used in (D). (D) Population histogram of d′ as a function of the difference in preferred orientations between LFP and MU (“close”, <20°, dark grey and “far”, >60°, light grey). Median d′ of the LFP is indicated by arrows and is larger for close than for far tuned neurons (for a detailed description, see main text). (E) Scatter plot of preferred orientations of LFP and single units isolated from the same tetrode as in Figure 5. Both are uncorrelated (for a detailed discussion, see text).

Figure 7

Correlation as a function of frequency. (A) Correlation between the preferred orientation of the MU and the LFP is shown as a function of frequency. Note that the correlation increases over at 100 Hz and saturates over 150 Hz at 0.8. Bins at 50 and 150 Hz are confounded by possible line noise and therefore very noisy. (B and C) Scatter plot of the preferred orientation of MU against the preferred orientation of the LFP at 125 and 195 Hz. Note the clearly visible increase in correlation from Figure 5. Most points fall along the main diagonal. Points in the lower right and upper left corner are also well correlated because of the circular nature of the distribution. The number of points is not the same as in Figure 5, because not so many LFP sites are tuned in these frequency bands.

Figure 8

Ocular dominance analysis. (A) Scatter plot of the ocular dominance index (ODI) of MU against the ODI of the LFP in the 32–55 Hz frequency range. While the correlation is not perfect, a linear trend is clearly (All sites: Spearman's ρ = 0.61, p < 10⁻²⁰; Monkey A: ρ = 0.51, p = 5 × 10⁻⁶; Monkey C, ρ = 0.65, p = 2.4 × 10⁻⁹). The black line shows the least square fit to the data of both monkeys. LFPs are generally less tuned to ocularity than MU activity. Grey dots represent data from monkey A, black dots from monkey C. (B) Correlation of the ODI between MU activity and LFP as a function of frequency. The correlation increases to a maximal level of about 0.6, saturating for frequencies greater than 60 Hz. Also compare to Figure 7A. The different spatial scales seem to be reflected in the saturation point as well.