Local field potential in cortical area MT: stimulus tuning and behavioral correlations

Abstract

Low-frequency electrical signals like those that compose the local field potential (LFP) can be detected at substantial distances from their point of origin within the brain. It is thus unclear how useful the LFP might be for assessing local function, for example, on the spatial scale of cortical columns. We addressed this problem by comparing speed and direction tuning of LFPs obtained from middle temporal area MT with the tuning of multiunit (MU) activity recorded simultaneously. We found that the LFP can be well tuned for speed and direction and is highly correlated with that of MU activity, particularly for frequencies at and above the gamma band. LFP tuning is substantially poorer for lower frequencies, although tuning for direction extends to lower frequencies than does tuning for speed. Our data suggest that LFP signals at and above the gamma band reflect neural processing on the spatial scale of cortical columns, within a few hundred micrometers of the electrode tip. Consistent with this notion, we also found that frequencies at and above the gamma band measured during a speed discrimination task exhibit an effect known as "choice probability," which reveals a particularly close relationship between neural activity and behavioral choices. In the LFP, this signature of the perceptual choice comprises a shift in relative power from low-frequency bands (alpha and beta) to the gamma band. It remains to be determined how LFP choice probability, which is a temporal signature, is related to conventional choice probability effects observed in spike rates.

Figures

Figure 1.

Schematic illustration of the behavioral tasks. A, Fixation task. The monkey was required to maintain fixation on the central cross while tuning curves were measured using visual stimuli positioned in the MU RF. B, Spatial 2-AFC speed discrimination task. The monkey fixated the central cross for 500 ms at the beginning of the trial, and the visual stimuli then appeared for 1 s. After the disappearance of the visual stimuli, the two saccade targets appeared inside the visual stimulus apertures. After another 200 ms, the fixation point disappeared, giving the monkey permission to execute the operant saccade.

Figure 2.

A, Filter characteristics (adapted from Krohn-Hite model 3384 user’s manual). The attenuation is plotted against frequencies normalized to the high- or low-cutoff frequency. B, The LFP signal in one trial during the measurement of a speed tuning curve. The top left panel shows the raw voltage trace; the bottom left panel shows the same trace with 60 Hz noise removed. The two vertical lines indicate stimulus onset and offset. The right panels illustrate the power spectra of the two single traces depicted in the left panels. C, Average power for all speed tuning data. Power was computed within 5 Hz bins for each individual trial, averaged over all trials within a block, normalized by the peak power within the block, and then averaged across all sites for which speed tuning curves were collected. We excluded sites that showed strong inhibitory effects (Fig. 8, Table 2), sites that were not tuned for speed, and sites in which the high-cutoff frequency for the LFP exceeded 150 Hz. The black trace represents the power spectrum during visual stimulus presentation. The gray trace represents the power spectrum of the spontaneous activity. Error bars indicate the SEM. Note that the spontaneous activity was also normalized by the peak of the spectrum during visual stimulus presentation, this is why the spontaneous activity is never close to unity. Each data point is the average power within a 5 Hz window. D, The dashed line is the modulation of LFP power attributable to the visual stimulus, which is simply the difference between the two traces in C. The scale of this trace is indicated on the y-axis at the left margin of the graph. The black line is the ratio of the two traces in C, which is the visual response gain as a function of frequency. The y-axis at the right margin of the plot shows the scale for the ratio values.

Figure 3.

An example of simultaneously recorded SU, MU, and LFP tuning curves. A, SU speed (left) and direction (right) tuning curves. B, MU tuning curves. C, LFP tuning curves. D, An illustration of the “iceberg effect.” We defined the width of a tuning curve as the full width of the curve at 90% of the best response, i.e., 0.9 × (Rmax − Rspont). More pronounced visual modulation (the black curve compared with the gray curve) increases Rmax − Rspont; consequently, the tuning width also becomes larger. MUA, MU activity; SUA, SU activity; A.U., arbitrary units.

Figure 4.

A, Comparison of direction tuning properties between MU and LFP recordings. Top, Preferred directions; bottom, tuning widths. B, A comparison of speed tuning properties between MU and LFP recordings; only bandpass sites are included in this analysis. Top, Preferred speeds; bottom, tuning widths.

Figure 5.

Speed tuning curves of different LFP frequency bands at a speed-tuned site, plotted as the total power within each frequency band as a function of stimulus speed. (Note that the y-axis scale differs among the plots.) We subtracted the spontaneous activity in each trial from the visually evoked activity before plotting the tuning curve. A.U., Arbitrary units.

Figure 6.

A, The percentage of speed-tuned sites among all sites in which LFP signals were recorded, as a function of frequency. The grayscale value (indicated in the color scale below the grid) of each pixel depicts the percentage of speed-tuned sites within a frequency band defined by the low-cutoff value indicated on the x-axis and the high-cutoff value indicated on the y-axis. B, The percentage of direction-tuned sites plotted in the same format as in A. In both A and B, each square represents the value of one frequency band, with the low-cutoff frequency indicated on the y-axis and the high-cutoff frequency indicated on the x-axis.

Figure 7.

Comparison of tuning parameters derived from MU data and from narrow spectral bands (20 Hz) within the LFP. Adjacent bands were 10 Hz apart. The x-axis indicates the center of each frequency band. A, Percentage difference between the preferred speeds derived from LFP and from MU activity, calculated as the absolute difference between preferred speed of LFP and preferred speed of MU, divided by the preferred speed of MU. Error bars indicate the SEM. B, Correlation coefficients of LFP preferred speeds and MU preferred speeds. The gray data points were not significantly different from 0 (p > 0.05). The insets show scatter plots of the MU preferred speed against LFP preferred speed, for one low-frequency and one high-frequency band. C, Mean absolute difference between the preferred directions derived from LFP and from MU data, with SEs. D, Correlation coefficients of LFP preferred directions and MU preferred directions. The gray data points were not significantly different from 0 (p > 0.05). The insets show scatter plots of the MU preferred directions against LFP preferred direction, for one low-frequency and one high-frequency band.

Figure 8.

A, A speed tuning site in which the LFP was inhibited at nonpreferred speeds. The MU tuning curve has the same shape as LFP, but there is no inhibition at any speed. B, The LFP and MU direction tuning curves recorded at the same site. The two curves are similar in shape, but the LFP curve is inhibitory at all directions. The format of A and B are the same as in Figure 3, in which the dashed lines indicate spontaneous activity. A.U., Arbitrary units; MUA, MU activity.

Figure 9.

Comparison of CP values calculated from LFP and MU data. A, LFP signals (0–150 Hz) did not exhibit significant CP (mean of 0.498). B, LFP signals above 40 Hz exhibit significant CP (mean of 0.533), which is also correlated with CP values calculated from MU data. C, CP values calculated from narrow frequency sub-bands within the LFP. Each band is 20 Hz wide, and adjacent bands were separated by 10 Hz. x-Axis indicates the center of each frequency band. The black data points are significantly different from 0.5 (t test, p < 0.05). D, CP values calculated from the relative power (the total power of each trial, summed across frequency bands, was normalized to unity) of narrow frequency sub-bands within the LFP. The format is the same as in C.