Dynamics of Ongoing Activity: Explanation of the Large Variability in Evoked Cortical Responses

See allHide authors and affiliations

Science  27 Sep 1996:
Vol. 273, Issue 5283, pp. 1868-1871
DOI: 10.1126/science.273.5283.1868


  • Fig. 1.

    Evoked activity in response to repetitive stimulation exhibits large variability. (A) Two individual responses (a and b) to a repeated visual stimulus [bottom trace in (B)]: The images (1a,b) show the activity in a 2 mm by 2 mm area of cortex, taken at different times from response onset. Activation above the mean level is coded in red, suppression in blue, as indicated by the color scale (right); full scale corresponds to a fractional change of ∼5 × 10−5). The small square in the first image marks the site, above the microelectrode, from which the optical traces (2a,b) were taken. Note the large variability in the evoked response, also reflected in the LFP (3a,b) and single-neuron spike trains (4a,b), both recorded simultaneously with the optical signals. The absence of slow components in the LFP is due to high-pass filtering above 3 Hz. (B) Average evoked response: The optical images and signals, LFP, and single-unit activity were averaged, triggered on the onset of 34 visual stimuli (drifting full-field grating) in the preferred orientation of the recorded unit.

  • Fig. 2.

    Cortical evoked activity is related to the initial state. (A) Scatter plot of optically measured evoked activity at a single cortical site 42 ms after response onset in 34 successive single trials versus the initial state at that site. Both axes have the same arbitrary units. The straight line depicts the result of linear regression (correlation coefficient R = 0.9). (B) Correlation coefficients [as in (A)] for all sites in the imaged cortical area. The arrow marks the site, selected in (A). The statistical significance of correlation is indicated by color. (C) Correlation between the evoked LFP 28 ms after response onset and the initial state. (D) Correlation between the evoked spike rate, measured over an interval of 35 ms centered around 28 ms after response onset, and the initial state. The correlations in (C) and (D) are between a single site (microelectrode recording) and all optically measured sites.

  • Fig. 3.

    Predicting the cortical evoked response. (A) A single-trial response to a stimulus was predicted by summing the reproducible response and the ongoing activity, approximated by the initial state. (B) Comparison of the predicted and measured responses. (Top trace) Averaged evoked response (34 trials), measured from a single optical channel above the microelectrode site (small square in top-left frame). (First row) Averaged evoked activity pattern (after subtraction of frame 0), shown at five different times after response onset, indicated by the arrows. All other rows show single-trial responses. (Second row) Initial state, approximating ongoing activity during the response. (Third row) Predicted response, obtained by adding the frames in the first and second rows. (Fourth row) Measured response.

  • Fig. 4.

    Quality of prediction of the response. (A) Three consecutive single-trial responses (1 through 3) to the same visual stimulus, showing the initial state, the measured response 28 ms later, and the predicted response at that time. Subtracting the initial state from the measured response yielded the net pattern [M − I]. (B) Quality of prediction, assessed by the correlation coefficient between predicted and optically measured activity patterns as a function of time from response onset. The curve shows the mean correlation; the error bars denote the standard error of the mean (n = 35 recording sessions). (C) Autocorrelation of optically measured activity patterns, triggered on the response onset (time 0). The right-hand curve shows the correlation coefficient between the ongoing activity at time 0 (just before response onset) and the evoked activity. The left-hand curve shows the correlation coefficient between the same ongoing activity at time 0 and the ongoing activity before stimulus onset. After calculating the correlation coefficient for each pixel in the matrix at a certain delay, we simply summed all the pixels (because we did not see any consistent temporal differences between the different pixels). The insets in (B) and (C) show the correlations over prolonged time.