Figure 2. The difference in neural activity between hit and miss trials can be partly explained by consistent hit-associated increases in activity of specific neurons, and somewhat better by trial-by-trial population-wide fluctuations, but these mean-based approaches fall short of being fully descriptive.

(a) Recorded traces from 10 randomly selected neurons over four subsequent trials. For further analysis, we took the mean dF/F₀ per neuron over the visual stimulation period (thick colored lines) as single mean neural activity measure per trial. (b) Data of one entire recording block consisting of 74 tuned and simultaneously recorded neurons over 336 trials. Blue rectangle shows the four trials depicted in panel (a). (c) In an example animal, the detection of stimuli (green) with test contrasts (0.5–32%) correlated with a modest increase in preferred population dF/F₀ over undetected stimuli (red) (two-sample t-test, p<0.05) but none of the individual contrasts reached statistical significance (resp. vs. no resp., two-sample t-tests, FDR-corrected p>0.05). (d) As (c), but for mean over all animals the graph shows a small, but consistent overall difference of dF/F₀ with visual detection (test contrasts 0.5–32%, n=8 animals, p<0.05). (c, d) Shaded areas show the standard error of the mean. (e) The hit-associated increase in neural activity per neuron for all hit trials of test contrast stimuli (panel I) can be partly explained by specific neurons showing consistent dF/F₀ increases or decreases across trials (panel II) partly by trial-by-trial population-wide fluctuations regardless of neuronal identity (panel III) and somewhat better by both (panel IV). (f) Control analysis by shuffling neuronal identities (IDs) per trial (n=1000 iterations, black distribution) shows that the population activity is more predictable based on consistent hit modulations per neuron (top panel) and more neurons are significantly hit-modulated (bottom panel) than can be expected by chance. (g) Analyses as in (f), but across animals; comparison versus shuffle-based R²-expectation showed above-chance (at α=0.05) predictability of hit modulations using neuron ID, trial ID or both for respectively 7/8, 8/8, and 8/8 animals (left panel). The fraction of significantly hit-modulated neurons was above chance (at α=0.05) for 7/8 animals (right panel).

DOI: http://dx.doi.org/10.7554/eLife.10163.005

Figure 2—figure supplement 1. Several control analyses reveal no confounding effects of motor preparation, modulatory feedback, and eye movements.

(a–f) Traces of population dF/F₀ averaged over animals (n=8) for different contrasts show the difference between response and no-response trials over time. (a) Probe trials (0% stimulus contrast). Note the absence of any neuronal activity during responses to 0% contrast probe trials (green line, t-test vs. 0, p=0.549) and (b) 0.5% contrast, (c) 2% contrast, (d) 8% contrast, (e) 32% contrast, and (f) 100% contrast. (a–f) Response trials (green) show quicker offsets because the stimulus turns off when the animal makes a licking response. Also note the absence of any motor-related neural activity in panel (a) supporting the interpretation that the observed correlates are unrelated to motor activity, preparation or reward expectancy. (g, h) Removal of locomotion trials does not qualitatively affect neural correlates of stimulus detection. (g) When computed only on trials where animals were not moving (89.9% of trials), the mean population dF/F₀ as a function of stimulus contrast shows little difference with the original analysis (compare to Figure 2d). Paired t-test over test contrasts (0.5–32%) showed a significant difference between response and no-response trials (p<0.05). Our original results are very similar to the current analyses and are therefore not dependent on movement-induced modulations. (h) As (g) but for heterogeneity (still trials only); the overall paired t-test for hit/miss differences grouping 0.5–32% contrasts was highly significant (p<0.001), suggesting that our main results are not due to locomotion-related artifacts. (i, j) Population correlates of visual detection are not dependent on motor-related and/or feedback signals. (I) As Figure 2d; mean population dF/F₀ as a function of contrast, now using only the first ~400 ms (394 ms; 10 frames) after stimulus onset. Mean reaction time over animals and contrasts was ~1.2 s (see Figure 1f) leaving on average about 0.8 s between the last data point included in this analysis and the subsequent licking response.(j) As (i) but for heterogeneity (compare with Figure 3d). Results were qualitatively similar to our original analysis for heterogeneity, but not for dF/F₀ (paired t-test over intermediate contrasts for dF/F₀, p=0.543, for heterogeneity, p<0.005). (k, l) Population correlates are not explained by eye blinks or saccades. (k) As Figure 2d; mean population dF/F₀ as a function of contrast using only trials in which the animal’s eye position remained fixed and no blinks were detected during the entire stimulus period. (l) As (k) but for heterogeneity (compare with Figure 3d). Our results are qualitatively and quantitatively similar for dF/F₀ and heterogeneity (hit–miss paired t-test over test contrasts; p<0.05 and p<0.005, respectively). All panels: shaded areas show the standard error of the mean. Asterisks indicate statistical significance: *p<0.05; **p<0.005; ***p<0.001.

DOI: http://dx.doi.org/10.7554/eLife.10163.006