Figure 3. Increasing target activation is matched by background network suppression.
(a) The proportion of neurons activated across all neurons (targets and background) increases as more target neurons are activated. Inset right: average activation across all trial types is increased on stimulus trials compared to catch. (b) The proportion of neurons suppressed across all neurons (targets and background) increases as more target neurons are activated. Inset right: average suppression across all trial types is increased on stimulus trials compared to catch. (c) The ratio of activation and suppression is similar to that observed on catch trials (inset right) and is not strongly modulated by the number of activated target neurons. (d) Stimulation of target neurons causes mild activation of background neurons (targets excluded; inset right) but this is not modulated by the number of target neurons activated. (e) Stimulation of target neurons causes suppression of background neurons (targets excluded; inset right) which increases as more target neurons are activated. All data are hit:miss matched to remove potential lick signals (see Figure 3—figure supplement 1 and Materials and methods). For all plots N = 11 sessions, 6 mice, 1–2 sessions each. Some trial types from some sessions are excluded for having too few hits or misses to be able to match the hit:miss ratio. Error bars and shading are s.e.m; data points, error bars and linear fits are stimulus trials, shading is catch trials; grey data points: individual trial types, individual sessions; coloured error bars: data averaged within trial type (number of target zones) across sessions; linear fits are to individual data points; fits are reported ± 95% confidence intervals.
Figure 3—figure supplement 1. Comparison of network activity on hits and misses for both threshold go trials and catch trials in an effort to quantify and account for lick responses.
(a) The proportion of background neurons activated when different numbers of target neurons are activated on hits/false alarms (green) and misses/correct rejects (grey) on stimulus trials (error bars) and catch trials (shading). Background neurons become more active on stimulus hits and catch false alarms and there does not seem to be much modulation of the amount of background activation by the number of activated target neurons. Note that the background neuron activation response on stimulus trials overlaps with the response on catch trials across the full range of number of target neurons activated. (b) The proportion of activated background neurons is higher on stimulus hit trials than miss trials. Note that we used the 50 target zone stimulation trial type as it has a roughly equal number of hits and misses. (c) The proportion of activated background neurons is greater on catch false alarms (FA; when animals lick) then on catch correct rejects (CR; when the animal do not lick). (d) The proportion of activated background neurons does not differ between 50 target zone stimulus trial hits and catch trial false alarms raising the concern that this response is due to licking, not photostimulation. (e) The proportion of background neurons suppressed when different numbers of target neurons are activated. Same colour conventions as in (a). Stimulus hits are associated with decreased levels of background suppression than misses, and this appears to be somewhat modulated by the number of activated target neurons. Catch trial false alarms are associated with marginally more suppression than catch trial correct rejections. (f) The proportion of suppressed background neurons is lower on hits than misses. (g) The proportion of suppressed background neurons does not differ significantly between false alarms and correct rejects on catch trials. (h) The proportion of suppressed background neurons does not differ between 50 target zone stimulus trial hits and catch trial false alarms. (i) Fluorescence trace from an example neuron with photostimulation trial epochs removed (black trace) with licks denoted below (red ticks). This neuron had a lick correlation of 0.14. The fluorescence trace has been Gaussian smoothed (σ = 1 s) for display only. (j) Distribution correlation values across all neurons recorded. (k) Distribution of p-values for correlation values reported in (j). 46 ± 11% of neurons are lick modulated (α = 0.05). (l) Top: Average traces from neurons binned into 10 equal bins by their lick correlation (trace colours) aligned to the onset of spontaneous licking (red line). Bottom: average across all spontaneous lick bouts. (m) Lick – fluorescence cross-correlation for all lick-responsive neurons positively modulated (yellow) or negatively modulated (blue) by licking. (n) Distribution of the time of maximum |cross-correlation| for positively modulated (yellow) and negatively modulated (blue) neurons in a 4 s window centred on the 0 time-lag of the cross-correlation. p=2.75 × 10−160 Mann Whitney U-Test, N = 9547 positively lick-modulated neurons and 4365 negatively modulated neurons. (o) Schematic illustrating the procedure used to ensure a 50:50 ratio of hits:misses for each trial type. This is done to remove the effect of increased recruitment of licking, and thus lick evoked neural responses, by stimulating target ensembles of increasing size. For a given trial type, say five target zones, we find the number of hits and misses and find the minority response type. For five target zone trials, which aren’t reliably detected by animals, this will likely be hits. We then take all hit trials and combine them with random resamples of miss trials of the same number as hit trials. For example if we only have two hits and six misses then for each random resample we would take all two hits and two random misses. In this way, we ensure that we have an equal number of hit and miss trials. We then calculate neuron response metrics across the subset of hits and misses in each resample and average these metrics across resamples. All error bars and shading are s.e.m. N = 11 sessions, 6 mice, 1–2 sessions each.
Figure 3—figure supplement 2. Neuropil subtraction has a small effect on response amplitude but it is not the sole cause of negative going responses.
(a) Distribution of trial-wise responses on 50+ neuron stimulation trials calculated from raw (Raw; black) and neuropil subtracted (Sub; red) traces (N = 30,894 neurons, 1312 trials). A large fraction of negative responses are observed even in raw traces without neuropil subtraction. Inset: Difference between each neuron’s trial-wise response when responses were calculated from neuropil subtracted or raw traces (Sub – Raw). Responses are slightly reduced when traces are neuropil subtracted (−0.16 ± 0.58 ΔF/σF, p=0 Wilcoxon signed-rank test). (b) Neuropil-subtracted trial-wise responses are highly correlated with raw trial-wise responses (R2 = 0.87, p=0), although slightly reduced (β0 = -0.10 ± 5.32 x 10−4, p=0) with this reduction mainly limited to large positive going responses (β1 = 0.82 x 10−1±3.31 × 10−4, p=0). Red: linear fit to data, Black: neuropil-subtracted data binned into deciles by raw trial-wise response and averaged within decile. Data are mean ± s.d. (c) Trial-wise responses classified as suppressed, activated and no response using neuropil-subtracted traces also show negative (−1.75 ± 1.05 ΔF/σF, p=0 Wilcoxon signed-rank test), roughly zero (0.16 ± 0.94 ΔF/σF, p=0, Wilcoxon signed-rank test) and positive (3.72 ± 3.18 ΔF/σF, p=0 Wilcoxon signed-rank test) trial-wise responses calculated from raw traces in the same trials. These groups also differed significantly from each other (p=0 Kruskal-Wallis test; Suppressed vs No response, p=0, Suppressed vs Activated, p=0, No response vs Activated, p=0, Bonferroni corrected for multiple comparisons) indicating that our thresholds on neuropil subtracted traces reliably separate different classes of trial-wise responses even in the absence of neuropil-subtraction. (d) Example raw (light red, light blue) and neuropil-subtracted (dark red, dark blue) traces for neurons reliably activated (red) and suppressed (blue) across trials (using neuropil-subtracted trial-wise magnitudes). Neuropil subtraction slightly decreases response magnitudes in both groups, though does not induce large changes in response time-course. Negative responses can be seen with and without neuropil-subtraction. Shading is mean ± s.e.m across neurons. Traces have been linearly interpolated across the photostimulation epoch (grey box) to avoid the photostimulation artefact for display only. All box-plots are median (mid-line) with 25th and 75th percentiles (box) and 5th and 95th percentiles (whiskers). N = 3.68 × 106 trial-wise responses, 30894 neurons, 11 sessions, 6 mice, 1–2 sessions each.
Figure 3—figure supplement 3. Activation and suppression have different spatial profiles.
(a) Lateral and axial spatial profile of activation (left) and suppression (right) relative to nearest target site co-ordinate: for each neuron plot its distance to nearest target site co-ordinate vs the proportion of trials it was activated or suppressed. Bin spatial distances and then average across all neurons in each bin. Average this value across stimulus trial types. (b) Quantification of spread of activation and suppression with lateral distance, collapsed across all axial planes. All data in all panels are hit:miss matched (see Figure 3—figure supplement 1 and Materials and methods) and have had catch trial values subtracted. All tests in (b) are Wilcoxon signed-rank tests, Bonferroni corrected for the number of lateral bins. * denotes p<0.05. N = 10 sessions, 6 mice, 1–2 sessions each. Note one session was excluded as there were no hits on catch trials so we were unable to subtract a hit:miss matched neural response rate on catch trials from the stimulus trial data. Error bars are s.e.m. and spatial bin widths are 10 µm.