Dissociation of task engagement and arousal effects in auditory cortex and midbrain
- Oregon Hearing Research Center, Oregon Health and Science University, United States
- Neuroscience Graduate Program, Oregon Health and Science University, United States
- Department of Mathematics and Statistics, Washington State University, United States
Figures
Figure 1 with 1 supplement
Pupil size correlates with task engagement.
(A) Schematic of behavioral setup, including free-field speaker for sound presentation, piezo spout to register licks and deliver reward, and infrared video camera for pupillometry. (B) Spectrogram of example trial (top) and task structure (bottom). False alarms (FAs) before the target window were punished with a timeout and hits resulted in a liquid reward. (C) Swarm plot showing behavioral performance (hit rate or FA rate) for each animal. Each point corresponds to a behavioral block. Red lines indicate mean and standard error of the mean (SEM) of the sensitivity (d') for each animal. All animals performed above chance on average (d’ > 1). (D) Mean time course of pupil size aligned to trial onset for active and passive blocks during neurophysiological recordings, averaged across all animals and recording days, normalized to maximum size per day. Shading indicates SEM. Gray dashed lines indicate sound onset. (E) Average time course of change in pupil size, normalized to 0.35 s pre-trial period for active and passive behavioral blocks. Shading indicates SEM. Gray dashed lines indicate sound onset. (F) Scatter plot compares pre-trial pupil size and mean change per trial (3 s post-onset window). Each dot represents a behavioral block. Passive and active blocks within the same session are connected by a line.
Figure 1—figure supplement 1
Pupil size correlates with task engagement and performance.
(A) Histogram of pupil size during passive and active blocks, averaged across all animals and physiological recording sessions (n = 35 sessions). Pupil size is normalized to maximum per session. (B) Mean time course of pupil size during behavior, grouped according to trial outcome (hit, miss, or false alarm [FA]), averaged across sessions. Mean pupil size was not significantly different for FA and hit trials (p = 0.220, hierarchical bootstrap), but it was smaller during miss trials (p = 0.0183). Shading indicates SEM. (C) Trial time course of average pupil change during active behavioral blocks and grouped according to performance, plotted as in B. Evoked pupil size did not differ between any of the trial outcome conditions (p > 0.05, hierarchical bootstrap).
Figure 2
Simultaneously recorded units showing task versus pupil-related changes.
(A) Firing rate (gray) of one unit in primary auditory cortex (A1) and time-aligned pupil size (green) during reference stimuli over one entire experiment (total recording time ~1 hr, inter-trial and target periods removed). Purple shading highlights active blocks. Dashed lines delineate boundaries between half of each behavioral block. (B) Peri-stimulus time histogram (PSTH) responses to reference noise averaged separately across the first and second half of each behavior block (black). Dashed gray lines are responses predicted by the null model, i.e., the PSTH averaged across all blocks. Model prediction accuracy is indicated by fraction variance explained (r2). (C) Top: PSTH response for the example unit averaged across passive and active blocks (left, dark and light purple, solid lines) and averaged across large and small pupil size trials (right, dark and light green, solid lines). Predictions of the task-only model are overlaid (same color pattern, dashed lines). Middle: Active/passive and large/small PSTH responses plotted with predictions by the pupil-only model, which only accounts for pupil-related changes. Bottom: Active/passive and large/small PSTH responses plotted with predictions by the full model, accounting for both task and pupil-related changes. (D) Firing rate and pupillometry for a second A1 unit, recorded simultaneously to the first example. (E and F) PSTH responses for the second unit, plotted as in B and C.
Figure 3 with 1 supplement
Task and pupil-related modulation of firing rates in primary auditory cortex (A1) and inferior colliculus (IC).
(A) Doughnut plots indicate number of units significantly modulated by task engagement, pupil size, or both in A1 (left) and IC (right, p < 0.05, jackknifed t-test). Total number of recorded units reported in the center. Purple and green: significant unique modulation by task or pupil, respectively. Black: unique modulation by both task and pupil. Dark gray: ambiguous task or pupil modulation. Light gray: no significant change in accuracy between full and null models. (B) Scatter plot of variance explained (r2) in single-trial activity by full model versus null model. Each symbol represents a unit in AC (left) or IC (right). Colors as in A. (C) Unique variance explained by pupil size (horizontal axis) plotted against unique variance explained by task engagement (vertical axis) for each unit. Stars correspond to examples in Figure 2A (purple) and D (green).
Figure 3—figure supplement 1
Unique effects of task and pupil on neural activity did not differ across single units (SU) and multiunits (MU).
(A) Unique variance explained by task separated for units categorized as MU (dark blue circles) and SU (light blue circles) in primary auditory cortex (A1) and inferior colliculus (IC). Median variance explained was not different between unit categories in either area (A1, SU median: 0.011, MU median: 0.010, p = 0.374; IC, SU median: 0.016, MU median: 0.021, p = 0.410, hierarchical bootstrap). (B) Data for unique variance explained by pupil, plotted as in A. Again, median effects were not different between unit categories in either area (A1, SU median: 0.006, MU median: 0.006, p = 0.453; IC, SU median: 0.015, MU median: 0.018, p = 0.583, hierarchical bootstrap).
Figure 4 with 1 supplement
Task- and pupil-related modulation of firing rate as a function of behavioral performance in primary auditory cortex (A1) and inferior colliculus (IC).
(A) Scatter plot of unique variance explained by task plotted against behavioral sensitivity (d’) in A1 (left) and IC (right). Each point represents a unit/target pair. When a unit was recorded across blocks with different target conditions, d’ and variance explained were measured separately in each block. Different colors refer to different animal subjects. Points with the same d’ align on the horizontal axis because they belong to experiments in which multiple units were recorded at the same time using an array (0.015 standard deviation jitter added to d' to facilitate visualization). Regression analysis shows a significant correlation between unique variance explained by task and performance in A1 (n = 132 unit/target pairs, r = 0.303, *p = 0.007, hierarchical bootstrap) but not in IC (n = 85 unit/target pairs, r = 0.069, p = 0.318). (B) Scatter plot of unique variance explained by pupil plotted against performance for each unit in A1 (left) and IC (right) as in A. Regression analysis shows no significant correlation between unique variance explained by pupil and performance in either A1 (r = 0.113, p = 0.158) or IC (r = −0.17, p = 0.552).
Figure 4—figure supplement 1
Variability of neural behavior state effects and task performance across animals.
(A) Unique variance explained by task engagement (r2 task-unique) in primary auditory cortex (A1; left) and inferior colliculus (IC; right) for each unit, grouped by animal. Mean variance explained did not differ significantly between animals in either area (A1 animal R: 0.025, T: 0.023, B: 0.017, F = 0.606, p = 0.547; IC animal R: 0.028, B: 0.021, L: 0.022, F = 0.221, p = 0.803, ANOVA). (B) Task performance (d’) for each behavior block, grouped by animal and recordings in A1 (left) and IC (right). Mean d’ differed between animals in both groups of experiments (A1 animal R: 1.55, T: 1.35, B: 0.93, F = 15.8, p < 0.0001; IC animal R: 1.52, B: 1.61, L: 2.48, F = 23.9, p < 0.0001, ANOVA). Interactions between animal and d’ on neural behavior state effects are reported in Figure 4 in the main text. (C) Unique variance explained by changes in pupil size (r2 pupil-unique), plotted as in A. Mean variance explained differed between animals in A1, but not in IC (A1 animal R: 0.011, T: 0.030, B: 0.012, F = 10.9, p < 0.0001; IC animal R: 0.018, B: 0.038, L: 0.022, F = 1.82, p = 0.170, ANOVA). (D) Scatter plots show r2 pupil-unique as a function of variance of pupil size during each experiment. This relationship is significant in both A1 (left, r = 0.341, p ≤ 10−5, hierarchical bootstrap) and IC (right, r = 0.409, p = 0.02). Dashed line shows regression fit. Multivariate regression shows that r2 pupil-unique depends significantly on both animal and pupil variance but not on d’ in A1 (animal: F = 11.5, p = 0.000021; pupil variance: F = 11.4, p = 0.00092; d’: F = 0.145, p = 0.703, ANOVA). In IC, r2 pupil-unique depends significantly only on pupil variance (animal: F = 2.23, p = 0.116; pupil variance: F = 16.2, p = 0.00016; d’: F = 0.032, p = 0.86, ANOVA).
Figure 5 with 1 supplement
Changes in pupil size account for apparent task engagement effects.
(A) Active versus passive modulation index, MIAP task-only, computed from responses predicted by the task only model, in which pupil size is shuffled, plotted against MIAP task-unique, in which pupil-dependent modulation is regressed out, for primary auditory cortex (A1; left, n = 132 unit/target pairs) and inferior colliculus (IC; right; circles for NCIC, n = 24 unit/target pairs, triangles for ICC units, n = 61). Colors indicate significant unique variance explained by one or two state variables, as in Figure 3 (p < 0.05, jackknife t-test). (B) Overlaid histograms of MIAP task-only and MIAP task-unique, sorted according to their magnitude for each unit in A1 and IC. Accounting for pupil size reduced the absolute magnitude of MIAP by about 33% in both areas (A1: p < 10−5; IC: p = 0.0140, hierarchical bootstrap). The prevalence of gray shading on the right side of the horizontal axis indicates that units with large, positive MIAP were most affected by this adjustment, and MIAP shifted to more negative values on average (A1 median MIAP task-only = 0.069, MIAP task-unique = 0.027, p = 0.0005; IC: median MIAP task-only = −0.010, MIAP task-unique = −0.037, p = 0.049, hierarchical bootstrap).
Figure 5—figure supplement 1
Task-related changes in central versus external inferior colliculus.
Histograms of MIAP task-only and MIAP task-unique sorted according to their magnitude for each unit in ICC (left) and NCIC (right). The median magnitude of MIAP task-unique was reduced by 43% relative to MIAP task-only in ICC (p = 0.0020, hierarchical bootstrap). Units in NCIC also showed a trend toward reduction, but the decrease was not significant (median decrease 31%, p = 0.231). Among ICC units with significant unique task modulation, n = 9/13 showed increased activity (MIAP task-unique > 0) during behavior, and the median was significantly different from zero (median 0.012, p = 0.010, hierarchical bootstrap). Among NCIC units, only n = 8/36 were enhanced during behavior, but median was not significantly different from zero (median −0.049, p = 0.07).
Figure 6 with 2 supplements
Unique task and pupil modulate independent neural populations.
Scatter plot compares firing rate modulation attributed to task engagement (MIAP task-unique) against modulation attributed to pupil (MILS pupil-unique) for each unit in primary auditory cortex (A1; left) and inferior colliculus (IC; right). The two quantities showed a weak negative correlation in A1 (r = −0.281, p = 0.040, hierarchical bootstrap) and were uncorrelated in IC (r = −0.104, p = 0.191). Colors indicate significant model performance as in Figure 3, and stars indicate examples from Figure 2.
Figure 6—figure supplement 1
Frequency tuning versus task-related modulation of neural activity.
Strip plots compare task-unique changes in activity (MIAP task-unique) between units with best frequency similar to the target tone (on-BF,<0.5 octave difference, blue) and different from the target tone (off-BF,>0.5 octave difference, orange) in primary auditory cortex (A1; left) and inferior colliculus (IC; right). Data are further broken down by units with no task-unique changes (left) and significant changes (p < 0.05, permutation test, right). In A1, there was no difference in median MIAP task-unique between modulated on- and off-BF units (on-BF: median 0.082, n = 28; off-BF: −0.012, n = 24; p = 0.221, hierarchical bootstrap) or unmodulated units (on-BF: −0.002, n = 44; off-BF: −0.019, n = 77; p = 0.136). Similarly, in IC, there were no differences between on- and off-BF units in the modulated (on-BF: median 0.05, n = 9; off-BF: 0.088, n = 6; p = 0.771, hierarchical bootstrap) or unmodulated groups (on-BF: 0.006, n = 25; off-BF: 0.005, n = 27; p = 0.347). The relatively small number of significantly modulated IC units makes it difficult to interpret the absence of a significant on- versus off-BF difference (Slee and David, 2015), although it confirms that units with task-unique modulation are relatively rare in IC (Figure 3A).
Figure 6—figure supplement 2
Task difficulty versus task-related modulation of neural activity.
Swarm plots compare task-unique changes in activity (MIAP task-unique) broken down by task difficulty, area (primary auditory cortex [A1] or inferior colliculus [IC]) and significance of task-unique modulation, plotted as in Figure 6—figure supplement 1. We found no difference in MIAP task-unique between any pair of difficulty conditions in any group (p>0.05, hierarchical bootstrap).
Figure 7 with 1 supplement
Task- and pupil-related modulation in primary auditory cortex (A1) and inferior colliculus (IC).
Mean variance in firing rate explained by behavioral state variables, broken down by portions attributed to task (purple), attributed to pupil (green), and not uniquely attributable to either state variable (gray) for A1 (left) and IC (right). Data are grouped in quintiles by r2 for the null model performance, a measure of auditory responsiveness. Variance explained by state was not correlated with null model performance in A1 (r = −0.134, p = 0.130, hierarchical bootstrap), but these values were negatively correlated in IC (r = −0.323, p = 0.0129). Colors as in Figure 3.
Figure 7—figure supplement 1
Auditory responsiveness predicts state-dependent modulation in inferior colliculus (IC).
(A) Scatter plots compare variance explained by the null model, a measure of auditory responsiveness for each neuron, against additional variance explained by the full model, which incorporates modulation by task engagement and pupil size. In primary auditory cortex (A1; left), the additional variance explained by the full model is uncorrelated with null model performance (n = 129, r = −0.134, p = 0.130, hierarchical bootstrap). In IC, the change is negatively correlated (IC: n = 66, r = −0.323, p = 0.0129), indicating a larger behavior effect in units with weaker auditory responses. Dashed diagonal line shows the theoretical limit on additional variance explained by the full model. If the additional variance explained is normalized by this limit, the data are positively correlated in A1 (r = 0.301, p = 0.0360) and not correlated in IC (r = 0.124, p = 0.178). (B) Scatter plots compare log signal-to-noise ratio (SNR) of response evoked by reference noise-evoked, an alternative measure of auditory responsiveness, against additional variance explained by the full model. Again, there is no significant correlation in A1 (r = −0.121, p = 0.335), but the correlation is negative in IC (r = −0.252, p = 0.0367). (C) Variance explained by the null model versus additional variance explained by a behavior-dependent model for larger populations of neurons. Some data did not include pupillometry and the model accounted only for task engagement. Again, there was no correlation in A1 (n = 254, r = −0.106, p = 0.0761), but there is a negative correlation in IC (n = 201, r = −0.317, p < 10−5).
Figure 8
Pupil size explains some persistent task-related modulation.
(A) Top: activity of ICC unit and concurrent pupil size recorded over passive and active blocks, plotted as in Figure 2. Bottom: peri-stimulus time histogram (PSTH) response to reference noise averaged over each half block (black) with null model prediction overlaid (dashed gray). The response was enhanced slightly during the post-passive (P2) block relative to pre-passive (P1). Without accounting for effects of pupil size, this suggests a sustained firing rate increase after behavior, MIP1P2 block-only = 0.25. After accounting for changes in pupil, the task-related effect is reduced, MIP1P2 block-unique = 0.02. (B) Data from example NCIC unit, plotted as in A. The PSTH shows a weak, suppressive response to the reference stimuli, but a large increase in mean firing rate that persists into the first half of the post-passive block (MIP1P2 block-only = 0.37). Again, the persistent change in firing rate can be accounted for by the very large pupil during the first half of P2 (MIP1P2 block-unique = −0.04). (C) Scatter plot compares MIP1P2 block-only and MIP1P2 block-unique for primary auditory cortex (A1; left) and inferior colliculus (IC; right). Colors as in Figure 3. Accounting for pupil reduces MI attributed to persistent effects after behavior in both areas (A1: p < 10−5, IC: p < 10−5, hierarchical bootstrap).
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Dissociation of task engagement and arousal effects in auditory cortex and midbrain
eLife 10:e60153.
https://doi.org/10.7554/eLife.60153
