Single neuron responses to perceptual difficulty in the mouse auditory cortex

Abstract

Perceptual learning leads to improvement in behavioral performance, yet how the brain supports challenging perceptual demands is unknown. We used two photon imaging in the mouse primary auditory cortex during behavior in a Go-NoGo task designed to test perceptual difficulty. Using general linear model analysis, we found a subset of neurons that increased their responses during high perceptual demands. Single neurons increased their responses to both Go and NoGo sounds when mice were engaged in the more difficult perceptual discrimination. This increased responsiveness contributes to enhanced cortical network discriminability for the learned sounds. Under passive listening conditions, the same neurons responded weaker to the more similar sound pairs of the difficult task, and the training protocol by itself induced specific suppression to the learned sounds. Our findings identify how neuronal activity in auditory cortex is modulated during high perceptual demands, which is a fundamental feature associated with perceptual improvement.

Fig. 1.. PL—Protocol, behavior, and imaging.

(A) Schematic illustration of the experimental setup. Created using Biorender.com. (B) Trial structure of the Go-NoGo task. Stimulus duration is 100 ms, followed by a 2-s response window. CR, correct reject; FA, false alarm. (C) Schematic illustration of the PL protocol. Mice start by training to discriminate an easy pair of sounds (light blue box, novice). Then, the Go stimulus is gradually changed toward the NoGo sound until discrimination performance is slightly above the perceptual threshold (expert). (D) Block-switching protocol. The protocol is composed of alternating blocks (~50 trials each) of easy and hard discriminations. Black and white pixel sequences represent the Go and NoGo sounds during the easy task, respectively. Gray and white pixel sequences represent the Go and NoGo sounds during the hard task, respectively. (E) Representative example from one mouse illustrating its behavioral d′ during the block-switching protocol, total of 10 blocks. Error trials are indicated below. (F) Average behavioral d′ in easy and hard blocks (n = 6 mice; 47 easy and 47 hard blocks, easy: 3.02 ± 0.75, hard: 1.1 ± 0.68, mean ± SD). P = 2.7 × 10⁻²², unpaired t test. (G) Average behavioral d′ in the first and second halves of the easy and hard blocks. Easy: 2.55 ± 0.72 and 2.82 ± 0.72 for first and second halves, respectively. P = 7.8 ×10⁻⁴, paired t test. Hard: 1.05 ± 0.68 and 1.17 ± 0.94 for first and second halves, respectively. P = 0.39, paired t test. Easy vs. Hard: P = 3 × 10⁻¹⁷ and P = 2 × 10⁻¹⁵, for first and second halves, respectively. (H) Representative calcium responses of single neurons (n = 125) evoked by the NoGo stimulus from one mouse engaged in the task during an easy and hard discrimination blocks (E and H, respectively). (I) Same as (H), but in the passive listening session of the same mouse. n.s., not significant. **P < 0.01, ***P < 0.001.

Fig. 2.. GLM—Design and results.

(A) Schematic description of the encoding model used to quantify the relationship between behavioral variables and average neuronal activity. (B) Relative contribution of each predictor to the explained variance of neural activity for each neuron. (C) Examples of predicted and actual ΔF/F signal for four neurons in each trial type and difficulty level and the relative contribution of each predicator to the model. Neuron 1 is mainly responsive to “easy GO,” neuron 2 is mainly responsive to “hard GO,” neuron 3 is responsive to the NoGO (the punishment variable contributes in the negative direction), and neuron 4 is an example for the contribution of perceptual difficulty. Scale, 0.04 ∆F/F, 1 s. (D) Histogram of the difficulty coefficient of all neurons that were modulated significantly by it. Pie chart showing the proportion of neurons with positive and negative coefficient (red and blue, respectively). Most neurons increase their response to the NoGo stimulus in the hard blocks. (E) Heatmaps of the average responses of all neurons with positive difficulty coefficient that were detected by the model. Below: Average response of all the 366 neurons in each trial type and difficulty level. Scale, 0.02 ∆F/F, 1 s.

Fig. 3.. Increased perceptual difficulty is accompanied by higher responsiveness—Analysis of responses to NoGo sounds.

(A) A representative example of calcium responses from a neuron during CR trials in the hard (H) and easy (E) blocks along the session. Scale, 0.1 ∆F/F, 1 s. (B) Calcium responses to the NoGo sound in CR trials during the easy and hard blocks. Neurons that changed their responses significantly are marked red. The arrow points to the neuron shown in (A). (C) On average, more neurons increased their responses during the hard task. P = 1.36 × 10⁻⁴⁴, paired t test. (D) Twenty-seven percent of neurons respond statistically stronger during the hard task while, only 1% responds weaker. (E) Representative example of single neuron responses to the “NoGo” sound during the passive listening session (pseudo-hard and pseudo-easy blocks are represented by dark and light blue, respectively). Scale, 0.1 ∆F/F, 1 s. (F) Calcium responses to the NoGo sound during the passively listening protocol in the pseudo-easy (x axis) and pseudo-hard (y axis) protocol. Neurons that changed their responses significantly are marked red. The arrow points to the neuron shown in (E). (G) On average, more neurons decreased their responses during the pseudo-hard task. P = 0.01, paired t test. (H) Sixteen percent of neurons show suppression in the pseudo-hard protocol, while only 1% shows stronger responses. *P < 0.05, ***P < 0.001.

Fig. 4.. Increased perceptual difficulty is accompanied by higher responsiveness—Analysis of responses to Go sounds.

(A) Representative example of calcium responses from a neuron during hit trials in the hard (dark blue) and easy (light blue) blocks along the session. Scale, 0.1∆F/F, 1 s. (B) Calcium responses of individual neurons to the Go sound in hit trials during the easy and hard blocks. Neurons that changed their responses significantly are marked red. The arrow points to the neuron shown in (A). (C) On average, neurons increased their responses during the hard task. P = 9.5 × 10⁻⁷, paired t test. (D) Pie chart describing the category of changes. Twenty-seven percent of neurons respond statistically stronger during the hard task, and 15% respond weaker. (E) Representative example of single-neuron responses to the Go sound during the passive listening session (pseudo-hard and pseudo-easy blocks are represented by dark and light blue, respectively). Scale, 0.1 ∆F/F, 1 s. (F) Calcium responses to the “Go” sound during the passively listening protocol in the pseudo-easy (x axis) and pseudo-hard (y axis) protocol. Neurons that changed their responses significantly are marked red. The arrow points to the neuron shown in (E). (G) On average, neuronal responses in the pseudo-hard protocol are weaker. P = 1.1 × 10⁻¹², paired t test. (H) Pie chart describing the category of changes. Thirty percent of neurons show suppression in the pseudo-hard protocol, while only 6% show stronger responses. (I) Venn diagram describing the fraction and overlap of neurons and their respective changes between easy and hard blocks in the task engaged state. (J) Venn diagram describing the fraction and overlap of neurons and their respective changes between pseud-easy and pseudo-hard blocks in the passive listening state. ***P < 0.001.

Fig. 5.. The subset of neurons with increased responsiveness improves discriminability.

(A) On average, single-neuron discriminability between the Go and NoGo is higher in the hard blocks as compared to the easy blocks. Data from mice during the engaged state. P = 4.6 × 10⁻²⁵, paired t test. (B) Representative examples of calcium responses from two neurons. Top two traces: A neuron that did not increase its discriminability yet has higher responses in CR trials during the hard task; bottom two traces: a neuron that increased its discriminability and also has higher responses to the Go sounds in the hard blocks. The traces show responses during CR and hit trials during the hard and easy blocks along the session. Scale: 0.1 ∆F/F, 1 s. (C) Subset of neurons that increased their responsiveness in hit and CR trials during the hard task (red in Figs. 3B and 4B) have statistically higher DI values. Easy versus hard: P = 8.2 × 10⁻⁴, 3.3 × 10⁻²⁵, and 5.3 × 10⁻¹¹ (paired t test for unmodulated, increased hit, and CR, respectively). Within easy: P > 0.05, unpaired t test. Within hard: P = 7.2 × 10⁻¹⁴ and 1.3 × 10⁻⁸ (unmodulated versus increased hit and CR, respectively). Increased hit versus increased CR: P > 0.05, unpaired t test. (D) On average, single-neuron discriminability in the hard blocks is significantly lower from that in the easy blocks during the passive state. P = 0.017, paired t test. (E) SVM decoder accuracy for discrimination between hit and CR trials with and without the subset of increasing neurons in hit and CR trials. All neurons, easy versus hard: P = 5.3 × 10⁻⁸⁷. All neurons, hard versus without random neurons, P = 0.64. Without “increased” versus all neurons/without random neurons, P = 6.3 × 10⁻³⁰ and P = 6.5 × 10⁻⁷, respectively, unpaired t test. Black horizontal lines indicate the distributions’ means. ***P < 0.001.

Fig. 6.. Sound exposure during training suppresses responses in ACx.

(A) Top: Representative micrographs from one mouse as a novice and as an expert (width: 260 μm). Middle: Illustration of the NoGo sound pair that we compared in (B) to (D). Bottom: Representative example of a calcium response in CR trials for a novice and an expert in one mouse. Scale: 0.2∆F/F, 1 s. (B) Single-neuron responses to the NoGo sound in CR trials in novices and experts (n = 3 mice). Red denotes significantly different responses. Arrow denotes the neuron whose traces are shown in (A). (C) Neuronal response is lower in experts. P = 2 × 10⁻¹³, paired t test. (D) Distribution of changes between novices and experts. (E) Experimental timeline before discrimination training, corresponding to (F) and (G). “i,” imaging session. (F) Representative example of a neuron responding to the Go sound and the surrounding stimuli during “baseline” and association stages. Note the specific suppression of the response to the Go stimulus (5.65 kHz). Scale: 0.5∆F/F, 1 s. (G) Average response of all neurons (n = 414, six mice) to the Go and surrounding stimuli for baseline and association stages. Bottom: Fraction of neurons that significantly increased (red) or decreased (blue) their response between baseline 2 and association stages. P= 2 × 10⁻³⁸, paired t test. (H) Experimental timeline of the initial discrimination training, corresponding to (I) and (J). (I) Representative example of responses from two neurons to the Go and NoGo sounds in association (green) and novice (blue) stages. Scale: 0.5(“1”) or 0.2(“2”)∆F/F, 1 s. (J) Average response of neurons (n = 414, six mice) to the Go and NoGo sounds in association and novice stages. Responses to the Go sound are not different. Responses to the NoGo sound are suppressed. P= 1.3 × 10⁻²⁴, paired t test. ***P < 0.001.