Audiovisual task switching rapidly modulates sound encoding in mouse auditory cortex

Curation statements for this article:
  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This study establishes for the first time that selective auditory attention reduces activity in the auditory cortex, similar to effects produced by engagement in a behavioral task. Moreover the study establishes the diversity of cortical modulations generated by attention.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

This article has been Reviewed by the following groups

Read the full article See related articles


In everyday behavior, sensory systems are in constant competition for attentional resources, but the cellular and circuit-level mechanisms of modality-selective attention remain largely uninvestigated. We conducted translaminar recordings in mouse auditory cortex (AC) during an audiovisual (AV) attention shifting task. Attending to sound elements in an AV stream reduced both pre-stimulus and stimulus-evoked spiking activity, primarily in deep-layer neurons and neurons without spectrotemporal tuning. Despite reduced spiking, stimulus decoder accuracy was preserved, suggesting improved sound encoding efficiency. Similarly, task-irrelevant mapping stimuli during inter-trial intervals evoked fewer spikes without impairing stimulus encoding, indicating that attentional modulation generalized beyond training stimuli. Importantly, spiking reductions predicted trial-to-trial behavioral accuracy during auditory attention, but not visual attention. Together, these findings suggest auditory attention facilitates sound discrimination by filtering sound-irrelevant background activity in AC, and that the deepest cortical layers serve as a hub for integrating extramodal contextual information.

Article activity feed

  1. Evaluation Summary:

    This study establishes for the first time that selective auditory attention reduces activity in the auditory cortex, similar to effects produced by engagement in a behavioral task. Moreover the study establishes the diversity of cortical modulations generated by attention.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

  2. Reviewer #1 (Public Review):

    In this study the authors have investigated the modulation of auditory coding in the mouse auditory cortex related to the shift of attention between sounds and visual stimuli in a sensory discrimination task.

    They employ a previously developed behavioral paradigm, which was used to investigate the same question in the mouse visual cortex. Mice most of the time receive both sounds and visual objects to guide their decision but the task alternates between blocks where sounds are informative and blocks where vision is actually informative. Multielectrode recordings are used to probe the activity of auditory cortex neurons during task performance. These recordings evidence that despite a large variety of effects in cortical neurons, overall the population is inhibited during the attentional shift towards sounds. This effect is similar to what is observed during task performance, if compared to non-engaged, passively listening animals. However this paper is the first one to demonstrate the effect during a real attentional shift independent of engagement. It is also interesting that despite lower firing rates the level of information stays the same, indicating that attention renders cortical activity more specific. Finally one important results is the fact that attention impacts activity outside the stimulus window, pointing at a network state change, rather than a focally targeted effect.

    The results are solid and detailed. The analysis across layers and spike waveform types is appreciable. One may regret that no population analysis is done as maybe some results can be buried in the noise of single neurons. Yet, the present analysis is sufficient for the conclusions drawn.

    One analysis seems to be missing however to connect these results with other studies. The authors have not really analysed modulation with respect to tuning. Are the neurons specifically tuned to one of the two expected stimuli modulated in a direction different from the rest of the neuronal population? If so this could reconcile the present findings with studies which found that attention boosts stimulus response.

  3. Reviewer #2 (Public Review):

    This paper will be of interest to neuroscientists studying mechanisms of selective attention, and those interested in how active listening shapes auditory processing. Animals are engaged in an across modality task switching paradigm with the key result being that actively attending to sounds causes widespread suppression in auditory cortex.

    In this study mice are trained in a modality-switching task, learning a simple go-no go paradigm with sounds (high versus low frequency tone clouds) and visual stimuli (the direction of moving gratings). Both tasks are clearly suprathreshold (when unisensory) and performed to a high level in both unisensory blocks and in blocks where both modalities are presented, with the to-be-attended modality cued by the modality of the preceding unisenory block. Impressively, the mice can switch between modalities within a session, allowing the authors to examine single unit responses to both selective attention conditions.

    Single unit activity is recorded across the cortical depth during task performance. The across-modality attention switching allows the same sensory stimuli to be presented in two contexts - one in which it is to be ignored, and the other in which it is to be attended. This raises this study above previous attempts to understand how active listening shapes cortical processing, where the comparison is animals doing a task to those not doing a task - here in both cases the animal is actively engaging with stimuli, motivated to perform a task and being rewarded for doing so.

    A decoding approach is used to determine how well the two acoustic categories can be distinguished in both attend-a and attend-v conditions. Their key finding is that overall activity is suppressed in auditory cortex during attention to sounds. This confirms and builds on several previous studies that reached the same conclusion by comparing neural activity during passive listening and active task performance. The authors extend these findings taking an information theory approach to ask whether neural responses remain equally informative in both conditions, and using an 'information per spike' metric of efficiency demonstrate that a subpopulation of neurons in the deeper layers become more efficient in their representation of information (that is, they convey equivalent information with fewer spikes).

    Overall the work included is technically demanding and performed to a high standard. I was surprised that the population of neurons in deep auditory cortex in which firing rates and information content increased were excluded from subsequent analysis with the authors choosing to focus on the 'efficiency' improvement in the suppressed population. I would like to see the analysis in figure 8, which seeks to link changes in firing rate to behaviour, repeated for this population. An alternative explanation (that is potentially supported by the observation in S1 that the firing rate changes are driven mostly by untuned neurons) is that during attention relevant neural populations are enhanced such that their activity is better read out and that homeostatic mechanisms act to suppress other, less informative, activity. It seems likely the authors have the data to rule this in or out.

    I would like to see the behavioral data presented in a way that better demonstrates how well the mice were selectively attending to the relevant modality. Figure 2 presents the overall performance (d') on this task, which seem robust. However, in figure 8 a more careful analysis of the false alarm rates to the AuVr stimulus (which in the attend-a condition is really the most critical condition to confirm that the animals are performing a selective attention task and not dividing their attention across modalities) look really very high (near 100% in some sessions). I might have missed it, but I couldn't find any information about the distribution of trial types. If they are not equiprobable (i.e. there are fewer of the conflict trials) then overall d' is a poor measure of selective attention. Rather than an overall d' criterion for performance one that assesses the ability of the animal to selectively attend - such as the false alarm rate on the AuVr attend-a and ArVu on the attend-v - would be more appropriate. In figure 8 it would be helpful to know how the data are ordered - by mouse? I think the conclusion we're meant to take away from figure 8 is that increased firing rate predicts lapses of attention. However enhanced firing could indicate higher arousal and therefore intention to lick. A comparison of correct reject trials and hit trials as well as the hits/fa analysis included would be meaningful here to see whether this is a (pre) motor effect or an attentional effect.

    The paper relies heavily on Wilcoxon rank sum tests without any multiple comparisons correction. Often the p values are very small and therefore the lack of correction is inconsequential however there are exceptions. Specifically, this is problematic, especially for the pupillometry section where the AVv_v comparison would not survive multiple comparisons (unlike the Ava_a comparison its range includes zero). This begs the question as to whether the significant effect on pupil size simply comes about from the presence of visual stimulus rather than the AV conditions. There is also no justification provided for why this is a one-tailed test.

  4. Reviewer #3 (Public Review):

    The authors measure modality-specific and modality unspecific attentional influences in the various layers and in putative inhibitory and excitatory neurons in the primary auditory cortex of mice.

    The authors find that attention to the auditory stimuli specifically decreases responses in the deep layers and that this decrease predicts behavioural accuracy during modality-specific (auditory) attentional task.

    The methodology used is rigorous and the authors document technically in a proper way their main findings. However, it is difficult to provide a mechanistic value to the main findings, in relation to how they are interpreted. Enhancement of sensory responses has been reported in association sensory areas (e.g. V4, Reynolds, Pasternak and Desimone, 2000), and normalization mechanisms possibly involving activation of inhibitory circuits are involved in attentional modulation of sensory responses in primary areas (work of Maunsell, 2017).

    The authors should make an effort to put their findings in the biological context provided by the literature on attentional mechanisms affecting cortical sensory processing, and provide a more explicit interpretation of some of their critical findings in relation to the role of attentional mechanisms:

    - How do they interpret the finding that deep layer suppression is modality specific? Why paying attention to the visual stimulus is not affecting auditory responsiveness in the same way? Indeed, the authors report that the spike reductions (occurring also during pre-stimulus) correlate with performance in acoustic but not in visual attentional tasks.

    - What is the functional meaning and the interpretation that these findings extend also during the intertrial periods (upon presentation of task-irrelevant stimuli)? it would be important to clarify how these findings are consistent with modality and task-specific attentional mechanisms.