Response outcome gates the effect of spontaneous cortical state fluctuations on perceptual decisions
Curation statements for this article:-
Curated by eLife
eLife assessment
Reao et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? The data demonstrate that the outcome of the previous trial, specifically a miss, allows these associations to be seen - while a correct response appears less likely to do so. and this is a valuable advance in our understanding of the relationship between brain state, behavioral state, and performance. Technically, the study is solid, ie, the methods, data and analyses broadly support the claims, with some weaknesses remaining.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Neuroscience (eLife)
Abstract
Sensory responses of cortical neurons are more discriminable when evoked on a baseline of desynchronized spontaneous activity, but cortical desynchronization has not generally been associated with more accurate perceptual decisions. Here, we show that mice perform more accurate auditory judgments when activity in the auditory cortex is elevated and desynchronized before stimulus onset, but only if the previous trial was an error, and that this relationship is occluded if previous outcome is ignored. We confirmed that the outcome-dependent effect of brain state on performance is neither due to idiosyncratic associations between the slow components of either signal, nor to the existence of specific cortical states evident only after errors. Instead, errors appear to gate the effect of cortical state fluctuations on discrimination accuracy. Neither facial movements nor pupil size during the baseline were associated with accuracy, but they were predictive of measures of responsivity, such as the probability of not responding to the stimulus or of responding prematurely. These results suggest that the functional role of cortical state on behavior is dynamic and constantly regulated by performance monitoring systems.
Article activity feed
-
-
Author Response
Reviewer 1 (Public Review):
In this paper, Reato, Steinfeld et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? They record spiking activity in the auditory cortex of head-fixed mice as the animals performed a tone frequency discrimination task. They then measure both overall activity and the synchronization between neurons, and link this ’baseline state’ (after removing slow drifts) of cortex to decision accuracy. They find that cortical state fluctuations only affect subsequent evoked responses and choice behavior after errors. This indicates that it’s important to take into account the behavioral context when examining the effects of neural state on behavior.
Strengths of this …
Author Response
Reviewer 1 (Public Review):
In this paper, Reato, Steinfeld et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? They record spiking activity in the auditory cortex of head-fixed mice as the animals performed a tone frequency discrimination task. They then measure both overall activity and the synchronization between neurons, and link this ’baseline state’ (after removing slow drifts) of cortex to decision accuracy. They find that cortical state fluctuations only affect subsequent evoked responses and choice behavior after errors. This indicates that it’s important to take into account the behavioral context when examining the effects of neural state on behavior.
Strengths of this work are the clear and beautiful presentation of the figures, and the careful consideration of the temporal properties of behavioral and neural signals. Indeed, slowly drifting signals are tricky as many authors have recently addressed (e.g. Ashwood, Gupta, Harris). The authors are well aware of the difficulties in correlating different signals with temporal and cross-correlation (such as in their ’epoch hypothesis’). To disentangle such slow trends from more short-lived state fluctuations, they remove the impact of the past 10 trials and continue their analyses with so-called ’innovations’ (a term that is unusual, and may more simply be replaced with ’residuals’).
The terms ‘innovations’ and ‘residuals’ are sometimes used interchangeably. We used innovations because that’s how they were introduced in the signal processing literature (i.e., Kailath, T (1968). ”An innovations approach to least-squares estimation–Part I: Linear filtering in additive white noise.” IEEE transactions on automatic control). We try to be explicit in the text about the formal definition of this quantity, to avoid problems with terminology.
I do wonder if this throws out the baby with the bathwater. If the concern is statistical confound, the ’session permutation’ method (Harris) may be better suited. If the concern is that short-term state fluctuations are more behaviorally relevant (and obscured by slow drifts), then why are the results with raw signals in the supplement (Suppfig 8) so similar?
The concern was statistical confound, although this concern is ameliorated when using a mixed model approach and focusing on fixed effects. However, our approach allowed us to assess the relative importance of slow versus single-trial timescales in the predictive relationship between cortical state (and arousal) and behavior, revealing that, in the conditions of our experiment, only the fast timescales are relevant. Because of this, we think that the baby wasn’t thrown out with the bathwater as, qualitatively, no new phenomenology was revealed when the slow components of the signals were included. In hindsight, it is true that the results we obtained suggest that maybe the effort we made to isolate the fast component of the signals was unjustified. However, this can only be known after both options have been tried, as we did. Moreover, we started using innovations based on the results in Figure 2 where, as we show, the use of innovations does make a difference, even at the level of fixed effects in a mixed model. We agree that we could have used the ‘session permutation’ method, but given the depth at which we have explored this issue in the manuscript already, and the clarity of the results, we think that adding a third method would only make reading the manuscript more difficult without adding any substantially new content.
While the authors are correct that go-nogo tasks have drawbacks in dissociating sensitivity from response bias, they only cursorily review the literature on 2AFC tasks and cortical state. In particular, it would be good to discuss how the specific method - spikes, EEG (Waschke), widefield (Jacobs) and algorithm for quantifying synchronization may affect outcomes. How do these population-based measures of cortical state relate to those described extensively with slightly different signals, notably LFP or EEG in humans (e.g. work by Saskia Haegens, Niko Busch, reviewed in https://doi.org/10.1016/j.tics.2020.05.004)? This review also points out the importance of moving beyond simple measures of accuracy and using SDT, which would be an interesting improvement for this paper too.
We thank the reviewer for pointing us towards the oscillation-based brain-state literature in humans. We have expanded the paragraph in the discussion where we compare our results with previous work in order to (i) elaborate on the literature on 2AFC tasks, (ii) specifically address the literature linking alpha power in the pre-stimulus baseline and psychophysical performance, and (iii) mention different methods for assessing desynchronization. Our view is that absence of lowfrequency power is a robust measure which can be assessed using different types of signals (spikes, imaging, LFP, EEG). That said, the relationship between desynchronization and behavior appears subtle and variable, specially within discrimination paradigms. These issues are discussed in the paragraph starting in line 527 in the text.
Regarding the use of SDT, we had already established that our main finding could be expressed as a significant interaction between FR/Synch and the stimulus-strength regressor, when predicting choice after errors (Supplementary Fig. 4A in original manuscript), which is equivalent to a cortical state-dependent increase in d′ after the mice made a mistake. In order to consider a possible effect of cortical state on the ‘criterion’ (i.e., an effect on the bias of the mice towards either response spout), we re-run this GLMM but adding the cortical state regressors as main effects. The results show that the FR-Synch predictor is only significantly greater than zero as an interaction after errors (p = 0.0025). As a main effect, it’s not significantly different from zero neither after errors (p = 0.28), nor after correct trials (p = 0.97). We have included this analysis as Figure 3-figure supplement 1B (replacing the previous Supplementary Fig. 4A) and commented on them in the text (lines 222-225).
Reviewer 2 (Public Review):
The relationship between measures of brain state, behavioral state, and performance has long been speculated to be relatively simple - with arousal and engagement reflecting EEG desynchronization and improved performance associated with increases in engagement and attention. The present study demonstrates that the outcome of the previous trial, specifically a miss, allows these associations to be seen - while a correct response appears less likely to do so. This is an interesting advance in our understanding of the relationship between brain state, behavioral state, and performance.
This is probably just a typo, but we would like to clarify that the relevant outcome in the previous trial is not a miss, but an incorrect choice in an otherwise valid trial (i.e., a trial with a response within the allowed response window).
While the study is well done, the results are likely to be specific to their trial structure and states exhibited by the mice. To examine the full range of arousal states, it needs to be demonstrated that animals are varying between near-sleep (e.g. drowsiness) and high-alertness such as in rapid running. The fact that the trials occurred rapidly means that the physiological and neural variables associated with each trial will overlap with upcoming trials - it takes a mouse more than a few seconds to relax from a previous miss or hit, for example. Spreading the rapidity of the trials out would allow for a broader range of states to be examined, and perhaps less cross-talk between adjacent trials. The interpretation of the results, therefore, must be taken in light of the trial structure and the states exhibited by the mice.
We thank the reviewer for the positive assessment of our work and also for raising this point in particular. This motivated us to look more carefully at this issue, with results that, we believe, strengthen our study.
-
eLife assessment
Reao et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? The data demonstrate that the outcome of the previous trial, specifically a miss, allows these associations to be seen - while a correct response appears less likely to do so. and this is a valuable advance in our understanding of the relationship between brain state, behavioral state, and performance. Technically, the study is solid, ie, the methods, data and analyses broadly support the claims, with some weaknesses remaining.
-
Reviewer #1 (Public Review):
In this paper, Reato, Steinfeld et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? They record spiking activity in the auditory cortex of head-fixed mice as the animals performed a tone frequency discrimination task. They then measure both overall activity and the synchronization between neurons, and link this 'baseline state' (after removing slow drifts) of cortex to decision accuracy. They find that cortical state fluctuations only affect subsequent evoked responses and choice behavior after errors. This indicates that it's important to take into account the behavioral context when examining the effects of neural state on behavior.
Strengths of this work are the clear and …
Reviewer #1 (Public Review):
In this paper, Reato, Steinfeld et al. investigate a question that has long puzzled neuroscientists: what features of ongoing brain activity predict trial-to-trial variability in responding to the same sensory stimuli? They record spiking activity in the auditory cortex of head-fixed mice as the animals performed a tone frequency discrimination task. They then measure both overall activity and the synchronization between neurons, and link this 'baseline state' (after removing slow drifts) of cortex to decision accuracy. They find that cortical state fluctuations only affect subsequent evoked responses and choice behavior after errors. This indicates that it's important to take into account the behavioral context when examining the effects of neural state on behavior.
Strengths of this work are the clear and beautiful presentation of the figures, and the careful consideration of the temporal properties of behavioral and neural signals. Indeed, slowly drifting signals are tricky as many authors have recently addressed (e.g. Ashwood, Gupta, Harris). The authors are well aware of the difficulties in correlating different signals with temporal and cross-correlation (such as in their 'epoch hypothesis'). To disentangle such slow trends from more short-lived state fluctuations, they remove the impact of the past 10 trials and continue their analyses with so-called 'innovations' (a term that is unusual, and may more simply be replaced with 'residuals').
I do wonder if this throws out the baby with the bathwater. If the concern is statistical confound, the 'session permutation' method (Harris) may be better suited. If the concern is that short-term state fluctuations are more behaviorally relevant (and obscured by slow drifts), then why are the results with raw signals in the supplement (Suppfig 8) so similar?
While the authors are correct that go-nogo tasks have drawbacks in dissociating sensitivity from response bias, they only cursorily review the literature on 2AFC tasks and cortical state. In particular, it would be good to discuss how the specific method - spikes, EEG (Waschke), widefield (Jacobs) and algorithm for quantifying synchronization may affect outcomes. How do these population-based measures of cortical state relate to those described extensively with slightly different signals, notably LFP or EEG in humans (e.g. work by Saskia Haegens, Niko Busch, reviewed in https://doi.org/10.1016/j.tics.2020.05.004)? This review also points out the importance of moving beyond simple measures of accuracy and using SDT, which would be an interesting improvement for this paper too.
-
Reviewer #2 (Public Review):
The relationship between measures of brain state, behavioral state, and performance has long been speculated to be relatively simple - with arousal and engagement reflecting EEG desynchronization and improved performance associated with increases in engagement and attention. The present study demonstrates that the outcome of the previous trial, specifically a miss, allows these associations to be seen - while a correct response appears less likely to do so. This is an interesting advance in our understanding of the relationship between brain state, behavioral state, and performance.
While the study is well done, the results are likely to be specific to their trial structure and states exhibited by the mice. To examine the full range of arousal states, it needs to be demonstrated that animals are varying …
Reviewer #2 (Public Review):
The relationship between measures of brain state, behavioral state, and performance has long been speculated to be relatively simple - with arousal and engagement reflecting EEG desynchronization and improved performance associated with increases in engagement and attention. The present study demonstrates that the outcome of the previous trial, specifically a miss, allows these associations to be seen - while a correct response appears less likely to do so. This is an interesting advance in our understanding of the relationship between brain state, behavioral state, and performance.
While the study is well done, the results are likely to be specific to their trial structure and states exhibited by the mice. To examine the full range of arousal states, it needs to be demonstrated that animals are varying between near-sleep (e.g. drowsiness) and high-alertness such as in rapid running. The fact that the trials occurred rapidly means that the physiological and neural variables associated with each trial will overlap with upcoming trials - it takes a mouse more than a few seconds to relax from a previous miss or hit, for example. Spreading the rapidity of the trials out would allow for a broader range of states to be examined, and perhaps less cross-talk between adjacent trials. The interpretation of the results, therefore, must be taken in light of the trial structure and the states exhibited by the mice.
-
