Causal roles of prefrontal cortex during spontaneous perceptual switching are determined by brain state dynamics

  1. Takamitsu Watanabe

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    By combining real-time closed-loop EEG-TMS and computational modelling, this study ambitiously examined the causal role of prefrontal cortex in resolving perceptual ambiguity. It impressively demonstrates brain-state-dependent effects on bistable perception.

    (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.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

The prefrontal cortex (PFC) is thought to orchestrate cognitive dynamics. However, in tests of bistable visual perception, no direct evidence supporting such presumable causal roles of the PFC has been reported except for a recent work. Here, using a novel brain-state-dependent neural stimulation system, we identified causal effects on percept dynamics in three PFC activities—right frontal eye fields, dorsolateral PFC (DLPFC), and inferior frontal cortex (IFC). The causality is behaviourally detectable only when we track brain state dynamics and modulate the PFC activity in brain-state-/state-history-dependent manners. The behavioural effects are underpinned by transient neural changes in the brain state dynamics, and such neural effects are quantitatively explainable by structural transformations of the hypothetical energy landscapes. Moreover, these findings indicate distinct functions of the three PFC areas: in particular, the DLPFC enhances the integration of two PFC-active brain states, whereas IFC promotes the functional segregation between them. This work resolves the controversy over the PFC roles in spontaneous perceptual switching and underlines brain state dynamics in fine investigations of brain-behaviour causality.

Article activity feed

  1. Version published to 10.7554/elife.69079 on eLife
  2. eLife

    Author Response:

    Reviewer #1 (Public Review):

    Watanabe presents a set of EEG-TMS experiments to show that brain stimulation in specific frontal regions affect both perception and brain states during bistable perception. The patterns of results appear interesting and potentially significant. The work uses relatively idiosyncratic methodologies in terms of data analysis and modelling, which makes the work harder to relate to extant literature. This situation requires authors to "go the extra mile" in explaining their approach and ensuring that readers can easily understand the findings in the light of what they're likely to already know - and here, I find steps could be taken.

    We are sorry for our unclear original manuscript that required extra efforts to read. As shown in Responses to the essential revisions, we have now modified it along with the reviewer's helpful and thoughtful suggestions. We hope that such modification will address the reviewer's concerns.

    Specific comments;

    • The author has an idiosyncratic definition of DLPFC; especially pDLPFC seems to coincide with iPCS retinotopic regions as found by Mackey et al, and with rIFJ from earlier work such as that by Sterzer and Kleinschmidt. When you look up DLPFC on wikipedia, this shows a region more 'superior' than both regions designated DLPFC in the present work, closer to aDLPFC than pDLPFC. Perhaps the author could try to more explicitly connect his nomenclature to the literature?

    We can understand the reviewer's concern on this terminology issue. In accordance with the reviewer's suggestion, we have now changed: aDLPFC to DLPFC and pDLPFC to IFC. Also we have added references to support such anatomical labels as follows:

    Methods section (lines 8-15 on page 21).

    "As in our previous work, the seven ROIs consisted of the right FEF (x = 38, y = 0, z = 60 in MNI coordinates), DLPFC (x = 44, y = 50, z = 10), IFC (x = 48, y = 24, z = 9), anterior superior parietal lobule (aSPL; x = 36, y = –45, z = 44), posterior superior parietal lobule (pSPL; x = 38, y = –64, z = 32), lateral occipital complex (LOC; x = 46, y = –78, z = 2) and V5 (hMT/V5; x = 47, y = –72, z = 1) (Fig. 1b). These coordinates are based on the following previous studies: a study by Sterzer and colleagues23 for the FEF; one by Knapen and colleagues24 for the DLPFC; one by Kleinschmidt and colleagues25 for the IFC; three studies by Kanai, Carmel and their colleagues26–28 for the aSPL and pSPL; one by Freeman and colleagues29 for the LOC and V5."

    • The possible relation of the individual brain state dynamics with the ongoing sequence of bistable perception apart from the TMS manipulation is not treated. This may feel self-evident to the author perhaps because this is the topic of previous studies, but it's confusing to a novice reader. To me, linking the sequence of bistable perceptual states to the sequence of brain states as found using the author's methodology is a fundamental step to allow interpretation of all of the subsequent results, because it speaks to the meaning and significance of the existence of these brain states. Without this step, I find it difficult to interpret figures 2a and 2b (which, I have to say, do indeed look like enticing patterns in the data). So, specifically, does the author replicate the brain-state vs behavior correlations that he reported in his earlier (2014) work on this topic? And, because this publication reported mainly across-observer correlations, what about relations between brain states and their transitions and perceptual events on a within-subject basis?

    Thank you for giving us to the opportunity to show detailed results that validate our application to the EEG data. Now we have inserted all such results into the first part of the Results section with new figures.

    • The exact analysis procedure that leads up to the brain state designation is not very transparent. What, for example, is the brain state that is "Frontal"? I would appreciate to see state-transition triggered time-frequency plots to be able to understand what exactly in the EEG the procedure picks up. The same holds for the TMS-triggered changes; is there any pattern in terms of TMS-induced time-frequency changes?

    In accordance with the reviewer's suggestion, we have now added time-frequency plots during brain state transitions as well as those before/after a TMS administration.

    • It would be a valuable addition if the author could clarify what he means with a brain state; this term means different things in different fields. The concept of brain state is now primarily based on high frequency EEG signatures, but there are likely many other possible measurements that could produce estimated brain state. How would the findings change if other measures were used as a basis for the same methods?

    As the reviewer stated, a "brain state" in this study is different from so-called "miscrostate" in EEG research but indicates an activity pattern of multiple (here, seven) brain regions or a group of such activity patterns. This concept of the brain state is based on the energy landscape analysis. In terms of generalisability of this analysis method, previous studies have demonstrated that such brain states are identifiable in both task-related and resting-state functional MRI data1,2,30,31.

    In the meantime, hidden Markov model (HMM) can also identify similar brain states32–35. However, few studies have applied the HMM to task-related data, and no work has used it to the neural signals during bistable visual perception.

    Given this, this study used the energy landscape analysis to identify the brain states and dynamics between them.

    We clarified this point by adding the following descriptions into the Methods section.

    Methods section (lines 14-18 on page 26)

    "Note that a “brain state” in this study is not a so-called “miscrostate” in conventional EEG research; it indicates an activity pattern of multiple (here, seven) brain regions or a group of such activity patterns. Although other analyses, such as hidden Markov model (HMM), can also identify brain states78–81, we adopted the energy landscape analysis in this study because it was previously used to identify the brain states underpinning the bistable visual perception20. "

    • Figure 1l. From methods and explanations it's not really clear how this figure is produced. If it is created from single-subject surface locations that were explicitly targeted, and these locations are then transformed into an average-subject surface, that would be correct. But weren't these locations targeted based on MNI coordinates? In this case one would expect more of a spread in specific locations because of the across-subject variability in surface folding. So, could the author please explain in more detail how this figure is generated?

    We are sorry for the insufficient description on this issue. We have now added the following explanation to the Methods section and the legend of the figure (Fig. 4a):

    Methods section (line 33 on page 22 – line 2 on page 23)

    "We confirmed that the coil did not substantially move throughout the experiment by re-measing the location with the neuro-navigation system at the end of each experiment day. The green circles in Fig. 4a show such end-of-the-day locations averaged across the four-day sessions in the main experiment."

    Fig. 4. a. We administered inhibitory TMS over the three PFC regions. "In one TMS condition, we placed the TMS coil over one of the PFC areas using a stereoscopic neuro-navigation system based on the MNI coordinates at the beginning of each experiment day. At the end of the day, we re-measured the coordinates of the TMS coil using the navigation system. Finally, we averaged the coordinates across the four-day sessions in the main experiment. The green circles represent such mean MNI coordinates of the stimulated brain site for each participant. The green circles were mapped closely onto the original coordinates (the centres of the yellow circles)."

    • Page 8, I appreciate the logic that "barrier heights are associated with the dwelling time in the brain states and inversely correlated with the transition frequency between them", but this needs to be fleshed out more. What are the numerical simulations here? These aren't described in the text and as a reader, I'm left having to believe the accuracy of the 'numerical simulations' without being given the opportunity to understand them. This explanation would be a nice opportunity to go into detail about how the author does (and, consequently, the audience should) understand and interpret both the brain states, and their transitions.

    We are sorry for our insufficient description on this issue. Now in accordance with the suggestion, we have now re-written the methods of the numerical simulation. Please see our response to (1)-(iii) in Responses to the essential revisions.

    Reviewer #2 (Public Review):

    The author tested the hypothesis that the causal influence of the PFC on bistable perception is dynamic and depends on the (fluctuating) state of the cortical networks. Using offline and online EEG measurements and a sophisticated analysis procedure, the author characterized their dynamic brain states when observers perceived a bistable rotating sphere defined by Structure-from-Motion with alternating perceived direction of rotation. TMS applied to different regions of the frontal, parietal, and visual areas had different effects on observers perceptual dynamics, depending on the dynamic state of the cortical networks. It's quite impressive to see the large effect size from TMS to the aDLPFC, and the opposite direction of the effects observed from aDLPFC and pDLPFC/FEF stimulation makes it more convincing that the PFC has specific and robust roles in bistable perception.

    Thank you for the reviewer's positive evaluation.

    Although the effect on bistable perception from state-dependent TMS of DLPFC is robust and very interesting, the functional mechanism of how different regions of DLPFC contribute to the perceptual dynamics remains unclear. I find it surprising that the author did not address the potential role of attention in mediating DLPFC's contribution to observers' perceptual dynamics. Given that attention does play a role in the dynamics of many forms of bistable perception, it is important to distinguish between an intrinsic contribution of DLPFC to bistable perception vs. an effect mediated by changes in attentional state. It is also useful for the author to discuss how and why certain brain states are linked to certain perceptual states.

    We agree with the reviewers on effects of attention on the DLPFC and IFC functions in the bistable perception. Also, we admit that we should have to state our inference on the neuropsychological functions of each brain state. Now, to address these concerns, we have added detailed discussions into the Discussion section.

    There are many forms of bistable perception, and their dynamics are controlled or influenced by shared as well as independent mechanisms (e.g., Cao T et al, Frontiers in Psychology 2018). It would be useful to discuss the generalizability and limitations of the current results in relation to different types of bistable stimuli. The methodological approach developed by the author will be quite useful in researching the neural mechanisms of other types of bistable perception.

    We appreciate for letting us know the nice behavioural study. Now we have extended our notion on the generalisability of the current observations as follows:

    Discussion section (line 30 on page 17 – line 2 on page 18)

    "These findings may not be directly applicable to other types of multistable visual perception, such as binocular rivalry, which is linked to lower-level brain architectures such as the visual cortex 32–38. In fact, a comprehensive behavioural study reported the relative dissimilarity in perceptual switching rate between the current SFM-induced bistable perception and the binocular rivalry61. In contrast, the same study found the similarity in between the SFM-induced bistable perception and other fluctuating perception triggered by spinning dancer62 and Lissajous-figure63. Given this, the current observations might be more applicable to types of bistable perception that requires construction of a 3D image from 2D motion compared to the other types such as the binocular rivalry."

    Reviewer #3 (Public Review):

    This is an ambitious study by a competent single author who has previously published highly innovative work on this topic. The study incorporates real-time closed-loop EEG-TMS and computational modeling to causally test the role of PFC in perceptual switching of bistable perception triggered by ambiguous visual input. While the work is technically impressive and involves a substantial amount of work spread over multiple experiments (especially notable in the context of a single-author manuscript), I have some major concerns as described below.

    1. The author's previous work on energy landscape in the context of bistable perception was conducted using fMRI. This current study employs EEG, and records time series from 7 ROIs (Fig. 1a-b). Some of these ROIs are very close together, less than a few centimeters (e.g., a-p SPL; a-p DLPFC; LOC-V5). The conventional thinking is that scalp EEG does not have the spatial resolution to separate signals from such closely spaced areas. While the author employs a Laplacian montage, validation data suggesting that the resulting signal had high SNR and could differentiate between neighboring regions is missing.

    To address this concern, we probed the data and found results to support the sensitivity and specificity of the current EEG system. We have added these new observations into the Methods section.

    1. The EEG analysis rests on gamma band (30-80 Hz) power. This should be explained in the main text. It is technically risky to record gamma band activity using scalp EEG, due to muscle, eye, and, most concerningly, microsaccade-related artifacts (see work by Yuval-Greenberg). Since the task employs a structure-from-motion stimulus, the effect of microsaccades is especially worrying. No control data was presented to suggest that these artifacts do not contribute to the analyses.

    We agree on the necessity of reducing the microsaccade-related artefacts on the gamma-band signals. We adopted a derivation method (i.e., Hjorth signal calculation) and ICA to reduce such artefacts (for the derivation method, see 5–8; for the ICA, see 9–11), but the original manuscript did not present explicit results to support effects of these signal processing methods. To address this situation, we conducted a new EEG experiment, in which 30 healthy adults underwent the same psychophysics paradigm. In the additional experiment, 28 EEG electrodes were placed around the seven regions of interests (ROIs) in the same manner as in the original experiment, whereas the other four electrodes were located around the eyes for electrooculography (EOG)5. Based on previous literature12–14, we used these EOG signals to infer the timings of the occurrences of microsaccades. In this experiment, we confirmed that the current preprocessing methods could reduce the artefacts induced by microsaccades, which was described in the Methods section. We also explicitly stated that we used the gamma-band EEG signals in the first paragraph in the Results section.

    1. P. 10 There is a concern here that the hypothesis testing is circular. The models were fit by using EEG data (and behavior?) to calculate the energy landscape, so is it trivially expected then that the dwell times seen behaviorally correlate with the energy barrier estimated by the model?

    The energy landscape analysis used no behavioural data to identify the brain state dynamics. We clarified this by adding a sentence into the Results section.

    1. It's not clear to me why pDLPFC's result was interpreted as "functional diversity".

    To clarify this, we have modified the description on the IFC function as follows:

    Abstract

    "Moreover, these findings indicate distinct functions of the three PFC areas: in particular, the DLPFC enhances the integration of two PFC-active brain states, whereas IFC promotes the functional segregation between them."

    Discussion section (lines 18-21 on page 15)

    "Moreover, the current findings suggest distinct functions of the PFC regions in terms of the brain state dynamics: the activation of DLPFC enhances the functional integration between the Frontal and Intermediate state, whereas the IFC activity promotes the functional segregation between the two brain states; the FEF activity stabilises Frontal state."

    1. The pDLPFC region here would be more accurately referred to as inferior frontal gyrus (IFG) or ventral frontal cortex (VFC), or inferior frontal cortex (IFC). It is not part of the classic DLPFC.

    In accordance with the reviewer's suggestion, we have now replaced pDLPFC with IFC throughout the manuscript.

  3. eLife

    Evaluation Summary:

    By combining real-time closed-loop EEG-TMS and computational modelling, this study ambitiously examined the causal role of prefrontal cortex in resolving perceptual ambiguity. It impressively demonstrates brain-state-dependent effects on bistable perception.

    (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.)

  4. eLife

    Reviewer #1 (Public Review):

    Watanabe presents a set of EEG-TMS experiments to show that brain stimulation in specific frontal regions affect both perception and brain states during bistable perception. The patterns of results appear interesting and potentially significant. The work uses relatively idiosyncratic methodologies in terms of data analysis and modelling, which makes the work harder to relate to extant literature. This situation requires authors to "go the extra mile" in explaining their approach and ensuring that readers can easily understand the findings in the light of what they're likely to already know - and here, I find steps could be taken.

    Specific comments;

    - The author has an idiosyncratic definition of DLPFC; especially pDLPFC seems to coincide with iPCS retinotopic regions as found by Mackey et al, and with rIFJ from earlier work such as that by Sterzer and Kleinschmidt. When you look up DLPFC on wikipedia, this shows a region more 'superior' than both regions designated DLPFC in the present work, closer to aDLPFC than pDLPFC. Perhaps the author could try to more explicitly connect his nomenclature to the literature?

    - The possible relation of the individual brain state dynamics with the ongoing sequence of bistable perception apart from the TMS manipulation is not treated. This may feel self-evident to the author perhaps because this is the topic of previous studies, but it's confusing to a novice reader. To me, linking the sequence of bistable perceptual states to the sequence of brain states as found using the author's methodology is a fundamental step to allow interpretation of all of the subsequent results, because it speaks to the meaning and significance of the existence of these brain states. Without this step, I find it difficult to interpret figures 2a and 2b (which, I have to say, do indeed look like enticing patterns in the data). So, specifically, does the author replicate the brain-state vs behavior correlations that he reported in his earlier (2014) work on this topic? And, because this publication reported mainly across-observer correlations, what about relations between brain states and their transitions and perceptual events on a within-subject basis?

    - The exact analysis procedure that leads up to the brain state designation is not very transparent. What, for example, is the brain state that is "Frontal"? I would appreciate to see state-transition triggered time-frequency plots to be able to understand what exactly in the EEG the procedure picks up. The same holds for the TMS-triggered changes; is there any pattern in terms of TMS-induced time-frequency changes?

    - It would be a valuable addition if the author could clarify what he means with a brain state; this term means different things in different fields. The concept of brain state is now primarily based on high frequency EEG signatures, but there are likely many other possible measurements that could produce estimated brain state. How would the findings change if other measures were used as a basis for the same methods?

    - Figure 1l. From methods and explanations it's not really clear how this figure is produced. If it is created from single-subject surface locations that were explicitly targeted, and these locations are then transformed into an average-subject surface, that would be correct. But weren't these locations targeted based on MNI coordinates? In this case one would expect more of a spread in specific locations because of the across-subject variability in surface folding. So, could the author please explain in more detail how this figure is generated?

    - Page 8, I appreciate the logic that "barrier heights are associated with the dwelling time in the brain states and inversely correlated with the transition frequency between them", but this needs to be fleshed out more. What are the numerical simulations here? These aren't described in the text and as a reader, I'm left having to believe the accuracy of the 'numerical simulations' without being given the opportunity to understand them. This explanation would be a nice opportunity to go into detail about how the author does (and, consequently, the audience should) understand and interpret both the brain states, and their transitions.

  5. eLife

    Reviewer #2 (Public Review):

    The author tested the hypothesis that the causal influence of the PFC on bistable perception is dynamic and depends on the (fluctuating) state of the cortical networks. Using offline and online EEG measurements and a sophisticated analysis procedure, the author characterized their dynamic brain states when observers perceived a bistable rotating sphere defined by Structure-from-Motion with alternating perceived direction of rotation. TMS applied to different regions of the frontal, parietal, and visual areas had different effects on observers perceptual dynamics, depending on the dynamic state of the cortical networks. It's quite impressive to see the large effect size from TMS to the aDLPFC, and the opposite direction of the effects observed from aDLPFC and pDLPFC/FEF stimulation makes it more convincing that the PFC has specific and robust roles in bistable perception.

    Although the effect on bistable perception from state-dependent TMS of DLPFC is robust and very interesting, the functional mechanism of how different regions of DLPFC contribute to the perceptual dynamics remains unclear. I find it surprising that the author did not address the potential role of attention in mediating DLPFC's contribution to observers' perceptual dynamics. Given that attention does play a role in the dynamics of many forms of bistable perception, it is important to distinguish between an intrinsic contribution of DLPFC to bistable perception vs. an effect mediated by changes in attentional state. It is also useful for the author to discuss how and why certain brain states are linked to certain perceptual states.

    There are many forms of bistable perception, and their dynamics are controlled or influenced by shared as well as independent mechanisms (e.g., Cao T et al, Frontiers in Psychology 2018). It would be useful to discuss the generalizability and limitations of the current results in relation to different types of bistable stimuli. The methodological approach developed by the author will be quite useful in researching the neural mechanisms of other types of bistable perception.

  6. eLife

    Reviewer #3 (Public Review):

    This is an ambitious study by a competent single author who has previously published highly innovative work on this topic. The study incorporates real-time closed-loop EEG-TMS and computational modeling to causally test the role of PFC in perceptual switching of bistable perception triggered by ambiguous visual input. While the work is technically impressive and involves a substantial amount of work spread over multiple experiments (especially notable in the context of a single-author manuscript), I have some major concerns as described below.

    1. The author's previous work on energy landscape in the context of bistable perception was conducted using fMRI. This current study employs EEG, and records time series from 7 ROIs (Fig. 1a-b). Some of these ROIs are very close together, less than a few centimeters (e.g., a-p SPL; a-p DLPFC; LOC-V5). The conventional thinking is that scalp EEG does not have the spatial resolution to separate signals from such closely spaced areas. While the author employs a Laplacian montage, validation data suggesting that the resulting signal had high SNR and could differentiate between neighboring regions is missing.

    2. The EEG analysis rests on gamma band (30-80 Hz) power. This should be explained in the main text. It is technically risky to record gamma band activity using scalp EEG, due to muscle, eye, and, most concerningly, microsaccade-related artifacts (see work by Yuval-Greenberg). Since the task employs a structure-from-motion stimulus, the effect of microsaccades is especially worrying. No control data was presented to suggest that these artifacts do not contribute to the analyses.

    3. P. 10 There is a concern here that the hypothesis testing is circular. The models were fit by using EEG data (and behavior?) to calculate the energy landscape, so is it trivially expected then that the dwell times seen behaviorally correlate with the energy barrier estimated by the model?

    4. It's not clear to me why pDLPFC's result was interpreted as "functional diversity".

    5. The pDLPFC region here would be more accurately referred to as inferior frontal gyrus (IFG) or ventral frontal cortex (VFC), or inferior frontal cortex (IFC). It is not part of the classic DLPFC.

  7. Version published to 10.1101/2021.04.16.440188v1 on bioRxiv