Ubiquitous predictive processing in the spectral domain of sensory cortex
Curation statements for this article:-
Curated by eLife
eLife Assessment
The authors analyzed spectral properties of neural activity recorded using laminar probes while mice engaged in a global/local visual oddball paradigm. They found solid evidence for an increase in gamma (and theta in some cases) for unpredictable versus predictable stimuli, and a reduction in alpha/beta, which they consider evidence towards a "predictive routing" scheme. The study is overall important because it addresses the basis of predictive processing in the cortex, but some of the analytical choices could be better motivated, and overall, the manuscript can be improved by performing additional analyses.
This article has been Reviewed by the following groups
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
- Evaluated articles (eLife)
Abstract
The appearance at the anatomical level of a canonical laminar microcircuit suggests that each six-layer column of granular cortex may mediate a canonical computation. Hypotheses for such computations include predictive coding, predictive routing, efficient coding, and others. However, single-neuron recordings capture only the individual elements of the hypothesized laminar microcircuit, while local field potentials (LFPs) from a laminar probe offer insight into the broader population activity. Through the Allen Institute’s OpenScope Brain Observatory, data in mice performing a visual oddball task during multi-area laminar recording was used to test predictive processing hypotheses in the spectral domain. Histological labeling of the cortical laminae enabled a fine-grained examination of their roles in the task, and frequency bands capturing both feedforward and feedback effects were analyzed. ɣ-band local-field potential (LFP) oscillations conveyed feedforward prediction errors in lower sensory areas of cortex; ⍺/β-band oscillations weakened in unpredictable conditions compared to predictable ones; and θ-band oscillations additionally signalled slower, longer-scale temporal prediction errors. In combination with the previous findings, predictive routing explains these experiments where neither ubiquitous predictive coding nor feedforward adaptation can.
Article activity feed
-
-
-
eLife Assessment
The authors analyzed spectral properties of neural activity recorded using laminar probes while mice engaged in a global/local visual oddball paradigm. They found solid evidence for an increase in gamma (and theta in some cases) for unpredictable versus predictable stimuli, and a reduction in alpha/beta, which they consider evidence towards a "predictive routing" scheme. The study is overall important because it addresses the basis of predictive processing in the cortex, but some of the analytical choices could be better motivated, and overall, the manuscript can be improved by performing additional analyses.
-
Reviewer #1 (Public review):
Summary:
The authors recorded neural activity using laminar probes while mice engaged in a global/local visual oddball paradigm. The focus of the article is on oscillatory activity, and found activity differences in theta, alpha/beta, and gamma bands related to predictability and prediction error.
I think this is an important paper, providing more direct evidence for the role of signals in different frequency bands related to predictability and surprise in the sensory cortex.
Comments:
Below are some comments that may hopefully help further improve the quality of this already very interesting manuscript.
(1) Introduction:
The authors write in their introduction: "H1 further suggests a role for θ oscillations in prediction error processing as well." Without being fleshed out further, it is unclear what role …
Reviewer #1 (Public review):
Summary:
The authors recorded neural activity using laminar probes while mice engaged in a global/local visual oddball paradigm. The focus of the article is on oscillatory activity, and found activity differences in theta, alpha/beta, and gamma bands related to predictability and prediction error.
I think this is an important paper, providing more direct evidence for the role of signals in different frequency bands related to predictability and surprise in the sensory cortex.
Comments:
Below are some comments that may hopefully help further improve the quality of this already very interesting manuscript.
(1) Introduction:
The authors write in their introduction: "H1 further suggests a role for θ oscillations in prediction error processing as well." Without being fleshed out further, it is unclear what role this would be, or why. Could the authors expand this statement?
(2) Limited propagation of gamma band signals:
Some recent work (e.g. https://www.cell.com/cell-reports/fulltext/S2211-1247(23)00503-X) suggests that gamma-band signals reflect mainly entrainment of the fast-spiking interneurons, and don't propagate from V1 to downstream areas. Could the authors connect their findings to these emerging findings, suggesting no role in gamma-band activity in communication outside of the cortical column?
(3) Paradigm:
While I agree that the paradigm tests whether a specific type of temporal prediction can be formed, it is not a type of prediction that one would easily observe in mice, or even humans. The regularity that must be learned, in order to be able to see a reflection of predictability, integrates over 4 stimuli, each shown for 500 ms with a 500 ms blank in between (and a 1000 ms interval separating the 4th stimulus from the 1st stimulus of the next sequence). In other words, the mouse must keep in working memory three stimuli, which partly occurred more than a second ago, in order to correctly predict the fourth stimulus (and signal a 1000 ms interval as evidence for starting a new sequence).
A problem with this paradigm is that positive findings are easier to interpret than negative findings. If mice do not show a modulation to the global oddball, is it because "predictive coding" is the wrong hypothesis, or simply because the authors generated a design that operates outside of the boundary conditions of the theory? I think the latter is more plausible. Even in more complex animals, (eg monkeys or humans), I suspect that participants would have trouble picking up this regularity and sequence, unless it is directly task-relevant (which it is not, in the current setting). Previous experiments often used simple pairs (where transitional probability was varied, eg, Meyer and Olson, PNAS 2012) of stimuli that were presented within an intervening blank period. Clearly, these regularities would be a lot simpler to learn than the highly complex and temporally spread-out regularity used here, facilitating the interpretation of negative findings (especially in early cortical areas, which are known to have relatively small temporal receptive fields).
I am, of course, not asking the authors to redesign their study. I would like to ask them to discuss this caveat more clearly, in the Introduction and Discussion, and situate their design in the broader literature. For example, Jeff Gavornik has used much more rapid stimulus designs and observed clear modulations of spiking activity in early visual regions. I realize that this caveat may be more relevant for the spiking paper (which does not show any spiking activity modulation in V1 by global predictability) than for the current paper, but I still think it is an important general caveat to point out.
(4) Reporting of results:
I did not see any quantification of the strength of evidence of any of the results, beyond a general statement that all reported results pass significance at an alpha=0.01 threshold. It would be informative to know, for all reported results, what exactly the p-value of the significant cluster is; as well as for which performed tests there was no significant difference.
(5) Cluster test:
The authors use a three-dimensional cluster test, clustering across time, frequency, and location/channel. I am wondering how meaningful this analytical approach is. For example, there could be clusters that show an early difference at some location in low frequencies, and then a later difference in a different frequency band at another (adjacent) location. It seems a priori illogical to me to want to cluster across all these dimensions together, given that this kind of clustering does not appear neurophysiologically implausible/not meaningful. Can the authors motivate their choice of three-dimensional clustering, or better, facilitating interpretability, cluster eg at space and time within specific frequency bands (2d clustering)?
-
Reviewer #2 (Public review):
Summary:
Sennesh and colleagues analyzed LFP data from 6 regions of rodents while they were habituated to a stimulus sequence containing a local oddball (xxxy) and later exposed to either the same (xxxY) or a deviant global oddball (xxxX). Subsequently, they were exposed to a controlled random sequence (XXXY) or a controlled deterministic sequence (xxxx or yyyy). From these, the authors looked for differences in spectral properties (both oscillatory and aperiodic) between three contrasts (only for the last stimulus of the sequence).
(1) Deviance detection: unpredictable random (XXXY) versus predictable habituation (xxxy)
(2) Global oddball: unpredictable global oddball (xxxX) versus predictable deterministic (xxxx), and
(3) "Stimulus-specific adaptation:" locally unpredictable oddball (xxxY) versus …
Reviewer #2 (Public review):
Summary:
Sennesh and colleagues analyzed LFP data from 6 regions of rodents while they were habituated to a stimulus sequence containing a local oddball (xxxy) and later exposed to either the same (xxxY) or a deviant global oddball (xxxX). Subsequently, they were exposed to a controlled random sequence (XXXY) or a controlled deterministic sequence (xxxx or yyyy). From these, the authors looked for differences in spectral properties (both oscillatory and aperiodic) between three contrasts (only for the last stimulus of the sequence).
(1) Deviance detection: unpredictable random (XXXY) versus predictable habituation (xxxy)
(2) Global oddball: unpredictable global oddball (xxxX) versus predictable deterministic (xxxx), and
(3) "Stimulus-specific adaptation:" locally unpredictable oddball (xxxY) versus predictable deterministic (yyyy).
They found evidence for an increase in gamma (and theta in some cases) for unpredictable versus predictable stimuli, and a reduction in alpha/beta, which they consider evidence towards the "predictive routing" scheme.
While the dataset and analyses are well-suited to test evidence for predictive coding versus alternative hypotheses, I felt that the formulation was ambiguous, and the results were not very clear. My major concerns are as follows:
(1) The authors set up three competing hypotheses, in which H1 and H2 make directly opposite predictions. However, it must be noted that H2 is proposed for spatial prediction, where the predictability is computed from the part of the image outside the RF. This is different from the temporal prediction that is tested here. Evidence in favor of H2 is readily observed when large gratings are presented, for which there is substantially more gamma than in small images. Actually, there are multiple features in the spectral domain that should not be conflated, namely (i) the transient broadband response, which includes all frequencies, (ii) contribution from the evoked response (ERP), which is often in frequencies below 30 Hz, (iii) narrow-band gamma oscillations which are produced by large and continuous stimuli (which happen to be highly predictive), and (iv) sustained low-frequency rhythms in theta and alpha/beta bands which are prominent before stimulus onset and reduce after ~200 ms of stimulus onset. The authors should be careful to incorporate these in their formulation of PC, and in particular should not conflate narrow-band and broadband gamma.
(2) My understanding is that any aspect of predictive coding must be present before the onset of stimulus (expected or unexpected). So, I was surprised to see that the authors have shown the results only after stimulus onset. For all figures, the authors should show results from -500 ms to 500 ms instead of zero to 500 ms.
(3) In many cases, some change is observed in the initial ~100 ms of stimulus onset, especially for the alpha/beta and theta ranges. However, the evoked response contributes substantially in the transient period in these frequencies, and this evoked response could be different for different conditions. The authors should show the evoked responses to confirm the same, and if the claim really is that predictions are carried by genuine "oscillatory" activity, show the results after removing the ERP (as they had done for the CSD analysis).
(4) I was surprised by the statistics used in the plots. Anything that is even slightly positive or negative is turning out to be significant. Perhaps the authors could use a more stringent criterion for multiple comparisons?
(5) Since the design is blocked, there might be changes in global arousal levels. This is particularly important because the more predictive stimuli in the controlled deterministic stimuli were presented towards the end of the session, when the animal is likely less motivated. One idea to check for this is to do the analysis on the 3rd stimulus instead of the 4th? Any general effect of arousal/attention will be reflected in this stimulus.
(6) The authors should also acknowledge/discuss that typical stimulus presentation/attention modulation involves both (i) an increase in broadband power early on and (ii) a reduction in low-frequency alpha/beta power. This could be just a sensory response, without having a role in sending prediction signals per se. So the predictive routing hypothesis should involve testing for signatures of prediction while ruling out other confounds related to stimulus/cognition. It is, of course, very difficult to do so, but at the same time, simply showing a reduction in low-frequency power coupled with an increase in high-frequency power is not sufficient to prove PR.
(7) The CSD results need to be explained better - you should explain on what basis they are being called feedforward/feedback. Was LFP taken from Layer 4 LFP (as was done by van Kerkoerle et al, 2014)? The nice ">" and "<" CSD patterns (Figure 3B and 3F of their paper) in that paper are barely observed in this case, especially for the alpha/beta range.
(8) Figure 4a-c, I don't see a reduction in the broadband signal in a compared to b in the initial segment. Maybe change the clim to make this clearer?
(9) Figure 5 - please show the same for all three frequency ranges, show all bars (including the non-significant ones), and indicate the significance (p-values or by *, **, ***, etc) as done usually for bar plots.
(10) Their claim of alpha/beta oscillations being suppressed for unpredictable conditions is not as evident. A figure akin to Figure 5 would be helpful to see if this assertion holds.
(11) To investigate the prediction and violation or confirmation of expectation, it would help to look at both the baseline and stimulus periods in the analyses.
-
Reviewer #3 (Public review):
Summary:
In their manuscript entitled "Ubiquitous predictive processing in the spectral domain of sensory cortex", Sennesh and colleagues perform spectral analysis across multiple layers and areas in the visual system of mice. Their results are timely and interesting as they provide a complement to a study from the same lab focussed on firing rates, instead of oscillations. Together, the present study argues for a hypothesis called predictive routing, which argues that non-predictable stimuli are gated by Gamma oscillations, while alpha/beta oscillations are related to predictions.
Strengths:
(1) The study contains a clear introduction, which provides a clear contrast between a number of relevant theories in the field, including their hypotheses in relation to the present data set.
(2) The study provides a …
Reviewer #3 (Public review):
Summary:
In their manuscript entitled "Ubiquitous predictive processing in the spectral domain of sensory cortex", Sennesh and colleagues perform spectral analysis across multiple layers and areas in the visual system of mice. Their results are timely and interesting as they provide a complement to a study from the same lab focussed on firing rates, instead of oscillations. Together, the present study argues for a hypothesis called predictive routing, which argues that non-predictable stimuli are gated by Gamma oscillations, while alpha/beta oscillations are related to predictions.
Strengths:
(1) The study contains a clear introduction, which provides a clear contrast between a number of relevant theories in the field, including their hypotheses in relation to the present data set.
(2) The study provides a systematic analysis across multiple areas and layers of the visual cortex.
Weaknesses:
(1) It is claimed in the abstract that the present study supports predictive routing over predictive coding; however, this claim is nowhere in the manuscript directly substantiated. Not even the differences are clearly laid out, much less tested explicitly. While this might be obvious to the authors, it remains completely opaque to the reader, e.g., as it is also not part of the different hypotheses addressed. I guess this result is meant in contrast to reference 17, by some of the same authors, which argues against predictive coding, while the present work finds differences in the results, which they relate to spectral vs firing rate analysis (although without direct comparison).
(2) Most of the claims about a direction of propagation of certain frequency-related activities (made in the context of Figures 2-4) are - to the eyes of the reviewer - not supported by actual analysis but glimpsed from the pictures, sometimes, with very little evidence/very small time differences to go on. To keep these claims, proper statistical testing should be performed.
(3) Results from different areas are barely presented. While I can see that presenting them in the same format as Figures 2-4 would be quite lengthy, it might be a good idea to contrast the right columns (difference plots) across areas, rather than just the overall averages.
(4) Statistical testing is treated very generally, which can help to improve the readability of the text; however, in the present case, this is a bit extreme, with even obvious tests not reported or not even performed (in particular in Figure 5).
(5) The description of the analysis in the methods is rather short and, to my eye, was missing one of the key descriptions, i.e., how the CSD plots were baselined (which was hinted at in the results, but, as far as I know, not clearly described in the analysis methods). Maybe the authors could section the methods more to point out where this is discussed.
(6) While I appreciate the efforts of the authors to formulate their hypotheses and test them clearly, the text is quite dense at times. Partly this is due to the compared conditions in this paradigm; however, it would help a lot to show a visualization of what is being compared in Figures 2-4, rather than just showing the results.
-
