Neural responses during natural vision are action-timed rather than locked to the onset of stable foveal input

Reviewer #1 (Public review):

Summary:

This manuscript describes a study examining MEG responses to participants free-viewing natural visual images. The vast majority of our knowledge of visual processing in the brain comes from studies where visual input is presented during fixation and the neural response is measured relative to stimulus onset. Even studies that include eye movements tend to either analyze the data relative to the start of each new fixation, or ignore saccades as noise. The current study simultaneously measures MEG and eye-tracking during active vision, and conducts a variety of analyses testing which of the saccade-related events produce the best alignment to the neural data. Five human participants viewed thousands of complex natural scene images while freely moving their eyes. MEG data were then binned as a function of saccade duration and aligned to different fixation and saccade events. M100 responses were better aligned with the preceding saccade onset than the current fixation onset. An additional analysis showed that when MEG signals were decomposed into independent components, the majority of the components showed more alignment and variance explained from saccade-related events (saccade onset, peak velocity, peak visual motion energy, and peak saccade curvature) compared to fixation-onset-defined events; the strongest performing of these factors was the time of peak saccade curvature. A final analysis compared the similarity of MEG topographies measured from stimulus onset (as would be standard in a static design) to those linked to peak saccade curvature and fixation onset, showing that stimulus onset responses were quite dissimilar to the active vision aligned events.

Strengths:

Overall, I think this is a fundamentally important research question, taking a novel and interesting approach. I very much like the idea behind this study. My enthusiasm is somewhat tempered by the weaknesses described below. However, at the very least I think this study would be valuable as a key launching point for future explorations, and for pushing the field into a much-needed new direction.

Weaknesses:

In its current state, the manuscript seems preliminary/incomplete in terms of both data analysis and engagement with the prior literature.

(1) In terms of the theoretical contribution, there are several potential contributions, some supported more by the data than others, and some more novel than others. In my rough assessment, from most general to most specific:
a. Static vision is not the same as active vision. Supported somewhat by the analyses. Not novel (there are several studies both recent and older making this point, aside from the vaguely referenced sink-source sentence in the discussion), but this is still an understudied/underappreciated area.
b. Neural responses are better aligned to saccade-related events than fixation-related events. Supported pretty compellingly by the analyses, and pretty novel. An important theoretical contribution.
c. Peak saccade curvature is the saccade-related event explaining most variance. An extremely novel finding, but not well supported by the current data. At best, this seems a preliminary, exploratory hint of something to investigate further. It's intriguing but lacking in both empirical support (e.g. is this even consistent across subjects?) and theoretical discussion (what would it mean / what would be the mechanisms of such a link?).

(2) There is a small number of subjects, and for several main analyses, the data are pooled across them. Small N's can be reasonable in cases where there is large data for each subject. But it is standard to show the subjects individually to confirm reliability. Figure 1 does this nicely, but then for the main analyses examining the ICs and variance explained by the different saccade-related events (Figures 2C-F), the data were pooled across subjects. Strong conclusions are being drawn from the pooled data (e.g. highest proportion of explained variance from the peak curvature event), but it's unclear if this is consistent across subjects or potentially dominated by 1 or 2 subjects. Indeed, when the "best" score is presented for each participant (Fig 2E), only 2 of the 5 subjects showed peak saccade curvature as the best. And these results look strikingly different across subjects (P5 doesn't even look anything like an M100 response).

(3) Several parts of the results and methods are hard to follow. I had to read the paper several times to understand it. In many cases, the methods text doesn't even link with the results (e.g. the term "M100" is not anywhere in the methods).

(4) Several parts of the results felt under-explored:
a) The analysis in Figure 1E is very interesting, but it's not reported in enough detail. There are no quantitative results here, just a visual of a distribution and a description of it being broad. I would be particularly interested in seeing the mean alpha reported for the best sensor for each participant (i.e. linking with the rest of that figure).
b) How consistent is the timepoint of peak saccade curvature? It appears to increase with saccade duration, but is it a fixed / consistent percentage of saccade duration? If not, what factors cause it to vary? How similar is this timepoint to the optimal alpha from the analysis in Figure 1E? Would binning the data based on peak saccade curvature instead of saccade duration produce even better alignments for Figure 1D?
c) For the Figure 3 analysis comparing static scene-onset responses to the saccade- and fixation-related responses: I am wondering how much of the difference is actual saccade-related activity vs a true difference in visual processing. It seems the interpretation is that "visual processing", when measured in static contexts, is very different from when measured in active contexts. But what's being compared is not visual processing specifically, but the entire whole-brain MEG response. I think in order to make this conclusion more compelling, there needs to be some way of filtering out these influences. E.g., a study that presents a simulated saccade condition, where a participant keeps their eyes fixated but views snapshots of the visual scene mimicking the exact saccade sequence of another subject.

(5) The discussion felt too thin. See some specific points below. In general, combined with the fact that the results were often hard to follow and sparse, I was left with the impression that this report was being forced into a shorter format than necessary.

(6) How do microsaccades and other types of eye movements fit into this story?

Reviewer #2 (Public review):

Summary:

Although our visual system is continuously analyzing the current visual scene, its processing proceeds in discrete episodes separated by brief eye movements (saccades). It has generally been assumed that the analysis of the next visual snapshot begins in earnest when the eyes land on a new fixated location just after a saccade, but there have been various studies indicating that at least some amount of processing occurs earlier, as the system anticipates the impending eye movement. Here, the authors use magentoencephalography (MEG) measurements to record visually-driven responses and determine at what point exactly the processing of a new visual snapshot begins.

Strengths:

(1) The work is concise and to the point, and the techniques used are a good way to answer the underlying question about visual processing, since they reflect widespread activity in the brain (rather than activity at a particular location or structure).

(2) The use of natural images and extensive data collection from 5 participants is a nice feature of the experimental design which permits characterization of the common effects and of variance across individuals.

(3) The data are analyzed rigorously, but the results are also understood intuitively; for instance, by visual comparison of responses aligned on fixation onset versus saccade onset.

(4) The results provide a clean characterization of when visual analysis begins relative to saccade onset under natural viewing conditions.

Weaknesses:

(1) There were questions about how the scene-onset condition was established, and how data were selected for it.

(2) The significance of the results is slightly overstated; the text would benefit if some of the claims were phrased with a bit more carefully.

(3) In particular, the issue of how motor-related processes (versus stimulus-related content) may determine the processing of the next visual snapshot should be discussed with a bit more nuance.

These are minor weaknesses. Overall, I found the work to be novel and instructive, as it bridges neurophysiological and psychophysical findings in a satisfactory way.

Reviewer #3 (Public review):

Summary:

This manuscript addresses a fundamental question in cognitive neuroscience: which event should serve as the temporal reference for neural processing during natural vision? While fixation onset has traditionally been treated as the analogue of stimulus onset in free-viewing experiments, the authors convincingly demonstrate that this assumption is incomplete.

The study utilizes a remarkable natural-viewing dataset consisting of simultaneous MEG and eye-tracking recordings collected during the exploration of thousands of natural scenes. The authors compare several candidate eye-movement events and evaluate which event best explains the timing of the early M100 response. Across several complementary analyses, saccade-related events consistently outperform fixation onset, with peak saccade curvature emerging as the event that best predicts neural response timing.

Strengths:

A particular strength of the work is that the conclusions do not rely on a single analytical approach. Instead, multiple independent analyses converge on the same interpretation, increasing confidence that the observed timing relationships are robust rather than analysis-specific. The comparison between natural-viewing responses and classical stimulus-onset responses is especially compelling and highlights qualitative differences in their spatiotemporal organization. Of particular conceptual importance, the findings support the broader perspective that perception is intrinsically linked to action and internally generated sensorimotor processes. This aligns well with growing evidence that oculomotor action and active sampling play central roles in perception. The work contributes to an important ongoing shift in how natural vision should be studied experimentally and interpreted theoretically.

Weaknesses:

I identified no major weaknesses in the study. The main limitation is the relatively small number of participants, despite the exceptionally rich dataset. Future work in larger cohorts and across complementary electrophysiological recording modalities will help establish the generalizability of the reported temporal relationships.

Author response:

We would like to thank the editor and reviewers for their constructive and thoughtful feedback. We appreciate the reviewers' assessment that our work addresses a fundamentally important research question through a novel approach. We are also glad the reviewers found our data to be rigorously analysed, and that they valued our focus on the whole cortex rather than localised regions or electrodes. We are encouraged by the overall assessment of our work and welcome the suggestions for improving the manuscript. Below, we summarise how we plan to address the reviewers' comments in our revision:

Analyses

- We will include quantitative results for Fig. 1E that describe the distribution of the optimal alpha values across sensors and participants.

- For Figures 2 and 3 we will include analyses of individual participants in the Appendix.

- We will provide more detailed descriptions on how the timing of peak saccade curvature relates to saccade onset and to the identified optimal alpha value.

Presentation of Methods and Results

- We will phrase our claims and conclusions more carefully and nuanced throughout, ensuring direct coverage by the data and analyses.

- We will revise the currently complex sections of the Methods and Results to improve clarity and readability.

- We will be more explicit about how the data for scene onset were selected.

Revision of the Discussion

- We will extend the Discussion section to address possible mechanisms linking the timing of peak saccade curvature and ERF initiation. We will also provide a more thorough discussion of existing and more recent literature on the topic.

- We will emphasise the main takeaway of the study: the observation that saccade-related processes are more important to the M100 than previously thought, and, reversely, that this component may be less directly related to fixation-locked responses. We will also present our observation of peak saccade curvature as a starting point for future research, as it was not intended as conclusive mechanistic insight into how and why this process relates to early cortical responses.