Visual field map clusters in human frontoparietal cortex

Essential revisions:

1) Please rerun the pRF analysis without smoothing, as requested by Reviewer 1. The Reviewing Editor (Jack Gallant) believes that smoothing should be avoided if possible and only done when absolutely necessary, but would be satisfied if the smoothed data was retained as long as the unsmoothed data is also provided.

We now include the unsmoothed data as requested. Importantly for future studies attempting to map these association areas using our pRF procedures, we compared the solutions from using the grid fit, which involves smoothing, and search fit, which does not directly. In comparison, we find that the grid fit solutions cross-validate slightly better than the search fit solutions, but both are robust and there is not statistical difference between the two (Figure 7—figure supplement 1). Critically, the map structures themselves appear almost indistinguishable; the use of the grid or search fit has no discernible impact on the structure of the visual field maps. This is not terribly surprising because the search fit is seeded by the first pass grid fit solution, and constrained to solutions near the seeds. For researchers interested in these details, we now include a supplemental figure of these comparisons and more importantly, we make the model solutions for both grid and search fits for all subjects available in the public data repository. Moreover, we include more details of the fitting procedures, including rationale and results in the text comparing the two fitting procedures.

2) Please undertake the test-retest analysis suggested by Reviewer 2. The Reviewing Editor believes that it would be preferable to use test-retest procedures everywhere. However, Reviewer 2 is most concerned about the anterior parietal findings. It is felt that if the authors cannot do the test-retest, they should back off on their parietal claims because the result appears to be somewhat inconsistent across the small group of subjects used here.

We made some minor edits to the freesurfer segmentations of a few of the hemispheres and this cleaned up the topography in PPC somewhat. Now, 9/10 hemispheres show a clear lateral foveal to medial peripheral organization that was described in Swisher et al., 2007. The remaining hemisphere also has a lateral foveal to medial peripheral organization, but appears to have some foveal representation at the medial end of the map as well. If we consider that hemisphere to be “half” consistent, then overall 9.5/10 is pretty darn consistent. We feel that this level of consistency is no longer equivocal. Moreover, in response to the reviewers we hustled to collect more data on our Allegra system in the midst of the massive construction and renovations for our new Prisma scanner. Luckily, we were able to get additional data, which allowed us to conduct test-retest analyses on two of the subjects. We strategically rescanned the two subjects whose anterior IPS eccentricity maps were the least compelling. As shown in the new figure (Figure 7), the eccentricity, as well as angle, gradient structure are indistinguishable across the two sessions, in both PCS and IPS. Therefore, we now report highly consistent eccentricity maps along IPS and we show that they are reliable across independent test-retest sessions. Remarkably, the one subject who showed what appears to be two foveal representations in IPS2/3 in session 1 also showed this in session 2, confirming both the stability of these maps and the reliability of our measurements and models. Moreover, in the first version we mainly discussed our cross-validation tests as a formal way to compare linear, non-linear, and simple laterality models. In this revision, we additionally draw attention to the fact that these cross-validation results provide additional support for the reliability of the models.

3) Please more fully analyze the iPCS as suggested by Reviewer 2.

We now include plots for iPCS for each individual subject (Figure 5—figure supplement 2), similar to sPCS, as well as a more in-depth description in the manuscript. Any further differences in the amount of space dedicated to sPCS over iPCS is only due to the discovery of coherent organization in sPCS, which seems to be lacking in iPCS. For example, although there seems to be a single confluent foveal representation in iPCS, the anatomical location varies between subjects. Additionally, there seems to be no obvious consistent organization in the polar angle representation across subjects.

4) Please address the Swisher IPS results mentioned by Reviewer 2, correcting the description and improving the analysis and interpretation of the IPS maps if possible.

First of all, we apologize to David Somers for mistaking his claims about the IPS0/1 eccentric organization. Frankly, we just mixed up our description, which flipped the direction of his reported gradient. We now correctly describe the direction of the foveal-peripheral map and we remove the comment about only showing this in a single subject, as in the manuscript they write that it was consistent in other subjects. We also expand our discussion of these findings.

5) Either the authors should provide more analysis of the sPCS1/2 maps as suggested by Reviewer 2, or they should provide a more extensive discussion of these maps and potential confounds (e.g., RF size of the constituent neurons).

We have included new figures (Figure 5—figure supplement 1 and Figure 6D) and analysis to address these concerns, as well as added an acknowledgement of the problem with our size estimation and suggested ways in which future studies might go about addressing this problem.

6) The Reviewing Editor and Senior Editor also suggest that some discussion be added to evaluate the results in the light of the recent Glasser et al. study multi-modal mapping study.

We have added discussion framing our results in relation to the Glasser et al., 2016 multi-modal parcellation (MMP) brain mapping study. Specifically, we spent some time figuring out how to transform the MMP to each subjects’ native space so that we could make a direct comparison. We did the same thing for the probabilistic maps of visual topography (PMVT; Wang et al., 2012). We have a new figure (Figure 8) that displays the overlap in our subjects. Now, we discuss the alignment between our retinotopic visual maps and those in the MMP and PMVT. In a nutshell, there is some interesting areas of overlap between all three mapping schemes, but these do not seem to be particularly consistent across participants. We acknowledge that the MMP is not an atlas, and was not designed to work on an individual based solely on anatomical landmarks. Nonetheless, readers may find it useful to “ball park” these existing maps with our frontal and parietal visual maps.

7) Although the reviewers agree that the sPCS1/sPCS2 result is solid (apart from the smoothing issue), they request that you provide further analysis of the orientation of the eccentricity and polar angle axes.

This was very good advice. We now further describe how the eccentricity-angle angle is not perfectly orthogonal. We also include a new figure that shows maps of visual field coverage plots in sPCS1/2 for individual subjects (Figure 6B). These clearly show that despite non-orthogonality the voxels in the map tile visual space. If the angle and eccentricity maps were aligned, then one would predict that the visual field coverage plot would form a spiral of the pRF centers. As can be seen, none of the subjects show anything remotely resembling a spiral. The pRFs are regularly dispersed in mostly the contralateral VF.

Reviewer #1:

[…] However, there is one potentially problematic aspect of the present analysis, which may undercut the impact of the results. As the authors describe, they use only the "grid fit" of the pRF modeling method, which as they state, is a "coarse fit [that] was solved on time series that were spatially blurred (approximating a Gaussian kernel of 5 mm width at half height) and temporally decimated (2x)". In other words, it appears that the data were smoothed spatially and temporally, which in the original pRF modeling method is done simply to save computational resources and accelerate processing time (Dumoulin & Wandell, Neuroimage 2008). This "grid fit" was not intended to be used as the final fit. Previous reports of phase-encoded methods that employed spatial or temporal smoothing came to inaccurate conclusions about the organization of visual cortex (e.g., Hasson et al., Neuron 2002; Hadjikhani et al., Nature Neurosci 1998), which were later revised by employing better methods that did not rely on such smoothing (Larson & Heeger, J Neurosci 2006; Brewer et al., Nature Neurosci 2005). While it may be an acceptable choice to employ only the smoothed "grit fit" of the pRF method in this case, perhaps to compensate for other suboptimal or difficult-to-measure aspects of the dataset, it is not the standard pRF analysis. It is appropriate that the authors did not bury this information exclusively in the methods section of the article, but the authors should further clarify that this is not simply a choice between equally accurate pRF modeling options.

Thank you, this is a great point, as we did indeed have reasoning for choosing the grid fit but unfortunately did not sufficiently explain our reasoning in the manuscript. We have added a new cross-validation analysis comparing the search and grid fit (Figure 7—figure supplement 5B). For this analysis, both the grid fit and the search fit are cross-validated against the original, unsmoothed left-out test data. Interestingly, the grid fit is at least as accurate as the search fit for all visual areas (though the differences are not statistically significant). Although the grid fit is trained on less data, it does not suffer from reduced accuracy. We believe this is due to the search fit being more susceptible to noise, while the grid fit imposes a “prior” or inherent structure in the maps by sampling every other voxel. See above in “Essential Revisions” for a description of our response to this issue. Overall, we are pleased that the map structure is similar to the two types of solutions, and we believe that the inclusion of the search fit adds further credibility to the results and conclusions.

Reviewer #2:

[…] Major Concerns:

1. The visually sensitive region in the inferior precentral sulcus (iPCS) is given inconsistent treatment across the manuscript. As with sPCS, several prior publications have reported greater visual sensitivity for contralateral than ipsilateral visual stimuli and at least one prior publication (Hagler & Sereno, 2006) reports consistent polar angle map representations in this region. In the present manuscript, iPCS is not given the full analysis given to sPCS, but oddly shows up in a broad cross-area comparison in the final figure (Figure 6). The amount of evidence presented for iPCS is insufficient to warrant inclusion in this manuscript. Either iPCS should be analyzed in the same detail as sPCS or it should be explicitly stated that no original claims are being made for this region (and any such claims should be excised). It would appear that the analysis of iPCS may suffer from the low N of this study.

We now include plots for iPCS for each individual subject, similar to sPCS, as well as a more in-depth description in the manuscript. As the cross validation of our results was high in this area for all subjects, we don’t believe our failure to identify a consistent organization is due to low power. See above in “Essential Revisions” for a description of our response to this issue.

2. […] We reported 'we reliably find a continuous gradient of eccentricity response phase along the IPS0/IPS1 border. This gradient reveals a mediolateral eccentricity representation, with a laterally position foveal representation in the fundus of the IPS, moving through parafoveal representations on the medial sulcal wall to representations of the peripheral extent of stimulation on the adjacent gyrus.' And go on to state that this organization 'defines a map complex or cluster by virtue of the confluent foveal representation.' In regards, to the IPS0/1 border, the present manuscript seems to provide only supporting evidence for these claims. While the present IPS0/1 map cluster finding should not be considered novel, the scarcity of published IPS eccentricity maps still makes this replication a worthwhile contribution to the literature. However, the discussion in the present manuscript misstates our prior finding on two points. First, it incorrectly characterizes our statement of the fovel-to-periphery gradient as medial-to-lateral, when in fact we show and state the reverse. Secondly, the manuscript states that we found this result only in a single subject; while it is true that Figure 4 displays only one subject, the finding for IPS0/1 was robust across subjects. I request that the authors revise their discussion to correctly characterize our prior findings.

Thank you for identifying our mistake. Our updated maps, test-retest results, and rewritten description now perfectly align with Swisher et al., 2007. See above in “Essential Revisions” for a description of our response to this issue.

With regards, to the more anterior IPS map cluster, the Swisher et al paper was more equivocal: 'Although the representation of eccentricity within IPS3/4 is less clear, a parsimonious interpretation of the data suggests that these anterior regions continue the pattern of foveal responses laterally and peripheral responses medially seen in the more posterior IPS, likely forming a second distinct foveal representation.' The present manuscript also presents equivocal results for the more anterior IPS foveal representation – in most subject hemispheres the (IPS2/3) foveal representation appears lateral as reported in Swisher et al, but in some subject hemispheres, a foveal representation appears at the medial edge of the IPS2/3 border. Moreover, in some subjects both lateral and medial foveal representations appear. It is not clear what can be concluded from the functional organization of this more anterior fovea representation. Larger N and/or test-retest analysis might help to clarify this issue. Alternately, claims regarding this anterior foveal cluster should be more strongly tempered.

See above in “Essential Revisions” for a description of our response to this issue.

3. For the reasons stated in point 2, the parietal lobe findings reported in this manuscript and thus the claims of 'frontoparietal visual field map clusters' appear to have lesser impact than the well-founded report of distinct sPCS1/sPCS2 regions. While parietal findings still belong in this manuscript (with revisions), the authors might wish to consider revising the title of the manuscript.

We have made revisions to the discussion of the parietal findings to address any perceived issues with inconsistency in organization and feel the parietal results, although not as novel, are still an important part of the paper. Therefore, we wish to keep the title.

4. One concern regarding the sPCS1/2 maps is that the eccentricity maps appear rather parallel to the polar angle maps, unlike in posterior regions where there is a clear orthogonal relationship between these two spatial axes. Although the discussion mentions that the two axes are 'not perfectly orthogonal,' this important issue still feels a bit swept under the rug. More quantitative analysis of the relationship between the two axes and how this varies across individuals is needed. This is an important issue as prior claims of retinotopic areas (e.g. area V8) have been attacked and eventually undermined on this issue.

As now seen in the new figures, none of the subjects have axes that are orthogonal, but none are parallel either. See above in “Essential Revisions” for a description of our response to this issue.