Development of visual cortex in human neonates is selectively modified by postnatal experience
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
Overall, this study will make significant contributions to developmental neuroscience and vision science as they attempt to study how prenatal and postnatal maturation influence structural-functional measurements in the early and high-level visual cortex. These results will be of broad interest as it is a novel attempt to study processes that might be innate or genetically wired and those that emerge due to worldly experiences within the sensory systems. The authors are addressing an important and timely question based on a large and impressive infant database.
(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 #2 agreed to share their name with the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Experience-dependent cortical plasticity is a pivotal process of human brain development and essential for the formation of most cognitive functions. Although studies found that early visual experience could influence the endogenous development of visual cortex in animals, little is known about such impact on human infants. Using the multimodal MRI data from the developing human connectome project, we characterized the early structural and functional maps in the ventral visual cortex and their development during neonatal period. Particularly, we found that postnatal time selectively modulated the cortical thickness in the ventral visual cortex and the functional circuit between bilateral primary visual cortices. But the cortical myelination and functional connections of the high-order visual cortex developed without significant influence of postnatal time in such an early period. The structure–function analysis further revealed that the postnatal time had a direct influence on the development of homotopic connection in area V1, while gestational time had an indirect effect on it through cortical myelination. These findings were further validated in preterm-born infants who had longer postnatal time but shorter gestational time at birth. In short, these data suggested in human newborns that early postnatal time shaped the structural and functional development of the visual cortex in selective and organized patterns.
Article activity feed
-
-
Author Response
Reviewer #1 (Public Review):
Using a large neonatal dataset from the developmental Human Connectome project, Li and colleagues find that cortical morphological measurements including cortical thickness are affected by postnatal experience whereas cortical myelination and overall functional connectivity of ventral cortex developed significantly were not influenced by postnatal time. The authors suggest that early postnatal experience and time spent inside the womb differentially shape the structural and functional development of the visual cortex.
The use of large data set is a major strength of this study, furthermore an attempt to examine both structural and functional measures, and connectivity analysis and separating these analyses based on the pre-and full-term infants is impressive and strengthens the claims …
Author Response
Reviewer #1 (Public Review):
Using a large neonatal dataset from the developmental Human Connectome project, Li and colleagues find that cortical morphological measurements including cortical thickness are affected by postnatal experience whereas cortical myelination and overall functional connectivity of ventral cortex developed significantly were not influenced by postnatal time. The authors suggest that early postnatal experience and time spent inside the womb differentially shape the structural and functional development of the visual cortex.
The use of large data set is a major strength of this study, furthermore an attempt to examine both structural and functional measures, and connectivity analysis and separating these analyses based on the pre-and full-term infants is impressive and strengthens the claims made in the paper. While I find this work theoretically well-motivated and the use of the large dHCP dataset very exciting, there are some concerns, that need to be addressed.
There is a bit of confusion if the authors really compared the structural-functional measures in the final analysis. If the authors wish to make claims about the relationship, then there must be a compelling analysis detailing these findings.
Thanks for the suggestions. We have added analysis to directly investigate the relationship between the development of homotopic connection and corresponding structural measurements in the area V1 (Page 13 Line 5-16):
“The above results revealed that structural and functional properties of the ventral visual cortex both developed with PMA, but were differently influenced by the in-utero and external environment (Table 1). We further investigated the relationship between structural and functional development based on area V1, which showed a strong developmental effect in both structural and functional analyses. Mediation analysis was employed to see whether the development (GA or PT) of the homotopic connection between bilateral V1 was mediated by the structural properties (CT or CM). We found that the PT had a significant direct effect on the homotopic function that was not mediated by CT or CM (Fig 6a-b). In contrast, the direct effect of GA on the homotopic connection was not significant but the indirect effect of GA through CM on the connection was significant (Fig 6c-d).”
There is also a bit of confusion in the terminology used in the study regarding ages; the gestational age, premenstrual age, and postnatal time. I think clarifying and simplifying it down to GA and postnatal time will help the reader and avoid confusion.
Thank you for the suggestion. We have made extensive revision regarding the terminology throughout the paper and simplified it down to GA and PT. Please see the response to the 1st major concern in the Essential Revisions (for the authors) section above.
*Reviewer #2 (Public Review):
The authors utilize the publicly available dHCP dataset to ask an interesting question: how does postnatal experience and prenatal maturation influence the development of the visual system. The authors report that experience and prenatal maturation differentially contribute to different aspects of development. Namely, the authors quantify cortical thickness, myelination, and lateral symmetry of function as three different metrics of development. The homotopy and preterm infant analyses are strengths that, on their own, could have justified reporting. However, I have concerns about the analytic approaches that were used and the conclusions that were drawn. Below I list my major concerns with the manuscript.
PMA vs. GA vs. PT
The authors seek to understand the contribution of experience and prenatal development, yet I am unsure why the authors focused on the variables they did. There are three variables of interest used throughout this study: Gestational age at birth (GA), postnatal time (PT), and postmenstrual age at the time of scan (PMA). The last metric, PMA, is straightforwardly related to GA and PT since PMA = GA + PT. In most (but not all) of the manuscript, the authors use PMA and PT, with GA used without justification in some cases but not in others.
It is unclear why PMA is used at all: PMA is necessarily related to PT and GA, making these variables non-independent. Indeed, the authors show that PMA and PT are highly correlated. The authors even say that "the contribution of postnatal experience to the development was not clarified because PMA reflects both prenatal endogenous effect and postnatal experience." So, why not use GA at birth instead of PMA? Clearly, GA is appropriate in some cases (e.g., Figure S4 or in some of the ANOVA applications), and to me, it seems to isolate the effect the authors care about (i.e., duration of prenatal development). Perhaps there is some theoretical justification for using PMA, but if so, I am unaware.
That said, I expect that replacing all analyses involving PMA with GA will substantially change the results. I do not see this as a bad thing as I think it will make the conclusions stronger. As is, I am left unsure about what the key takeaways of this paper are.
We appreciate the suggestions, and we have replaced the related analyses involving PMA with GA in the manuscript. Please see the Response to the 1st major concern in the Essential Revisions (for the authors) section above for more detail.
Using GA instead of PMA will have several benefits: 1) It will be much simpler to think of these two variables since they contrast the duration of fetal maturation and time postnatally. 2) This will help the partial correlation analyses performed since the variance between the variables is more independent. It will also mean that the negative relationships observed between PT and cortical thickness when controlling for PMA (e.g., Figure 2h) might disappear (reversed signs for partial correlations are common when two covariates are correlated). 3) this will allow the authors to replace Figure 1a with a more informative plot. Namely, they could use a scatter of GA and PT, giving insight into the descriptive statistics of both dimensions.
We have revised the manuscript throughoutly following the reviewer’s suggestion. However, we thought it would be necessary to show the overall development of CT and CM across the general age (PMA) in Figure 1. Therefore, we didn’t replace the figure 1a but added a scatter figure between GA and PT in Figure 2-figure supplement 1 and added descriptive statistics of them in the manuscripts: “The mean GA of the neonates was 39.93 weeks (SD = 1.26) and the mean PT was 1.21 weeks (SD = 1.25), the correlation between them was not significant (r = - 0.08, p > 0.1; Figure 2-figure supplement 1).” Moreover, the negative relationships between PT and CT when controlling for PMA disappeared in the revised results as the reviewer’s predicted.
I suspect that one motivation for the use of PMA over GA is for the analysis in Figure 6. In this analysis, the authors pick a group of term infants with a PMA equal to the preterm infants. Since PMA is the same, the only difference between the groups (according to the authors) is the amount of postnatal experience. However, this is not the only difference between the groups since they also vary in GA (and now PT and GA are negatively correlated almost perfectly). I don't know how to interpret this analysis since both the amount of prenatal maturation and postnatal experience vary between the groups.
We appreciate the reviewer’s opinion that both GA and PT were different between preterm and term-born neonates. Then any of the differences between the two groups might came from the combined effect of GA and PT in our results, and unfortunately, we might not able to separate them in this analysis. However, the preceding results indicated that the CT was significantly influenced by PT and GA while CM was significantly influenced by GA, which So we discuss the preterm and term-born comparison in the context of these findings (Page 19 Line 26-29 and Page 20 Line 1-5): “We found CT in the ventral cortex was generally lower in the term-born than preterm-born infants, while the CM showed the opposite trend in the two groups. Since the preterm babies have longer PT but shorter GA compared to full-term infants at the same PMA, this result supported the above analysis that CT was preferably influenced by PT while CM was largely dependent on GA during the neonatal period”. Furthermore, we added a description in the limitation section to stress the caveat (Page 20 Line17-19): “Meantime, both GA and PT were different between preterm and term-born neonates. Then any of the differences between the two groups might came from the combined effect of GA and PT, and unfortunately, we were not able to separate them in this study.”
Justification of conclusions and statistical considerations
I had concerns about some of the statistical tests and conclusions that the authors made. I refer to some of these in other sections (e.g., the homotopy analyses), but I raise several here.
I am not sure what evidence the authors are using to make this claim: "we found that the cortical myelination and overall functional connectivity of ventral cortex developed significantly with the PMA but was not directly influenced by postnatal time." Postnatal time is significantly correlated with cortical myelination, as shown in Figures 2g, 2h, 3b, 3c, and postnatal time is significantly correlated with functional connectivity, as shown in Figures 4h, 5c, 5d, and 5e. Hence, this general claim that "the development of CT was considerably modulated by the postnatal experience while the CM was heavily influenced by prenatal duration" doesn't seem to be supported: both myelination and thickness are affected by postnatal experience and prenatal duration (as measured by PMA). A similar sentiment is expressed in the abstract. Perhaps the authors suggest different patterns in the strength of change for PMA vs. PT across these metrics, but if so, then statistical tests need to support that conclusion, and the claims need to reflect that sentiment.
Interestingly, Figure S4 presents a compelling ANOVA that does support this conclusion. Still, this result is relegated to the supplement, and it also uses GA, rather than PMA, making it hard to reconcile with the other claims made in the main text. Moreover, it uses ANOVAs, which dichotomizes a continuous variable. Here and elsewhere in the manuscript (e.g., Figures 3d, 3e), the authors split the infants into quartiles and compare them with ANOVAs. Their use for visualization is helpful, but it is unclear what the statistical motivation for this is rather than treating these as continuous variables like is possible with linear mixed-effects models. Moreover, it is unclear why the authors excluded half the data from the study (i.e., quartiles 2 and 3) in this ANOVA when all four quartiles could be used as factors.
We appreciate the reviewer’s comments. We have clarified our results and conclusion in the revised manuscript based on the new analyses that replaced PMA with PT and GA (See the response to the 1st major concern in the Essential Revisions). The previous claims have been changed as following:” the postnatal time could modulate the cortical thickness in ventral visual cortex and the functional circuit between bilateral primary visual cortices. But the cortical myelination, particularly that of the high-order visual cortex, developed without significant influence of postnatal time in such early period” (Page 2, Lines 8-12). This claims could be supported by the results in figure 2. Moreover, to support the claims about the comparison of the influence between GA and PT on structural development, we replaced the ANOVA analysis with a linear mixed-effect model as the reviewer mentioned.
To compare the influence of GA against PT on the structural development in the whole ventral visual cortex (Page 7 Line 15-19), “We applied a linear mixed-effect model to test whether the CT (or CM) of the whole ventral cortex were differently influenced by the GA vs. PT, and found that the GA had a significantly stronger effect on the CM than PT (interaction between GA and PT, p < 0.05) but no significant difference was found of the effect on the CT between the ages (p > 0.6).”
To compare the influence of GA against PT on the structural development in the area V1 and VOTC, we applied a similar linear mixed-effect model analysis for the two ROIs (Page 8 Line 17-18 and Page 9 Line 1-4): “Moreover, we applied a linear mixed-effect model to test the developmental influence of GA vs. PT on the cortical structure , and the results showed that the CT in two ROIs showed non-significantly different influences from GA against PT (p > 0.3), but CM showed at least marginally significant results in both two ROIs (V1: p < 0.01 and VOTC: p < 0.09).”
It is unclear what the evidence is to support the following claim: "Both CT and CM show higher correlation with PMA in the posterior than anterior region, and higher correlation in the medial than lateral part within the anatomical mask (Figure 2a and Figure S2b-c [sic])" From Figure 2 or Figure S2, I don't see a gradient. From Figure S3, there might be a trend in some plots, but it is hard to interpret since it is non-monotonic. More generally, is there a statistical test to support this claim?
We added a correlation analysis between the diction (x: lateral to medial; y: posterior to anterior) and measurements (CT and CM) in the ventral visual cortex, and the resulting coefficient was all significant (r = 0.7/-0.8 for CT along x/y axis, and r = 0.91/-0.83 for CM along x/y axis; p < 0.001). See Figure 1-figure supplement 2. However, the consideration provided by the reviewer still exists that such significance was driven by part of the areas and the gradient was non-monotonic. Therefore, we replaced the original claim with the following sentence (Page 6 Line 3-8): “In addition, we found distinct spatial variation along ventral cortex, e.g. posterior-anterior and medial-lateral directions (Figure 1-figure supplement 2a-b). Generally, both CT and CM showed higher correlation with PMA in the posterior than anterior region (r = -0.8 and -0.83; p < 0.001), and higher correlation in the medial than lateral part within the ventral visual cortex (r = 0.7 and 0.91; p < 0.001; Figure 1-figure supplement 2c-d).”.
"and the interaction [sic] was more prominent in CM (simple effect: t = 10.98, p < 10-9) that in than CT (t = 2.07, p < 0.05)." Does 'more prominent' mean it is 'significantly stronger'? If not, then the authors should adjust this claim
The claim ‘more prominent’ did express ‘significantly stronger’ since we found that the interaction between CM and CT along PMA or PT was significant in the ANOVA analysis. This analysis has been removed because we thought that the comparison between two structural measurements is not very relevant to the conclusion of the paper. We now applied a linear mixed-effect model to compare the influence of GA against PT on specific structural development. So this result and claim have been removed from the new manuscript.
Are the authors Fisher Z transforming their correlations? In numerous places, correlation values seem to be added together or used as the input to other correlation analyses. It is unclear from the methods whether the authors are transforming their correlation values to make that use appropriate.
We are sorry for the confusion. All the statistical analyses involving correlation coefficients were Fisher-Z transformed. We have added a clear description in the manuscripts involving the Fisher-Z transformation (Page 25 Line 16-18).
Homotopy analyses
The homotopy section is a strength of the paper, but I have doubts about the approach taken to analyze this data and some of the conclusions drawn. I don't expect any of my suggestions to change the takeaway of this section, but I do think they are essential criticisms to address.
I do not think that the non-homotopic control condition is appropriate. In Arcaro & Livingstone (2017), the authors had 3 categories for this analysis: homotopic pairs (e.g., left V1 vs. right V1), adjacent pairs (e.g., left V1 vs. right V2), and distal pairs (e.g., left V1 vs. right PHA1). In the homotopy analysis performed by Li and colleagues, they compare homotopic pairs with all other pairs. I don't think that is generous to the test since non-homotopic pairs include adjacent pairs that should be similar and distal pairs that shouldn't be similar. This may explain why some non-homotopic distribution overlaps with the homotopic distribution in Figure 4c.
Thanks for these suggestions. In the revised manuscript, we reanalyzed the data by dividing the connections into three groups for each subject. See Page 26 Line 24-29: “For each subject, Pearson correlations were carried out on the ROI-averaged time series within and across the left and right ventral cortex. The resulting connections were divided into three groups, namely the homotopic connection (the connection between two paired areas in two hemispheres. e.g. right and left V1), adjacent connection (e.g., right V1 and left V2 since V1 and V2 are adjacent) and distant connections (two areas that were not the paired or adjacent)”.
Regardless of this decision, I think the authors should reconsider their statistical test. I think the authors are using a between samples t-test to compare the 34 homotopic pairs with the hundreds of non-homotopic pairs. This is statistically inappropriate since the items are not independent (i.e., left V1 vs. right V1 is not independent of left V1 vs. right V2, which is also not independent of left V3 vs. right V2). This means the actual degrees of freedom are much lower than what is used. Moreover, I am unsure how the authors do this analysis across participants since this test can be done within participants. The authors should clarify what they did for this analysis and justify its appropriateness.
Thank you for the suggestion. In the previous manuscript, we first averaged the connection matrix across subjects and then calculated the homotopic (or non-homotopic) connections between areas, and therefore, statistical analysis could not be performed. In the revised paper, we calculated the three groups of connections for each subject before the average. We applied a non-parameter statistical analysis (Wilcoxon signed-rank) to address the issue of the independent comparison among the connections, and found the homotopic connections were significantly stronger than the adjacent or distant connections.
See (Page 26 Line 29 and Page 27 Line 1-3): “Independent-sample T-test was used to test whether the homotopic correlation was significantly greater than zero across subjects. To compare the correlation among the three types of connections, we applied a non-parameter statistical analysis (Wilcoxon signed-rank) across subjects”.
The results showed that (Page 9 Line 17-21) “the homotopic connections in all ROIs of ventral cortex were significant (mean r = 0.13– 0.43, t > 12.87, s < 10-9; Fig 4a-b), and were significantly higher than adjacent connections (0.29 ± 0.12 vs. 0.19 ± 0.10, Wilcoxon signed rank test on the Fisher-Z transformed r value: z = 16.32, p < 10-9) and distal connections (0.04 ± 0.06, z = 16.32, p < 10-9; Fig. 4c)”.
Could the authors speculate on why the correlations in homotopic regions are so much lower than what Arcaro and Livingstone (2017) found. I can think of a few possibilities: higher motion in infants, less rfMRI data per participant, different sleep/wake states, and different parcellation strategies. Regarding the last explanation, I think this is a real possibility: the bilateral correlation may be reduced if the Glasser atlas combines functionally heterogeneous patches of the cortex. Hence, the authors should consider this and other possible explanations.
Thank you for the suggestion. The neonates included in this study were all under natural sleep during the scan, so sleep/wake states would not be one of the causes. We added some possible reasons for this difference following the related results (Page 19 Line 9-13): “However, the present homotopic connections in the human neonates were lower than those in neonate macaca mulattas (Arcaro and Livingstone, 2017). This difference might relate to the higher motion in human infants, less r-fMRI data in the present study, coarser parcellation in the visual cortex used in this work, and the developmental difference between primates and humans in the neonatal period.”
The authors assume that the homotopic analyses mean that there are lateral connections between hemispheres (e.g., "Furthermore, the connections among the ventral visual cortex have developed during this early stage. Specifically, the homotopic connections between bilateral V1 and between bilateral VOTC both increased with GA, indicating an increased degree of functional distinction"). While this might be true, it doesn't need to be. Functional connectivity can be observed between regions that lack anatomical connectivity. Instead, two regions could both be driven by another region. In this case, the thalamus might drive symmetrical activity in the visual cortex.
We agree with the reviewer’s view that the development of functional connectivity might be driven by other regions like thalamus. So we added this interpretation in the discussion section (Page 19 Line 23-25): “It is worth noting that the increased homotopic connection can be direct or indirect, e.g., the effect might be driven external regions with enhanced connection to both of the areas (e.g. thalamus)”.
Miscellaneous
I am not sure what the motivation of this line is: "Moreover, those studies did not fully control the visual experience in the first few weeks of the subjects, thus cannot give a clear conclusion whether the innate functional connectivity is unrelated to postnatal visual experience." Arcaro, Schade, Vincent, Ponce, & Livingstone (2017) did control the visual experience of subjects. Moreover, the research here doesn't control infant experience in the way this sentence implies: it implies an experiment manipulation (i.e., fully control) rather than a statistical control that is done here. Consider rephrasing
We have rephrased this sentence in the introduction section (Page 5 Line 2-5): “Moreover, the human infants participating in a previous study (Kamps et al., 2020) were around one month old (mean age: 27 d; range from 6 to 57 d), who might already acquire some visual experience, and thus this study could not exclude postnatal visual experience on the innate functional connectivity”.
I am not sure why this claim is made: "Area V1 was selected because this region is the most basic region for visual processing and probably is the most experience-dependent area during early development". Is there evidence supporting this claim? Plasticity is found throughout the visual cortex, and I think which region is most plastic depends on the definition of plasticity. For instance, most people have the same tuning properties to gabor gratings (e.g., a cardinality bias), but there is enormous variability in face tuning across cultures.
We have removed this claim in the manuscript.
The abstract says 783 infants were included in this study, but far fewer are actually used. The authors should report the 407 number in the abstract if any number at all.
We have revised the number accordingly.
Any comparisons of preterms and terms ought to be given the caveat that the preterm environment can be very different than the term environment: whereas a term infant goes home and sees friends and family without restriction, the preterm environment can be heavily regulated if they are in a NICU. Authors should either provide details about the environments of the preterms in their study, or they should consider how differences in the richness of visual experience - regardless of quantity - may affect visual development.
We agree with the reviewer’s concern, and added a paragraph in the limitation section to stress the caveat (Page 20 Line 12-16): “One limitation of this study is the comparison between preterm and term-born infants did not consider the different visual experience in these infants. The preterm-born neonates may experience very different environment than those of the term-born, e.g. the preterm environment can be heavily regulated if they were in a NICU, but we didn’t have detailed information about the postnatal environment to control for it.”
Reviewer #3 (Public Review):
The authors use a large neonatal dataset to examine how development may occur differently based on whether on not the neonate spent that time in gestation or out of the womb accruing potentially accruing visual experience. In this manner, the authors hope to tease apart those aspects of development that are biologically programmed versus those that occur in response to experience within the visual cortex. They show structurally that cortical thickness is affected by postnatal experience while cortical myelination is not, and functionally they find regional differentiation present between visual areas at birth and that their connectivity changes with development and postnatal experience. The conclusions seem well supported by the data and analyses and provide some insight into which aspects of brain structure at birth are sculpted more by postnatal experience and which are more determined by endogenous developmental timelines.
The analyses are based on a large sample of infants, and the authors were careful to statistically separate which aspects of an infant's age, gestational or postnatal, are driving brain development, providing a deeper picture of infant brain development than previous publications. Overall, the findings seem well supported by the data as the analyses are relatively straightforward.
Visualization of the data and findings could be improved, as a few figures are difficult to interpret without having to read the methods.
We have extensively revised the figures in the manuscript to improve the readability. See updated Figures 2-7.
The acronyms regarding gestation, postnatal, and post-menstrual time are a little distracting. Please consider explicitly writing "gestational time" etc when referring to these numbers to improve readability.
We have replaced the analyses involving PMA with gestational age (GA) or postnatal time (PT) in the revised manuscript to simplify the terminology. Please see the Response to the 1st major concern in the Essential Revisions (for the authors) section above. We believe this change makes the paper easier to follow even with the abbreviations.
Because the cortical ribbon of infants is so thin at birth, there seems to be a possibility that partial-volume effects could be more prevalent in less-developed infants and impact myelin metrics. If not modeled or estimated, it should at least be discussed.
In fact, the cortical thickness of the neonatal brain is not thinner than that of the adult. Particularly, the average cortical thickness of infants aged 0-5 months is around 2-2.5 mm (Wang et al., 2019), which is similar to adults (Fjell et al., 2015). Therefore, the partial-volume effect for cortical gray matter is not a special concern for infants.
Nevertheless, we agree that the partial-volume effects might have different influences on infants of different ages. We added this consideration in the limitation section (Page 20 Line 20-24). “Another concern was about the partial-volume effect on the cortical measurements. The changing thickness of cortical ribbon during development may changes the degree of partial-volume effect, and thus may affect the cortical myelination measurement and may contribute to the myelination difference observed between preterm and term-born groups.”
Structural and functional development could be more formally compared using quantitative models if the authors want those points more strongly related; the two are only qualitatively discussed at present.
We have added a formal analysis to investigate the relationship between structural and functional development. Please see the Response to the 1st concern of Reviewer 1 (public review).
-
Evaluation Summary:
Overall, this study will make significant contributions to developmental neuroscience and vision science as they attempt to study how prenatal and postnatal maturation influence structural-functional measurements in the early and high-level visual cortex. These results will be of broad interest as it is a novel attempt to study processes that might be innate or genetically wired and those that emerge due to worldly experiences within the sensory systems. The authors are addressing an important and timely question based on a large and impressive infant database.
(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 #2 agreed to share their name with the authors.)
-
Reviewer #1 (Public Review):
Using a large neonatal dataset from the developmental Human Connectome project, Li and colleagues find that cortical morphological measurements including cortical thickness are affected by postnatal experience whereas cortical myelination and overall functional connectivity of ventral cortex developed significantly were not influenced by postnatal time. The authors suggest that early postnatal experience and time spent inside the womb differentially shape the structural and functional development of the visual cortex.
The use of large data set is a major strength of this study, furthermore an attempt to examine both structural and functional measures, and connectivity analysis and separating these analyses based on the pre-and full-term infants is impressive and strengthens the claims made in the paper. …
Reviewer #1 (Public Review):
Using a large neonatal dataset from the developmental Human Connectome project, Li and colleagues find that cortical morphological measurements including cortical thickness are affected by postnatal experience whereas cortical myelination and overall functional connectivity of ventral cortex developed significantly were not influenced by postnatal time. The authors suggest that early postnatal experience and time spent inside the womb differentially shape the structural and functional development of the visual cortex.
The use of large data set is a major strength of this study, furthermore an attempt to examine both structural and functional measures, and connectivity analysis and separating these analyses based on the pre-and full-term infants is impressive and strengthens the claims made in the paper. While I find this work theoretically well-motivated and the use of the large dHCP dataset very exciting, there are some concerns, that need to be addressed.
1. There is a bit of confusion if the authors really compared the structural-functional measures in the final analysis. If the authors wish to make claims about the relationship then there must be a compelling analysis detailing these findings.
2. There is also a bit of confusion in the terminology used in the study regarding ages; the gestational age, premenstrual age, and postanal time. I think clarifying and simplifying it down to GA and postnatal time will help the reader and avoid confusion. -
Reviewer #2 (Public Review):
The authors utilize the publicly available dHCP dataset to ask an interesting question: how does postnatal experience and prenatal maturation influence the development of the visual system. The authors report that experience and prenatal maturation differentially contribute to different aspects of development. Namely, the authors quantify cortical thickness, myelination, and lateral symmetry of function as three different metrics of development. The homotopy and preterm infant analyses are strengths that, on their own, could have justified reporting. However, I have concerns about the analytic approaches that were used and the conclusions that were drawn. Below I list my major concerns with the manuscript.
PMA vs. GA vs. PT
1. The authors seek to understand the contribution of experience and prenatal …
Reviewer #2 (Public Review):
The authors utilize the publicly available dHCP dataset to ask an interesting question: how does postnatal experience and prenatal maturation influence the development of the visual system. The authors report that experience and prenatal maturation differentially contribute to different aspects of development. Namely, the authors quantify cortical thickness, myelination, and lateral symmetry of function as three different metrics of development. The homotopy and preterm infant analyses are strengths that, on their own, could have justified reporting. However, I have concerns about the analytic approaches that were used and the conclusions that were drawn. Below I list my major concerns with the manuscript.
PMA vs. GA vs. PT
1. The authors seek to understand the contribution of experience and prenatal development, yet I am unsure why the authors focused on the variables they did. There are three variables of interest used throughout this study: Gestational age at birth (GA), postnatal time (PT), and postmenstrual age at the time of scan (PMA). The last metric, PMA, is straightforwardly related to GA and PT since PMA = GA + PT. In most (but not all) of the manuscript, the authors use PMA and PT, with GA used without justification in some cases but not in others.
It is unclear why PMA is used at all: PMA is necessarily related to PT and GA, making these variables non-independent. Indeed, the authors show that PMA and PT are highly correlated. The authors even say that "the contribution of postnatal experience to the development was not clarified because PMA reflects both prenatal endogenous effect and postnatal experience." So, why not use GA at birth instead of PMA? Clearly, GA is appropriate in some cases (e.g., Figure S4 or in some of the ANOVA applications), and to me, it seems to isolate the effect the authors care about (i.e., duration of prenatal development). Perhaps there is some theoretical justification for using PMA, but if so, I am unaware.
That said, I expect that replacing all analyses involving PMA with GA will substantially change the results. I do not see this as a bad thing as I think it will make the conclusions stronger. As is, I am left unsure about what the key takeaways of this paper are.
2. Using GA instead of PMA will have several benefits: 1) It will be much simpler to think of these two variables since they contrast the duration of fetal maturation and time postnatally. 2) This will help the partial correlation analyses performed since the variance between the variables is more independent. It will also mean that the negative relationships observed between PT and cortical thickness when controlling for PMA (e.g., Figure 2h) might disappear (reversed signs for partial correlations are common when two covariates are correlated). 3) this will allow the authors to replace Figure 1a with a more informative plot. Namely, they could use a scatter of GA and PT, giving insight into the descriptive statistics of both dimensions.
3. I suspect that one motivation for the use of PMA over GA is for the analysis in Figure 6. In this analysis, the authors pick a group of term infants with a PMA equal to the preterm infants. Since PMA is the same, the only difference between the groups (according to the authors) is the amount of postnatal experience. However, this is not the only difference between the groups since they also vary in GA (and now PT and GA are negatively correlated almost perfectly). I don't know how to interpret this analysis since both the amount of prenatal maturation and postnatal experience vary between the groups.
Justification of conclusions and statistical considerations
I had concerns about some of the statistical tests and conclusions that the authors made. I refer to some of these in other sections (e.g., the homotopy analyses), but I raise several here.
4. I am not sure what evidence the authors are using to make this claim: "we found that the cortical myelination and overall functional connectivity of ventral cortex developed significantly with the PMA but was not directly influenced by postnatal time." Postnatal time is significantly correlated with cortical myelination, as shown in Figures 2g, 2h, 3b, 3c, and postnatal time is significantly correlated with functional connectivity, as shown in Figures 4h, 5c, 5d, and 5e. Hence, this general claim that "the development of CT was considerably modulated by the postnatal experience while the CM was heavily influenced by prenatal duration" doesn't seem to be supported: both myelination and thickness are affected by postnatal experience and prenatal duration (as measured by PMA). A similar sentiment is expressed in the abstract. Perhaps the authors suggest different patterns in the strength of change for PMA vs. PT across these metrics, but if so, then statistical tests need to support that conclusion, and the claims need to reflect that sentiment.
Interestingly, Figure S4 presents a compelling ANOVA that does support this conclusion. Still, this result is relegated to the supplement, and it also uses GA, rather than PMA, making it hard to reconcile with the other claims made in the main text. Moreover, it uses ANOVAs, which dichotomizes a continuous variable. Here and elsewhere in the manuscript (e.g., Figures 3d, 3e), the authors split the infants into quartiles and compare them with ANOVAs. Their use for visualization is helpful, but it is unclear what the statistical motivation for this is rather than treating these as continuous variables like is possible with linear mixed-effects models. Moreover, it is unclear why the authors excluded half the data from the study (i.e., quartiles 2 and 3) in this ANOVA when all four quartiles could be used as factors.
5. It is unclear what the evidence is to support the following claim: "Both CT and CM show higher correlation with PMA in the posterior than anterior region, and higher correlation in the medial than lateral part within the anatomical mask (Figure 2a and Figure S2b-c [sic])" From Figure 2 or Figure S2, I don't see a gradient. From Figure S3, there might be a trend in some plots, but it is hard to interpret since it is non-monotonic. More generally, is there a statistical test to support this claim?
6. "and the interaction [sic] was more prominent in CM (simple effect: t = 10.98, p < 10-9) that in than CT (t = 2.07, p < 0.05)." Does 'more prominent' mean it is 'significantly stronger'? If not, then the authors should adjust this claim
7. Are the authors Fisher Z transforming their correlations? In numerous places, correlation values seem to be added together or used as the input to other correlation analyses. It is unclear from the methods whether the authors are transforming their correlation values to make that use appropriate.
Homotopy analyses
The homotopy section is a strength of the paper, but I have doubts about the approach taken to analyze this data and some of the conclusions drawn. I don't expect any of my suggestions to change the takeaway of this section, but I do think they are essential criticisms to address.
8. I do not think that the non-homotopic control condition is appropriate. In Arcaro & Livingstone (2017), the authors had 3 categories for this analysis: homotopic pairs (e.g., left V1 vs. right V1), adjacent pairs (e.g., left V1 vs. right V2), and distal pairs (e.g., left V1 vs. right PHA1). In the homotopy analysis performed by Li and colleagues, they compare homotopic pairs with all other pairs. I don't think that is generous to the test since non-homotopic pairs include adjacent pairs that should be similar and distal pairs that shouldn't be similar. This may explain why some non-homotopic distribution overlaps with the homotopic distribution in Figure 4c.
9. Regardless of this decision, I think the authors should reconsider their statistical test. I think the authors are using a between samples t-test to compare the 34 homotopic pairs with the hundreds of non-homotopic pairs. This is statistically inappropriate since the items are not independent (i.e., left V1 vs. right V1 is not independent of left V1 vs. right V2, which is also not independent of left V3 vs. right V2). This means the actual degrees of freedom are much lower than what is used. Moreover, I am unsure how the authors do this analysis across participants since this test can be done within participants. The authors should clarify what they did for this analysis and justify its appropriateness.
10. Could the authors speculate on why the correlations in homotopic regions are so much lower than what Arcaro and Livingstone (2017) found. I can think of a few possibilities: higher motion in infants, less rfMRI data per participant, different sleep/wake states, and different parcellation strategies. Regarding the last explanation, I think this is a real possibility: the bilateral correlation may be reduced if the Glasser atlas combines functionally heterogeneous patches of the cortex. Hence, the authors should consider this and other possible explanations.
11. The authors assume that the homotopic analyses mean that there are lateral connections between hemispheres (e.g., "Furthermore, the connections among the ventral visual cortex have developed during this early stage. Specifically, the homotopic connections between bilateral V1 and between bilateral VOTC both increased with GA, indicating an increased degree of functional distinction"). While this might be true, it doesn't need to be. Functional connectivity can be observed between regions that lack anatomical connectivity. Instead, two regions could both be driven by another region. In this case, the thalamus might drive symmetrical activity in the visual cortex.
Miscellaneous
12. I am not sure what the motivation of this line is: "Moreover, those studies did not fully control the visual experience in the first few weeks of the subjects, thus cannot give a clear conclusion whether the innate functional connectivity is unrelated to postnatal visual experience." Arcaro, Schade, Vincent, Ponce, & Livingstone (2017) did control the visual experience of subjects. Moreover, the research here doesn't control infant experience in the way this sentence implies: it implies an experiment manipulation (i.e., fully control) rather than a statistical control that is done here. Consider rephrasing
13. I am not sure why this claim is made: "Area V1 was selected because this region is the most basic region for visual processing and probably is the most experience-dependent area during early development". Is there evidence supporting this claim? Plasticity is found throughout the visual cortex, and I think which region is most plastic depends on the definition of plasticity. For instance, most people have the same tuning properties to gabor gratings (e.g., a cardinality bias), but there is enormous variability in face tuning across cultures.
14. The abstract says 783 infants were included in this study, but far fewer are actually used. The authors should report the 407 number in the abstract if any number at all.
15. Any comparisons of preterms and terms ought to be given the caveat that the preterm environment can be very different than the term environment: whereas a term infant goes home and sees friends and family without restriction, the preterm environment can be heavily regulated if they are in a NICU. Authors should either provide details about the environments of the preterms in their study, or they should consider how differences in the richness of visual experience - regardless of quantity - may affect visual development.
-
Reviewer #3 (Public Review):
The authors use a large neonatal dataset to examine how development may occur differently based on whether on not the neonate spent that time in gestation or out of the womb accruing potentially accruing visual experience. In this manner, the authors hope to tease apart those aspects of development that are biologically programmed versus those that occur in response to experience within the visual cortex. They show structurally that cortical thickness is affected by postnatal experience while cortical myelination is not, and functionally they find regional differentiation present between visual areas at birth and that their connectivity changes with development and postnatal experience. The conclusions seem well supported by the data and analyses and provide some insight into which aspects of brain structure …
Reviewer #3 (Public Review):
The authors use a large neonatal dataset to examine how development may occur differently based on whether on not the neonate spent that time in gestation or out of the womb accruing potentially accruing visual experience. In this manner, the authors hope to tease apart those aspects of development that are biologically programmed versus those that occur in response to experience within the visual cortex. They show structurally that cortical thickness is affected by postnatal experience while cortical myelination is not, and functionally they find regional differentiation present between visual areas at birth and that their connectivity changes with development and postnatal experience. The conclusions seem well supported by the data and analyses and provide some insight into which aspects of brain structure at birth are sculpted more by postnatal experience and which are more determined by endogenous developmental timelines.
The analyses are based on a large sample of infants, and the authors were careful to statistically separate which aspects of an infant's age, gestational or postnatal, are driving brain development, providing a deeper picture of infant brain development than previous publications. Overall, the findings seem well supported by the data as the analyses are relatively straightforward.
- Visualization of the data and findings could be improved, as a few figures are difficult to interpret without having to read the methods.
- The acronyms regarding gestation, postnatal, and post-menstrual time are a little distracting. Please consider explicitly writing "gestational time" etc when referring to these numbers to improve readability.
- Because the cortical ribbon of infants is so thin at birth, there seems to be a possibility that partial-volume effects could be more prevalent in less-developed infants and impact myelin metrics. If not modeled or estimated, it should at least be discussed.
- Structural and functional development could be more formally compared using quantitative models if the authors want those points more strongly related; the two are only qualitatively discussed at present. -
