Network segregation is associated with processing speed in the cognitively healthy oldest-old
Curation statements for this article:-
Curated by eLife
eLife assessment
This study provides empirical support for how brain function at the system level, particularly network segregation, influences cognitive abilities even in the oldest-old range of human aging. The findings are potentially interesting to understand successful aging.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
The brain is organized into systems and networks of interacting components. The functional connections among these components give insight into the brain’s organization and may underlie some cognitive effects of aging. Examining the relationship between individual differences in brain organization and cognitive function in older adults who have reached oldest-old ages with healthy cognition can help us understand how these networks support healthy cognitive aging. We investigated functional network segregation in 146 cognitively healthy participants aged 85+ in the McKnight Brain Aging Registry (MBAR). We found that the segregation of the association system and the individual networks within the association system (the fronto-parietal network , cingulo-opercular network, and default mode network), has strong associations with overall cognition and processing speed. We also provide a healthy oldest-old (85+) cortical parcellation that can be used in future work in this age group. This study shows that network segregation of the oldest-old brain is closely linked to cognitive performance. This work adds to the growing body of knowledge about differentiation in the aged brain by demonstrating that cognitive ability is associated with differentiated functional networks in very old individuals representing successful cognitive aging.
Article activity feed
-
-
-
Author Response
Reviewer #1 (Public Review):
In this study, Sims et al. evaluate how system-level brain functional connectivity is associated with cognitive abilities in a sample of older adults aged > 85 years old. Because the study sample of 146 normal older adults has lived into advanced years of age, the novelty here is the opportunity to validate brain-behavioral associations in aging with a reduced concern of the potential influence of undetected incipient neuropsychological pathology. The participants afforded resting-state functional magnetic resonance imaging (rs-fMRI) data as well as behavioral neuropsychological test assessments of various cognitive abilities. Exploratory factor analysis was applied on the behavioral cognitive assessments to arrive at summary measures of participant ability in five cognitive domains …
Author Response
Reviewer #1 (Public Review):
In this study, Sims et al. evaluate how system-level brain functional connectivity is associated with cognitive abilities in a sample of older adults aged > 85 years old. Because the study sample of 146 normal older adults has lived into advanced years of age, the novelty here is the opportunity to validate brain-behavioral associations in aging with a reduced concern of the potential influence of undetected incipient neuropsychological pathology. The participants afforded resting-state functional magnetic resonance imaging (rs-fMRI) data as well as behavioral neuropsychological test assessments of various cognitive abilities. Exploratory factor analysis was applied on the behavioral cognitive assessments to arrive at summary measures of participant ability in five cognitive domains including processing speed, executive functioning, episodic memory, working memory, and language. rsfMRI data were submitted to a graph-theoretic approach that derived underlying functional nodes in brain activity, the membership of these nodes in brain network systems, and indices characterizing the organizational properties of these brain networks. The study applies the classification of the various brain networks into a sensory/motor system of networks and an association system of network, with further sub-systems in the latter that includes the frontoparietal network (FPN), the default-mode network (DMN), the cingulo-opercular network (CON), and the dorsal (DA) and ventral (VA) attention networks. Amongst other graph metrics, the study focused on the extent to which networks in these brain systems were segregated (i.e., separable network communities as opposed to a more singular large community network). Evaluation of the brain network segregation indices and cognitive performance metrics showed that in general higher network functional segregation corresponds with higher cognitive performance ability. In particular, this association was seen between the general association system with overall cognition, and the FPN with overall cognition, and processing speed.
The results worthy of highlighting include the documentation of oldest-old individuals with detectable brain neural network segregation at the level of the association system and its FPN sub-system and the association of this brain functional state notably with general cognition and processing speed and less so with the other specific cognitive domains (such as memory). This finding suggests that (a) apparently better cognitive aging might stem from a specific level of neural network functional segregation, and (b) this linkage applies more specifically to the FPN and processing speed. These specific findings inform the broader conceptual perspective of how human brain aging that is normative vs. that which is pathological might be distinguished.
We appreciate this comment and we have added these points to the conclusion more explicitly.
To show the above result, this study defined functional networks that were driven more by the sample data as opposed to a pre-existing generic template. This approach involves a watershed algorithm to obtain functional connectivity boundary maps in which the boundary brain image voxels separate functionally related voxels from unrelated voxels by virtue of their functional covariance as measured in the immediate data. This is also a notable objective and data-driven approach towards defining functional brain regions-of-interest (ROIs), nodes, and networks that are age-appropriate and configured for a given dataset as opposed to using network definitions based on other datasets used as a generic template.
The sample size of 146 for this age group is generally sufficient.
For the analyses considering the significance of the effect of the brain network metrics on the cognitive variables, the usage of heirarchical regression to evaluate whether the additional variables (in the full model) significantly change the model fit relative to the reduced model with covariates-only (data collection site, cortical thickness), while a possible approach, might be problematic, particularly when the full model uses many more regressors than the reduced model. In general, adding more variables to regression models reduces the residual variance. As such, it is possible that adding more regressors in a full model and comparing that to a reduced model with much fewer regressors would yield significant changes in the R^2 fit index, even if the added regressors are not meaningfully modulating the dependent variable. This may not be an issue for the finding on the FPN segregation effect on overall cognition, but it may be important in interpreting the finding on the association system metrics on overall cognition.
Critically, we should note that the correlation effect sizes (justified by the 0.23 value based on the reported power analyses) were all rather small in size. The largest key brain-behavior correlation effect was 0.273 (between DMN segregation and Processing Speed). In the broader perspective, such effects sizes generally suggest that the contribution of this factor is minimal and one should be careful that the results should be understood in this context.
The recent, highly publicized paper from Marek and colleagues (citation below) offers some support for the assertion that these effect sizes are on the order that would be expected for ‘true’ signals in the brain. While the study reported here is not a “BWAS” as described in the Marek article (BWAS is a brain-wide association study, examining, without a priori hypotheses of brain network, all possible associations), and therefore our study does not fall prey to some of the multiple comparisons issues described in that paper, the general expected effect sizes based on that paper should be relevant here.
Marek and colleagues suggest that 1) effect sizes in the range of 0.273 are on the order of the larger brain-behavior relationships that can be expected to be replicable, and 2) samples that remove some drivers of individual variability are beneficial to the capability of a study to identify an effect. Relevant to the latter point, by reducing the variability in our sample due to age (our age range is tight) and early signs of neurological disease (these were screened out in our sample), this leaves a sample that is homogeneous along these variables, meaning that brain variability associated with cognitive performance can be more easily pulled from the data.
Our data have large variability on the behavior end, and large variability on the brain end, allowing better power for seeing effects between them.
Marek, S., Tervo-Clemmens, B., Calabro, F.J., Montez, D.F., Kay, B.P., Hatoum, A.S., Donohue, M.R., Foran, W., Miller, R.L., Hendrickson, T.J., et al. (2022). Reproducible brain-wide association studies require thousands of individuals. Nature 603, 654–660. 10.1038/s41586-022-04492-9.
Overall, the findings based on hierarchical regressions that evaluate the network segregation indices in accounting for cognition and the small correlation magnitudes are basically in line with the notion that more segregated neural networks in the oldest-old support better cognitive performance (particularly processing speed). However, the level of positive support for the notion based on these findings is somewhat moderate and requires further study.
The addition of a control analysis (sensorimotor network) in the newer version of the paper showed that these effects are not present in brain networks not thought to relate to cognition. We agree that further study of these questions is necessary for stronger claims to be made, but the current study advances the field by showing clearly that segregation of the association network and its components relates to behavior even in this oldest old cohort.
Reviewer #2 (Public Review):
The authors capitalised on the opportunity to obtain functional brain imaging data and cognitive performance from a group of oldest old with normative cognitive ability and no severe neurophysiological disorders, arguing that these individuals would be most qualified as having accomplished 'healthy ageing'. Combined with the derivation of a cohort-specific brain parcellation atlas, the authors demonstrated the importance of maintaining brain network segregation for normative cognition ability, especially processing speed, even at such late stage of life. In particular, segregation of the frontoparietal network (FPN) was found to be the key network property.
These results bolstered the findings from studies using younger old participants and are in agreement with the current understanding of the connectomme-cognition relationship. The inclusion of a modest sample size, power analysis, cohort-specific atlas, and a pretty comprehension neuropsychological assessment battery provides optimism that the observed importance of FPN segregation would be a robust and generalisable finding at least in future cross-sectional studies. The fact that FPN segregation is relatively more important to cognition than other associative networks also provides novel insight about the possible 'hierarchy' between age-related neural and cognitive changes, regardless of what mechanisms lead to such segregation at such an advanced age. it is also interesting that processing speed remains to be the 'hallmark' metric of age-related cognitive changes, indirectly speaking to its long assumption fundamental impact on overall cognition.
As laid out by the authors, if network differentiation is key to normative cognitive ability at old age, intervention and stimulation programs that could maintain or boost network segregation would have high translational value. With advent in mobile self-administrable devices that target behavioural and neural modifications, this potential would have increasing appeal.
However, I feel that a few things have prevented the manuscript to be a simple yet impactful submission
- Interpretation and the major theme of discussion. While the authors attempted to discuss their findings with respect to both the compensatory and network dedifferentiation hypotheses, the results and their interpretation do not readily provide any resolution or reconciliation between the two, a common challenge in many ageing research. The authors did not further elaborate how the special cohort they had may provide further insights to this.
While the results certainly are in line with the dedifferentiation hypothesis, why 'this finding does not exclude the compensation hypothesis' (Discussion) was not elaborated enough. In particular, the authors seemed to suggest that maintained network specialisation may be in such a role, but the results and interpretations regarding network specialisation were not particularly focused on throughout the manuscript. In addition, both up regulation within a network and cross-network recruitment can both be potential compensatory strategies (Cabeza et al 2018, Rev Nat Neurosci). Without longitudinal data or other designs (e.g. task) it is quite difficult to evaluate the involvement of compensation. For instance, as rightly suggested by the authors, the two phenomena may not be mutually exclusive (e.g., maintenance of the FPN differentiation at such old age could be a result of 'compensation' that started when the participants were younger).
The reviewer makes some excellent points that we have taken to heart in this revision. We agree that the data as described do not directly address the compensation hypothesis, and therefore de-emphasized our descriptions of that hypothesis in service of a simpler, more impactful manuscript.
As described above in our response to the essential revisions “In the original submission, we noted relevant literature which describe both the dedifferentiation hypothesis and the compensation hypothesis of aging. Our original aim was to include more of a literature review of cognitive aging theories in the introduction and discussion, but that choice made it too confusing (and honestly left out much important literature). In responding to the reviews we realize that bypassing this cursory literature review here is preferable for the readability of the manuscript. Instead, we cite a literature review, and focus on the dedifferentiation hypothesis.
“The data we show here addresses the dedifferentiation hypothesis specifically since we are using the segregation metric- a reflection of dedifferentiation of network organization. The reviewers’ comments caused us to do a great deal of thinking on this topic, and we have a forthcoming review with our colleague Ian McDonough that covers this topic in more detail (McDonough, Nolin, and Visscher, 2022). We have substantially rewritten the relevant sections in the discussion (especially section 3.2) to be more clear for readers.”
As also described in our response to essential revisions 2c, we have added to the discussion regarding the utility of studying the oldest-old; this is in the second paragraph of the discussion, and reproduced above. Additionally, in the Introduction, we also briefly address the importance of this cohort. We state “Prior work has mostly been done in younger-old samples (largely 65-85 years old). Studying the younger-old can be confounded by including pre-symptomatic disease, since it is unknown which individuals may be experiencing undetectable, pre-clinical cognitive disorders and which will continue to be cognitively healthy for another decade. The cognitively unimpaired oldest-old have lived into late ages, and we can be more confident in determining their status as successful agers. A further benefit of studying these successful cognitive agers is that because of their advanced age and the normal aging and plasticity processes associated with it, there is greater variance in both their performance on neurocognitive tasks, and in brain connectivity measures than there is in younger cohorts (Christensen et al., 1994). This increased variance makes it easier to observe across-subject relationships of cognition and brain networks (Gratton, Nelson, & Gordon, 2022). We provide new insight into the relationship between the segregation of networks and cognition by investigating this relationship in an oldest old cohort of healthy individuals.”
- Some further clarity about the data and statistical analyses would be desirable. First, since scan length determines the stability of functional connectivity, how long was the resting-state scan? Second, what is the purpose of using both hierarchical regression and partial correlation? While they do consider different variances in the dataset, they are quite similar and the decision looks quite redundant to me as not much further insights have been gained. [the main insight to including a regression is to be able to compare the different networks to each other.]
The resting-state fMRI scan is 8 minutes in length. This has been added to the text. After considering the redundancy the reviewer notes between hierarchical regression and correlations, we have simplified our statistical approach and only included correlations in the main body of the manuscript. We have put the regressions in the supplemental materials so if interested readers would like to be able to see those results, they are still available.
-
eLife assessment
This study provides empirical support for how brain function at the system level, particularly network segregation, influences cognitive abilities even in the oldest-old range of human aging. The findings are potentially interesting to understand successful aging.
-
Reviewer #1 (Public Review):
In this study, Sims et al. evaluate how system-level brain functional connectivity is associated with cognitive abilities in a sample of older adults aged > 85 years old. Because the study sample of 146 normal older adults has lived into advanced years of age, the novelty here is the opportunity to validate brain-behavioral associations in aging with a reduced concern of the potential influence of undetected incipient neuropsychological pathology. The participants afforded resting-state functional magnetic resonance imaging (rs-fMRI) data as well as behavioral neuropsychological test assessments of various cognitive abilities. Exploratory factor analysis was applied on the behavioral cognitive assessments to arrive at summary measures of participant ability in five cognitive domains including processing …
Reviewer #1 (Public Review):
In this study, Sims et al. evaluate how system-level brain functional connectivity is associated with cognitive abilities in a sample of older adults aged > 85 years old. Because the study sample of 146 normal older adults has lived into advanced years of age, the novelty here is the opportunity to validate brain-behavioral associations in aging with a reduced concern of the potential influence of undetected incipient neuropsychological pathology. The participants afforded resting-state functional magnetic resonance imaging (rs-fMRI) data as well as behavioral neuropsychological test assessments of various cognitive abilities. Exploratory factor analysis was applied on the behavioral cognitive assessments to arrive at summary measures of participant ability in five cognitive domains including processing speed, executive functioning, episodic memory, working memory, and language. rsfMRI data were submitted to a graph-theoretic approach that derived underlying functional nodes in brain activity, the membership of these nodes in brain network systems, and indices characterizing the organizational properties of these brain networks. The study applies the classification of the various brain networks into a sensory/motor system of networks and an association system of network, with further sub-systems in the latter that includes the frontoparietal network (FPN), the default-mode network (DMN), the cingulo-opercular network (CON), and the dorsal (DA) and ventral (VA) attention networks. Amongst other graph metrics, the study focused on the extent to which networks in these brain systems were segregated (i.e., separable network communities as opposed to a more singular large community network). Evaluation of the brain network segregation indices and cognitive performance metrics showed that in general higher network functional segregation corresponds with higher cognitive performance ability. In particular, this association was seen between the general association system with overall cognition, and the FPN with overall cognition, and processing speed.
The results worthy of highlighting include the documentation of oldest-old individuals with detectable brain neural network segregration at the level of the association system and its FPN sub-system and the association of this brain functional state notably with general cognition and processing speed and less so with the other specific cognitive domains (such as memory). This finding suggests that (a) apparently better cognitive aging might stem from a specific level of neural network functional segregation, and (b) this linkage applies more specifically to the FPN and processing speed. These specific findings inform the broader conceptual perspective of how human brain aging that is normative vs. that which is pathological might be distinguished.
To show the above result, this study defined functional networks that were driven more by the sample data as opposed to a pre-existing generic template. This approach involves a watershed algorithm to obtain functional connectivity boundary maps in which the boundary brain image voxels separate functionally related voxels from unrelated voxels by virtue of their functional covariance as measured in the immediate data. This is also a notable objective and data-driven approach towards defining functional brain regions-of-interest (ROIs), nodes, and networks that are age-appropriate and configured for a given dataset as opposed to using network definitions based on other datasets used as a generic template.
The sample size of 146 for this age group is generally sufficient.
For the analyses considering the significance of the effect of the brain network metrics on the cognitive variables, the usage of heirarchical regression to evaluate whether the additional variables (in the full model) significantly change the model fit relative to the reduced model with covariates-only (data collection site, cortical thickness), while a possible approach, might be problematic, particularly when the full model uses many more regressors than the reduced model. In general, adding more variables to regression models reduces the residual variance. As such, it is possible that adding more regressors in a full model and comparing that to a reduced model with much fewer regressors would yield significant changes in the R^2 fit index, even if the added regressors are not meaningfully modulating the dependent variable. This may not be an issue for the finding on the FPN segregation effect on overall cognition, but it may be important in interpreting the finding on the association system metrics on overall cognition.
Critically, we should note that the correlation effect sizes (justified by the 0.23 value based on the reported power analyses) were all rather small in size. The largest key brain-behavior correlation effect was 0.273 (between DMN segregation and Processing Speed). In the broader perspective, such effects sizes generally suggest that the contribution of this factor is minimal and one should be careful that the results should be understood in this context.
Overall, the findings based on hierarchical regressions that evaluate the network segregation indices in accounting for cognition and the small correlation magnitudes are basically in line with the notion that more segregated neural networks in the oldest-old support better cognitive performance (particularly processing speed). However, the level of positive support for the notion based on these findings is somewhat moderate and requires further study.
-
Reviewer #2 (Public Review):
The authors capitalised on the opportunity to obtain functional brain imaging data and cognitive performance from a group of oldest old with normative cognitive ability and no severe neurophysiological disorders, arguing that these individuals would be most qualified as having accomplished 'healthy ageing'. Combined with the derivation of a cohort-specific brain parcellation atlas, the authors demonstrated the importance of maintaining brain network segregation for normative cognition ability, especially processing speed, even at such late stage of life. In particular, segregation of the frontoparietal network (FPN) was found to be the key network property.
These results bolstered the findings from studies using younger old participants and are in agreement with the current understanding of the …
Reviewer #2 (Public Review):
The authors capitalised on the opportunity to obtain functional brain imaging data and cognitive performance from a group of oldest old with normative cognitive ability and no severe neurophysiological disorders, arguing that these individuals would be most qualified as having accomplished 'healthy ageing'. Combined with the derivation of a cohort-specific brain parcellation atlas, the authors demonstrated the importance of maintaining brain network segregation for normative cognition ability, especially processing speed, even at such late stage of life. In particular, segregation of the frontoparietal network (FPN) was found to be the key network property.
These results bolstered the findings from studies using younger old participants and are in agreement with the current understanding of the connectomme-cognition relationship. The inclusion of a modest sample size, power analysis, cohort-specific atlas, and a pretty comprehension neuropsychological assessment battery provides optimism that the observed importance of FPN segregation would be a robust and generalisable finding at least in future cross-sectional studies. The fact that FPN segregation is relatively more important to cognition than other associative networks also provides novel insight about the possible 'hierarchy' between age-related neural and cognitive changes, regardless of what mechanisms lead to such segregation at such an advanced age. it is also interesting that processing speed remains to be the 'hallmark' metric of age-related cognitive changes, indirectly speaking to its long assumption fundamental impact on overall cognition.
As laid out by the authors, if network differentiation is key to normative cognitive ability at old age, intervention and stimulation programs that could maintain or boost network segregation would have high translational value. With advent in mobile self-administrable devices that target behavioural and neural modifications, this potential would have increasing appeal.
However, I feel that a few things have prevented the manuscript to be a simple yet impactful submission
- Interpretation and the major theme of discussion. While the authors attempted to discuss their findings with respect to both the compensatory and network dedifferentiation hypotheses, the results and their interpretation do not readily provide any resolution or reconciliation between the two, a common challenge in many ageing research. The authors did not further elaborate how the special cohort they had may provide further insights to this.
While the results certainly are in line with the dedifferentiation hypothesis, why 'this finding does not exclude the compensation hypothesis' (Discussion) was not elaborated enough. In particular, the authors seemed to suggest that maintained network specialisation may be in such a role, but the results and interpretations regarding network specialisation were not particularly focused on throughout the manuscript. In addition, both up regulation within a network and cross-network recruitment can both be potential compensatory strategies (Cabeza et al 2018, Rev Nat Neurosci). Without longitudinal data or other designs (e.g. task) it is quite difficult to evaluate the involvement of compensation. For instance, as rightly suggested by the authors, the two phenomena may not be mutually exclusive (e.g., maintenance of the FPN differentiation at such old age could be a result of 'compensation' that started when the participants were younger).
- Some further clarity about the data and statistical analyses would be desirable. First, since scan length determines the stability of functional connectivity, how long was the resting-state scan? Second, what is the purpose of using both hierarchical regression and partial correlation? While they do consider different variances in the dataset, they are quite similar and the decision looks quite redundant to me as not much further insights have been gained.
-
-
-