Dissecting the chain of information processing and its interplay with neurochemicals and fluid intelligence across development
Curation statements for this article:-
Curated by eLife
eLife assessment
This important study combines behavioral and imaging experiments to understand how levels of important brain chemicals shape the processing of information in the brain in children and young adults. The sample size and data quality are outstanding and some of the data are quite convincing. However, the calculation and interpretation of the brain chemical concentration measurements as well as the interpretation of the model-based behavioral parameters are not fully justified and support for the overall conclusions is incomplete. This work will be of interest to neuroscientists, psychologists, and neuroimaging researchers investigating the developing brain in health and disease.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Previous research has highlighted the role of glutamate and gamma-aminobutyric acid (GABA) in perceptual, cognitive, and motor tasks. However, the exact involvement of these neurochemical mechanisms in the chain of information processing, and across human development, is unclear. In a cross-sectional longitudinal design, we used a computational approach to dissociate cognitive, decision, and visuomotor processing in 293 individuals spanning early childhood to adulthood. We found that glutamate and GABA within the intraparietal sulcus (IPS) explained unique variance in visuomotor processing, with higher glutamate predicting poorer visuomotor processing in younger participants but better visuomotor processing in mature participants, while GABA showed the opposite pattern. These findings, which were neurochemically, neuroanatomically and functionally specific, were replicated ~21 mo later and were generalized in two further different behavioral tasks. Using resting functional MRI, we revealed that the relationship between IPS neurochemicals and visuomotor processing is mediated by functional connectivity in the visuomotor network. We then extended our findings to high-level cognitive behavior by predicting fluid intelligence performance. We present evidence that fluid intelligence performance is explained by IPS GABA and glutamate and is mediated by visuomotor processing. However, this evidence was obtained using an uncorrected alpha and needs to be replicated in future studies. These results provide an integrative biological and psychological mechanistic explanation that links cognitive processes and neurotransmitters across human development and establishes their potential involvement in intelligent behavior.
Article activity feed
-
-
eLife assessment
This important study combines behavioral and imaging experiments to understand how levels of important brain chemicals shape the processing of information in the brain in children and young adults. The sample size and data quality are outstanding and some of the data are quite convincing. However, the calculation and interpretation of the brain chemical concentration measurements as well as the interpretation of the model-based behavioral parameters are not fully justified and support for the overall conclusions is incomplete. This work will be of interest to neuroscientists, psychologists, and neuroimaging researchers investigating the developing brain in health and disease.
-
Reviewer #1 (Public Review):
Zacharopoulos et al. present a multi-modal investigation into the developmental trajectories of cognitive processing, decision-making processing, and visuomotor processing in children and young adults, and attempt to relate them to neuroimaging measures of functional brain connectivity and neurotransmitter concentrations in two distinct brain regions.
Results suggest specific interactions between neurotransmitter concentrations and visuomotor task performance. Interestingly, GABA and Glu levels appear to have different relationships with task performance if the participant group is trichotomized into older, 'mean-age', and younger participants. These findings appear consistent across three different visuomotor processing tasks and replicate well between two time points at which task performance and MRS …
Reviewer #1 (Public Review):
Zacharopoulos et al. present a multi-modal investigation into the developmental trajectories of cognitive processing, decision-making processing, and visuomotor processing in children and young adults, and attempt to relate them to neuroimaging measures of functional brain connectivity and neurotransmitter concentrations in two distinct brain regions.
Results suggest specific interactions between neurotransmitter concentrations and visuomotor task performance. Interestingly, GABA and Glu levels appear to have different relationships with task performance if the participant group is trichotomized into older, 'mean-age', and younger participants. These findings appear consistent across three different visuomotor processing tasks and replicate well between two time points at which task performance and MRS measures were established for each participant (1.5 years apart). Visuomotor connectivity (assessed with resting-state-fMRI) also showed age-group-specific relationships with neurotransmitter levels. Finally, the authors present evidence that visuomotor processing mediates the relationship between neurochemical levels and scores of fluid intelligence, but only for older participants.
STRENGTHS
The study has an astonishing sample size in the context of MRS research, a field that has historically struggled to aggregate large datasets because of a severe lack of methodological standardization. Longitudinal MRS data from close to 300 participants means that this is one of the largest MRS datasets to date, enabling the group to add another exciting piece of work to their six previously published manuscripts on relationships between cognitive performance and neurochemical measures from this powerful resource. MRS data quality appears excellent, owed to state-of-the-art acquisition and raw data processing. The authors are further to be commended for making the raw MRS data publicly available - they will serve as a fantastic resource for method developers and applied researchers in the field.
WEAKNESSES
There is generally little to no consideration or discussion concerning age trajectories of MRS-derived metabolite estimates during childhood and early adulthood, which are not clearly established at all. There is evidence for increasing GABA+macromolecules during childhood (Porges et al, eLife 2021, https://elifesciences.org/articles/62575), although it may be ascribed to macromolecules rather than GABA itself (Bell et al, Sci Rep 2021, https://pubmed.ncbi.nlm.nih.gov/33436899/). The findings should at least be discussed in the context of this literature, but I suggest going a step further. The authors have all the data to make a major contribution to the scarce body of evidence on metabolite changes between 6 and 18 years by examining whether GABA and Glu estimates actually appear to change systematically across the age range of their dataset (especially exciting since they have longitudinal data)! It would be immensely valuable to see an analysis like this.
With that said, a methodological weakness concerns the computation of neurochemical concentrations presented here. Firstly, the authors can provide more detail about the acquisition and data processing/modeling decisions. Secondly, and more importantly, MRS-derived estimates of concentration can never be absolute, and always require several assumptions about the relative contributions of tissue classes (GM, WM, CSF) to the measurement volume, tissue water content, water and metabolite MR relaxation times, MR visibility, etc. Quantitative MRS estimates therefore need to be interpreted with caution, especially when these confounding factors are likely to vary between observed groups, or with age, pathology, etc. - there is plenty of reason to assume that cortical maturation, iron accumulation, etc. contribute to changes in relative GM/WM/CSF fractions or relaxation time changes. The authors present two different correction methods to account for some of these aspects, but only present the results of one, stating that "The results showed the same general pattern across all quantification methods.", which is insufficient to assess what changed and what didn't. Interestingly, the authors have presented no less than *four* different quantification methods in a similar manuscript using the same dataset (Zacharopoulos et al, Human Brain Mapp 2021; https://onlinelibrary.wiley.com/doi/10.1002/hbm.25396), but they do not mention normalization to the internal creatine signal in this present work, or whether it yielded different results (which might indicate that their method of tissue correction introduces a confounder rather than correcting for it). There is no mention of whether any further analysis of the water T2 relaxation time estimates was performed, but it would be vital to understand whether they themselves change with age, since this would establish that they are likely to confound GABA and Glu estimation. Generally, the choice to perform additional subject-specific acquisitions to allow corrections for water T2 relaxation is understandable, but not clearly motivated or explained in the experimental section. The authors should further clarify whether the relative tissue volume fractions of GM, WM, and CSF are stable across the age range, or whether there is a systematic tissue composition change with age that may also confound the Glu and GABA estimation.
Finally, I am surprised to find no discussion of limitations at all. It is important to point out the methodological limitations of MRS, which are widely discussed in the MRS literature, but probably less obvious to those readers less intimately familiar with it. This concerns not only the confounding factors for quantification that I described above but also the challenges of the comparably low spectral resolution at 3 Tesla. Even with high-quality data as presented here, it remains unclear whether the small GABA signal can be reliably separated from glutamate, glutamine, and glutathione, all of which exhibit substantial spectral overlap with each other and other strong signals as well as the underlying macromolecular background. The limitations (and how they impact interpretation) ought to be mentioned and discussed in the context of the vast amount of literature. They should provide the reader with the appropriate context and the awareness that all MRS measures are extremely sensitive to many different experimental factors and modeling decisions.
-
Reviewer #2 (Public Review):
Zacharopoulos et al. investigated the relationship between MR spectroscopy-detected neurotransmitter concentrations (GABA/glutamate) in the intra-parietal sulcus (IPS) and middle frontal gyrus (MFG), behaviourally measured indices of sensory and cognitive processing, and fMRI measured functional connectivity within the frontoparietal network. They find that increased IPS glutamate concentration is related to poorer visuomotor processing in younger participants and better performance in older participants, while IPS GABA predicts the opposite pattern. They further show that these relationships are mediated by frontoparietal functional connectivity. Finally, they show that IPS GABA and glutamate concentration are related to fluid intelligence and that this relationship is mediated by visuomotor processing and …
Reviewer #2 (Public Review):
Zacharopoulos et al. investigated the relationship between MR spectroscopy-detected neurotransmitter concentrations (GABA/glutamate) in the intra-parietal sulcus (IPS) and middle frontal gyrus (MFG), behaviourally measured indices of sensory and cognitive processing, and fMRI measured functional connectivity within the frontoparietal network. They find that increased IPS glutamate concentration is related to poorer visuomotor processing in younger participants and better performance in older participants, while IPS GABA predicts the opposite pattern. They further show that these relationships are mediated by frontoparietal functional connectivity. Finally, they show that IPS GABA and glutamate concentration are related to fluid intelligence and that this relationship is mediated by visuomotor processing and moderated by the developmental stage. These data add to our understanding of the dynamic role of excitatory and inhibitory neurotransmitter systems in cognitive processes throughout development.
Strengths:
The study employs an impressively large cross-sectional, multimodal, dataset, with almost 300 participants ranging from 6 to 18+ years old.
The main finding (i.e., the interaction between GABA/Glu, visuomotor processing, and age) is found across three behavioural tasks and replicated in a second dataset collected 1.5 years after the first.
The authors extensively report the results of the numerous analyses performed in the supplementary material.
Weaknesses:
Pre-registration of experimental and analytical plans should be the norm, e.g., to reduce so-called 'p-hacking'. I am by no means asserting that this behaviour has occurred in the current study; however, it is disappointing that there is no reported pre-registration for such a large-scale study, where the selection and order of analyses (and the subsequent corrections applied) can meaningfully influence the pattern of results.
Many tests were performed in the study using frequentist statistics, and the way the results are reported makes it difficult to discern how distinguishable those that were reported as meaningful are from those that were disregarded.
Insufficient analyses were conducted to describe the relationships between a) GABA and glutamate, b) repeated behavioural measures, and c) test-retest reliability. This reduces the strength of the claims, some of which could be accounted for by simpler, potentially less interesting, explanations.
-
Reviewer #3 (Public Review):
This study sought to identify relations between parameters of the diffusion decision model (DDM) and concentration of the neurotransmitters glutamate and GABA, as measured by magnetic resonance spectroscopy, and to evaluate the possibility that age moderates these relations in a developmental sample spanning middle childhood through young adulthood. The authors find a set of age-by-neurotransmitter concentration interaction effects indicating that lower levels of glutamate and greater levels of GABA in the intraparietal sulcus are related to faster non-decision times (lower values of the Ter parameter of the DDM) for "younger" participants but have the opposite relations in "older" participants (although given the way that the results are reported, the reader has little indication of what age group the terms …
Reviewer #3 (Public Review):
This study sought to identify relations between parameters of the diffusion decision model (DDM) and concentration of the neurotransmitters glutamate and GABA, as measured by magnetic resonance spectroscopy, and to evaluate the possibility that age moderates these relations in a developmental sample spanning middle childhood through young adulthood. The authors find a set of age-by-neurotransmitter concentration interaction effects indicating that lower levels of glutamate and greater levels of GABA in the intraparietal sulcus are related to faster non-decision times (lower values of the Ter parameter of the DDM) for "younger" participants but have the opposite relations in "older" participants (although given the way that the results are reported, the reader has little indication of what age group the terms "younger" and "older" refer to). The authors find similar interaction effects regarding relations between neurotransmitter concentration and connectivity in a visuomotor network and between neurotransmitter concentration and a fluid intelligence test. They then test moderated mediation models to determine whether functional connectivity in the visual-motor network mediates relations between neurotransmitter concentration and Ter, and whether Ter mediates the relation between neurotransmitter concentration and intelligence.
Strengths of the study include the relatively large sample size and the unique combination of brain and behavioral measures. The reported bivariate associations indicate an intriguingly consistent pattern of age-related moderation effects on the relation between neurotransmitter concentrations and several variables relevant to cognition (Ter, visuomotor connectivity, intelligence test scores) that could provide valuable insights to the field about the interplay between neurotransmitters and cognitive processes across development. However, the inferences that can be drawn from this work are seriously limited by an array of conceptual and methodological concerns.
A major conceptual issue is that the study is motivated by the premise that the nondecision time parameter (Ter) of the DDM is a major mechanistic underpinning of intelligence during child and adolescent development. There are several reasons why this premise is not well-supported. Although Ter is sometimes found to have weak correlations with scores on intelligence tests, the clearest pattern of findings across multiple studies is instead that individual differences in intelligence are primarily related to the DDM's drift rate (v) parameter (e.g., Schmiedek et al., 2007; Schulz-Zhecheva et al., 2016; Schubert & Frischkorn, 2020). The authors highlight the earlier finding that Ter mediates the effect of age on intelligence in the Krause et al. (2020) paper, but this paper is of questionable relevance to the current study. Krause et al. (2020) investigated cognitive changes in aging (ages 18-62) which are quite different from the current study's focus on development from middle childhood through young adulthood (ages ~ 7-24). Aging in older adults is known to have limited and task-specific effects on drift rate but strong effects on boundary and Ter whereas development from childhood through young adulthood coincides with the rapid maturation of drift rate (as shown in both prior research and in the current study's supplemental plots). Beyond the relatively weak evidence that Ter is a major contributor to intelligence during development, it is important to note that Ter is a nonspecific "residual" parameter (Schubert & Frischkorn, 2020) that is, theoretically, the summation of a wide array of different processes that are difficult to dissociate (perceptual encoding, visual search, motor responding). Therefore, in contrast to the drift rate and boundary parameters, it is difficult to interpret Ter as indexing a unitary mechanistic process, which is consistent with earlier findings that Ter shows limited evidence of psychometric validity as a task-general trait (Schubert et al., 2016). Finally, it is notable that bivariate tests of DDM parameters' relations with intelligence in the current study's sample (Supplemental File 9) suggest that Ter does not show a robust relation with intelligence, whereas drift rate shows relatively strong relations with intelligence in every group except for "older" individuals, who have likely fully matured and may therefore have less variance in both v and intelligence.
There are several opaque and potentially problematic features of the EZ DDM analysis. The tasks have relatively few trials spread across multiple different conditions within each task and it is unclear whether the DDM parameters were estimated separately in each condition or were estimated from trial-level data that were collapsed across conditions. This is especially concerning for the ANT, which has 96 trials distributed across 12 conditions, or apparently only 8 trials per design cell. Given the relatively low number of trials (both per design cell and overall) it is also concerning that parameter recovery studies do not appear to have been completed to ensure that this number of trials is sufficient to reliably estimate DDM parameters. In addition, accuracy rates and other behavioral summary statistics are not reported for any of the tasks. As ceiling levels of accuracy (i.e., few error RTs) can also cause prevent accurate estimation of parameters, this is another indication that assessing parameter recovery could be critical for inferences in this study.
A broader concern related to the measurement of DDM parameters is that they are each assumed to reflect the same mechanistic process across the three different tasks, but this assumption is not explicitly modeled (e.g., as a latent factor). Although the fact that Ter parameters across all three tasks have similar patterns of results is consistent with this assumption, constructing a latent factor using EFA or CFA would provide an explicit, and critical, test of the assumption. Latent factors formed from DDM parameters across the three tasks would also have several key methodological advantages over single-task measures, including separating variance in the cross-task mechanism of interest from task-specific "method variance", increasing statistical power by improving measurement of the latent mechanism, and reducing the number of multiple comparisons that need to be corrected for. The last two points are particularly critical for this study because it is possible that poor (i.e., single task) measurement and the large number of comparisons that were corrected for may have resulted in a consequential Type II error, such as a failure to detect effects involving DDM parameters other than Ter (drift rate and boundary). Related to these points, it is generally difficult for readers to judge the study's claims about the cross-task relevance of each DDM parameter or about dissociations between the different parameters because no intercorrelations between the parameters (within and between tasks) are reported or discussed.
Although the setup of the bivariate and moderation tests of relations between neurotransmitter concentrations and other variables is generally rigorous, it is concerning that only linear effects of age appear to have been considered. There appears to be clear evidence of nonlinear age-related trends in the scatterplots of parameter values displayed in Supplementary File 2.
The mediation analyses are central to the study's claims, but their results are particularly difficult to draw conclusions from due to several problematic methodological details. First, the confidence interval (CI) used to evaluate the significance of effects in these analyses was a 90% CI, essentially changing the alpha level for these tests from the conventional p<.05 to p<.10. Although this is claimed, in the Methods section, to be justified because these were "follow-up analyses based on the significant results obtained in advance", it is a highly unusual change that appears likely to have been made post hoc. Another apparent post hoc change is also mentioned in Methods: "we additionally removed cases that fell beyond three standard deviations after running the multiple regression models". It is not clear what variable or residual score this statement refers to and it is also not clear why this outlier trimming step was carried out only at the mediation stage, and not prior to the earlier bivariate analyses. The combination of an unusually high effective alpha level and potential post hoc adjustments to researcher degrees of freedom seriously undermines confidence in the mediation results. Even if the statistical hypothesis tasks are taken at face value, though, the evidence for mediation still appears very weak because the standardized effects are small and only significant in one age group (either "older" or "younger" but not both).
A broader challenge for readers is that neither the absolute ages nor the general developmental period of participants is mentioned anywhere in the main text or main plots of the paper. "Younger" and "older" mean very different things when referring to an aging sample (e.g., in the cited Klaus et al. study) versus a developing sample that spans middle childhood through young adulthood. Even if the range of the current study's sample is known, plots that split groups up by plus or minus one SD of age obscure age-related trends because the ages of the subgroups are not known. It may be easier to interpret findings if groups are instead split by neurotransmitter levels and plotted by absolute age.
-
