Article activity feed

  1. Evaluation Summary:

    This manuscript will be of interest to cognitive and developmental neuroscientists who are interested in brain oscillations and their changes with development. This study decomposes the EEG alpha power, demonstrating the confound of aperiodic activity in true oscillatory power and elucidating opposing relation of periodic and aperiodic components with age. The main approach of this paper is well motivated, and the main conclusions are supported by the analysis, which is applied to multiple large datasets, though there are some minor issues with some of the follow up analyses. Overall, this manuscript makes a timely and important case for the consideration of aperiodic signals in future research.

    (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 and Reviewer #3 agreed to share their names with the authors.)

    Was this evaluation helpful?
  2. Reviewer #1 (Public Review):

    In this study a large dataset of EEG recordings from children, adolescents and young adults between 6 and 22 years was analysed regarding developmental changes of alpha amplitude. Previous literature on this issue is incoherent, sometimes suggesting age-related decrease sometimes increase of alpha activity. The authors, therefore, separated periodic and aperiodic activity, suspecting that age-related decrease of alpha activity might be spuriously driven by aperiodic signals. And indeed this is what they could demonstrate nicely. In fact, when cleaning EEG from aperiodic signals alpha activity even shows an increase during development. Aperiodic-adjusted alpha activity is, moreover, reduced in participants with ADHD in the sample.

    What is particularly nice is that in a sub-sample where MRI scans were available alpha changes were associated with thalamic volume and thalamic white matter integrity. Only the latter, as was shown in the study, is associated with aperiodic-adjusted alpha power. This suggests that developmental changes of alpha activity might be driven by thalamic maturation.

    This is a really extremely well-done study. The idea is elaborated, and hypotheses well justified. The large sample provides good statistical power. The statistical modelling is a very smart way of analysing the data. The findings are definitely clarifying a lot of questions that can arise given the incoherent previous literature on this topic. I particularly like that a fundamental mechanism, namely maturation of the thalamus, was included to provide an explanation for developmental changes of alpha activity instead of simply speculating on the causes of the described EEG effect.

    All in all I think this is an amazing manuscript that will surely have a strong impact on the field.

    Was this evaluation helpful?
  3. Reviewer #2 (Public Review):

    Electrophysiological oscillations are a fundamental neural process that has rich implications in behavior. A major index of such activities is power (for a given frequency), which has substantiated a massive body of work in multiple domains. Conventional indices of power contain both the periodic component (reflecting oscillatory activity) and the aperiodic component (reflecting the exponential 1/f decay of power across the full spectrum). Recently, this confound in power indices has caught increasing attention. Therefore, it is time to address this potentially serious issue in the extant literature and promote the consideration of aperiodic signals in future research.

    Here, the authors decomposed the periodic and aperiodic components in the EEG alpha power measure. They demonstrated the confound of aperiodic signals in the total alpha power and elucidated the opposing relation of the periodic and aperiodic components with age. The approach is rigorous, including a large sample (N > 2500), validation in an independent sample, careful consideration of the effects of ADHD/clinical diagnoses. The results are largely supportive of the authors' claims.

    My main issue with the paper concerns the focus on "total" as opposed to "relative" power in the study. It has become standard in the literature to use "relative power", i.e., normalize power by dividing the total power by the average power of the full power spectrum. This normalization process effectively removes contamination/confounds of skull thickness and skin conductivity from total power. Therefore, the focus on a contaminated and thus rather obsolete measure of alpha power here is somewhat misplaced, undercutting the methodological and theoretical impact of the study.

    Another problem with the focus on total power is that its main contamination (skull thickness/skin conductivity) is shared by an aperiodic index, the aperiodic intercept. Therefore, the positive correlation between skull thickness and age can drive the negative correlation between age and both total power and aperiodic intercept.

    Relative alpha power and aperiodic slope are largely free of such signal contamination and should deserve more focused examination. Relatedly, it would shed useful light on the inherent association among these four major indices (total/relative alpha power, aperiodic intercept/slope) by running simple correlation among them. In fact, relative alpha power shows similar correlation with age as the aperiodic-adjusted alpha power. While the simulation results suggest that the former can lead to significant distortions, it is unclear, in real datasets and here in development, how the two indices of alpha power are related and whether the aperiodic confound merely weakens the validity of relative power, causing null findings. Such findings will inform the field as to how "distorted" the extant claims are and whether the extant literature suffers from false negatives or there is a high likelihood of alarming contradictions to true oscillatory activities such that a "new look" of previous findings is warranted.

    Was this evaluation helpful?
  4. Reviewer #3 (Public Review):

    In this paper, Troendle et al investigate changes in alpha oscillation across childhood and adolescence. The main goal of this investigation is to examine how alpha oscillations change across these age ranges, by investigating a large open dataset and adopting new methods that should help to address methodological limitations of many previous analyses. In particular, a key goal is to examine changes in periodic alpha power, and control for potential confounds due to changes in peak frequency and/or aperiodic activity. To do so, they employ a novel spectral parametrization method, and systematically compare measures of isolated periodic alpha activity to conventional measures. Overall, they find that they can replicate the age-related decrease of total alpha power when using conventional methods. However, when explicitly measuring and controlling for aperiodic activity, they find that periodic alpha activity actually increases with age. They suggest this discrepancy can be explained by changes in aperiodic activity, as the aperiodic slope and intercept are found to systematically change across age, in a way that likely drives the finding decrease of total alpha power, while the periodic alpha power actually increases. There are also some follow up analyses, including relating alpha power to anatomical measures of the thalamus, and to performance on an attention task.

    Strengths of this investigation include that it analyzes multiple, large datasets with well motivated methods. I think the goal of this paper addresses an important question, in terms of seeking to clarify some basic patterns of oscillation changes across development, and doing so in a rigorous way, both in terms of employing methods that are robust to estimating different features of the data, and in terms of using multiple, large datasets, including an internal replication of the main findings. I find the main goal and analysis compelling in terms of examining how alpha activity changes across this age range.

    I also find some limitations to some aspects of this paper and analysis that could be improved, as they do not always clearly describe the context or support the claims that are made for some of the follow-up analyses, as described in the following.

    1. Framing and prior literature

    I find some limitations in the organizing of this paper and it's relationship to prior work that could be improved, as I find that the paper could do better situating the analyses here with prior work, in particular in relation to the methodological issues it is addressing, and prior work on aperiodic activity.

    For example, in the abstract it is stated that "simulations in this study show that conventional measures of alpha power are confounded". Despite this statement, simulations are not a core feature of this study. There are a couple simulated examples in the supplement, which are referred to in lines 89-95, however it's worth nothing noting that while this section does not include any citations, the described issues, and related simulations, are very similar to points that have been made previously in the literature, that seem like they should be cited here:
    - Donoghue, T., Dominguez, J., & Voytek, B. (2020). Electrophysiological Frequency Band Ratio Measures Conflate Periodic and Aperiodic Neural Activity. ENeuro, 7(6), ENEURO.0192-20.2020. https://doi.org/10.1523/ENEURO.0192-20.2020
    - Donoghue, T., Schaworonkow, N., & Voytek, B. (2021). Methodological considerations for studying neural oscillations. European Journal of Neuroscience, ejn.15361. https://doi.org/10.1111/ejn.15361

    The paper also understates previous work on aperiodic activity, and the degree to which it is known to vary with age, in line 116-117 stating "there is insufficient evidence for the reported significant association between age and aperiodic signal components". This seems to ignore the large number of studies that have replicated this finding, including (some non-exhaustive examples):
    - Thuwal, K., Banerjee, A., & Roy, D. (2021). Aperiodic and Periodic Components of Ongoing Oscillatory Brain Dynamics Link Distinct Functional Aspects of Cognition across Adult Lifespan. Eneuro, 8(5), ENEURO.0224-21.2021. https://doi.org/10.1523/ENEURO.0224-21.2021
    - Voytek, B., Kramer, M. A., Case, J., Lepage, K. Q., Tempesta, Z. R., Knight, R. T., & Gazzaley, A. (2015). Age-Related Changes in 1/f Neural Electrophysiological Noise. Journal of Neuroscience, 35(38), 13257-13265. https://doi.org/10.1523/JNEUROSCI.2332-14.2015
    Perhaps this claim is supposed to more specifically reflect the age-range analyzed here, in which case recent studies examining this (in relatively large datasets) are also not mentioned here, including, for example:
    - Donoghue, T., Dominguez, J., & Voytek, B. (2020). Electrophysiological Frequency Band Ratio Measures Conflate Periodic and Aperiodic Neural Activity. ENeuro, 7(6), ENEURO.0192-20.2020. https://doi.org/10.1523/ENEURO.0192-20.2020
    - Hill, A. T., Clark, G. M., Bigelow, F. J., Lum, J. A. G., & Enticott, P. G. (2022). Periodic and aperiodic neural activity displays age-dependent changes across early-to-middle childhood. Developmental Cognitive Neuroscience, 54, 101076. https://doi.org/10.1016/j.dcn.2022.101076

    The notes above do not undermine the utility of examining alpha oscillations in detail, but I think the specific contribution of this work could be better contextualized in terms of other existing work. In the introduction, for example, the following review is an important piece of work that could be cited when introducing aperiodic activity:
    - He, B. J. (2014). Scale-free brain activity: Past, present, and future. Trends in Cognitive Sciences, 18(9), 480-487. https://doi.org/10.1016/j.tics.2014.04.003

    2. Model quality control

    A limitation to the methods employed in this study is a lack of description of if and how model fit quality was evaluated. For the method of parametrizing neural power spectra that is employed, it is important to validate that models fit the data well, otherwise the estimated parameters may be unreliable. This is especially important in developmental and clinical data, as analyzed here, as this data can be quite noisy, and differences in levels of noise across ages or between clinical groups could plausibly lead to differences in model fit quality. Useful quality checks for this kind of analysis would be to report the average r-squared (or model error) for the parametrized data, and to examine whether model fit quality is significantly related to age, or clinical status.

    Note that there is also a detailed guide for how best to apply spectral parametrization to developmental datasets, including notes on quality control, that may be useful:
    - Ostlund, B., Donoghue, T., Anaya, B., Gunther, K. E., Karalunas, S. L., Voytek, B., & Pérez-Edgar, K. E. (2022). Spectral parameterization for studying neurodevelopment: How and why. Developmental Cognitive Neuroscience, 54, 101073. https://doi.org/10.1016/j.dcn.2022.101073

    Not reporting any quality control metrics of the model fits also deviates from the analysis of the validation dataset as described in the pre-registered analysis (https://osf.io/7uwy2), which includes the note that the plan is for data to be excluded from the analysis if there is a bad model fit (R-squared < 0.9). It is unclear from the manuscript if this was done at all - and if so, why it was not described, and if not, why this deviates from the pre-registration. Note that though examining and reporting model fit quality is important, it is unclear where the value of 0.9 in the pre-registration came from, and it is unclear if this is an appropriate threshold for these specific datasets.

    3. The analysis of the relationship between the aperiodic intercept and aperiodic exponent

    There is an analysis in this paper that attempts to evaluate whether the change in aperiodic intercept that is observed is more than expected due to the measured change in aperiodic exponent. The approach taken for this analysis is ill-posed, and the interpretations made of this analysis are not supported. The issue is that the degree to which the intercept changes due to a change in exponent depend on the rotation frequency, which is not acknowledged or addressed in the analysis employed here.

    For example, for spectra rotated at 0 Hz, there is no measured change in offset from a change in exponent, whereas for a rotation at 100 Hz, there is a large influence of exponent on the change in offset, with different degrees of impact in between. The results of this analysis are therefore heavily influenced by the rotation frequency that is used. The analysis by the authors uses a rotation frequency of 19 Hz, however, there is no justification provided for this value. It is noted as being the middle point of the analyzed range, however, this itself is unrelated to whether it is an appropriate rotation frequency (since which frequency the spectrum rotates at is unrelated to the experimenter's decision of which frequency range to analyze).

    In real data, we don't a priori know what the rotation frequency point is, and in general it need not be a single, consistent point, and between subjects, is difficult to measure. To get a sense of what it might be, anecdotally, we can see in Figure 2C that in this particular subset, the rotation point is not at 19 Hz, and appears to be at a higher frequency. If the rotation point is actually higher than 19 Hz, then the analysis employed will systematically under-estimate the impact of the measured exponent change - leading to the conclusion that intercept is changing over and above the influence of the exponent. However, this conclusion is only valid if the rotation point of 19 Hz is accurate, and we would likely arrive at a different conclusion by picking a different rotation point. This analysis, by itself, is therefore invalid. Such an analysis would require a clear motivation of having measured the correct rotation frequency to be interpretable.

    4. Flanker Analysis

    Also relating to organization (similar to point 1) it is unclear why the analysis of the Flanker task, which is alluded to in the abstract, is only mentioned in the Discussion section. Given that this appears to be a key analysis, it is unclear why it is not presented in detail in the Results. The Flanker task and analysis is also not described in much detail in the methods. An issue with the Flanker analysis only being mentioned in the Discussion, with a link to supplemental table, is that the details of the results are somewhat obfuscated from the reader. When looking at these results, two key features seem notable - the first that though it is significant effect of aperiodic-adjusted alpha power, the beta value is very small (many times smaller than the coefficients for age and gender), and second, that although it doesn't quite pass significance, the estimated beta value for the total alpha power has the same magnitude as for the individualized alpha power. Between these two features, it is not clear if the relationship between aperiodic-adjusted alpha power and the Flanker performance is of sufficient magnitude to interpret that alpha power is related to attentional performance, and it's not clear that aperiodic-adjusted alpha power is more related to attentional performance than total alpha power (since a difference in significance does not necessarily imply a significant difference in the parameters). I think this analyses, as presented, therefore does not clearly support the claim made in the abstract that alpha power is found to relate to improved attentional performance.

    Was this evaluation helpful?