Literature analysis of aperiodic activity investigated using M/EEG and ECG. A) Illustration of different types of aperiodic activity in the time and frequency domain. BC) We analyzed 489 abstracts indexed on PubMed using LISC21, a package for collecting and analyzing scientific literature. B) This analysis revealed that changes in aperiodic activity are related to similar traits, states and disorders in measures of both neural and cardiac activity. C) We further noted a tiny overlap of studies (N=4) that refer to both cardiac and cortical aperiodic activity in their abstracts. Yet, none of these studies considers confounding influences of cardiac aperiodic activity on the measurement of cortical aperiodic activity. D) We additionally found a steep increase related to the investigation of neural aperiodic activity in the 2020s highlighting the current interest of the topic in the neuroscience community. EF) We further downloaded and analyzed freely available full-texts of M/EEG studies investigating aperiodic activity to see to which extent and how cardiac activity was handled. This analysis revealed that only 17.1% of EEG studies remove cardiac activity and only 16.5% measure ECG (for MEG 45.9% removed cardiac; 31.1% mention ECG was measured). We were further interested in determining which artifact rejection approaches were most commonly used to remove cardiac activity, such as independent component analysis (ICA22), singular value decomposition (SVD23), signal space separation (SSS24), signal space projections (SSP25) and denoising source separation (DSS26). We found that the most commonly applied method both in EEG and MEG recordings was independent component analysis (ICA). GH) An arbitrary selection of previous studies (N = 60) shows a vast amount of different frequency ranges are used to investigate aperiodic activity. While a significant amount of studies looked into a range between ∼0.1-50 Hz (∼30%), most studies used a unique frequency range. I) Not only do the upper and lower bounds vary between studies, but also the general width of the fitting range can vary from 0.9 to 290 Hz.

Aperiodic signals recorded using ECG are related to aging and heart rate variability. A) grand average power spectra plotted separately per Dataset and the associated aperiodic power spectra for the lower (0.25-20 Hz) and higher (10-145 Hz) frequency range. BC) Age was used to predict the spectral slope using different upper and lower slope limits for higher (10 - 145 Hz) and lower (0.25 - 20Hz) frequency ranges. Significant effects, i.e. effects with credible intervals not overlapping with a region of practical equivalence (ROPE; see Methods - Statistical Inference), are highlighted in red or blue (see colorbar). Null effects, which were defined as effects with credible intervals completely within a ROPE, are highlighted in green. Results where no decision to accept or reject (see40) an effect could be made, are masked using hatches. D) To understand whether aperiodic cardiac activity also relates to common measures of heart rate variability we predicted the spectral slope using 90 different measures of heart rate variability. We find consistent (yet different) associations with mostly fractal and time domain measures in both lower and higher frequency ranges.

Cardiac activity is captured in EEG and MEG recordings. AB) Cardiac activity is captured at zero time lags in concurrent MEG and EEG recordings, if the ECG signal is rejected via ICA this effect disappears in MEG, but not completely in EEG data. CD) Reconstruction of the ECG signal was impaired, but remained possible even after rejecting the ICA signal using ECG (both in MEG and EEG data). Notably, reconstruction of the ECG signal (after ICA) worked better in EEG than MEG data. A * indicates a “significant” effect (see Methods - Statistical Inference). E) To illustrate how aperiodic activity recorded using ECG might impact neural aperiodic activity we simulated cardiac and neural time series data. The neural time series data was simulated as in Gao et al.37 with an EI ratio of 1:2. The cardiac time series consists of a PQRST-Complex and different types of 1/f noise. G) Combining both cardiac and neural time series data shows that even if the PQRST-Complex is barely visible in the combined time domain signal, the resulting power spectrum can be heavily affected by simulated changes in aperiodic cardiac activity (GH).

Age-related changes in aperiodic brain activity are most pronounced in cardiac components. Age was used to predict the spectral slope at rest in three different conditions (ECG components not rejected, ECG components rejected and ECG components only) per channel across a variety of frequency ranges. A) Standardized beta coefficients either per channel averaged across all frequency ranges (left) or per frequency range (right) averaged across all channels. Age-related B) steepening, C) flattening and D) null effects in the spectral slope were observed and visualized in a similar manner as in A). EF) We further show the direction of results where we didn’t find enough evidence to support either a steepening, flattening or null effect. G) Summary of all observed findings in %. H) at some frequency ranges neural and cardiac aperiodic activity change independently with age (see also BC). I) At other frequency ranges cardiac and neural aperiodic activity are similarly modulated by age. We used cardiac and neural aperiodic activity as predictors for age in a multiple regression model to test whether both explain unique variance in aging. This analysis reveals that when adding both MEGECG component and MEGECG rejected as predictors age-related flattening effects were reduced, yielding no longer significant flattening results (see Figure 5B; upper panel). However, in case of the observed steepening effects, significant effects for MEGECG components remained in 98.75% of the tested frequency ranges (see Figure 5B; lower panel).

event-related spectral parametrization of working-memory in the ECG. A) Subjects were asked to either “listen” to or “memorize” a sequence of 5,9 and 13 digits (adapted from 63,64). Spectra in the “Baseline” period were compared to the “Delay” period of the “memorize” condition. B) The averaged evoked difference in the aperiodic spectrum between baseline and delay periods. The spectra were reconstructed from the aperiodic parameters of the spectral fits and plotted as a function of frequency past the average knee frequency (∼5Hz; C). The spectral slope of the ECG signal was significantly flatter during the “Delay” compared to the “Baseline” period (D). E) The flattening of the spectral slope relative to “Baseline” was strongest in conditions with higher working memory load. Error bars indicate standard errors of the mean. A * indicates a “significant” effect (see Methods - Statistical Inference).

ECG Spectra + Knee Frequency:

Aperiodic signals recorded using ECG can be associated with aging. A) grand average power spectra plotted separately per Dataset B) indicating a strong “knee” from ∼15Hz. CD) The spectral slope recorded using ECG was calculated using different upper frequency limits and correlated with age. This analysis shows that the association between age and spectral slope increased until ∼145 Hz in 3 of 4 datasets.

Raw Spectra + Goodness of fit metrics:

Grand average power spectra for the MEG data recorded at Cambridge and split in the three conditions MEGECG not rejected, MEGECG rejected, MEGECG components. A) Spectra averaged across channels split in older and younger subjects (median split). B) Aperiodic activity was extracted from the power spectra using the IRASA method 65 implemented in the YASA package 84. * indicate a “significant” effect (see Methods - Statistical Inference). C) Goodness of fit was assessed using R2 for both the IRASA and FOOOF model fit (see Supplementary S2/S3 for FOOOF fits). D) The spectral exponent obtained from FOOOF was compared to the spectral slope extracted from IRASA showing that both are highly related.

- Replication FOOOF:Age-related changes in aperiodic brain activity can be explained by cardiac components A) Age was used to predict the spectral slope at rest in three different conditions (ECG components not rejected [blue], ECG components rejected [orange], ECG components only [green]. B) Age distribution in the sample recorded at the University of Cambridge. C) Comparison of standardized beta coefficients shows that the strongest association with age is present on the data reflecting only ECG components * indicate a “significant” effect (see Methods - Statistical Inference).

Replication FOOOF Comparison of Knee vs Fixed aperiodic modes:

Comparison between Fixed and Knee model fits using FOOOF A) Age was used to predict the spectral slope at rest in three different conditions (ECG components not rejected [blue], ECG components rejected [orange], ECG components only [green]. B) Goodness of fit was assessed using R2 and compared between “Fixed” and “Knee” models. C) Comparison of the “Knee” and “Fixed” model fits shows that not fitting the knee was offering on average the better balance between goodness-of-fit and model complexity as suggested by the Bayesian Information Criterion (BIC). Error bars indicate confidence intervals.

Replication Salzburg Sample:

Age-related changes in aperiodic brain activity can be explained by cardiac components A) Age was used to predict the spectral slope at rest in three different conditions (ECG components not rejected [blue], ECG components rejected [orange], ECG components only [green]. B) Age distribution in the sample routinely recorded as part of MEG measurements at the University of Salzburg. C) Comparison of standardized beta coefficients shows that the strongest association with age is present on the data reflecting only ECG components * indicate a “significant” effect (see Methods - Statistical Inference).

SSS Maxfilter Analysis:

Age-related changes in aperiodic brain activity are explained by cardiac components AC) Age was used to predict the spectral slope (fitted at 0.1-145Hz) at rest in three different conditions (ECG components not rejected [blue], ECG components rejected [orange], ECG components only [green]. BD) The analysis in AC) was repeated for different fitting ranges with lower limits starting at 0.5 Hz in 1Hz steps ranging until 10 Hz and upper limits starting at 45 Hz in 5 Hz steps ranging until 145 Hz. Significant effects, i.e. effects with credible intervals not overlapping with a region of practical equivalence (ROPE; see Methods - Statistical Inference), are highlighted in red or blue (see colorbar). Null effects, which were defined as effects with credible intervals completely within a ROPE, are highlighted in green. Results where no decision to accept or reject (see40) an effect could be made, are masked using hatches.

ICA Thresholds:

A) The relationship between age and the aperiodic slope was compared within three conditions (MEGECG not rejected [blue], MEGECG rejected [orange] and MEGECG component [green] across different thresholds to select cardiac components from independent components of the MEG signal via ICA. Importantly, we always detected a significant relationship between age and aperiodic activity for the ECG component. While we did not always detect a significant relationship between age and aperiodic activity in MEG ECG not rejected and MEG ECG rejected conditions, the standardized β coefficients are heavily overlapping within each condition. Notably using a high threshold (r > 0.8) failed to identify cardiac components in >50% of the subjects. We therefore opted for a lower threshold of 0.4 for all related analysis in the main manuscript. Significant effects, i.e. effects with credible intervals not overlapping with a region of practical equivalence (ROPE; see Methods - Statistical Inference), are highlighted using a star. B) The amount of extracted ECG components per subject as a function of different ICA thresholds. On average, less than cardiac 2 components were extracted per subject irrespective of the used threshold. However, using high thresholds e.g. > 0.8 only allowed for the detection of cardiac components in less than 50% of the subjects.

SSS Maxfilter Analysis:

SSS Maxfilter Analysis: Steepening and flattening of the spectral slope with age is dependent on the sensor location and the investigated frequency range. Age was used to predict the spectral slope at rest in three different conditions (ECG components not rejected, ECG components rejected and ECG components only) per channel across a variety of frequency ranges (see Figure 1B). A) Standardized beta coefficients either per channel averaged across all frequency ranges (left) or per frequency range (right) averaged across all channels. Age-related B) steepening, C) flattening and D) null effects in the spectral slope were observed and visualized in a similar manner as in A). EF) We further show the direction of results where we didn’t find enough evidence to support either a steepening, flattening or null effect. G) Summary of all observed findings in %.

Head Movement, EOG control analysis:

Steepening and flattening of the spectral slope with age is dependent on the recording site and the investigated frequency range, when controlling for head movements. A) Head movement velocity was estimated using information obtained from the cHPI measurement. B) Spectral slopes were obtained also for horizontal and vertical EOG electrodes on the same frequency ranges as in Figure 4. Age was used to predict the spectral slope at rest in three different conditions (ECG components not rejected, ECG components rejected and ECG components only) per channel across a variety of frequency ranges. C) Standardized beta coefficients either per channel averaged across all frequency ranges (left) or per frequency range (right) averaged across all channels. Age-related D) steepening, E) flattening. Significant effects, i.e. effects with credible intervals not overlapping with a region of practical equivalence (ROPE; see Methods - Statistical Inference), are highlighted in red or blue (see colorbar). Null effects, which were defined as effects with credible intervals completely within a ROPE, are highlighted in green. Results where no decision to accept or reject (see40) an effect could be made, are masked using hatches.