Prevalent and sex-biased breathing patterns modify functional connectivity MRI in young adults

Abstract

Resting state functional connectivity magnetic resonance imaging (fMRI) is a tool for investigating human brain organization. Here we identify, visually and algorithmically, two prevalent influences on fMRI signals during 440 h of resting state scans in 440 healthy young adults, both caused by deviations from normal breathing which we term deep breaths and bursts. The two respiratory patterns have distinct influences on fMRI signals and signal covariance, distinct timescales, distinct cardiovascular correlates, and distinct tendencies to manifest by sex. Deep breaths are not sex-biased. Bursts, which are serial taperings of respiratory depth typically spanning minutes at a time, are more common in males. Bursts share features of chemoreflex-driven clinical breathing patterns that also occur primarily in males, with notable neurological, psychiatric, medical, and lifespan associations. These results identify common breathing patterns in healthy young adults with distinct influences on functional connectivity and an ability to differentially influence resting state fMRI studies.

Trial registration: ClinicalTrials.gov NCT01031407.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1. Gray plots of scans containing deep breaths and bursts.

Gray plots containing a deep breaths and b bursts. Upper panels show the z-scored respiratory belt traces in blue (y-ticks at left are z = −1 and 1), and 3 commonly derived respiratory measures (ENV, RV, and RVT, with vertical offsets to enable non-overlapping visualization; scales are identical in all figures). In the grayscale heat maps, all in-brain fMRI signals are shown organized by anatomical compartment, with a green line separating gray matter from white matter and ventricle signals. In a, three deep breaths are indicated by arrows, with major decreases (vertical black bands) in fMRI signal following each of the breaths. In b, over two dozen bursts are present (arrows mark several), with accompanying modulations of fMRI signals. Supplementary Figs. 1 and 2 show comprehensive versions of each scan. Additional exemplars of deep breaths (c) and bursts (d) are shown at the bottom. The code to produce gray plots was published in ref. , and comprehensive gray plots of all 17,640 scans are shown in Supplementary Movie 1.

Fig. 2. Examples of deep breaths and bursts.

Plots are formatted as in Fig. 1. Note in deep breaths a the slowness of the breaths, the possibility of transient apnea afterward, the incongruence of the respiratory measures ENV, RV, and RVT, and the presence of fMRI signal decreases in each instance. Note in bursts b the repeated, serial modulation of breathing amplitude, the congruency of ENV, RV, and RVT, and the rhythmic correlates in fMRI signals.

Fig. 3. Properties of deep breaths and bursts.

a Heat maps at the top illustrate 90-s segments surrounding visually identified events. In different subjects, 35 examples each of bursts, deep breaths, and isolated non-respiratory motions were identified from respiratory belt traces and motion traces. In the motion-exhibiting subjects, a random set of timepoints was also selected. Illustrated are respiratory belt traces, three respiratory measures (ENV, RV, and RVT), the global (gray matter average) fMRI signal, head motion (framewise displacement, following filtering, and 4-TR calculation as in ref. ), and DVARS (z-scored). A gray/red heatmap represents statistically significant differences from the random events (two-sample t-test, two-tailed) beyond p < 0.001, illustrated on a logarithmic scale capped at p < 1e−10. The basis of bursts and deep breaths are apparent in the respiratory belt images. b Mean signals of ENV, RV, and RVT congruently mark bursts, but deep breaths display differences across the respiratory measures, with RVT having little positive deflection. c Mean global fMRI signals differ across patterns: deep breaths have brief signal increases and marked signal decreases with nadir near 15 s, and 30 s to resolution, on average. Bursts have more marked positive deflections, and slower timecourses on average, resolving near 40 s on average. Motion produces no global fMRI signal changes. Shade plots reflect mean and std. d Deep breaths display considerable motion and DVARS changes time-locked to event onsets; bursts have smaller time-locked modulations that do not achieve significance. Source data are provided as a source data file.

Fig. 4. Rater scorings and algorithmic indices detect sex effects and global functional connectivity influences.

a Plots of total scans (of 4) with bursts and deep breaths for both raters, with score correlations and Cohen’s kappas inset, for N = 399 subjects (all panels display results from N = 399 subjects except d and e, which concern the three 21-subject groups). b Histograms of scores across subjects for both patterns, showing raters by color. c Bar graph of percent scans of each sex displaying patterns. Chi-squared tests of bursts yield p = 3.4e−8 and 8.8e−7 for J.D.P. and C.J.L. (denoted by ***), effects unchanged by excluding members of the three groups. No significant differences are seen by sex in deep breath scores. d Bar plots showing mean values with std error bars of the ratings in a clean, burst, and deep breath groups (each with N = 21 unrelated subjects). B and D denote burst and deep breath. Desired respiratory properties are found in each group. e Algorithm indices of the three 21-subject groups, corroborating rater scores and confirming desired breathing patterns (compare with d directly above). Box plots show median and 25th and 75th percentiles as boxes, whiskers encompass 99% of normally distributed data, outliers are individually marked (all box plots in later panels follow this format). f Algorithm indices of breathing patterns by sex, with significant differences by two-sample t-test in bursts but not deep breaths (compare with c directly above). g Box plots of algorithm indices for each pattern as a function of mean rater score, demonstrating significant Pearson correlations of humans and algorithm ratings. h Box plots of gFC as a function of mean pattern scores, showing much stronger effects of bursts on gFC, quantified by Pearson correlation. i Betas of multiple linear regression of pattern scores in gFC (gFC = b0 + b1*burst_score + b2*deep_breath_score), performed in each sex separately, showing much stronger effects of bursts. Bars show mean values, error bars show 95% confidence intervals; fits do not differ by sex (both n.s. by two-sample two-sided t-test, uncorrected for multiple comparisons). j Box plots of gFC and head size (intracranial volume, ICV) by sex, both significantly different by sex by two-sample two-sided t-test (p = 9.1e−9 and <1e−20, respectively). k Color chart of significance of main effects of multiple ANCOVA models. Sex effects become insignificant when both head size and respiratory variables are modeled. Source data are provided as a source data file, though group identity is redacted.

Fig. 5. Spatiotemporal properties of bursts and deep breaths.

a Color legend of network locations and colors from ref. , with text labeling of the networks of particular interest for this paper (full legend in Supplementary Fig. 11). b Correlation matrices are derived from spans of −10 to 40 s about the event onsets shown in Fig. 3 in minimally preprocessed data (red dotted lines in c), and show mean differences of 35 bursts and 35 deep breaths compared to 35 random onsets, only coloring cells significant at p < 0.05 by 10,000 permutation tests (nearly all cells are significant; non-significant cells are colored gray). In grayscale matrices at right, matrices of bursts were contrasted to deep breaths via 10,000 permutation tests, and the top and bottom 2.5% of actual differences among permutation ranks are illustrated (in white and black; gray is insignificant) in matrices for minimally preprocessed (MP), FIX-ICA-denoised (Post-FIX), and minimally preprocessed time series plus global signal regression (GSR). Differences exist under each processing strategy, prominently including visual, auditory, and somatomotor cortex (blue, pink, orange, and cyan). c Mean signals of 35 patterns from each kind of time series in b, with mild smoothing, colored by the legend above. Dotted ovals encircle the peaks and troughs of the aforementioned sensorimotor networks. d Surface representations of the events, the same data in b and c, in minimally preprocessed time series. Supplementary Movie 5 animates these patterns and those of non-respiratory motion onsets and random onsets.

Fig. 6. Functional connectivity associated with breathing patterns.

All images color only contrasts or differences significant at p < 0.05 by 10,000 permutation tests (gray cells are insignificant). a Contrasts of the three groups. The top row shows mean differences between groups. The bottom row shows the rank of mean group correlations amidst random, unrelated groups drawn from the entire cohort. b Mean within-subject differences between scans without breathing patterns (B−D−) and scans with bursts (B+D−, top row) or deep breaths (B−D+, bottom row) (using only scans where both raters fully agreed). c Betas of multiple linear regression performed separately in each sex, using unrelated, non-group-member subjects only. Regressors were z-scored, and all betas are shown for minimally preprocessed (MP) data. Betas for bursts alone are shown for several other data processing strategies (Supplementary Fig. 11 shows full sets of betas). In each of these three main analyses (a, b, c), bursts strongly associated with an increase in sensorimotor correlations (yellow dotted circles), and deep breaths lack elevation in these regions (blue dotted circles).

Fig. 7. Comparison of periodic breathing waveforms and bursts.

At left, single-subject waveforms of periodic breathing in opioid use, stroke, heart failure, at high altitude, and in newborns. These are all conditions and situations in which periodic breathing is commonly encountered. At right, illustrations of bursts in 11 HCP subjects. All plots are on the same time scale, and the green lines measure 1 min. Note the long cycle times seen in patients with heart failure (in the Stroke and Heart failure sections), reflecting, in part, an exaggerated delay in central detection of changes in arterial gas tensions (see Supplementary Discussion for more detail). The stroke example illustrates shows the added effect of delayed circulatory time in heart failure. Images at left modified from,– with permission. Images from ref. adapted with the permission of the American Thoracic Society. Copyright © 2020 American Thoracic Society. All rights reserved. The American Journal of Respiratory and Critical Care Medicine is an official journal of the American Thoracic Society. Readers are encouraged to read the entire article for the correct context at https://europepmc.org/article/med/15665317. The authors, editors, and The American Thoracic Society are not responsible for errors or omissions in adaptations.