Beyond unimodal data: an fMRI-electrophysiological-hormonal account of brain signal variance

2415 

Abstract Submission 

Giulia Baracchini¹, Jason da Silva Castanheira¹, Laura Pritschet², Tyler Santander², Hannah Grotzinger², Emily Jacobs², Mallar Chakravarty³, R. Nathan Spreng¹

¹Montreal Neurological Institute and Hospital, McGill University, Montreal, Quebec, ²University of California Santa Barbara, Santa Barbara, CA, ³Brain Imaging Centre, Douglas Research Centre, Montreal, Quebec

Giulia Baracchini  
Montreal Neurological Institute and Hospital, McGill University
Montreal, Quebec

Jason da Silva Castanheira  
Montreal Neurological Institute and Hospital, McGill University
Montreal, Quebec

Laura Pritschet  
University of California Santa Barbara
Santa Barbara, CA

Tyler Santander  
University of California Santa Barbara
Santa Barbara, CA

Hannah Grotzinger  
University of California Santa Barbara
Santa Barbara, CA

Emily Jacobs  
University of California Santa Barbara
Santa Barbara, CA

Mallar Chakravarty, PhD  
Brain Imaging Centre, Douglas Research Centre
Montreal, Quebec

R. Nathan Spreng  
Montreal Neurological Institute and Hospital, McGill University
Montreal, Quebec

Within-subject variability of fMRI BOLD signals is often considered as mere noise[1], yet it is associated with behavior, age and clinical status[2-3]. The promising translational value of BOLD signal variability is constrained by a lack of understanding of its properties and drivers[4]. fMRI data alone cannot be used to gain insight into the sources of brain signal variance, given the limitations posed by the physics of fMRI BOLD. More broadly, the interplay between brain and body signals may be of particular importance. Here, we merge across neuroimaging and brain-body modalities to build a framework for BOLD signal variability that has the potential to accelerate discovery.

In the first experiment (Fig 1), we bridged across neuroimaging modalities to identify which signal properties drive brain signal variability. We leveraged empirical ranges of amplitude, frequency and background 'noise' of fMRI and electrophysiological data, to fit Fast Fourier Transforms and simulate multi-modal timeseries. We manipulated, one at a time, the afore-mentioned parameters (1000 iterations) and, for each iteration, we quantified signal variability via the root Mean Square of Successive Differences (rMSSD)[5]. In the second and third experiments (Fig 1), given the clinical applicability of fMRI, we aimed at identifying the within-subject reliability of BOLD signal variability, and quantified the interplay between brain-body signal variance. We leveraged an open-source, multi-modal, deeply phenotyped, micro-longitudinal (n=1, healthy female, age 23) dataset of fMRI BOLD and hormonal data, collected over 30 consecutive days[6]. fMRI data were preprocessed and denoised following previous studies[6-7]. For each day, we obtained rMSSD measures and hormonal concentrations of estrogen, progesterone, LH, FSH and testosterone from blood samples. To assess reliability of BOLD variability, we used product-to-moment correlations between whole-brain rMSSD measures across the 30 days. To quantify the interplay between daily rMSSD and hormonal scores, we used Multiple Factor Analysis (MFA)[8] – a multivariate method similar to Principal Component Analysis that aims at explaining the shared variance between repeated measures of separate sets of variables. Network-level rMSSD values were entered in MFA, as defined by the Schaefer 400 regions-7 network atlas[9].

fMRI BOLD signal variability was driven by changes in signal amplitude and background 'noise', such that greater variability was obtained at higher amplitude and lower background 'noise' levels. Negligible effects were observed when manipulating BOLD signal frequency (Fig 2A). Electrophysiological simulations showed that signal variability can reflect changes in frequency, when a broader frequency range is provided than the one observed with fMRI. BOLD signal variability exhibited high test-retest reliability (median r=0.93; Fig 2B) and showed a positive association with daily fluctuations in estrogen levels (r=0.50, p=0.005). Daily fluctuations in brain signal variability strongly loaded together with daily hormonal fluctuations (PC1 38% variance; Fig 2C), highlighting how variations in the BOLD signal are rooted in complex interactions between brain and body. Estrogen variations preferentially mapped onto brain signal variations from modulatory networks (Salience and Default; PC2 31% variance; Fig 2C), and brain-body interactions additionally clustered based on the phases of the menstrual cycle (menses vs ovulatory window; Fig 2C).

Altogether, our study demonstrates the significance of conducting cross-modal investigations on brain signal dynamics, as a way to more fully capture the complexity of a fundamental property of biological signals -variance-, and maximize the insights and translational applicability of brain signal variability.

EEG/MEG Modeling and Analysis

Multivariate Approaches

Task-Independent and Resting-State Analysis ²

Multi-Modal Imaging ¹

Neurophysiology of Imaging Signals

Computational Neuroscience

Data analysis

ELECTROPHYSIOLOGY

FUNCTIONAL MRI

Modeling

MRI PHYSICS

Multivariate

Statistical Methods

Systems

Other - brain signal variance