Resting-state fMRI signals contain spectral signatures of local hemodynamic response timing

Abstract

Functional magnetic resonance imaging (fMRI) has proven to be a powerful tool for noninvasively measuring human brain activity; yet, thus far, fMRI has been relatively limited in its temporal resolution. A key challenge is understanding the relationship between neural activity and the blood-oxygenation-level-dependent (BOLD) signal obtained from fMRI, generally modeled by the hemodynamic response function (HRF). The timing of the HRF varies across the brain and individuals, confounding our ability to make inferences about the timing of the underlying neural processes. Here we show that resting-state fMRI signals contain information about HRF temporal dynamics that can be leveraged to understand and characterize variations in HRF timing across both cortical and subcortical regions. We found that the frequency spectrum of resting-state fMRI signals significantly differs between voxels with fast versus slow HRFs in human visual cortex. These spectral differences extended to subcortex as well, revealing significantly faster hemodynamic timing in the lateral geniculate nucleus of the thalamus. Ultimately, our results demonstrate that the temporal properties of the HRF impact the spectral content of resting-state fMRI signals and enable voxel-wise characterization of relative hemodynamic response timing. Furthermore, our results show that caution should be used in studies of resting-state fMRI spectral properties, because differences in fMRI frequency content can arise from purely vascular origins. This finding provides new insight into the temporal properties of fMRI signals across voxels, which is crucial for accurate fMRI analyses, and enhances the ability of fast fMRI to identify and track fast neural dynamics.

Individual variability in the relationship between physiological and resting-state fMRI metrics

“…This suggests that either (a) there exists a common driver, such as the BOLD contrast itself, that impacts the two measures, but RSF and CVR are differently impacted over time, or (b) the impact of a common driving property is smaller than originally thought, or should be reconsidered altogether. This interpretation is in line with recent evidence of a stronger link between RSF and haemodynamic responses, specifically neurovascular coupling, than to CVR measured through a BH task (Bailes et al, 2023): while both RSF and CVR might describe a component related to neurovascular coupling, they are not necessarily linked to the same component. Hence, interpreting our results based on our unique dense sampling data, we join the point of view of Lipp et al (2015) and De Vis et al (2018) and express caution in adopting RSF measures interchangeably with CVR.…”

“…Fesharaki et al (2021) suggest that lower agreement between RSF measures and BH-induced CVR might be attributed to a lack of optimal thresholding of the corresponding maps. More recently, Huck et al (2023) showed that ALFF has a bias to the properties of venous vasculature, which might confirm the hypothesis that these RSF metrics are biased by resting cerebral blood volume (De Vis et al, 2018), although Bailes, Gomez, Setzer and Lewis (2023) attributes such bias more to the neurovascular coupling than to the CVR mechanism, since the latter does not include the same metabolic factors driving the former (D’Esposito, Deouell & Gazzaley, 2003; Iadecola, 2017; Pinto et al, 2021).…”

The temporal specificity of BOLD fMRI is systematically related to anatomical and vascular features of the human brain

“…Since functional connectivity estimates are delay-sensitive, it could be that the exact boundaries at which networks segregate are influenced by differences in their hemodynamic response properties even if the underlying neuronal activity were synchronous. Although prior studies have not specifically tested this effect, their results often display a spatial pattern that resembles our observed delays (Amemiya et al, 2020;Bailes et al, 2023;Lewis et al, 2016, fig. 7).…”

“…A challenge for identifying precise timing information in fMRI is that the temporal properties of hemodynamic responses are highly heterogeneous across the brain (Bailes et al, 2023; Handwerker et al, 2004; Pfeuffer et al, 2003) and across experimental contexts (Chen et al, 2021; Friston et al, 1998; Handwerker et al, 2012). BOLD responses are influenced by the vascular architecture of the cortex: they are weaker but have an earlier onset in the parenchyma, and stronger in the large draining veins that collect large volumes of deoxygenated blood from multiple capillary beds (Turner, 2002).…”

“…A challenge for identifying precise timing information in fMRI is that the temporal properties of hemodynamic responses are highly heterogeneous across the brain (Bailes et al, 2023;Handwerker et al, 2004;Pfeuffer et al, 2003) and across experimental contexts (Chen et al, 2021;Friston et al, 1998;Handwerker et al, 2012).…”