BOLD fractional contribution to resting-state functional connectivity above 0.1 Hz
Introduction
Conventional fMRI investigations of brain resting-state (RS) typically focus on functional connectivity (FC) below 0.1 Hz, and have contributed consistent and significant findings about the baseline brain function (Biswal et al., 1995, Fox et al., 2005, Fransson, 2005, Greicius et al., 2003, Zang et al., 2007). The rationale behind the great interest in the low-frequency fluctuations and the band-pass filtering (0.01–0.08/0.1 Hz) step employed in routine preprocessing of RS data is mainly threefold: (1) spontaneous signals associated with major RS networks have been found to be dominated by frequency components below 0.1 Hz (Damoiseaux et al., 2006, Fransson, 2005); (2) cardiac/respiratory-cycle-locked physiological noise components typically reside in frequency bands above 0.1 Hz, where neural-activity-relevant signal is believed to be minimal; and (3) conventional MR techniques only support whole brain acquisition at the time scale of seconds, which potentially limits the capability to observe spontaneous activity at a broader frequency spectrum.
Recent advances in MR techniques have allowed the examination of brain FC at faster temporal scales with improved signal to noise ratio (SNR) (Feinberg et al., 2010, Hennig et al., 2007, Larkman et al., 2001, Lin et al., 2006, Moeller et al., 2010, Zahneisen et al., 2011), and emerging evidence has shown that spontaneous activity may persist in frequency bands above 0.1 Hz (Boyacioglu et al., 2013, Gohel and Biswal, 2014, Niazy et al., 2011, Wu et al., 2008) and even up to at least 0.8 Hz (Boubela et al., 2013, Lee et al., 2013). Growing interest in the higher frequency behavior of spontaneous activity has yielded several interesting discoveries regarding RSFC. For instance, some groups reported frequency specificity of the spatial patterns associated with different RS networks, and the preliminary interpretations were linked with similar frequency-dependent behavior of spontaneous activity using electrophysiological recordings (Gohel and Biswal, 2014, Wu et al., 2008). Using a sliding window approach, Lee et al. (2013) observed more stable FC patterns in the visual/sensorimotor cortex in the 0.5–0.8 Hz band compared to 0.01–0.1 Hz, which may relate to the fact that high-frequency spontaneous activity can equilibrate in shorter time windows while low-frequency components could exhibit spuriously large dynamics if the minute-long window fails to encompass a few complete 2π cycles.
However, cautious optimism should be taken towards the advantages and potential opportunities brought by the observable high-frequency fluctuations: as an indirect measure of neuronal activity, blood-oxygenation-level dependent (BOLD) signal results from an inherently slow hemodynamic process, which in fact might be too slow to accommodate the observed high-frequency FC. Widely adopted models of task-evoked hemodynamic response functions (HRFs) (for instance, Glover, 1999, canonical HRF in SPM8, Wellcome Trust Centre for Neuroimaging, University College London, UK), have been tacitly assumed to apply equally well to either task-based or RS analysis, for example in de-convolving the true neural activity from RS BOLD responses (Niazy et al., 2011, Tagliazucchi et al., 2012, Wu et al., 2013) or establishing direct links between electrophysiological recordings and BOLD signals (Liu et al., 2012, Sadaghiani et al., 2010). However such HRF models only support the persistence of BOLD contrast at frequencies up to ~ 0.3 Hz, which thereby seems inconsistent with recent observations. Without questioning the validity of these HRF models, distinctions between task and rest have been widely accepted to lie in mental functionality instead of the underlying slow hemodynamic nature, which is thought to be limited by the process of perfusion through the venous compartment. Thus, it has become of critical importance to investigate whether the observed high-frequency FC originates from neural activity (through a BOLD mechanism) or other un-identified sources.
Recently, fMRI acquisitions with multiple echoes have been applied to differentiate between BOLD and non-BOLD components of fMRI datasets (Kruger and Glover, 2001, Kundu et al., 2012, Peltier and Noll, 2002), based on the fact that percent signal change of BOLD signal should be linearly dependent on TE due to R2⁎ (transverse relaxation rate) decay. Similarly, we can utilize the property of TE-dependence to examine whether the observed RSFC above 0.1 Hz also has a BOLD-like origin.
In the present study, we collected RS data at different TEs, attempting to gauge the relative contributions of BOLD and non-BOLD components to RSFC at different frequency scales (with TR = 1 s, we were able to resolve spontaneous activity up to 0.5 Hz). Resting-state HRFs were simulated by evaluating Buxton's balloon model (Buxton et al., 1998, Mildner et al., 2001) in the equilibrium state to heuristically estimate the qualitative changes of HRF waveforms that may accommodate the elevated frequency responses, and possibly the quantitative upper bound of frequency ranges that these changes may promise. Network patterns at two non-overlapping frequency bands (< 0.1 Hz) and (0.2–0.4 Hz) were further compared to assess the frequency dependence of the spatial patterns associated with two RS networks: the default-mode network (DMN) and the executive-control network (ECN).
Section snippets
Theory
Assuming mono-exponential decay, fMRI signals can be modeled as:where S0 is the initial signal amplitude at TE = 0, and R2⁎ is the inverse of relaxation time 1/T2⁎. Accordingly, the normalized signal change (the 1st order derivative of raw fMRI signal divided by the baseline amplitude S) should be an additive mixture of BOLD component — R2⁎ change (linearly dependent on TE), and non-BOLD component — small changes in S0 (Kruger and Glover, 2001, Kundu et al., 2012):
TE-dependence of correlated signal amplitude
Fig. 2 plots the correlated signal amplitude as a function of TE in the six different frequency bands: 0–0.5 Hz (B0), 0.01–0.1 Hz (B1), 0.1–0.2 Hz (B2), 0.2–0.3 Hz (B3), 0.3–0.4 Hz (B4), and 0.4–0.5 Hz (B5). Signals across all frequency bands exhibited a significantly linear dependence on TE, demonstrating persisting BOLD-like contributions to RSFC at frequency bands above the conventional 0.1 Hz. However, intercepts of the fitted linear models deviated from the theoretical zero, and were frequency
BOLD-like contributions to RSFC above 0.1 Hz
With TR = 1 s, we observed salient linear dependence of correlated signal amplitudes on TE at frequency bands up to 0.5 Hz within the DMN/ECN, demonstrating persistence of BOLD-like RSFC at frequencies above 0.1 Hz.
The apparent contradiction between predictions from the conventional HRF model (Fig. 4B) and BOLD-like signals at frequency bands up to 0.5 Hz (Figs. 4E, F) implies that canonical task HRFs may not be applicable to rest conditions, where the cerebral blood flow exhibits only small
Conclusion
This fMRI study provides further evidence supporting the persistence of RSFC at frequency bands higher than 0.1 Hz and the differences in network patterns across different frequencies. With acquisition at different TEs, we have observed BOLD-like linear dependence of spontaneous activity on TE, supporting neural relevance of the RSFC in extended frequency bands and implying that HRF models should be modified for rest compared to traditional task-based models. We have also demonstrated
Acknowledgment
Funding of this study is supported by NIH P41 EB015891. The authors gratefully acknowledge two anonymous reviewers for their constructive comments, which have substantially improved the quality of the manuscript.
References (58)
Modelling with independent components
NeuroImage
(2012)- et al.
The effect of scan length on the reliability of resting-state fMRI connectivity estimates
NeuroImage
(2013) Dynamic models of BOLD contrast
NeuroImage
(2012)- et al.
Mapping and correction of vascular hemodynamic latency in the BOLD signal
NeuroImage
(2008) - et al.
Influence of heart rate on the BOLD signal: the cardiac response function
NeuroImage
(2009) - et al.
EEG correlates of time-varying BOLD functional connectivity
NeuroImage
(2013) - et al.
In contrast to BOLD: signal enhancement by extravascular water protons as an alternative mechanism of endogenous fMRI signal change
Magn. Reson. Imaging
(2010) - et al.
To smooth or not to smooth? Bias and efficiency in fMRI time-series analysis
NeuroImage
(2000) - et al.
Inflow effects on functional MRI
NeuroImage
(2012) Deconvolution of impulse response in event-related BOLD fMRI
NeuroImage
(1999)