Mapping and correction of vascular hemodynamic latency in the BOLD signal
Introduction
There is a growing interest in using fMRI to study distributed networks of interacting regions across the whole brain (Marrelec et al., 2006, Rogers et al., 2007, Varela et al., 2001). Currently, time series modeling techniques are applied to regional BOLD signals in order to infer synchrony and directional influence among neural populations, both within specific tasks (see, e.g., Marrelec et al., 2006, Rogers et al., 2007 for review) and, more recently, during task-free resting state (see Fox and Raichle, 2007 for review).
A fundamental assumption of most fMRI correlation and causality studies is that the relative timing of BOLD signals from different regions of cortex reflects the timing of their underlying neural activity. However, the BOLD signal results from a coupling between neuronal activity and vascular responses (Bandettini et al., 1992, Kwong et al., 1992, Logothetis et al., 2001, Ogawa et al., 1992). Thus, differences in the timing (latency) of vascular responses between brain regions will affect the relative timing of the BOLD signals, thereby confounding the ability to infer neural connectivity and causality. The dynamics, location, and magnitude of the BOLD signal are heavily influenced by the vasculature in each voxel; large vessel effects can cause delays of up to 4 s relative to capillary effects (Bandettini, 1999, Lee et al., 1995). Widely disparate hemodynamic response functions (HRFs) have been found across the cortex (Bandettini, 1999, Buckner et al., 1998, Handwerker et al., 2004, Robson et al., 1998, Schacter et al., 1997, Thomason et al., 2005), and even adjacent voxels can exhibit latency differences of up to 2 s, which are most likely of vascular origin(Miezin et al., 2000). Latency differences may become more pronounced at higher spatial resolutions due to greater variability of partial volume effects from vascular compartments. Therefore, it is critical to model and correct for non-neural latency differences in the BOLD signal before attempting to quantify inter-regional correlations or causality.
While several researchers have quantified HRFs and timing delays in focal (sensory) regions prior to modeling their interactions (e.g. Chen and Desmond, 2005, Menon et al., 1998, Miezin et al., 2000), the vascular latency confound is largely overlooked in studies of whole-brain functional connectivity. One reason is due to the difficulty of characterizing HRFs across the whole brain, as most tasks used to probe HRFs activate only a small set of regions robustly and with adequate signal-to-noise ratio, and virtually all such HRF measurements have been limited to sensory regions. Even if such measurements could be made with sufficient reliability and accuracy, it is notclear that they generalize to other regions because of known regional differences in the vascular system. A second reason is that whole-brain, resting-state studies are less amenable to analyses such as condition-dependent modulation of connectivity/causality (e.g. Roebroeck et al., 2005, Sun et al., 2007) which essentially subtracts away the common vasomotor component of the HRF delay and thus circumvents the need for latency correction.
In the present study, we investigated whether a hypercapnic challenge (breath holding, or BH) can be used to quantify vascular latency differences across the whole brain. BH causes local up-regulation of blood flow when reduced perfusion, in the presence of ongoing baseline metabolism, leads to vasodilation of capillaries that is characteristic of local reactivity (Corfield et al., 2001, Kastrup et al., 1999aKastrup et al., 1999b, Kastrup et al., 1999c, Kastrup et al., 1998, Liu et al., 2002, Nakada et al., 2001). Since BH modulates cerebral blood flow to all vascularized brain regions without an accompanying change in CMRO2 (Kastrup et al., 1999a), it is a simple and robust method for assessing regional vascular reactivity properties, uncoupled from neural activation (Bandettini and Wong, 1997, Cohen et al., 2004, Thomason et al., 2005, 2007). We hypothesized that the latency of a voxel in the BH task would reflect the vascular component of its HRF latency, and that adjusting its time series accordingly would reduce confounds in further analysis.
Although previous studies have found an empirical relationship between BH and task-activation magnitudes (Handwerker et al., 2007, Thomason et al., 2007), it is not known whether BH timing delays willbe directly proportional to activation-induced vascular delays. The BH latency may incorporate the transit time of blood flow from major artieries in addition to the intrinsic reactivity of local capillary beds, and it is possible that the former may dominate the latency measurement in BH. In an analysis of time-to-peak and coherence phase, (Handwerker et al., 2007) found that latency in a BH task wasnot reliably predictive of latency in a visuomotor saccade task, though occasional significant correlations were observed; however, the relationship may have been weakened due to behavioral variability in motor response time during the saccade task. In the current study, we investigated the agreement between latency measured by BH and that from a paradigm designed to evoke simultaneous activation of primary sensory (auditory, visual, and motor) regions, with the expectation that a correspondence between latency values across the 2 tasks would provide evidence that BH latency primarily reflects local vasomotor reactivity.
As described previously, regional differences in latency would impact functional connectivity maps of networks such as the default-mode network (DMN). The DMN encompasses a set of regions that exhibit low-frequency correlated signals in task-free resting state (Greicius and Menon, 2004, Raichle et al., 2001), and collectively down-regulate during a wide range of cognitively demanding tasks (Binder et al., 1999, Gusnard et al., 2001, Mazoyer et al., 2001, McKiernan et al., 2003, Shulman et al., 1997). DMN connectivity has been shown to vary across groups (Baliki et al., 2008, Damoiseaux et al., 2008, Garrity et al., 2007, Greicius et al., 2007Greicius et al., 2004, Uddin et al., 2008) and cognitive states (Esposito et al., 2006, Waites et al., 2005), and an increasing number of studies seek to make inferences about behavior (Clare Kelly et al., 2008, Daselaar et al., 2004, Hampson et al., 2006) and dysfunction (Garrity et al., 2007, Greicius et al., 2007Greicius et al., 2004, Uddin et al., 2008) from measurements of DMN connectivity. Removal of non-neural latency differences is important for obtaining interpretable, quantitative results, particularly for between-group studies in which hemodynamics may pose a major confound.
Latency is also an important consideration in detecting task activation, particularly when event-related designs are employed. Analysis of event-related designs is sensitive to time shifts between the actual and modeled voxel responses, and methods to estimate andcompensate for potential delays within the framework of the general linear model have been proposed (Calhoun et al., 2004, Friston et al., 1998, Liao et al., 2002, Worsley and Taylor, 2006). In the present study, we also examined the effect of a BH latency correction on activation maps in an event-related working memory (WM) task.
Lastly, to illustrate the potential impact of BH latency correction oncausality analysis, we examined correction-induced changes in Granger causality among a set of regions in the WM task for one subject.
Section snippets
Subjects
Participants included 10 healthy adults (4 female) between the ages of 22 and 42 (mean age = 27.6), including 1 left-handed male. All subjects provided written, informed consent, and all protocols were approved by the Stanford Institutional Review Board.
Physiological monitoring
In the BH scan, subjects were monitored by a respiratory belt (TSD 201, Biopac Systems, Santa Barbara, CA) placed snugly around their upper thorax. The belt's electrical conductance is nominally proportional to the belt circumference and thus
Motion
Subjects had minimal overall and task-correlated motion in the BH,SM, and WM tasks (Table 1). Over the combined scans (BH + SM, BH + WM, and BH + Rest), a drift of less than one voxel was observed for all but one subject, who showed a 6.9 mm drift across the BH andWM tasks combined. It was found that results from this subject were consistent with the others, so the data were not excluded. A second subject had excessive motion only during the SM task (he found the stimulus tones amusing and laughed
Discussion
We have proposed the use of a hypercapnic challenge (BH task) as a method of measuring and correcting for relative hemodynamic latency differences across the whole brain. Such a task allows the interrogation of most gray-matter regions of the brain (Fig. 3) and is thus not limited to specific regions, as is the more typically employedvisual (or other sensory) task. Our results indicate that BH is a robust method for assessing non-neural, vasoreactivity-based latency differences across
Acknowledgments
This research was supported by NIH grants P41-RR09784 to GG and T32-GM063495 to CC.
References (76)
- et al.
Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI
NeuroImage
(2006) - et al.
Functional-anatomic study of episodic retrieval. II. Selective averaging of event-related fMRI trials to test the retrieval success hypothesis.
NeuroImage
(1998) - et al.
Latency (in)sensitive ICA. Group independent component analysis of fMRI data in the temporal frequency domain.
NeuroImage
(2003) - et al.
fMRI analysis with the general linear model: removal of latency-induced amplitude bias by incorporation of hemodynamic derivative terms
NeuroImage
(2004) - et al.
Cerebrocerebellar networks during articulatory rehearsal and verbal working memory tasks
NeuroImage
(2005) - et al.
Hypercapnic normalization of BOLD fMRI: comparison across field strengths and pulse sequences
NeuroImage
(2004) - et al.
Does hypercapnia-induced cerebral vasodilation modulate the hemodynamic response to neural activation
NeuroImage
(2001) - et al.
When less means more: deactivations during encoding that predict subsequent memory
NeuroImage
(2004) - et al.
Independent component model of the default-mode brain function: assessing the impact of active thinking
Brain Res. Bull.
(2006) - et al.
Motion correction algorithms may create spurious brain activations in the absence of subject motion
NeuroImage
(2001)