Biophysical and physiological origins of blood oxygenation level-dependent fMRI signals

Abstract

After its discovery in 1990, blood oxygenation level-dependent (BOLD) contrast in functional magnetic resonance imaging (fMRI) has been widely used to map brain activation in humans and animals. Since fMRI relies on signal changes induced by neural activity, its signal source can be complex and is also dependent on imaging parameters and techniques. In this review, we identify and describe the origins of BOLD fMRI signals, including the topics of (1) effects of spin density, volume fraction, inflow, perfusion, and susceptibility as potential contributors to BOLD fMRI, (2) intravascular and extravascular contributions to conventional gradient-echo and spin-echo BOLD fMRI, (3) spatial specificity of hemodynamic-based fMRI related to vascular architecture and intrinsic hemodynamic responses, (4) BOLD signal contributions from functional changes in cerebral blood flow (CBF), cerebral blood volume (CBV), and cerebral metabolic rate of O(2) utilization (CMRO(2)), (5) dynamic responses of BOLD, CBF, CMRO(2), and arterial and venous CBV, (6) potential sources of initial BOLD dips, poststimulus BOLD undershoots, and prolonged negative BOLD fMRI signals, (7) dependence of stimulus-evoked BOLD signals on baseline physiology, and (8) basis of resting-state BOLD fluctuations. These discussions are highly relevant to interpreting BOLD fMRI signals as physiological means.

Figures

Figure 1

Relationship between neurovascular components and BOLD fMRI. (A) Cells are differentiated in this in-vivo rat brain image from two-photon laser scanning microscopy (provided by Alberto Vazquez at the University of Pittsburgh), where neuronal cell bodies and processes appear green (Oregon Green BAPTA-1AM), astrocyte cell bodies and processes are yellow (Sulforhodamine 101), and capillaries appear as dark bands. Increased activity in the neurons and astrocytes near and surrounding the capillaries induces an increase in capillary blood flow and dilation of upstream arterioles. (B) Vascular structure is depicted in this image from human cortex obtained by vascular casting and electron scanning microscopy, as adapted from Figure 4 in Reina-De La Torre et al (1998). The region outlined by the red rectangle approximates the size of the cellular image in (A). A dense network of small-diameter intracortical vessels appears below the dashed horizontal line, while a smaller number of large-diameter pial vessels appear above the dashed horizontal line. Blood flows from the pial arteries to the intracortical arteriolar branches, then into the dense capillary network, draining back through the intracortical venules and pial veins. (C) Whole-brain human BOLD fMRI studies depict functional responses localized to multiple brain regions (Kim and Ugurbil, 1997b). Functional MRI was acquired by the gradient-echo echo planar imaging technique with spatial resolution of 3.1 × 3.1 × 5 mm3 at 4 T. The size of each single fMRI pixel is approximated by the boundaries of the vascular image in (B); thus, there are numerous vessels of different sizes within each fMRI pixel, and contributions from these vessels must be considered to understand BOLD biophysical mechanisms. BOLD, blood oxygenation level dependent; MRI, magnetic resonance imaging; fMRI, functional MRI.

Figure 2

Vascular architecture relevant to hemodynamic-based fMRI. (A) This schematic illustrates a section from the cortical surface (top) to white matter (bottom) with the various types of arterial (A) and venous (V) intracortical vessels as classified by penetration depth (adapted from Figure 27 in Duvernoy et al, 1981). Group 6 (arteries only) extends to white matter with no cortical branches, group 5 also extends to white matter, but has cortical branches, group 4 extends to cortical layer 6 (highlighted in green), group 3 extends to layers 3 to 5 (layer 4 highlighted in yellow), and groups 1 and 2 extend only to the upper cortical layers. (B) This venogram is a 3-D T2*-weighted coronal MR image of cat brain obtained at 9.4 T with 78 mm isotropic resolution, 1.25 mm-thick slab selection, and minimum intensity projection performed to enhance the contrast of venous vessels; data acquisition and processing methods are reported elsewhere (Park et al, 2008). The bottom image is an expansion of signal within the dashed yellow box. White matter (outlined by dashed blue lines) is clearly differentiated from gray matter. Venous vessels draining from white matter (indicated by red arrows) belong to group 5. (C) This schematic of a tangential section (parallel to the cortical surface) shows venous vascular units (blue dots with black traces) with their arterial rings (red dots) (adapted from Figure 63 in Duvernoy et al, 1981), where the size of a single venous vascular unit (typically 1 to 2 mm in diameter) corresponds to the amount of tissue it drains. MR, magnetic resonance; MRI, MR imaging; fMRI, functional MRI.

Figure 3

Vascular responses to neural activity and expected changes to MR-related parameters. (A) This schematic adapted from Kim and Fukuda (2008) shows oxyhemoglobin (red dots) and deoxyhemoglobin (blue dots) in blood flowing through arteries, arterioles, capillaries, venules, and finally to veins. Oxygen delivered via oxyhemoglobin diffuses into extravascular tissue, and where it is used as a metabolic substrate. At prestimulus baseline conditions, blood oxygen saturation is ∼100% in arteries, while it is ∼60% in veins, although actual values vary with physiological condition. Increases in neural activity trigger an increase in blood velocity (indicated by the size of arrows) and dilation of vessels. The resulting increase in perfusion exceeds what is required by the increase in oxygen consumption rate, and the consequence is higher oxyhemoglobin levels and lower deoxyhemoglobin levels, such that capillary and venous oxygenation levels actually increase. (B) This table graphically illustrates the expected changes in the MR-related parameters of equation (1) for different compartments within each pixel due to vascular responses to neural activity. For simplicity, it is assumed that capillaries are composed of both arterial and venous blood. Upward arrows indicate an increase during stimulus, downward arrows indicate a decrease, and horizontal arrows indicate no change. BOLD, blood oxygenation level dependent; MR, magnetic resonance.

Figure 4

High-resolution fMRI of cat visual cortex. GE BOLD, SE BOLD, and CBV fMRI studies were performed at 9.4 T with 156 μm in-plane resolution and 2 mm slice thickness (Zhao et al, 2006); CBV fMRI was obtained after intravascular injection of long half-life iron oxide nanoparticles. During full-field binocular visual stimulus, it is expected that layer 4 will have the highest metabolic and CBF responses (Lowel et al, 1987), as well as the highest synaptic density and highest cytochrome oxidase activity (Woolsey et al, 1996). If an fMRI technique is highly specific to metabolic response and neural activity, then the middle of the cortex should therefore shows the highest signal change. In these studies, the highest CBV change (C) indeed occurred at the middle of the visual cortex in layer 4. GE BOLD fMRI (A) has the highest percent signal change at the cortical surface, where large pial vessels are located (green contours), while large vessel contributions are suppressed in SE BOLD (B). BOLD, blood oxygenation level dependent; fMRI, functional magnetic resonance imaging; GE, gradient echo; SE, spin echo; CBV, cerebral blood volume; TE, echo time; CBF, cerebral blood flow.

Figure 5

Dynamic changes in stimulus-induced CBF, BOLD, CBV, and CMRO2-related fMRI signals. Stimulus starts at time=0 second and stimulus periods are indicated by black horizontal bars. (A) Dynamic CBF and BOLD responses to forepaw stimulus were obtained in the rat somatosensory cortex (Silva et al, 2000). These time courses for regions of interest from the cortical ‘surface' and the middle of the cortex (‘deep') show that the CBF response is earlier than the BOLD response. (B) Dynamic BOLD, CBV, and CMRO2-related responses were obtained in cat visual cortex (Nagaoka et al, 2006). BOLD and CBV data were obtained at normal physiological conditions, while the CMRO2-related change was acquired with infusion of a vasodilator to suppress stimulus-induced CBF and CBV changes. The CBV response was obtained after intravascular injection of iron oxide nanoparticles. Peak intensity from the time course of each response type was used for normalization. The BOLD response starts slightly later than CBVt and CMRO2 responses, but BOLD and CBVt peak times are similar during 10 seconds stimulation, with a lag in CMRO2 peak time. Note that a small BOLD dip was observed. (C) The amplitude and dynamics of BOLD responses in human visual cortex (Cohen et al, 2002) are closely dependent on baseline CBF levels. When baseline CBF is low due to hypocapnia, the BOLD response is of higher amplitude and faster versus normocapnia. Conversely, when baseline CBF is high due to hypercapnia, the BOLD response is of lower amplitude and sluggish versus normocapnia. BOLD, blood oxygenation level dependent; fMRI, functional magnetic resonance imaging; CBV, cerebral blood volume; CBF, cerebral blood flow; CMRO2, cerebral metabolic rate of oxygen utilization; CBVt, total CBV.

Figure 6

Compartment-specific CBV changes during sensory stimulus. (A) Corresponding total, arterial, and venous CBV changes were measured by fMRI in the cat visual cortex (Kim and Kim, 2011). The arterial CBV change was obtained by fMRI using the varied magnetization transfer effect (Kim et al, 2008), while total CBV was determined with intravascular infusion of iron oxide nanoparticles. The venous CBV change was then estimated by the difference in total CBV and arterial CBV changes. The arterial CBV response is rapid like the total CBV response, while the venous CBV change is relatively slow. (B–D) Imaging of the rat somatosensory cortex followed isoflurane anesthesia and craniotomy (provided by Alberto Vazquez at the University of Pittsburgh). Intrinsic optical image with signal weighted to red blood cells shows pial vessels as hypointense structures (B). Baseline fluorescent microscopic images (C) enhanced by intravascular injection of Texas red dye show arterial and venous vessels (indicated by ‘A' and ‘V', respectively) as hyperintense structures. In the functional change map (D) determined by the difference between baseline images and images acquired during 4 seconds of forepaw stimulus, the highest signal intensities indicate stimulus-induced increases in dye content due to dilation. Only arterial vessels dilate, while venous vessels do not. fMRI, functional magnetic resonance imaging; CBV, cerebral blood volume.