Cortical depth-specific microvascular dilation underlies laminar differences in blood oxygenation level-dependent functional MRI signal

Abstract

Changes in neuronal activity are accompanied by the release of vasoactive mediators that cause microscopic dilation and constriction of the cerebral microvasculature and are manifested in macroscopic blood oxygenation level-dependent (BOLD) functional MRI (fMRI) signals. We used two-photon microscopy to measure the diameters of single arterioles and capillaries at different depths within the rat primary somatosensory cortex. These measurements were compared with cortical depth-resolved fMRI signal changes. Our microscopic results demonstrate a spatial gradient of dilation onset and peak times consistent with “upstream” propagation of vasodilation toward the cortical surface along the diving arterioles and “downstream” propagation into local capillary beds. The observed BOLD response exhibited the fastest onset in deep layers, and the “initial dip” was most pronounced in layer I. The present results indicate that both the onset of the BOLD response and the initial dip depend on cortical depth and can be explained, at least in part, by the spatial gradient of delays in microvascular dilation, the fastest response being in the deep layers and the most delayed response in the capillary bed of layer I.

Neuroglial activation is accompanied by release of vasoactive mediators that dilate and constrict the surrounding arterioles (1, 2) and capillaries (3, 4). These changes in diameter in turn lead to changes in blood flow throughout the vascular matrix and can be detected on the macroscopic level as a positive blood oxygenation level-dependent (BOLD) functional MRI (fMRI) signal when blood flow response exceeds oxygen consumption (5–7). Under the assumption of local neurovascular coupling, the onset of the changes in diameter is determined by the following three factors, any of which may differ as a function of the cortical depth and branching order within the vascular tree: (i) the onset and peak time of the neuronal activity evoking the response; (ii) the time needed to release a vascular messenger [e.g., prostaglandin or NO (8)]; and (iii) the time needed for the target vessel to respond. However, in addition to local neurovascular coupling, vascular responses can propagate within the arteriolar/capillary networks (3, 9, 10). Indeed, propagation of dilation and constriction has been observed on the cortical surface (11–15), in excised cerebral vessels, and in noncerebral preparations (16, 17).

Previous studies with single-vessel resolution in vivo have been limited to the cortical surface, but recent improvements in two-photon microscopy technology allow direct imaging of single-vessel diameters and flow velocities within a 3D geometry of vascular trees (1, 2, 18, 19). In the present study, we used this technology to examine microvascular responses to sensory stimulation down to 550 μm below the cortical surface in the rat primary somatosensory cortex (SI). We then compared the results with high-resolution BOLD fMRI to investigate the extent to which laminar BOLD profiles reflected the underlying microvascular dynamics. Specifically, we focus on the following questions. (i) What is the location of the fastest dilation onset within the 3D branching arteriolar/capillary tree? (ii) Can we observe a gradient of microvascular onset and peak times as a function of the cortical depth or branching order? (iii) What features of the cortical depth-resolved BOLD response can be explained by the microscopic vascular measures?

Results

Spatial Gradient of Dilation Onset and Peak Times Along the Trunks of Diving Arterioles Suggests Propagation Toward the Pial Surface.

We used two-photon microscopy in combination with intravascular injection of a fluorescent contrast agent (fluorescein-dextran) to follow 90 diving arterioles down to 550 μm below the cortical (pial) surface in the rat SI. All measurements were acquired within ∼1.5 mm from the center of the neuronal response, mapped before two-photon imaging using surface potential recordings (1, 2). First, we address the issue of propagation of dilation along main diving arterioles toward the surface. Fig. 1A shows a typical example of the cortical microvasculature. In the expanded view on the right, surface arterioles (red) dive down (red arrows) to supply oxygen and glucose to the capillary bed. Surfacing venules (blue arrows) bring the deoxygenated blood from the capillary bed to draining surface venules (blue). For every arteriole, diameter change in response to a forepaw stimulus (2 s, 3 Hz, ∼1 mA) was measured at multiple depths along the diving trunks and their lateral branches. The majority of diving arterioles and their branches (82%) exhibited 5–25% dilation, with no significant dependence on the depth (Fig. S1). An example of a set of measurements acquired along an individual arteriolar trunk, including the parent surface arteriole measured close to the diving point, is shown in Fig. 1B. The dilation time-courses measured at different depths were normalized by the peak amplitude (time-courses without normalization are shown in Fig. S1). Trunk measurements for different depth categories, averaged across all measured arterioles (population-averaged), are overlaid in Fig. 1C and Fig. S1A. The number of different vessels in each category is listed in Table S1. We found that the delay in vascular response increased with decreasing cortical depth. For each measurement, we extracted the onset time (Fig. 1D) and time-to-peak (Fig. 1E). The onset was estimated by fitting a line to the rising slope between 20 and 80% to the peak and calculating an intercept with the prestimulus baseline (Inset, Fig. 1D). Although estimation of the onset potentially can be biased by differences in the signal-to-noise ratio of measurements in different layers, the prestimulus SD in our sample was independent of the cortical depth (Fig. S1E). Both the onset and time-to-peak decreased significantly with an increase in the cortical depth (P < 0.05). We estimated a propagation speed of 1,100 and 600 μm/s for the onset and time-to-peak, respectively, based on the linear regression slope.

Fig. 1.

Relative timing of dilation response along diving trunks. (A) (Left) Image calculated as a maximum intensity projection (MIP) of an image stack 0–300 μm in depth using a 4× objective. Individual images were acquired every 10 μm. (Right) Detailed view of the region within the white square on the left, calculated as a MIP of a stack 0–400 μm in depth acquired with 2-μm resolution. (B) An example of a set of temporal diameter change profiles acquired from an individual arteriolar tree. Each curve is an average of eight stimulus trials. The curves are normalized by the peak amplitude. (C) Population-averaged and peak-normalized time-courses of diameter change for the arteriolar trunks from different cortical depths. Insets in B and C show zoomed-in views of the first 3 s following the stimulus and define color coding. There was no statistically significant change in peak amplitude with depth; the same time-courses without normalization are shown in Fig. S1. (D) Onset time of the arteriolar trunk dilation as a function of the cortical depth. Inset shows an example of fitting a line to the rising slope for the estimation of dilation onset. (E) Time-to-peak of the arteriolar trunk dilation as a function of the cortical depth. In both D and E, the straight line depicts the trend obtained from linear regression fitting. The number of measurements for each category is listed in Table S1.

Spatial Gradient of Dilation Onset and Peak Times as a Function of Branching Order Suggests Propagation into Capillary Beds.

We followed branching arteriolar trees and measured diameter changes evoked by a forepaw stimulus while keeping track of the cortical depth and branching order. An example of a set of measurements is shown in Fig. 2A. Fig. 2 B and C and Fig. S1 B and C show averaged time-courses of arteriolar diving trunks (black), their first-order branches (magenta), and higher-order branches (yellow) for layer I (<150 μm) and layer II/III (150–550 μm). The distribution of baseline diameters for different branching orders is shown in Fig. S2. As in the previous case (Fig. 1), we extracted the onset time and time-to-peak for each of the measurements (Fig. 2 D and E). Similar to the trunks (Fig. 1 D and E), there was a gradual delay in the response with a decrease in the cortical depth. In addition, for each depth category, the response of the branches was delayed relative to the trunks at the same depth (Fig. 2 B and C). Two-way ANOVA showed that the onset and time-to-peak were affected significantly by both the cortical depth (onset: P < 0.001, F = 45; time-to-peak: P < 0.001, F = 41) and branching order (onset: P < 0.001, F = 8; time-to-peak: P < 0.001, F = 30). For the onset, the effect of depth interacted with the effect of branching order (P = 0.025, F = 4). Indeed, high-order branches in layer I were visibly delayed relative to those in layer II/III (Inset, Fig. 2C).

Fig. 2.

Relative timing of dilation as a function of branching order. (A) (Left) MIP of an image stack with a depth of 0–200 μm. The red and blue arrows show the direction of flow in a diving arteriolar tree and in one of the surfacing venules connected to the same capillary bed. (Right) Time-courses of the arteriolar trunk and two first-order branches at different depths are overlaid. Each curve is an average of eight stimulus trials. The curves are normalized by the peak amplitude. (Center) Schematic drawing of branching within the top 200 μm. (B) Population-averaged time-courses of arteriolar trunks and their branches in layer I (<150 μm). (C) Population-averaged time-courses of arteriolar trunks and their branches for layer II/III (150–550 μm). (Inset) The time-courses of higher-order branches from the two depth categories are overlaid. There was no statistically significant change in peak amplitude with depth; the same time-courses without normalization are shown in Fig. S1. (D) Onset time as a function of the cortical depth. (E) Time-to-peak as a function of the cortical depth. In both D and E, the three straight lines (black, magenta, and yellow) depict the trend obtained from linear regression fitting of the trunks, first and higher order branches, respectively.

Together with Fig. 1, these results demonstrate that the fastest dilation occurred at or below the penetration limit of our two-photon measurements. The observed gradients of onset and peak times were consistent with propagation upstream along the arteriolar trunks and into the local capillary bed via the lateral branches. The largest delays were observed for the smallest branches (capillaries) in layer I (Fig. S3).

Laminar Differences in BOLD Transients Might Reflect the Underlying Depth-Specific Microvascular Dynamics.

To investigate the extent to which the observed microvascular dynamics might be reflected in BOLD fMRI signals, we performed cortical depth-resolved BOLD fMRI. fMRI experiments were done in a separate group of animals using an identical stimulus protocol. Fig. 3 A–C shows an example from a single representative animal. Fig. 3C shows BOLD ratio images using a single slice cutting through the center of the evoked response. The ratio images were calculated relative to the prestimulus baseline to enable comparison with intrinsic optical imaging from previous studies (Methods). The same data are presented as P-value maps in Fig. S4. The positive response (red and yellow colors) corresponds to the forepaw area in SI, contralateral to the stimulated paw. In agreement with our previous studies using optical (spectral) measurements of blood oxygenation and laser speckle imaging of blood flow (1, 20, 21), there was a sustained negative BOLD response (blue colors) in the surrounding region and ipsilateral SI (Fig. S5).

Fig. 3.

Depth-resolved fMRI. (A) (Upper) A raw EPI image of the slice that cuts through the center of evoked activity and a zoomed ROI in the cortex contralateral to the stimulated forepaw (1 mm along the dorsal curvature). The ROI is divided into six 200-μm color-coded slabs. (Lower) The same image as in A thresholded at 40% of the maximum intensity to reflect the sensitivity of the surface RF coil (for display purposes only). (B) BOLD signal time-courses from the same subject using the ROI indicated in A. Color-coded time-courses from different slabs are superimposed. The red arrowheads here and in D and E point to the initial dip. Inset shows a zoomed segment between −2 s and 4 s relative to the stimulus onset. (C) Ratio images of BOLD contrast for the same subject. The images are thresholded as in A. The color scale is expressed as percent signal change relative to the prestimulus baseline. Time (in seconds) relative to stimulus onset is indicated above the images. Note that the upper and lower rows are sampled at different Δt. (D) Time-courses averaged across subjects. For each subject, the time-courses were normalized by the peak of the surface response before averaging. Inset shows timing differences between two slabs corresponding to the top 200 μm (L1) and 600–800 μm (L4). Error bars indicate SE. (E) Time-courses averaged across subjects. Different slabs are normalized by peak amplitude. Error bars shown on the L1 slab (red) and the L4 slab (blue) indicate SE.

We defined a region of interest (ROI) in the following way: After tracing the cortical surface and drawing a grid (Methods), we defined the surface borders of the ROI as ±0.5 mm from the center of the BOLD response. The ROI extended down to 1.2 mm below the surface and was divided into six 200-μm slabs (color-coded in Fig. 3). To examine potential laminar differences, we extracted time-courses from all voxels within a particular slab (Fig. 3B). Fig. 3 D and E shows the average across subjects. The first three slabs (red, green, and cyan) lay within the depth penetration of the two-photon measurements. The first detectable contralateral BOLD response was negative and was most pronounced in the surface slab (red arrowhead, Fig. 3); this response may correspond to an initial brief decrease in oxygenation that often is observed in optical imaging studies (21–24) and with O₂ electrode measurements (25, 26) but rarely by fMRI (6, 27). This initial surface negativity, the “initial dip,” was visible in time-courses extracted from the ROI in each subject (Fig. 3B) but did not reach a voxelwise statistical significance threshold (P < 0.01) within single subjects (Fig. S4). Across-subject ROI analysis revealed a significant initial dip in the surface slab (red curves in Fig. 3 D and E; P < 0.05; error bars represent SE). After the dip, there was an increase in BOLD signal in all slabs that lasted beyond the stimulus duration (2 s). The biggest increase in BOLD signal was observed at the cortical surface. Examination of rise times of the normalized time-courses across the slabs revealed that surface responses were delayed relative to the deeper slabs, whereas the response in the middle slabs (L4 in Fig. 3 D and E) preceded the response in other layers. The fast rise of the positive BOLD signal in the middle slabs is in agreement with a previous high-resolution fMRI study by Silva and Koretsky (28). At each depth, the onset of the BOLD signal lagged behind the dilation of the higher-order branches by >0.5 s, as expected because of the extra time needed to wash out deoxyhemoglobin and reoxygenate the blood (Fig. S6).

Discussion

We used in vivo depth- and vascular compartment-resolved two-photon imaging of single blood vessels and high-resolution BOLD fMRI to investigate the possible contribution of depth-specific microvascular dynamics to laminar-resolved BOLD signal changes. Our results indicate that both the positive BOLD onset and initial dip depend on cortical depth and can be explained, at least in part, by the spatial gradient of microvascular dilation delays, with the fastest response in the deep layers and the most delayed response in the capillary bed of layer I.

Our two-photon results indicate that the deepest measured arterioles and capillaries had the fastest dilation. Within the top 550 μm, we observed a clear gradient of onset and peak times consistent with the propagation of dilation up along the diving trunks. Upstream propagation of vasodilation has been observed in vivo on the cortical surface (11–15) but not in the depth domain. At each depth, dilation latency increased monotonically with increasing branching order. As a result, dilation response of the capillary bed in layer I was most delayed, by ∼1 s relative to the diving arterioles at 500 μm. This behavior was paralleled by the delayed rise of the positive BOLD response in the top cortical layers and the initial dip that was most pronounced in the superficial layer. These findings are consistent with the original idea that the initial dip results from an increase in oxygen consumption in the tissue that precedes the blood flow response (24, 29, 30). Indeed, the largest dip colocalized with the slowest onset of vascular response along the depth axis. In addition to oxygen consumption in the tissue, the initial dilation of the capillary bed on its own can contribute to the initial dip before being counteracted by an increase in flow leading to the washout of deoxygenated blood. In our case, this effect is unlikely, because the dip largely preceded the dilation. However, a confounding factor of the initial volume increase has been reported in awake monkeys, a model in which the transit times are longer than in the rat and the attentive state might produce an increase in cardiac output and cholinergic modulation (31). The laminar specificity of the initial dip observed in our study can explain the long-standing discrepancy between optical and fMRI studies (27, 32), because optical imaging signals are strongly weighted toward the superficial layers, whereas the fMRI signal is weighted more uniformly throughout the voxels. With the typical voxel sizes used in human fMRI studies, the partial volume effects would be expected to result in minimal sensitivity to the initial dip.

Our results indicate that, within the depth range of two-photon measurements, the onsets of positive BOLD response and microvascular dilation decreased with depth at a comparable rate. The onset of the positive BOLD response depends on the timing of dilation (an increase in blood volume), the time-course of oxygen metabolism, and the rate of deoxyhemoglobin washout. We previously have described a mathematical modeling framework to predict the macroscopic hemodynamic response based on microscopic vessel dilation and oxygen metabolism parameters (33, 34). The current findings with regard to BOLD onset timing are in good agreement with the prediction of this vascular anatomical network model, with the onset of positive BOLD lagging behind the dilation onset at each depth by >0.5 s.

Laminar differences in the BOLD response have been demonstrated in an earlier high-resolution fMRI study (28). The authors attributed these differences to neuronal propagation from layer IV, the cortical input layer, to the superficial and deeper layers. However, neuronal activity is known to spread throughout the cortical depth within a few milliseconds, too fast to explain the observed laminar hemodynamic differences of 500 ms or more. For example, in our previous publications we reported tight synchronization of the onset and peak of neuronal activity (both spiking and synaptic currents) throughout the cortical depth evoked by the forepaw or whisker stimulation on a time scale of <10 ms (1, 2, 22, 35). The present study provides an alternative explanation based on laminar differences in the timing of dilation onset.

Our findings do not imply that propagation within the vascular wall is the only mechanism behind the observed timing gradients and do not rule out local neurovascular communication in the upper layers. In the presence of both local and propagated signaling, the onset of dilation would be determined by the faster of the two processes. Specifically, substantial delays observed in the superficial arteriolar branches and capillaries suggest that local neurovascular communication in layer I, if it exists, has slower kinetics. Further investigations are needed to address the mechanisms behind this phenomenon; such mechanisms might include a laminar gradient in the release of vasoactive substances (dependent, for example, on laminar differences in the amplitude of the neuronal response within a particular population of neurons), diffusion of vasoactive mediators (e.g., NO), or laminar differences in the intrinsic properties of the local microvasculature.

Previous studies in vivo have demonstrated that both arterioles and capillaries can change diameter in response to stimulation (1, 2, 4, 36). Arteriolar dilation is not expected to contribute substantially to BOLD contrast because of the high oxygenation level of the arteriolar blood and the small volume of arterioles relative to other (capillary and venous) compartments. Capillary dilation that coincides temporally with flow increase would partially mask the positive BOLD signal because it leads to an increase in the partial volume of blood within an MRI voxel. The magnitude of this effect cannot be estimated from our results, because all capillaries in our sample were connected directly to precapillary arterioles, and their dilation might not reflect the average behavior of the capillary bed.

A potential limitation of the study is that the measurements of different modalities were performed in separate groups of animals. Nevertheless, the response was consistent across subjects, and the key features were evident in each subject. Another potential limitation is the use of anesthesia, including possible interference of α-chloralose with vascular dynamics (37, 38). However, two-photon measurements and fMRI were performed under the same anesthesia conditions and therefore are expected to be subject to the same effect, rationalizing the comparison across the measurement modalities.

In the future, systematic measures of the microvascular response together with direct measurements of intra- and extravascular oxygenation (39) will allow the fitting of mechanistic models of the BOLD response, with a firm grounding in the underlying biophysical parameters. Such models ultimately may allow quantitative estimation of physiological parameters, including vascular and metabolic activity, from noninvasive macroscopic measures in health and disease.

Methods

Animals.

We used 19 Sprague-Dawley rats (130–200 g) for two-photon experiments and six Sprague-Dawley rats for fMRI experiments. All experimental procedures were approved by the University of California at San Diego Institutional Animal Care and Use Committee. All animal procedures were performed as described in refs. 1 and 2.

Stimulus and Synchronization with Data Acquisition.

The stimulation lasted 2 s and consisted of a train of six electrical pulses (3 Hz, 300 μs, ∼1 mA) with an interstimulus interval of 20–25 s. The intensity was adjusted to provide stimulation below the movement threshold.

In both optical and fMRI experiments, stimulation was presented using a separate computer that also acquired transistor--transistor logic (TTL) timing signals for data acquisition (“trigger out” TTLs for each line or frame during two-photon acquisition and for each slice during fMRI) using a National Instruments IO DAQ interface controlled by a home-written software in Matlab. The TTL data were used to determine the timing of each line/frame/slice relative to the stimulus onset during data analysis performed in Matlab.

Two-Photon Microscopy.

Two-photon microscopy was performed as described in ref. 1. Fluorescein-conjugated dextran (FD-2000; Sigma) in physiological saline was injected i.v (40). Images were obtained with an Ultima two-photon microscopy system from Prairie Technologies using 4× (Olympus XLFluor4×/340, NA = 0.28) and 40× (Olympus, NA = 0.8) objectives. Line scans up to 1 mm long were acquired across multiple vessels (up to six) with a scan rate of 80–170 Hz. The scan resolution was 0.5 μm or less. Diving arterioles were measured in the frame mode at five to eight frames/s.

fMRI.

fMRI was performed on a 7T/21-cm BioSpec 70/30 USR horizontal bore scanner (Bruker) equipped with a B-GA 12S gradient set (Bruker) with a maximum strength of 50 G/cm and minimum rise time of 100 μs. A 10-mm-diameter surface RF coil was used to transmit and receive the radiofrequency signal. Magnetic field homogeneity was optimized using high-order shimming over the ROI. BOLD functional data were acquired using a single-shot gradient-echo echo planar imaging (EPI) pulse sequence with the following parameters: TE = 10 ms, flip angle = 30°, matrix = 80 × 80, slice thickness = 1 mm, TR = 1 s, five adjacent slices. The field of view was adjusted in each experiment to minimize ghosting artifacts and varied among subjects (Table S2). The images were collected in the coronal orientation. A quick functional scan was acquired at the start of the experiment and analyzed online to determine the center of the evoked BOLD response. The position of the stack was adjusted to allow the middle slices to cut through the center of activity. High-resolution rapid acquisition with relaxation enhancement (RARE) T1-weighted anatomical images (matrix size 200 × 200) were obtained at the same slice to identify brain structures.

EPI images were loaded in Matlab and spatially resampled at 256 rows using a cubic spline interpolation. The stimulus onset in our fMRI experiments was uniformly jittered relative to the MRI sampling intervals (TRs). Thus, in different stimulus trials the MRI images were acquired at different time-points relative to the stimulus onset, enabling estimation of the BOLD time-course with sub-TR temporal resolution. Different voxels (depths) with a given slice experienced the same jitter. For each trial, we used a cubic spline interpolation procedure to resample the data to a 200-ms step. After the resampling, the trails were averaged, and time-courses were extracted from the defined ROIs. This procedure resulted in an effective temporal resolution of 200 ms, higher than the TR for individual trials. The exact timing of each slice acquisition was available from the auxiliary timing file recorded by a separate computer. Timing differences between slices were corrected. Ratio images were calculated relative to the prestimulus baseline (an average of 4 s prestimulus). Multiple trials were averaged per subject (Table S2). A cortical ROI was defined as follows: We traced the cortical surface along the dorsal curvature of the brain surface from the center of the evoked activity ±2 mm on the cortical surface. The curvature was divided into eight equal segments. At the border of each segment a 1.2-mm line was drawn perpendicular to the surface throughout the cortical depth. The deep ends were connected, creating eight sectors. The depth then was divided into six equal 200-μm segments, creating an 8 × 6 grid. Time-courses were extracted from individual cells in the grid as the mean of all pixels within the cell in a temporal image series. The spatial resampling procedure had no significant effect on the signal time-courses extracted across depths (Fig. S7).

The use of a small surface transmit-and-receive RF coil in conjunction with rapid TR can introduce T1-weighted inflow effects into the BOLD signal (41–43). The contribution from inflow effects is expected to be relatively small for EPI gradient-echo sequences (41). This effect is reduced further at high field with reduction of T2* (44).