Activation resembling ocular dominance or orientation columns has been mapped with high-resolution functional magnetic resonance imaging (fMRI). However, the neuronal interpretation of these functional maps is unclear because of the poor sensitivity of fMRI, unknown point spread function (PSF), and lack of comparison with independent techniques. Here we show that cerebral blood volume (CBV)-weighted fMRI with a blood plasma contrast agent (monocrystalline iron oxide nanoparticles), in combination with continuous temporally encoded stimulation, can map columnar neuronal activity in the cat primary visual cortex with high sensitivity, selectivity, and reproducibility. We examined hemodynamic response PSF by comparing these CBV-based signals with oxygen metabolism-based negative blood oxygenation level-dependent signals. A significant positive correlation exists between CBV- and metabolism-based iso-orientation maps, suggesting that the hemodynamic PSF is narrower than intercolumn distances. We also compared CBV-based fMRI with optical intrinsic signal (OIS) imaging, a technique that identifies sites of increased neuronal activity, to investigate neuronal correlation. Iso-orientation maps obtained by fMRI and OIS were well matched, indicating that areas of the highest orientation-selective CBV signals correspond to sites of increased neural activity. Using CBV-based fMRI, we successfully mapped orientation-selective functional architecture in the medial bank of the visual cortex, an area inaccessible to OIS imaging. Thus, we conclude that contrast agent-based fMRI, in combination with continuous temporally encoded stimulation, is a highly sensitive technique capable of mapping neural activity at the resolution of functional columns without depth limitation.
Activation patterns resembling ocular dominance and orientation columns have been demonstrated with high-resolution functional magnetic resonance imaging (fMRI) in human and cat primary visual cortex in response to monocular eye stimulation (Menon et al., 1997; Menon and Goodyear, 1999; Dechent and Frahm, 2000; Cheng et al., 2001; Goodyear and Menon, 2001) and orientation-selective stimulation (Kim et al., 2000; Duong et al., 2001; Zhao et al., 2005). However, the utility of high-resolution fMRI is still problematic because of its poor sensitivity, unknown point spread function (PSF), and a lack of confirmation by independent techniques.
Although an increase in fMRI signal is closely correlated with an increase in neuronal activity on a large supramillimeter scale (Heeger et al., 2000; Ress et al., 2000; Logothetis et al., 2001), this finding may not be applicable on a submillimeter columnar scale. When a single cortical column is activated (Kennerley et al., 2005), the highest fMRI signal will appear at the site of increased neuronal activity (referred to as an “active column”) regardless of the PSF (see Fig. 1A). However, if multiple columns are activated, successfully determining the location of active columns within fMRI maps will be dictated by the PSF width relative to the distance between neighboring active columns (see Fig. 1B). Because of this limitation, it is conceivable that the highest signals induced by bars of 0° (90°) orientation may appear at the location of 90° (0°) orientation columns. Therefore, it is absolutely imperative to confirm our interpretation of submillimeter-resolution fMRI columnar maps by comparing them with conventional techniques such as optical intrinsic signal (OIS) imaging, 2-deoxyglucose (2-DG) imaging, or multiple-unit recording. Of these, OIS imaging is the best choice. OIS maps can serve as a reference to a spatial pattern of neuronal activity; a good correlation between spiking activity and OIS has already been observed (Grinvald et al., 1986; Maldonado et al., 1997; Bosking et al., 2002) (see also Blasdel and Salama, 1986). In contrast, the 2-DG technique cannot be performed with multiple stimulus configurations, and multiple-unit recording lacks sufficient field of view because of the point-by-point measurement.
In the present studies, we performed high-resolution fMRI at the resolution of orientation columns in the visual area of isoflurane-anesthetized cats. To achieve high specificity and sensitivity for orientation column mapping, we adopted two techniques: the injection of a blood plasma contrast agent [monocrystalline iron oxide nanoparticles (MION)] for cerebral blood volume (CBV) weighting (Kennan et al., 1998; Mandeville et al., 1998, 2004; van Bruggen et al., 1998; Zhao et al., 2006) and a paradigm of temporal encoding in conjunction with continuous stimulation (Engel et al., 1994; Sereno et al., 1995; Kalatsky and Stryker, 2003). Three major issues were examined: (1) reproducibility and spatial specificity of the orientation-selective fMRI signals, (2) PSF of the hemodynamic fMRI signal, which was investigated by comparing iso-orientation maps obtained from CBV-weighted fMRI with cerebral metabolic rate of oxygen (CMRO2)-based fMRI maps, and (3) the neural interpretation of iso-orientation fMRI maps, which was tested by correlation with OIS maps.
Materials and Methods
With approval of the Institutional Animal Care and Use Committee at the University of Pittsburgh, a total of 13 cats (1.2–2.3 kg, 9–23 weeks old) was studied by fMRI; each cat was used for multiple experiments (Table 1), and six of these cats were also studied by optical imaging for comparison. Each cat was initially treated with atropine sulfate (0.05 mg/kg, i.m.). Anesthesia was induced by a mixture of ketamine (10–20 mg/kg, i.m.) and xylazine (1 mg/kg, i.m.); a mixture of air and oxygen maintaining 30–35% O2 was used with 2% isoflurane during surgery and with either 0.8–1.2% isoflurane (n = 11) or α-chloralose (n = 2; 30 mg/kg, i.v.; maintenance, 5 mg/kg, i.v.) during fMRI and OIS studies, except for the comparison study between the CMRO2-based fMRI and CBV-weighted fMRI. In this experiment, a mixture of 30–35% O2 and 65–70% N2O was used with 0.8–1.2% isoflurane. The cat was paralyzed with pancuronium bromide (0.2 mg · kg−1 · hr−1, i.v.), and appropriate contact lenses were used. For MRI studies, the cat was placed in a cradle and secured in a normal postural position by a custom-designed head frame. The cranium was exposed for placement of a surface coil. After completion of MRI experiments, the inner contour of the coil was traced on the cranium as a spatial reference before removal. The cat was then moved to the room adjacent to the MR system for the purpose of optical imaging and was secured in a stereotaxic apparatus. A cranial window was made over the marked region, the dura was incised and reflected, and the exposed cortical surface was covered by agarose (∼1% in saline). Throughout the experiments, rectal temperature, end-tidal CO2, and mean arterial blood pressure were maintained within a normal physiological range [38.1 ± 0.2°C, 3.5 ± 0.1%, and 99.6 ± 12.1 mmHg (n = 13), respectively].
Visual responses were induced by binocularly presenting 100% contrast square-wave gratings, which were generated using either software from Cogent Graphics (John Romaya, Wellcome Department of Imaging Neuroscience, London, UK; http://www.vislab.ucl.ac.uk/Cogent/index.html) or VSG2/5 (Cambridge Research Systems, Kent, UK). Spatial frequency of the gratings was 0.1 cycle/° (0.3 cycle/° was used in comparison studies when noted). The gratings drifted at 2 cycles/s, with the direction of motion reversed every 0.5 s. Stimulation was synchronized with image acquisition in fMRI and OIS experiments. An identical visual stimulation paradigm was used in both fMRI and OIS measurements. The 80 s cyclical continuous stimulation paradigm was used in most studies, during which eight orientations were presented [0° (horizontal) to 157.5°, 22.5° increments, 10 s each] without gaps between stimuli. In two cats (Table 1), the 80 s cyclical intermittent stimulation paradigm (similar to block-design stimulation presentation) was additionally used, in which four orientations [0 (horizontal), 45, 90, and 135°, 10 s each] were alternated with 10 s homogeneous gray. In both stimulation paradigms, the duration of one full stimulation cycle was 80 s, and it was repeated 10 times for a total presentation time of 800 s (i.e., one scan).
All MRI experiments were performed on a 9.4 tesla horizontal magnet with a clear boar size of 31 cm (Varian, Palo Alto, CA) using a surface radio frequency (RF) coil [inner diameter, 2.5 cm (n = 9) or 1.6 cm (n = 4)] placed over the primary visual cortex. The position of the functional imaging slice was determined based on a flow-compensated, gradient-recalled, three-dimensional (3-D) venographic image [repetition time (TR), 50 ms; echo time (TE), 20 ms; data matrix, 512 × 256 × 256; field of view (FOV), ∼4.0 × 2.0 × 2.0 cm (dependent on brain size); isotropic resolution, ∼78 μm] (Park and Kim, 2005). Based on vascular patterns and gyral curvatures from the 3-D venogram, a location of the 1-mm-thick fMRI slice was chosen in the visual cortex in which vessels were perpendicular to the slice. For direct comparison, a two-dimensional (2-D) gradient-recalled anatomic image was also acquired (TR, 50 ms; TE, 15–20 ms; data matrix, 256 × 256; a typical FOV, 2 × 2 cm) with the same FOV, slice thickness, and position as fMRI to avoid a potential spatial misregistration resulting from the use of an anatomical image reconstructed from 3-D venogram.
For CBV-weighted fMRI, a bolus of dextran-coated MION contrast agent (typically 10 mg Fe/kg body weight; obtained from the laboratory of Dr. Ralph Weissleder, Massachusetts General Hospital, Boston, MA) was injected into either the femoral or cephalic vein. The intravascular injection of MION contrast agent enhances the sensitivity of fMRI in microvessels and induces a high susceptibility effect in large vessels, which reduces large vessel contributions (Kennan et al., 1998; Mandeville et al., 1998; van Bruggen et al., 1998). In our previous studies with similar imaging parameters at 9.4 T, the use of 10 mg Fe/kg MION improved the fMRI sensitivity in the cat visual cortex (Zhao et al., 2006). MION has a long intravascular half-life and distributes throughout the blood plasma in the cortical vessels; fMRI signal with iron oxide is mainly weighted by changes in plasma CBV in microvessels (Kennan et al., 1998; Mandeville et al., 1998; van Bruggen et al., 1998). CMRO2-based fMRI (without contrast agent) was obtained by reducing stimulus-induced cerebral blood flow (CBF) and CBV with the aid of intravenous infusion of a vasodilator, sodium nitroprusside (sNP) (0.36–2.81 mg/kg for one scan) (Fukuda et al., 2006; Nagaoka et al., 2006). All functional images were obtained during 800-s-long visual stimulation using a gradient-echo four-shot echo planar imaging (EPI) sequence; typical imaging parameters were as follows: FOV, 2 × 2 cm; data matrix, 128 × 128 (i.e., in-plane resolution of 156 × 156 μm/pixel), slice thickness, 1 mm; TR, 2 s; TE, 10 ms for CBV-weighted fMRI. The CMRO2-based fMRI columnar mapping was performed on four cats before MION injection with exactly the same imaging parameters used for the CBV-weighted fMRI studies, except TE was 18 ms.
After the fMRI studies, OIS imaging was performed on a custom-built imaging system (Moon et al., 2004). The camera (FOV, 1.9 × 1.4 cm; 29 × 29 μm/pixel) was positioned above the MR imaging area so that the pial vessel pattern closely matched that of the MR image. The focus plane was 0–800 μm below the cortical surface. The wavelength used was 570 ± 10 nm [an isosbestic point of oxyhemoglobin and deoxyhemoglobin (dHb)]. Images were acquired with temporal resolution of 1/30 s.
All data analysis was performed using custom-made programs with visual C++ (Microsoft, Seattle, WA) and Matlab (MathWorks, Natick, MA). Differences attributable to anesthetics, stimulation paradigm, RF coil, and optical imaging focus plane essentially could not be observed in the results. Thus, we pooled all data together.
To obtain a spatial resolution of fMRI similar to OIS maps, all of the fMRI data were interpolated by zero filling in k-space from 128 × 128 matrix to 512 × 512 matrix, resulting in a typical nominal in-plane resolution of 39 × 39 μm/pixel. Before zero filling, the high-frequency components in k-space data were apodized to avoid the ringing artifact that resulted from subsequent zero filling. These procedures did not change the activation patterns (see Fig. 4E). To match the temporal resolution of OIS to fMRI data, we averaged each of 60 consecutive OIS frame images during 800 s recording, resulting in a temporal resolution of 2 s/image.
Generation of temporally encoded iso-orientation maps.
To generate temporally encoded maps, Fourier analysis was applied to signals continuously recorded for 800 s (10 stimulation cycles) on a pixel-by-pixel basis (Engel et al., 1997; Kalatsky and Stryker, 2003). The signal was modeled using sinusoidal functions with multiple-frequency components, with magnitude and phase described by where S(x,y,t) is the signal intensity at position x and y at time t, ai(x,y) is the magnitude of the ith component, fi is the ith frequency component, and ϕi(x,y) is the phase of the ith component at the position x, y. To obtain an “activation map” at time t, S(x,y,t) was calculated from the magnitude and the phase of a frequency of the ith component. We refer to the activation map at the orientation-specific frequency as the iso-orientation map (Fig. 1). The temporally encoded iso-orientation map obtained from the sinusoidal curve fitting is essentially “differential” map. Both fMRI and OIS iso-orientation maps were determined by correcting for delays related to hemodynamic responses (see Results). The correlation of the fMRI time course with a reference waveform could be calculated (Boynton et al., 1996; Engel et al., 1997) to set a statistical threshold. Correlation value normalized by a reference waveform (or statistical value) is not linearly related to the magnitude of signal changes. Because it is necessary to use absolute signal changes for determining percentage change, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and reproducibility, all functional images were shown without any statistical threshold.
Quantitative analysis of iso-orientation maps.
Several quantitative analyses were performed within a region of interest (ROI): percentage signal change, SNR, CNR, intercolumnar distance, reproducibility, stimulus spatial frequency dependency, linear correlation, overlap, and density of singularities. To calculate SNR, magnitude at the orientation-specific frequency was divided by SD of magnitudes across frequencies above the orientation-specific frequency (Warnking et al., 2002). CNR is defined as the ratio of SNR within the activation area to that within the non-active area. Because the ROI of the activation area contains singularities, our CNR may be underestimated. Intercolumnar distance in the iso-orientation maps was determined within a 5 × 5 mm activation area by autocorrelation analysis (Duong et al., 2001; Fukuda et al., 2005). To compare images obtained from different modalities, global subtraction was performed on all fMRI data to remove a slow signal fluctuation, and spatial filters (Table 1) were applied to remove spatial inhomogeneity. Properties of the filter were determined by the difference of two Gaussian functions where σ1 and σ2 determine low-pass frequency and high-pass frequency, respectively. Singularity was visually determined in reference to magnitude and phase maps and counted within the maps of 5 × 5 mm.
Data are presented as mean ± 1 SD, and statistical significances (p < 0.05) were evaluated by Student's t test and multiple comparisons of Fisher's protected least significant difference after ANOVA. p value of correlation coefficient was evaluated by the Student's t test after Bonferroni's correction.
Slice selection and image coregistration based on intracortical veins
Intracortical veins were non-invasively visualized for fMRI slice selection and for coregistration of anatomical and functional images acquired by different MRI methods. Anatomical landmarks (Fig. 2A) were determined with a 3-D vessel-weighted imaging technique (venogram, ∼78 μm isotropic resolution). In a coronal 2-D reconstruction of the dataset (Fig. 2B), most intracortical emerging veins appear as dark lines perpendicular to the pial surface. This pattern of intracortical vessels was used to select a tangential imaging slice along the marginal gyrus, which contains a relatively large flat section of primary visual cortex. In a tangential 2-D reconstruction of the same dataset (Fig. 2C), intracortical veins appear as dark spots, especially within the ROI (a 5.8 × 6.5 mm red box in Fig. 2C), which is the area we used for determining correspondence between fMRI and OIS maps. A single-slice 2-D anatomic image (Fig. 2D) was also acquired with the same FOV, slice thickness, and position as fMRI for coregistration between fMRI and anatomic images. A single image from the EPI dataset acquired for fMRI (Fig. 2E) again shows intracortical veins as dark spots (see ROI). Visual comparisons of vessel-weighted anatomical reference images with EPI data showed that the center positions of dark spots did not shift by >150 μm, which is equivalent to the size of one pixel before zero padding. Thus, we concluded that good coregistration was achieved within the ROI (Fig. 2F–H).
Temporally encoded iso-orientation CBV-weighted fMRI maps
To efficiently map iso-orientation columns with fMRI, we used MION as a contrast agent and adopted a temporally encoded method in which data are collected during continually cycled stimulation (Engel et al., 1997; Kalatsky and Stryker, 2003); the response to the orientation-specific stimulation cycle can be decomposed from the other frequency components by Fourier analysis. Because CBV-weighted fMRI signals were obtained during 80 s cyclical stimulation, the peak of the fMRI response specific to a particular orientation can therefore be observed every 80 s (Fig. 3A); a power spectrum of CBV-weighted fMRI signal (Fig. 3B) showed a remarkable peak at the orientation-specific stimulation frequency [1/(80 s) = 0.0125 Hz]. In contrast, the response at the frequency corresponding to each orientation stimulation cycle [1/(10 s) = 0.1 Hz, orientation-nonspecific stimulation frequency] was small because continuous stimulation saturates the orientation-nonspecific response and eventually attenuates the power at the orientation-nonspecific frequency. To calculate the percentage change of the orientation-specific frequency component for CBV-weighted fMRI, we divided the magnitude at the orientation-specific frequency by the sum of magnitudes at direct current and orientation-nonspecific frequencies. The average CBV-weighted orientation-specific signal change in a single 800 s scan was 0.62 ± 0.23% (n = 5 cats) in marginal gyrus, which is consistent with values from a previous report [∼0.7% in block design stimulation; note that orientation-nonspecific CBV-weighted fMRI response is significantly larger, ∼2% (Zhao et al., 2005)]. Peaks also appear at several other frequencies (Fig. 3B); components less than ∼0.01 Hz (e.g., arrowhead 1) result from slow baseline drift, whereas those between ∼0.02 and 0.1 Hz (e.g., arrowheads 3, 4) are related to vasomotion (Mayhew et al., 1996; Obrig et al., 2000). The frequency of respiration (∼0.3 Hz) and cardiac pulsation (∼3 Hz) were out of range for this spectrum because of the 2 s sampling rate. However, they were aliased into the spectrum (e.g., arrowheads 6, 7). Activation maps at these frequencies were computed using Equation 1 at the 13 s time point in the 80 s stimulation cycle (see Materials and Methods). Orientation column-like patterns are seen only at the orientation-specific frequency (Fig. 3C). We will refer to activation maps at the orientation-specific frequency as iso-orientation maps.
Magnitude (Fig. 4A) and phase (Fig. 4B) maps at the orientation-specific frequency encode orientation tuning strength and orientation preference for each pixel, respectively. Hemodynamic response is always delayed from the stimulation onset time, inducing a shift in phase values (referred to as hemodynamic response time). To accurately assign orientation preference to phase maps, the hemodynamic response time must be determined. The hemodynamic response time can be determined using two different orientation presentation orders (Kalatsky and Stryker, 2003): one is the “forward” direction starting at 0° and increasing in orientation by 22.5°; the other is the “backward” direction starting at 157.5° and decreasing in orientation by 22.5° (Fig. 4C schematic). The hemodynamic response time is the same in the forward and time-reversed backward stimulation data, but its direction is opposite; therefore, the effect of the hemodynamic response time can be removed from the phase maps. By averaging phase maps of these forward and backward data (Fig. 4C, left and middle, respectively), an absolute phase map (right), in which the 0° stimulus presentation corresponds to a time of 0 s (i.e., no delay), can be generated. The difference between absolute phase (Fig. 4C, right) and phase in forward or time-reversed backward data (Fig. 4C, left or middle) indicate the hemodynamic response time. To determine the distribution of hemodynamic response times, the forward phase map was compared with the absolute map on a pixel-by-pixel basis (Fig. 4D). We found that the hemodynamic response time was typically ∼13 s in CBV-weighted fMRI with our continuous stimulation paradigm (Fig. 4D, indicated by a red arrow on vertical axis), which corresponds to the time-to-peak of hemodynamic response. Most pixels are distributed equally across the dotted yellow line, which has been shifted 13 s from the dashed diagonal, suggesting that hemodynamic response times are similar across pixels. Similar average hemodynamic response time and distribution were also observed in OIS data (data not shown). Because the hemodynamic response time is almost spatially homogeneous, the average hemodynamic response time across pixels can be used to assign stimulus orientation. This method was chosen for the current study and used to correct the hemodynamic response times for all fMRI and optical imaging signals (Table 1). In most studies, 13 s was used as the hemodynamic response time; the iso-orientation map of 0° stimulus orientation presented between 0 and 10 s was obtained from the 0.0125 Hz signals at the time of 13 s in the 80 s stimulation cycle; that of the 22.5° stimulus orientation presented between 10 and 20 s was obtained from the 0.0125 Hz signals at 23 s, etc. Note that errors in hemodynamic response time will lead to errors in orientation assignment; however, 4 s error (i.e., ± TR) will result in only (4 s/80 s cycle) × 180° = 9° error, which is relatively small compared with our orientation resolution of 22.5°.
After correcting for the 13 s hemodynamic response time, we then calculated the iso-orientation map of 0° stimulus presented between 0 and 10 s from the magnitude (A) and corrected phase (B) maps (Fig. 4E, right and left, before and after interpolation, respectively). Interpolation of raw 128 × 128 matrix data into 512 × 512 matrix did not change the appearance of functional maps. Prominent patchy and band-like activation patterns were only seen in marginal gyrus. Black patches in the 0° activation maps indicate a decrease in CBV-weighted fMRI signals and thus an increase in CBV changes during 0° stimulation. The average interval between the black patches or bands in the marginal gyrus was 1.2–1.6 mm (1.4 ± 0.2; n = 5 cats), which is similar to the distance between iso-orientation columns (1.2–1.4 mm) reported in a 2-DG study (Lowel et al., 1987). These observations would indicate that these patches and bands likely represent iso-orientation columns.
Reproducibility and selectivity of CBV-weighted fMRI maps
To determine the reproducibility of activation patterns obtained from the continuous stimulation paradigm, we compared the iso-orientation maps obtained from five different 800 s CBV-weighted fMRI scans. CNR (see Material and Methods) of the orientation-specific CBV-weighted fMRI signal was 2.1 ± 0.6 (n = 5 cats) in a single scan, which is sufficiently high to perform scan-by-scan comparisons. Activation patterns corresponding to 0° stimulus orientation were consistent across repeated scans (Fig. 5A). Pixelwise correlation analysis (Fig. 5B) showed that signal changes within the primary visual cortex (solid white “activation area” rectangle in Fig. 5A) are highly reproducible (r = 0.83 for first vs third scan and 0.83 for first vs fifth scan; p < 0.001). In comparison, the correlation coefficient within a brain region outside of the primary visual cortex (dotted white “non-active area” rectangle in Fig. 5A) is close to 0 (r = 0.06 for first vs third scan and 0.10 for first vs fifth scan). These observations were consistent for all eight orientations. The average pixelwise correlation coefficient for all eight orientations within ROI in the activation area (e.g., r = 0.75 ± 0.10 at first vs second scans; n = 5 cats) was significantly higher (ANOVA, F(1,32) = 641.430; p = 9.67 × 10−23) than those in the non-activation area (e.g., r = 0.06 ± 0.06 at first vs second scans; n = 5 cats) across five scans (Fig. 5C). The correlation coefficients were not significantly different across five scans (ANOVA, F(3,12) = 2.207, p = 0.140 for activation area; F(3,12) = 2.093, p = 0.155 for non-active area), suggesting that a similar activation pattern is observed for ∼67 min (i.e., 800 s/scan × 5 scans).
To confirm that the observed activation patterns are in fact within area 18, we obtained CBV-weighted fMRI responses to both low (0.1 cycle/°) and high (0.3 cycle/°) spatial frequency stimulation (Movshon et al., 1978) (Fig. 6A). Iso-orientation maps obtained from one 800 s scan for all eight stimulation orientations show a smooth change in activation pattern with a change in orientation (top row), and the patterns are complementary at half the orientation-specific stimulation cycle [e.g., compare 0° (13 s) and 90° (53 s) maps]. Most orientation-specific signals in marginal gyrus with low-spatial-frequency stimulation (top row) were not present with high-spatial-frequency stimulation (bottom row), which verifies that these signals originate from area 18. In contrast, activation in the posteromedial part of the marginal gyrus was stronger during high-spatial-frequency stimulation, suggesting an association with area 17. It should be noted that the area 17 in this slice is perpendicular to the cortical surface, so the assignment of stimulus orientation is distorted because of contributions of neighboring columns in the 1 mm slice thickness; thus, typical black and white patterns are not represented.
The distinct response to low and high-spatial-frequency stimulation was much clearer in the magnitude maps (Fig. 6B). To quantitatively determine how spatial frequency stimulation affected the magnitude of signal change, we compared histograms of magnitude within the 2.1 × 2.1 mm ROI shown in Figure 6B (red and green boxes). As shown in Figure 6C, in area 18, low-spatial-frequency stimulation induced higher signal changes than high-frequency stimulation (compare red solid line vs red dotted line), whereas in area 17, high-frequency stimulation induced higher signal changes than low-frequency stimulation (compare green dotted line vs green solid line). This was consistently observed in all five cats tested. In area 18, the average cumulative curve obtained with low-spatial-frequency stimulation was significantly shifted to the right (ANOVA, F(1,98) = 11.849; p = 0.001) compared with the curve from high-frequency stimulation in the same area (Fig. 6C, inset, D); this demonstrates that the low-spatial-frequency stimulation induced higher magnitude changes. In area 17, the average cumulative curve obtained with high-spatial-frequency stimulation was not statistically different (ANOVA, F(1,58) = 2.566; p = 1.115) from the curve with low-frequency stimulation (Fig. 6D, inset). These results indicate that the activation patterns observed in the low-spatial-frequency study are primarily attributed to area 18 of visual cortex (Movshon et al., 1978).
CBV-weighted versus CMRO2-based fMRI iso-orientation maps
Our CBV-weighted iso-orientation fMRI maps clearly show reproducible and selective activation patterns. To examine the PSF of hemodynamic response, we compared CBV-weighted iso-orientation fMRI maps with CMRO2-based fMRI iso-orientation maps. Metabolism-based fMRI signals (Kim et al., 2000) can be extracted from blood oxygenation level-dependent (BOLD) response by the intravenous infusion of sNP (Nagaoka et al., 2006), which suppresses evoked hemodynamic (CBF and CBV) responses without changing evoked spiking activity (Fukuda et al., 2006). As seen from the images (Fig. 7A) (for another example, see supplemental Fig. 1, available at www.jneurosci.org as supplemental material), most patches with higher CMRO2 changes have higher CBV responses; iso-orientation maps of the CMRO2-based fMRI signal were significantly correlated to those of CBV-weighted fMRI (r = 0.71, p < 0.001 in Fig. 7B; r = 0.66 ± 0.10, n = 4 cats in Fig. 7C). Some mismatch is likely attributable to poor CNR of the single-scan CMRO2-based fMRI signal (1.4 ± 0.3), which is significantly smaller (p = 0.0097, two-tail paired t test; n = 4 cats) than CNR of the single-scan CBV-weighted fMRI (1.9 ± 0.3). This is also supported by average correlation coefficient between first versus second scans, which yields 0.32 ± 0.26 (n = 4) for CMRO2-based fMRI and 0.68 ± 0.13 for CBV-weighted fMRI. To increase the CNR of fMRI signals, all repeated scans were averaged (Fig. 7C). The significant positive correlation between CBV-weighted and CMRO2-based fMRI maps suggests that (1) the PSF of CBV response is narrower than the intercolumn distance as discussed in Figure 1B, and (2) active columns in CBV-weighted fMRI maps likely indicate sites of increased neuronal activity because oxidative metabolism is known to be highly correlated with neuronal activity (Thompson et al., 2003).
CBV-weighted fMRI versus optical imaging iso-orientation maps
To more convincingly determine whether the observed bands and patches in CBV-weighted fMRI maps mark the actual sites of increased neuronal activity, we recorded OIS data after we collected CBV-weighted fMRI data (n = 6 cats). We compared the plasma-based CBV-weighted fMRI signal with CBV-weighted 570 nm OIS, which indicates a change in total hemoglobin, or red blood cell (RBC) counts, thus addressing a change in total blood volume at the resolution of functional columns. Previously, neural correlates to OIS in visual cortex have only been demonstrated with dHb-weighted wavelengths (Grinvald et al., 1986; Shmuel and Grinvald, 1996; Maldonado et al., 1997; Bosking et al., 2002). However, good agreement between iso-orientation maps of 570 nm OIS and dHb-weighted OIS (Fukuda et al., 2005) (supplemental Fig. 2, available at www.jneurosci.org as supplemental material) suggests that 570 nm OIS is also a good neural correlate.
Functional MRI and optical images were coregistered using pial vessel patterns; because the slice chosen for fMRI studies was below the cortical surface (Fig. 8A, red rectangle), the pial vessel pattern above the fMRI slice (Fig. 8A, green rectangle) was visualized using a 2-D MR image reconstructed from the 3-D venogram (Fig. 8B). Then, the optical image (Fig. 8C) was linearly transformed manually so that pial vessel patterns matched the MRI patterns within the ROI (Fig. 8B,C, green boxes). Based on visual inspection after coregistration, we estimate that any remaining mismatch between pial vessel locations is ≤200 μm within the 5.8 × 6.5 mm ROI (Fig. 8, compare D, E); this mismatch may be attributable to a slight difference in imaging plane orientations between MRI and optical imaging. Taking into account the possible coregistration error between MR anatomical images and fMRI data, the overall coregistration error between fMRI maps and OIS maps is ≤350 μm. Because the width of iso-orientation columns (i.e., half of the intercolumn distance) is 700 μm, criterion for a good congruence between fMRI and OIS columns will therefore appear as an overlap of >350 μm (>50%).
Iso-orientation maps of CBV-weighted fMRI were similar to those of 570 nm OIS in all six cats tested (Fig. 8F) (for another example, see supplemental Fig. 3, available at www.jneurosci.org as supplemental material); activation borders, columnar patterns (black and white patches and bands), and most loci show a good match between the two modalities. To quantify the results, we converted the iso-orientation maps of CBV-weighted fMRI and 570 nm OIS into binary images. Assuming minimal distortion of columnar shape in regions in which the image is perpendicular to intracortical emerging vessels, an ROI was appropriately chosen from the vessel-weighted MR anatomical image (Fig. 8B, inset, yellow rectangle) (for sizes of ROIs, see Table 1). Because of the differences in pixel resolution and signal magnitude between fMRI and OIS maps, pixelwise correlation analysis was not performed. Instead, the percentage overlap was calculated. Pixels with a value smaller than the mean value within this ROI (Fig. 8F, yellow rectangle) were assigned as active for a given orientation, whereas others were assigned as inactive for the same orientation. Then, we calculated the percentage of overlap for both active and inactive pixels between the OIS and fMRI binary maps from all eight orientations. There were significant differences for the percentage of overlap (ANOVA, F(4,20) = 147.32; p = 1.55 × 10−14) across the stimulation angle differences from 0° orientation. The highest percentage of overlap was found when the stimulus orientation was the same between OIS and fMRI maps (Fig. 8G) (72.5 ± 4.7%; n = 6 cats at 0° angle difference; p = 2.6 × 10−4 to 8.00 × 10−5 compared with other angle differences). Because this value meets our criterion for good congruence (>50%), we conclude that CBV-weighted fMRI iso-orientation maps agree with OIS maps. Thus, the highest CBV-weighted fMRI signal change is expected at the site of increased neuronal activity.
Mapping orientation columns inside cortical sulcus
We demonstrated that MION-based CBV-weighted fMRI techniques can reliably map iso-orientation columns on dorsal cortical surface regions. Thus, with this highly sensitive MION-based CBV-weighted fMRI, functional columns can be mapped in cortical regions deeply embedded in sulci to which OIS imaging cannot be applied. We attempted to record orientation-specific responses on the medial bank of cortical area 17 (suprasplenial and splenial gyri) in two cats, an area that is inaccessible with OIS imaging. We successfully mapped iso-orientation columns in this area (Fig. 9A). The activation pattern was highly reproducible; the pixelwise correlation coefficient between two successive scans within a region of activity was 0.89 and 0.66 (p < 0.001) at 0° iso-orientation maps for two cats. Spatially distinct activation patterns on a sagittal imaging slice appear as irregular patches and bands (Fig. 9A). Patchy patterns appear in the region between dashed lines 2 and 3, in which the imaging plane is parallel to the cortical surface and in which the vessel-weighted anatomical image shows a high density of tangential intracortical veins. Band-shaped patterns appear in regions in which the imaging plane is perpendicular to the cortical surface and in which intracortical veins run parallel to the slice orientation (between dashed lines 1 and 2 and between lines 3 and 4). The arrangement of iso-orientation columns for all orientations (0–180°) was represented in the orientation preference map by color coding. Assignment of stimulus orientation in the regions perpendicular to the cortical surface (between dashed lines 1 and 2 and between 3 and 4) was distorted because of contributions of neighboring columns in the 1 mm slice thickness. Similarly, the assignment of stimulus orientation was inaccurate in the most anterior and posterior regions because of low signal sensitivity. Detail of the orientation column structure is seen in a region selected between dashed lines 2 and 3 (rectangle in the magnitude map), an area in which signal sensitivity is high and cortical vessels run perpendicular to the slice. Consistent with findings by OIS imaging from the exposed cortical surface (marginal gyrus), the orientation preferences change abruptly at singularities (Bonhoeffer and Grinvald, 1993) and at fracture-like structures (Blasdel, 1992; Rao et al., 1997) (Fig. 9B), which appear as regions of low signal intensity in the magnitude map (dark spots and lines). There were 2.0–2.5 singularities/mm2 in the medial bank of area 17 (n = 2), which is higher than the density of singularities (1.3–1.8/mm2; n = 5 cats) in the dorsal part of area 18. These values are similar to results obtained from OIS studies in the dorsal region of area 17 (2.1–2.4/mm2) (Bonhoeffer et al., 1995; Rao et al., 1997) and area 18 (1.2–2.1/mm2) (Bonhoeffer and Grinvald, 1993; Womelsdorf et al., 2001). Our results further suggest that fMRI can be used to map submillimeter-scale functional structures without depth limitation.
Our major finding is that CBV-based fMRI iso-orientation maps are similar to CMRO2-based fMRI and OIS maps on a columnar level; even when multiple columns are simultaneously activated, the highest signal change in CBV-weighted fMRI maps generally corresponds to the highest signal change in CMRO2-based fMRI and OIS maps. This leads us to infer that most sites of the highest orientation-selective CBV signals correspond to the sites of increased neuronal activity. Thus, the PSF of CBV response is sufficiently narrow to resolve iso-orientation columns (Fig. 1B, schematic profile of 1); the highest CBV changes induced by bars of 0° (90°) orientation are likely to indicate 0° (90°) orientation columns. However, this finding cannot simply be applied to neural interpretation of BOLD columnar maps because the BOLD signal is directly related to a mismatch between CBF and CMRO2 responses. Even if CBV (or CBF) and CMRO2 responses are localized to active columns (Fig. 7), the highest changes in BOLD fMRI signals may not indicate active columns (Duong et al., 2000; Cheng et al., 2001); thus, additional studies are needed.
Improvement of high-resolution fMRI specificity and sensitivity
Imaging techniques must have high specificity and sensitivity to be useful for neuroscience research. Because oxygen metabolism occurs at the site of increased neuronal activity (Thompson et al., 2003), imaging the metabolic change should yield high spatial specificity (Malonek and Grinvald, 1996). Thus, after the presentation of a stimulus, the early negative BOLD signal (referred to as an “early dip”) (Kim et al., 2000), which likely indicates an early oxygen consumption change (Nagaoka et al., 2006), should provide high specificity to active cortical columns. However, to detect this early response and use the high spatial specificity of the early dip signal, images must be acquired with high spatial and temporal resolution, which results in a significant reduction in sensitivity. Although this can be overcome using a vasodilator, which induces a CMRO2-based fMRI signal (Nagaoka et al., 2006), the vasodilator cannot be easily used for human studies. Therefore, it is currently difficult to achieve sufficient sensitivity for metabolic mapping of multiple orientation columns within a reasonable time frame without the assistance of vasodilator.
To improve the specificity and sensitivity of functional signals, we injected MION, a blood plasma contrast agent. Compared with the conventional BOLD fMRI technique, MION-based CBV fMRI has high sensitivity of functional signals (Kennan et al., 1998; Mandeville et al., 1998; van Bruggen et al., 1998) and high specificity to microvessels (Mandeville and Marota, 1999; Zhao et al., 2006). Sensitivity enhancement is dependent on many parameters, including magnetic field strength and dose of contrast agent (Kim and Ugurbil, 2003). Because an increased BOLD signal at higher magnetic fields counteracts the effect of MION, CBV-weighted fMRI at lower magnetic fields results in a greater improvement in sensitivity. However, even at 9.4 T, we found that functional sensitivity with 10 mg Fe/kg MION was improved by >1.6 times over the conventional gradient-echo BOLD fMRI in the cat visual cortex (Zhao et al., 2006). The high sensitivity of MION-based CBV fMRI leads to excellent reproducibility for small functional signals (Fig. 5). Thus, CBV-weighted fMRI is a superior alternative to conventional BOLD fMRI in brain research with animal models (Dubowitz et al., 2001; Vanduffel et al., 2001; Leite et al., 2002; Fize et al., 2003; Tsao et al., 2006); this technique, however, cannot be easily used in human studies because a large amount of exogenous contrast agent is required. Alternatively, non-invasive CBV-based fMRI, such as the vascular-space-occupancy technique (Lu et al., 2003, 2004), can be used if it has sufficient sensitivity.
To further improve the specificity and sensitivity of functional signals, we used the temporally encoded method with a gradual shift in orientations during continuous stimulation. This approach has been routinely used for fMRI retinotopic mapping (Engel et al., 1994; Sereno et al., 1995; Engel et al., 1997; Tootell et al., 1998), and it was recently applied to OIS imaging to generate orientation-preference maps (Kalatsky and Stryker, 2003). There are several advantages to continuous stimulation with temporal encoding. The fMRI response after the orientation-specific stimulation cycle can be easily separated from other frequency components induced by vasomotion, respiration, heart beat, etc., by means of frequency analysis. Also, the orientation nonspecific signal induced by each stimulus presentation is almost saturated by continuous stimulation (Fig. 3B, C, component 5), essentially similar to the removal of orientation nonspecific signal, which occurs with cocktail blank subtraction (Bonhoeffer and Grinvald, 1993). Finally, the signal responds to the orientation changes in a gradual manner, which is well fitted by our cosine function model. Thus, the Fourier analysis method with continuous stimulation (Kalatsky and Stryker, 2003) is more efficient to detect functional response compared with a conventional block-design paradigm (Kim et al., 2000; Duong et al., 2001; Zhao et al., 2005). However, determining hemodynamic response time is not so straightforward for data obtained from continuous stimulation. If a pixel is activated from multiple stimulations (e.g., 0 and 90°), it will be difficult to assign orientation preference (Zepeda et al., 2004). Furthermore, if the hemodynamic response time is not spatially homogeneous, determining pixelwise hemodynamic response time to assign orientation preference would require forward and backward stimulation (Fig. 4C) (Sereno et al., 1995; Warnking et al., 2002; Kalatsky and Stryker, 2003). In our studies, however, the hemodynamic response time is similar across pixels in area 18 (Fig. 4D), thus eliminating the need to perform the backward stimulation.
Neurovascular control at a columnar level
The congruence of CBV-weighted fMRI maps and 570 nm OIS maps suggests an increase in total blood volume at a columnar level. Because CBV-weighted fMRI with a contrast agent and 570 nm OIS techniques are sensitive to plasma volume (Kennan et al., 1998; Mandeville et al., 1998; van Bruggen et al., 1998) and RBC count (Frostig et al., 1990; Malonek et al., 1997; Sheth et al., 2004; Vanzetta et al., 2004; Suh et al., 2005; Fukuda et al., 2006), respectively, our data can be used to obtain an insight into neurovascular control on a columnar level. During orientation-selective stimulation, RBCs might redistribute across cortical columns. According to in vivo microscopic studies (Villringer et al., 1994; Hudetz, 1997), ∼10–20% of capillaries (i.e., microvessels of ∼5 μm diameter) carry only plasma during resting conditions, although all of these capillaries carry a mixture of RBCs and plasma during stimulation conditions. If total microvascular blood volume remains the same over the cortex (Wei et al., 1993), it would suggest a reduction of plasma volume in active columns and an increase of plasma volume in inactive columns. However, our CBV-weighted fMRI and OIS data show that both plasma volume and RBC count increase within the same iso-orientation columns. Similar colocalization of activity-dependent changes in plasma volume and RBC count is also reported in studies comparing OIS imaging at the wavelength of the hemoglobin isosbestic points and optical imaging of intravascular plasma tracers in monkey ocular dominance column (Frostig et al., 1990), rodent barrel field (Narayan et al., 1995), and cat orientation column (Fukuda et al., 2005; Vanzetta et al., 2005). Combining our data with those of others, it can be concluded that the major mechanism of columnar-specific CBV response is not a redistribution of “RBC to plasma ratio” across cortical columns but rather dilation of microvessels within active columns. This could be explained by recently demonstrated hemodynamic regulation mechanisms via the activity of astrocytes (Zonta et al., 2003; Takano et al., 2006) and pericytes (Howarth et al., 2005).
This work was supported by National Institutes of Health (NIH) Grants NS44589, EB003324, and EB003375 and the McKnight Foundation. The 9.4 T MRI system was funded in part by NIH Grant RR17239. We thank Sung-Hong Park for development of MR venography, Michelle Tasker for animal preparation, Dr. Amiram Grinvald for encouraging fMRI versus optical imaging experiments, and Dr. Dae-Shik Kim for insightful comments on a previous version of this manuscript. We are also grateful to Kristy Hendrich for critical reading and editing of our manuscript.
- Correspondence should be addressed to Dr. Seong-Gi Kim, Departments of Radiology and Neurobiology, University of Pittsburgh, 3025 East Carson Street, Pittsburgh, PA 15203.