Evaluation of activity-dependent functional pH and T_1ρ response in the visual cortex

Abstract

Recent experiments suggest that T₁ relaxation in the rotating frame (T_1ρ) detects localized metabolic changes in the human visual cortex induced by a flashing checkerboard task. Possible sources of the T_1ρ signal include pH, glucose, and glutamate concentrations as well as changes in cerebral blood volume. In this study we explored the relationship of the T_1ρ signal changes related to cerebral blood volume changes by employing inferior saturation pulses. Our hypothesis was that there would be a contribution of cerebral blood volume to the functional T_1ρ signal, but a majority of the signal would correspond to metabolic changes. In addition, the relationship between T_1ρ and pH was explored by manipulating the frequency of the flashing checkerboard and imaging with T_1ρ, BOLD, and ³¹P spectroscopy. We hypothesized that T_1ρ and pH changes would be sensitive to the stimulation frequency. To test this hypothesis, we used a full-field visual flashing checkerboard and varied the frequency between 1, 4, and 7 Hz. Supporting our hypotheses, we found that approximately 73% of the measured signal change corresponds to metabolism in vivo and that increasing stimulation frequency increased responses measured by all three imaging modalities. The activation area detected by T_1ρ overlapped to a large degree with that detected by BOLD, although the T_1ρ response area was significantly smaller. ³¹P spectroscopy detected a greater acidosis with the higher stimulation frequencies. These observations suggest that, similar to the BOLD response, the magnitude of the T_1ρ and pH response depends on stimulation frequency and is thus likely to be activity-dependent.

Introduction

While pH regulation is often assumed to be homeostatic, local dynamic pH fluctuations occur in the brain with neural activity. There are a number of potential source of acid in the brain. Carbohydrate metabolism induced by increased neural activity produces lactic acid, CO₂, and H⁺ as end-products. In addition, the synaptic release of protons from neurotransmitter vesicles generates a localized acidosis to modulate pH-sensitive channels and other receptors in the pre and post-synaptic membrane. Such transient local pH changes have the potential to dramatically alter pathophysiology and behavior through a number of pH-sensitive receptors in the brain. For example, acid-sensing ion channel-1a (ASIC1a) plays a critical roles in fear, anxiety-related behaviors (Coryell et al., 2007, 2008, 2009; Wemmie et al., 2002, 2003, 2004, 2006), and CNS diseases (Wemmie et al., 2002, 2006). In addition, significantly elevated baseline lactate concentrations have been found in bipolar disorder (Dager et al., 2004) as well as functionally elevated lactate responses in panic disorder (Maddock et al., 2009). These findings suggest that local or more global acidosis exists within these disorders possibly resulting from altered metabolic regulation.

The ability to non-invasively measure pH dynamics in vivo provides a unique insight into better understanding brain function as well as neurological and psychiatric disorders. Various magnetic resonance (MR) approaches have been developed to measure brain pH with pH-dependent chemical shift changes using MR spectroscopy (Alger et al., 1989; Allen et al., 1992; Bhujwalla et al., 1998; Gillies et al., 1994; Sappey-Marinier et al., 1992; van Sluis et al., 1999; Vermathen et al., 2000), pH-sensitive contrast agents (Raghunand et al., 2002, 2003), proton exchange properties (Aime et al., 2002; Ward and Balaban, 2000; Zhang et al., 2003; Zhou et al., 2003a,b), and by combining paramagnetic contrast agents with proton exchange (Liu et al., 2012; Longo et al., 2012; Takayama et al., 2012). ³¹P MR spectroscopy methods have been widely used to measure intracellular pH using the pH-dependent chemical shift between inorganic phosphate (Pi) and phosphocreatine (PCr) (Patel et al., 2000; Stubbs et al., 1992). The disadvantage of the ³¹P spectroscopic techniques is their limited spatial resolution (on the order of 10–30 cm³ volume), long acquisition times (on the order of 5–10 min) for a single measurement, and the requirement of special hardware that is typically not available on clinical scanners. Molecules with pH-sensitive ¹H resonance have also been reported (Garcia-Martin et al., 2001; Gasparovic et al., 1998; Pan et al., 1988; Rothman et al., 1997). 2-Imidazole-1-yl-3-ethoxycarbonylpropionic acid (IEPA) has been used to estimate pH, which has a pH-dependent chemical shift due to the protonation and deprotonation of the amine group. pH-dependent relaxation properties have also been shown using an injection of exogenous contrast agents (van Sluis et al., 1999; Vermathen et al., 2000). pH-sensitive gadolinium complexes have shown the possibility of measuring pH with a spatial resolution comparable to standard MRI (Raghunand et al., 2002, 2003). However, the enhancement observed on an image depends on the local concentration of the agent as well as pH, leading to difficulty in quantification of the environmental parameter of interest.

¹H based MR imaging pulse sequences have been used to probe pH changes in vivo using proton exchange properties (Magnotta et al., 2012; Ward and Balaban, 2000; Zhou et al., 2003a,b). These techniques include amide proton transfer (APT), a specific variant of chemical exchange saturation transfer (CEST), and T_1ρ. APT can probe the interaction between bulk water and amide protons in the peptide bonds to enable the assessment of microenvironment properties such as pH and temperature. It has been shown that the amide proton exchange rate is base-catalyzed for pH within the physiological range (Englander et al., 1972). This approach has been used to image pH changes in the ischemic rat brain (Zhou et al., 2003a) and brain tumors in human (Jones et al., 2006; Zhou et al., 2008). The T_1ρ signal reflects water-protein interactions in tissue, which has been shown to be pH sensitive. The sensitivity of T_1ρ to the surrounding pH has been well documented. Several studies have shown pH-dependency of T_1ρ in phantoms and animal models (Jin et al., 2011; Kettunen et al., 2001, 2002; Magnotta et al., 2012). In addition, T_1ρ positively correlates with lactate concentration, suggesting that pH modulates T_1ρ in ischemic tissue (Kettunen et al., 2001). Recently, we found that T_1ρ contrast can detect dynamic fluctuations in brain activity resulting from a visual flashing checkerboard (Magnotta et al., 2012). ³¹P measurement6s revealed a local acidosis in the visual cortex associated with brain activity that was associated with an increase in the T_1ρ relaxation time. In addition, a significant increase in the lactate signal using ¹H spectroscopy within the visual cortex was observed. The BOLD signal originates in the venous system, which results in spatial displacement of the signal from the actual site of brain activity. Therefore functional T_1ρ imaging may provide improved localization of brain activity since metabolic changes and lactic acid concentration changes are likely more localized to brain activity as compared to blood oxygenation.

In addition to pH, T_1ρ relaxation rates have been shown to be sensitive to the concentrations of other metabolites including glucose and glutamate (Jin and Kim, 2013). In addition, they found that the T_1ρ dispersion of these metabolites was similar to pH thus providing other possible sources that may contribute to the functional T_1ρ responses that have been observed. In this study we propose to further explore the functional T_1ρ signal dynamics in vivo. It has been previously suggested that the functional T_1ρ signal may reflect changes in blood volume and not metabolism. Jin and Kim eliminated the blood volume compartment in animals through the administration of superparamagnetic iron oxide particles (MION) in cats. For human imaging, two non-invasive techniques have been frequently used to eliminate the intravascular blood component of the signal: 1) application of saturation pulses and 2) addition of small velocity sensitizing gradients. For the first portion of this study, we choose to utilize saturation pulses to investigate the magnitude of the cerebral blood contribution to the T_1ρ signal since it would not sensitize the signal to diffusion processes and allowing short echo-times to be utilized minimizing T₂ contributions. Our hypothesis is that blood volume will correspond to a portion of the T_1ρ signal in vivo but metabolic contributions will provide the dominant source of the T_1ρ signal change. This is based on the recent work conducted by Jin and Kim in animals (Jin and Kim, 2013). Second, the relationship between the frequency of the visual stimulation and the functional response assessed using ³¹P spectroscopy and T_1ρ imaging was investigated. A number of previous positron emission tomography (PET) and functional MRI (fMRI) BOLD studies have shown the frequency dependence of the blood flow response within the visual cortex response to a flashing checkerboard. Previous studies have found the maximum blood flow response to occur with stimulation frequencies in the range of 6–8 Hz (Fox and Raichle, 1984; Lin et al., 2008; Singh et al., 2003; Vafaee et al., 1999). In addition, a previous study using ¹H spectroscopy has also shown frequency dependence in lactate signal (Lin et al., 2010). In this study, we hypothesized that the measurements of metabolism assessed using ³¹P spectroscopy and T_1ρ will exhibit similar frequency dependence with the greatest response occurring near 8 Hz. To test this hypothesis, dynamic imaging was performed using T_1ρ, BOLD, and ³¹P spectroscopy while viewing a full field flashing checkerboard with different temporal frequencies (1, 4, and 7 Hz).

Methods

Data acquisition

Two MR imaging experiments were conducted on a 3T Siemens Trio scanner (Siemens Medical Solutions, Erlangen, Germany) using a 12 channel head-coil for the functional T_1ρ imaging and quadrature transmit receive dual tune ³¹P/¹H coil (RapidMRI, Columbus, Ohio) for the ³¹P spectroscopic acquisitions. Before being enrolled into either of the two experiments, all subjects provided written informed consent to a research protocol approved by the Institutional Review Board at the University of Iowa Institutional.

The first experiment was designed to evaluate changes in the functional T_1ρ signal and pH associated with stimulation frequency in the visual cortex. Ten subjects (Seven men and three women, 25–35 years of age) underwent MRI, fMRI, and ³¹P MRS study. High-resolution anatomical T₁-weighted images were acquired using a 3D MP-RAGE sequence using the following parameters: TR = 2530 ms, TE = 2.8 ms, TI = 909 ms, flip angle = 10°, FOV = 256 × 256 × 256 mm, matrix size = 256 × 256 × 256, bandwidth = 180 Hz/pixel. BOLD imaging was performed using a $T_2^{*}$-weighted gradient-echo sequence with the following acquisition parameters: TR = 2000 ms, TE = 30 ms, flip angle = 90°, FOV = 220 × 220 mm, matrix size = 64 × 64, bandwidth = 2004 Hz/pixel, 25 slices and slice thickness/gap = 5.0/1.0 mm. Functional T_1ρ imaging was performed using an echo-planar spin-echo sequence with an additional T_1ρ spin-lock encoding pulse (Fig. 1) to collect 15 slices. The sequence parameters were TR = 2000 ms, TE = 12 ms, FOV = 220 × 220mm, matrix size = 64 × 64, bandwidth = 1954 Hz/pixel, and slice thickness = 5.0 mm. Two spin-lock pulses were used (10 and 40 ms) with a spin-lock frequency of 350 Hz. The final portion of the study acquired ³¹P spectroscopic data. The subjects were briefly removed from the scanner and the 12 channel head-coil was replaced with a dual tune ³¹P/¹H coil. Imaging for the spectroscopic portion of the study consisted of a localizer followed by a 2D ³¹P spectroscopic acquisition. ³¹P CSI spectroscopy was collected using a free induction decay acquisition with the following parameters: TR = 4000 ms, TE = 2.3 ms, FOV = 240 × 240mm, matrix size = 8 × 8, thickness = 30 mm, bandwidth = 1000 Hz, vector size = 1024. A volume of 40 mm isotropic centered on the occipital cortex was selected as the shim volume and was utilized as the CSI voxel of interest for metabolic measurements.

Fig. 1.

T_1ρ-weighted pulse sequence with EPI readout. Two non-selective 90° pulses are separated by a pair of spin locking pulses with opposite phase. Crusher gradients (shaded) are used to destroy any residual magnetization after the spin-lock preparation period. T_1ρ images are acquired by conventional spin-echo EPI readout. The events within the brackets are repeated N times within a given TR where N is the number of slices acquired. M multiple spin-lock pulse durations can be selected.

It should be noted that the functional T_1ρ imaging was performed by applying a spin-lock pulse for each of the acquired slices in the echo-planar sequence (Fig. 1) effectively reducing the TR for application of the T_1ρ pulse. The TR for application of the spin-lock pulses was approximately 133 ms. This approach greatly increases the temporal resolution of the technique at the expense of quantitative T_1ρ accuracy. With such an approach, data from the first four acquired slices needs to be excluded from the analysis since they have not reached steady state. The remaining 11 slices exhibit a steady state response with the percent signal change in T_1ρ relaxation time being linear for each slice. The estimated T_1ρ percent signal is underestimated for the slices of interest used in this work as compared to a fully relaxed condition.

The second imaging experiment was designed to determine the influence of cerebral blood volume on the functional T_1ρ signal. Three healthy control subjects (all male between the ages of 29–34) were recruited into the study. Single slice functional T_1ρ imaging was performed using the same parameters as used in the previous experiment except only a single slice of data was acquired. Two repetitions of the T_1ρ measurements were obtained while stimulating the visual cortex using a full field 7 Hz flashing checkerboard as described below. In one of the runs, an inferior spatial saturation pulse with slab thickness of 100 mm and offset 5 mm from the imaging slice was applied while in the other run the data was acquired without the saturation pulse.

Stimulation paradigms

The experimental paradigm used was a full field visual flashing checkerboard presented with a block design of 18 scans (36 s) “OFF” and 18 scans (36 s) “ON” following the first. For the first imaging experiment, a visual flashing checkerboard was presented for the functional imaging using a block design as shown in Fig. 2A. Each block of visual stimulation used one of three frequencies (1, 4, or 7 Hz) of the flashing checkerboard, which were presented in random order. The baseline condition was a fixation cross. Every 4 s during the flashing checkerboard condition, a red square was shown in the center of the checkerboard. Subjects were asked to press a button on the fiber optic response system (Lumina LP-400, Cedrus Corporation, San Pedro, USA) when the square appeared. This was done to ensure that the subjects were on task during the entire study. This was especially important during the spectroscopic portion of the study where the activation period was lengthy. For BOLD and T_1ρ imaging, 7 cycles of activation and visual fixation were presented with a 72 second period. The task began and ended in the baseline condition. Two runs of the BOLD and T_1ρ measurements were obtained. The second experiment used only 7 Hz flashing checkerboard for each stimulation block as shown in Fig. 2B. Single slice functional T_1ρ imaging data was acquired with and without the application of the spatial presaturation RF pulse to determine the influence of cerebral blood volume on the functional T_1ρ signal. For ³¹P spectroscopy, the session consisted of four blocks each lasting 5 min 20 s. The visual stimulation was presented to the subjects using MATLAB (The MathWorks, Inc., Natick, MA, USA) and the Psychophysics Toolbox (Brainard, 1997; Pelli, 1997).

Fig. 2.

Flashing checkerboard paradigms for the functional imaging experiments. (A) T_1ρ imaging without a spatial presaturation pulse (w/o preSAT) and with the presaturation pulse (w/ preSAT) were performed to determine the influence of cerebral blood volume on the functional T_1ρ signal. VF = visual fixation. (B) BOLD and T_1ρimaging were performed twice in an interleaved manner, followed by ³¹P spectroscopy study. ST1, ST2, and ST3 represent visual stimulation with a full field flashing checkerboard at different temporal frequencies of 1, 4, and 7 Hz. The temporal frequencies were presented in a random order. During the ³¹P study, the visual stimuli (ST1, ST2, and ST3) were randomly presented.

Image analysis

The anatomical T₁-weighted images for each subject were placed into the Montreal Neurological Institute (MNI) coordinate system using AFNI. All functional imaging data were analyzed using a combination of AFNI (Cox, 1996) and MATLAB.

Experiment 1: Frequency dependence of functional signals

For BOLD data, pre-processing of $T_2^{*}$-weighted time series data included three-dimensional motion correction, slice timing correction, linear trend removal, and spatial smoothing with a Gaussian filter (6 mm FWHM). The data from each run was concatenated before statistical analysis. The analysis was performed for all voxels within the brain for each subject using a general linear model (GLM). A linear regression was performed to estimate the signal contributed by the blood oxygenation signal corresponding to the experimental design. An estimate of the hemodynamic response for each stimulation frequency was estimated by convolving the experimental design with a gamma-variate function (Cox, 1996). Nuisance regressors were included for the subject motion parameters and baseline fluctuations using a third order polynomial. A t-test was employed to generate statistical parametric maps. The mean $T_2^{*}$-weighted image was estimated and used to co-register the images with the MNI aligned T₁-weighted anatomical image using an affine transformation. The resulting transform was used to place the statistical parametric maps into MNI space.

The individual spin-lock images from the functional T_1ρ runs were aligned to a single time point using three-dimensional motion correction. Each pair of spin-lock times was then analyzed using MATLAB to generate a T_1ρ map using a log-linear regression of the voxel signal intensity with the spin-lock time. The T_1ρ maps were spatially smoothed using a Gaussian filter (6 mm FWHM). The resulting T_1ρ maps were analyzed using multiple linear regression that included separate regressors for each stimulation frequency and motion parameters as nuisance regressors. For functional T_1ρ imaging, the experimental design was used as the regressor of interest since the impulse response function of this signal has not yet been characterized. The mean 40 ms spin-lock image was estimated and used to co-register the T_1ρ images with the MNI aligned T₁-weighted anatomical image. The resulting transform was used to place the statistical parametric maps into MNI space.

The frequency dependence of the BOLD and T_1ρ responses were analyzed to identify voxels that whose response exhibited a significant relationship for the following contrasts: 7 Hz-1 Hz, 7 Hz-4 Hz, and 4 Hz-1 Hz. The threshold for significance in the contrast map was set at p < 0.05 corrected for multiple comparisons using a false discovery rate correction.

The ³¹P data was analyzed using the Siemens Syngo software to determine the chemical shift of the inorganic phosphate (Pi) and phosphocreatine (PCr) peaks in the spectra. The analysis included frequency filtering, frequency and phase correction, baseline correction, and curve fitting with prior knowledge. The pH within the brain was estimated using the following formula proposed by Patel et al. (2000).

$$\text{pH}=6.77+\log \left\{\left(\delta -3.29\right)/\left(5.68-\delta \right)\right\}$$

where δ is the chemical shift in ppm between Pi and PCr. The brain pH in the visual cortex estimates according to the corresponding condition where compared using an ANOVA model in SPSS.

Experiment 2: Blood volume contribution to functional T_1ρ signal

The individual spin-lock images from the functional T_1ρ runs were aligned to a single time point. For this experiment two-dimensional correction was used and T_1ρ maps generated as described above. The T_1ρ maps were spatially smoothed using a Gaussian filter (6 mm FWHM). The resulting smoothed T_1ρ maps were analyzed using multiple linear regression using the experimental design as the regressor of interest and motion parameters as nuisance regressors. The resulting statistical maps were thresholded at a p < 0.01, uncorrected. The union of the two activation maps was calculated and the signal change within this region was calculated.

Results

Statistically significant activation maps corresponding to the flashing checkerboard were generated for each frequency (1, 4, and 7 Hz) using both T_1ρ and BOLD functional imaging. Fig. 3 shows the group activation maps of the T_1ρ and BOLD imaging in MNI space overlaid on the average T₁-weighted anatomical image. The activated voxels with a significant activation (p < 0.05, corrected) are shown. The positive T_1ρ activation is shown in red and the positive BOLD activation is in blue. In Fig. 3, it is evident that the area of activation increases with stimulation frequency for both BOLD and T_1ρ functional imaging. The size of the activated region as well as the median and maximum t-statistic also reflects this frequency dependency (Table 1). It is evident that the BOLD response is substantially larger than the T_1ρ response, however there is a high degree of overlap (green voxels shown in Fig. 3) between the two responses. The average of the T_1ρ and BOLD time course as shown in Fig. 4 was obtained in the ROI (3 × 2) within the visual cortex across subjects as shown with a white square in Fig. 3C. It is observed that the magnitude of the T_1ρ response depends on the temporal frequency. In addition, the peak percentage change for T_1ρ functional imaging (~2%) was approximately half of the change found using BOLD imaging (~4%). Fig. 5 shows contrast maps of BOLD and T_1ρ response between stimulus frequencies. Both BOLD and T_1ρ response significantly increased with the visual stimulation frequency. The ³¹P data showed a reduction in the estimated pH value in all subjects during the activated state within the voxel encompassing the visual cortex. Representative ³¹P spectroscopy spectra during different conditions (VF, 1 Hz, 4 Hz, and 7 Hz) within the visual cortex are presented in Fig. 6A. Fig. 6B shows brain pH values within the visual cortex estimated by ³¹P spectroscopy. During 4 Hz and 7 Hz visual stimulation, brain pH significantly decreased (acidosis) relative to the visual fixation (REST) and there was a trend relative to the 1 Hz visual stimulation (p = 0.08). pH was slightly reduced at 1 Hz relative to REST, but this was not statistically significant. PCr/Pi and ATP/Pi ratios decreased with stimulation frequency but were not statistically significant (Table 2).

Fig. 3.

BOLD and T_1ρ functional activation maps resulting from the different stimulus temporal frequencies. Group analysis of the functional imaging data (p < 0.05, corrected) for the (A) BOLD and (B) T_1ρ at flashing checkerboard frequencies of 1, 4, and 7 Hz, respectively. Two axial slices (MNI z = 6 and z = 10) are shown. Different color scales are used for the BOLD (sky/blue) and T_1ρ (yellow/red) to display the positive activation. (C) The bottom row shows the overlap (green) between the BOLD and T_1ρ results. Voxels responding only to BOLD are shown in blue and those only responding with T_1ρ are shown in red.

Table 1.

MNI coordinates of the activated cluster center of mass, cluster size, and significance.

		MNI coordinates			Cluster size	Median t-value	Maximum t-value	p
		X	y	z
BOLD	1 Hz	4	−90	−12	13,064	8.7	13.7	<0.05
	4 Hz	2	−88	−15	14,116	8.6	17.5	<0.05
	7 Hz	2	−88	−15	21,321	10.1	30.4	<0.05
T1p	1 Hz	−26	−99	−8	26	2.8	4.0	<0.05
	4 Hz	22	−91	0	162	4.4	4.6	<0.05
		−2	−101	3	35
	7 Hz	4	−85	−3	359	4.8	6.4	<0.05
		−18	−99	−10	128

Fig. 4.

Percentage changes of BOLD signal and T_1ρ times during different stimulus temporal frequencies. Group-average percent changes of the BOLD (blue line) and T_1ρ (red line) time courses obtained in the ROI (3 × 2) within the visual cortex across subjects as shown in white colored square in Fig. 3C. Error bars depict standard error of the mean.

Fig. 5.

BOLD and T_1ρ contrast maps comparing response based on flashing frequency (7 Hz-1 Hz, 7 Hz-4Hz, and 4 Hz-1 Hz). Axial sections of statistical contrast maps (p < 0.05, corrected) for (A) BOLD and (B) T_1ρresponse between stimulus frequencies of 7 Hz, 4 Hz, and 1 Hz. Three slices (MNI z = −8, z = −4, and z = 4) are shown.

Fig. 6.

Brain pH measurements obtained using ³¹P spectroscopy during different conditions (VF, 1 Hz, 4 Hz, and 7 Hz) within the visual cortex.(A) Representative ³¹P spectroscopy spectra during the different conditions (PE = phosphoethanolamine, Pi = inorganic phosphate, PCr = phosphocreatine, GPE = glycerophosphoethanolamine, GPC = glycerophosphocholine). (B) Brain pH obtained from the chemical shift between PCr and Pi peaks measured. Error bars depict standard error. There was a statistically significant difference between pH during the stimulation at 4 and 7 Hz as compared to REST.

Table 2.

Phosphate ratios and intracellular pH during different stimulus flashing frequencies.

Rate	PCr/Pi(% REST)	ATP/Pi (% REST)	pH
REST	100 ± 10.7	100 ± 10.0	7.061 ± 0.011
1 Hz	83.9 ± 13.4	88.67 ± 13.4	7.045 ± 0.015
4 Hz	79.06 ± 11.3	88.41 ± 16.6	7.007 ± 0.010*
7 Hz	76.57 ± 9.3	78.17 ± 9.6	7.008 ± 0.009*

^*(p < 0.05).

Statistically significant activation was observed in the primary visual cortex for T_1ρ imaging with and without the spatial presaturation RF pulse as shown in Fig. 7. The activated voxels with a significant activation (p < 0.01, uncorrected) are shown. Areas of the T_1ρ activation with the presaturation pulse were similar with those of the T_1ρ activation without the presaturation pulse. However, a percentage signal change was decreased when the presaturation pulse was applied. It was found that the blood volume contribution to activity-evoked T_1ρ change was about 27% (34.8%, 24.1%, and 22.8% for each subject) as shown in Table 3.

Fig. 7.

T_1ρ functional activation maps with/without the spatial presaturation RF pulse for three subjects in response to a full field flashing checkerboard (7 Hz). Colors indicate the percent signal change from baseline for voxels exhibiting a statistical threshold of p < 0.01, uncorrected. The rightmost images show the overlap (green) between the T_1ρ results without and with the presaturation pulse. Voxels responding only to T_1ρ without the presaturation pulse are shown in blue and those only responding with T_1ρ with the presaturation pulse are shown in red.

Table 3.

T_1ρ signal percent changes in response to the flashing checkerboard frequency of 7 Hz with/without the spatial presaturation RF pulse in three subjects.

Subject	% changes (without spatial preSAT)	% changes (with spatial preSAT)	% decrease
S1	2.27%	1.48%	34.8%
S2	2.53%	1.92%	24.1%
S3	1.36%	1.05%	22.8%

Discussion

These results replicate the findings of our previous work that suggested an acidosis within the visual cortex in response to a full field flashing checkerboard using ³¹P spectroscopy (Magnotta et al., 2012), which also corresponded to a similar response using T_1ρ functional imaging. The work herein extends that previous study by finding that the magnitude of the pH and T_1ρ response depends on the stimulation frequency and thus likely depends on the associated changes in brain activity (Singh et al., 2003). ³¹P spectroscopy detected a greater acidosis during visual stimulation with the higher temporal frequencies (4 and 7 Hz). It should be noted that the ³¹P voxel of interest was fairly large (27 cm³) and likely only a fraction of the tissue within the voxel was active and produced a local acidosis. Therefore, the local pH change within the activated tissue may be significantly greater than what was found here. The T_1ρ response also showed a graded response based on stimulation frequency and in a direction that would be consistent with an acidosis. However, this correlation does not prove that the functional T_1ρ response is driven by acidosis, and other metabolites that change with brain activity may also contribute to the response.

Our BOLD data agree with the findings of Vafaee and Gjedde (2000) and Singh et al. (2003) suggesting that the cluster size of activated voxels as well as the magnitude of BOLD signal is increased with stimulation frequency within the range studied (1–7 Hz). Neurovascular responses of the visual pathway as a function of stimulus frequency have been described for visual stimulation (Pawela et al., 2008). It is known that the electrical response amplitude induced by the visual stimulus relies on the total number of cortical neurons responding to the stimulus, with the greatest number of neurons responding to stimulus frequency rates (approximately 8 Hz, 1/125 msec) matching the “activity-recovery cycle” duration (125 ms) of the pathway from retina to visual cortex. The response amplitude decreases as the temporal frequency of flicker exceeds 8 Hz. The stimulus rate-dependent increases in the BOLD signal are also consistent with the PET observations for CBF (Fox and Raichle, 1984).

Previous studies by Hulvershorn et al. (2005) suggested that a local change of cerebral blood volume mainly affects functional T_1ρ signal because blood T_1ρ is longer than tissue T_1ρ. In addition, the concentration of other metabolites such as glucose and glutamate may also affect functional T_1ρ signal (Jin and Kim, 2013). In order to suppress the intravascular signal, a spatial presaturation RF pulse was applied inferior to the imaging slice to minimize the blood contribution to the observed signal intensity. Our result shows the vascular contribution to T_1ρ signal is about 27% in the brain. Therefore, a majority of T_1ρ signal likely comes from the tissue compartment. A similar cerebral blood volume fraction of the T_1ρ signal was also found by Jin and Kim (2013). It should be noted that the first experiment uses a multi-slice approach for acquiring data and may have different cerebral blood volume contributions as compared to the single slice acquisition. Future studies are needed to fully understand the physiological sources that drive the functional T_1ρ signal. Our previous study in sheep blood phantoms suggested a double disassociation between T_1ρ and BOLD; T_1ρ was not sensitive to blood oxygenation while BOLD was not sensitive to pH (Magnotta et al., 2012). Here we also found significant differences in the activation patterns detected by T_1ρ and BOLD. While the T_1ρ activation area was contained mostly within the BOLD activation area, the T_1ρ activation area was significantly smaller. It is well known that the BOLD signal shows significant venous contributions in draining blood from the activated areas; thus, BOLD contrast may occur at sites significantly distant from the actual sites of neuronal activity. This is why some investigators apply small motion sensitizing gradients to the echo-planar gradient echo sequence commonly used to acquire BOLD data. Therefore, we speculate that the smaller T_1ρ changes may represent local metabolic changes that are more localized to the area of brain activity as compared to the BOLD signal. We found that the signal change was half of the BOLD magnitude and may require additional temporal sampling to achieve the same power as BOLD imaging.

Lactate is one of the potential sources of acid that influences pH. Prichard et al. (1991), Maddock et al. (2009) and Magnotta et al. (2012) observed that lactate concentration significantly increases during visual stimulation. Lin et al. (2010) also observed a stimulation rate-dependent increase of CBF correlated with lactate production. These observations suggest that anaerobic glycolysis may help drive the pH and CBF responses detected here. Increased activity requires greater blood flow to supply glucose and oxygen and to remove metabolic waste including CO₂. Prior studies have shown that local acidosis is a critical determinant of cerebrovascular tone (Kety and Schmidt, 1948). Consequently, activity-evoked acidosis may be a key mechanism underlying neurovascular coupling. Nevertheless, current functional imaging paradigms remain dependent on blood flow, blood volume and the vascular anatomy, which limits their temporal and spatial resolution. If pH helps drive the vascular response, then T_1ρ could potentially be a more direct and precise method of imaging brain function. If so, one might expect the local metabolic changes detected by T1p to be more localized to the activated region. In addition, the T_1ρ response might precede the hemodynamic response detected by BOLD and CBF measures. In a parallel imaging study, we have found that the T_1ρ signal precedes the temporal changes seen using BOLD and arterial spin-labeling (Heo et al., 2013) providing further evidence that the T_1ρ signal is sensitive to functional metabolic changes.

The activity-evoked pH changes detected here by ³¹P-spectroscopy required substantially longer imaging times compared to BOLD and T_1ρ. This additional imaging time may lead to greater habituation and attenuation of the neuronal activity (Ho and Berkley, 1988; Janz et al., 2001; Maddess et al., 1988), and may help explain why ³¹P did not detect a pH change evoked by 1 Hz or why the pH changes evoked by 4 and 7 Hz did not differ from each other. Consistent with this possibility, with continuous visual stimulation Sappey-Marinier et al. (1992) observed a decline in lactate concentration from an initial maximum level. In our ³¹P study, we found a decrease in brain pH during the visual stimulus in all subjects. However, recent ³¹P studies have indicated a trend towards a slightly alkaline pH or no changes in brain pH during visual stimulation (Rango et al., 2001; Sappey-Marinier et al., 1992). In our study design, we required subjects to remain attentive and vigilant during the entire activation period by having them respond with a button press to a visual cue that appeared every 4 s. This was important due to the long acquisition times required for the ³¹P protocol. In addition, differences in field strength, acquisition parameters, age of the subjects, and/or the visual stimulus may account for differences in the results.

In conclusion, our study suggests that the functional T_1ρ response depends on the degree of brain activity. Furthermore, this work suggests a strong relationship between T_1ρ and pH estimated by ³¹P spectroscopy. Consequently, our findings support the hypothesis that dynamic T1p imaging detects activity-evoked pH changes. Recent study by Jin and Kim (2013) suggested that tissue T_1ρ signal in parenchyma is mainly caused by an activation-induced change of tissue metabolism such as glucose and glutamate while tissue acidosis is unlikely the major contributor to the activity-evoked T_1ρ signal. Glucose and pH are tightly coupled through the adenosine triphosphate (ATP) cycle, therefore the present study is unable to determine the source of the T_1ρ changes and warrants future studies. T_1ρ imaging has the potential to provide a new functional imaging marker that may be more specific to the area of brain activity. Therefore, it is possible that by non-invasively detecting metabolic changes in the human brain, T_1ρ MRI could offer a novel, more direct approach to map brain function. A number of psychiatric and neurological disorders could potentially benefit from the ability to study dynamic metabolic changes.