Detecting activity-evoked pH changes in human brain

Abstract

Localized pH changes have been suggested to occur in the brain during normal function. However, the existence of such pH changes has also been questioned. Lack of methods for noninvasively measuring pH with high spatial and temporal resolution has limited insight into this issue. Here we report that a magnetic resonance imaging (MRI) strategy, T₁ relaxation in the rotating frame (T₁ρ), is sufficiently sensitive to detect widespread pH changes in the mouse and human brain evoked by systemically manipulating carbon dioxide or bicarbonate. Moreover, T₁ρ detected a localized acidosis in the human visual cortex induced by a flashing checkerboard. Lactate measurements and pH-sensitive ³¹P spectroscopy at the same site also identified a localized acidosis. Consistent with the established role for pH in blood flow recruitment, T₁ρ correlated with blood oxygenation level-dependent contrast commonly used in functional MRI. However, T₁ρ was not directly sensitive to blood oxygen content. These observations indicate that localized pH fluctuations occur in the human brain during normal function. Furthermore, they suggest a unique functional imaging strategy based on pH that is independent of traditional functional MRI contrast mechanisms.

To what degree pH changes during normal brain function is unclear (1). However, neuronal activity could cause transient, localized pH changes via several mechanisms. Increased neuronal activity enhances carbohydrate metabolism producing the pH-lowering by-products lactic acid and CO₂ (2). Activity-evoked HCO₃⁻ transport can alter pH (3). Local field potentials produced by ion fluxes could change pH (4). In addition, acidic synaptic vesicles release protons during neurotransmission (5). Such dynamic pH fluctuations have the potential to dramatically alter physiology and behavior through a number of pH-sensitive receptors and channels (6). Acid-sensing ion channels, for example, play critical roles in synaptic plasticity, learning, memory, pain, and neurodegeneration (7–10). Superimposed on activity-dependent brain pH changes and the potential physiological effects are several buffering systems. Principal among these is the CO₂/HCO₃⁻ system. In a reversible reaction, CO₂ combines with water to form carbonic acid, which readily dissociates into HCO₃⁻ and H⁺. Raising HCO₃⁻ shifts the equilibrium away from H⁺ and increases pH. Conversely, raising CO₂ shifts the equilibrium toward H⁺, thereby lowering pH. The ability to measure these pH changes in the functioning brain is key for gaining insight into this poorly understood dimension of CNS physiology and pathophysiology.

Routinely measuring pH in the brain would require novel noninvasive methods. Traditionally, ³¹P spectroscopy has been used to estimate brain pH (11); however, ³¹P is limited by poor spatial resolution (typically 10- to 30-cm³ volumes), long acquisition times (often 5–10 min for a single measurement), and the need for special hardware not typically available on clinical scanners. Recently, ¹H MRI pulse sequences have been shown to detect H⁺ exchange between water and proteins and thus can be highly pH sensitive. These techniques include amide proton transfer (APT) and T₁ in the rotating frame (T₁ρ) (12–15). APT detects H⁺ exchange by taking advantage of differences in resonances between amide and water protons. The spin-lock preparation pulse used in T₁ρ imaging sensitizes the magnetic resonance (MR) signal to relaxation effects arising from H⁺ exchange between free water protons and those bound to proteins and macromolecules. Here, we focused on T₁ρ because of its pH sensitivity, high spatial and temporal resolution, and potential to detect dynamic pH changes during brain function.

Results

Validation of pH Sensitivity in Buffered Phantoms.

To evaluate T₁ρ sensitivity to pH in the physiological range, we first studied phantoms [3.5% agar (wt/vol)], 8% BSA (wt/vol), 0.1 M phosphate buffered saline pH-adjusted with HCl and NaOH to values ranging from pH 6.0 to pH 8.0 (Fig. 1A). They were imaged by using a fast spin-echo sequence with a spin-locking preparation pulse, which created a B₁ field of 1,000 Hz, and four spin-lock times (10, 20, 40, and 60 ms). T₁ρ times were inversely proportional to the measured pH (R² = 0.98) (Fig. 1A), suggesting that T₁ρ is sensitive to pH in the physiological range.

Fig. 1.

T₁ρ and T₂* sensitivity to pH and pO₂ manipulations. (A) T₁ρ maps for agar phantoms with pH adjusted to different levels (6.0–8.0). The relationship between T₁ρ times and pH was linear within this range. The estimated T₁ times were calculated within a central 5 × 5 region. (B) T₁ρ maps in sheep blood phantoms (Upper) with corresponding mean and SD in a central 5 × 5 region of interest plotted in Lower. The sheep blood phantoms are arranged from left to right as follows: unaltered (control), acidified, and hyperoxygenated. (C) Corresponding T₂* maps in same sheep blood phantoms (Upper) with mean and SD in a central 5 × 5 region of interest plotted in Lower.

T₁ρ Insensitivity to Oxyhemoglobin Content.

Because current blood oxygenation level-dependent (BOLD) imaging relies on T₂* contrast to detect changes in blood oxyhemoglobin content, we compared the specificity of T₁ρ and T₂* for pH and oxygen content in fresh sheep blood. Blood pH and O₂ content were varied in three test conditions: (i) unaltered, (ii) acidified (25 mM HCl), and (iii) oxygenated with 100% O₂. Data from a single axial slice were collected by using a T₁ρ fast spin-echo sequence and T₂*-weighted gradient-echo sequence. We found that T₁ρ was sensitive to pH, but not O₂ content (Fig. 1B). Conversely, T₂* was sensitive to O₂ content, but not pH (Fig. 1C). These data suggest that T₁ρ changes are unlikely due to changes in oxyhemoglobin content.

pH Detection in Mouse Brain.

To assess pH sensitivity of T₁ρ in vivo, measurements were simultaneously obtained with T₁ρ and a custom-made MRI-compatible pH sensor (pHOptica; World Precision Instruments; detection range pH 5–9) implanted into the amygdala of mice. Data were collected from anesthetized mice under three conditions: (i) room air inhalation, (ii) 20% CO₂ inhalation, and (iii) following HCO₃⁻ injection (5 mmol/kg, i.p.). Fig. 2A shows examples of T₁ρ maps obtained from a single mouse brain. Direct pH measurements varied from animal to animal, most likely due to differences in respiratory suppression from anesthesia. In all cases, CO₂ inhalation lowered pH relative to air and prolonged T₁ρ times throughout the brain, including at the pH-sensing probe tip. HCO₃⁻ injection produced the opposite effect, raising pH and shortening T₁ρ times. Fig. 2B shows the relationship between T₁ρ at the sensor tip and pH measured from the sensor across all mice and conditions (R² = 0.77).

Fig. 2.

T₁ρ detects pH changes in the anesthetized mouse brain evoked by CO₂ inhalation or i.p. HCO₃⁻ injection. (A) T₁ρ maps from a mouse brain after given bicarbonate (Left), exposed to room air (Center), and exposed to 10% CO₂ (Right). The position of the fiber-optic pH sensor is shown by the white arrow, and the open square indicates the location of the ROI used for estimating the T₁ρ times. (B) The relationship between the mean T₁ρ times and the corresponding direct pH measurements obtained from the fiber-optic sensor across four mice. Each mouse is represented by a different symbol.

Systemic pH Changes in Human Brain Induced by CO₂ and Hyperventilation.

Qualitatively similar T₁ρ responses were observed in the human brain when pH was manipulated through breathing. A research participant was imaged on a Siemens 3T scanner under three conditions: (i) normal breathing of room air, (ii) breathing 5% CO₂, and (iii) paced hyperventilation of room air (27 breaths per minute). These manipulations changed average end-tidal CO₂ (EtCO₂) measurements from 4.3% to 7.7% CO₂ (Fig. 3B). Consistent with hypercarbic acidosis, CO₂ inhalation produced a widespread increase in T1ρ relaxation time (Fig. 3 A and B). Paced hyperventilation produced the opposite effect, and consistent with a respiratory alkalosis reduced T₁ρ times (Fig. 3 A and B). Fig. 3C shows the subtracted T₁ρ maps between hyperventilation and room air, and between 5% CO₂ and room air, suggesting that these systemic manipulations change pH throughout the brain.

Fig. 3.

T₁ρ measurements throughout the human brain are responsive to EtCO₂ manipulation. (A) T₁ρ maps of the human brain varied with EtCO₂ concentration during hyperventilation, breathing room air, and 5% CO₂ challenge. The open box identifies a 7 × 7 region of interest in white matter where T₁ρ time estimates were obtained. (B) The estimated T₁ρ times in white matter varied with measured EtCO₂ concentrations (module CO2100C; Biopac). (C) T₁ρ subtraction images between the hyperventilation and air conditions (Left) and 5% CO₂ and air condition (Right).

Localized pH Changes Induced by Flashing Checkerboard.

With the above data suggesting that T₁ρ is sensitive to brain pH, we hypothesized that T₁ρ might detect localized pH changes evoked by brain activation. To test this hypothesis, we used a visual flashing checkerboard task (Fig. 4A). Six participants were presented with a visual flashing checkerboard (22 × 22 squares) alternating at 8 Hz in a block design (Fig. 4A) (16). For comparison, two runs of BOLD were interleaved with three runs of T₁ρ. Fig. 4B shows the group activation maps for the T₁ρ and BOLD imaging overlaid on the average T₁-weighted anatomical image. Voxels exhibiting significant activation (P < 0.05, corrected) are shown. The flashing checkerboard significantly increased T₁ρ times in the occipital cortex. A similar activation region was measured by functional MRI (fMRI) BOLD. Consistent with the observation that T₁ρ and BOLD detect mutually independent phenomena, a difference was observed between the size of the T₁ρ- and BOLD-responsive areas. The reason for this apparent difference is not clear. It is possible that T₁ρ detects more focal changes than BOLD, given that the vascular tree underlying the hemodynamic response is larger than regions of neural activity. For example, venous drainage of brain areas can extend the BOLD signal tens of millimeters from the activation site (17). The well-established effect of acidic pH on blood flow (18) suggests localized acidosis might even help drive the BOLD response.

Fig. 4.

Flashing checkerboard alters T₁ρ, BOLD, lactate, and ³¹P measurements in the visual cortex. (A) Flashing checkerboard paradigms for functional T₁ρ, BOLD, and MRS studies. For the T₁ρ/BOLD studies, three runs of the T₁ρ block sequence where interleaved with two runs of BOLD. (B) T₁ρ and BOLD functional activation maps (P < 0.05, corrected) resulting from the visual flashing checkerboard stimulus. Four contiguous slices are shown. (C) Lactate to total creatine ratio increased significantly during visual stimulation relative to both baseline and the poststimulus recovery phase. *P < 0.05, paired t test. (D) ³¹P spectroscopy estimates of pH in the visual cortex were significantly reduced during flashing checkerboard presentation relative to both baseline and recovery. *P < 0.05, paired t test.

Because lactic acid is one potential source of localized pH change, we measured lactate by ¹H MR spectroscopy (MRS) in a voxel positioned at the BOLD site. Consistent with a previous report (19), we found that the flashing checkerboard significantly increased the lactate-to-creatine ratio (Fig. 4C). If the T₁ρ and lactate responses indeed reflect a localized acidosis, we hypothesized that we should be able to detect the pH change with an alternative method. We chose ³¹P spectroscopy, a widely accepted measure of pH, and used the T₁ρ data to guide voxel positioning. In six additional subjects, the flashing checkerboard altered visual cortex ³¹P estimates of pH (Fig. 4D). Like T₁ρ, these changes indicate a transient, activity-dependent localized acidosis.

Discussion

These results indicate that neuronal activity can change local pH in the human brain during normal function. Neuronal activation may lower pH through a variety of processes (20) acting individually or together to produce the acidosis observed here; further study will be required to identify the mechanisms underlying this pH change. Because monocarboxylate transporters cotransport protons with lactate (21), our studies suggest that lactate production and transport might contribute to the localized acidosis.

The localized acidosis observed here might have a number of consequences for brain physiology and pathophysiology (6, 8). The reduced pH could activate some ion channels and receptors and inhibit others, thereby influencing brain function and behavior (5, 7, 8, 22). A reduced pH has also been implicated in ischemic stroke, neurodegenerative disease, seizures, and respiratory control (9, 10, 23–27). Interestingly, in patients with panic disorder, lactate induction by the flashing checkerboard was abnormally elevated (19). Others have also suggested lactate and pH-buffering abnormalities in panic disorder (28). These observations coupled with our findings support the possibility that abnormal pH dynamics may contribute to panic disorder (2).

Additional advances in our knowledge of brain pH dynamics might come from the improved spatial and temporal resolution provided by T₁ρ. The T₁ρ sequence used here has an isotropic spatial resolution of ∼4 mm and temporal resolution of 6 s, whereas the spatial and temporal resolutions of ³¹P spectroscopy are 30 mm and several minutes, respectively. Further improvements in the T₁ρ scan time may be possible by acquiring only two spin-lock times and by using parallel imaging. Comparable temporal and spatial resolution might also be possible with APT in an echo-planar sequence if only two frequency-offset pulses were applied about the center imaging frequency.

One limitation of both T₁ρ and APT is that they depend on H⁺ exchange with proteins and amide groups. Thus, either T₁ρ or APT could be affected by appreciable changes in local protein concentration as well as pH. The observation that T₁ρ correlated closely with direct pH measurements in the mouse brain and with secondary pH imaging methods in human brain argues that the T₁ρ changes observed here were due at least in part to pH.

In addition to improving the ability to measure brain pH, this study has unique implications for functional imaging in general. Because T₁ρ showed a linear response to pH, T₁ρ may be more quantifiable than BOLD fMRI, which is not quantifiable other than percent change and does not address baseline conditions. In addition, the spatial resolutions of BOLD fMRI and ¹⁵O-water positron emission tomography depend on blood flow and oxyhemoglobin content and are thus limited by the vascular anatomy. Although T₁ρ provided a similar pattern of activation as BOLD, T₁ρ changes were independent of blood oxygenation. Thus, by measuring functional pH changes, T₁ρ MRI might provide a means for more precisely localizing brain activity.

Methods

Sheep Blood Phantom Imaging.

All animal care met National Institutes of Health standards, and the University of Iowa Animal Care and Use Committee approved all procedures. T₁ρ data were collected by using a fast spin-echo sequence with four spin-lock times (10, 20, 40, and 60 ms) and B₁ = 400 Hz. T₂*-weighted imaging, which is sensitive to blood oxygenation, was obtained from a single axial slice by using a gradient-echo sequence with eight echo-times (1.7, 2, 3, 6, 9, 12, 14, and 16 ms). pH, pO₂, and pCO₂ levels in the phantoms were confirmed with a blood gas analyzer before and after imaging (Radiometer ABL 5).

Mouse Brain pH Measurements.

pH sensors (pHOptica) were custom-clad in MRI-compatible PEEK tubing (PlasticsOne) and assembled by World Precision Instruments and PreSens Inc. Sensors were implanted into the amygdala as described (7). Twenty-four hours postimplantation, mice were anesthetized with ketamine/xylazine and imaged on a Varian 4.7-T scanner. CO₂ (10% and/or 20%) was administered by nasal cannula to lower brain pH as described (7), and NaHCO₃ (5 mmol/kg, i.p.) was administered to raise pH as described (7, 29). T₁ρ images were collected by using a fast spin-echo sequence [time to echo (TE) = 12 ms, time to repetition (TR) = 2,000 ms, field of view (FOV) = 30 × 30 mm, imaging matrix size = 256 × 128, slice thickness = 1 mm] with spin-lock durations of 10, 20, 40, and 60 ms and B₁ = 1,000 Hz. T₁ρ maps were generated for each condition, and a 5 × 5 region of interest was placed at the tip of the fiber-optic probe to study the relationship between T₁ρ times and pH measured via the fiber optic sensor.

Functional Brain Imaging (BOLD, T₁ρ, and ¹H MRS).

All human research protocols were approved by the University of Iowa Institutional Review Board. Multimodal functional imaging was performed on six subjects (four males and two females, age 28–35 y). Functional T₁ρ images were collected by using an echo-planar spin-echo sequence (TE = 12 ms, TR = 2,200 ms, FOV = 220 × 220 mm, matrix size = 64 × 64, and slice thickness/gap = 4/1.0 mm) with three spin-lock pulses (10, 30, and 50 ms) and a B₁ frequency of 400 Hz. This sequence had a temporal resolution of 6.6 s per T₁ρ measurement. BOLD imaging was performed by using a T₂* weighted echo-planar gradient-echo sequence (TE = 30 ms, TR = 2,000 ms, FOV = 220 × 220 mm, matrix size = 64 × 64, and slice thickness/gap = 4.0/1.0 mm). For BOLD imaging, seven cycles of flashing checkerboard and visual fixation were presented with an 80-s period. For functional T₁ρ imaging, five cycles were collected with a 72-s period. The ¹H MRS data were acquired by using a single-voxel point-resolved spin-echo sequence with water suppression. For functional ¹H spectroscopic imaging, the task began in the baseline task followed by the visual activation condition and returning to the baseline condition. For all activation studies, attention was ensured by asking subjects to press a button in response to a red square presented in the center of the screen every 4 s.

All BOLD fMRI data were analyzed by using standard preprocessing steps, including motion correction, slice timing correction, and spatial smoothing. A general linear model was used to generate individual statistical maps and calculate signal change. T₁ρ data were preprocessed by first performing motion correction followed by T₁ρ map generation. T₁ρ data were spatially smoothed, statistical maps were generated by using a general linear model, and estimates of T₁ρ time changes were computed. BOLD percent signal change and T₁ρ time changes were mapped to MNI space where a t test was performed across the subjects and corrected for multiple comparisons by using a false discovery rate analysis.

The two ¹H spectroscopic measurements obtained for each condition were frequency and phase corrected and averaged, and the resulting spectral data were analyzed by using LCModel. Ratios of Lactate/Cr and Lac/NAA were obtained and compared between (i) baseline, (ii) activation, and (iii) recovery periods using ANOVA.

³¹P Spectroscopic Functional Measures.

Functional data were acquired in six subjects (male/female = 4/2; ages = 22–33 y). A 2D ³¹P spectroscopic sequence used a free induction decay acquisition (TE = 2.3 ms, TR = 4,000 ms, FOV = 240 × 240 mm, matrix = 8 × 8, thickness = 30 mm, averages = 16, vector size = 1,024). This acquisition was repeated three times (baseline, activation, baseline) with each measurement taking 10 min 24 s. ³¹P data were analyzed by using the Siemens Syngo software to determine the chemical shift of the inorganic phosphate (Pi) and phosphocreatine (PCr) peaks in the ³¹P spectra. The analysis included frequency filtering, frequency and phase correction, baseline correction, and curve fitting with prior knowledge. Brain pH was estimated by using the proposed equation (30):

where δ is the chemical shift between in ppm between Pi and PCr. The pH estimates for the baseline, activated, and recovery phases were compared by using ANOVA.