|Home | About | Journals | Submit | Contact Us | Français|
To validate 23Na twisted projection magnetic resonance imaging (MRI) as a quantitative technique to assess local brain sodium concentration ([Na+]br) during rat focal ischemia every 5.3 minutes.
The MRI protocol included an ultrashort echo-time (0.4 msec), a correction of radiofrequency (RF) inhomogeneities by B1 mapping, and the use of 0–154 mM NaCl calibration standards. To compare MRI [Na+]br values with those obtained by emission flame photometry in precision-punched brain samples of about 0.5 mm3 size, MR images were aligned with a histological three-dimensional reconstruction of the punched brain and regions of interest (ROIs) were placed precisely over the punch voids.
The Bland–Altman analysis of [Na+]br in normal and ischemic cortex and caudate putamen of seven rats quantitated by 23Na MRI and flame photometry yielded a mean bias and limits of agreement (at ±1.96 SD) of 2% and 43% of average, respectively. A linear increase in [Na+]br was observed between 1 and 6 hours after middle cerebral artery occlusion.
23Na MRI provides accurate and reliable results within the whole range of [Na+]br in ischemia with a temporal resolution of 5.3 minutes and precisely targeted submicroliter ROIs in selected brain structures.
The apparent diffusion coefficient (ADC) of tissue water is an established MRI marker for initial ischemic damage to the brain (1). Recently, 23Na MRI has been explored as a complementary MRI technique for ischemic stroke characterization, and proposed as a means to determine precisely the stroke onset time for establishing patient eligibility for thrombolytic therapy (2). 23Na MRI timing of stroke is based on the linear increase in brain sodium concentration ([Na+]br) in affected areas (2–4) in the first several hours (“ticking clock” concept). It reflects metabolic disruptions, in particular, in the functionality of Na/K-ATPase and other ion pumps. The possibility to address brain tissue viability after stroke by 23Na magnetic resonance imaging (MRI) has also been considered elsewhere (5,6). To obtain better insight into physiological and functional changes in brain tissue when sodium concentration reaches a certain threshold, quantitative techniques are necessary.
MRI is the only noninvasive experimental technique suitable to establish [Na+]br as a potential biomarker because it enables the study of concentration dynamics in vivo in a single animal. Early studies have shown an agreement between [Na+]br values measured by 23Na MRI and 22Na radionuclide dilution assay in a normal rat brain and in a rat glioma model (7,8). These studies used long 23Na imaging times (≈1–3 hours) to improve the signal-to-noise ratio, and either the whole brain or several-millimeter-thick slabs were required for the radionuclide assay.
Stroke applications present 23Na MRI quantitation with new challenges. Ischemic stroke is characterized by a broad range of [Na+]br values (which may reach close to the [Na+] of 140 mM in blood plasma). Spatial inhomogeneity of sodium accumulation in cerebral ischemia (2) calls for the [Na+]br quantitation over small regions. Regional measurements of [Na+]br with MRI and radionuclide techniques were performed in brain tumors (8), but these regional values were not compared directly. Thus, to our knowledge, absolute sodium concentration mapping has not been addressed in small brain regions, and not in rat cerebral ischemia. Last but not least, to follow the dynamics of the [Na+]br increase during evolving acute stroke, the imaging time should not exceed several minutes.
The present work was undertaken to validate and calibrate 23Na MRI as a quantitative technique suitable for rat brain ischemia with submicroliter regions of interest (ROIs) and a temporal resolution of 5.3 minutes. A systematic comparison with the postmortem [Na+]br determination was performed in precisely the same ROIs using a gold standard technique, emission flame photometry (9). This makes the scope of this study fundamentally different from that of previous investigators (2,3,8). To be able to assess physiological and functional changes in the brain occurring at certain sodium concentrations, the issue of quantitation, as addressed here, is instrumental.
Approval for animal use was obtained from the appropriate institutional committee and was consistent with the “Principles of Laboratory Animal Care” (NIH publication No. 86-23, revised 1985). Seven normally fed male Sprague–Dawley rats weighing 300 ± 16 g (mean ± SD) were used. Anesthesia was induced with 3% isoflurane and maintained with 1.0%–2.5% isoflurane, 30% oxygen, and balance nitrous oxide, administered via endotracheal tube and artificial respiration (Model 681, Harvard Apparatus, South Natick, MA) Femoral arterial and venous catheters were inserted. Inside the magnet an MR-compatible ventilator (MRI-1, CWE, Ardmore, PA) was used. An appropriate maintenance level of isoflurane was determined by monitoring the blood pressure response to tail pinch. Body temperature was maintained at 37°C by a servocontrolled system consisting of a rectal temperature probe and a heating blanket outside the magnet or a thermostated water jacket inside the magnet. Immobilization was implemented with 0.4 mg/kg pancuronium bromide injected intramuscularly (IM) at 60-minute intervals (on the bench) or continuously infused intravenously (IV) at 0.4 mg/kg/h (delivered at 1 mL/h) inside the magnet. Arterial blood pH and gases (PaCO2, PaO2) were measured (ABL-3, Radiometer America, Westlake, OH) at up to 11 timepoints to ensure physiological stability. Arterial blood pressure was continuously monitored from a femoral artery using a strain gauge transducer (DT-XX, Viggo Spectramed, Miami, FL) and recorded on a polygraph (Gould, Cleveland, OH). Focal cerebral ischemia was produced by middle cerebral artery occlusion (MCAO) using insertion of an intraluminal suture. The 3-0 monofilament poly-L-lysine-coated nylon suture was inserted 20–21 mm through the internal carotid artery and further into the circle of Willis, occluding the middle cerebral artery (MCA) at its origin (10).
For 23Na/1H MRI, the animal’s head was positioned inside a 5-cm-diameter, 5-cm-long dual-tuned dual-quadrature birdcage transmit/receive RF coil (11) in the animal cradle with a recirculating water bed and fittings for respiratory and anesthesia gas supply. Cylinder tubes containing NaCl solutions at different concentrations (0, 77, 116, and 154 mM) were placed next to the animal’s head and served as external position and concentration references. Images were obtained on a 3 T whole body scanner (General Electric Medical Systems, Milwaukee, WI) within a field of view (FOV) of 50 × 50 × 50 mm. 1H diffusion-weighted multislice spin-echo images (repetition time/echo time (TR/TE) 2000/140 msec, in-plane resolution 0.2 mm, 8 slices 3.2-mm thickness each, diffusion weighting b-factor values of 0, 93, 372, and 837 s/mm2, scan time per b-factor 4.7 minutes), with the diffusion-sensitizing gradient applied along each of the Cartesian axes, were used for reconstruction of ADC trace maps. For 23Na MRI, a three-dimensional (3D) twisted projection imaging (TPI) scheme (12) with a voxel size of 0.48 mm3 was applied. Every 5.3 minutes, eight acquisitions were averaged for each of 398 projections using TR/TE of 100/0.4 msec. The inhomogeneity correction of the B1 field was performed by RF mapping as described elsewhere (13). The 23Na MRI series typically spanned 2–4 hours within the 1–6-hour window after ischemia.
After the end of MRI scanning (typically, 4.2–5.9 hours after MCAO), rats were decapitated and their heads were immediately frozen and kept at −80°C in order to preserve the spatial characteristics of the brain inside the skull for further juxtaposition with MR images. The brain was chipped out of the skull in a −20°C cold box and mounted into the cryostat (−8°C). Twelve to 18 samples of ≈0.5 mg wet weight were punched from the ipsilateral and contralateral brain (9) at 2–3 coronal levels, typically between + 1 and −4 mm from bregma. The micro-puncher inner diameter was 0.53 mm, the sampling depth was determined by examining coronal brain cuts taken every 40 µm, and the samples were precision-weighed using a Cahn model C-44 microbalance (ATI Orion, Boston, MA). The sampling location was guided by ADC and 23Na maps of the brain (Fig. 1a,e) and the change in surface reflectivity of ischemic tissue (14) to make sure that homogeneous samples of ischemic and normal brain in a wide range of Na content were obtained. Cut-face photographs of the brain were taken at several levels, including punched surfaces before and after sampling (Fig. 1b,c). Na content in punched samples was determined by emission flame photometry at 589 nm using an IL943 flame photometer (Instrumentation Laboratory, Lexington, MA). The 40-µm-thick coronal sections of the brain at different levels from bregma were mounted on glass slides, digitized, and reflective changes were used to outline the infarct area, as shown in Fig. 1d.
Parametric 1H ADC maps were generated pixel-wise by exponential fitting of the diffusion-weighted image intensity versus the b value (15) in MatLab (MathWorks, Natick, MA). 23Na MR images were reconstructed, corrected for inhomogeneity of the B1 field and for the nonzero noise baseline in the magnitude mode reconstruction (13), and stacked in 4D (including the time dimension) using C and C++ scripts in the UNIX environment. Digitized brain slices (taken every 400 µm at different levels from bregma) were stacked and registered using ImageJ (16) (available from W.S. Rasband, ImageJ, National Institutes of Health, Bethesda, MD; http://rsb.info.nih.gov/ij/, 1997–2007) to render volumetric reconstructions of the brain. MR images were aligned with histological 3D images and cut-face photographs and analyzed using AMIDE software (17). To compare [Na+]br values obtained by MRI and flame photometry, cylindrical ROIs in the MR images were placed exactly at the positions of punch holes on histological and cut-face images, as shown in Fig. 1. Average 23Na MR image intensities over selected ROIs in the brain were referenced to those from ROIs placed over the reference tubes and corrected for partial signal saturation in the NaCl solution due to its longer T1 (T1 = 60 msec, correction factor 1 – exp(−TR/T1)). Tissue 23Na with T1 of 10–30 msec is fully relaxed in these conditions. To account for partial volume effects, the ROIs over the reference tubes extended beyond the tube diameter in the transverse plane, thus ensuring that the entire signal coming from solution in the tube was integrated (18). The integrated signal was subsequently normalized to the actual cross-section of the tube, so that no underestimation of the mean occurs. Although tubes with different NaCl concentrations were placed within the FOV with the intent to average [Na+]br values obtained by MRI using different reference NaCl concentrations, eventually all final values were obtained using only the 154 mM solution for consistency. The Bland–Altman analysis of agreement (19) was performed using MedCalc for Windows, v. 188.8.131.52 (MedCalc Software, Mariakerke, Belgium).
Physiological variables at different phases of the experimental protocol were within the normal range for all animals, as summarized in Table 1. Although one-way analysis of variance (with a post-hoc Bonferroni correction for multiple comparisons) reveals statistically significant fluctuations in MABP, pH, and paCO2, these fluctuations were not relevant for the purpose of comparison of [Na+]br obtained by MRI and elemental analysis. No correlation between fluctuations in physiological variables and 23Na was observed in the course of the experiment.
The changes in 23Na signal intensity after MCAO were analyzed in the ipsilateral and homotopic contralateral frontal cortex, parietal cortex, and caudate putamen (Fig. 1). To obtain the [Na+]br values by MRI and to compare them with corresponding [Na+]br values obtained from the same ROIs by a standard analytical technique, emission flame photometry, reference tubes with NaCl solutions of known concentrations placed within the FOV were used for calibration, as described in Materials and Methods.
23Na MRI intensity showed a linear increase in ischemic brain and no statistically significant changes in contralateral ROIs over time (Fig. 2). To estimate [Na+]br in ipsilateral ROIs at the end of the experiment, 23Na image intensities were extrapolated to the decapitation time using linear regression, as shown in Fig. 2. These “final” [Na+]br values showed a satisfactory correlation with the flame photometry [Na+]br data, as demonstrated in Fig. 3a. The agreement between the two techniques may be better assessed using the Bland–Altman plot (20) plotting the differences between the two measurements as percentage of the averages (19). The plot in Fig. 3b shows good agreement between 23Na MRI and flame photometry data. The mean value of the relative difference between flame photometry and MRI results (mean bias) and limits of agreement (at ± 1.96 SD) were 2% ± 43% of average, and 95% confidence intervals were ±4% of average for the mean bias and ±8% of average for the upper and lower limits of agreement.
The results of this study present a first documented validation of the absolute quantitation of distribution and accumulation of sodium in ischemic rat brain by 23Na MRI. Monitoring of ion fluxes is a promising approach to understand the pathophysiology of stroke, and Na+ is of particular interest for noninvasive MRI monitoring because of the high NMR sensitivity of 23Na. Quantitation of the sodium signal, however, may be significantly biased if the 23Na fast biexponential T2 relaxation is not taken into account (21). This requires TE below 0.5 msec. In earlier quantitative studies, two short-TE imaging protocols were tested: projectional MRI with total imaging time of 2.84 hours (7) and TPI with imaging time of 53 minutes (8) or 26 minutes (unpubl. data), and the whole rat brain or large fragments of a brain tumor were used for a radionuclide biochemical [Na+]br assay.
In comparison with those earlier reports (7,8), currently validated quantitative methodology possessed much shorter temporal resolution of only 5.3 minutes and precisely targeted very small ROIs in selected brain structures rather than the whole brain. The MRI protocol included the ultrashort TE (0.4 msec) to minimize a quantitation bias, the B1 mapping to correct for RF inhomogeneities, and the use of calibration standards. The robustness of quantitation with respect to the susceptibility artifacts around the calibration standards was assured by a combination of the ultrashort TE, 3D volume excitation, FID acquisition mode, and by integration of the signal from the standards beyond the borders of the tube. A TR of 100 msec resulted in the absence of signal saturation in both normal and ischemic brain. Although the key features of the MRI protocol have been developed in earlier publications (11–13), the feasibility of 23Na MRI comparison against the gold standard, emission flame photometry, in small ROIs hinged upon two crucial advances in the present study: 1) precise directed sampling of normal and ischemic cortex and caudate putamen, and 2) alignment of MR images with a 3D reconstruction of the punched brain for precise ROI placement. Our results in the rat model demonstrate that 23Na MRI provides accurate and reliable results within the whole range of [Na+]br in ischemia. The mean bias of 2% of average is close to zero, and difference values do not show significant systematic variation over the range of measurement (Fig. 3b). One may notice a tendency of slightly lower 23Na MRI values at higher [Na+]br. This residual quantitation bias cannot be attributed to relaxation time differences, because the combination of TR/TE ensured full recovery of the signal. Rather, its sources may include higher noise contribution at low signal intensity due to feasible inaccuracies in estimation of the nonzero noise baseline in the magnitude mode MR reconstruction, as well as higher MR signal loss attributable to the partial volume effects in the sites with elevated sodium content than in the rest of the brain.
In agreement with earlier studies (2–4), a linear increase in [Na+]br was observed during evolution of cerebral ischemia between 1 and 6 hours after MCAO (Fig. 2). The precision of intersection of linear regressions for ischemic and control ROIs (i.e., of the stroke onset time error) was previously estimated as 1 ± 4 minutes in a model of cortical stroke (2).
In conclusion, good agreement was achieved between 23Na MRI and direct determination of [Na+]br by flame photometry in a small animal model with a small voxel size. It is reasonable to expect that even more accurate regional quantitation of [Na+]br is possible in the clinical setting where the requirements to the signal-to-noise ratio imposed by the voxel size are less demanding.
The authors thank Jayashree Kanchana for early efforts in data processing and Jayjayantee Dasgupta for technical support.
Contract grant sponsor: National Institutes of Health; Grant number: NS30839.