Sodium MRI revisited

It has been 35 years since Sadek Hilal and colleagues demonstrated the feasibility of brain sodium MRI in the cat.1 The early years of 23Na MRI were filled with excitement over the potential to non-invasively investigate, in vivo, electro-physiologic, metabolic processes crucial to life. Unfortunately, early studies suggested this goal would be difficult to achieve due to the low signal to noise of 23Na MRI and theoretical and experimental findings demonstrating that intracellular and extracellular sodium signals could not be well differentiated without relatively toxic shift reagents. Recent years have seen revitalized interest in 23Na MRI due to major technical advances, however in many cases the 23Na MRI basic principles demonstrated early on have not been taken into consideration, and literature misstatements and data misinterpretations are unfortunately common. Here we restate these principles, discuss how literature errors slow progress, and point towards continued opportunities for 23Na MRI in medicine. The small T2 time constant found in many sodium MRI studies is a consequence of the electric quadrupolar nature of the sodium nucleus; very transient interactions between the 23Na electric quadrupolar moment and tissue molecular electric field gradients catalyze NMR relaxation. These electric field effects are much stronger than local magnetic field interactions (responsible for water proton relaxation) and hence result in very small T2 values. The corresponding fast relaxation rate constants do not imply “binding” of sodium in the chemical kinetic sense. The use of the adjectives “bound,” “restricted,” and “mobile” (Supporting Information Material) connote molecular conditions that are generally not present. This should be discontinued. 23Na has a nuclear spin quantum number of 3/2. There are three single quantum transitions between the four nuclear spin states, and theory prescribes the T2 relaxation of the magnetization of a single pool of tissue sodium will exhibit two different time constants. (In saline, the latter are equal, with an apparent single T2 value.) Because of the quantum mechanical properties of the nuclear spin states, 60% of the magnetization will relax as the faster (smaller T2) relaxation component and 40% as the slower (larger T2) component. Again, this is true for a single “pool” of Na+ ions. The relaxation of sodium magnetization will be a superposition of multiple biexponential curves if homeostatic sodium exchange between multiple environments is sufficiently slow. When this relaxation is arbitrarily force-fitted with a single biexponential decay, the relative percentages of the two components can be any value and should not be constrained to 60:40. Indeed, this was noted even for intracellular 23Na alone,2 measurements which were enabled through the use of shift reagents (SRs). Extracellular SRs are paramagnetic lanthanide chelates that shift the extracellular 23Na resonance frequency, allowing for direct studies of the intracellular sodium signal in tissues where the shift reagent penetrates the extracellular space (e.g., see previous works3, 4). The characteristics of SR-free extracellular sodium can be obtained by subtracting the intracellular signal (obtained with SR) from the SR-free total signal.5, 6 Because of the strong calcium binding of SRs, and their required doses, they are too toxic to be used for humans in vivo, yet they allow informative measurements in excised tissues and in animals; for example, such studies revealed that specific inhibition of Na+,K+-ATPase (NKA) activity (e.g., with ouabain) causes an increase in intracellular Na+.4, 7, 8 Initial studies of cells and isolated organs in suspension with SR demonstrated small T2 for intracellular sodium.2, 9 The assumption that extracellular sodium would have large T2 (closer to saline) led to the prospect that T2 could be used to identify intracellular sodium in vivo. Unfortunately, accumulated evidence suggests that tissue extracellular sodium T2 values are very similar to intracellular. Interstitial 23Na in a perfused heart was shown to have a small T2 component value of approximately 2 ms,5 the same as the intracellular.2 Cartilage, essentially an extracellular matrix, has a small T2 value, on the order of 1 ms.10 Even a simple 8% agarose gel has a small T2 of 2 ms.11 It was hoped that double- and triple-quantum 23Na coherences would discriminate intra- and extracellular signals. However, the labile macromolecular interactions that result in similar intra- and extracellular small T2 values are expected to result in similar multiple quantum coherences from both the intra- and extracellular spaces. The contribution of extracellular sodium to multiple quantum coherences has been demonstrated experimentally.12-14 In retrospect, and with appreciation for the complexity of the extracellular environment, the similarities of intra- and extracellular relaxation properties are not surprising. To our knowledge, there is no evidence (that does not violate one of the principles above) showing that relaxation characteristics can unambiguously distinguish intra- from extracellular sodium MR resonances in biological tissues. Because intracellular and extracellular sodium cannot be differentiated in human studies, the common, and relatively strong, single-quantum sodium MRI signal gives the “tissue sodium concentration” (TSC).15 This is not a physical concentration; there is no physical space that actually has that value of sodium concentration. Rather, TSC is the volume fraction-weighted average of the intra- and extracellular sodium concentrations. Image voxels generally contain more than one million cells. Changes in TSC reflect changes in intracellular sodium concentration, extracellular sodium concentration, intra-/extracellular volume fractions, or (usually) several of these interdependent and possibly competing factors (the “Hilal ambiguity”16, 17). For example, increases in intracellular Na+ concentration and intracellular volume fraction by themselves cause an increase and decrease, respectively, in TSC. But these usually occur together (commonly as a consequence of NKA compromise), with the former osmotically driving the latter. Therefore, while TSC may be extremely useful as a bulk tissue parameter, care must be taken both in its quantitation and interpretation of underlying physiological implications, particularly of TSC changes (see Supporting Information Material). As in any MRI experiment involving quantitation, care must be taken with technical considerations such as B0 and rf homogeneity and calibration standards. With respect to sodium concentration, additional care must be taken because of the T2 relaxation characteristics, and the possibility of magnetization that is undetectable even with very short echo times must be kept in mind. Another aspect of quantitation is the “partial volume effect”; when a voxel contains two (or more) tissues. Particular care must be taken if there is fluid (e.g., cerebrospinal fluid or synovial) with very high sodium within a voxel. One suggestion has been to invert the magnetization and image at a null point in the fluid signal recovery.18 However, the tissue will now be averaged within a voxel with a partial null signal, also incurring error. Many studies attempt to utilize TSC along with assumptions about certain tissue compartments to calculate other tissue parameters such as intracellular sodium concentration. However, these assumptions might incur large errors, given the over an order of magnitude larger extracellular sodium concentration than intracellular. A small extracellular volume fraction change can yield TSC changes comparable to large intracellular sodium concentration changes, and thus assumptions about these compartmental parameters can lead to large errors. Therefore, it is imperative that when such assumptions are made, they are explicit, and potential errors associated with the assumptions be calculated and discussed. The Supporting Information Material accompanying this letter presents a number of specific literature examples from this past year, demonstrating the persistent misinterpretation of concepts delineated above. There are a number of problems associated with the continued misstatements and misinterpretations: First, having errors in the literature should as a practice be avoided and corrected. Second, it leads to suboptimal study design; for example, the continued development of relaxation time measurements with the motivation that it will better yield information on compartmental sodium concentrations. Finally, modeling based on inappropriate assumptions might prevent investigators from considering physiological explanations that might not align with those assumptions. The Supporting Information Material has an example of how rigorous interpretation of 23Na TSC changes might have informed a realistic metabolic mechanism for human cerebral stroke some years ago. The main goal of this letter is to review the basic principles of 23Na MRI, which have been misstated in the literature over the years. It is not meant to downplay the numerous studies demonstrating the potential of 23Na MRI. While the initial promise of utilizing sodium MRI to directly monitor electrophysiological events has not yet been realized, new noninvasive water proton MRI techniques for quantifying cell volume fraction19 have potential to be combined with sodium TSC data to better obtain physiological and pathophysiological information. Aside from direct electrophysiological information, many recent reports point toward other clinical applications of 23Na MRI. The Supporting Information Material section points to some of these from the past year. Our opinion is that the potential of 23Na MRI will be best fulfilled by studies which design protocols and interpretations with the above 23Na MR principles in mind, and which focus on specific and practical implementations that have a clear path to impact.20 Through this, we will best utilize resources for targeting unmet medical needs, and responding to those with the unique, and now practical, technology of sodium MRI. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.