CONTRIBUTIONS TO MAGNETIC SUSCEPTIBILITY OF BRAIN TISSUE

Abstract

This review discusses the major contributors to the subtle magnetic properties of brain tissue and how they affect MRI contrast. With the increased availability of high field scanners, the use of magnetic susceptibility contrast for the study of human brain anatomy and function has increased dramatically. This not only has led to novel applications, but also has improved understanding of the complex relationship between MRI contrast and magnetic susceptibility. Chief contributors to magnetic susceptibility of brain tissue have been found to include myelin as well as iron. In the brain, iron exists in various forms with diverse biological roles, many of which are only now starting to be uncovered. An interesting aspect of magnetic susceptibility contrast is its sensitivity to the microscopic distribution of iron and myelin, which provides opportunities to extract information at spatial scales well below the MRI resolution. For example, in white matter, the myelin sheath that surrounds axons can provide tissue contrast that is dependent on axonal orientation and reflect the relative size of intra- and extra-axonal water compartments. Extraction of such ultra-structural information, together with quantitative information about iron and myelin concentrations is an active area of research geared towards characterization of brain structure and function and their alterations in disease.

INTRODUCTION

MRI is exquisitely sensitive to the subtle variations in magnetic susceptibility across the human brain, especially when using so-called “susceptibility-weighted” (SW) techniques like gradient-echo MRI at high field. These techniques are increasingly being used to study the brain’s vasculature, and regional variations in tissue myelin and iron content. At high field, they provide exquisite anatomical detail not available with other techniques (Figure 1) (1,2). SW contrast also forms the basis of blood oxygenation level dependent (BOLD) fMRI, which has revolutionized the study of brain activity by recording the susceptibility changes resulting from changes in blood oxygenation (3).

Figure 1.

SW MRI of ex-vivo hippocampal tissue. Signal amplitude (left) and phase (right) show distinctly different contrast. Regions such as dentate gyrus (DG), cornu ammonis (CA), subiculum (S), and parahippocampal gyrus (PHG) are readily identifiable. Strong contrast is seen in and around vessels (yellow arrows), at pyramidal cell layers (red arrow) and in white matter (speckles indicated with white arrow) attributed to iron in ferritin and deoxyhemoglobin. Measurements performed at 7 T, TE=30ms, 50x50x500mm resolution. Scale bar = 2 mm.

Along with the optimization and increased use of SW MRI, major advances have been made in the understanding of its contrast mechanisms. For example, it is now well understood that susceptibility contrast is not only affected by tissue composition, but also by the orientational order resulting from its cellular and molecular structure. At the same time, further details about the complex biological role of iron, a primary contribution to magnetic susceptibility, are becoming available. Here, we will review recent progress in the understanding and interpretation of magnetic susceptibility contrast and how this is applied to study changes with disease. The review is written in the context of previous literature that has reviewed this topic in the past (4–7).

TISSUE MAGNETIC SUSCEPTIBILITY

Magnetic susceptibility refers to the magnetizability of a material when placed in a magnetic field B₀. It is indicated with the symbol χ, which represents a proportionality constant between the material’s macroscopic magnetization M and local magnetic field B ( $M=\chi \frac{B}{{\mu }_0}\approx \chi \frac{B_0}{{\mu }_0}$, with μ₀ a physical constant representing the magnetic permeability of vacuum, and equal to 4π. 10⁻⁷ H m⁻¹). A material’s χ is dependent on its molecular constituents, and originates from spins and motions of nuclei and their electrons. χ can be positive or negative, reflecting whether magnetization aligns with the field (paramagnetism), or opposes it (diamagnetism). Paramagnetism generally originates from field-induced alignment of unpaired electron spins, whereas diamagnetism is associated with field-induced alteration of electron orbits (8).

Each electron of an atom has a spin of s = ± ½ and the total spin, S, is the sum of the spins of each individual valence electron. The vast majority of biological molecules have even numbers of electrons and all their spin-up electrons are paired with spin-down electrons giving each of these molecules a total spin S = 0. These are called closed shell molecules and have a very feeble negative magnetizability and are called diamagnetic. For many purposes (but not in MRI) this effect can be ignored and the molecules are then called nonmagnetic.

A very small fraction of biological molecules, especially those containing transition metal elements such as iron, copper and manganese, have unpaired valence electrons. This gives these open shell molecules and ions a magnetizability of the opposite sign and roughly a thousand times stronger than that of closed shell structures. Using heme in hemoglobin as an example the magnetic behavior of ferric and ferrous iron ions is indicated in Figure 2. The d-shell of iron provides five energy levels or orbitals each potentially containing one spin-up and one spin-down electron. Spherical harmonics corresponding to these orbitals are indicated on the left of the figure. Here the z-axis is perpendicular to the plane of the heme and r² = x² + y². The spin degree of freedom leads to the d-shell providing a total of ten spin orbitals that may be occupied by valence electrons. Ferrous iron has six valence electrons to distribute over these ten spin orbitals and ferric iron has five. A paramagnetic molecule or ion will have a total magnetic moment, m, given by m = μ_effμ_B where μ_B = 9.274 10⁻²⁴ J T⁻¹ is the Bohr magneton. The effective number of Bohr magnetons, μ_eff, is often given quite accurately by the spin-only approximation, ${\mu }_{\mathit{eff}}=2\sqrt{S(S+1)}$ (8). Ions such as iron in the brain do not exist in isolation but interact with their environment. That is, they will be bonded to some degree to adjacent ions or molecules and these ligands will modulate the energy levels and the μ_eff of the spin orbitals. Therefore, the energy levels and spin values shown in Figure 2 are approximations and will be modified somewhat in a real situation depending on the details of the ligands bonded to the iron atom.

Figure 2.

Correlation of iron oxidation states with spin states. Using heme as an example, the diagram shows how the magnetic properties, determined by the total spin S, correlate with the oxidation states to produce high spin and low spin versions of both ferrous and ferric iron. Hgb refers to hemoglobin and cyt c⁺⁺ refers to cytochrome c. Modified from: Perutz MF. Molecular anatomy, physiology, and pathology of hemoglobin. In: Stamatoyannopoulos G, editor. The molecular basis of blood diseases. Philadelphia: Saunders; 1987. p 127–178.

Human brain tissues have a rather weak (diamagnetic) magnetic susceptibility that is dominated by that of water (χ = −9.05 ppm) due to its large abundance (70–85% of brain tissue is water). Across grey and white matter of healthy brain, it varies within a range from about −9.2 to −8.8 ppm. This variation is primarily caused by variations in iron and myelin content, which both have sufficient concentration and susceptibilities sufficiently different from water to contribute substantially. Lipids in myelin are diamagnetic (relative to water), and as a result, heavily myelinated white matter is generally the most diamagnetic tissue in healthy brain. Somewhat outside the range for χ mentioned above is blood inside the venous vasculature, where iron in deoxyhemoglobin renders χ slightly more paramagnetic (fully deoxygenated blood: χ = −7.9 ppm). A wider range for χ may also be found under pathological conditions. For example, calcifications may result in a more diamagnetic χ(9), whereas accumulation of iron (e.g. associated with neurodegeneration or as a result of microbleeds) may render χ more paramagnetic. After a brief review of the effect of magnetic susceptibility on MRI, further details about the biological role, magnetic properties, and brain distribution of iron and myelin χ will be discussed below.

EFFECTS OF SUSCEPTIBILITY ON MRI

The effects of tissue susceptibility variations on the magnetic field can be studied with SW MRI through acquisition of one or multiple gradient echo signals, from which amplitude and frequency are extracted. Tissue susceptibility can affect both amplitude and frequency through the resonance frequency f of water ¹H protons, which depends on the field B they “sense” according to $f=\frac{\gamma }{2\pi }B$. B is a combination of the applied field (static field B₀ of the MRI scanner) and the field shift ΔB created by the magnetization M of the object. With the latter relating to χ according to $M\approx \chi \frac{B_0}{{\mu }_0}$, we have ΔB = μ₀M = χB₀. Although variations in χ are generally small (on the order of 0.1 ppm), they can lead to easily detectable frequency shifts that scale with static field strength: at 7 T, typical frequency shifts are on the order of a few Hz.

The field shift ΔB resulting from a spatially varying M(r) can be found by (spatially) convolving M with the magnetic field generated by a point dipole (an infinitely small magnet with unit magnetization). For the simple case of an infinitely extending uniform magnetization M this dipole convolution leads to a uniform field shift ΔB = μ₀M; an outer boundary that is geometrically simple (e.g. cylinder, sphere), leads to an additional field shift that can be taken into account by the use of so-called “demagnetization”, “demagnetizing”, or “shape” factors (here indicated with S), which e.g. for a sphere results in a subtraction of a factor $\frac{1}{3}{\mu }_{0}M$ (10).

Unfortunately, further complication results from the fact that, while χ and M are defined macroscopically, mechanistically they result from (sub-) atomic scale dipoles. Thus, only with gross simplification (and ignoring statistical mechanics) is it possible to characterize the actual field ΔB sensed by a water ¹H proton. Commonly used is the so-called “Lorentz sphere” approach, which models the sensing nucleus to reside in a spherical cavity, surrounded by a macroscopic continuum magnetization (see e.g. (10,11)) (Figure 3). Again, considering uniform magnetization and using shape factors, the spherical inner boundary of the magnetization (outer boundary of cavity) leads to $\mathrm{\Delta }B=\frac{1}{3}{\mu }_{0}M$ (see e.g. (11)). Interestingly, in the aforementioned situation of a finite object with spherical outer boundary, we would have $\mathrm{\Delta }B=\frac{1}{3}{\mu }_{0}M-\frac{1}{3}{\mu }_{0}M=0$, leading to the somewhat counter-intuitive finding that the average, magnetization-related field shift sensed by water protons (ΔB)is zero inside a spherical, uniformly magnetized object!

Figure 3.

Simplified model of magnetic environment of water proton (white sphere at center of each image) used to calculate the magnetic field it senses and the resulting MRI resonance frequency. Magnetic environment is assumed to result from nearby water dipoles, while effects from bonds (represented in chemical shift) are ignored. When considering the extensive averaging over the many sensing protons in a voxel, and the many rapidly moving magnetic sources of water molecules that surround them, the magnetic environment can be approximated by a continuum with a spherical cavity that for calculation purposes can be as small as the sensing proton. Molecular model of water courtesy of Michael Bruist, University of the Sciences, Philadelphia.

To estimate the magnitude of frequency shifts in brain tissue, we consider the simple case of a long cylinder of uniform magnetization aligned with the direction of B₀ (e.g. a blood vessel or white matter fiber bundle), in which case we have S = 0 (identical to the case of an infinitely extending object), resulting in $\mathrm{\Delta }B=\frac{1}{3}{\mu }_{0}M$, and leading to $\mathrm{\Delta }f=\frac{\gamma }{2\pi }\frac{1}{3}\chi B_0$, which at 7 T is about 100 Hz for each ppm in susceptibility. Since SW MRI can easily pick up shifts as small as 0.1 Hz (2), susceptibility shifts of only 1 ppb are in principle detectable! This offers unique opportunities to quantify and map subtle variations in tissue iron and myelin content.

Of course, this is not the entire story. Other than macroscopic object boundaries and atomic-scale structure, tissues generally possess structure at various intermediate scales, which complicates interpretation of field shifts in terms of voxel-averaged χ values and underlying variation in tissue constituents. To the extent that this structure is captured by the MRI resolution, its effect can be accounted for by the use of shape factors, or more generally by deconvolution methods, both based on information available from the MRI image. The latter allows one to directly translate frequency distributions into images of χ. This so-called “quantitative susceptibility mapping” (QSM) is a very active area of research and one of tremendous recent progress (for reviews, see (12,13)). Structure that exists on the sub-voxel scale and is not captured by MRI may also confound the interpretation of frequency shifts, as has been emphasized recently (14). Accounting for such structure requires assumptions about its geometry and orientation, and necessarily requires gross simplification. An established approach is to consider the average magnetic environment of water protons in a voxel (10,15,16). Due to the extensive spatial and motional averaging, this average environment may be approximated to have a relatively simple shape (e.g. cylindrical, ellipsoidal, or spherical), in which case their effect on ΔB may be estimated by (again) the use of shape factors (see also the section on tissue structure below).

There are also a couple of caveats with the measurement of frequency shifts: In MRI, field estimates are based on the voxel-averaged resonance frequency of water protons, and thus weighted by their NMR visibility. The latter can be affected by the local environment through relaxation times T₁, T₂, and T₂*, diffusional motion, and magnetization transfer. Secondly, MRI frequency measurements may be sensitive to factors other than susceptibility including e.g. temperature and image artifacts caused by motion, drift, respiration etc.

Field variations are not only reflected in the signal phase and frequency, but may affect amplitude as well. In the absence of motional averaging of the field sensed by protons, field variations within a voxel will result in a distribution of resonance frequencies, leading to a decoherence and signal loss of the proton signals that increases with echo time. By comparing signals acquired at different echo times, the signal decay rate can be quantified. Generally, single-exponential signal decay (rate indicated with R₂*) is assumed, and the decay rate is interpreted as (partly) reflecting the strength of the susceptibility variations.

Both χ and R₂* can be used to study the concentration of susceptibility perturbers such as iron, myelin and deoxyhemoglobin. It is also possible to directly (without calculating χ and R₂*) combine signal amplitude and phase into a single image (17): this may facilitate the visual identification of focal changes in susceptibility and has been applied clinically to visualize vascular abnormalities and local iron accumulation in a range of pathologies (18).

Signal amplitude, phase (and frequency) and extracted parameters R₂* and χ all have been used to improve visualization of detailed anatomical structures, predominantly at 7 T. This includes neo-cortical sub-structure (19–21), the hippocampal and perirhinal cortex (22–24), and many other GM regions including the subthalamic nucleus (25), the zona incerta (26,27), the basal ganglia (28,29), and the substantia nigra (30,31). Many of these sub-cortical GM regions are important targets and surgical landmarks for deep brain stimulation (32).

BRAIN DISTRIBUTION OF IRON

Iron in Brain Tissue

Much of MRI visible iron in tissue is located in ferritin and hemosiderin, associated with proteins that facilitate storage and transport (33). Excessive amounts outside such storage forms may be toxic, result in oxidative stress, and lead to disease (34). In healthy adult brain, total iron concentration varies from 0–≈200 μg per gram tissue, i.e. 200 ppm, with typical white matter (WM) and cortical grey matter (GM) regions being at the lower end of this range (< 60 ppm) (35,36). At the microscopic scale, strong variation exists across cell types, with microglia and oligodendrocytes staining most consistently for iron (36–40). Across WM, macroscopically, variable iron staining is observed, with subcortical WM often showing a “patchy” or “speckled” appearance (see e.g. Figs. 1 and 5)(37,39), and major fiber bundles such as the corpus callosum and optic radiations showing little staining (40). In GM, highest iron concentrations (up to ≈ppm) are found in the dentate nuclei of the cerebellum, and in central brain regions such as the basal ganglia and red nuclei, and often associated with glial cells (37,41). The lower concentrations found in cortex are often associated with oligodendrocytes (37,41).

Figure 5.

Co-localization of iron and myelin, and appearance in SW MRI.

A. In cortical GM of the human occipital lobe, laminar variation in both iron and myelin is seen that is reflected in MRI R₂*. Substantial variation in iron is also seen in WM, with highest iron at the GM-WM boundary and otherwise a generally patchy distribution. These characteristics are also seen in MRI R₂*. Iron stain: Perls’/DAB; Myelin stain: luxol fast blue; MRI: 7 T, TE=15/30 ms, scale 0–100 s⁻¹. Reproduced from Fukunaga et al., PNAS 2010.

B. Varying contribution of iron and myelin to R₂*. Myelin appears to dominate MRI R₂* in the major WM fiber of the optic radiation (black arrows), whereas the reverse is true in the WM immediately adjacent to the cortex (white arrow). Iron stain: Perls’/DAB; Myelin stain: luxol fast blue; MRI: 7 T, TE=15/30 ms, scale 0–80 s⁻¹. Reproduced from Fukunaga et al., ISMRM 2011, 12.

C. Investigation of patchy iron distribution in subcortical WM. Patchy iron distribution appears to be reflected in signal phase of SW MRI R₂* (from same tissue but slightly shifted area). Iron stain: Perls’/DAB; MRI: 11.7 T, TE=20 ms, 12.5 mm resolution. Scale bar = 100 mm. Reproduced from T-Q Li et al., ISMRM 2010, 2304.

Most pathological conditions are associated with increased tissue iron content, primarily in the form of hemosiderin and ferritin. Iron in hemosiderin near sites of prior vascular injury may originate from extravasated blood. More widespread increases occur with iron storage diseases such as aceruloplasminemia, where 2–3 fold increases in iron have been found across much of the brain (42). In many neurological diseases, iron accumulates preferentially in the basal ganglia (for reviews see (34,43,44)), regions where also other metals such as manganese (45) and gadolinium (46) have been found to accumulate. It is possible that the brain uses these regions as storage or dump for materials is does not need. In the rather common neurological disorder, restless leg syndrome, a decrease in iron tissue accumulation is present (47).

Iron in Blood

In blood, iron is predominantly associated with deoxyhemoglobin and normally present at concentrations of about 500 ppm, which is several-fold higher than in healthy tissue. Iron’s paramagnetism changes with blood oxygenation state: fully deoxygenated, it renders blood susceptibility χ_deoxy about 1.2 ppm paramagnetic relative to water (48,49), whereas this paramagnetic behavior vanishes at 100% oxygenation. Thus MRI voxels within or near large vessels experience phase shifts and changes in R₂* that depend on blood oxygenation level (50). This phenomenon provides the mechanism for BOLD contrast in fMRI (3). It also provides an opportunity to look at venous oxygenation and possibly oxygen extraction fraction by studying susceptibility shifts in large veins calculated from the phase shift created in their vicinity. This can be based on R₂* (51,52) or χ (53)

The oxygenation of blood (and thus the oxygenation state of hemoglobin) varies considerably across the brain’s vasculature: where arterial and capillary blood oxygenation may reach 80–90%, in the venous vasculature this may drop to 60% or lower, which corresponds to Δχ_vein ≈ 0.5 ppm. This makes SW MRI preferentially sensitive to the venous vasculature and explains its success in venographic applications (54).

In addition to the oxygenation level, the contribution of blood to the MRI signal depends on its local concentration (i.e. blood volume). Averaged over a size scale of a typical MRI voxel, cerebral blood volume (CBV) may range from 2–4%. Assuming 60% oxygenation, this contributes about 10–20 ppb to the overall susceptibility of tissue. As will be seen in the following sections, this is generally small relative to the contribution of other substances and therefore may contribute little to contrast observed between tissue types observed in SW MRI studies of anatomy (55).

Nevertheless, CBV can be locally significantly higher and is known to substantially vary over brain regions (56,57). Extreme examples are large veins such as the sagittal sinus, or the choroid plexus. As the size of these structures is similar or larger than the typical MRI resolution, apparent CBV values may approach 100%. Thus, the sensitivity of BOLD fMRI and vascular contrast in venography based on SW MRI will vary over brain regions and may accentuate larger veins. This is especially the case for vessels angled with B₀, where field changes extend to surrounding tissue and increase the contrast.

Pathological conditions can not only result in changes in vascular anatomy and oxygenation that may be picked up with SW MRI (58), but may also lead to extravasation of blood. For example, hemorrhagic stroke and traumatic brain injury may cause bleeding into tissue surrounding vessels, and the associated paramagnetic effects can be detected with SW MRI (59). With time, extravasated blood will break down and its iron will be sequestered in hemosiderin, which also has paramagnetic properties and often remains detectable with SW MRI.

EFFECT OF IRON ON MRI CONTRAST

The contribution of iron to the magnetic susceptibility of brain tissue is not only dependent on concentration, but potentially also on form and associated type of magnetism. For ferritin, theoretical estimates indicate that the contribution of iron to χ (Δχ_iron) is 1.4 ppb for each ppm weight fraction of iron (i.e. for each μg iron per gram tissue, see Table 1) (49). This is based on a theoretical estimate of 520 ppm for χ of a ferritin particle fully loaded with 4500 Fe⁺⁺⁺ ions. Magnetometry and MRI measurements with QSM on ferritin phantoms, in-vivo, and ex-vivo brain indicate a range for Δχ_iron of 0.55–1.3 ppb per ppm iron, somewhat below the theoretical estimate (60–67). Although the reason for this is unknown, it appears that χ of a ferritin particle may be somewhat below 520 ppm as would be expected if the ferritin iron loading is incomplete.

Table 1.

Useful conversion factors (mM = millimol/liter = 6.02x10¹⁷ atoms/cc). A brain tissue density of 1g/cc is assumed.

1 μg Fe/g fresh tissue weight	1 ppm weight fraction Fe	1 μg Fe/cc	17.91 μM Fe
1 μg Cu/g fresh tissue weight	1 ppm weight fraction Cu	1 μg Cu/cc	15.74 μM Cu

Due to field inhomogeneities across an MRI voxel associated with iron (e.g. ferritin particles), also the apparent signal decay rate R₂* (as well as R₂′, which excludes effects on the decay other than those caused by field inhomogeneities) may provide a measure of iron concentration (68). Theoretical estimates that assume spherical particles, and ignore effects of water diffusion, suggest R₂′ increases by about 0.11 s⁻¹ per Tesla static field for each ppb in Δχ_iron (69). Assuming Δχ_iron to be 1 ppb for each ppm weight fraction of iron (see above), we would expect R₂′ to change by 0.11 s⁻¹ T⁻¹ per ppm iron. This is about two-fold higher than inferred from measurements, which indicate a range of 0.037 to 0.055 s⁻¹ T⁻¹ per ppm iron for R₂* (which should increase by at least as much as R₂′) (66,70,71). Similarly, studies of the relationship between R₂* and Δχ_iron found a proportionality constant of around 0.05 (range 0.047–0.052) s⁻¹ T⁻¹ per ppb in Δχ_iron (72,73), again about half of the theoretical estimate of 0.11 s⁻¹ T⁻¹ per ppb in Δχ_iron.

A possible explanation for the lower than expected effect of iron on R₂* is the phenomenon of diffusion averaging. When spins diffuse through field gradients around a ferritin particle, differences (i.e. decoherence) between their accumulated phases may be reduced, reducing R₂*. To evaluate the likelihood of this effect contributing, one can consider practical values for diffusion constant D, the particle radius R, and its susceptibility Δχ (69,74). Substantial phase accumulation near a particle occurs at time $\tau =\frac{1}{\mathrm{\Delta }\omega }$, which at 7 T equals 3 μs (Δχ = 520 ppm). With D ≈ 1 μm² ms⁻¹, the diffusion distance in time τ is $\sqrt{\tau D}=55\text{nm}$, which is substantially larger than the radius of an isolated ferritin particle, i.e. ~ 6 nm (summarized: diffusion substantially affects R₂* when $\frac{D}{\mathrm{\Delta }\omega R^2}>1$). Since the latter is indicative of the spatial extent of the particle’s field gradients, this suggests that diffusion indeed is likely to reduce R₂*. By how much will depend on the practical conditions, including specifics of the diffusion process, and the uniformity of the particles’ distribution including their potential clustering. Nevertheless, it should be realized that, due to various reasons (see below), not only R₂* but also frequency shifts, on which QSM is based, may deviate from the actual field shifts.

BIOLOGICAL ROLE OF IRON

As mentioned above, brain iron exists in various forms and in varying concentrations that change with age and disease (for review, see e.g. (33,43,75)). The ability of iron to facilitate redox reactions in which an electron is added to a substrate molecule (reduction) or extracted from a substrate (oxidation) is a key aspect of iron’s immense biological significance. Important biological roles include the facilitation of neurotransmitter (e.g. dopamine) cycling, enzyme and mitochondrial function, ATP and DNA synthesis, and the generation of myelin (43,75). We will briefly discuss the various aspects of iron metabolism in the human brain.

The Proteins of Iron Metabolism

At one time there were four main proteins identified with iron metabolism in mammals; (i) hemoglobin, the major protein of the red blood cell, (ii) transferrin, a transport protein capable of binding two iron atoms and moving them through the blood and within tissues, (iii) transferrin receptor, a cell-membrane-bound-protein capable of binding transferrin and translocating it and its two iron atoms from extracellular regions to the intracellular space and (iv) ferritin, an iron-storage protein capable of storing up to 4500 iron atoms in a mineralized form. It was more or less tacitly assumed that random diffusion of iron atoms and these various proteins was a major factor determining the body’s iron economy.

In the last twenty-five years or so a very large array of additional proteins involved in one aspect or another of iron metabolism and subject to intricate control mechanisms has been identified (76,77). It is likely that additional insights will result from ongoing research and these insights may lead to major revisions of current understanding. Random diffusion as a driving force has, at least in part, given way to a concept of tightly controlled (or ‘chaperoned’) forms of movement and chemical processes involving iron atoms.

A partial list of recently identified proteins involved with iron metabolism includes the iron regulatory protein elements IRP1 and IRP2 (78), cellular iron importers (e.g., the divalent metal transporter DMT1), exporters (ferroportin), the regulatory hormone hepcidin (79), various oxidases (e.g., ceruloplasmin and hephaestin) and reductases (DCYTB, STEAP3 see list of abbreviations for full names), etc. In addition the mitochondria have distinct iron-related molecules including iron importers (mitoferrin) and transporters (frataxin). As discussed below very recently two important additions have been made to this array of proteins involved in iron metabolism. These are the PCBP (poly(rC)-binding protein) family of iron chaperones and NCAO4, the cargo receptor mediating ferritinophagy.

The Labile Iron Pool (LIP) and the Storage Iron Pool (SIP)

In water solution both Fe⁺⁺ and Fe⁺⁺⁺ ions are liganded (solvated) by a shell of six water molecules. Fe⁺⁺⁺ is highly reactive with these surrounding water molecules and almost immediately hydrolyses some waters in its coordination shells to form insoluble iron oxides and hydroxides (i.e., rust) which then precipitate. To avoid the generation of these iron precipitates the concentration of Fe⁺⁺⁺ aqua-ions in the cytosol is kept at essentially zero by mechanisms described below. However, the cytosol must contain a certain level of readily reactive iron continuously available to carry out essential functions. For example, there are now hundreds of enzymes known that are produced as apoproteins in the cytosol and that require the insertion of iron-containing cofactors to function properly. There are three classes of these cofactors; heme, iron-sulfur clusters (both produced in the mitochondria) and iron aqua-ions (Fe⁺⁺ · 6 H₂O and Fe⁺⁺⁺ · 6 H₂O) (80,81). Early on, based on biochemical and thermodynamic arguments (82), it was suggested that there is a labile iron pool (LIP) within the cytosol to support crucial metabolic processes. The LIP could be based on Fe⁺⁺ aqua-ions and/or small molecular weight liganding molecules and should have a concentration in thevicinity of roughly 1 μM. Such a concentration would be sufficient to provide a constant supply of Fe⁺⁺ to force the many iron-requiring enzymes in the cytosol to remain in the active, metallated form rather than the inactive apoenzyme form (83). A candidate liganding compound needs to ensure proper enzyme function and to prevent autoxidation of Fe⁺⁺ to Fe⁺⁺⁺ as well as to prevent free radical generation via the Fenton reaction. In the cytosol it would need to serve these various functions at pH = 7 and in the presence of a dissolved O₂ concentration of 2 – 5 μM (84–86). A strong candidate recently proposed as the dominant molecule in the LIP is a conjugate of Fe⁺⁺ with the tripeptide antioxidant glutathione (87). An alternative based on a citrate molecule liganded to Fe⁺⁺ has also been recently proposed (88).

A Storage Iron Pool (SIP) also exists within the cytoplasm of most cells. Iron storage in brain can involve iron linked to neuromelanin, lipofuscin, metallothioneins and hemosiderin as well as to ferritin. However, as storage based on ferritin is the most thoroughly studied and appears to be the most prevalent iron storage mode in the brain, only this will be discussed here. When an excess of iron in the cytoplasm is detected by the IRP proteins, a process is initiated leading to the synthesis of H and L ferritin chains. Twenty-four of these chains self-assemble into a hollow spherical protein shell with pores connecting the interior with the cytosol. Excess Fe⁺⁺ in the LIP moves into the pores and encounters a ferroxidase site associated with the H subunits where it is oxidized to Fe⁺⁺⁺. Further, incompletely understood, steps lead to incorporation of the ferric ions into an insoluble mineralized iron oxide core capable of holding up to 4500 Fe⁺⁺⁺ ions (89). When a deficit of iron in the cytosol is sensed this process is somehow reversed, the mineral is dissolved, the Fe⁺⁺⁺ ions are reduced and the resulting Fe⁺⁺ ions are released to the cytosol.

The LIP and the SIP coexist within the cytosol and their relationship is important in assessing the clinical significance and interpretation of MR images of iron-rich brain tissues. The LIP has a very small iron concentration (roughly on the order of 0.3 – 1.6 μM) (80) and is ‘below the radar’ - it is far too small to register on conventional MRI or QSM. On the other hand, the SIP in brain regions such as the basal ganglia has a much higher iron concentration and is a prominent feature of high field MRI and QSM images. The molar SIP iron concentration in iron-rich brain regions can be estimated. On average, normal humans in the range of 30–70 years of age have roughly 200 μg iron per gram of tissue in the globus pallidus (35). Using Table 1 it is seen that this corresponds to an iron concentration of 3760 μM. In fact, this is probably a considerable underestimate in iron-rich cells as the tissue volumes studied by Hallgren and Sourander included not only cell cytoplasm but also other intracellular and extracellular regions. Also, histochemical iron staining indicates that the iron loading in these regions is not uniformly spread among the various cells but tends to be concentrated in specific cells. It is clear that iron-rich cells such as those in the globus pallidus have thousands of Fe⁺⁺⁺ ions in the SIP for each Fe⁺⁺ ion in the LIP and that MR imaging provides direct information on the SIP but not on the LIP.

A major goal of SW MRI research is to develop clinically useful information derived in part from measured brain iron concentrations. MRI has already demonstrated very clear imaging correlates of brain iron deposition with disease in a group of rare, but devastating, inherited disorders of iron metabolism (90). There is a substantial ongoing effort to accomplish this in more common neurological and psychiatric disorders such as Alzheimer’s and Parkinson’s diseases and depression (e.g., (91–95)). In most cases it is unclear from the MRI information alone whether the imaged brain iron effects are directly responsible for the pathology or are the result of pathological changes elsewhere that affect iron deposition in the SIP indirectly. In either case the MRI findings may be clinically relevant but this relevance would be enhanced if it was accompanied by a more or less complete understanding of the underlying iron-dependent processes but this is not currently available. Note that MRI information on brain iron reflects the status only of the SIP. However, the Fe⁺⁺⁺ ions in the SIP are ‘hors de combat’: locked in a mineralized, insoluble, and unreactive state and unable to take part in either constructive or destructive biological processes unless released back into cytosol and the LIP.

On the other hand, the Fe⁺⁺ ions in the LIP are continually involved with important processes such as Fenton reactions generating free radicals and the activities of possibly hundreds of iron-dependent enzymes. Thus, these ions are far too small in numbers to register on MRI but they are likely candidates for impacting both physiological and pathological brain processes. Therefore, it is to be hoped that an complete understanding of the relation between these two iron pools in the brain cytoplasm will eventually be developed as an aid in extending the clinical relevance of QSM. Two recent advances in relating the LIP and SIP involve the identification of iron metallochaperones and a process known as ferritinophagy.

Metallochaperones

When a simple metal salt (e.g., CuSO₄ or FeCl₂) is dissolved in water the metal ion becomes hydrated and diffuses freely through the solution. In cells there is a significant concentration of several different metals such as copper, iron and zinc. Many enzymes are utilized that require a particular divalent metal ion, e.g., Fe⁺⁺, at the active site. In many cases it is found that Cu⁺⁺ or Zn⁺⁺ would bind more avidly than Fe⁺⁺ to many iron-dependent enzymes and inactivate them if Cu⁺⁺ and Zn⁺⁺ metal ions as well as Fe⁺⁺ were freely available in the cytoplasm. This and related problems can be resolved using the concept of metallochaperones. By studying the insertion of copper into an antioxidant enzyme (Cu-Zn superoxide dismutase) in yeast, it was determined that the copper ion does not diffuse freely through the cytoplasm but isinstead tightly bound to a protein that prevents it from reacting with random apoproteins and directs its insertion only into the correct substrates. This binding is sufficiently tight and the metallochaperone protein is present in sufficient concentration to assure that there is less than one ‘free’ copper aquaion per cell (96,97). Similar chaperone protein have also been identified for Zn⁺⁺ (98). Thus there appears to be no pool of free copper or zinc ions in the cytoplasm.

The situation with iron is somewhat different. As was seen above, the LIP contains Fe⁺⁺ at approximately the 1 μM level. The method by which cytosolic iron is directed to ferritin was unknown until 2008 when a human protein (PCBP1), expressed in yeast, was found to bind to both iron and ferritin and to enhance iron loading into ferritin. It therefore functions as a metallochaperone for iron (99). PCBP1 is a member of a protein family also containing PCPB2, PCPB3 and PCPB4. Each of these proteins has been demonstrated to have iron chaperone properties. Not only do these metallochaperones direct iron to ferritin but also to iron-requiring enzymes as well (80,100–106).

Ferritinophagy

Ferritin has been assumed to serve two purposes in tissues, (i) to detoxify tissues in the presence of excess iron and (ii) to store iron for future use when required for cellular needs. The discovery of iron metallochaperones was an important step toward explaining how the first function is performed but there has long been some uncertainty as to exactly how the oxidized Fe⁺⁺⁺ ions in ferritin are retrieved and reduced to Fe⁺⁺ when it is required to support cell functions (89). Autophagy is a process in which cellular components (proteins and organelles – called cargo) are enveloped within double membrane structures called autophagosomes and delivered to lysosomes for degradation and recycling of their components. It can be viewed as analogous to phagocytosis except that the object being digested has its origin internal, rather than external, to the cell, It is believed that this process may play a significant role in several important disorders including Alzheimer’s, Parkinson’s and Huntington’s diseases (107–112).

There is now evidence indicating that the autophagy pathway has a major role in iron release from ferritin, a process called ferritinophagy. Recent studies (113,114) have identified the protein NCOA4 (nuclear receptor coactivator 4) as a cargo receptor for ferritin. This means that, in the presence of low cellular iron availability, NCOA4 binds to the ferritin H subunit, marking the ferritin molecule for transfer to a lysosome and leading to the eventual degradation of the ferritin protein shell. It has been demonstrated that NCOA4 is required for delivery of ferritin to lysosomes and that cells deficient in NCOA4 have decreased cellular iron availability. This represents a major advance in the understanding of the overall function of ferritin in cells. However, it still remains to clarify how, once the protein shell is digested, the mineralized Fe⁺⁺⁺ ions are removed from the oxide core and reduced to Fe⁺⁺ to reenter the LIP.

Implication

SW MRI has great promise in providing new approaches to imaging brain processes, particularly those involving iron. However, it is restricted to providing information on iron in the metabolically silent Storage Iron Pool. Recent developments in the cell biology of iron have indicated how an understanding of the Labile Iron Pool and its interactions with the Storage Iron Pool may aid the interpretation of SW MRI.

EFFECT OF MYELIN ON MRI CONTRAST

Myelin ensheaths most of the longer axons in the brain and serves to accelerate nerve conduction (for review, see e.g. (115)). It is formed by oligodendrocytes and consists of multiple (5–20 or more) phospholipid bilayers, which alternatingly flank intra and extra-cellular space of oligodendrocytes. Other than lipids, the bilayer contains mostly proteins and little (<20%) water whereas the opposite is true for the space between bilayers. On the millisecond-timescale of T₂ relaxation, water can be thought of as being contained in 3 distinct compartments: the axonal space, the interstitial space, and between the myelin wraps (Figure 4). The water in the latter space has long been a measurement target of methods aiming at estimating brain myelin content (116), as most hydrogen protons in the phospholipids and proteins of myelin are not directly visible with MRI. There is a substantial molecular order in the phospholipid bilayers (117), which has relevance to myelin’s magnetic properties, as will be discussed below. Myelin is several-fold more abundant in WM compared to GM, because of WM’s high density of myelinated fibers.

Figure 4.

Schematic cross section of axon showing location of myelin lipid bilayer and water. Within the myelin sheath, water is constrained to an approximately 4 nm wide space between successive bilayer wraps. Distinction between intra- and extracellular space of oligodendrocyte membrane is ignored in this schematic.

Myelin is nearly absent at birth and increases rapidly during the first 6–10 months of life (118,119). More incremental changes continue during childhood and in some regions into adolescence (119,120). In healthy adult brain, WM myelin content is about 25–35%, and it consists of about 40% water, 50% lipid and 10% protein (121,122). Abnormal myelination has been implicated with various psychiatric diseases (123), and there are a number of diseases that lead to demyelination, including amyotropic lateral sclerosis (ALS) and multiple sclerosis (MS).

Field effects and QSM of SW MRI suggest that WM is diamagnetic relative to water-like CSF (124–127), and this susceptibility shift has been attributed primarily to phospholipids in myelin. QSM-based estimates place Δχ_myelin (contribution of myelin to the susceptibility of WM) in the range of −13 to −34 ppb (66,126,128,129), not including orientation dependence of χ (see below).

As is the case with iron, the susceptibility variations introduced by myelin cause spatial variations in the magnetic field, which increase R₂*. Unfortunately, in WM, iron and myelin often colocalize (20,130,131) and can contribute with similar magnitude to R₂* (20), making it difficult to infer local myelin content from R₂* alone (Fig. 5a–c). Due to their opposing effects on tissue susceptibility however, combined analysis of R₂* and susceptibility may allow identifying the individual contributions of iron and myelin and quantify their concentration. (20,66,131,132).

Nevertheless, it has proven difficult to explain frequency and R₂* distributions in the brain based on the voxel-averaged concentrations of iron and myelin alone: across WM, substantial contrast variations have been observed that appear to reflect the orientation of major fiber bundles (133–137), although a few early studies found no clear orientation dependence (138,139). In fact, this dependence has been shown to allow prediction of fiber orientation (136). It has been attributed to the cellular structure of WM, which is highly anisotropic and makes the distribution of frequencies and their average within a voxel dependent on the structure’s orientation in the static field (14). Proper interpretation of susceptibility contrast in WM requires taking the anisotropy and orientation of this microstructure into account (see next section) (14,69).

Theoretical estimates show that the structure of myelin, when modeled in very crude approximation as a set of parallel cylindrical sheaths with axis at angle θ with the field, would increase R₂* by: $\mathrm{\Delta }{\mathrm{R}}_2^{\ast }=0.5{\mathit{sin}}^2\theta \mathrm{\gamma }\mathrm{\Delta }{\chi }_{\mathit{myelin}}{\mathrm{s}}^{-1}{\mathrm{T}}^{-1}$ (69). Assuming a range of Δχ_myelin of [−13, −34] ppb (66,126,128,129), this leads to $\mathrm{\Delta }{\mathrm{R}}_2^{\ast }=[1.7-4.5]{\mathit{sin}}^2\theta {\mathrm{s}}^{-1}{\mathrm{T}}^{-1}$. This does not take into account effects of diffusion and orientation dependence of the susceptibility itself (see below). The experimentally determined estimates of this sin²θ coefficient are in the range of [1–2] s⁻¹ T⁻¹ (134–136), somewhat lower than predicted by theory. This suggests that diffusion has a substantial effect on the R₂* of white matter. Using considerations similar as used for the iron particles above, with D~1 μm² ms⁻¹, R~1 μm, and Δω = [8–21] s⁻¹ (calculated by converting susceptibility of WM to that of pure myelin, assuming a 20% myelin volume fraction), we have $\frac{D}{\mathrm{\Delta }\omega R^2}$ in the range of [48–125] which is much greater than one, indicating a substantial effect of diffusion on R₂*.

A further complication is that myelin’s susceptibility itself has been found to be orientation dependent as well (140,141), a phenomenon that can be attributed to the orientational order of its constituent lipid molecules. Phospholipid molecules have an anisotropic susceptibility (i.e. it depends on their orientation relative to the magnetic field) (142); due to their organized arrangement in myelin’s bilayers (117,143), the organized arrangement of the bilayers around the axon, and the organization of axons into fiber bundles, this anisotropy has effects on the magnetic field that are observable on the macroscopic scale (i.e. on the scale of MRI voxels) (140,141). To accurately describe the effect of anisotropic χ on the magnetization, a tensor representation of χ can be used, with χ_|| and χ_⊥ as diagonal elements representing χ parallel and perpendicular to the fiber bundle respectively. MRI measurements put WM anisotropy (expressed here as χ_|| − χ_⊥) in the range of 10 to 25 ppb (141,144–148), although some of these measurements may be inaccurate due to confounding compartmental shape effects (see below). Independent estimates based on magnetic torque balance measurements (~15ppb) however fall within this range (149).

While the orientation dependencies of susceptibility and R₂* in WM complicates quantitative interpretation in terms of myelin and iron content, they may offer opportunities to map fiber orientation (136,150). As will be seen in the following, they may also offer opportunities to obtain cellular compartment specific information.

EFFECTS FROM TISSUE MICROSTRUCTURE

As hinted at above, not only the amount of substances that alter or “perturb” tissue susceptibility (e.g iron, deoxyhemoglobin and myelin) but also their microscopic distributions can affect the MRI signal. Even when susceptibility itself is isotropic, a non-uniformity at any sub-voxel spatial scale can increase R₂* and shift the frequency. The former effect can be understood by considering that perturber distribution affects field distribution, to which R₂* is sensitive (see e.g. (69)). The effect of “perturber” distribution on frequency can be understood by considering that the MRI frequency is not a faithful representation of the voxel-averaged field, in part because water proton concentration and visibility is not uniform over the voxel (e.g. where there is myelin or ferritin, there is (almost) no water); as a result, the MRI signal represents biased sampling of field. These effects need to be accounted for when calculating R₂* and susceptibility from MRI data, and when interpreting their values in terms of tissue composition.

For example, in WM, a biased sampling of the field occurs because of the compartmentalization of water around myelin, i.e. the measured frequency does not correspond to the average field in the voxel because there is little or no contribution from within the myelin sheath. An established approach (analogous to the Lorentz sphere approach discussed above) is to estimate this field is by considering average magnetic environment of a sensing proton on a water molecule (10,15); this approach has been successfully used to predict frequency shifts in biological materials such as muscle fibers, trabecular bone, and red blood cells (16,151,152)). It is based on the equivalence between averaging fields versus averaging sources that has its basis in the linearity of the Maxwell equations (10), and the incredible amount of averaging that occurs across the large number (at least 10¹⁴) polarized water protons in a typical MRI voxel. For the average frequency to be reflective of the average field, substantial diffusion averaging also needs to take place, which appears a realistic assumption in brain tissue (see above, $\frac{D}{\mathrm{\Delta }\omega R^2}\gg 1$).

Fig. 6 shows application of this concept to the simple case of uniformly distributed spherical or oriented ellipsoidal particles. Since our goal here is to estimate the effect of the particles, we ignore here (sub) atomic scale effects that are common between ¹H protons in pure water and ¹H protons in a mix of water and particles (e.g. chemical shift, dipolar coupling, and any other effects of the OH group in the water molecule or neighboring water molecules). In this simplification, the average magnetic environment seen by the proton resembles the shape of the particle: spherical particles will lead to a spherical average magnetic environment, whereas ellipsoidal particles lead to an ellipsoidal magnetic environment. This greatly simplifies estimation of the average local field at the ¹H nucleus, as it can be approximated by the use of one or more shape factors (see above) (10).

Figure 6.

Simulated (2D) average magnetic environment M_av of a water proton in the presence of randomly distributed spherical (top row) or ellipsoidal (bottom row) magnetic particles. Single particles (left column) and M_av (right column) are shown at 8-fold expanded scale. Brightness represents magnetization level; M_av images include horizontal (thick line) and vertical (thin line) profiles through sensing nucleus at center. With the spherical particles, and with the uniformly oriented ellipsoidal particles, the average magnetic environment resembles the particle shape.. M_av(x,y) was calculated by adding the environments for all water pixel locations (i,j), and dividing the result by the number of water pixels N: M_av(x,y) = Σ_i,j M(x + i, y + j)/N.

In WM, the magnetic environment at the sub-voxel scale can be approximated to be a continuum with cylindrical symmetry (Fig. 7, top row), and its effect on the resonance frequency (which we assume here to correspond to the average field) can be approximated by a cylindrical shape factor. As a result, in the axonal and interstitial space of WM, the predicted frequency shift, like R *, has an angular dependence according to $\mathrm{\Delta }f=0.5{\mathit{sin}}^2\theta \frac{\gamma }{2\pi }\mathrm{\Delta }{\chi }_{\mathit{myelin}}{\mathrm{s}}^{-1}$, which for the example of a fiber with Δχ_myelin = 20 ppb and a 90° angle with a 7 T field equates to &Delta;f ≈ 3 Hz. This seems inconsistent with (69) where, as it appears incorrectly, zero frequency shift was predicted, but analogous to (14) that uses the so-called “generalized Lorentzian” model.

Figure 7.

Simulated (2D) average magnetic environment of a water proton inside and outside myelinated axons running perpendicular to the magnetic field (and the image plane). For isotropic susceptibility, both environments have cylindrical symmetry and lead to similar local fields (and thus frequencies). For anisotropic susceptibility, an asymmetry is observed that is reversed between the two water environments. This will lead to frequency differences between environments. Thick and thin plot lines show horizontal and vertical profiles respectively. Spatial scale of environment is 7-fold expanded compared to that of magnetization in left column. The average magnetic environment was calculated as described in caption of Fig. 6.

The situation described above changes when considering anisotropy of the susceptibility. Computational modeling and analytical approaches suggest different shifts in different compartments in WM, specifically the spaces of axonal, extracellular, and myelin water (128,144,153), something not observed when considering only isotropic susceptibility. All these frequency shifts maintain a sin²θ dependence (128). Again, as with the isotropic case, frequency shifts can be estimated by considering the average magnetic environment of a water proton in each compartment, and use of a corresponding shape factor (Fig. 7, bottom row). The latter must be a tensor in order to account for the orientation dependent nature of the magnetic environment.

The notion of compartment specific relaxation and frequency shifts has recently been confirmed experimentally (128,144). In WM, R₂* decay is triple-exponential and dependent on orientation (128,144,154), consistent with the existence of three slowly interchanging (on the R₂* time scale) water compartments (155) within and distinctly different magnetic environments. Thus, measuring the decay curve and fitting it to a sum of 3 exponentials with different frequencies would allow extraction of cellular compartment-specific information (144). For example, one could extract an estimate of the size of the myelin water compartment, which could then be interpreted as being indicative of local myelin content. This refines an approach used previously based on R₂* differences alone (156), and forms a potentially more specific alternative to myelin measurement methods based on magnetization transfer and proton density (157). As such, it may facilitate detection of myelin loss in demyelinating diseases such as MS (158). Another possible application is the use of the orientation dependence of the frequency shifts to map fiber orientation (159). Of course, compartment-specific frequency shifts are problematic for quantitative methods such as QSM and susceptibility tensor imaging and need to be accounted for (145).

SUSCEPTIBILITY CHANGES WITH DISEASE

While the potential of magnetic susceptibility contrast for the study of brain pathology had been recognized since the early days of MRI, much of the early work, performed at low field (at or below 1.5 T) focused primarily on visualizing gross changes in susceptibility due to calcifications with substantial mass, or vascular abnormalities, often associated with tumors and hemorrhages (160,161). Study of tissue iron accumulation initially was limited to the regions with highest concentration and preferentially done with T₂ contrast (38), while its specificity to iron outside these regions was rather limited (68). Transition to higher field strength and the increasing use of high-sensitivity coil arrays expanded the use of susceptibility contrast to include more subtle changes in iron and deoxyhemoglobin, and most recently, studies have started to apply SW MRI changes in brain myelin content and structure. Although many of these applications have been summarized previously (e.g. (6,18,75,162,163)), some of the major findings are highlighted below.

In the case of malignant brain tumors, SW MRI has been used to aid diagnosis and treatment based on the detection of local increases in deoxyhemoglobin and hemosiderin due to angiogenesis and vascular leakage (164–166). Additionally, SW MRI may aid in monitoring the effects of radiation therapy by detecting hemosiderin associated with micro bleeds that may result from the induced vascular damage (167). Similarly, SW may also be able to detect the microbleeds associated with traumatic brain injury, supplementing other MRI techniques in its diagnosis (168).

Both normal aging and numerous pathologies can lead to brain iron accumulation, the pattern of which may be specific to the type of pathology. Characterization of this pattern and the severity of the accumulation with SW MRI may thus help in diagnosis and possibly with monitoring of treatment. A number of neurodegenerative disorders lead to iron increases in the sub-cortical GM of the basal ganglia (34,44,169), and occasionally is constrained to specific (sub) regions. For example in CADASIL, selective accumulation has been observed in the putamen (170), whereas in Parkinson’s Disease, the substantia nigra may be selectively affected (31,171). Widespread iron accumulation in cortical GM has been reported in Alzheimer’s Disease (172), whereas in ALS more focal cortical accumulation has been reported (173,174). Focal iron increases also occur within or adjacent to MS lesions (175–182), and such accumulation may facilitate the detection of cortical lesions that often escape detection with other MRI contrasts (177–179,181). An interesting possibility is the detection of MS pathology based on structural changes in myelin: to the extent that these precede myelin loss, detection of such structural changes may offer an opportunity for early diagnosis in improved assessment of treatment response (144,180,183).

UNRESOLVED ISSUES

Over the last few years, tremendous progress has been made in the understanding of susceptibility contrast, in particular in WM. Additionally, much progress has been made in extracting quantitative information with potential clinical relevance, including susceptibility, R₂*, and the distribution of water between various cellular compartments. These parameters may allow quantification of biological and biophysical parameters such as tissue myelin and iron content, and the oxygenation state of tissue. Nevertheless, there a number of questions and issues that remain unresolved or not fully clarified, some of which will be discussed below.

A question that has been around for some time is the biological relevance of tissue iron and its particular distribution of the brain. Why is iron high in the basal ganglia, and what is the causal relationship with pathology? Why does iron co-localize with myelin in some fibers, but is absent in others? If iron is required for myelin generation and turnover, why does it vary so much over WM? Does some of the myelin turn over relatively rapidly? While at this point we can only speculate, it is possible that detailed mapping of iron distribution throughout the brain and over a range of pathologies may lead to some clarification in the near future. Of course, improved understanding of the relationship between LIP and SIP (see above) will be critical in this.

Another unresolved issue is the relationship between R₂* and susceptibility changes related to iron. Observed R₂* changes are generally below those predicted by theory based on a random, uniform distribution of ferritin particles. It is unclear of this reduction is entirely explained by diffusion averaging, and this is difficult to determine conclusively due to uncertainty about the uniformity of particles’ distribution: clustering of particles or other types of ordering (e.g. alignment with fibers, or preference for specific cellular compartments) may substantially alter R₂*. This has relevance for quantification of tissue iron and myelin content from joint analysis of R₂* and susceptibility data. Nevertheless, a surprisingly linear relationship has been reported across GM of the basal ganglia (see above).

Puzzling also is the frequency shift attributed to the myelin water compartment for fibers perpendicular to the magnetic field in the analysis of multi-exponential relaxation in WM. Theoretical modeling suggests a strong paramagnetic frequency shift +100 ppb, whereas experimentally, a weak diamagnetic shift of about −10 ppb is observed (184). It appears therefore as if the myelin water experiences the magnetic field attributed to myelin’s lipid bilayer. Full understanding is required for a confident quantification the myelin water fraction from SW data.

Lastly, it needs to be emphasized that throughout this paper, as well as the published literature, it is generally assumed that MRI signal frequency directly reports on the average local field. As indicated above, this is not necessarily true as the voxel-averaged frequency depends on the MRI-visibility of protons that contribute to the signal. T₁, T₂, and T₂* relaxation, MT, and diffusion all may affect a proton’s visibility and thus its contribution to the amplitude and frequency of the average signal. This potential difference between voxel-averaged frequency and voxel-averaged field may affect the interpretation of SW MRI, as is apparent from the triple-exponential R₂* decay observed in WM. It also means that the extracted parameters may depend somewhat on the details of the MRI pulse sequence. With further research, it will become clear to what extent this issue affects quantification.

ABBREVIATIONS

ALS

Amyotrophic Lateral Sclerosis

BOLD

Blood Oxygen Level Dependent

CADASIL

Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy

CBV

Cerebral Blood Volume

DAB

Diaminobenzidine

DCYTB

Duodenal Cytochrome B

DMT

Divalent Metal Transporter

GM

Grey Matter

IRP

Iron Regulatory Protein

LIP

Labile Iron Pool

MS

Multiple Sclerosis

NCOA

Nuclear Receptor Coactivator

PCBP

Poly(rC) Binding Protein

QSM

Quantitative Susceptibility Mapping

SIP

Storage Iron Pool

STEAP

Six-Transmembrane Epithelial Antigen of the Prostate family

SW

Susceptibility Weighted

WM

White Matter