The influence of iron oxidation state on quantitative MRI parameters in post mortem human brain

A variety of Magnetic Resonance Imaging (MRI) techniques are known to be sensitive to brain iron content. In principle, iron sensitive MRI techniques are based on local magnetic field variations caused by iron particles in tissue. The purpose of this study was to investigate the sensitivity of MR relaxation and magnetization transfer parameters to changes in iron oxidation state compared to changes in iron concentration. Therefore, quantitative MRI parameters including R1, R2, R2∗, quantitative susceptibility maps (QSM) and magnetization transfer ratio (MTR) of post mortem human brain tissue were acquired prior and after chemical iron reduction to change the iron oxidation state and chemical iron extraction to decrease the total iron concentration. All assessed parameters were shown to be sensitive to changes in iron concentration whereas only R2, R2∗ and QSM were also sensitive to changes in iron oxidation state. Mass spectrometry confirmed that iron accumulated in the extraction solution but not in the reduction solution. R2∗ and QSM are often used as markers for iron content. Changes in these parameters do not necessarily reflect variations in iron content but may also be a result of changes in the iron's oxygenation state from ferric towards more ferrous iron or vice versa.


Introduction
Iron plays an important role in the human brain for maintaining normal metabolism and is involved in several biophysiological processes, such as oxygen transport (Pauling, 1948), the production of myelin (Todorich et al., 2009), and mitochondrial respiration (Pinero and Connor, 2000). In the healthy brain, iron accumulates with age depending on the brain region. For example, in the globus pallidus iron accumulates until the 4 th decade of live, where as in the putamen it gradually accumulates until the 8 th to 9 th decade of live (Hallgren and Sourander, 1958). The highest iron levels, up to 250 mg/kg tissue, are found in deep gray matter structures (Krebs et al., 2014). A disturbed iron metabolism has been linked to a variety of neurological disorders (Pinero and Connor, 2000;Ward et al., 2014), such as Alzheimer's disease (AD) (Quintana et al., 2006;Schenck et al., 2006), multiple sclerosis (MS) (Hametner et al., 2013;Stankiewicz et al., 2014) or Parkinson's disease (PD) (Martin et al., 2008;Sofic et al., 1988).
Magnetic resonance imaging (MRI) is the most important imaging technique to assess and map brain iron in vivo (Ropele and Langkammer, 2017). Several MRI techniques to quantify iron have been proposed, such as relaxation time mapping (Haacke et al., 2005;Langkammer et al., 2010;Ogg and Steen, 1998), phase imaging (Hammond et al., 2008;Hopp et al., 2010), quantitative susceptibility mapping (QSM) (Kames et al., 2018;Langkammer et al., 2012b;Sun et al., 2015;Wang and Liu, 2015;Wisnieff et al., 2015), magnetic field correlation (Jensen et al., 2009(Jensen et al., , 2006 and direct saturation imaging (Smith et al., 2009). In principle, iron sensitive MRI techniques sense local magnetic field variations caused by iron particles. These local magnetic field variations depend on the magnetic properties of the iron compounds and affect image contrast, via quantitative relaxation parameters as well as the phase of the complex MR signal (Brooks et al., 1998;Drayer et al., 1986).
The four main iron compounds in the human brain are: (I) ferrous iron (Fe 2þ ) bound to haemoglobin in the blood, (II) ferric iron (Fe 3þ ) stored as ferritin, (III) ferric (Fe 3þ ) and ferrous (Fe 2þ ) iron bound in magnetite (Fe 3 O 4 ) and maghemite (γ-Fe 2 O 3 ) and (IV) ferric iron (Fe 3þ ) bound to transferrin (Brem et al., 2006a;Theil, 2013). Most of the non-heme ferric iron is stored as ferritin in the human brain. Ferritin is a storage protein consisting of a protein shell and a core containing up to 4500 mainly ferric iron ions as hydrated iron oxide nanocrystals (5 Fe 2 O 3 9H 2 O) called ferrihydrite (Dobson, 2001;Gossuin et al., 2009;Harrison and Arosio, 1996).
Iron's ability to change the oxidation state from ferrous iron (Fe 2þ ) to ferric iron (Fe 3þ ), and vice versa, through electron exchange is critically important for maintaining the balance between iron intake and storage. A disturbed iron metabolism can lead to abnormal iron levels and an increased release of reactive oxygen species which results in oxidative stress leading to cell death and neurodegeneration (Ward et al., 2014). There is evidence that pathological processes are associated with altered iron content and oxidation state. For example, an increase in iron content and a change in iron oxidation state from ferrous to ferric iron was observed in PD (Sofic et al., 1988). In contrast, other studies observed no difference in total iron content and no alteration of the iron oxidation state between PD patients and controls (Galazka-Friedman et al., 1996;Wypijewska et al., 2010) but an increase in labile iron concentration resulting in higher reactive oxygen species in the substantia nigra of PD subjects (Wypijewska et al., 2010). Quintana et al. reported differences in the composition of the iron compounds in AD brains compared to healthy brains (Quintana et al., 2006). In the healthy brain, ferric iron is the major iron compound, whereas in AD mixed ferric-ferrous iron is the major iron compound (Quintana et al., 2006(Quintana et al., , 2004. In AD, ferrous iron was found to be increased in senile plaques and associated with neurofibrillary tangles (Sayre et al., 2000;Smith et al., 2004).
While the effects of changes in iron content on the MRI signal have been studied extensively, research on the role of oxidation state has been limited. Recently, Dietrich et al. proposed a technique to differentiate between ferrous and ferric iron based on their different relaxation and magnetic properties (Dietrich et al., 2017). In particular, lower R 2 * but similar susceptibility values were observed in ferrous iron chloride solutions compared to ferric iron chloride solutions with the same iron concentration.
The purpose of this study was to investigate the sensitivity of MR relaxation rates, magnetic susceptibility and magnetization transfer ratio (MTR) to (I) changes in iron oxidation state and to (II) changes in total iron content in post mortem human brain tissue. Therefore, quantitative MRI, was performed prior and after (I) the change in oxidation state from ferric towards more ferrous iron and (II) the change in total iron content by chemical iron extraction.

Brain tissue
Coronal brain slices from four formalin fixed human brains (female/ male ¼ 2/2) without history of neurological disorders or neurological cause of death and a mean age at death of 71 years (range ¼ 60-81 years) were included in this study. For brain tissue fixation 4% neutral buffered formalin (SAV LP GmbH, Flintsbach, Germany) was used with a mean fixation duration of 122 days (range ¼ 91-161 days). The post mortem interval between death and autopsy was between 9 and 24 h (mean ¼ 17 h). The right hemisphere of each brain was cut into approximately 1 cm thick coronal slices. From each brain, two slices, containing white matter and cortical gray matter were used for this study. After initial MRI acquisition, one slice of each brain was assigned to iron reduction and the other to iron extraction. Each slice was further cut in half where as one half underwent either the iron reduction or the extraction procedure and the other half was kept in formalin as reference.

Iron reduction and iron extraction
To change the oxidation state of iron in the brain, sodiumdithionite was used as a reducing agent. Therefore, 1 mM sodiumdithionite, as final concentration, was dissolved in phosphate-buffered saline (PBS) at pH 7.4 to reduce ferric to ferrous iron without changing the overall iron content (Watt et al., 2006). To extract the iron from the brain slices, extraction was performed according to an established procedure (Fukunaga et al., 2010;Oh et al., 2013;Schenck et al., 2006) by using a solution of 2 mM ferroxamine and 1 mM sodiumdithionite, as final concentration, in PBS at pH 7.4.

Inductively coupled plasma mass spectrometry of the solutions
Total iron concentrations in the extraction and reduction solutions were determined with inductively coupled plasma mass spectrometry (ICPMS, Agilent 7700x, Agilent Waldbronn Germany). The iron concentration was determined at m/z ¼ 56 in the He-mode to reduce polyatomic interferences. Germanium (Ge) at m/z 74 was used as internal standard. The NIST CRM 1640a (trace elements in water) was successfully used for quality control. The solutions in which the brain slices were stored were refreshed with new solutions every other day over 11 days. Additionally, we determined the iron concentrations in the PBS buffer, reduction solution and in the extraction solution prior to tissue treatment.

MRI
Image acquisition was performed on a 3T MRI system (MAGNETOM PRISMA, Siemens Healthineers, Erlangen, Germany) at room temperature (23 C) prior and after 11 days of iron reduction or iron extraction. All tissue slices were placed in a plastic container and immersed in PBS for at least 3 h to minimize embedding effects (Dusek et al., 2018). All sequences were acquired with an identical resolution of 0.86 Â 0.86 mm, a slice thickness of 2 mm and 20 slices. For R 1 mapping, a turbo inversion recovery (TIR) sequence with TR ¼ 8000 ms, TE ¼ 6.4 ms, six inversion times (TI ¼ 100, 200, 400, 800, 1600 and 3200 ms) and turbo factor ¼ 9 was acquired. For R 2 mapping, a turbo spin echo (TSE) sequence with TE ¼ 10, 73, 115 ms and TR ¼ 4000 ms was acquired. A 2D RF spoiled gradient echo (GRE) sequence with six echos (TE ¼ 4.92,10.34,15.76,21.18,26.60,32.02 ms), TR ¼ 300 ms and a flip angle of 15 was used for R 2 * mapping. To calculate MTR, a MT-weighted GRE sequence with identical parameters as used for R 2 * mapping but with an off-resonant RF saturation pulse (offset frequency ¼ 1.2 kHz, duration ¼ 10 ms and flip angle ¼ 500 ) was acquired.
Maps of R 1 , R 2 and R 2 * were calculated on a voxel by voxel basis, assuming mono-exponential relaxation. QSM was calculated according to a previous proposed method by Kames et al. (2018). MTR was calculated according to MTR ¼ (M 0 -M S )/M 0 , where M S is the image with and M 0 without additional RF saturation pulse. R 1 , R 2 , R 2 * and MTR were assessed in multiple regions of interest (ROIs) which were manually drawn in white matter and the cortex of each brain slice.
Statistical analysis was performed using R (R Foundation for Statistical Computing, Vienna, Austria). To test if the data is normal distributed, the Shapiro Wilk test was used. A Wilcoxon rank sum test was used to test if the difference between Formalin and iron extraction and Formalin and iron reduction, respectively was significant. P-values < 0.05 were assumed to be significant.

Results
Overall, iron extraction affected all relaxation parameters and MTR, whereas changing the iron oxidation state by reducing ferric to ferrous iron affected only R 2 and R 2 *, as shown in Fig. 1 and Supplementary  Figure 1.

Effects of iron reduction
A change in iron oxidation state from ferric to ferrous iron showed no influence on R 1 (Fig. 2) and MTR (Fig. 5), both in white matter and cortex. R 2 decreased by 18% from 11 s À1 to 9 s À1 (p ¼ 0.03) in white matter and by 13% from 7 s À1 to 6 s À1 (p ¼ 0.04) in the cortex, as shown in Fig. 3. R 2 * decreased by 11% from 24.4 s À1 to 21.7 s À1 (p ¼ 0.049) in white matter and by 11% from 16.9 s À1 to 15.0 s À1 (p ¼ 0.007) in the cortex (Fig. 4).

Effects of iron extraction
In general, iron extraction resulted in a decrease of R 1 by 40% from 3.5 s À1 to 2.1 s À1 (p < 0.001) in white matter and by 37% from 1.9 s À1 to 1.2 s À1 (p < 0.001) in the cortex. R 2 decreased by 25% from 12 s À1 to 9 s À1 (p < 0.001) in white matter and by 24% from 7.9 s À1 to 6 s À1 (p < 0.001) in the cortex. R 2 * decreased by 20% from 25 s À1 to 20 s À1 (p < 0.001) in white matter and by 26% from 18 s À1 to 13 s À1 (p < 0.001) in the cortex. After iron extraction, MTR increased by 34% from 0.26 to 0.34 (p < 0.001) in white matter and by 48% from 0.19 to 0.28 (p < 0.001) in the cortex.
All values of R 1 , R 2 , R 2 * and MTR prior and after chemical iron reduction and iron extraction for both white matter and cortex are summarized in Supplementary Table 1.

Effects of iron extraction and iron reduction on QSM
Qualitatively, QSM decreased due to both iron extraction and iron reduction as shown in Supplementary Figure 2. The quality of the susceptibility maps was not sufficient for a reliable quantitative ROI based analysis.

Mass spectrometry of the iron extraction and iron reduction solutions
Mass spectrometry of the iron extraction solution, after treating the tissue slices, showed an exponential decrease in iron content over time. As expected, the largest amount of iron was extracted from brain slices during the first days of the experiment. With decreasing tissue iron content due to extraction, less iron accumulated in the extraction solution. After 11 days of iron extraction, a mean of 165 AE 25 μg iron was extracted from each tissue slice. In the iron reduction solutions, no changes in iron content were observed, as shown in Fig. 6.

Discussion
In this study we investigated the effect of changes in iron content and iron oxidation state from ferric to ferrous iron on MR relaxation rates and Fig. 1. Representative R 1 , R 2, R 2 * and MTR maps after 11 days of iron extraction (top row) and iron reduction (bottom row). The upper part of each brain slice underwent the iron extraction or reduction procedure whereas the lower part was kept in formalin as reference. A decrease in iron content after iron extraction affected all relaxation times and MTR. A change in iron oxidation state from ferric (Fe 3þ ) towards more ferrous (Fe 2þ ) iron by chemical iron reduction affected only R 2 and R 2 *. Fig. 2. R 1 relaxation rates of white matter and cortex decreased after iron extraction and remained unchanged after iron reduction from ferric to ferrous iron. (* p < 0.05, ** p < 0.01, *** p < 0.001) MTR in post mortem human brain tissue. Overall, our results showed that all assessed parameters are sensitive to changes in total iron content, whereas only R 2, R 2 * and QSM were also sensitive to changes in iron oxidation state.
Conversion of ferrous to ferric iron, or vice versa, alters the magnetic properties and electron spin relaxation times of the iron ions (Gore et al., 1984). Gore et al. reported that a conversion of ferrous to ferric iron ions in water solutions will alter nuclear magnetic resonance (NMR) relaxation times as relaxation of water molecules near Fe 3þ ions is much more efficient than near Fe 2þ ions (Gore et al., 1984). The large difference in relaxation times of ferrous and ferric iron based on their different correlation times are larger than one would expect based on their magnetic moments or static magnetic susceptibility (Gore et al., 1984). This holds true for water solutions, as the dipole-dipole coupling is the dominant source of magnetic interaction (Abragam, 1961).
In human post mortem brain tissue, iron reduction did not lead to a significant difference in R 1 relaxation rate, which is in contrast to experiments with ferrous and ferric agarose gel phantoms where changes in T 1 relaxation properties were observed (Tokuhiro et al., 1996). The reason for the different observations in human brain tissue could be caused by the different composition of the iron cores in ferritin bound iron compared to pure ferric or ferrous iron ions as used in most phantom experiments (Brem et al., 2006b). In the human brain, iron is mainly stored as ferritin bound iron (Quintana et al., 2006). In the ferritin core, iron is present in more than one magnetic phase, like ferrihydrite, magnetite or hematite (Brooks et al., 1998;Galvez et al., 2008). In ferrihydrite, more than 90% of the iron is stored as Fe 3þ , whereas magnetite consists of an alternating sublattice of Fe 2þ and Fe 3þ which are antiferromagnetically coupled (Brem et al., 2005;Quintana et al., 2006). To fully understand the magnetic properties of ferritin and its influence on susceptibility sensitive MRI, more specific experiments are needed.
Our results showed changes in R 2 * after both chemical iron extraction and iron reduction. The different sensitivity of R 2 * to ferrous and ferric iron is in line with a previous work of Dietrich et al., where R 2 * showed higher values for ferric than ferrous iron chloride solutions which was further evident as R 2 * showed a stronger increase with increasing iron concentration in ferric than ferrous iron chloride (Dietrich et al., 2017). Fig. 3. R 2 relaxation rates decreased significantly after iron extraction and iron reduction, both in white matter and cortex. (* p < 0.05, ** p < 0.01, *** p < 0.001) Fig. 4. Both a decrease in iron content after iron extraction and a shift in iron oxidation state from ferric to ferrous iron after iron reduction resulted in a significant decrease of R 2 * relaxation rates in both white matter and cortex. (* p < 0.05, ** p < 0.01, *** p < 0.001) In general, R 2 and R 2 * are mainly influenced by the relatively abundant stored iron, which is to a large part ferric iron captured inside ferritin proteins. The chemically more active ferrous iron has a much lower concentration in the brain than the stored iron inside ferritin. Overall, R 2 and R 2 * are sensitive to both ferrous and ferric iron, but due to the dominating contribution of ferric iron, R 2 and R 2 * reflect mainly the stored iron rather than the active iron in the brain.
Theoretically, the combination of R 2 * and R 1 can distinguish changes in oxidation state from changes in iron content. However, in a situation where both, the iron content and oxygenation state changes, no differentiation can be made. Furthermore, such distinction would be only possible if the underlying myelin content does not change, as R 1 is sensitive to both iron and myelin in white matter (Stüber et al., 2014).
In light of our observations and previous literature, changes in R 2 * may not solely represent changes in total iron content. A change in iron oxidation state can increase or decrease R 2 * in addition to any changes related to variations in iron content. Thus, it is possible that changes in R 2 * observed in various neurological disorders such as MS, AD and PD could partly be caused by changes in iron oxidation state in addition to changes in iron content. This would be in line with observations in post mortem brain tissue of PD where an increase in total iron content but also a shift in iron oxidation state from ferric to more ferrous iron was observed (Sofic et al., 1988). Both effects, an increase in iron and the shift in iron oxidation state, would result in an increase of R 2 *. In contrast, other post mortem studies showed no significant difference in iron content and iron oxidation state of the substantia nigra of PD and control subjects (Chwiej et al., 2007;Galazka-Friedman et al., 1996;Wypijewska et al., 2010). Thus, an increase in R 2 * as observed in PD (Langkammer et al., 2012b;Lewis et al., 2018) may not purely reflect an increased iron content. However, it is currently not possible to provide an estimate how strongly changes in iron oxidation state affect relaxation parameters in vivo.
The observed increase in MTR after iron extraction is in agreement with the reduction in R 1 due to iron extraction. This finding is also in line with a previous work by Langkammer et al., where MTR was negatively correlated with iron content assessed by mass spectrometry (Langkammer et al., 2012a). Due to reduced T 2 relaxation times in formalin fixed post mortem tissue, direct saturation increases and might additionally influence the MTR measurement. Thus, our findings for MTR can not be translated directly to the in vivo situation due to the reduced T 1 and T 2 relaxation times in formalin fixed brain tissue (Birkl et al., 2016).
Our study has some limitations. Due to the chemical procedure of extracting or reducing iron in the whole tissue samples, no regional quantification of the absolute amount of extracted respectively reduced iron can be made. Our experiments are designed to investigate the different sensitivity of relaxation rates and MTR to changes in iron content and iron oxidation state in general. Future and more specific experiments will be needed to directly correlate the absolute change in iron concentration and oxidation state with the observed changes in relaxation parameters and MTR, independently for different brain regions. Furthermore, as can be seen from the iron concentrations in the extraction solutions shown in Fig. 6, not all iron was removed from the tissue. The effects of complete iron removal were therefore not assessed with the present study. The reference brain slices were kept in formalin while the other brain slices were treated with the iron extraction or iron reduction solution for 11 days. Although the samples were immersed in PBS prior to MRI, the additional storage of the reference brain slices for 11 days might introduce a small bias as the formalin fixation duration has an influence on relaxation times. Furthermore, the gradual washing-out of formalin in the brain slices treated with the iron extraction or iron reduction solution could also influence the relaxation times compared to the reference brain slices. However, in a recent study investigating the effect of iron extraction on the estimation of myelin water fraction, no bias in the reference brain tissue was observed .
Several studies showed that bi-exponential T 1 relaxation can be observed in brain white matter (Gochberg and Gore, 2003;Koenig et al., 1990;van Gelderen et al., 2016) and that the short T 1 component is dominated by the contribution of myelin . Due to the use of formalin fixed brain tissue, the overall T 1 is reduced by up to 70% compared to in vivo (Birkl et al., 2016) which limits the possibility to analyse a potential bi-exponential T 1 decay.
In recent years, many studies used QSM as measure for iron in the brain. Overall, QSM was shown to be sensitive to iron in deep gray matter (Langkammer et al., 2012b;Sun et al., 2015) and influenced by iron and myelin in white matter (Langkammer et al., 2012a;Wiggermann et al., 2017). In our study, we computed QSM to investigate if QSM is sensitive to both changes in iron content and iron oxygenation state as one could expect from theory. We were able to qualitatively observe a decrease in susceptibility after iron extraction and iron reduction, which is in agreement with the findings for R 2 * (Supplementary Figure 2). However, the quality of the susceptibility maps was not sufficient for a quantitative analysis in line with the other assessed parameters. The reason why QSM quality is poorer than the R 2 * quality, even though they are derived from the same scan, is that R 2 * is a more local measure of tissue properties than phase/QSM. Background field inhomogeneities due to air bubbles or the irregular shape of the tissue samples have low spatial frequencies compared to the voxel size. Therefore, the loss of coherence across a voxel is small and R 2 * is only weakly affected by such field inhomogeneities (Reichenbach et al., 1997). Even if the field is rather inhomogenous, R 2 * still reflects local tissue properties, as long as the voxels are small enough that the field inside the voxel can be regarded as homogeneous (Reichenbach et al., 1997). In QSM, on the other hand, for each voxel the field from regions outside that particular voxel also plays a role. Therefore, field inhomogeneities with spatial frequencies larger than the voxel dimensions still affect the QSM. In future studies, the effects from irregular shape and air bubbles can be alleviated by using larger tissue samples.
The focus of our study was to investigate the effect of iron on relaxation properties and MTR in white matter. Thus, the tissue samples used in the present study did not include deep gray matter, which is very rich in iron, in order to avoid saturating the solution or even washing iron from the deep gray matter into the white matter. Future studies will need to also focus on brain tissue with deep gray matter structures, which will be more challenging as the amount of iron in deep gray matter structures is up to 250 mg/kg tissue (Hallgren and Sourander, 1958;Krebs et al., 2014). Achieving a proper iron extraction and reduction across a whole brain slice including basal ganglia structures is much more time consuming based on the fast saturation of the iron extraction solution with iron dissolved out of the iron rich deep gray matter structures. Orientational effects caused by the underlying white matter microstructure were ignored in our analysis as these orientation dependent effects are mainly caused by myelin and not by the iron particles (Birkl et al., 2015;Kor et al., 2019). It was shown previously that neither iron extraction nor the change in oxidation state has an influence on histologically assessed myelin content .
Our results do not undermine the value of R 2 , R 2 * and QSM as marker for iron as they showed a significant correlation with overall iron content (Langkammer et al., 2012b(Langkammer et al., , 2010Sun et al., 2015;Walsh et al., 2013). The present work, however, highlights the different sensitivity of relaxation properties, magnetic susceptibility and MTR to changes in iron content and oxidation state.

Conclusion
All assessed quantitative parameters, were affected by a change in total iron content, whereas a change in iron oxidation state from ferric towards ferrous iron only affected R 2 , R 2 * and QSM. Therefore, a decrease in R 2 or R 2 * can be caused either by a decrease of the total iron content, the reduction of ferric iron to ferrous iron or a combination of both. Thus, care must be taken if R 2 , R 2 * or QSM is used to assess iron in the human brain, especially in neurological disorders where a disturbed iron metabolism is present.