Biophysical contrast sources for magnetic susceptibility and $\text{R}{2}^{*}$ mapping: A combined 7 Tesla, mass spectrometry and electron paramagnetic resonance study

Abstract

Iron is the most abundant trace metal in the human brain and consistently shown elevated in prevalent neurological disorders. Because of its paramagnetism, brain iron can be assessed in vivo by quantitative MRI techniques such as ${{\text{R}}_2}^{*}$ mapping and Quantitative Susceptibility Mapping (QSM). While Inductively Coupled Plasma Mass Spectrometry (ICP-MS) has demonstrated good correlations of the total iron content to MRI parameters in gray matter, the relationship to ferritin levels as assessed by Electron Paramagnetic Resonance (EPR) has not been systematically analyzed. Therefore, we included 15 postmortem subjects (age: 26–91 years) which underwent quantitative in-situ MRI at 7 Tesla within a post-mortem interval of 24 h after death. ICP-MS and EPR were used to measure the total iron and ferritin content in 8 selected gray matter (GM) structures and the correlations to ${{\text{R}}_2}^{*}$ and QSM were calculated. We found that ${{\text{R}}_2}^{*}$ and QSM in the iron rich basal ganglia and the red nucleus were highly correlated with iron (R² > 0.7) and ferritin (R² > 0.6), whereas those correlations were lost in cortical regions and the hippocampus. The neuromelanin-rich substantia nigra showed a different behavior with a correlation with total iron only (R² > 0.5) but not with ferritin. Although qualitative results were similar for both qMRI techniques the observed correlation was always stronger for QSM than ${{\text{R}}_2}^{*}$. This study demonstrated the quantitative correlations between ${{\text{R}}_2}^{*}$, QSM, total iron and ferritin levels in an in-situ MRI setup and therefore aids to understand how molecular forms of iron are responsible for MRI contrast generation.

1. Introduction

The use of magnetic susceptibility (χ) contrast in MRI (Wang and Liu, 2015) provides unique information about tissue’s structure as it reflects the magnetic property of the underlying tissue composition. It has been applied to study the human brain during healthy aging (Liu et al., 2016) and during neurodegenerative diseases such as Parkinson’s Disease (Lotfipour et al., 2012; Barbosa et al., 2015; He et al., 2015; Langkammer et al., 2016; Acosta-Cabronero et al., 2017; Chen et al., 2019), Alzheimer’s Disease (Acosta-Cabronero et al., 2013; Moon et al., 2016; Kim et al., 2017; Du et al., 2018) and Multiple Sclerosis (Langkammer et al., 2013; Chen et al., 2014; Li et al., 2016; Stüber et al., 2016; Zhang et al., 2016; Castellaro et al., 2017). Specifically, quantitative MRI techniques, such as the ${{\text{R}}_2}^{*}$ mapping and Quantitative Susceptibility Mapping (QSM), have shown high sensitivity to differentiate healthy subjects from patients with neurological diseases.

Most of these studies have attributed the positive contrast of QSM solely to iron, however the interpretation of the observed contrast on QSM and ${{\text{R}}_2}^{*}$ maps in the brain are difficult since it is rather related to the bulk χ distribution, independent of possible individual contributions from biomolecules included in the voxel. Several validation studies using data from postmortem subjects showed a strong correlation of iron with both ${{\text{R}}_2}^{*}$ and QSM (Langkammer et al., 2010; 2012 Zheng et al., 2013; Hametner et al., 2018). While these studies show that iron in biological concentration is enough to generate χ -induced contrast, its spatial influence is not fully understood yet (Brammerloh et al., 2022). The Basal Ganglia (BG) structures are known to have higher iron concentration compared to other structures (Hallgren and Sourander, 1958), which would explain their superior ${{\text{R}}_2}^{*}$ and QSM sensitivity to iron and the strong visual similarity of these structures with histological staining. Myelin, on the other hand, has an increased anisotropic diamagnetic contribution, which is strongly evident in white matter (WM) structures. Other components, such as calcium and copper, may also affect the local magnetic susceptibility.

Quantification of absolute metal concentration in the human body has already been investigated in the literature with a wide range of techniques (Grochowski et al., 2019). Specifically, the Inductively Coupled Plasma Mass Spectrometry (ICP-MS) provides sufficient sensibility for quantifying different trace elements in the range commonly found in the human brain, while also enabling simultaneous quantification of these elements (Krebs et al., 2014).

Apart from identifying iron contribution to the ${{\text{R}}_2}^{*}$ and QSM contrast, the knowledge about its molecular form is also important since it may improve interpretation of these maps. Although ferritin (Ft) has been suggested as the main storage of non-heme iron (Birkl et al., 2020; Harrison and Arosio, 1996), there may be other iron biomolecules that could potentially influence the contrast as well.

Quantification of molecular forms of iron is challenging as it requires specific targeting for a specific molecule, which is hardly applicable for a simultaneous analysis of different elements as in ICP-MS. Electron Paramagnetic Resonance (EPR) can simultaneously measure paramagnetic ions in a sample, while also differentiating its molecular structure (Cammack and Cooper, 1993). Previous works on EPR applied to human brain tissue showed that it can consistently measure at least four paramagnetic compounds, with two peaks associated to iron, one to copper and one to an organic radical (Otsuka et al., 2022). Further analysis and comparison to the Ft EPR spectra also indicates that the characteristic broad feature measured in EPR corresponds to the ferritin as well (Bossoni et al., 2023; Weir et al., 1985).

Therefore, by combining information from both ICP-MS and EPR techniques, further knowledge about tissue composition can be achieved and when comparing to quantitative MRI maps (qMRI) such as ${{\text{R}}_2}^{*}$ and QSM, the biophysical interplay between tissue’s magnetic composition and MRI signal can be inferred. In this study, the relationship between the ${{\text{R}}_2}^{*}$ and QSM contrast of eight GM structures to their corresponding metal and paramagnetic ions’ concentration was made by using data from eight GM structures. Analysis including all region of interest (ROI) were made and compared to the literature, while individual analysis of each ROI was also performed and investigated for regional differences in the GM’s susceptibility contrast sources.

2. Materials and methods

2.1. Subjects

15 postmortem subjects were included in this study, recruited from the Death Verification service of the Capital (SVOC) after signed consent by the next of kin. Subjects were between 36 and 91 years old (mean age = 66.53 +− 15.91 years old). Details about the subjects are included in Supplementary Material (ST1), which indicates the age, sex, postmortem interval (PMI), body temperature at the time of MRI experiment, death cause and fixation interval.

2.2. Image acquisition

In situ images from 15 postmortem subjects were obtained using a 7T MR scanner (Magnetom, SIEMENS Healthineers, Erlangen, Germany) with a 32-channel receiver head coil (Nova Medical, Wilmington MA, USA). A multi-echo 3D gradient recalled echo sequence (3DGRE) was used, with flip angle of 10°, 5 echoes (1st echo of 5 ms and echo time interval of 4 ms), repetition time of 25 ms, resolution of 0.5 × 0.5 × 1.0 mm³, and a Field of View of 203 × 224 × 128 mm³.

2.3. Image processing

The pipeline for in situ image processing is shown in Fig. 1. All the processing pipeline was performed on Matlab R2021a (The Mathworks Inc.). Phase images were coil-combined using the Virtual reference Coil approach (Parker et al., 2014), with linear weighting of magnitude images (Otsuka et al., 2023). Magnitude images were obtained by Sum-of-Squares. A combined magnitude image was generated by summing the square of the magnitude along each echo. This resulted in an improved image with an appreciable contrast-to-noise ratio compared to an image at a specific echo. Brain mask was generated using the BET algorithm in FSL software (Smith, 2002). After visual inspection, an additional erosion was performed in cases where the mask was poorly segmented. Phase images were combined following a complex fitting (Liu et al., 2013). Unwrapping procedure was performed using a 3D region-growing algorithm (Jenkinson et al., 2002). Background field filter was performed using the Projection onto Dipole Fields (PDF) algorithm (Liu et al., 2011) and dipole inversion using MEDI-L1 algorithm (Liu et al., 2012) in order to calculate the QSM map. Furthermore, ${{\text{R}}_2}^{*}$ maps were obtained by fitting the magnitude data to a single exponential, using the ARLO algorithm (Pei et al., 2015).

Fig. 1.

Image processing pipeline used in this work. Phase images were reconstructed by means of the Virtual Reference Coil approach, while the magnitude images were reconstructed via Sum-of-Squares. Phase images were echo-adjusted using a nonlinear complex fitting (MEDI toolbox), phase unwrapping using a path-following approach (SEGUE), background filter using PDF (MEDI toolbox) and dipole inversion using MEDI-L1 (MEDI toolbox). As for magnitude images, an echo averaging was performed in order to obtain an image with optimal CNR, brain mask was extracted using BET (FSL software) and refined in the dipole inversion algorithm. Finally, ${{\text{R}}_2}^{*}$ maps were calculated using the ARLO algorithm.

2.4. Image segmentation

ROIs were manually segmented using the ITK-SNAP software (Yushkevich et al., 2006) by comparing the MRI slices with the brain cuts at the ROI collection (Fig. 2). The following ROIs were segmented bilaterally along 5 consecutive slices to match the thickness of the collected samples: Caudate Nucleus (CN), Pre-Central Cortex (CPR), Entorhinal Cortex (ENT), Globus Pallidus (GP), Hippocampus (HIP), Putamen (PUT), Red Nucleus (RN) and Substantia Nigra (SN). The choice of the ROIs was made based on their composition and variability in the iron concentration.

Fig. 2.

ROI segmentation on magnitude images (top row) based on histological slices (bottom row), example in Subject 14 Table ST1. Images and photos are shown as in radiological convention.

2.5. Sample acquisition and preparation

After MR imaging, the brains were extracted and fixed in formaldehyde solution with a mean fixation time of 34.44 +− 7.17 months. Individual values for fixation time are included in Table ST1 as well. After this period, the brains were sliced in coronal sections using the pre-central gyrus as anatomical reference. Each slice was approximately 1 cm thick. During dissection photographs were taken of each slice of the brain for visual comparison to MRI (Fig. 2). Selected brain regions were cut and extracted by an experienced neuropathologist (RDR) by selecting slices of the brain which contained higher volumes of the specific structure. In total, 8 cerebral regions were extracted for this study.

In some cases, some cerebral regions could not be properly identified, and therefore were excluded from the study.

Brain samples were weighted (wet mass) and then freeze-dried and weighted again (dry mass). The freeze-dried samples were then grinded to powder, placed in closed tubes and stored in fridge for further reading.

For EPR readings, samples were carefully put into quartz tubes, and their mass was measured by subtracting the weight of the empty tube before putting the samples. It should be noted that some parts of the sample adhered to the tubes’ wall due to dipolar interactions. EPR is measured by placing the sample inside a cavity, which has a limited length of a few centimeters at which the spectrometer can measure, any additional sample outside of this range will not be accounted by the spectrometer. Therefore, an error associated to the mass that was effectively measured in the EPR cavity should be taken into consideration.

After EPR readings, each sample was separated into three tubes containing approximately 10 mg of sample. Samples were then solubilized following the methodology described in Batista et al. (2009) for ICP-MS reading.

During the whole procedure of sample acquisition and preparation, any contact to metal sources was avoided by using ceramic utensils.

2.6. EPR experiment and spectra processing

EPR experiments were carried out using an X-band spectrometer (JEOL, model JES-FA-200) with a microwave frequency of 9.45 GHz at room temperature. Due to some possible impurities in the cavity, the empty tube spectrum was acquired before every measurement with the sample, and then, subsequently subtracted to eliminate its contributions to the sample spectra.

EPR spectra that displayed a different behavior were excluded from analysis due to impossibility of further processing and quantifying. The resulting total number of samples used for further analysis was 206.

Due to the superposition of different signals, EPR spectra were processed via an adapted methodology previously described in Otsuka et al. (2022). In short, each signal was individually processed using EasySpin toolbox (Stoll and Schweiger, 2006). Narrow features were processed by windowing the spectra into the field range covering each peak. A polynomial baseline was found to give a reasonable baseline correction for the narrow peaks. Then, each narrow peak was fitted individually, and their contributions removed from the original spectra.

As for the broad feature, due to its large magnetic field range, usual baseline correction methods were found to be unreliable. Therefore, it was used a different approach which accounts for the shape of the absorption spectrum. It was found that this method results on a consistent corrected spectrum, without introducing any spectral distortions. It assumes that the first derivative of the spectrum is that of an absorption spectrum, and therefore it searches for the optimum linear baseline that returns an absorption shape of the first derivative of the spectrum. The baseline corrected spectra of the broad signal were fitted by implementing the two-spin system model, as described in Bossoni et al. (2023).

The spectra processing pipeline allowed the relative quantification of three paramagnetic peaks, named based on previous works:

• [LIP]: high-spin Fe(III) at rhombic environment ascribed to LIP or weakly bound iron (Kumar et al., 2016)

• [Cu_EPR]: Cu(II) with no specific molecule

• [Ft]: broad Fe(III) feature associated to ferrihydrite core of ferritin (Otsuka et al., 2022).

2.7. ICP-MS experiment

ICP-MS readings were performed on a spectrometer equipped with a reaction cell (DRC-ICP-MS ELAN DRCII, PerkinElmer, SCIEX, Norwalk, CT, USA). Full specifications of the spectrometer as well as the whole procedure for sample preparation and analysis are described in Batista et al. (2009). Absolute concentrations of [Fe], [Zn], [Cu], [Mn] and [Al] were measured on each sample since these metallic elements were quantified in previous studies and could influence the local magnetic susceptibility at some degree.

2.8. Statistical analysis

First, it was evaluated the contrast agreement between ${{\text{R}}_2}^{*}$ and QSM. For this purpose, a correlation test and linear regression was performed by including all segmented ROIs (206 ROIs in total). Next, an individualized analysis for each ROI (approximately 14 points per ROI) was also performed to evaluate the correspondence of contrast among different cerebral regions.

Then, results from MRI were compared against spectroscopic data by using four general linear models (GLMs) using all 206 samples, without discretizing by ROI. The rationale to use GLM was to explain variations in quantitative MRI maps in terms of all quantified elements (metallic trace elements or paramagnetic ions). Each model is represented in the equations below (Models 1 to 4).

Model 1:

$${{\text{R}}_2}^{*}={\alpha }_{Fe}[\mathit{F}\mathit{e}]+{\alpha }_{Zn}[\mathit{Z}\mathit{n}]+{\alpha }_{Cu}[\mathit{C}\mathit{u}]+{\alpha }_{Mn}[\mathit{M}\mathit{n}]+{\alpha }_{Al}[\mathit{A}\mathit{l}]+{\alpha }_0$$

Model 2:

$$\chi ={\beta }_{Fe}[\mathit{F}\mathit{e}]+{\beta }_{Zn}[\mathit{Z}\mathit{n}]+{\beta }_{Cu}[\mathit{C}\mathit{u}]+{\beta }_{Mn}[\mathit{M}\mathit{n}]+{\beta }_{Al}[\mathit{A}\mathit{l}]+{\beta }_0$$

Model 3:

$${{\text{R}}_2}^{*}={\alpha }_{LIP}[\mathit{L}\mathit{I}\mathit{P}]+{\alpha }_{Cu/EPR}\left[\mathit{C}{\mathit{u}}_{EPR}\right]+{\alpha }_{Ft}[\mathit{F}\mathit{t}]+{\alpha }_0$$

Model 4:

$$\chi ={\beta }_{LIP}[\mathit{L}\mathit{I}\mathit{P}]+{\beta }_{Cu/EPR}\left[\mathit{E}\mathit{C}{\mathit{u}}_{EPR}\right]+{\beta }_{Ft}[\mathit{F}\mathit{t}]+{\beta }_0$$

Where ${\alpha }_N$ represents the linear coefficients of the corresponding element N which explains the ${{\text{R}}_2}^{*}$ contrast, and ${\beta }_N$ represents the linear coefficients of the corresponding element N which explains the QSM contrast. The index N = 0 represents the intercept on each case. The elements included in each model were defined as the elements quantified via ICP-MS (Models 1 and 2) and via EPR (Models 3 and 4).

For each model, the quantitative parameter from MRI (${{\text{R}}_2}^{*}$ and QSM) was set as the dependent variable, and each quantified element (metallic trace elements from ICP-MS; paramagnetic ions from EPR) were used as independent variables to explain the contrast. Statistical analyses were corrected for multiple comparisons and p < 0.01 was chosen as statistically significant. No distinction between ROI was made at this step.

Subsequent analyses were performed only for the elements which were statistically significant from the GLMs. Linear regression and correlation tests were then performed between MRI (${{\text{R}}_2}^{*}$ and QSM) and the selected elements’ concentration. In this analysis, all 206 samples were included, which we refer to as global-level analysis.

Lastly, the heterogeneity in the concentration of trace elements and paramagnetic ions in the brain suggests that these elements may present variable contributions to the observed contrast. Therefore, linear regression and correlation tests were also performed individualized for each ROI (approximately 14 samples per ROI).

3. Results

3.1. qMRI vs spectroscopy: global analysis

Fig. 3.A shows a depiction of ${{\text{R}}_2}^{*}$ and QSM maps from the same random subject at the same slice. The graph on Fig. 2B displays the data plot from the mean values of ${{\text{R}}_2}^{*}$ and QSM extracted from the same ROI in each subject, totalizing 8 different ROIs per subject. A good linear correlation was obtained between maps (R² = 0.72; p < 0.01) and the linear regression yielded a slope of 0.53 s⁻¹/ppb using all ROIs.

Fig. 3.

(A) Example of ${{\text{R}}_2}^{*}$ and QSM calculated from the same axial image in a 56 years old female subject. (B) Data plot between ${{\text{R}}_2}^{*}$ and QSM showing the linear regression (shaded area represents the confidence interval) as well as the quality of the fitting (R² and p-value).

From the GLM models, it was found that only [Fe] and [Ft] showed statistical significance (Table ST2 in Supplementary information) and were able to explain the contrast. Fig. 4 shows the data plot of ${{\text{R}}_2}^{*}$ and QSM versus [Fe] and [Ft] including data point from all GM structures. The slope of the linear regression between ${{\text{R}}_2}^{*}$ and QSM versus [Fe] yielded a value of 0.53 s⁻¹/[Fe] and 0.81 ppb/[Fe], respectively.

Fig. 4.

Data plot of qMRI (${{\text{R}}_2}^{*}$ and QSM) versus spectroscopic data ([Fe] and [Ft]) showing the linear regression (shaded area represents the confidence interval) as well as the quality of the fitting (R² and p-value).

3.2. qMRI vs spectroscopy: individual analysis per structure

Fig. 5 shows the linear regression and correlation tests between ${{\text{R}}_2}^{*}$ and QSM individualized per ROI, where variable correlation can be identified.

Fig. 5.

Data plot between ${{\text{R}}_2}^{*}$ and QSM individually for each ROI showing the linear regression (shaded area represents the confidence interval) as well as the quality of fitting (R² and p-value). Red plots represent statistically significant (p < 0.05) results.

The heterogeneous distribution of [Fe] and [Ft] and the results indicated in Fig. 5 may suggest a variable contribution to the bulk magnetic susceptibility. To investigate the contribution of [Fe] and [Ft] on the contrast of each structure, an individual analysis was performed (Supplementary Data). Since some structures shared similar results regarding their correlation to both [Fe] and [Ft], for simplification of representation of data, these structures were grouped into three separate groups of structures, according to their correlations to [Fe] and [Ft] (Fig. 6).

Fig. 6.

Graphs from ROI-analysis of the relationship between qMRI (${{\text{R}}_2}^{*}$ and QSM) and spectroscopic data (ICP-MS and EPR). Images below indicated ROI locations, where each ROI is color-coded considering its group (Groups 1, 2 and 3).

Group 1 included structures with high iron concentration: CN, GP, PUT and RN. The high correlation found for both [Fe] and [Ft] indicates that the contrast of these structures is highly dominated by these elements, as expected due to their higher concentration. The estimated slopes for the structures of Group 1 between ${{\text{R}}_2}^{*}$ and QSM and [Fe] were closer to the obtained slope for the global analysis (Figs. SF1 and SF2, Supplementary Information), indicating that these structures are the main contributors for the global correlation.

Group 2 contained low iron concentration structures: CPR, ENT and HIP. These structures did not show any correlation to [Fe] or [Ft], suggesting that iron is not the dominant source of contrast on those regions. This could be explained due to their low iron concentrations.

Lastly, the Group 3 only contained SN, with a slope of 0.40 s⁻¹*[Fe] for ${{\text{R}}_2}^{*}$ and 0.86 ppb*[Fe] for QSM. However, it was not correlated to [Ft], suggesting a different molecular form of iron contributing to the contrast of this structure.

Fig. 6 shows the behavior of each group of structures in relationship do [Fe] and [Ft]. Results for each individual region are shown in Supplementary data (Figs. SF1–SF4).

4. Discussion

An evaluation of the influence of different metals and paramagnetic ions on the quantitative maps of MRI was performed by using both spectroscopic data (ICP-MS and EPR) with qMRI (${{\text{R}}_2}^{*}$ and QSM). This assessment was performed at two levels: global-level to compare with previous works and ROI-level to verify possible spatial heterogeneity of the main contrast source. Furthermore, the use of EPR allowed the quantification of specific Fe(III) structures, which are ascribed to [Ft] and [LIP], according to literature spectra of these biomolecules (Boas and Troup, 1971; Weir et al., 1985; Cammack and Cooper, 1993; Bou-Ab-dallah and Chasteen, 2008; Bossoni et al., 2023).

Starting by comparing results from both MRI techniques, a comparison between ${{\text{R}}_2}^{*}$ and QSM maps was performed by considering all included ROIs (Fig. 2). The high correlation coefficient indicates that both maps share similar contrasts at a global level. Previous reports of this slope at different field strength and conditions can be compared: 0.37 s⁻¹/ppb at 7T in vivo (Deistung et al., 2013), 0.22 s⁻¹/ppb at 4.7T in vivo (Sun et al., 2015) and 0.23 s⁻¹/ppb at 3T postmortem in situ (Barbosa, 2017).

The disparity of the reported values is unlikely to be due the differences between scanners and platforms (Ropele et al., 2014; Rua et al., 2020). However, both maps are strongly dependent on the processing algorithm used (Bilgic et al., 2021). Furthermore, ${{\text{R}}_2}^{*}$ is linearly dependent on B₀ in the range of 1.5T to 7T (Peters et al., 2007), resulting in an increase of approximately 2.33 and 1.49 when rescaling the values from 3T and 4.7T to 7T, respectively. Therefore, the rescaled slopes to 7T are: 0.37 s⁻¹/ppb (Deistung et al., 2013), 0.33 s⁻¹/ppb (Sun et al., 2015) and 0.57 s⁻¹/ppb (Barbosa, 2017). Therefore, the reported value of 0.53 s⁻¹/ppb is consistent with previously reported postmortem in situ values (0.57 s⁻¹/ppb) and higher compared to values found for in vivo conditions (0.37 s⁻¹/ppb and 0.33 s⁻¹/ppb).

The in vivo and in situ postmortem conditions also presents considerable differences in both ${{\text{R}}_2}^{*}$ and QSM, probably due to the strong influence of fully deoxygenated blood. The postmortem interval (PMI) of this study was around 13.60 +− 3.96 h (individual values are indicated in Table ST1 in Supplementary information), in which all the blood is already fully deoxygenated. From the results of this study, and comparing to the literature, it seems that the ${{\text{R}}_2}^{*}$ values are mostly affected by the differences between in vivo versus in situ conditions, compared to QSM.

A theoretical model of ${{\text{R}}_2}^{*}$ as a linear function $\chi$ is given by Yablonskiy and Haacke (1994), in which it considers a homogeneous static dephasing regime:

$${{\text{R}}_2}^{*}=\frac{2\pi }{9\sqrt{3}}\gamma B_0\chi +{R_{2,0}}^{*}$$

Where γ is the gyromagnetic ratio for the hydrogen proton, $B_0$ is the applied magnetic field and ${R_{2,0}}^{*}$ comprises the ${{\text{R}}_2}^{*}$ effects non-related to $\chi$. This would give a slope of 0.75 s⁻¹/ppb at 7T. It should be noted that the theoretical value is dependent on the geometric shape of the source, which in the above equation was assumed to be spherical. Furthermore, the multi-compartment nature of the ${{\text{R}}_2}^{*}$ should also be taken into consideration.

Compared to ${{\text{R}}_2}^{*}$, QSM is a relatively recent technique. Nonetheless, both techniques have successfully demonstrated a correlation to iron. In this study, a similar trend was observed for the GM structures. The white matter structures (WM) were not included in this study as they present features which would require additional methodological considerations, out of the scope of this study.

Comparing the results of QSM vs [Fe] to the ones reported in literature (Table 2), it was observed that the obtained slope in this study is in the range of the reported slopes: 0.80 ppb/[Fe] to 1.03 ppb/[Fe]. The differences in values may be attributed to the effect of temperature on QSM values (Birkl et al., 2015), QSM processing pipeline and iron quantification methods.

Table 2.

Information and results of the linear regression between ${{\text{R}}_2}^{*}$ and [Fe] from different studies

${{\text{R}}_2}^{*}$ vs Iron studies
R 2	Slope (s−1/[56Fe])	Intercept (s−1)	Iron Estimation method	Field Strength (T)	Sample	Refs.
0.90	0.27	14.28	ICP-MS	3	7 postmortem (in situ)	Langkammer et al. (2010)
0.84	0.15	17.97	ICP-MS	3	3 postmortem (in situ)	Barbosa (2017)
0.97	0.05	1.00	ICP-MS	3	Ferritin (horse spleen)	Barbosa (2017)
0.42	0.66	43.96	Colorimetry	7	6 postmortem (in situ)	Hametner et al. (2018)
0.83	0.53	47.91	ICP-MS	7	15 postmortem (in situ)	This paper

A theoretical estimate of the iron nucleated in Ft’s contribution to χ is given in Schenck (1992)and yields a value of 1.40 ppb/[Fe] (where [Fe] is measured in μg/g) by assuming Bohr magneton to be 3.78 and brain density of 1.04 g/cm³ at 15^oC and ignoring other contributions, such as from myelin and other possible sources. Previous experimental data using solutions with different Ft concentrations and their respective QSM maps have reported slopes of 1.11 ppb/[Fe] (Zheng et al., 2013) and 1.31 ppb/[Fe] (Barbosa, 2017) from the linear regression (temperature information not informed by these studies).

The measured slope (QSM vs [Fe]) in this study was found to be around 27 % and 38 % lower than the results with pure ferritin phantoms (Table 1), and 42 % lower than the theoretical result. This indicates that the portion of the susceptibility contrast related to iron is not entirely derived from ferritin, indicating that there are other sources of iron contributing to the measured susceptibility. This observation was also corroborated by our EPR analysis, where the correlation to the Ft signal was lower than that observed for total iron concentration.

Table 1.

Information and results of the linear regression between QSM and [Fe] from different studies.

QSM vs Iron studies
R 2	Slope (ppb/[Fe])	Intercept (ppb)	Iron Estimation Method	Field Strength (T)	Sample	Refs.
0.84	0.89	−22	ICP-MS	3	7 postmortem (in situ)	Langkammer et al. (2012)
0.76	0.80	10.8	XRF	3	1 postmortem (frozen)*	Zheng et al. (2013)
0.99	1.11	−32.36	ICP-MS	3	Ferritin (phantom)	Zheng et al. (2013)
0.86	0.71	−16	ICP-MS	3	3 postmortem (in situ)	Barbosa (2017)
0.99	1.31	38	ICP-MS	3	Ferritin (phantom)	Barbosa (2017)
0.40	1.03	3.8	Colorimetry	7	6 postmortem (in situ)	Hametner et al. (2018)
0.70	0.81	68.8	ICP-MS	7	15 postmortem (in situ)	This paper

^*Multiple Sclerosis patient’s brain.

Comparing the results of ${{\text{R}}_2}^{*}$ vs [Fe] to the ones reported in literature (Table 2), it was observed that the obtained slope in this study is in the range of the reported slopes: 0.35 s⁻¹/[Fe] to 0.66 s⁻¹/[Fe], values adjusted after accounting for the linear B₀-dependence of ${{\text{R}}_2}^{*}$ maps.

Experimental data suggests a linear dependence between ${{\text{R}}_2}^{*}$ and iron with slope of 0.12 s⁻¹/[Fe] with a horse spleen ferritin phantom (Barbosa, 2017) when rescaling the values to 7T. This value is considerably lower than that found for brain tissue, however, the phantom used in Barbosa (2017) also presented ${{\text{R}}_2}^{*}$ values considerably lower than the human brain tissue (roughly 4 to 5 times lower).

As an accurate Ft content indicator, EPR was used to estimate relative concentration of paramagnetic Fe(III) at different sites, namely the [LIP] and [Ft]. The missing correlation between [LIP] and ${{\text{R}}_2}^{*}$ and QSM suggests that this form of iron in the sample is not involved in the main contrast mechanism of gray matter tissue. On the other hand, [Ft] showed good correlation to both ${{\text{R}}_2}^{*}$ and QSM, indicating that the Fe(III) ions in the Ft core may be one of the major sources of iron-related contrast in these techniques.

Calibration of EPR data is much more complex than the ICP-MS as it requires the use of standard sample with similar composition as that of the measured paramagnetic peak. The choice of standard sample is not straightforward as it requires that the standard is measured together with the sample itself. This is not feasible for the brain EPR data due to the broad range of the Ft signal, which would result in a superposition with the standard sample. Therefore, to simplify the measurements, a relative concentration was opted.

The global-level results suggest strong influence of iron and ferritin to the contrast of ${{\text{R}}_2}^{*}$ and QSM techniques. However, given the highly heterogeneous distribution of iron in the brain (Hallgren and Sourander, 1958), it remains to be evaluated if this influence is also variable throughout the brain. To test the heterogeneity in the correlation between ${{\text{R}}_2}^{*}$ and QSM vs Fe and Ft, an individual analysis per ROI was performed.

For simplicity of displaying the results, structures were grouped into three groups according to the observed correlations (strong correlation to Fe and Ft; no correlation to Fe and Ft; partial correlation fo Fe and no correlation do Ft). Results for each ROI separately are shown in supplementary data (Figs. SF1–SF4).

These results indicate variable influence of [Fe] to the contrast of both QSM and ${{\text{R}}_2}^{*}$ in GM, which seems to be dependent on the [Fe] concentration of the tissue. This can be understood as an increase in the relative contribution of other sources when [Fe] levels are low, since bulk χ distribution also depends on the concentration of the substance in addition to its intrinsic χ value. Furthermore, by using EPR data to further specify the iron state and molecule as Ft, we observed that SN, although well correlated to Fe, was not correlated to Ft, suggesting that there exists another molecular form of iron contributing to its contrast.

While the number of subjects included in this study is considerably higher compared to previous similar studies, this study has its own limitations. The long fixation time may have affected the iron distribution in the brain (Schrag et al., 2010), however, the results from this study were shown to agree with previous reports from literature, indicating that, at least for the BG structures, the prolonged fixation time did not interfere with the results. However, the lack of literature data concerning cortical structures poses a question whether these structures are affected or not by the fixation time effects.

The choice of one from various QSM pipelines has considerable impact on the resulting values, which should be considered when comparing different studies (Langkammer et al., 2018; Bilgic et al., 2021). Also, the choice of quantification methodology may have considerable impact on the measured iron concentration. In this study, the ICP-MS was chosen as a trace element’s quantification due to its high sensibility, while also allowing simultaneous quantification of multiple metallic elements.

Although the inclusion of cortical structures introduced additional information regarding GM contrast, background field artifacts are amplified at the boundaries of the brain mask, affecting the estimated χ values on structures such as the CPR. However, ${{\text{R}}_2}^{*}$ maps are more stable in these regions and demonstrated similar results as QSM regarding [Fe] and [Ft]. Additionally, the low sensibility of the coils on the CPR region could also have contributed to a lack of observed correlation, due to its low SNR.

As for the SN, there was no differentiation between its sub-structures (pars compacta and pars reticulata) for our analysis. It is known that these sub-structures have different composition (Snyder and Connor, 2009). However, dividing the SN into two sub-structures would result in a smaller sample for the EPR measurements, rendering lower SNR spectra.

Finally, EPR is blind to Fe(II) and Cu(I), which means that EPR cannot detect all ionic states of these metals and therefore. Furthermore, oxidation processes in the brain were shown to have some influence on the contrast of the images (Birkl et al., 2020), and although no correlation was found to [LIP], other iron forms formed as a byproduct of oxidation cannot be excluded. However, these forms of iron are expected to be in a less stable state, contributing mostly to the [LIP] signal instead of [Ft].

Nonetheless, iron was shown to have variable contributions to the qMRI maps, which is highly dependent on its relative concentration in the structure. While it is still left to be explored whether qMRI contrast changes due to diseases have any relationship to these elements, the assignment of these changes entirely to iron overload should be carefully made, especially in structures with low iron concentration.

5. Conclusion

By combining data from qMRI maps based on magnetic susceptibility (${{\text{R}}_2}^{*}$ and QSM) to spectroscopic data for investigation of metallic trace elements (ICP-MS) and paramagnetic ions (EPR) in human brain tissue, further understanding about the biophysical relationship between qMRI contrast and metals’ concentration in the tissue was gained. It was found that both ${{\text{R}}_2}^{*}$ and QSM display varied behavior among different gray matter structures, with the agreement possibly being highly influenced by iron concentration distribution in the tissue as observed by individual analysis of each ROI. By analyzing the relationship of qMRI to ICP-MS and EPR data it was possible to group structures into three groups according to their correlations. The first group is represented by iron-rich structures with strong correlation to both [Fe] and [Ft], the second group is represented by low iron structures with no correlation to neither [Fe] and [Ft] and the third group was the SN, which contain high levels of iron and was correlated to [Fe], however not to [Ft]. The distinct behavior of the SN structure indicates that there may be another contributor to the contrast of this structure, possibly the NM. This study sets a standard baseline of the relationship between qMRI and spectroscopic data, which could be used as a comparison in future studies applied to diseased brain samples to investigate any differences in the patterns we found throughout this study.

Data availability

Data will be made available on request.