In Vivo Assessment of Age-related Brain Iron Differences by Magnetic Field Correlation Imaging

Abstract

Purpose

To assess a recently developed Magnetic Resonance Imaging (MRI) technique called Magnetic Field Correlation (MFC) imaging along with a conventional imaging method, the transverse relaxation rate (R2), for estimating age-related brain iron concentration in adolescents and adults. Brain region measures were compared to non-heme iron concentrations (C_PM) based on a prior postmortem study.

Materials and Methods

Asymmetric spin echo (ASE) images were acquired at 3T from 26 healthy individuals (16 adolescents, 10 adults). Regions of interest (ROIs) were placed in areas in which age-related iron content has been estimated post-mortem: globus pallidus (GP), putamen (PUT), caudate nucleus (CN), thalamus (THL) and frontal white matter (FWM). Regression and group analyses were conducted on ROI means.

Results

MFC and R2 displayed significant linear relationships to C_PM when all regions were combined. Whereas MFC was significantly correlated with C_PM for every individual region except FWM and detected significantly lower means in adolescents than adults for each region, R2 detected significant correlation and lower means for only PUT and CN.

Conclusion

Our results support the hypothesis that MFC is sensitive to brain iron in GM regions and detects age-related iron increases known to occur from adolescence to adulthood. MFC may be more sensitive than R2 to iron-related changes occurring within specific brain regions.

INTRODUCTION

Iron is the most abundant metal in the human body (1). Normal human brain contains an unusually high concentration of iron, with some regions having up to 5 mg Fe/100 g fresh tissue more iron than the primary storage site in liver (2). Although the role of iron in the brain is not fully understood, several important neurodevelopmental and functional processes are known to require iron, including neurotransmitter synthesis, myelin production and maintenance, brain oxygen transport, as well as the Fenton reaction, which is crucial to oxidative stress (3, 4).

Endogenous iron is classified as either heme or non-heme iron; heme iron is found in oxygen transporting hemoglobin and represents two-thirds of all iron in the body while brain iron exists mainly as non-heme iron primarily in the iron-storage molecules, ferritin and hemosiderin (5). Postmortem studies have revealed the regional distribution of iron in the brain to be heterogeneous and higher in gray matter (GM) than white matter (WM) by a ratio ranging from 2:1 to 4:1 (3, 6, 7). During typical development and aging, brain iron concentration is extremely low at birth but progressively accumulates until the end of the fourth decade of life with varying rates depending on the region. Starting from the first few years of life, the highest iron concentrations occur in the basal ganglia.

The disruption of iron homeostasis in the brain has been implicated in numerous neuropathologies and is hypothesized to play a role in neurodevelopmental disorders. In Parkinson’s and Alzheimer’s diseases, regional elevations in iron concentration are found, whereas iron is deficient in restless legs syndrome as well as in preliminary findings on attention-deficit hyperactivity disorder (8 – 11). The study of such iron-related pathologies may, therefore, benefit from in vivo imaging methods for characterizing brain iron. Since both ferritin and hemosiderin iron are strongly magnetic and have non-uniform microscopic spatial distributions in brain (7, 12 – 14), they generate microscopic magnetic field inhomogeneities (MFIs) that significantly increase the proton transverse nuclear magnetic resonance relaxation rates R2 and R2* (15 – 19). Consequently, the magnetic resonance imaging (MRI) measurement of R2 and R2* have become common methods to quantify brain iron (20 – 30). However, the specificity and reliability of these methods is limited because both R2 and R2* are also affected by other molecular relaxation mechanisms unrelated to iron (31). The derived quantity R2′ = R2* – R2 may be more specific to iron than R2 or R2* because molecular relaxation mechanisms are expected to affect R2 and R2* in the same way, thus not affecting R2′ (23). However, taking the difference of R2* and R2 also diminishes the contribution of MFIs, thus reducing the sensitivity of R2′ to iron induced MFIs. Alternative MRI methods, such as the measurement of field dependent R2 increase (FDRI) and susceptibility weighted imaging (SWI), have been developed to bypass the limitations of these relaxation rate measures (5, 32, 33, 34). FDRI measurement is a sensitive method for quantifying brain iron content but is practically limited due to requiring measurement at two field strengths. On the other hand, SWI is not quantitative, although current work aimed at quantifying magnetic susceptibility via Quantitative Susceptibility Mapping (QSM) is being pursued (35, 36).

Another approach for quantifying microscopic MFIs is with the magnetic field correlation (MFC), which has a direct and well-defined relationship to MFIs generated by spatial variations in magnetic susceptibility such as those caused by iron rich glial cells (37–44). Compared to R2 and R2*, MFC has the advantage of being independent of molecular relaxation mechanisms unrelated to iron (e.g., dipolar interactions with proteins) thus allowing for a clearer physical interpretation. The goal of this study is to investigate the application of MFC imaging to assess the age-related intra-regional differences in brain iron and to compare these results with those obtained using the transverse relaxation rate, R2.

Both MFC and R2 values were measured in 26 healthy participants consisting of 16 adolescents and 10 adults. To focus on regions with pre-existing estimates of age-dependent iron content, a region of interest (ROI) analysis was conducted for the globus pallidus (GP), putamen (PUT), caudate nucleus (CN), thalamus (THL) and frontal white matter (FWM). MFC and R2 measures for these regions were compared to non-heme iron concentrations (C_PM) obtained from the seminal postmortem work of Hallgren and Sourander (6). As intra-regional differences in postmortem brain iron between adolescents and adults have been well documented in the aforementioned regions (6), we examined how well either metric (MFC and R2) was able to detect these known differences in vivo. This analysis provides a specific test of the relative sensitivities of MFC and R2, which may help inform the application of these metrics for assessing intra-regional iron differences due to pathology (20, 24, 25, 27, 29, 38, 41).

MATERIALS AND METHODS

Theory

The MFC is defined by the following equation (37)

$$\text{MFC}(t-t^{\prime })\equiv {\gamma }^2\langle \delta B(t)\delta B(t^{\prime })\rangle .$$ [1]

Here δB(t) represents the difference between the local magnetic field experienced by a water proton at a time t and the uniform background field, γ is the proton gyromagnetic ratio, and the angle brackets indicate an averaging over all the water molecules within a voxel. MFC(t) provides a well-defined metric for MFIs and is typically a monotonically decreasing function in time (37, 43), reflecting a loss of correlation due to water diffusion. It has been demonstrated that MFC can be estimated with MRI using an asymmetric spin-echo (ASE) pulse sequence with multiple refocusing pulse time shifts (37).

To extract MFC values at a given echo time (TE), the signal intensities as a function of the 180° refocusing pulse time shift (t_s) were fitted to

$$S(\text{TE},t_s)=a_{1}exp(-2a_2{t}_s^2),$$ [2]

where S(TE, t_s) is the signal intensity corresponding to the refocusing pulse time shift of t_s at TE. The two fitting parameters a₁ and a₂ were determined by nonlinear least-squares minimization using the Levenberg-Marquardt algorithm (45), and a₂ provided an estimate for MFC at time t = TE/2.

Participant Inclusion and Exclusion Criteria

Twenty-six healthy individuals comprised of 16 adolescents (13.2 – 17.9 years, mean = 15.9, standard deviation (SD) = 1.6) and 10 adults (21.2 – 43.9 years, mean = 32.1 years, SD = 7.4) were recruited from our institution’s Child Study Center, Medical Center and the local community. Parent and child informed consent and assent, respectively, were obtained as approved by our Institutional Review Board. Absence of known neurological or chronic medical diseases was required of all subjects. Diagnosis of any DSM-IV axis I psychiatric disorders including psychotic, major depressive, conduct, tic, and pervasive developmental disorders were exclusionary. For adolescents, inclusion required an estimated full-scale IQ > 80 measured by the Wechsler Abbreviated Scale of Intelligence (WASI) (46), scores in the normal range (<65) on the parent-reported Child Behavior Checklist (47) and lack of DSM-IV psychiatric diagnoses based on the child version of the Anxiety Disorders Interview Schedule (48).

Putative Non-Heme Iron Concentration

In order to validate and assess the relationship between our measured MFC and R2 values compared to putative non-heme iron concentrations (C_PM), we employed a C_PM comparison used extensively in the literature - the C_PM values obtained in brain tissue from 98 post-mortem subjects between the ages of 0 – 100 years by Hallgren and Sourander (6, 8, 11, 14, 23, 24, 26, 27, 29, 32, 35, 39, 49–52). The C_PM for GP, PUT, CN and FWM for each subject’s age was calculated from their published regression equations for age-related changes in regional brain iron concentration. This was done using the empirical age regression formula

$${\text{C}}_{\text{PM}}=\alpha \left[1-{exp}^{-\beta t}\right]+\delta$$ [3]

where C_PM is the non-heme iron concentration (mg Fe/100 g fresh weight) at age t (years), α + δ is the asymptotic iron concentration for long times, β is the time-constant for the accumulation rate of iron and δ is the iron concentration at age zero. Regional formula values for α, β, δ are found in Hallgren and Sourander (6). In the case of the THL, no such equation was provided; thus we derived the regression formula for the postmortem non-heme iron versus age using a least squares fit of this region’s postmortem data from Hallgren and Sourander to a biexponential functional form, yielding the regression formula

$${\text{C}}_{\text{PM}}=14.19exp[-0.013t]-13.44exp[-0.0439t].$$ [4]

Image Acquisition

Participants were scanned at 3T (Allegra, Siemens Medical Solutions, Erlangen, Germany) with a quadrature head coil. To estimate MFC, ASE images were acquired with a segmented echo-planar-imaging (EPI) sequence using five different refocusing pulse timings having time shifts t_s = 0, −3, −6, −9, −12 ms. Here the negative sign indicates a reduction of the time interval between the initial 90° excitation pulse and a 180° refocusing pulse from the typical value of TE/2 in the case of a symmetric spin echo. The choice of maximum pulse time shift is a compromise between precision and accuracy, since the magnitude of the signal change relative to noise increases with shift size (improving precision) but the accuracy of the fitting model of Eq. [2] decreases. Prior MFC studies of brain (37 – 39, 41 – 44) have used maximum shifts of 12 to 16 ms for the data fitting, in order to attain both reasonable precision and accuracy. Other parameters included: repetition time (TR)/TE = 2000/50 ms, flip angle = 90°, field of view (FOV) = 230 × 230 mm², matrix = 128 × 128, oblique axial slices (number/thickness/gap) 20/1.8 mm/2.7 mm, EPI factor = 43 (three segments per image), bandwidth = 1775 Hz/pixel, 25 averages with fat saturation, total scan time = 12 minutes 30 seconds. An isotropic dimension of 1.8 × 1.8 × 1.8 mm³ was chosen as a balance between minimizing the effect of macroscopic gradients and maintaining an adequate signal-to-noise ratio (SNR). To estimate R2, additional zero shift images (i.e., t_s = 0 ms) with the same parameters were acquired but at TE = 32 ms. Total scan time for R2 zero shift images = 5 minutes.

All imaging sequences were obtained with partial brain coverage in order to focus on iron rich deep cortical nuclei and to reduce scan time. The acquisition orientation was determined mid-sagittally by aligning the axial slice grid to include the middle rostrum and splenium of the corpus callosum in the central slice for each subject. Gaps were included to ensure the images were not contaminated by slice crosstalk. We note here that the most current established MFC protocol allows for whole brain coverage without gaps in a scan time of about 1.5 minutes per average and requires only 4 averages at 3T for a total scan time of approximately 6 minutes (42, 44).

Image Processing

Prior to MFC calculation, all zero-shift images were coregistered to the first zero-shift (i.e., spin echo) image (TE = 50 ms) using SPM8 (Statistical Parametric Mapping, Wellcome Department of Imaging Neuroscience, University College London, UK) and the rigid-body transformation obtained from coregistering each zero-shift image was applied to all the ASE images corresponding to that average. Parametric maps of the MFC were generated on a voxel-by-voxel basis with a nonlinear least squares fit to Eq. [2], using software running in MATLAB (version 7.9.0.529, R2009b; Mathworks, Inc., Natick, MA, USA) (39). MFC maps were thresholded to have a minimum value of 0 s⁻² and a maximum value of 2000 s⁻², which included the expected physical range.

For the estimation of R2, the additional zero shift images acquired with TE = 32 ms were also co-registered to match zero shift images with TE = 50 ms in SPM8. The R2 parametric maps were derived on a voxel-by-voxel basis from the zero shift signal intensities S(TE, 0) for TE₁ = 32 ms and TE₂ = 50 ms using the following equation

$$\text{R}2=\frac{1}{{\text{TE}}_2-{\text{TE}}_1}ln\left[\frac{S({\text{TE}}_1,0)}{S({\text{TE}}_2,0)}\right].$$ [5]

In order to avoid collecting values from voxels with partial volume effects, cerebrospinal fluid (CSF) was removed from all parametric maps. This was done in SPM8 by tissue segmenting the zero shift image (TE = 50 ms) into GM, WM and CSF. A CSF binary mask was created from the segmented CSF image (voxels > 0.10 = 0, rest = 1) and applied to the MFC and R2 parametric maps.

Region of Interest Analysis

For each subject, MRIcron (Version 1, April 2010) was used to manually specify bilateral ROIs on the zero shift image (TE = 50 ms) for the GP, PUT, CN, THL and FWM (Fig. 1). Each ROI was delineated according to anatomical boundaries with the exception of FWM which was sampled with square ROIs. Each region was sampled in every slice where the region was visible, except for the two most inferior slices where EPI distortion was prevalent and for the FWM where the last superior slice sampled was the slice prior to the emergence of the corpus callosum body. The ROIs were then applied over the MFC and R2 maps to extract metric means and standard errors.

Fig. 1. Regions of interest.

Bilateral regions of interest (ROIs) were manually drawn on the zero shift image (TE = 50 ms) for the globus pallidus (GP), putamen (PUT), caudate nucleus (CN), thalamus (THL) and frontal white matter (FWM). Each ROI was delineated according to anatomical boundaries with the exception of FWM which was sampled with square ROIs. Each region was sampled in every slice where the region was visible except for the two most inferior slices where echo planar imaging (EPI) distortion was prevalent and for the FWM where the last superior slice sampled was the slice prior to the emergence of the corpus callosum body. The ROIs were then applied over the MFC and R2 maps (that had CSF removed) to extract metric means and standard errors.

Statistical analyses were performed using SPSS (Version 17.0.0). Linear regression analysis on each metric versus C_PM measures was conducted globally for all regions combined and for individual regions, including calculation of Pearson’s correlation coefficients. For each region, individual MFC and R2 means were plotted as a function of age along with regional age-dependent growth curves defined by Hallgren and Sourander (6) adjusted by our obtained linear regression equation for correct scaling. Group comparison between adolescent subjects versus adult subjects was investigated for each region using two-sample t-tests (2-tailed).

RESULTS

Linear regression analyses of individual metric means for all the regions combined versus C_PM were significant for both MFC (r = 0.75, p < 0.001, Fig. 2A) and R2 (r = 0.92, p < 0.001, Fig. 3A). However, individual regional analysis found MFC to be significantly correlated with C_PM for all regions (THL r = 0.39, p = 0.046, Fig. 2C; CN r = 0.61, p = 0.001, Fig. 2D; PUT r = 0.64, p < 0.001, Fig. 2E; GP r = 0.66, p < 0.001, Fig. 2F) except FWM (r = 0.31, p = 0.128, Fig. 2B) whereas R2 was significantly correlated only for CN (r = 0.80, p < 0.001, Fig. 3D) and PUT (r = 0.84, p < 0.001, Fig. 3E). A summary of all linear regression correlation coefficients and p-values is shown in Table 1. For combined brain regions, R2 has been previously expressed as a linear combination of non-heme iron concentration, macromolecular mass fraction (f_M) and a region independent constant term (29, 50). For the regions considered, f_M varies from 0.1857 (CN) to 0.2936 (FWM) (50). Accounting for f_M in a multiple linear regression analysis for combined regions increased the correlation coefficient for R2 (R2 = 0.6(C_PM) + 25.7(f_M) + 7.2; r = 0.96, p < 0.001) but had no effect on the correlation coefficient for MFC (MFC = 20.2(C_PM) + 270.1(f_M) + 8.65; r = 0.75, p < 0.001).

Fig. 2. Linear regression of putative postmortem iron concentration versus MFC means.

A. Global linear regression of all regions. B. Linear regression of frontal white matter (FWM, diamonds). C. Linear regression of thalamus (THL, squares). D. Linear regression of caudate nucleus (CN, upright triangles). E. Linear regression of putamen (PUT, circles). F. Linear regression of globus pallidus (GP, downward triangles). For each subplot: X-axis is the putative postmortem non-heme iron concentration (C_PM), Y-axis is MFC mean, the linear regression equation, Pearson’s correlation coefficient and p-value are reported.

Fig. 3. Linear regression of putative postmortem iron concentration versus R2 means.

A. Global linear regression of all regions. B. Linear regression of frontal white matter (FWM, diamonds). C. Linear regression of thalamus (THL, squares). D. Linear regression of caudate nucleus (CN, upright triangles). E. Linear regression of putamen (PUT, circles). F. Linear regression of globus pallidus (GP, downward triangles). For each subplot: X-axis is putative postmortem non-heme iron concentration (C_PM), Y-axis is R2 mean, the linear regression equation, Pearson’s correlation coefficient and p-value are reported.

Table 1.

MFC and R2 Correlation with Postmortem Iron

Regions	MFC (s−2)	R2 (s−1)
All	0.75, < 0.001	0.92, < 0.001
GP	0.66, < 0.001	0.37, 0.063
PUT	0.64, < 0.001	0.84, < 0.001
CN	0.61, 0.001	0.80, < 0.001
THL	0.39, 0.046	0.24, 0.242
FWM	0.31, 0.128	− 0.06, 0.778

Pearson’s r, corresponding p-value, n = 26

Table 2 summarizes the C_PM, MFC and R2 regional means for our cohort age groups. Group analyses examining age-related differences between adolescents and adults found similar trends. For each of the regions, MFC means were significantly lower for adolescents than adults corresponding to significant group differences observed in C_PM means (Table 3). R2 regional means were also significantly lower for adolescents but only for PUT and CN (Table 3). Metric means plotted against age and overlaid with adjusted projected growth curves obtained from Hallgren and Sourander show similar fits for both MFC and R2 means for all ROIs except FWM (Fig. 4).

Table 2.

C_PM, MFC and R2 Regional Means for Adolescents, Adults and the Entire Subject Population

Group	Age	Region	CPM*	MFC (s−2)	R2 (s−1)
Total (n = 26)	22.1 ± 2	GP	18.0 ± 0.4	440 ± 38	25.3 ± 0.3
		PUT	8.7 ± 0.4	195 ± 12	17.4 ± 0.2
		CN	6.5 ± 0.2	245 ± 14	16.2 ± 0.1
		THL	5.2 ± 0.1	139 ± 8	16.7 ± 0.1
		FWM	3.7 ± 0.1	197 ± 10	16.8 ± 0.1
Adolescents (n = 16)	15.9 ± .4	GP	16.6 ± 0.2	337 ± 26	25.0 ± 0.3
		PUT	7.3 ± 0.1	170 ± 10	16.8 ± 0.1
		CN	5.6 ± 0.1	213 ± 14	15.8 ± 0.1
		THL	4.7 ± 0.1	125 ± 8	16.7 ± 0.1
		FWM	3.4 ± 0.03	180 ± 9	16.8 ± 0.2
Adults (n = 10)	32.1 ± 7	GP	20.3 ± 0.3	605 ± 62	25.9 ± 0.6
		PUT	10.9 ± 0.4	235 ± 21	18.3 ± 0.3
		CN	7.9 ± 0.2	297 ± 22	16.9 ± 0.2
		THL	5.8 ± 0.1	161 ± 13	16.8 ± 0.3
		FWM	4.1 ± 0.04	225 ± 20	16.8 ± 0.2

mean ± standard error,

^*(mg Fe/100 g fresh weight)

Table 3.

Group Analysis of Regional Means

Group	Age	Region	CPM*	MFC (s−2)	R2 (s−1)
Adolescents v. Adults	< 0.001	GP	< 0.001	0.002	0.161
		PUT	< 0.001	0.016	0.001
		CN	< 0.001	0.002	0.001
		THL	< 0.001	0.017	0.507
		FWM	< 0.001	0.032	0.938

two sample t-test (2-tailed) p-value,

^*(mg Fe/100 g fresh weight)

Fig. 4. MFC and R2 means with age.

Left column graphs are MFC means plotted over age for globus pallidus (GP, downward triangles), putamen (PUT, circles), caudate nucleus (CN, upward triangles), thalamus (THL, squares) and frontal white matter (FWM, diamonds) (top to bottom graphs, respectively). Right column graphs are R2 means plotted over age for GP, PUT, CN, THL and FWM (top to bottom graphs, respectively). For all subplots: X-axis is age in years, left Y-axis is the MFC (left column graphs) or R2 (right column graphs) metric means corresponding to the individual points, and right Y-axis is putative postmortem non-heme iron concentration (C_PM) corresponding to the adjusted growth curve from Hallgren and Sourander (6) (solid line).

To provide an additional comparison, MFC was recalculated with just two time shifts (t_s = 0, −12 ms). Linear regression analysis of the recalculated MFC and C_PM retained all significant correlations as the MFC analysis with five time shifts except for THL (all regions: r = 0.79, p < 0.001; GP: r = 0.65, p < 0.001; PUT: r = 0.62, p = 0.001; CN: p = 0.66, p < 0.001; THL: r = 0.26, p = 0.207; FWM: r = 0.28, 0.164). Group analysis of regional means retained significantly lower adolescent MFC for the GP (p = 0.006) and CN (p = 0.010) with a lower trend for PUT (p = 0.055).

While there are several methods for estimating the signal-to-noise ratio (SNR) of in vivo images, as a simple indicator we applied the method suggested by Kaufmann et al. (53) whereby SNR is estimated from the mean signal for an ROI within the subject divided by the mean signal for air. For the within subject ROI, we chose CSF, as this is less affected by T2 signal decay, and for the air signal, we chose an ROI outside the head in the read encoding direction to avoid ghosting artifacts. This procedure was applied to the images acquired with TE = 50 ms and zero shift for the 180° refocusing pulse, which are used for both the MFC and R2 calculations, yielding an SNR of 103.5 (SEM = 2.1). Similar results were obtained for the other images used for MFC and R2 calculations.

DISCUSSION

Given that iron is essential in many important processes of normal brain development and aging together with the fact that it has been implicated in certain neuropathologies (3, 4, 8 – 11), there is great interest in developing and validating in vivo imaging methods for characterizing brain iron levels. The investigation of neuropathologies typically involves comparing brain iron content between patient and control groups within specific regions (11, 20, 24, 25, 27, 29, 38, 41). To corroborate the sensitivity of both MFC and R2 to such intra-regional differences in pathology, our current study examined how well either metric was able to detect the intra-regional brain iron differences known to exist between adolescents and adults (6). This is the first study to investigate the age-related intra-regional differences in brain iron using MFC and R2.

The MFC theory and imaging methodology has been discussed in detail by Jensen et al. (37, 39). As previously described, MFC has a well-defined direct relationship to MFIs generated by spatial variations in magnetic susceptibility. These variations arise from microscopic magnetic tissue structures with length scales of roughly 1 to 100 μm (e.g., iron rich glial cells, extracellular paramagnetic contrast agents, capillaries) and to a lesser extent, from larger structures (e.g., sinus air cavity) that produce macroscopic field gradients. These macroscopic contributions are typically small except in brain regions near air/tissue interfaces, bone and large veins. Compared to R2, MFC is a more specific metric of microscopic MFIs because it depends only on MFIs and weakly on water diffusion (which acts as a lowpass spatial filter). R2, in contrast, depends strongly on molecular dipole-dipole interactions generated by tissue constituents unrelated to iron. Thus acquiring MFC, along with R2 measures, provides complementary information that is more specific to iron-related MFIs, which allows for a clearer physical interpretation of observed group differences.

Although our focus is in detecting intra-regional brain iron differences, several studies have demonstrated strong inter-regional correlations between R2 and iron in the GM structures of healthy human and non-human primate brains (16, 23, 26, 49, 54, 55). However, this correlation lessens when analyses of WM or neuropathology are included (5) and may reflect the fact that R2 is affected by tissue properties, unrelated to iron, that contribute to molecular relaxation mechanisms. These tissue properties (i.e. water and protein content, ratio of free to bound water, macromolecular mass fraction, water compartmentalization and microstructure density) differ between GM and WM and are modified in normal aging and pathology (56, 57).

A well documented characteristic of iron deposition throughout normal brain development and aging is its dynamic age-related increase (6). Adolescence is marked by a steep rate of increase whereas the iron concentration peaks in adulthood, reaching a plateau around the fourth decade. Our MFC measures were able to detect significant age-related differences between the adolescent and adult groups in the GP, PUT, CN, THL and FWM, as reported in C_PM post-mortem values. Within all regions, the MFC means in adolescents were significantly lower than in adults. Although both MFC and R2 measures generally followed the exponential growth curves defined by Hallgren and Sourander (6), R2 measures only showed significantly lower means in adolescents for the PUT and CN. In contrast, a study by Schenker et al. investigating age-dependent iron changes with R2 in GP and PUT was able to detect significant changes in both regions but for a much broader age range (51). Other MRI methods have also been able to detect such age-related changes in iron with varying sensitivity (32, 49, 58, 52).

In accord with previous R2 studies, we observed a strong correlation of R2 with C_PM when all regions were assessed together (3, 23, 26, 28, 29). This correlation was strengthened when interregional macromolecular mass fraction (f_M) differences were accounted for corroborating Mitsumori et al.’s previous observations (29, 50). Our R2 values are similar to those of another R2 study conducted at 3T (23) and, as expected from R2’s documented field dependence, higher than those performed at 1.5 T (26, 27, 59). MFC measures also showed a significant correlation with C_PM for the combined regions, but this was not improved by adding f_M as an independent variable. That the inclusion of f_M improves the correlation for R2 but not for MFC is consistent with the view that R2 reflects multiple tissue properties while MFC, as implied by Eq. [1], is a specific measure of MFIs. However, whereas MFC was significantly correlated to C_PM in each individual GM region investigated, R2 was only significantly correlated in the PUT and CN. The observation that MFC correlated with iron in a greater number of individual regions, as well as with all regions combined, may also be a consequence of MFC being a more specific measure of iron-induced MFIs.

Although R2 displayed a higher correlation with C_PM with combined regions (Table 1), it is known that R2 is also affected by other molecular relaxation mechanisms unrelated to iron which may enhance or diminish the correlation (31). Such a lack of specificity is possibly reflected in the loss of R2’s correlation significance within the GP and THL, both gray matter regions known to have high iron content (6). On the other hand, R2 correlated with C_PM in the CN and PUT to a higher degree than MFC. This variability could, perhaps, stem from the distinct microscopic tissue properties of each ROI that may also influence R2. Specifically, the CN and PUT have been shown to have very similar microstructure (small satellite/fusiform neurons with short axons) whereas both the GP and THL have an excessive number of pervading nerve fiber bundles (60).

House et al. have shown that for R2 to be dominated by iron induced MFIs, the sampled tissue must have an iron concentration of at least 5 mg Fe/100 g fresh weight (28). In contrast to GM, WM is known to have lower C_PM (< 5 mg Fe/100 g fresh weight), less water content and more protein content. Consequently, the correlation between R2 and brain iron is less strong. Similar to other studies, our R2 measures in the FWM failed to correlate significantly with C_PM (3, 26 – 28). This was also found in MFC measures of FWM. These results may reflect the small iron changes amongst dynamic myelin changes found in WM during this age range (6, 61). Table 2 shows that the expected FWM differences between adolescents and adults for our study is only about 0.7 mg Fe/100 g fresh weight, which is substantially smaller than the expected differences for the GM regions. Other MRI techniques have been suggested for iron measurement in WM. A previous study of FWM, temporal WM and occipital WM found that R2* was able to show a significant linear correlation with C_PM when R2 was not (3), while Li et al. demonstrated the ability of R2* to significantly distinguish heterogeneity between major WM fiber bundles due to differences in myelin content as well as iron distribution (62).

A few anomalous observations should be noted. For R2 measures of FWM plotted as a function of age, the adjusted growth curve from Hallgren and Sourander could not be meaningfully plotted because the R2 measures were not significantly correlated to C_PM and had a negative slope. Additionally, the growth curve fit for MFC measures of GP plotted with age had an impossible negative MFC y-intercept at age zero as a result of the linear regression of GP with C_PM. The linear regression equation had a significant negative y-intercept of −712 (p = 0.013) with a standard error of 266. While this may simply be a statistical artifact, it could also conceivably reflect a true nonlinearity in the dependence of MFC on the iron concentration. Such a nonlinearity is theoretically possible if the typical iron loading per iron rich cell within the GP varies with age. This is analogous to the quadratic dependence of MFC on contrast agent concentration observed in yeast phantoms (40). Altogether, the MFC regional measures in this study are similar to those previously reported with some variance due to methodology differences such as cohort age range, image resolution, sequence parameters (e.g, TE, t_s) and ROI delineation (38, 39).

A limitation of this study is the measurement of two time points for our R2 calculation versus five time shifts for MFC. Although the number of averages is the same, this may bias toward higher R2 standard deviations. Overall, reanalysis of MFC using two time shifts still resulted in significant correlations to C_PM in more individual GM regions than R2 and equaled R2 in detecting significant group differences within regions. This suggests that although acquiring MFC with more time points than R2 did reduce noise in the analysis, it did not account for all of MFC’s higher intra-regional specificity to iron-induced MFIs.

In conclusion, this is the first study to investigate MFC imaging for the in vivo detection of age-related changes in iron concentration within specific regions. In comparison to the commonly used MRI measurement of R2, MFC was able to detect significantly lower measures of iron content in adolescents versus adults in all regions investigated as predicted from C_PM estimates. Conversely, R2 detected age-related differences in only the PUT and CN. Furthermore, MFC measures showed significant linear correlations with C_PM in the GP, PUT, CN and THL versus only PUT and CN for R2. More individual regions displayed significant MFC correlation with C_PM than R2, even when only two MFC time shifts were used. In regards to the investigation of neuropathologies, comparing iron levels between patient and control groups within specific regions is of interest. MFC’s ability to detect expected age-related differences within all brain regions between adolescent and adult groups corroborates its sensitivity for also detecting brain iron differences due to pathology (38, 41). Due to the small sample size and cross-sectional design of this preliminary study, our results should be interpreted with caution. Nonetheless, these results do support the hypothesis that MFC is sensitive to brain iron in GM regions, and it appears to consistently give information about intra-regional changes in iron-induced MFIs beyond that provided by R2. Thus, in vivo measurement of brain iron may benefit from acquiring MFC in tandem with R2 or R2*. This is feasible given that whole brain MFC maps can be acquired within a few minutes (42, 44).

While postmortem work has established gold standard values for non-heme iron measures in various brain regions throughout the lifespan, improved MRI methods with significant iron specificity, such as MFC imaging, may help to advance the understanding of iron-related pathogenesis, disease progression and allow future treatment monitoring. Given these clinical implications, several MRI methods for measurement of brain iron that attempt to bypass the limitations of conventional relaxation rate measures are actively being investigated. As previously mentioned, these include FDRI, SWI, and the more recent QSM (5, 32 – 36). These various approaches are in effect quantifying different aspects of brain iron deposition. Future work will need to be done to directly compare MFC imaging with these other advanced methods to assess what congruence or divergence exist in the measures and to highlight potential synergies that may provide the most useful assessment of brain iron in vivo.