Human Brain Iron Mapping using Atlas-based T₂ Relaxometry

Abstract

Several in vivo quantitative magnetic resonance imaging (qMRI) techniques have been proposed as surrogate measures to map iron content in the human brain. The majority of in vivo qMRI iron mapping methods utilized the age-dependent iron content data based on postmortem data. In this work we fused atlas-based human brain volumetry obtained on a large cohort of healthy adults using FreeSurfer with T₂ relaxation time measurements. We provide a brain atlas-based T₂ relaxation time map which was subsequently used along with published postmortem iron content data to obtain a map of iron content in subcortical and cortical gray matter. We have also investigated the sensitivity of the linear model relating transverse relaxation rate with published iron content to the number of regions used. Our work highlights the challenges encountered upon using the simple model along with postmortem data to infer iron content in several brain regions where postmortem iron data are scant (e.g. corpus callosum, amygdale).

Introduction

Iron is present in several complex chemical forms in human brain and is shown to be important in brain tissue metabolism (1–3), but its abnormal accumulation has been implicated in several pathologies (3–6). To date, several quantitative magnetic resonance imaging methods (qMRI) methods were used (1–5) in combination with phantom or postmortem (5–11) measurements to provide in vivo surrogate markers of iron content in the human brain. In general, iron mapping methods can be grouped into three categories (3,4): (I) relaxation time (3,8–10,12–18), (II) magnetic field correlation (MFC) which uses asymmetric spin echo (19, 20), and (III) phase-based or susceptibility-weighted methods (3,10,21–24). Relaxation-based methods for iron mapping (3) use spin-lattice relaxation rate (R₁ = 1/T₁), or spin-spin relaxation rate (R₂ = 1/T₂) or gradient echo methods (R₂* = 1/T₂* = R₂ +R₂’), where T₂ and T₁ are the transverse and longitudinal relaxation times, respectively. An important relaxation rate based method is the field dependent R₂ increase method (FDRI) in which R₂ measurements are performed on the same subjects at two magnetic fields (B₀) (24–26).

Several investigations (see Table 5 in Ref. 3) have shown that R₁ (3, 13, 14), R₂ (3, 8, 9, 10, 12, 15, 16, 18), R₂* (3, 8, 12, 17) or R₂’ (R₂’ = R₂* - R₂) correlated strongly with brain tissue iron content using histological or postmortem brain iron measurements (6,8–11). The majority of iron mapping reports to-date used the caudate, putamen, and globus pallidus, as representative of extrapyramidal gray matter structures (3). Langkammer at al. (10) and Gelman et al. (15) showed that both R₂ and R₂’ correlated with iron content while Vymazal et al. (14) showed that R₁ and R₂ correlated with iron content. Hikita et al. (16) reported that both R2’ and R2 can be used for mapping iron in the basal ganglia, but since R₂’ is affected by macroscopic magnetic field inhomogeneity, R₂ was recommended for iron mapping. Several other works reported strong correlation between R₂ with iron (see Table 5 in Ref 3).

Phase sensitive or susceptibility-weighted imaging (SWI) methods have also been used to map iron (3,11,22–24). In two recent reports, phase sensitive methods did not correlate with iron content compared with R₂* (11) or FDRI methods (24).

We are not aware of a unified model (2) that can be used to explain the apparent paradox of iron correlation with R₁, R₂, R₂*, R₂’ and FDRI measurements as reported in the literature.

Compared to gradient-echo methods, the R₂ maps are less sensitive to macroscopic field inhomogeneities and are computationally less intensive compared to other methods (16). In this work we used spin-echo measurements with long repetition time to obtain regional R₂ values. This work has two major goals. First, we present the regional heterogeneity of T2 relaxation values on subcortical, deep and cortical gray matter in addition to deep and lobar white matter using atlas-based anatomical labels. Second, we attempt to correlate our R₂ measurements at 3.0 T with iron content using the postmortem age vs. iron content data reported in 1958 by Hallgren and Sourander (6). To verify the validity of the R₂ relaxation rate approach used in this report as a surrogate marker of iron, we also demonstrate the association between other in vivo qMRI methods such as MFC (20) and FDRI (26) and between published FDRI and R₂ measurements using in vivo brain atlas-based methods.

Materials and methods

Participants

This retrospective study has been approved by our institutional review board and is fully compliant with the Health Insurance Portability and Accountability Act (HIPAA). A written informed consent was obtained from each participant prior to data acquisition. All participants were neurologically normal based on the review of medical history. All scans were read as “normal” by a board certified radiologist. The participants included 86 right-handed healthy adults with 42 males (age range = 18.7–56.8 years; mean age ± SD = 35.4 ± 10.3 years) and 44 females (age range = 19.0–56.0; mean age ± SD = 37.2 ± 9.8 years).

MRI Data Acquisition

We acquired whole-brain conventional MRI (T1-weighted and dual spin-echo) data using a Philips 3.0 T Intera system with a dual quasar gradient system with a maximum gradient amplitude of 80 mT/m, maximum slew rate of 200 mT/m/ms and an eight-channel SENSE compatible head coil (Philips Medical Systems, Best, Netherlands).

T1-weighted data

T1-weighted data were acquired using 3D fast field-echo sequence (3D-FFE) with a field-of-view (FOV) of 240 × 240 mm² and isotropic voxel size of 0.9375 mm covering the entire brain. The flip angle, repetition time (TR), echo time (TE), and scan time were 6°, 10.3 ms, 2.0 ms and 322 sec, respectively.

T2-weighted data

T2-weighted data acquired using 2D dual turbo spin-echo (TSE) sequence with scan time ~ 367 seconds, TE₁/TE₂/TR=8/90/6800 ms in the axial plane (3mm slice thickness, square FOV = 240x240 mm² at 44 axial sections).

Processing

A pictorial flowchart of the computational procedures applied to the MRI data is illustrated in Fig. 1. The MRI data processing pipeline used in this work is described in more details elsewhere (27).

Figure 1.

A pictorial illustration of the MRI data processing pipeline used in this study. High resolution T1-weighted data were segmented and parcellated using FreeSurfer. The early and late DSE data were used to estimate T₂ relaxation time map which was aligned in each subject atlas-space. A listing of all brain regions is provided in the Supplement.

Segmentation and Parcellation of T1-weighted Data

The T1-weighted brain data were automatically segmented into cortical and sub-cortical regions using FreeSurfer software library (28; http://surfer.nmr.mgh.harvard.edu/). Using the cortical and sub-cortical segmentations provided by FreeSurfer (28, 29), an atlas consisting of about hundred and eighty white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) regions was generated for each subject in the T1-weighted native space (29). A tabulation of the brain anatomical labels (non-CSF) provided by FreeSurfer and used in this work is provided in the Supplementary Table. Due to its complex structure and poor white and gray matter contrast on T1-weighted data, the thalamus segmented by FreeSurfer contained both white and gray matter.

Computation of Relaxation Time from Dual Spin Echo Data

The T₂ relaxation times were estimated from the early and late echoes (TE₁, TE₂) volumes according to standard spin-echo procedures assuming a single compartment model (30,31):

$$S_i=k*\mathit{\text{PD}}*\left(1-\text{exp}\left(-\frac{\mathit{\text{TR}}}{T_1}\right)\right)*\text{exp}(-\frac{{\mathit{\text{TE}}}_i}{T_2})$$ [1]

where S_i is the signal intensity of the ith echo; the T₂ relaxation time is obtained as (30, 31):

$$T_2=\frac{{\mathit{\text{TE}}}_2-{\mathit{\text{TE}}}_1}{\text{ln}\left(\frac{S({\mathit{\text{TE}}}_1)}{S({\mathit{\text{TE}}}_2)}\right)}$$ [2]

Fusion of T2 Relaxation Time and T1-weighted Data

The T₂ relaxation time map for each subject was registered to the T1-weighted data using advanced normalization tools (ANTs) with symmetric normalization (32). The ANTs tool has also been independently tested and compared against 14 different registration programs (33). A detailed description of the methods used in the registration is presented previously (27). Since the T1-weighted volume and the brain atlas are in the same space, the atlas was used as a look-up table to locate chosen brain structures from all other image modalities registered to the T1-weighted space.

Brain iron concentration using Hallgren and Sourander Postmortem data

As described in the Introduction, several previous reports (3,12–18,22–26) have used the postmortem (6) best fit age vs. regional brain iron content data to estimate iron content using in vivo qMRI surrogates such as R₁ (13, 14), R2* (12, 15), and R₂ (15, 18) relaxation rates. In brief, regional iron content for each of the 12 brain regions (r) in Hallgren and Sourander (6) was estimated for each subject (n) as function of age using Equation [3]:

$$\mathit{\text{Iron}}(n,r)=A(r)*(1-\text{exp}(-B(r)*\mathit{\text{Age}}(n)))+C(r)$$ [3]

Statistical analysis

Age comparisons between males and females were conduced using independent t-test. Regional comparisons in R₂ and estimated iron content were evaluated using paired t-tests applied to compare qMRI data obtained from the entire cohort on several related regions (e.g. motor and sensory cortex). The t-test results on two variables as used in this work are equivalent to one way analysis-of-variance (F = t²). The correlations between regional R₂ and iron content using the Hallgren and Sourander postmortem data (6) or the published MFC and FDRI data were computed using the Pearson correlation coefficient (32). All statistical analyses were performed using MATLAB R12.1 Statistical Toolbox v 3.0 (The Mathworks, Natick, MA).

Results

The healthy men and women in our cohort did not differ in age (p=0.40, F=0.72). Given previous reports of insignificant gender-based differences in brain volumetry (35) and to simplify the analyses we group-averaged microstructural brain data obtained from age-matched men and women used in this study bilaterally.

Quantification and Visualization of T₂ Relaxation Time

Table 1 summarizes the group average volume and corresponding T₂ relaxation time of 28 white and gray matter regions that cover the entire human brain subcortical, cortical GM, deep and lobar white matter. Table 2 covers 12 additional cortical and lobar regions such as prefrontal, visual, motor and sensory cortices and corresponding white matter lobes. We included the prefrontal, motor and sensory regions as iron content data were provided in the Hallgren and Sourander post mortem study (6) and these functionally important cortical subdivisions could be looked up in our brain atlas. The qMRI average and standard deviation were obtained by volume-weighted averaging of the corresponding brain atlas labels or subvolumes tabulated in Supplemenray Table. Figure 2 shows a multiview of the T₂ relaxation time map (Fig. 2A) at a level selected for visualizing the basal ganglia, cerebral cortex, corpus callosum, amygdale and hippocampus.

Table 1.

Atlas-based bilateral volume in mL and corresponding volume-averaged T2 relaxation time of 28 gray and white matter brain regions that cover the entire brain (see Supplementary Table for a list and grouping of all brain regions).

#	Brain Structure	Volume (mL) Mean ± SD	T2 (ms) Mean ± SD
1	Thalamus Proper	14.80 ± 1.44	95.18 ± 3.87
2	Caudate	7.21 ± 1.07	111.51 ± 7.56
3	Putamen	11.34 ± 1.34	81.64 ± 3.18
4	Globus Pallidus	3.29 ± 0.43	66.89 ± 3.29
5	Hippocampus	8.77 ± 0.81	118.56 ± 4.95
6	Amygdala	3.32 ± 0.44	112.10 ± 3.50
7	Accumbens area	1.20 ± 0.22	102.73 ± 3.88
8	Frontal Cortex	165.08 ± 18.58	134.85 ± 8.06
9	Temporal Cortex	100.95 ± 11.22	113.71 ± 3.82
10	Parietal Cortex	108.58 ± 10.93	126.62 ± 8.49
11	Occipital Cortex	41.25 ± 5.28	107.42 ± 4.54
12	Cingulate Cortex	18.27 ± 2.64	115.22 ± 4.20
13	Insular Cortex (GM)	12.09 ± 1.42	121.54 ± 4.67
14	Cerebellar Cortex (GM)	99.23 ± 11.36	121.61 ± 5.23
15	Frontal Lobe WM	144.51 ± 17.96	95.83 ± 3.15
16	Temporal Lobe WM	63.67 ± 8.30	90.71 ± 2.43
17	Parietal Lobe WM	97.16 ± 11.41	95.38 ± 3.43
18	Occipital Lobe WM	42.71 ± 6.58	96.57 ± 3.57
19	Cingulate Lobe WM	23.37 ± 2.61	93.20 ± 2.40
20	Insular WM	17.41 ± 1.70	90.22 ± 1.71
21	Cerebellar WM	29.83 ± 3.85	99.46 ± 2.50
22	Brain Stem WM	20.80 ± 2.20	113.58 ± 4.65
23	Periventricular WM	69.15 ± 13.27	96.26 ± 2.00
24	CC Anterior WM	0.82 ± 0.14	94.35 ± 4.77
25	CC Mid Anterior	0.54 ± 0.16	112.74 ± 7.73
26	CC Central	0.51 ± 0.13	115.91 ± 6.49
27	CC Mid Posterior	0.49 ± 0.10	122.39 ± 8.44
28	CC Posterior	0.93 ± 0.15	96.86 ± 4.45

Table 2.

Atlas-based bilateral volume in mL and corresponding volume-averaged T₂ relaxation time of 12 cortical gray matter and corresponding lobar white matter subregions.

#	Brain Structure	Volume (mL) Mean ± SD	T2 (ms) Mean ± SD
1	GM caudal middle frontal	12.75 ± 2.40	143.75 ± 11.19
2	GM precentral @ Frontal (Primary Motor)	26.77 ± 3.07	135.56 ± 10.66
3	GM entorhinal Temporal	3.17 ± 0.57	117.34 ± 5.81
4	GM postcentral @ Parietal (Primary Sensory)	18.23 ± 2.48	138.55 ± 11.09
5	GM precuneus @ Parietal	17.80 ± 2.10	115.03 ± 6.54
6	GM cuneus @ Occipital (Primary Visual Cortex)	4.79 ± 0.93	110.70 ± 7.18
7	WM caudal middle frontal	12.89 ± 2.32	97.73 ± 3.81
8	WM precentral	27.59 ± 3.39	100.37 ± 4.00
9	WM entorhinal	1.45 ± 0.38	99.77 ± 5.22
10	WM postcentral	15.00 ± 2.27	100.05 ± 4.29
11	WM precuneus	18.84 ± 2.59	93.85 ± 3.00
12	WM cuneus	5.06 ± 1.07	93.65 ± 3.41

Figure 2.

(a) A visual representation of cohort means regional T₂ relaxation time (ms). Table 1 and Table 2 provide information about the average volume and T2 values. (b) Cohort average iron content map (µg [Fe]/g wet weight) estimated volume-wise using the age-projected regression of the postmortem measurements against our volume-based R₂ measurements (see Table 4). Key brain regions are shown in (c) and color coded according to the iron content values.

The T₂ relaxation time average values (Table 2; Table 3; Fig 2A) are smallest in the globus pallidus (T₂ =66.9 ± 3.3 ms) whereas the largest T₂ values are measured in the frontal gray matter (T₂ =134.9 ± 8.1 ms). The lobar white matter T₂ values are significantly smaller than the corresponding cortical gray matter (e.g. T₂(occipital cortex) > T₂(occipital white matter); p<1×10⁻²⁰; F=839). The average T₂ values in the occipital cortex are significantly smaller than the average values in the parietal cortex (p<1×10⁻²⁰; F=937). The average T₂ values in the motor precentral cortex of the frontal lobe (T₂ =135.6 ± 10.7 ms) are significantly smaller than the average values (T₂ =138.54 ± 11.1 ms) in the sensory postcentral cortex of the parietal lobe (p=1×10⁻⁷; F=34). The average T₂ values in the hippocampus are significantly greater than that in the amygdala (p<1×10⁻²⁰; F=187). The deep periventricular (non-callosal) white matter T₂ relaxation time values are significantly smaller than the average values in the caudate (p<1×10⁻²⁰; F =355). Note that the T₂ relaxation time is significantly smaller in the anterior (genu) CC compared with the posterior (splenium) CC (p=0.0002; F=15).

Table 3.

The parameters used for calculating regional brain iron concentration from ref. #6 using Equation [3]. The regional R₂ values are obtained from the volume-based T₂ measurements in Table 1 and Table 2. The estimated iron along with the regressed FDRI from ref #26 and the MFC from ref #20 are also listed for further analysis (see Table 4 and Figure 2).

#	Brain Region	A µg/g Wet Wt.	B Rate per year	C µg/g Wet Wt.	Extrapolated Iron Content µg/g Wet Weight. µ(σ)a	Current Study R2 = 1/T2 (sec −1)	FDRI µ (σ)b	MFC µ (σ)c
1	Globus Pallidus	210.41	0.09	37	206.12 (9.46)	14.984 (0.720)	4.359 (0.088)	699.7 (145.0)
2	Putamen	140.62	0.04	46	113.87 (14.16)	12.267 (0.476)	2.363 (0.227)	387.0 (106.4)
3	Caudate	90.66	0.05	33	82.21 (8.38)	9.009 (0.615)	2.068 (0.146)	NA
4	Frontal Cortex (Motor)	40.79	0.05	40	3.13 (4.15)	7.422 (0.579)	NA	NA
5	Prefrontal cortex (caudal middle)	20.43	0.07	58	27.71 (1.55)	6.999 (0.552)	NA	NA
6	Temporal Cortex (superior)	20.70	0.07	55	29.84 (1.72)	8.076 (0.462)	NA	NA
7	Parietal Cortex (Sensory)	30.97	0.07	49	40.69 (2.53)	7.264 (0.581)	NA	NA
8	Parietal Cortex (precuneus)	30.31	0.06	60	34.67 (2.49)	8.721 (0.491)	NA	NA
9	Occipital cortex (cuneus or visual)	40.03	0.06	72	42.11 (3.03)	9.070 (0.562)	NA	NA
10	Cerebellar Cortex	20.70	0.09	68	32.10 (1.31)	8.238 (0.355)	NA	NA
11	Thalamus	NA	NA	NA	~60 (Fig. 8. Ref 6)	10.523 (0.415)	1.265 (0.091)	189.3 (71.6)
12	Frontal WM (Superior)	30.95	0.1	31.0	40.98 (1.44)	10.129 (0.405)	1.688 (0.116)	91.5 (41.9)
13	Genu CC	NA	NA	NA	NA	10.626 (0.535)	1.239 (0.066)	174.1(45.5)
14	Splenium CC	NA	NA	NA	NA	10.345 (0.460)	0.970 (0.06	103.0 (34.4)
15	Hippocampus	NA	NA	NA	NA	8.449 (0.347)	0.689 (0.029)	NA

^(a)The A, B and C values were obtained from the best fit postmortem iron content or [Fe] vs. age data in Hallgren and Sourander paper (Ref 6; N=81; age = 0–100 years; Ref 6). Mean and standard deviation values of age-expected iron content for all our subjects (n=1 to N=86). Iron content was estimated for each subject n and region r as Iron (n, r) = A(n) * [1-exp (−B(r)*Age(n))] + C(r). The average iron content in the thalamus (TH) was estimated from Fig. 8 in Ref. 6.

^(b)Computed mean and standard deviation for each subject in our cohort based on the age-regressed FDRI data on both men and women in Bartzokis et al. (Ref 24).

^(c)Based on the tabulated magnetic field correlation (MFC) data on 14 healthy controls (N=14; mean age = 39 & range 23–55 years) in Ge et al. (Ref 20).

Estimation of Iron Content FDRI and MFC in our cohort

The A, B and C parameters in Equation [3] were obtained from (6) to estimate the iron content in 12 brain regions for each subject in our cohort. The iron content mean and standard deviation on our cohort is provided in Table 3 along with the corresponding R₂ values from Table 2 and Table 3. To compare the methods (see section below), Table 3 includes the estimated average and standard deviation FDRI values on our cohort. The FDRI values were estimated using the ROI-based age vs. FDRI best linear fit parameters tabulated in Bartzokis et al. (26) on 8 regions that included the hippocampus, genu and splenium of the corpus callosum. Note that the postmortem iron content data in Hallgren and Sourander did not include these three regions. Table 3 also includes the average normative MFC values reported by Ge et al. (20) on six brain regions that did not include the caudate.

Correlation between our R₂ and Iron Measurements

Using the tabulated parameters in Table 3, we conducted a linear regression analysis of the volume-averaged and atlas-based R₂ measurements vs. the iron content on some selected groups of regions listed in Table 1 and Table 2 (regression model: R₂ = α + β*[Iron]). Table 4 provides the intercept (α), slope (β) and the corresponding Pearson correlation coefficient (r) and statistical significance (p) of several possible comparisons of the data in Table 3. Using the 12 regions including thalamus and frontal white matter the correlation coefficient between R₂ and iron was significant (r =0.906; p=0.00005).

Table 4.

Pearson correlation coefficient (r) and statistical significance (p) of R₂ vs. estimated iron (see Table 1) and the correspondence between FDRI, MFC, R₂ and estimated iron. The r and p values are listed for several combinations of brain regions.

Linear Fit Model	Number of brain regions (I, II, III)	Intercept α (S.D)	Slope β (S.D)	Pearson Correlation r	Correlation Significance p
Iron vs. R2a	12 included (eS+Crtx+Fwm+TH)	6.835 (0.480)	0.041 (0.006)	0.906	4.8×10−5
Iron vs. R2a	10 excluded (TH, Fwm)	6.444 (0.412)	0.042 (0.005)	0.951	2.4×10−5
Iron vs. R2a	9 excluded (CN; Fwm, TH)	6.507 (0.405)	0.043 (0.005)	0.959	4.3×10−5
Iron vs. R2a	8 excluded (CN, cuneus; Fwm, TH)	6.371 (0.427)	0.044 (0.005)	0.965	1.0×10−4
Iron vs. R2a	7 excluded (CN, cuneus, precuneus; Fwm, TH)	6.172 (0.447)	0.045 (0.005)	0.973	2.3×10−4
Iron vs. R2a	7 excluded (eS; Fwm, TH)	7.226 (1.983)	0.021 (0.055)	0.168	0.720
FDRI vs. R2b	8 included (eS, HC; Fwm, TH, gCC, sCC)	8.014 (0.766)	1.517 (0.360)	0.865	0.006
FDRI vs. R2b	7 excluded (CN)	8.209 (0.449)	1.579 (0.210)	0.959	6.6×10−4
FDRI vs. R2b	5 excluded (CN, TH, Fwm)	8.289 (0.529)	1.583 (0.225)	0.971	0.006
R2vs. MFCc	6 included (GP, PT; Fwm, TH, gCC, sCC)	9.288 (0.137)	0.008 (0.000)	0.995	3.5×10−5
Iron vs.FDRId	5 included (DGM; Fwm, TH)	0.552 (0.31)	0.018 (0.003)	0.968	0.007
Iron vs. MFCe	4 included (GP, PT; Fwm, TH)	−39.178 (17.824)	3.621 (0.145)	0.998	0.002
FDRI vs. MFCf	5 included (GP, PT; Fwm, TH, gCC, sCC)	−39.856 (17.284)	171.869 (7.219)	0.997	1.6×10−4

^IAbbreviations: CN = Caudate Nuclei; Fwm= Frontal white matter; gCC, sCC = genu and splenium of the corpsus callosum; GP = Globus Pallidus; HC = Hippocampus; PT = Putamen; TH = Thalamus Proper; eS = extended striatum = GP + PT + CN.

^IIIron refers to the regional brain [Fe] concentration of our cohort estimated from Hallgren and Sourander data which did not include splenium CC, genu CC, or Hippocampus (see Table 3). The R₂, FDRI and MFC average values of the selected brain regions are provided in Table 3.

^IIIThe linear regression models tested are (see Methods):

^(a)R₂ (sec⁻¹) = α(sec⁻¹) + β (sec⁻¹/(µg Fe/g ww))* [Iron Content]

^(b)R₂ = α₁β₁ * FDRI

^(c)R₂ = α₂ + β₂ * MFC

^(d)FDRI = α₃ + β₃ * [Fe];

^(e)MFC = α₄ + β₄ *[Fe](Iron content units are µg [Fe]/g Wet Weight)

^(f)MFC = α₅ + β₅ *FDRI

Consistent with several previous reports (12,14,15), the thalamus (12) and frontal white matter measurements clearly deviated from the best fit upon visual inspection of the data. After removing the thalamus and frontal white matter measurements (Table 4), the correlation between iron content and R₂ became stronger (R₂ = (6.444 ± 0.412) + (0.042 ± 0.005)* [Iron]); r=0.951, p=0.00002). To test the sensitivity of these results due to the number of regions included, Table 4 provides some results using 5, 7, 8 and 9 regions. Note that excluding the three brain regions with the highest iron content (i.e. globus, putamen and caudate) resulted in statistically insignificant correlations (r=0.168, p=0.72).

Visual Representation of Extrapolated Brain Iron Content

Using the best linear fit parameters obtained on the first 10 regions excluding frontal white matter and thalamus (Table 4), we inferred the regional iron content using the measured R₂ values in all gray matter structures assuming the linear relationship [Iron] = (−131.40 ± 23.42) + (21.36 ± 2.46) * R₂. The group-averaged brain iron content map is presented in Fig. 2b. The iron concentration is shown in units of part-per-million (ppm) or (µg [Fe]/g wet weight) allowing the presentation of the scalar and positive iron content output data in bytes (range 0 – 255). Figure 2b shows the iron content values of several gray matter regions that were not included in Hallgren and Sourander (6). For example, our data shows that the amygdale contains higher iron concentration (60.07 ± 5.92 µg [Fe]/g wet weight) than the hippocampus (49.83 ±7.39 µg [Fe]/g wet weight) (p<1×10⁻²⁰; F = 203).

Comparison between FDRI, MFC, R₂ and estimated Iron Content

Table 4 provides the linear regression analyses results (intercept, slope, standard deviations, correlation coefficient and significance) of the estimated iron content, estimated FDRI, published control MFC and our R₂ measurements as summarized in Table 3. The correlation between FDRI and R₂ measurements was significant (r = 0.865; p= 0.006) using the 8 structures including thalamus, frontal WM. The FDRI vs. R₂ correlation was stronger upon excluding our caudate measurements (r=0.959; p=0.00066). Our R₂ measurements also correlated strongly with the normative MFC values (r=0.995; p=0.000035). The estimated FDRI and published normative MFC measurements correlated strongly (r=0.997; p=0.00016; see Table 4).

Discussion

To the best of our knowledge, this is the first human brain atlas-based report of age-extrapolated iron content of the cortical gray matter. We presented the spatial heterogeneity of water T₂ relaxation time on a large cohort of healthy adults. We adopted FreeSurfer for atlas-based volume measurements (28, 35). Our methods adapted validated procedures to fuse and register microstructral MRI metrics such as T₂ relaxation time with high resolution T1-weighted atlases in each subject native space (27). We presented subcortical, cortical gray matter, deep and lobar white matter to illustrate spatial and interregional heterogeneity. To verify our R₂-based results, we have also contrasted our approach with two alternative qMRI methods that were used to provide markers of brain iron such as FDRI and MFC. We found strong correspondence between published MFC and FDRI methods on one hand, and a strong association between the current R₂ approach and the FDRI approach, on the other hand. We tabulated the volume of each brain region used as our quantitative measurements are averaged over the entire brain region volume (12,30,31). This is important when comparing our findings with previous reports that adopted ROI (14–16,18), or voxel-based approaches (12).

The relative average values of the normative adult brain volumes in our work (e.g. frontal cortex > temporal > occipital > putamen > caudate > hippocampus> amygdala) are consistent with several volumetric reports using manual delineation or automated methods (12,30,35). We have used dual fast spin echo and adopted a single compartment model to estimate the T₂ relaxation time in both gray and white matter. The primary contribution of this work pertains to the heterogeneity of T₂ relaxation time measurements in gray matter and their potential as a marker for iron content using atlas-based methods.

In accordance with previous reports (30,31,36–38), we found significant region-dependent variations in transverse T₂ relaxation (T₂: globus pallidus < putamen < white matter < cortex). Consistent with a recent atlas-based study on healthy adults (30), the average T₂ relaxation time of the occipital cortical gray matter was significantly smaller than the temporal, frontal and parietal cortices (T₂: occipital < temporal < parietal < frontal). Our finding of significantly smaller average T₂ relaxation time in the precentral motor cortex compared to the postcentral sensory cortex is concordant with increased iron in the motor cortex compared with the sensory cortex as reported by Hallgren and Sourander (6). The finding of relatively low iron content in the hippocampus compared to the subcortical gray matter regions is consistent with previous MRI (8,26) and postmortem data (3,7).

Consistent with previous reports on the regional heterogeneity of the corpus callosum using T₂ relaxation time measurements (36,39,40), our data indicate that T₂ relaxation time values of the anterior CC are significantly smaller than that of the posterior CC (Fig. 2a). This inter-regional T₂ relaxation trend on the CC has also been reported using T₁ relaxation times (e.g. T₁(gCC) < T₁(sCC); see 36). The observation that T₂(gCC) < T₂(sCC) could not be explained exclusively by myelin since the anterior CC has been shown in human adult postmortem samples to be populated by slow-conducting, thin and lightly myelinated axons (41). In addition, the myelin water content of the genu has been shown to be significantly smaller than in the splenium (36).

This paradoxical result on the CC indicates that other biophysical factors in addition to myelin may contribute to the measured T2 relaxation time in compact white matter with reduced myelin. In our simple single compartment model we ignored the possible contribution of fast exchange between free water molecules in the genu with a potentially short-lived macomolecular pool (42–44). Therefore, exchange of water with this “solid-like” compartment would reduce the effective T₂ relaxation time measured using the two echoes (42).

A relatively larger iron content in the frontal CC than the posterior CC would offer another plausible explanation to this finding on the corpus callosum. To the best of our knowledge, there has been no comprehensive histopathological report (11) on the iron content of the different and functionally specialized subdivisions of the human CC. Scant postmortem data available indicate that the iron content of the CC is low (8–10). Higher iron content in the genu of the CC compared to the splenium CC may be inferred from the FDRI and MFC data (see Table 3; Table 4).

The iron content contribution to white matter relaxation rate has been debated (17, 18). A recent study by Li et al. (17) did not report strong association between iron content and relaxation rate in compact white matter tracts.

In this work, we presented the mean age and volume-averaged iron map of deep and cortical gray matter excluding the thalamus proper and frontal white matter. We noted along with previous reports (12,14,15,18,19) that white matter R₂ measurements deviated from the iron vs. R₂ inter-regional best least-squares fit. We applied a linear model relating R₂ and iron content to gray matter measurements and used as many regions we could match using our atlas labels with published data (6). In a recent report, Mitsumori et al. (18) adapted a model proposed by Gelman et al. (45) that incorporated the macromolecular fraction (e.g. 1 – free water fraction) to account for deviant R₂ measurements in white matter (e.g. R₂ = α + β * iron + γ * (1 – water fraction)). In a future extension of this work, other regional microstructural qMRI metrics such as the diffusivities obtained using diffusion tensor imaging will be examined in addition to the water content.

This retrospective study has several limitations. The R₂ relaxation rate approach used in our work as an in vivo surrogate marker of iron depended on postmortem measurements from an important study published in 1958 which did not include detailed information about gender, handedness, side and the exact location of the ROIs used. Brain structures such as hippocampus, amygdale and the corpus callosum subdivisions were not included (6). As noted in the Results (see Table 4), the approach adopted in this work to estimate iron was not sensitive upon excluding brain structures with the highest iron content. However, to the best of our knowledge, ours is the first to use 10–12 regions to obtain the best linear regression fit parameters (Table 4); almost all previous iron mapping studies used 3–9 regions (3,11,12) that included identifiable regions with high iron content such as putamen and globus palllidus.

The application of iron mapping methods may be challenged if water distribution and tissue microsture are altered and produce counter effects that reduce the relaxation rate sensitivity to iron (46). For example, natural aging may increase water content in the extracellular space and neuronal cells would decrease in size while iron may also increase in gray matter (5,31,46). If demyelination of white matter occurs due to aging (31), or pathology (47,48), T₂ may increase, but commensurate iron deposition due to pathology (6) would decrease T₂. In these scenarios, the measured relaxation rate may not be a direct and specific marker of iron (3,46). The presence of blood deposits (e.g. traumatic injury) gliosis (48,49), and microcalcifications in the volume-of-interest may also confound the measured relaxation rate (1,5).

The single compartment model adopted in this work could not separate the relative contributions from the intracellular and extracellular compartments in gray matter. The application of the single compartment model to white matter can not separate the intra-axonal, myelin water, extracellular and the semisolid macromolecular pool (50–54). To improve the specificity of relaxation measurements the acquisition of additional MRI contrasts (56–58) at different echo times and using multicompartment models would allow objective analysis of the sensitivity and specificity of different iron mapping methods.

Additional histopathological and in situ standardized quantitative MRI measurements of key human brain structures such as amgydala, hippocampus, visual, motor and sensory cortices, anterior and posterior CC are needed to help isolate or model the in vivo contributors to MRI microstructural metrics (2,3). In conclusion, our results underscore the utility of using standardized multimodal and atlas-based approaches and the need for sophisticated models to understand the yet unresolved biophysical contributors to MRI signal.