Cross-sectional and Longitudinal Assessment of Brain Iron Level in Alzheimer Disease Using 3-T MRI

Introduction

Iron is a prevalent element in the human body and an indispensable cofactor for major biochemical processes such as oxidative phosphorylation, oxygen transport, and neurotransmitter and DNA synthesis (1). Iron in neuronal tissue is mainly stored in the form of ferritin or hemosiderin and is essential for normal brain function (2). Abnormal iron accumulation has been reported in numerous neurodegenerative disorders (3). However, it is unclear if increased iron deposition contributes to the pathogenesis of these diseases or represents an epiphenomenon. Higher regional brain iron concentrations in patients with Alzheimer disease (AD) were associated with β-amyloid plaques (4) and neurofibrillary tangles (5). Iron accumulation was found to increase the production of amyloid precursor protein and thus might be directly involved in the formation of β-amyloid plaques (6,7).

Neuroimaging techniques allow in vivo detection of paramagnetic brain iron (2,8,9), and, among other approaches, R2* relaxation rate mapping has been validated to quantitatively assess iron deposition in brain tissue (10). The majority of AD studies have investigated iron in deep gray matter and have found higher levels in basal ganglia (BG) in patients with AD compared with healthy control (HC) participants (2,8,11–13). Only a few studies have evaluated iron deposition in AD outside the BG (8,13,14), which would be potentially clinically relevant but remains challenging to assess with gradient-echo MRI because of the proximity and effects of discontinuities of the magnetic susceptibility (15). Zhu and colleagues (8) found higher iron concentrations in the hippocampi, parietal cortex, BG, and dentate nuclei in 15 patients with AD. More recently, higher iron levels were described in 29 patients with AD in the frontal and temporoparietal lobes (14) and in 19 patients with AD in the left pre- and postcentral gyrus, hippocampi, amygdala, BG, cingulate cortex, entorhinal cortex, allocortex, and neocortex (13). However, all these studies used a cross-sectional design. To our knowledge, no published study has longitudinally examined the relationship between iron changes in the neocortex and cognitive decline in patients with AD.

Thus, in this study, we assessed brain iron level in a large clinical cohort of participants with AD by using R2* relaxation rate mapping with an advanced correction method that also allowed for R2* calculation in neocortical regions. We compared global and regional iron differences between patients with AD and age-matched HC participants, evaluated longitudinal R2* changes in participants with AD during a 17-month follow-up period, and determined the association of cross-sectional and longitudinal iron data with cognitive decline.

Materials and Methods

Study Participants

The trial protocol for this prospective study was registered at the National Library of Medicine (NCT02752750). Data generated or analyzed during the study are available from the corresponding author, on request. This prospective study was approved by the local medical ethics committee, and signed written informed consent was obtained from all study participants or their caregivers. Study participants were consecutively recruited from an ongoing longitudinal multicenter cohort study, the Prospective Dementia Registry–Austria of the Austrian Alzheimer Society, between 2010 and 2016 (16,17), including 819 participants diagnosed with dementia according to the criteria in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (18), without need for 24-hour care and with a caregiver who could provide information about their condition. In a prior study (17), we included 64 participants who were also included in our current study. This prior report evaluated the spatial distribution of white matter hyperintensities (WMHs), whereas in the present study we evaluated brain iron level by using R2* mapping. Inclusion criteria for all participants were a diagnosis of probable or possible AD according to the criteria of the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (19) and a complete MRI protocol including a three-dimensional multiecho gradient-echo sequence. Patients with insufficient MRI scan quality and outliers were excluded from the study (Fig 1). All study participants underwent a Mini-Mental State Examination (MMSE) (20). Individuals with AD also underwent testing with the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) battery (21) within 3 months of the MRI examination. CERAD total scores were calculated according to the method derived by Chandler et al (22) characterizing global cognitive performance. We also included the CERAD subtests for verbal memory (word list learning and word list delayed recall) and constructional praxis (figure drawing). The majority of participants with AD who were included at baseline (56 of 100) underwent a subsequent neuropsychological and neuroimaging assessment with an MRI protocol identical to the baseline protocol and a mean follow-up time and standard deviation of 1.39 years ± 0.55 (mean follow-up time, 17 months).

Figure 1:

Download as PowerPoint

The HC participants were selected from participants of the prospective Austrian Stroke Prevention Family Study in community-dwelling individuals with an MRI scan between 2010 and 2014 and were individually matched for age (within 3 years; 40 men, 60 women). This study was performed among the healthy elderly population of Graz, Austria. These volunteers were randomly selected from the community register, had a normal neurologic status, and were without cerebrovascular attacks and dementia (23).

Cardiovascular risk factors were recorded in both groups and were defined based on history and taken medications.

MRI acquisition.—MRI scans were obtained at 3 T (Magnetom Trio Tim; Siemens Healthcare, Erlangen, Germany) by using a 12-channel phased-array head coil. MRI included a structural T1-weighted magnetization-prepared rapid acquisition gradient-echo sequence with 1-mm isotropic resolution (repetition time msec/echo time msec/inversion time msec, 1900/2.19/900; flip angle, 9°) and a T2 fluid-attenuated inversion recovery sequence (10 000/69/2500; flip angle, 160°; resolution, 0.9 × 0.9 × 3 mm).

R2* relaxation data were acquired with a spoiled three-dimensional gradient-echo sequence (repetition time msec/echo time msec, 35/4.92; flip angle, 15°; six equally spaced echoes; interecho spacing, 4.92 msec; spatial resolution, 1 × 1 × 2 mm; 64 slices; field of view, 256 mm; bipolar readout; parallel imaging using generalized autocalibrating partial parallel acquisition, 2; acquisition time, 4 minutes 51 seconds).

MRI Analysis

R2* mapping.—R2* relaxation rate was estimated by fitting to a monoexponential model weighted by a sinc function to account for macroscopic B0 field inhomogeneities using Matlab software (version 9.1.0; MathWorks, Natick, Mass) (15,24). R2* maps of participants older than the main age range are shown in Figure 2.

Figure 2:

Download as PowerPoint

WMH segmentation and regional R2* analysis.—WMH masks were drawn semiautomatically, and the WMH volume was calculated as previously described and was normalized according to total intracranial volume (17).

Automated segmentation of cortical and deep gray matter structures based on the T1-weighted magnetization-prepared rapid acquisition gradient-echo scan was done with FreeSurfer software (version 5.3; https://surfer.nmr.mgh.harvard.edu) (25), and the processing steps are shown in Figure 3.

Figure 3:

Download as PowerPoint

The processing included the segmentation of the subcortical deep gray matter structures and parcellation of the cerebral cortex into regions, based on gyral and sulcal regions (Fig 4). The segmentations were frontal, parietal, temporal, and occipital lobes; caudate nucleus, globus pallidus, and putamen (summarized as global BG); and hippocampus and thalamus. Regional masks were visually checked by using an indicating data quality rating scale (from 0 for strong artifacts to 3 for no artifacts) for each MRI sequence, and MRI scans with a quality rating of 0 or 1 were excluded from the study and then were affinely registered to the R2* maps by using FSL-flirt with six degrees of freedom and trilinear interpolation (FSL, version 6; https://fsl.fmrib.ox.ac.uk). Finally, the transformed masks were binarized and, to prevent partial volume effects, were eroded with a restrictive threshold level according to the 75th percentile of the intensity range. To reduce the possible effect of voxels affected by susceptibility artifacts, median (instead of mean) R2* values were calculated for all regions of interest and were then used for further statistical analysis. In addition, all regional volumes and intracranial volume were determined with the FreeSurfer software.

Figure 4:

Download as PowerPoint

Statistical Analysis

Statistical analysis was performed with software (SPSS Statistics for Windows, version 25; IBM, Armonk, NY). Evaluation of normal distribution of data was performed with the Shapiro-Wilk test. WMH volume had a skewed distribution; therefore, the volumes were natural log transformed. Student t test and Mann-Whitney U test for continuous variables and Pearson χ² test for nominal variables were applied. We applied the one-sample Wilcoxon signed-rank test to check whether the median annualized percentage rate was different from zero. To study the differences in the R2* levels between groups, we first applied the Mann-Whitney U test, then a multivariable linear regression analysis, adjusted for WMH, regional brain and intracranial volume, sex, and smoking status. Smoking was defined as a possible cofounder because it was the only risk factor that differed between the AD and HC groups (Table 1). The association between annualized R2* changes and changes in cognition scores was evaluated by using Pearson correlation and multivariable linear regression models adjusted for age, sex, and annualized changes in WMH and intracranial and regional volumes. In addition, a linear mixed-effect model was calculated with the lmerTest package (25) in R Core Team (2020) (R: a language and environment for statistical computing; R Foundation for Statistical Computing, Vienna, Austria). Outliers, defined as larger than the mean ± 3 standard deviations, were excluded. P < .05 was considered to indicate a significant difference, and P values were corrected for multiple comparisons with the false discovery rate (26).

Results

Participant Characteristics

Our final cohort consisted of 100 participants with AD (80 probable AD, 20 possible AD) with a mean age and standard deviation of 73 years ± 9 (age range, 51–90 years). There were 42 men (mean age, 72 years ± 9; age range, 55–90 years) and 58 women (mean age, 73 years ± 9; age range, 51–86 years) (Table 1). Of the 100 participants with AD who were included at baseline, 56 (mean age, 71 years ± 10; age range, 51–87 years; 25 men, 31 women) underwent an identical subsequent assessment after mean follow-up of 17 months (median baseline MMSE score, 22 [interquartile range, 18.5–25]; median follow-up MMSE score, 18 [interquartile range, 13.5–22]; median baseline CERAD total score, 43 [interquartile range, 34–53]; median follow-up CERAD total score, 35 [interquartile range, 21–48]; median baseline natural log-transformed WMH, 3.85 mm³ [interquartile range, 3.33–4.36 mm³]; median follow-up natural log-transformed WMH, 3.92 mm³ [interquartile range, 3.54–4.46 mm³]). Drop-out analysis revealed no significant differences between baseline characteristics of 44 patients with AD who dropped out and 56 patients with AD who remained in the study (mean age, 74.5 years vs 71.3 years; P = .15; median MMSE score, 21.8 vs 21.5; P = .80, respectively).

Comparison of R2* Values between HC and AD Groups

Table 2 shows the comparison of R2* values in total and regional gray matter between AD and HC groups. After correction for sex, smoking, WMH, and intracranial and regional volumes, median R2* levels remained higher in the whole cortex including the total neocortex (HC group, 17.0 sec⁻¹; AD group, 17.4 sec⁻¹; P < .001) and deep gray matter (HC group, 29.0 sec⁻¹; AD group, 30.2 sec⁻¹; P = .01) in patients with AD as compared with HC participants. Regionally, median R2* levels in the temporal (HC group, 16.4 sec⁻¹; AD group, 18.1 sec⁻¹; P < .001) and occipital (HC group, 19.6 sec⁻¹; AD group, 20.2 sec⁻¹; P = .007) lobes and in the caudate nuclei (HC group, 22.9 sec⁻¹; AD group, 24.1 sec⁻¹; P = .007) were higher in patients with AD.

Longitudinal Changes of Regional Volumes and R2*

Table 3 summarizes the global and regional annualized percentage rates of R2* and normalized brain volume.

Regional brain volumes decreased over time in total cortex (annualized rate = −3.1%; P < .001), global BG (annualized rate = −1.5; P ≤ .001), frontal lobe (annualized rate = −2.8%; P < .001), temporal lobe (annualized rate = −4.9%; P < .001), parietal lobe (annualized rate = −2.9%; P < .001), occipital lobe (annualized rate = −1.5%; P < .001), caudate nuclei (annualized rate = −2.4% P < .001), putamen (annualized rate = −4.6%; P < .001), and hippocampus (annualized rate = −4.1%; P ≤ .001).

R2* levels increased significantly over time in global BG (annualized rate = 1.2%; P = .002), putamen (annualized rate = 1.1%; P = .01), and caudate nuclei (annualized rate = 1.6%; P = .02).

Association of R2* and Cognitive Deterioration in AD

Although R2* at baseline was not associated with MMSE score in the AD group, R2* changes in the occipital lobe were associated with changes in MMSE score during 17 months (after adjustment for changes in WMH, intracranial and brain volume, and age and sex [β = −1.45 score per sec⁻¹, P = .04]). However, after correction for false discovery rate, these differences were no longer significant.

Longitudinal changes in R2* in the temporal lobe over the observation period were associated with changes in CERAD total score before (r = −0.36; P = .01) and after adjustment for changes in WMH, intracranial and regional brain volume, age and sex, and false discovery rate correction (β = −3.23 score/sec⁻¹, P = .003) (Table 4). The association between R2* and CERAD total score in the temporal lobe was confirmed by using a mixed-effect model (β = −1.78 score/sec⁻¹ ± 0.82 [standard error]; P = .03) (Table E1 [online]).

Additional regression analysis of R2* changes in cortical lobes and CERAD subtests for verbal memory and constructional praxis showed that longitudinal changes in R2* in the temporal lobe were associated with changes in constructional praxis after adjustment for changes in WMH, intracranial and regional brain volume, and age and sex (β = −0.96 score/sec⁻¹; P = .04) and verbal memory (β = −0.86 score/sec⁻¹, P = .05). However, after correction for false discovery rate, these differences were no longer significant.

Discussion

We assessed neocortical and subcortical brain iron levels in a clinical cohort of patients with Alzheimer disease (AD) by means of R2* relaxation rate mapping. The use of an advanced correction method also allowed the calculation of R2* rates in neocortical regions, which are highly relevant clinically but are challenging to assess from the MRI perspective. Our study improved on previous investigations in terms of sample size and by exploring whether R2* at baseline related to longitudinal cognitive decline. We used 3-T R2* relaxometry with a correction technique for intravoxel dephasing and found higher R2* levels in the global basal ganglia (BG) (healthy control [HC] participants, 29.0 sec⁻¹; patients with AD, 30.2 sec⁻¹; P = .01) and total neocortex (HC participants, 17.0 sec⁻¹; patients with AD, 17.4 sec⁻¹; P < .001) in participants with AD compared with age-matched community-dwelling individuals without overt cognitive impairment. The regional differences in iron load between patients with AD and HC participants were most pronounced in the occipital (HC participants, 19.6 sec⁻¹; patients with AD, 20.2 sec⁻¹; P = .007) and temporal (HC participants, 16.4 sec⁻¹; patients with AD, 18.0 sec⁻¹; P < .001) lobes. It is interesting to note that in the longitudinal analysis, only the BG showed a significant increase in iron over time, but increased iron levels in the temporal lobe were associated with longitudinal changes in Consortium to Establish a Registry for Alzheimer’s Disease total score (β = −3.23 score/sec⁻¹, P = .003) independently of longitudinal changes in white matter hyperintensity and brain volume.

Previous MRI studies have also described higher iron levels in the BG (2,8,11–13), but only a few studies have investigated R2* in neocortical regions in patients with AD (8,13,14). Their results are in line with those of our study. In addition, a postmortem study observed higher iron concentrations in the temporal lobes of patients with AD, which also were strongly associated with the rate of cognitive decline (27).

To date, cross-sectional MRI studies examining the relationship between iron overload and cognition have had conflicting results. Some investigators have described higher iron levels in the hippocampus, caudate nucleus, and neocortex to be related to cognitive deterioration (2,9,28), but others did not find such associations (12).

We can only speculate on the cause of the relationship between R2* levels in the temporal lobe and cognition in our study. The temporal lobe is strongly affected by β-amyloid deposition (29), and iron is particularly present in β-amyloid plaques (30). Interactive effects between β-amyloid and iron in amyloid-rich brain areas, such as the temporal lobe, may be responsible for our finding, with already small increases of iron in these regions having substantial effects on the cognition. In an in vitro AD model, Teller and co-workers (31) observed a prominent degradation in spontaneous activity patterns, when β-amyloid and magnetite (iron oxide) acted together. In this context, in mammalian cell cultures, iron interfered with the aggregation of β-amyloid, delaying the formation of well-ordered aggregates, thus specifically enhancing the toxicity of β-amyloid (32). This evidence, together with the fact that iron is attached to β-amyloid, might underlie the relationship between R2* levels and cognition, at least to a certain extent.

Because iron accumulation in the deep gray matter is associated with healthy aging (33), increased iron levels in our study might also partially reflect the effects of age. The association between iron load in the temporal lobes and cognitive decline in the patients in our study was seen even though the temporal lobe R2* change was not significant. One explanation might be that even small insignificant increases in iron deposition in this region are sufficient to result in cognitive deterioration. Another explanation might be that iron deposition is not responsible per se for the cognitive decline but represents only the epiphenomenon of neurodegenerative processes. In this case, even small insignificant increases in iron deposition might be indicators of more rapid neurodegeneration.

We acknowledge the short follow-up interval and the lack of longitudinal data from HC participants as limitations of our study. We did not use cerebrospinal fluid biomarkers or amyloid/τ PET for the AD diagnosis, as our study was designed in 2008 when the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association criteria (19) were standard. It is noteworthy, however, that although new criteria that have been introduced during the past decade may increase diagnostic accuracy, the older criteria had a sensitivity of 81% and specificity of 70% across more than a dozen clinical-pathologic studies (34). Although iron has been shown to be the overwhelming determinant of R2* in gray matter, we cannot entirely exclude the possibility that confounding factors, such as myelin content, also contributed to the observed variations of R2* (35). However, storage iron cannot be removed from the brain and the myelination will not increase at this age, which render increased iron deposition highly probable, as the substrate of R2* increases in our study.

As a future perspective, quantitative susceptibility mapping, which is closely related to R2* rate mapping but yields complementary information regarding iron concentration and myelin content (36), may help further explore the pathologic changes underlying susceptibility changes over time in patients with AD. Previous studies have already demonstrated superior sensitivity of quantitative susceptibility mapping over R2* in the detection of pathologically induced changes in multiple sclerosis and Parkinson disease (37,38).

The strengths of our study included its longitudinal design together with the relatively large sample size. Even though the follow-up interval was rather short, our findings support the view that impaired iron homeostasis plays a role in the origin (or at least the propagation) of Alzheimer disease pathophysiology, alone or in combination with β-amyloid deposition.

Disclosures of Conflicts of Interest: A.D. disclosed no relevant relationships. L.P. disclosed no relevant relationships. M. Soellradl disclosed no relevant relationships. M. Sackl disclosed no relevant relationships. C.T. disclosed no relevant relationships. E.H. disclosed no relevant relationships. C.E. disclosed no relevant relationships. B.G. disclosed no relevant relationships. M.D. disclosed no relevant relationships. S.R. disclosed no relevant relationships. R.S. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: is a consultant for Axon Neuroscience, gave lectures for Ajro and Pfizer Austria. Other relationships: disclosed no relevant relationships. C.L. disclosed no relevant relationships.

Author Contributions

Author contributions: Guarantors of integrity of entire study, R.S., C.L.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, A.D., L.P., C.E., C.L.; clinical studies, A.D., C.E.; statistical analysis, A.D., L.P., M. Soellradl, M. Sackl, R.S., C.L.; and manuscript editing, A.D., L.P., M. Soellradl, M. Sackl, E.H., C.E., B.G., S.R., R.S., C.L.