Colocalization of cerebral iron with Amyloid beta in Mild Cognitive Impairment

Quantitative Susceptibility Mapping (QSM) MRI at 7 Tesla and 11-Carbon Pittsburgh-Compound-B PET were used for investigating the relationship between brain iron and Amyloid beta (Aβ) plaque-load in a context of increased risk for Alzheimer's disease (AD), as reflected by the Apolipoprotein E ε4 (APOE-e4) allele and mild cognitive impairment (MCI) in elderly subjects. Carriers of APOE-e4 with normal cognition had higher cortical Aβ-plaque-load than non-carriers. In MCI an association between APOE-e4 and higher Aβ-plaque-load was observable both for cortical and subcortical brain-regions. APOE-e4 and MCI was also associated with higher cortical iron. Moreover, cerebral iron significantly affected functional coupling, and was furthermore associated with increased Aβ-plaque-load (R2-adjusted = 0.80, p < 0.001) and APOE-e4 carrier status (p < 0.001) in MCI. This study confirms earlier reports on an association between increased brain iron-burden and risk for neurocognitive dysfunction due to AD, and indicates that disease-progression is conferred by spatial colocalization of brain iron deposits with Aβ-plaques.

1. To investigate a potential relationship between increased AD-risk, as reflected by MCI and APOE-e4 carrier-status, with cerebral iron-burden, as measured by whole-brain QSM at ultra-high field strength of 7T. 2. To estimate combined effects of MCI and increased cerebral iron-load on MPFC-coupling by resting state BOLD fMRI. 3. To characterize the relationship between iron-load and Aβ -plaque density in brain regions with altered MPFC-coupling by whole-brain QSM and 11C-PiB-PET data.

Methods
Participants. 37 study participants aged between 62 and 89 years (22 cognitively normal, 15 MCI) without evidence of significant medical illness, were recruited as part of an ongoing study at our hospital. The study was conducted in accordance with good clinical practice guidelines issued by the local ethics committee (Kantonale Ethikkommission Zürich), as well as with the declaration of Helsinki. All procedures were approved by the Kantonale Ethikkommission Zürich. Written informed consent was obtained from all participants before inclusion in the study. All participants received psychiatric examination and neuropsychological testing during screening for eligibility to participate in the current study and were categorized either as cognitively normal or MCI according to established criteria for diagnosis of MCI 42,43 50 .
Exclusion criteria for the current study were: severe cognitive deficits indicating dementia, significant medication or drug abuse with possible effects on cognition, 7T MRI exclusion criteria (such as history of claustrophobia, vertigo, seizure disorder, middle-ear disorder, double vision and the presence of metals in or on the body), MRI scans with the evidence of infection, infarction, or other focal lesions, clinically relevant changes in red blood cell count, exclusion criteria for PiB-PET, history of severe allergic reactions to drugs or allergens, serious medical or neuropsychiatric illness and significant exposure to radiation.
Carbon-11 based Pittsburgh compound B Positron Emission Tomography (PiB-PET) for estimation of brain Aβ-plaque density. PiB-PET based estimation was used to estimate individual brain Aβ -plaque-load 40,41 . Individual dose of 350 MBq of carbon-labelled PiB was injected into the cubital vein. Standard quantitative filtered back projection algorithm including necessary corrections was applied. Cerebral Aβ deposition values were extracted using PMOD Brain Tool software-package (PNEURO, Version 3.4, PMOD Technologies Ltd, Zürich, Switzerland). Late frame (minutes  values were standardized by the cerebellar gray matter average, resulting in 3D-volumes of PiB-PET retention (matrix = 128 × 128 × 47, voxel size = 2.3 × 2.3 × 3.3 mm). As a single measure of individual cortical Aβ -plaque-load, cortical PiB retention scores were determined by calculating a composite score using merged cortical PiB-PET intensity values, as reported earlier 51 .
Scientific RepoRts | 6:35514 | DOI: 10.1038/srep35514 MRI data processing. Quantitative susceptibility mapping (QSM) for measuring brain iron load. Multiple processing steps were performed to calculate from acquired MR phase images the quantitative susceptibility maps of which local cerebral iron load is inferred. First, phase unwrapping was performed using Laplacian based discrete phase unwrapping 26 . A brain mask was then obtained by skull-stripping the GRE magnitude image acquired at TE of 12 ms using FSL's brain extraction tool (BET, FMRIB Oxford, UK) with fractional threshold of 0.3. The unwrapped phase images were then divided by 2π *TE to obtain an image of the frequency shift in Hz for each echo. Subsequently, background fields were eliminated with the sophisticated harmonic artifact reduction for phase data (SHARP) 28 approach using a variable spherical kernel size with a maximum radius of 4 mm and a regularization parameter of 0.05 28 . After removal of background fields, the resulting images of the two echoes were averaged to obtain a higher SNR as compared to single echo reconstruction 52 . Inverse dipole calculations to obtain the susceptibility maps were performed using a LSQR based minimization 26,53 . From suitable reference regions such as white matter tracts and central cerebral spinal fluid (CSF) regions 23 , the region having the lowest standard deviation of mean susceptibility in all subjects was selected. In this sample the frontal central CSF region in the lateral ventricles was selected as a reference region for the final susceptibility quantification. All reported susceptibility values are then relative to the mean susceptibility value of this reference region. Classification of all subjects as "high" or "low" cerebral iron content was performed by a median split of the average cortical gray matter susceptibility of all subjects, in the same regions used for the determination of the individual cortical Aβ -plaque-load 51 .
Assessment of structure volumes and mean susceptibility. In order to assess atrophy and susceptibility differences between MCI subjects and controls, the T 1 -weighted image was co-registered to the GRE magnitude image. The co-registered T 1 image was then segmented using a multi-atlas matching approach developed as part of the Johns Hopkins University brain atlas, which is optimized for the parcellation of non-healthy brains 54,55 . ROIs were selected in the basal ganglia and several cortical gray matter structures for which mean susceptibility was calculated after eroding the ROI-masks with two pixels (1 mm) to account for partial volume effects and possible edge artifacts in cortical ROI's. To normalize different brain sizes across subjects, individual structural volume was corrected with the following approach: Corrected structure volume = Original structure volume × (group mean intracranial volume/subject intracranial volume). fMRI analysis. Pre-processing of the rs-fMRI data was performed using SPM12 (http://www.fil.ion.ucl.ac.uk/ spm/), the following steps were performed: realignment, slice time correction, co-registration of structural scan, segmentation, normalization and smoothing (FWHM = 4). The iron classification and MCI status were used as the covariates of interest in connectivity analysis using the CONN toolbox 56 . The signal was filtered with a band-pass filter using the default CONN setting of 0.01-0.1 Hz. Seed-to-voxel analysis was performed with the seed placed in the MPFC. Motion parameters (extracted using the Artifact Detection Tool, ART, https:// www.nitrc.org/projects/artifact_detect/), CSF, and white matter were regressed out, as variables of no interest. Connected voxels were included in the mask if they had a False Discovery Rate (FDR) corrected probability of p < 0.001. Using this mask gray matter susceptibility and PiB-PET retention values were extracted and averaged for each subject.

Statistics.
To examine the differences between groups 1-way MANCOVA was performed with the mean magnetic susceptibility or tissue volume of each brain structure as the outcome variable, while controlling for age and gender, followed by False Discovery Rate (FDR) multiple testing correction 57 . Effect sizes were calculated using Cohen's d. All statistical tests were performed using MATLAB R2014b (Mathworks, Natick, MA).

Results
Demographics of the study population. Demographic information for the investigated study population and neuropsychological test performance at time of inclusion are summarized in Table 1. MCI and healthy controls differed significantly in scores on the neuropsychological tests MOCA, VMLT, Boston Naming Test and WMS. Example PiB-PET images and QSM maps can be seen in Fig. 1. Cortical PiB-PET retention differed significantly (p = 0.006, effect size = 3.1) between the two groups. The frontal central CSF region in the lateral ventricles, which was used as a reference for susceptibility calculations, was significantly different in volume (healthy: 19.1 ± 2.0 ml, MCI: 24.2 ± 3.3 ml, p < 0.05) but not in absolute susceptibility ppb reading before referencing (healthy: 5.8 ± 1.1 ppb, MCI: 5.6 ± 1.1 ppb). For all subjects, the median split of the average cortical PiB-PET retention was found to be 1.13 and the median split of the average cortical susceptibility was 3.0 ppb. Accordingly, the study population was classified based on PiB-PET retention into "high" and "low" cortical PiB ("high": 7 healthy, 11 MCI) and susceptibility for iron -load ("high": 10 healthy, 8 MCI).
Effects attributable to MCI and APOE-e4 carrier status. Corrected volume was significantly different between controls and MCI subjects in the amygdala, hippocampus, thalamus and putamen with p < 0.001 and effect sizes of 0.80-1.2 (Table 2). However, no significant differences were found for the average susceptibility in any of these regions between the two groups.
Splitting the analysis based on MCI and APOE-e4 status showed no significant susceptibility differences in cortical regions of control subjects but strong significant increases in APOE-e4 carriers in the caudate nucleus (Table 3, p < 0.01, effect size = 1.03) and frontal, temporal, parietal and occipital cortices (p < 0.001, effect sizes = 0.67-1.11) for the MCI group. APOE-e4 positive subjects had significantly higher levels of Aβ -plaque-load in general, as indicated by PiB-PET retention (APOE-e4 positives: 1.56 ± 0.12, APOE-e4 negatives, 1.17 ± 0.04, p = 0.006). There was no significant effect of APOE-e4 status on the volume for any of the investigated cortical and subcortical regions (data not shown).   Table 1. Demographic data and clinical assessment scores for control subjects with normal cognition and MCI subjects at time of inclusion in the study. Data are presented as mean ± standard deviation. APOE-e4 status presented as N (percentage of group). Age and Education are in years. *Significant difference between controls and MCI with p < 0.05, **p < 0.01, ***p < 0.001.

Figure 1. Example images for a control subject (left) and MCI subject (right).
The top row shows PiB-PET images of Aβ -plaque-load in gray matter, which is highly increased in the frontal regions in the MCI subject, the signal in the white matter is non-specific to Aβ -plaque-load and is also observed in the control subject. The bottom row shows QSM maps of the same slices indicating regions with high iron load such as the basal ganglia. Susceptibility and Aβ-plaque-load correlate within brain regions defined by altered MPFCcoupling. The mask of the region with significantly increased coupling was applied to the individual PiB-PET images and QSM maps of all MCI subjects (Fig. 3). The Spearman's correlation between cortical PiB-PET  Table 3. Quantitative magnetic susceptibility (χ in ppb referenced to CSF) and PiB-PET retention (SUVR) separated by APOE-e4 status within the two groups. *Significant difference between APOE-e4 positive and negative with p < 0.05, **p < 0.01, ***p < 0.001. d indicates effect sizes (Cohen's d).
retention and susceptibility was found to be p < 0.001 (Spearman's rho = 0.86, R 2 -adjusted = 0.80). Analysis of the extracted values per group showed significant increases of cortical PiB-PET retention and susceptibility in the APOE-e4 carrier group of the MCI subjects (Fig. 3). Moreover, in the MCI group, the odds ratio for an APOE-e4 carrier to have "high" PiB-PET retention was 48 (p < 0.01, 95% confidence interval = 2.6-932.8) and 17.5 (p < 0.05, 95% confidence interval = 2.2-250.3) to be "high" iron compared to a non-carrier.

Discussion
In this study magnetic susceptibility was used as a MRI-based measure of cerebral iron load and combined with PiB-PET for measuring Aβ -plaque density in elderly subjects with normal cognition and MCI. For clarity and consistency with earlier studies, changes in susceptibility values will be referred to as changes in iron levels, due to the previously demonstrated correlation of susceptibility values with tissue iron levels in brain gray matter 24,[27][28][29] . The main finding of our study is the characterization of brain regions affected by high iron in MCI, within which a spatial colocalization of Aβ -plaques and iron was observable. This effect was associated with increased genetic risk for AD-dementia. As this colocalization is consistent with neuropathologic accounts on AD-signature brain regions 58 ,

MCI subjects
No APOE-e4 allele 1 APOE-e4 allele 2 APOE-e4 alleles our data may complement earlier considerations on the relationship between cerebral iron and AD-risk 10 . To our knowledge this is the first report on a significant impact of iron on functional network integrity in subjects with MCI. Although a previous smaller QSM-study reported higher iron load in AD 18 , our data does not show a general effect for MCI when compared to controls (Table 2) indicating such differences might occur in later stages of AD progression. However, MCI subjects with the APOE-e4 allele did show significantly higher iron levels in the neocortex (Table 3), which is a brain region affected by AD-pathology at early stages of disease progression 58 . Our finding of increased cortical iron may therefore support earlier considerations that MCI in APOE-e4 carriers may represent a prodromal stage of AD 6 . Increased cortical iron may be a more specific correlate of emerging neuro-cognitive dysfunction in prodromal AD than cortical Aβ -plaque-load, which was in our data associated with APOE-e4 independently from MCI (Table 3). This may be consistent with earlier data from MRI phase experiments that indicate significant change only for MCI subjects that progressed to dementia 59 and considerations on synergistic effects of Aβ and other aspects of neurodegeneration in AD 60,61 . The fact that reduced volume of subcortical nuclei including the hippocampal area was associated with MCI but not APOE-e4 most likely reflects heterogeneity of possible causes for MCI in the elderly.
It has been shown that the T2-prep BOLD method can achieve comparable contrast-to-noise ratio (CNR) as the conventional echo-planar-imaging (EPI) based BOLD approach, but has much reduced signal dropout and image distortion especially in brain regions close to air cavities such as some frontal and temporal areas 38 . Such dropout and distortion are particularly problematic at 7T where magnetic susceptibility gradients increase substantially at air-tissue boundary. At 3T or lower fields, where most clinical scans are conducted, such EPI artifacts are much reduced. Therefore, the T2-prep BOLD fMRI method in this study at 7T was adopted and it is expected that the findings are generalizable to studies using conventional EPI BOLD fMRI sequences at 3T. For this study a seed based approach investigating functional connectivity of the MPFC was chosen, as the MPFC is a central component of the DMN, which has been demonstrated to be impaired by Aβ pathology already in the preclinical stage of AD 33 . Our findings of iron load being associated with increased coupling in fronto-temporal brain regions is consistent with earlier reports on DMN-change in AD 62,63 and thus indicate that increased iron may contribute to the dysfunction of cognitive brain networks in subjects at risk for AD. However, as the current study investigated combined effects of MCI and iron load for definition of a brain region with particular liability for AD-associated brain change based on altered coupling to the MPFC 64 , our data does not support an independent role of iron for augmenting pathological decline in AD.
The reported correlation (Fig. 3) between cortical iron and Aβ -plaque-load within these functionally altered brain regions suggest that increased cerebral iron relates to regional accumulation of Aβ in subjects at risk for AD and reflects preclinical neuronal dysfunction in AD-signature regions. Additionally, our finding of significantly higher levels of both iron and Aβ in APOE-e4 carriers is consistent with earlier reports and suggests that the APOE-e4 allele may confer susceptibility to AD via brain iron accumulation 10 .
Our data furthermore suggest that the co-occurrence of iron and Aβ may be mediated by APOE-e4, which has been demonstrated to both promote cerebral Aβ accumulation by competing for the same clearance pathways 65 and increase cerebral iron retention by impaired lipoprotein trafficking due to low affinity of APOE-e4 to high-density lipoprotein 10 . While direct interactions between iron and Aβ may result in increased toxicity by production of redox-active iron forms and oxidative stress 20,66 , brain iron accumulation is also associated with microglial over-activation 22 , promoting neurodegeneration in AD 67 . Our observation of altered functional connectivity, may reflect these processes and thus indicate preclinical brain change with the potential of causing progressive neuronal damage, as reflected by worsening neurocognitive disorder. Although the sample size is small, the increased sensitivity at the high field strength of 7T with inherently better SNR in QSM, provides currently the most sensitive detection of in vivo gray matter iron levels 29 . When interpreting the current data it needs to be taken into account that the QSM-signal is biased by decreased myelin density 29,68 . However, the cortical and deep gray matter regions investigated in this study are low in myelin content and thus the myelin contribution to the susceptibility signal in this study was considered negligible. While spatial co-localization of microhemorrhages with Aβ -plaques may bias iron measures 69 , in the current study no microhemorrhages were observable within the brain regions investigated.
Considering that iron may reflect processes associated with Aβ related neurocognitive dysfunction, further studies are needed to investigate whether the efficacy of therapeutic strategies lowering brain Aβ -plaque-load for slowing down progression of AD 7,70 is affected by the extent of local iron accumulation 11,71 .