Lifestyle, biological, and genetic factors related to brain iron accumulation across adulthood

.


Introduction
Iron is an essential metal involved in several physiological processes, including neurotransmitter synthesis (Zecca et al., 2004) and production of adenosine triphosphate in mitochondria (Todorich et al., 2009).A tight regulation of iron balance is ensured by several proteins, including ferritin, which is used as an indicator of body iron storage, with higher serum levels reflecting high stores (Adams, 2015).
Transferrin is the main iron transporter through the plasma and binds to specific transferrin-receptors to deliver iron to cells.The demand for iron regulates the production of transferrin and the expression of transferrin-receptors, such that presence of low iron concentration increases the production and expression of these proteins (Khumalo et al., 1998).To ensure normal physiological function, iron homeostasis relies on the interplay between uptake, transport, storage, usage, and recycling.
Dysregulation of iron homeostasis, causing an increase in iron content, can trigger production of reactive oxygen-species that causes neurotoxicity and cell loss due to ferroptosis (Ward et al., 2014).
While in young age, iron is critical for brain development and linked to better cognition (Larsen et al., 2020), aging has been associated with accumulation of cellular iron in the brain (Hallgren & Sourander, 1958).Growing evidence demonstrates deleterious effects of iron accumulation on brain integrity and cognition in older age (Biel et al., 2021), with limited knowledge on the factors influencing brain iron content and accumulation.Whereas a few studies have investigated some potential factors in relation to brain iron, no study has used a longitudinal design to assess factors of brain iron accumulation with aging.The few cross-sectional studies have yielded either conflicting or non-replicated patterns.
Among modifiable factors, greater brain iron content has been linked to smoking (Li et al., 2021a;Pirpamer et al., 2016) and higher body mass index (BMI) (Li et al., 2021b;Pirpamer et al., 2016) in healthy adult populations.Greater iron content has been also related to alcohol habits among individuals with alcohol use disorder (Juhás et al., 2017).However, among healthy individuals without alcohol use disorder, the results regarding alcohol consumption have been mixed: one study did not find any association (Pirpamer et al., 2016), whereas another study reported that higher alcohol consumption was related to greater iron content (Topiwala et al., 2022).Dietary intake has also been investigated in relation to brain iron but yielded mixed results (Hagemeier et al., 2015;Zachariou et al., 2021).In addition, previous studies have suggested that hypertension may be associated with J o u r n a l P r e -p r o o f higher brain iron content (Rodrigue et al., 2011).Only one study found a significant link between brain and serum iron in a small sample of men (House et al., 2010), with higher serum iron levels being related to higher brain iron content (but see Pirpamer et al., 2016).Moreover, there are inconsistencies regarding which brain regions are affected and to which extent.
Besides modifiable factors related to lifestyle and vascular health, iron homeostasis is influenced by genetic factors, such as homeostatic iron (HFE) having been linked to iron levels in the brain (Kalpouzos et al., 2021).One study among patients with cognitive impairment showed that ApoE ε4 carriers had higher brain iron content compared to non-carriers (Yim et al., 2022), supporting the notion that different metals in the brain (e.g., iron, zinc, and copper) interact with proteins related to Alzheimer's disease (AD), contributing to the progression of the disease (Ayton et al., 2015;Xu et al., 2014).
Although the exact interaction between ApoE and brain iron is not established, the gene plays a role in several pathways that have been linked to elevated brain iron content, such as proteinopathy (amyloid and tau pathologies are rich in iron), metabolic factors (lipid metabolism and cardiovascular health), and neuroinflammation (Ayton et al., 2017;Kloske & Wilcock, 2020;Mahley, 2016;Spotorno et al., 2020).
In this longitudinal study, we aimed to assess the influence of various factors on brain iron content and accumulation in normal aging, for the purpose of hypothesis generation.We examined factors previously investigated in cross-sectional studies, such as lifestyle (i.e., alcohol, smoking, physical activity, diet), cardiovascular health, blood markers of iron metabolism (i.e., ferritin, transferrin, transferrin receptors), and the potential effect of ApoE ε4 status in relation to brain iron content and accumulation.The nature of this study was to a large extent exploratory but with some hypotheses: in line with the findings by House et al. (2010), we hypothesised that indicators of higher levels of peripheral (blood) iron would be related to more iron in the brain.Further, we reasoned that age would influence the direction of the relationship between iron and the factors: whereas in younger adults, more iron would be beneficial and associated with better lifestyle and indicators of health, within older adults, higher levels of iron would be detrimental and associated with poorer lifestyle and indicators of health.
We used data from a longitudinal healthy adult lifespan sample, and approximated iron concentration in the brain with magnetic resonance imaging (MRI)-based quantitative susceptibility mapping (QSM; Langkammer et al., 2012) in the iron-rich basal ganglia (i.e., caudate nucleus, putamen, and pallidum) and the cortex.

Participants
J o u r n a l P r e -p r o o f Data from the IronAge project were used, which consisted of 208 individuals (age range 20-79, 108 female) at baseline and 135 individuals at follow-up (mean days between MRIs: 1003, SD: 81).
Participants were healthy volunteers and reported no history of neurological or psychiatric conditions.
The protocol consisted of blood sampling, physiological and anthropometric measurements, cognitive testing, MRI of the brain, and questionnaires.

MRI acquisition and preprocessing
Acquisition.A structural 3D T1-weighted IR-SPGR image was obtained with the following parameters: repetition time (TR) = 6.96 ms, echo time (TE) = 2.62 ms, 176 axial slices with slice thickness of 1 mm, in-plane resolution = 0.94 mm x 0.94 mm, field of view (FOV) = 24 cm, flip angle = 12°.For brain iron quantification, a 3D multi-echo Gradient Recalled Echo (meGRE) sequence was used with the following parameters: TR = 27.2 ms, 124 axial slices of 1.2 mm thickness, in-plane resolution = 0.94 mm x 0.94 mm, FOV = 24 cm, flip angle = 17°.The first TE was 1.9 ms, and it was followed by 7 consecutive TEs with a constant interval of 3.18 ms between them.For assessment of white matter hyperintensities (WHMs), a 3D T2-weighted fluid-attenuated inversion recovery (FLAIR) sequence with the following parameters was obtained: TR = 8000ms, TI = 2259ms, effective TE = 114ms, echo train length 180 echoes, flip angle = 90°, FOV = 27cm, reconstructed into 292 sagittal slices with a 0.6mm thickness and an in-plane resolution of 0.53mm x 0.53mm.
Quantitative susceptibility mapping (QSM).Gustavsson et al. (2022).We used the recommended nonlinear variant of Morphology-enabled dipole inversion (MEDI) and the MEDI toolbox to calculate the QSM images (Liu et al., 2011); http://weill.cornell.edu/mri/pages/qsm.html).Initially, the total field map was estimated from the complex meGRE images by performing a nonlinear least square fitting on a voxel-by-voxel basis and was then spatially unwrapped using a guided region-growing unwrapping algorithm (Xu & Cumming, 1999).The background fields were eliminated using a nonparametric technique based on Projection onto Dipole Fields (Liu et al., 2011).Finally, the corrected frequency map was used as input for the field-to-source inverse problem to calculate the map of susceptibility.
To address the issue of relative susceptibility values, we implemented zero-referencing where a region is placed as a reference and its average susceptibility is subtracted from the susceptibility values across the brain.We selected a region in corticospinal tract 39]) as the reference region (Garzón et al., 2017).A region of 1000 voxels was centred on the coordinates using an in-house developed region-growing algorithm.The algorithm implements the white-matter mask to ensure the created region encompassed white-matter tissue only.The FMRIB Software Library (FSL, http://fsl.fmrib.ox.ac.uk) was applied to calculate non-linear transformation parameters (Andersson et al., 2007) and to obtain the white-matter mask (Zhang et al., 2001).

J o u r n a l P r e -p r o o f
Automated segmentation of cortical and deep grey-matter structures was performed on the T1weighted images with the Freesurfer longitudinal stream (Reuter et al., 2012) -version 7.1.0(http://surfer.nmr.mgh.harvard.edu/).The longitudinal processing pipeline creates an unbiased within-participant template space and image that is used for segmentation and surface reconstruction (Reuter et al., 2010;Reuter & Fischl, 2011).This was done for participants with both two time points of MRI data and a single time point, to leverage the valuable information of the whole cohort.
Next, QSM and the segmentation results were resampled to the native structural space.Then, statistics including average and standard deviation were computed on the QSM maps.We merged the segmented left and right caudate, putamen, and pallidum for each region to construct the basal ganglia.Baseline iron of nucleus accumbens did not correlate with age (r = 0.056, p = 0.42) nor was there any significant iron accumulation over time (F( 2 , 134 ) = 0.057, p = 0.8).Moreover, the reliability in QSM estimates in this nucleus was lower compared to caudate, putamen and pallidum (intraclass correlation coefficient ICC (accumbens) = 0.66, confidence interval CI = 0.53-0.76;ICC (caudate) = 0.90, confidence interval CI = 0.80-0.95;ICC (putamen) = 0.94, confidence interval CI = 0.86-0.97;ICC (pallidum) = 0.92, confidence interval CI = 0.88-0.94).Therefore, nucleus accumbens was not included in basal ganglia QSM estimates.The exclusion of the region is consistent with the literature on brain iron in relation to different disorders, including Huntington's disease (Domínguez D et al., 2016), parkinsonian diseases (Hanssen et al., 2019), psychotic spectrum disorders (Sui et al., 2022), Alzheimer's disease (Ghaderi et al., 2024), and normal aging (Hofer et al., 2022;Petok et al., 2024;Yablonskiy et al., 2021).Similarly, 31 segmented regions from left and right hemispheres were merged to form the cortex ROI (see Table S1 in supplementary materials for included regions).
Prior to computing statistics on the QSM maps, the boundary of segmentations was eroded by one voxel, and a fraction (15%) of the most extreme values was removed to avoid the influence of high signal from neighbouring vessels and obtain more robust estimates (Garzón et al., 2017).
The automated segmentation of lesions was performed with the longitudinal pipeline in the LST software using T1 and FLAIR images.First, the images were run through the Lesion Growth Algorithm with kappa threshold of 0.3, determined after visual quality inspection of a random sample of the lesion probability maps.Then, the segmented lesion maps from the 2 time points were compared in a consecutive and iterative manner and to decide if changes were significant or due to erroneous conclusions of lesions (Schmidt et al., 2019).The WMH volumes were adjusted for total intracranial J o u r n a l P r e -p r o o f volume (Jack et al., 1989) before being transformed for normality using a two-step normality transformation for analysis (Templeton, 2011).

Blood sampling and composite score of blood iron
Blood was collected through venepuncture before 10 AM while fasting since 8 PM the day before.The samples were brought to the Centre for Clinical Laboratory Studies for immediate analyses using standard procedures (Karolinska Hospital, Stockholm).For the present study, the following markers of iron metabolism were used: serum ferritin, plasma transferrin, and serum transferrin receptors.As these markers are key indicators of the regulation of iron metabolism, we combined them and transformed them into a T-score, creating a composite score -which will be referred to as the composite score of blood iron.To this end, the direction of ferritin was inverted to match the direction of transferrin and transferrin receptors.The variables were then converted into z-scores and averaged into one score, and lastly transformed into a T-score with a mean = 50 and standard deviation = 10.
The follow-up score was adjusted to baseline using the mean and standard deviation of the baseline score.

Genotyping
DNA was extracted from peripheral blood samples and stored at Karolinska Institutet Biobank.DNA samples were transferred on PCR plates and sent to the SNP&SEQ Technology Platform, Uppsala University (National Genomics Infrastructure (NGI), SciLifeLab Sweden).The genotyping was performed using a multiplexed primer extension (SBE) chemistry of the iPLEX assay with detection of the incorporated allele by mass spectrometry with a MassARRAY analyser from Agena Bioscience (Gabriel et al., 2009;Ross et al., 1998;Storm et al., 2019).Raw data from the mass reader was converted to genotype data using the Typer software (Agena Bioscience).ApoE was dichotomized into any ε4 allele (n = 56) and no ε4 allele (n = 152) based on the single-nucleotide polymorphisms rs429358 and rs7412.The genotype distribution for rs429358 were in Hardy-Weinberg equilibrium (χ2 = 2.63, ps > 0.1), but not for rs7412 (χ2 = 15.18,ps < 0.001).Our sample has a higher frequency of homozygotes and lower frequency of heterozygotes (C/C = 85.6%,T/T = 2.9%, C/T = 11.5%)than expected for equilibrium (C/C = 83.2%,T/T = 1%, C/T = 15.8%).

Lifestyle variables: Physical activity, smoking, and alcohol
All participants were assessed on lifestyle-related factors using structured questionnaires and consisted of physical activity and diet (Harris et al., 2013), smoking, and alcohol (Qiu et al., 2012).
Participants reported the frequency of their physical activity during the previous year (from never to several times or hours a day), based on five questions with predefined response categories.The watching tv or reading).The responses were calculated as an average (i.e., 1-6, with leisure time inactivity inversely scored) with higher score indicating a more active life.Smoking was measured as status (never/past/current), as cumulative pack-years, and as pack-years for the past 5 years.A packyear was defined as daily consumption of 20 cigarettes over the period of one year with recent packyears only taking the last 5 years into consideration (Pirpamer et al., 2016).Alcohol consumption was registered as both quantity measured as units (12g/unit) and frequency measured as never or occasional, light to moderate, and heavy (>7 drinks per week for females or >14 drinks per week for males (Breslow et al., 2013)).

Mediterranean-DASH Intervention for Neurodegenerative Delay (MIND) score
Diet was assessed based on data from a semi-quantitative food frequency questionnaire (FFQ; Harris et al., 2013).For each food item, the average frequency of consumption over the past year was determined using a 9-point scale that ranged from never to three or more times per day (Johansson et al., 2002).The MIND score was then computed using selected items from the FFQ.The score was based on the original MIND dietary pattern (Morris et al., 2015), but adapted for food choices and dietary recommendations in the Nordic countries.These dietary components included three unhealthy food groups (butter and margarine, red meat products, pastries and sweets) and 10 healthy food groups (berries, whole grain, beans, nuts, green leafy vegetables, other vegetables, cooking oils (olive and rapeseed), cheese, fish, poultry).Table S2 in supplementary materials describes the items and cut-offs.
For all the diet score components except cooking oils, we summed the frequency of consumption of each food item and ascribed a score of 0, 0.5, or 1. Consumption of cooking oils was ascribed a 1 if used for cooking and eating, and 0 for not consumed at all.In total 13 out of original 15 components were used; alcohol was excluded as it was analysed independently, and information on deep-fried food was not available.The total MIND diet score was summed over all 13 components scores with higher score corresponding to better dietary habits (range 0-13).

Cardiovascular health
As indicator of vascular health, we used the Life's simple 7 (LS7), a validated metric consisting of the following seven modifiable cardiovascular health factors reflecting lifestyle and biometric measures: diet, smoking, BMI, blood pressure, blood glucose level (glycated haemoglobin, HbA1c), total plasma cholesterol, and physical activity (Thacker et al., 2014).Each factor was categorised into three levels (defined as poor=0, intermediate=1, and optimal=2) according to the cut-off values used by Sabia et al. (2019).The participants' arterial blood pressure was measured twice, with a 5-minute interval J o u r n a l P r e -p r o o f between each measurement, on their left arm while sitting down using a sphygmomanometer.The mean of the two readings was then used.Medication for treating hypertension and cholesterol was self-reported and considered.Body Mass Index (BMI) was calculated by dividing an individual's weight in kilograms by their height in meters squared.For smokers, we split former smokers if they stopped within the last five years or more than five years ago.The assessment of physical activities was based on a questionnaire that estimated how many hours a week the participants spent doing different kinds of activities (Nevalainen et al., 2015).The activities were then categorised according to how effortful they were to perform, as defined by Sabia et al., (2017).We selected the most commonly consumed vegetables, fruits, fibre, and high-fibre bread into separate variables that were used to determine if the participants followed the dietary daily recommendations.The total LS7 score was obtained by summing up the points from each weighted item (from 0 to 14).The cut-off values are presented in supplementary Table S3.

Statistical analyses
The variables were assessed for normal distribution using Shapiro-Wilk test before being included in the models.When necessary, Templeton's two-step method was performed to transform any nonnormal variables into normally distributed ones (Templeton, 2011).
Structural equation modelling (SEM) in AMOS (IBM SPSS 26) was used to assess the relationship between the factors and brain iron content and accumulation (see Figure 1).Longitudinal changes for the factors and brain iron were modelled by estimating change regression models (McArdle, 2009).
Note that the change regression model accommodates the estimation of difference scores independent of the individuals' initial baseline levels.The model allows to attenuate potential effects pertaining to regression to the mean, a statistical phenomenon, which may be misinterpreted as true change.Regression to the mean may be reflected in strong correlations between baseline and change (Barnett et al., 2005).
First, we tested whether iron accumulated in the selected regions of interest over time in separate change regression models.This model served as a measurement model and defined the relationship between the observed measurements (i.e., iron at time point 1 (T1) and time point 2 (T2)) and the unobserved difference between the time points (difference = T2-T1).Figure 1:1 illustrates the measurement model and can be described using the following equation:

Model 1: Iron ~ age + sex + education + grey-matter change + white-matter hyperintensity load
Second, a combined model was constructed to examine the relationship between the two measurement models.To that end, the two measurement models were combined so that the J o u r n a l P r e -p r o o f relationship between every factor was separately assessed for every brain region.This allows for the relationship between iron and the factor to be assessed cross-sectionally and longitudinally within the same model, as baseline and follow-up data are included for both.Figure 1:2 illustrates the individual factor models and can be described using the following equation: Model 2: Iron ~ baseline factor + follow-up factor + interaction terms + covariates Third, an integrative model was constructed for each separate brain region in which all factors associated with the outcome were included into the combined model.To this end, the factor scores of the significant (on 10% level) factors were extracted from their respective models.Figure 1:3 illustrates the integrative factor model, and the model equations are defined in respective result section.
Factors age, sex, and education were included in the models as covariates of baseline and difference scores of iron and the factors.Sex was categorised with females as a reference category, smoking status and alcohol habits were categorised from 0-2 with "never" as reference group.Furthermore, WMH and volumetric changes in regional grey-matter change were included as covariates in the models due to their relation to age and iron (Daugherty et al., 2015;Dickstein et al., 2007;Vernooij, 2007;Yan et al., 2013).To assess possible moderator effects of age or sex on the iron-factor relationships, we included age and sex interactions in the models for relevant factors.In addition, in the models assessing the link between brain iron and ApoE, genotype status (no ε4 allele as reference group) and its interaction with age were included.The interaction variables were removed from the model if they were not significant at alpha 0.01.To follow up on significant age-interactions with posthoc analyses and for the purpose of illustration, we stratified the sample into three age-groups with equal age-range (younger: 20-39 years old, middle-aged: 40-59, older: 60-79).Similarly for sexinteractions, the sample was stratified by sex.
J o u r n a l P r e -p r o o f To accommodate for missing data due to the attrition between time point 1 and 2, full information maximum likelihood (FIML; (Finkbeiner, 1979;Schafer & Graham, 2002)) estimates were used for all models.FIML is a procedure which uses available information for estimating parameters that contain variables with missing values rather than imputing or omitting data.Using FIML allows for a more accurate population estimate and a better measurement model compared to other procedures dealing with missing data, such as listwise deletion (Schafer & Graham, 2002).The estimates are unbiased under the assumption of missing at random.That is, the likelihood of data from a variable missing may depend on other variables in the model rather than the variable itself (Little & Rubin, 2002).In our sample, only age was a significantly differentiating variable with dropouts being younger than returnees (t(206) = -2.1,p = 0.03), whereas other variables such as iron content, blood markers, cardiovascular score, and lifestyle did not predict missingness.We included covariates such as J o u r n a l P r e -p r o o f chronological age, sex, and education in our models to allow FIML to account for the missing data (Rubin, 1975;Schafer & Graham, 2002).The inclusion of these covariates optimises the likelihood of the model parameter values estimated by FIML to be accurate and unbiased when missing at random (Staudt et al., 2022).Both univariate (+/-3.29 SD; n = 2) and multivariate outliers (baseline and followup within brain regions; Mahalanobi's distance; p < 0.001 threshold for the X 2 value; n = 8) were excluded from analyses and treated as missing (Tabachnick & Fidell, 2013).
The following indices were used to evaluate whether the model provided a good representation of the data: the comparative fit index (CFI) and the root mean square error of approximation (RMSEA).Values greater or equal to 0.95 for CFI and values lower or equal to 0.08 for RMSEA were considered to indicate acceptable fit (Kline, 2005).The alpha level for the chi-squared difference test of the models was set to 0.05.Given the exploratory design of the study, the alpha level for statistical significance of the individual factors in the models was set to 0.1.In the integrative models, the alpha level was set to 0.05.The total number of models run was 36 (2 for assessing change in iron, 10 factors x 2 regions, 6 age+sex interactions x 2 regions, 2 integrative models).To ensure the stability and robustness of the results, and to avoid interpretation of spurious findings, we conducted bootstrapping analyses.The bootstrapping analyses were based on 5000 samples and bias-corrected 95% confidence intervals (CIs) of parameter estimates for the regression weights.The effects were considered reliable if 95% CIs for the regression weights did not include zeros.Additionally, given the number of analyses performed, we also report False Discovery Rate-adjusted p-values (Benjamini & Hochberg, 2000) for the integrative models.Additional analyses for caudate, putamen, and pallidum were performed and are presented in supplementary materials.

Results
Demographic information on indicator level is presented in Table 1.
Table 1.Baseline participant characteristics for the total sample and three age groups J o u r n a l P r e -p r o o f
Lower brain iron content at baseline was related to greater iron accumulation in both cortex and basal ganglia.Older age was significantly associated with higher brain iron at baseline for all models.

Composite score of blood iron
All the statistics for the composite score of blood iron analyses are reported in Table 3.There was an increase over time in the composite score of blood iron (estimated increase = 10, p = 0.003) and significant variance in change (p < 0.001).A higher score of the composite score at baseline was significantly associated with higher baseline iron and greater iron accumulation in basal ganglia.We did not observe any significant main effects of iron in cortex.An age interaction with the composite score on iron accumulation in basal ganglia was observed, such that older adults with more peripheral iron accumulated more basal-ganglia iron (β = 0.270, p = 0.036; younger adults β = 0.12, p = 0.4; middleaged adults β = 0.19, p = 0.16).Furthermore, a significant sex interaction with the composite score on baseline iron was observed in both basal ganglia and cortex.In basal ganglia, a higher composite score was associated with more iron in males only (β = 0.221, p = 0.008; females β = 0.02, p = 0.8).In cortex, however, none of the associations were significant when stratifying by sex (males: β = 0.16; females β = -0.05,ps > 0.05).

Cardiovascular factors and ApoE status
Results concerning cardiovascular health as well as ApoE status are reported in Table 3.There was an increase over time in the LS7 score (estimated increase = 4.2, p < 0.001) and significant variance in change (p < 0.001).For LS7, we observed that improved cardiovascular health (i.e., higher score) was related to less iron accumulation in both cortex and basal ganglia.We found a significant interaction between age and ApoE on iron accumulation in basal ganglia, such that ε4 carriers exhibited greater brain iron accumulation than non-ε4 carriers among younger adults only (β = 0.284, p = 0.031; middleaged β = -0.125,p = 0.35; older: β = -0.004,p = 0.97).

J o u r n a l P r e -p r o o f
Table 3. Individual-factor models with brain iron in each region assessing the cross-sectional and longitudinal associations with the composite score of blood iron, LS7, and ApoE status.

Lifestyle factors
The statistics for the analyses of the relationship among brain iron, iron accumulation, and lifestyle factors are presented in Table 4.The MIND score (estimated increase = 2.3, p < 0.001), physical activity (estimated increase = 1.1, p < 0.001), and alcohol habits (estimated increase of consumption with never as reference category, p < 0.001) significantly changed over time and had significant variance in change (p < 0.001).Weekly alcohol consumption and the smoking variables did not, and as such the changechange associations were omitted from the analyses.
A higher MIND score (i.e., higher adherence to healthy diet) at baseline was related to higher baseline iron content and greater iron accumulation in cortex.There was an interaction with age which demonstrated that younger adults with better diet at baseline had higher iron content in cortex (younger β = 0.32, p = 0.006; middle-aged β = 0.002, p = 0.9; older β = 0.16, p = 0.13).Furthermore, an interaction between age and the MIND score on baseline iron content in basal ganglia was observed.

J o u r n a l P r e -p r o o f
Consistent with the pattern observed in cortex, younger adults with a better MIND score had more iron at baseline in basal ganglia (younger β = 0.296, p = 0.007; middle-aged β = -0.096,p = 0.4; older β = 0.03, p = 0.8).We did not observe any significant interactions between sex and the MIND score.Regarding physical activity, we observed both a sex-and an age-interaction with physical activity on iron content and accumulation in cortex.More physical activity at baseline was related to greater iron accumulation in cortex among males (β = 0.165, p = 0.077; females β = -0.14, p = 0.2).When stratifying by age group, however, none of the associations were significant (younger β = 0.17, p = 0.2; middleaged β = 0.14, p = 0.2; older β = -0.02,p = 0.9).Lastly, we observed that a higher consumption of alcohol at baseline was related to greater iron accumulation in basal ganglia only (β = 0.136, p = 0.097).We did not observe any significant associations with the smoking variables.

Integrative models
Only factors individually associated with the iron outcomes at p < 0.1 were included in the integrative models.All the statistics are reported in Table 5.

J o u r n a l P r e -p r o o f
The model for cortex was constructed as follows:

Model 3a: iron ~ MIND + MIND x Age + Physical activities + Physical activities x Age + Change in LS7 + covariates
In the cortex model, lower baseline iron content was marginally related to increased iron accumulation.A healthier diet at baseline was related to greater iron accumulation (p uncorrected = 0.026).The association between improved cardiovascular health (LS7) and decreased iron accumulation was marginally significant (p uncorrected = 0.057, Figure 2A).The results are illustrated in Figure 2A-B.However, after FDR-correction all associations were at trend levels only and are presented in Table 5.
The model for basal ganglia was constructed as follows:

Model 3b: iron ~ Blood marker + Blood marker x Age + MIND + MIND x Age + ApoE + ApoE x Age + Change in LS7 + covariates
In the basal ganglia model, lower baseline iron content was related to increased iron accumulation.A higher composite score of blood iron was related to higher brain iron content (P FDR = 0.02).There was a marginally significant age-interaction with baseline composite score, such that higher level of the composite score of blood iron at baseline was related to increased iron accumulation among older adults (younger: β = 0.023, p = 0.8; middle-aged: β = 0.154, p = 0.24; older: β = 0.245, p = 0.043).We observed that a healthier diet was related to higher baseline iron content (P FDR = 0.024).This effect was driven by younger adults (younger: β = 0.329, p = 0.003; middle-aged: β = -0.12,p = 0.3; older: β = 0.01, p = 0.9).Lastly, there was a main effect of ApoE (P FDR = 0.03) and an interaction with age on iron accumulation, with ε4 carriers exhibiting greater brain iron accumulation than non-ε4 carriers among younger adults only (younger β = 0.316, p = 0.003; middle-aged β = -0.04,p = 0.7; older: β = 0.08, p = 0.5).The results are illustrated in Figure 2 (B-D).

Control analyses
To ensure the stability and robustness of the results, and to avoid interpretation of spurious findings, we conducted bootstrapping analyses.We based the bootstrapping analyses on 5000 samples and reported bias-corrected 95% confidence intervals (CIs) of parameter estimates for the regression weights.The effects were considered reliable if 95% CIs for the regression weights did not include zeros.In the cortex model, the association between a healthier diet and iron accumulation remained significant (β = 0.19, p = 0.003, BS 95% CI: [0.073, 0.29]).Further, the observed relationship between iron accumulation and decline in cardiovascular health remained significant as well (β = -0.15,p = 0.03, BS 95% CI: [-0.282, -0.017]).
Lastly, despite lower reliability in QSM estimates of the nucleus accumbens, we also performed additional analyses where this region was included in the basal ganglia and the results were similar, except for the association between changes in cardiovascular health and iron accumulation which was no longer significant (data not reported).

Discussion
In this longitudinal study, composite score of higher peripheral iron load was related to higher brain iron content and iron accumulation.Cardiovascular health, and diet, were also related to brain iron.
The potential influence of some of the factors was further modulated by age and sex.The results discussed here on the relationships between modifiable and non-modifiable factors, brain iron content and accumulation among healthy adults are based on the integrative models.
The accumulation of brain iron in normal aging has so far been elusive as few studies have investigated brain iron longitudinally in lifespan sample.Here, we replicate the accumulation of iron over time in basal ganglia and extend the findings to cortex (Daugherty et al., 2015;Gustavsson et al., 2022).In line J o u r n a l P r e -p r o o f with a previous study, across the investigated regions, greater iron content at baseline was related to older age, but also to less iron accumulation (Gustavsson et al., 2022).

Peripheral iron marker
One of our aims was to investigate the link between peripheral (blood) iron markers and brain iron.
Studies that have investigated this link have reported mixed results, as only one study demonstrated an association between blood iron markers and brain iron content (House et al., 2010), whereas Pirpamer and colleagues (2016) did not.Here, we showed that a composite of blood markers of iron, reflecting higher levels, was linked to higher iron content in basal ganglia.Ferritin is one of the major components in iron storage and has been implicated in transporting iron across the blood-brain barrier (Fisher et al., 2007).Under normal circumstances, the production of ferritin is triggered by the increase of free iron and decreased at low levels of iron (Arosio et al., 2009).The relationship between ferritin and the transferrin proteins is inverse, meaning that when there are higher levels of iron and therefore ferritin, there are lower levels of transferrin and transferrin receptors (Kim et al., 2020).As such, individuals with higher levels of iron have higher levels of ferritin, and lower levels of transferrin and transferrin receptors in the blood.Together, the three blood markers represent storage and transport of iron in the blood.The underlying reason for why higher levels of peripheral blood markers of iron would be associated with more accumulation of iron in the brain is unclear.One possible factor could be moderation by genetic profiles.There are several genes that are implicated in iron metabolism.In genes involved in iron transport and storage, such as HFE, TF, and SLC25A37, some polymorphisms have been related to higher iron load in basal ganglia nuclei (Elliott et al., 2018;Kalpouzos et al., 2021).
In particular, C282Y and H63D mutations on the HFE gene are risk factors for peripheral iron overload; we recently showed that having at least a mutation on C282Y or H63D was related to higher iron load in both the blood and putamen in the same sample used in the present study (Kalpouzos et al., 2021).
Another possible factor could be that the blood brain barrier becomes more permeable with older age (Verheggen et al., 2020).Astrocytes are essential for the formation of the blood-brain barrier and have also been suggested to play a critical role in the distribution of iron throughout the brain (Dringen et al., 2007).Increased permeability of the blood-brain barrier in combination with age-related astrocytic dysfunction (Duncombe et al., 2017), could result in disturbed iron distribution and increased iron uptake.Lastly, the opposite pattern where dysregulation of iron homeostasis can cause a decrease rather than increase in iron content has critical consequences for brain functions, especially during development (Lozoff & Georgieff, 2006), which highlights the age-varying difference of brain iron, and related associations with modifiable factors, such as peripheral iron.Although further studies are needed, the connection between peripheral blood iron opens up the possibility of using blood-based iron as a proxy of brain iron content and factor of brain iron accumulation.
J o u r n a l P r e -p r o o f

Genetic marker of ApoE
We found that ApoE ε4 carriers accumulated more iron, more specifically in the group of younger adults.Our finding gives longitudinal support to previous studies, which found that ε4 carriers had higher QSM susceptibility in both cortical and subcortical regions (Nir et al., 2021;Van Bergen et al., 2018).Our result is in contrast to the magnification hypothesis of old age, where genetic effects on white-matter structure and brain activity became evident in older age (Papenberg et al., 2015;Sambataro et al., 2009).It is possible that the accumulation of iron plateaus with increasing age, and thereby attenuating or nullifying the genetic effect of being an ε4 carrier on iron accumulation in older age.In both our investigated regions, greater iron content at baseline was related to both older age and less iron accumulation.Variability of elevated iron content as a function of age and region has been observed in previous cross-sectional studies, specifically in basal ganglia (Aquino et al., 2009;Hallgren & Sourander, 1958).Still, most studies published are cross-sectional and largely focus on older individuals, highlighting the need for longitudinal studies encapsulating the full lifespan.

Cardiovascular health
Although the association between iron accumulation and cardiovascular health was not significant after FDR-correction, we chose to discuss the uncorrected result as it was robust according to the bootstrap analyses.Declining cardiovascular health over time was related to increased iron accumulation, adding support for the relationship between cardiovascular health and brain iron, as previously demonstrated cross-sectionally in hypertensive patients (Rodrigue et al., 2011).
Cardiovascular disorders (i.e., hypertension, metabolic syndrome) play a major role in the modulation of neurogenic proinflammatory responses through pathways of the heart-brain axis (see review by Hu et al., 2023), which may increase the susceptibility of greater iron accumulation.This may have implications for future interventions since the results suggest that following better guidelines of cardiovascular health contributes to less iron burden and possibly better brain health (Fu et al., 2023).
We further demonstrated that a global assessment score such as LS7 can be useful in investigating the relationship with brain iron.Importantly, the novelty of change-change associations in relation to composite scores of cardiovascular health should be emphasized as they are rare (Stephen et al., 2021).

Nutrition
We observed that a healthier diet was related to more iron in basal ganglia.This association was driven by the younger adults, whereas middle-aged and older adults displayed patterns in an opposite direction (although non-significant).Furthermore, a healthier diet at baseline was related to more iron accumulation in cortex.Iron has a critical role in development (Pivina et al., 2019) and has been J o u r n a l P r e -p r o o f reported to be beneficial to cognition in young age (Larsen et al., 2020).At a young age, the role of iron is likely beneficial due to normal iron metabolism and a higher metabolic demand for iron.
Conversely, in older age, iron dyshomeostasis is likely to occur and iron overload/overaccumulation may be deleterious.Our observations support Zachariou et al. (2022), who showed that healthier food (in terms of healthy fats and antioxidants) was related to lower brain iron content in an older population.The divergent findings of diet and cardiovascular health may be dependent on early vs. late adulthood.At a younger age, higher levels of iron may not have reached a threshold causing detrimental effects (c.f.Kalpouzos et al. 2017).In late adulthood, the combination of age-related iron dyshomeostasis, declining cardiovascular health may result in increased iron burden.

Smoking and alcohol consumption
In contrast to previous studies, we did not observe any links between brain iron and smoking nor alcohol.The reason for our non-findings regarding smoking is likely due to lack of statistical power.The prevalence of daily smoking in Stockholm has dropped from 16% to 5% between 2002-2021(Region Stockholm, 2023).This is reflected in our sample, as only 6% were current smokers in comparison to the 26% current smokers in another study, where smoking was related to more brain iron in thalamus (Li et al., 2021a).Furthermore, the means for accumulated and recent pack-years were 6.4 and 0.5 (median 5.5, 0) respectively in comparison to the median of 33 in the study where pack-years was related to more brain iron (Pirpamer et al., 2016).Regarding alcohol, the mean units consumed per week was 5 in our sample, with 25% never or occasional drinkers.In comparison, Topiwala et al. (2022) reported a mean of 17 units per week and only 3% of never drinkers.It is therefore possible that the alcohol consumption in our sample was below a level where it would have a detrimental effect on health, or enough to influence brain iron levels.

Strengths and limitations
The main strengths of this study include the wide age range of the study population, which allowed exploring factors related to brain iron content across the adult lifespan; the study design, which enabled assessment of the effect of such factors on brain iron content changes; and the comprehensive assessment of the study participants, for whom multiple modifiable and nonmodifiable factors were assessed.Another strength is the analytical approach of our study.Whereas we chose to only discuss the results of the integrative models, the results from the exploratory individual-factor models can serve as a basis for specific investigations in future studies.
The study's main limitation is that our population is composed of healthy volunteers and pathologies such as diabetes were an exclusion criterion, which led to a reduced range in LS7.As such, our results may be an underestimation of the effects that could be seen in the general population.This might have Moreover, another limitation is the sample attrition at follow-up, which affected our statistical power.
The limited number of ApoE ε4 carriers may have resulted in potential small genetic effects being missed.Further, the departure from Hardy-Wienberg equilibrium of the SNP rs7412 may have had a minor impact on the effects.The heterogenous distribution of brain iron remains unclear.Future studies should consider broader exploration as not all regions where iron accumulates were included.
Lastly, future studies are needed for a thorough investigation of possible sex differences given the indications observed in this study.

Conclusions
Our study assessed independently and simultaneously modifiable and non-modifiable factors of brain iron content and brain iron accumulation in a longitudinal setting among healthy adult individuals.
Critically, we demonstrated the connection between peripheral body iron and brain iron, highlighting the possibility of considering blood-based iron as a proxy of brain iron content.Furthermore, we demonstrated that a better diet has a beneficial relationship with brain iron content and accumulation among younger adults.We also demonstrated that declining cardiovascular health was related to increased iron accumulation.This inverse relationship between brain iron and lifestyle/cardiovascular health as a function of age may reflect that the metabolic demand and role of iron could be agedependent.However, more research is needed to better understand these relationships.Lastly, we found that younger ApoE ε4 carriers accumulated more iron than non-carriers.This study opens the possibility of looking at each factor as a potential target in preventing brain iron overload in older age.
ETHICS STATEMENT.The present study, which involved human participants, was reviewed, and approved by the Regional Ethical Review Board in Stockholm.The participants provided their written informed consent to participate in this study prior to data collection.
J o u r n a l P r e -p r o o f

J
o u r n a l P r e -p r o o f questions assessed how often or for how long the participants engaged in different activity categories, including household work, work/occupation activity, walking/cycling, and leisure time inactivity (i.e.,

Figure 1 .
Figure 1.Graphical representation of models 1-3 used to assess the link between brain iron and the factors.1) A change regression model which estimates iron accumulation while adjusting for covariates in blue (age, sex, education, grey-matter (GM) volume change, and white-matter hyperintensities (WMH)).2) Model 2 examines the relationship between each factor (i.e., diet) and brain region (i.e., cortex).Further, for each factor we included interaction terms with age and sex (i.e., diet x age and diet x sex).3) An integrative model was constructed for each separate brain region in which all factors associated with the outcome were included into the combined model.Observed variables are indicated by boxes and latent factors are indicated by circles.Regressions are indicated with one-headed arrows and covariances with two-headed arrows.Covariates are indicated with blue boxes and dotted lines.
covariances from exploratory analysis of the composite score of blood iron, cardiovascular-health marker, and ApoE ε4 status.Reported values are standardised betas.*≤ 0.1 ** ≤ 0.05 *** ≤ 0.01.Main-effects are reported for the factor.Interactions were omitted from the model if non-significant.The directions of interactions are reported in the text.Estimated regressions and covariances from exploratory analysis of the composite score of blood iron, cardiovascular-health marker, and ApoE ε4 status.Reported values are standardised betas.*≤ 0.1 ** ≤ 0.05 *** ≤ 0.01.Main-effects are reported for the factor.Interactions were omitted from the model if non-significant.The directions of interactions are reported in the text.
covariances of the final models.Reported values are standardised betas.The directions of interactions are reported in the text.GM = Grey matter; WMH = White matter hyperintensities; LS7 = Life's Simple 7. Significant results are presented in bold and trends in italic.

Figure 2 .
Figure 2. A) Scatterplots of (A) Worsening of cardiovascular health increased iron accumulation, (B)greater iron content in basal ganglia and greater composite score of blood iron, (C) greater iron content in basal ganglia and higher MIND score according to age groups.D) Iron accumulation in basal ganglia according to ApoE ε4 status and age groups, with younger ε4 carriers accumulating more than younger non-carriers.

J
o u r n a l P r e -p r o o f limited the possibility to detect associations between health-related factors and brain iron content, as well as limit the generalizability of the results, which might not be applicable to adults with comorbidities.QSM-derived estimates are sensitive to confounding signals from diamagnetic elements such as myelin.Although we rigorously process our QSM images to minimize confounding signals and artefacts, there is still a possibility that our estimates might have been affected by myelin or other artefacts.Another limitation of the study is the number of time points.Although two time points allow us to estimate change in iron content, it does not allow for estimation of longitudinal trajectories or non-linear increase of iron accumulation.Following the participants over a long period and with additional data points would allow for better estimation of the trajectory of iron accumulation.

Table 1 .
Baseline participant characteristics for the total sample and three age groups.

Table 2 .
Results from measurement models for baseline iron and iron accumulation in each region of interest.

Table 3 .
Individual-factor models with brain iron in each region assessing the cross-sectional and longitudinal associations with the composite score of blood iron, LS7, and ApoE status.

Table 4 .
Individual-factor models with brain iron in each region assessing the cross-sectional and longitudinal associations with the lifestyle factors.Interactions were omitted from the model if non-significant.The directions of interactions are reported in the text.MIND = Diet; PA = Physical activity.

Table 5 .
Integrative models with brain iron for cortex and basal ganglia, and the significant factors individually associated with the iron outcomes at p < 0.1 were included.FDR corrected P-values are reported in brackets.