In developed countries, the proportion of the elderly population is increasing. Between 2021 and 2050, the ratio of Europeans aged above 65 should increase from 20.8% to 30%. In addition, the old age dependency ratio is projected to be 56.7% [European Community (EU), 2023]. This aging population is affected by multiple diseases and comorbidity such as hyperlipidemia, hypertension, diabetes, heart disease (Davis et al., 2011) and neurodegenerative diseases (Nichols et al., 2019).

The percentage of elderly people with an Alzheimer's disease (AD) increase with age. In the USA, it represents 5% of people aged between 65 and 74 y/o, and 13.1% of people aged between 75 and 84 y/o (Alzheimer's Association Report, 2023). In addition, Parkinson's disease (PD), the second most frequent neurodegenerative disease, represents 2% of people over 65 y/o. The major risk of neurodegenerative diseases, such as PD and AD, is aging, with only 5%–10% with an early onset before the age of 50. The development of new tools for the early diagnosis of neurodegenerative diseases remains a major challenge, as there is a lack of biomarkers for predicting brain disorders and mild cognitive impairment (Beach, 2017; Jeromin and Bowser, 2017), following up disease progression, supporting dementia affected people, or stratifying patients susceptible to respond to new therapies.

Iron is essential for the brain. In addition to its role in oxygen transport (hemoglobin), it is involved in several processes such as myelinization or neurotransmitters synthesis (Hare et al., 2013; Betts et al., 2016; Treit et al., 2021; Wang et al., 2017). Nevertheless, there is increasing evidence that iron accumulates heterogeneously in the brain throughout life and is involved in neurodegeneration. Indeed, an abnormal iron load in several brain regions, notably in deep gray matter (DGM) structures (such as putamen, caudate and globus pallidus), in older adults leads to cellular oxidative damages, inducing neuronal death (Ficiarà et al., 2022; Costello et al., 2004; Treit et al., 2021; Sousa et al., 2020). This iron-induced cell death, named ferroptosis, is the subject of intensive research (Dixon et al., 2012).

MRI is sensitive to iron and provides a means to quantify iron-related metrics, in vivo and in a non-invasive way (Ravanfar et al., 2021; Wang et al., 2022). Tissue (non heme) iron displays paramagnetic properties (Wood and Ghugre, 2008; Ropele and Langkammer, 2017) and, depending on its chemical form (e.g. ferrihydrite core of ferritin), can possess high magnetic susceptibility, which generates magnetic field perturbations (Schweser et al., 2016). MRI sequences can detect iron due to this physical property, allowing an estimation of its content in different tissues (Ravanfar et al., 2021). The effect of susceptibility in the tissue affects the apparent transverse relaxation time T2* and can be easily estimated by means of 3D Multi-echo Gradient Echo (MGRE) (Keuken et al., 2017; Ropele and Langkammer, 2017). T2* mapping analyzes the magnitude decay of the MRI signal. Interestingly, the phase of the signal also contains valuable information that reflects the tissue susceptibility effects by enabling to measure susceptibility-induced magnetic field deformations. This information is used by Quantitative Susceptibility Mapping (QSM) techniques to estimate a quantitative measure of the bulk susceptibility of a voxel (de Rochefort et al., 2009; Liu et al., 2012; Schweser et al., 2016; Ropele and Langkammer, 2017; Ruetten et al., 2019; Li Y. et al., 2021).

However, besides iron effects, the magnetic susceptibility-derived measures in the brain are also affected by some neuronal structures such as myelin, by calcification and by deoxyhemoglobin (Ficiarà et al., 2022). For instance, the presence of the myelin increase R2* but decreases the relative susceptibility (Ficiarà et al., 2022). In DGM structures, which are mainly affected by neurodegeneration, the susceptibility effects from myelin and other paramagnetic materials are minor, therefore the measures are mostly determined by the iron load content (Ficiarà et al., 2022; Ravanfar et al., 2021).

MRI acquisition parameters have to be set properly, as they affect the sensitivity and specificity of subsequent iron deposition measurements, for both R2* and for QSM (Wang et al., 2009). R2* quantification depends notably on the magnetic field strength, since faster relaxation rates are obtained at higher magnetic field strengths (Peters et al., 2007; Deistung et al., 2013; Ropele and Langkammer, 2017). However, QSM measurements are not majorly affected by the strength of the magnetic field (Li Y. et al., 2021; Nikparast et al., 2022). The magnetic property measured by QSM is inherent to the tissue and therefore independent of the field strength (Li et al., 2018; Li Y. et al., 2021).

Ultra-high magnetic field (UHF) strengths MRI (>3T) provide a higher signal-to-noise ratio (SNR). This allows the acquisition of images at a higher resolution within clinical-feasible scan times without the mitigation of the sensitivity (Vachha and Huang, 2021). However, these acquisitions can also be more vulnerable to artifacts or effects from strong field variations due to the air/tissue interface that need to be accounted for (Vachha and Huang, 2021; Wang et al., 2022; Daval-Frérot et al., 2022). Despite the challenge this presents, UHF has provided improved T2* contrast, which can be helpful in pathological situations (e.g. in differentiating histological types of cortical multiple sclerosis lesions) (Cohen-Adad et al., 2011).

Image resolution also plays a role in R2* and QSM values. High image resolution decreases partial volume effects, leading to a higher specificity (especially in the presence of small veins within a voxel) (Haacke et al., 2015). This is particularly important for compact structures such as small deep nuclei. In fact, slice thickness has been reported to reduce by 10% the mean susceptibility for small structures (Li Y. et al., 2021)

Several studies have analyzed iron load in vivo using R2* and/or QSM, either from the normal aging perspective or with a focus on neurodegenerative diseases. Higher QSM values for the regions implicated in diseases, especially DGM, have been reported (see Ravanfar et al., 2021 for an extensive review).

Iron accumulation with age in healthy subjects has also been reported previously, with the goal to describe normal reference values. This is of utmost importance for future applications in the evaluation of deviations in diseases by demonstrating higher values in several subcortical nuclei (Siemonsen et al., 2008; Li et al., 2014; Acosta-Cabronero et al., 2016; Treit et al., 2021). These studies differ either in terms of cohort characteristics (age range and size) or MRI acquisition parameters (magnetic field strength and image resolution, often larger than 1 mm isotropic). Moreover, a comprehensive description of the cohort's health condition is often absent (for instance, no information on risk factors of the subjects, despite their healthy cognitive state, is given). Also, these studies have been mainly performed at 3T, at rather low spatial resolution (for review, see Madden and Merenstein, 2023).

Slight higher values of iron load in DGM during old age have been described for QSM and R2* in lifespan studies (Li et al., 2014; Treit et al., 2021). This trend has also been described for young and middle-aged adults using QSM (Burgetova et al., 2021), as well as for elders (Gong et al., 2015; Li Y. et al., 2021; Li et al., 2023; Liu et al., 2016; Persson et al., 2015; Poynton et al., 2014). By means of R2* alone, correlations with age and higher values in certain regions of middle-aged adults and elder subjects have also been reported (Holz et al., 2022; Daugherty and Raz, 2016; Pirpamer et al., 2016).

QSM and Field-Dependent Relaxation Rate Increase (FDRI) have also shown higher iron concentrations in certain regions in older participants (Bilgic et al., 2012). In the spirit of documenting the differences in iron load measured by QSM, age-specific atlases have been proposed, reporting also higher values in specific regions (Lao et al., 2023; Zhang et al., 2018). Iron load differences through aging have also been described longitudinally for middle-aged and elder adults, evidencing a correlation between age and DGM when comparing two time-points (Li J. et al., 2021). Moreover, the consistency of QSM measurements over different vendor machines has been reported, whose results are also in agreement with the literature, showing a positive correlation with age (Li Y. et al., 2021).

Furthermore, although less often, 7T data has also been used to address this matter. At a higher image resolution [voxel size ≤ (0.8mm)³], the agreement between R2* and QSM, which describes age-related differences due to iron accumulation in subcortical regions, has also been reported (Betts et al., 2016; Keuken et al., 2017).

To date, there is no study using UHF that focuses exclusively on healthy cognitive aging from middle to very old age using a large, well-characterized cohort and accounting for common risk factors that are frequent in the aging population (such as hypertension, hyperlipidemia or diabetes).

In this work, the focus relies on quantifying QSM and R2* on healthy brain aging at old ages using UHF-MRI, leveraging on the well-characterized SENIOR database. The latter documents biological, psychological, and imaging data (including amyloid PET imaging) from healthy older adults. This allows for an accurate description of the population, confirming the absence of neurodegenerative diseases and limiting the sources of variation due to inherent risk factors.

Using the SENIOR database high-resolution imaging data acquired at 7T, R2* and QSM values are computed for the DGM. From these regions, we identified those that present differences: (i) regarding age, providing also their normal values; (ii) with respect to sex; (iii) related with an increased cardiovascular risk factor; and (iv) linked with overweight. The effect of the presence of ϵ4 allele of apolipoprotein E (ApoE) was also evaluated, as presumed genetic risk factor in the malfunction of brain iron homeostasis (Wood, 2015).

In this paper, we presented the study of the brain iron concentration differences in the normal aging by means of QSM and R2*, which serve as proxies for such differences, and using a well-characterized cohort with high quality data and a state-of-the-art reconstruction pipeline. The SENIOR cohort presents a rich database of particularly healthy and cognitive unimpaired elderly, documented by a complete neuropsychological evaluation and absence of amyloid deposits (on PET imaging). Medical history, blood pressure, clinical, biological and genetic susceptibility evaluations are also available, ensuring low or asymptomatic presence of main comorbidities, which are common to old age.

Due to the robustness of the pipeline, we were able to exploit the high quality MRI acquisitions at 7T, which provided a means to improve the description of the iron load differences in healthy cognitive aging. This along with the advanced and centralized image processing pipeline, which allowed us to improve the quality of the QSM and R2* maps, enabled us to establish a trend of normality for our cohort, according to age and sex for specific regions. For both methods, our results are consistent with the literature, specially for the regions that are known to particularly accumulate iron (i.e. putamen and caudate). In fact, the putamen is an iron-rich region whose iron deposition increase starts in the middle age, which has been evidenced also by means of susceptibility measurements (Khan et al., 2012). Linear regression models and correlations have been used to describe the relationship between QSM values and age for these regions for participants ranging 20–79 y/o (Acosta-Cabronero et al., 2016), between 20–90 y/o (Li Y. et al., 2021), 21–58 y/o (Burgetova et al., 2021), 25–78 y/o (Gong et al., 2015), 18–80 y/o (Lao et al., 2023), 50–80 y/o (Li J. et al., 2021), 10–70 y/o (Li et al., 2023), 20–69 y/o (Liu et al., 2016; Persson et al., 2015), 69–86 y/o (Poynton et al., 2014). When applying a linear regression model to our data, it can be observed that the slopes exhibit similar positive trends in terms of iron quantification values with respect to age. This can also be evidenced visually, from R2* and QSM images, as the voxels' intensity is higher, particularly for the putamen and caudate regions (see Figure 1).

Regarding the thalamus, a negative small slope has been described in the literature (Burgetova et al., 2021; Gong et al., 2015; Li Y. et al., 2021; Li J. et al., 2021), with which our results are also consistent. Notice that more similar values of slope are obtained when the age range of the cohorts are alike. In fact, the iron load changes at different rates through lifespan (Treit et al., 2021), therefore linear curve fitting should be performed by specific age ranges. For instance, it has been described that the thalamus presents an increase of iron load until the fourth decade of age, which then decreases (Burgetova et al., 2021). Moreover, results presented in our work suggest that for some regions (amygdala, thalamus and globus pallidus) there are differences between male and female participants, that should also be taken into account when computing models. From all the studied ROIs, all but the hippocampus exhibit differences, which are either age and/or sex related. This is why normality values should be compared considering these two variables.

In our analysis we also considered risk factors that are common in old age (such as diabetes, high blood pressure and dyslipidemia). Although our cohort is particularly healthy and only a few present certain comorbidities at some extent, we noticed that even if these are treated, differences can be observed in QSM and R2* values (e.g. for the putamen and globus pallidus when having at least 2 as risk factors score compared to 0, according to our score). Higher susceptibility values have been previously reported in the putamen and caudate in association with diabetes type 2, as well as in the thalamus for smokers (Li J. et al., 2021). Nevertheless, the relationship between cardiovascular risks and iron accumulation is not yet fully understood. Some hypotheses point to an iron accumulation due to microvascular hemorrhages (Li et al., 2020). Moreover, some suggest that brain iron is highly modulated by the diet (Hagemeier et al., 2015). BMI has also been previously reported to be related with higher iron levels measured by means of R2* in the caudate, putamen and globus pallidus (Holz et al., 2022) and the amygdala and hippocampus (Pirpamer et al., 2016). From our study we evidenced this difference particularly in the putamen of female participants by means of both QSM and R2*.

In general, the obtained values for QSM and R2* present a high variability (as it can be observed in the standard errors from the linear regressions). These variations can stem from a variety of sources, such as the inherent variability in the physiology of each individual subject, or coming from the acquisition parameters. In fact, the choice of imaging parameters can impact the quality of the measure, for instance partial volume effects due to a low resolution can have a great impact specially when measuring small structures as the DGM. Nevertheless, from our results there is a high correspondence between values obtained from QSM and R2*, which has been already described in the literature (Feng et al., 2018; Peterson et al., 2019; Deistung et al., 2013; Ghassaban et al., 2019). This confirms that the reconstructed QSM maps through our robust pipeline, although small singularities can still be noted specially near the sinuses, behaves as expected in regard of R2*, particularly for the DGM measurements. The reliability of the QSM reconstructions was also confirmed regarding contralateral structures, specially for those located in areas more prone to singularities (Supplementary Section 3.2). A great advantage of our pipeline is that it can be applied to any other database with MGRE acquisitions at its disposal.

As it is described by Wang et al. (2020), R2* relates to compartmentalized “inclusions” with a susceptibility offset compared to the surrounding tissue, which broaden the distribution of frequencies within a voxel. The GRE signal phase is driven by an average magnetic susceptibility in a voxel, which relates to the relative size of each susceptibility-shifted compartment. Thus, on one hand, the QSM and R2* contrasts rely essentially on the phenomenon, on the other hand they use different information and reconstructions to be extracted. From this statement, it is not surprising to see QSM and R2* techniques exhibit a similar behavior for most of the regions and a high correlation between these two proxies have been previously documented (Feng et al., 2018; Ghassaban et al., 2019).

Our results are in line with this concordance between QSM and R2*. In addition, we evidenced that they could also provide complementary information, as differences could be reflected by one or the other proxy. In fact, our results showed differences between these two techniques regarding the sex-based analysis. While we were able to detect sex-related differences for some regions in QSM and R2*, these regions differ between them. For QSM, differences were found for the thalamus and amygdala regions, as for R2* these were found for the globus pallidus. Moreover, we evidenced the joint effect of age and sex in regions such as the accumbens and hippocampus, suggesting that these two factors should be taken into account when performing a detailed analysis of the iron profile by means of both QSM and R2*. Sex-based differences in the brain iron load are still a source of debate. Sex-related differences have been already described in the literature for young subjects, from R2*, signaling lower iron deposition in boys than girls for several brain regions (Peterson et al., 2019). A lower ferritin dependence has been reported in women in regions such as the caudate and the thalamus (Bartzokis et al., 2007). Evidence of this sex-based differences has been highlighted by means of QSM for the thalamus (Gong et al., 2015), putamen, red nucleus and substantia nigra (Persson et al., 2015; Li et al., 2023), for which female subjects present lower susceptibility values. On the other hand, some studies have also reported no influence of sex in susceptibility values (Gong et al., 2015; Li J. et al., 2021; Holz et al., 2022; Li et al., 2015).

The underlying reasons of sex-based iron differences are not yet clear. Our data showed that hematocrit is higher in male than in females. As we noted both higher or lower QSM values in females, our data can not elucidate any relationship between systemic iron and brain iron measured by QSM. Previous studies often point to hormonal differences due to estrogen and to menstruation, as differences are related with menarche and menopause (Grubić Kezele and Ćurko-Cofek, 2020; Larsen et al., 2023).

Regarding our genetic analysis, we only included the presence of the ApoE ϵ4 allele, as it is the only one for which there is consensus as a risk factor in developing Alzheimer's disease. Therefore, it is interesting evaluating its influence in the predisposition in iron accumulation in healthy subjects. We evidenced that the presence of ϵ4 allele does not modify the iron accumulation in the DGM, which has also been previously reported (Li J. et al., 2021). Nevertheless, higher susceptibility has also been reported in the hippocampus and amygdala for carriers of ApoE ϵ4 allele, however this is also the case for ApoE ϵ2 carriers, particularly those under 65 y/o (Nir et al., 2022).

Regardless of the high quality of the data and the robustness of the pipeline, some limitations were present during our study. First, due to an MRI protocol update the sample size was reduced to a fraction of the total cohort, which limits the general population representation in order to establish a norm. This also limited our study to a cross-sectional analysis and prevented us from performing a longitudinal one at the current date. New time points are currently being acquired, therefore future research using our cohort will be centered in investing brain iron accumulation using a longitudinal design. Additionally, in our cohort, only some participants present an ApoE ϵ4 allele, which affects the statistical power of our analysis in this matter. Moreover, only one participant was homozygous for this allele. Small subgroup size presents also a limitation when performing statistical analysis for the participants with a high CRS, as our cohort is particularly healthy with only a few presenting comorbidities linked with an increase of cardiovascular risk. Another limitation regarding the CRS is the available information to be included when calculating the score. For instance, it might also be relevant to include information regarding the physical state and diet (Sacco, 2011). Moreover, our cohort is a particular case of a very healthy elderly population, which does not necessarily reflect the general population of older adults, therefore the results should be interpreted accordingly. Although the participants are biologically well-characterized, some parameters of interest such as serum ferritin and transferrin were not available. Its inclusion in future analysis should be considered, as R2* values in some DGM regions have been described to be correlated with serum transferrin in older subjects with no cognitive impairment (House et al., 2010). It is also important to note that quantitative results are difficult to directly compare between studies, due to the inherent differences in QSM and R2* maps that can arise from multi-centric bias (due to the machine, acquisition parameters and/or pipelines used). In this case, only the trend of the values extracted from the maps, as well as the differences between groups given a parameter, should be compared.

Moreover, due to our multiple statistical test, in order to reduce the type I risk (false positive test) we performed a Bonferroni correction. This led to decreased statistical power, as many of the differences caught by the tests do not pass this correction. Furthermore, it is worth keeping in mind that this type of correction can also increase type II risk (false negative test). Notice also the drastic reduction of the difference's significance by the Bonferroni correction might be also due the rather subtle variation of the factors evaluated, especially those related to cardiovascular risks given the healthy status of our cohort and the control of comorbidity by medical treatments, such as for arterial hypertension or diabetes.

Furthermore, this work was focused only in the study of most of the DGM, as many of these structures have been reported of having a particular incidence in specific pathologies. Even if other regions, such as the substantia nigra, would have been of major interest to segment and to study, they were not included in this work due to limitations in the segmentation tool used, which did not include this feature. Other available solutions require higher spatial resolutions (Manjón et al., 2020) or multi-contrasts (Langley et al., 2015; Xiao et al., 2012) that were not acquired in SENIOR cohort. Also, the gray matter arranged in the cortex has also been described to play a role in this matter. Susceptibility differences between controls and patients have been reported in the cortex, either in specific regions (e.g. in Parkinson's disease) or in a diffuse way all over it (e.g. in multiple sclerosis) (Ravanfar et al., 2021; Cohen-Adad et al., 2011). Despite the fact that cortical susceptibility information is essential for the better understating of these pathologies, QSM cortical reconstructions are still a challenge, essentially due to the B₀ field inhomogeneities (Cohen-Adad et al., 2011). Despite the limitations found in the input phase data (see Supplementary Figure S2), thanks to the developed QSM computation pipeline it was possible to reconstruct the values for the deep gray matter structures. However, cortex values are less reliable and often cropped out of the QSM maps, specially close to the brain boundaries. Additional phase pre-processing would be needed for mitigating the B₀ inhomogeneities effects in these boundary regions.

The developed pipeline allowed the addition of the iron load, measured by two proxies (QSM and R2*), as a biomarker to the SENIOR database. This made possible the establishment of a range of normal values during healthy aging for our cohort and evaluate if common comorbidities have an impact in the measurements. This allowed the delineation of a general trend for each region, reflecting the iron level changes during aging. Regardless of the inherent batch variability across studies mentioned before, these general trends could be useful for comparison purposes for other studies. Moreover, as some previous studies have also done, we show that iron load values should be evaluated according to the age and sex of the participants, as these two variables play a major role in the differences that are found between them. In addition to this information, we also provide insights regarding other parameters that should be looked into (CRS and BMI). High quality data and the use of QSM and R2* allow the detection of differences in the iron load that are proven to be significant, even though these are subtle. Notice that even if several of our results are in line with the literature, no gold standard values for either R2* nor QSM have been established. Therefore, results should be interpreted carefully, specially keeping in mind the variability of these two proxies. Following work will be focused in including in the study participants presenting Alzheimer's disease to assess the ability to detect differences with respect to our healthy participants. Moreover, as new time-points are being added to the database, a next step will be moving to a longitudinal study design.