Quantitative Susceptibility Mapping of Brain Iron in Healthy Aging and Cognition

Abstract

Quantitative susceptibility mapping (QSM) is a magnetic resonance imaging (MRI) technique that can assess the magnetic properties of cerebral iron in vivo. Although brain iron is necessary for basic neurobiological functions, excess iron content disrupts homeostasis, leads to oxidative stress, and ultimately contributes to neurodegenerative disease. However, some degree of elevated brain iron is present even among healthy older adults. To better understand the topographical pattern of iron accumulation and its relation to cognitive aging, we conducted a systematic review of 47 QSM studies of healthy aging, with a focus on five distinct themes. The first two themes focused on age-related increases in iron accumulation in deep gray matter nuclei versus the cortex. The overall level of iron is higher in deep gray matter nuclei than in cortical regions. Deep gray matter nuclei vary with regard to age-related effects, which are most prominent in the putamen, and age-related deposition of iron is also observed in frontal, temporal, and parietal cortical regions during healthy aging. The third theme focused on the behavioral relevance of iron content and indicated that higher iron in both deep gray matter and cortical regions was related to decline in fluid (speed-dependent) cognition. A handful of multimodal studies, reviewed in the fourth theme, suggest that iron interacts with imaging measures of brain function, white matter degradation, and the accumulation of neuropathologies. The final theme concerning modifiers of brain iron pointed to potential roles of cardiovascular, dietary, and genetic factors. Although QSM is a relatively recent tool for assessing cerebral iron accumulation, it has significant promise for contributing new insights into healthy neurocognitive aging.

1. Introduction

In vertebrates, iron exists in two distinct forms: heme iron, which is linked exclusively to circulating or accumulating blood, and non-heme iron, which is present in virtually all cells and is a contributor to essential biological processes in the brain such as oxygen transport, DNA synthesis, mitochondrial respiration, myelin synthesis, and neurotransmitter synthesis and metabolism (Gutteridge, 1992; Hentze et al., 2004; Koeppen, 1995 ; Rouault and Cooperman, 2006; Todorich et al., 2009). Early anatomical studies of post-mortem brain tissue reported that iron deposition was notable in deep gray matter regions related to motor control, particularly the globus pallidus, caudate, putamen, and substantia nigra, and that the amount of iron tended to increase from childhood to adolescence (Spatz, 1922). In a seminal study, Hallgren and Sourander (1958), conducted the first systematic analyses of age-related differences in iron across different regions of the brain. These authors conducted histological analyses of post-mortem tissue for 98 brains from individuals spanning infancy to 100 years of age. Hallgren and Sourander confirmed that non-heme iron concentration was particularly prominent in the deep gray matter regions, relative to cortical or white matter regions.

Elevated iron concentration in deep gray matter nuclei is associated with neurodegenerative disease, notably Parkinson’s disease, in which movement disorder is prominent (Dexter et al., 1987; Gerlach et al., 1994; Ghassaban et al., 2019; Gotz et al., 2004; Ke and Qian, 2003; Sayre et al., 2000). Elevated brain iron may also have a role in dementia. As early as 1953, postmortem histochemical analyses suggested that a disturbance in the cerebral metabolism of iron was an aspect of Alzheimer’s disease (AD) pathogenesis (Goodman, 1953). Connor et al. (1992), in a postmortem examination of regional tissue distribution of iron and iron-regulatory proteins, using immunoassay, concluded that alterations in iron-regulatory proteins are exacerbated in AD. In postmortem studies of AD patients, tissue histology has demonstrated that neuronal iron accumulation co-localizes with Aβ deposition and tau neurofibrillary tangles (Duce et al., 2010; Grundke-Iqbal et al., 1990; Lovell et al., 1998; Smith et al., 1997). Critically, however, the increased deposition of brain iron during later adulthood, observed by Hallgren and Sourander (1958), occurs in the absence of specific neurodegenerative disease and is one feature of the complex constellation of changes in central nervous system function typically associated with normal human aging (Martin et al., 1998; Sfera et al., 2018; Ward et al., 2014; Zecca et al., 2004).

The development and application of neuroimaging techniques, particularly magnetic resonance imaging (MRI), can provide a complementary perspective to the histological studies, by characterizing the properties of iron in the human brain in vivo as well as ex vivo. Thus, while a substantial literature exists, from both structural and functional MRI, regarding changes in the brain during later adulthood (Dennis and Cabeza, 2008; Fjell and Walhovd, 2010; Grady, 2012; Raz et al., 2010), relatively few studies have addressed the role iron deposition specifically. Our goal in this article is to review the contributions that different MRI methods, particularly quantitative susceptibility mapping (QSM), have made to understanding age-related differences in iron deposition and the relation of iron to cognitive aging, in healthy individuals. A central question that we are addressing is whether the age-related increase in brain iron, in otherwise healthy adults, contributes to the declines in some aspects of cognitive functioning that are observed during later adulthood.

Specialized structural MRI sequences, such as multiecho gradient echo sequences, are particularly informative regarding iron because variations in iron create local differences in magnetic susceptibility (Haacke et al., 2005; Liu et al., 2015a), and thus regions with increased iron also have higher susceptibility and a relatively fast transverse relaxation rate (R2) or short relaxation time constant (T2*). In these studies, the most frequently used quantitative MRI index of magnetic susceptibility is the relaxometry-based measure of R2*, which estimates iron content as a sum of relaxation due to spin-spin interaction (R2) and local susceptibility effects (R2’) (Brass et al., 2006; Langkammer et al., 2010). Although relaxometry-based estimates of iron correlate highly with chemically determined iron concentration obtained postmortem (Brass et al., 2006; Langkammer et al., 2010), the R2* index and related relaxometry measures are potentially influenced by background field inhomogeneity unrelated to iron content (Haacke et al., 2015; Wang and Liu, 2015). Further, R2* can be affected by either paramagnetic (e.g., iron) or diamagnetic sources.

QSM is another quantitative MRI index of magnetic susceptibility that has several advantages over relaxometry-based measures (Langkammer et al., 2012; Li et al., 2011; Liu et al., 2011; Liu et al., 2015a). QSM is a more direct measurement of the intrinsic property of tissue and is independent of magnetic field strength. As compared to relaxometry, QSM has improved contrast of deep gray matter nuclei (Barbosa et al., 2015; Liu et al., 2013) and better sensitivity to the effects of age (Bilgic et al., 2016; Li et al., 2014) and neurodegenerative disease (He et al., 2015). For example, different regional patterns of iron deposition, as assessed by QSM, are expressed in AD (Acosta-Cabronero et al., 2013; Ayton et al., 2013) and cerebrovascular small vessel disease (Moon et al., 2016; Sun et al., 2017). Our primary goal here is to consider the degree to which QSM may shed light on age-related differences in cognition during healthy aging. In particular, does brain iron contribute to cognitive aging, or alternatively, are the effects of age and iron on cognition independent, until some threshold of neuropathology is crossed?

2. Methodological Considerations for QSM Studies of Aging

QSM is primarily sensitive to iron content in the case of gray matter and myelin content in the case of white matter, corresponding to positive (paramagnetic) and negative (diamagnetic) susceptibility values, respectively (Deh et al., 2018; Li et al., 2012; Li et al., 2011; Liu et al., 2015b). Thus, throughout the articles discussed here, we consider positive versus negative susceptibility values to correspond primarily to the relative contribution of iron versus myelin. The neurobiological source of magnetic susceptibility, however, is complex and not a one-to-one correspondence (Liu et al., 2015b). For example, whereas iron deposition is highest within deep gray matter nuclei, contributing to positive susceptibility values, these regions are also myelinated, though to a lesser degree than cortical neurons. Similarly, the susceptibility of white matter is determined primarily by myelin concentration, but iron is also present, in the oligodendrocytes forming the myelin and in the mitochondria in the axons (Meguro et al., 2008). As a result, average susceptibility values within a region of interest represent different signal sources. Although positive susceptibility is likely dominated by iron, the negative values are more difficult to interpret. Methods for addressing this difficulty are still in development. One approach is to use the absolute value (i.e., unsigned) combination of positive and negative values in separate maps of gray matter and white matter (Betts et al., 2016). A second approach is to use biophysical modeling to separate the voxelwise distributions of paramagnetic (e.g., iron) and diamagnetic (e.g., myelin) susceptibility signals based on the frequency shift and transverse relaxation rates (Shin et al., 2021). A third algorithm, termed DECOMPOSE-QSM, uses the phase and magnitude from a gradient echo acquisition sequence to separately estimate paramagnetic susceptibility, diamagnetic susceptibility, and reference susceptibility within each voxel (Chen et al., 2021a), which has recently been applied to study neurodegeneration in AD (Ahmed et al., 2023).

QSM values, derived from the phase measured by gradient echo sequences, are influenced by the scan acquisition parameters, especially echo time (Sood et al., 2017). For example, the myelin water signal cannot be detected by later echo times (Liu et al., 2015b). For these reasons, it is preferable to use multiple echo times when acquiring a gradient echo sequence. Although it is possible to acquire single echo gradient echo sequences, they are much more prone to streaking artifacts (Liu et al., 2015b). After acquiring a gradient echo dataset, researchers also have several options for unwrapping the phase image, removing the background field, and ultimately constructing the susceptibility map (Ravanfar et al., 2021), such as the morphology enabled dipole inversion (Liu et al., 2012) or sparse linear equation and least-squares algorithm (Li et al., 2015b).

Researchers must also decide whether to use relative susceptibility values or to reference susceptibility values against a particular reference region. Although most researchers agree that susceptibility values should be referenced against a control region, there is considerable debate regarding the appropriate reference region (e.g., ventricles or white matter). This is a particularly important issue in aging, where the accuracy of tissue segmentation (e.g., avoiding iron-rich choroid plexus in the ventricles) and age-related differences in underlying tissue properties (e.g., degree of myelination, presence of lesions) can contribute to differences in susceptibility (Deistung et al., 2017; Ravanfar et al., 2021). However, prior studies have shown that the effect of referencing did not change the observed age-related differences in QSM values (Acosta-Cabronero et al., 2016; Li et al., 2014).

Variability among QSM studies in aging may also reflect differences in the analytical approach. Some studies use a region-of-interest approach, which can be conducted at the participant level, whereas others use a voxelwise approach, which can identify smaller clusters but requires group-level registration and some degree of spatial smoothing. Voxelwise analyses of cortical regions are particularly vulnerable to biased susceptibility measures in blood vessels, lesions, and regions near the air-tissue interface (Chen et al., 2021a; Liu et al., 2015b). One approach to address artifactual sources of susceptibility, especially around the edges of the brain and near the air-tissue interface, is to erode the edges of the susceptibility maps (Bhattarai et al., 2020; Howard et al., 2022). Another approach to reduce the number of artifactual susceptibility values is to threshold and remove the most extreme 15% of values (Garzón et al., 2017; Persson et al., 2020).

3. Scope of Review

We conducted an integrative literature review between March and April 2023 with PubMed searches using two search terms related to age (ag*ing; older adults) and three related to QSM (QSM; Quantitative Susceptibility Mapping; magnetic susceptibility). We conducted six searches, representing each combination of one age term combined via AND with one QSM term. These search queries identified a total of 114 unique publications. From these publications, we selected studies that met all of the following five criteria: 1) appeared in an English language journal, through 2023; 2) reported in vivo MRI scans of the human brain; 3) was an empirical research report (i.e., reviews were excluded); 4) included cognitively healthy human participants over 60 years of age; and 5) examined relations between a QSM measure of the brain and either age, cognitive performance, or other MRI measures (when limited to a sample of healthy older adults). Studies that examined relations between aging and R2*-based MRI measures were only included in the current review if these measures were examined in combination with QSM-based MRI measures.

Of the 114 originally identified publications, we excluded 20 for not including cognitively healthy adults older than 60 years of age, 13 for not being original research reports, six for not including human participants, one ex vivo study, and 27 that did not report some relation between a QSM measure of the brain and either age, cognitive performance, or another MRI measure within healthy older adults. Thus, a total of 47 published articles (identified with an asterisk in References) from the original search queries met the criteria for the current review. Eight of these involved a comparison of patients (e.g., Parkinson’s disease, AD, multiple sclerosis) and healthy controls, and in these instances, we report only the findings of the healthy participants, as our focus here is on cognitive aging in healthy adults. We discuss the 47 included studies in terms of five themes: deep gray matter susceptibility; cortical susceptibility; the relation of susceptibility to neurocognitive function; multimodal imaging studies; and moderators of susceptibility in healthy aging. The studies associated with each theme, and the various forms of evidence supporting the themes, are presented in Tables 1–5. Note that findings from an individual study may contribute to more than one theme.

Table 1:

Age-Related Differences in QSM for Deep Gray Matter Regions

First author (year)	N	Age	CN	PT	SN	GP	RN	TH	DN	STN
Li et al. (2023)	220	10-70	+	++	+	+	+		+
Gong et al. (2015)	42	25-78	+	++	+	+	+	0
Zhou et al. (2020)	213	43-80	+	+	+	+	+	+	+
Acosta-Cabronero et al. (2016)	116	20-79	+	+	+	0	+	0	+
Betts et al. (2016)	20	22-28
	20	64-75	+	+	+	0	+	+	+	+
Guan et al. (2022)	100	M = 24
	189	M = 61	+	+	+	+	+	+
Li et al. (2014)	191	1-83	+	+	+	+	+		+
Ghassaban et al. (2019)	24	M = 63	+	+	+	0	+	0	0
Liu et al. (2016	174	20-69	+	+	+	0	+	+
Chen et al. (2013)	27	60-82	+	+		+		+
Kan et al. (2020)	17	M = 28
	19	M = 71	+	+	+	+	+		+
Keuken et al. (2017)	64	19-75	+	+	+	0	+			0
Persson et al. (2015)	183	21-69	+	+		+	0	+	0	0
Reeves et al. (2022)	170	9-81	+		+		+
Taege et al. (2019)	33	M = 52	+	+			+	+	−
Treit et al. (2021)	498	5-90	+	+			+		−
Zachariou et al. (2022)	35	66-86	+	+			+		+
Kalpouzos et al. (2021)	208	20-79	+	+			+
Persson et al. (2020)	21	M = 36
	15	M = 70	+	+			+
Moon et al. (2016)	18	M = 47	+	+			0		0
Zachariou et al. (2021)	73	61-86	+	+			0
Xu et al. (2008)	78	22-78	0	+	0		0	+	0
Li et al. (2015a)	132	43-80	0	+	0		0	0		0
Poynton et al. (2015)	11	21-29
	12	64-86	0	+		+	+	+	−	0
van Bergen et al. (2018a)	80	55-96	0	+			0		0
Bilgic et al. (2012)	11	21-29
	12	64-86	0		+	+	+	+	0	0
Howard et al. (2022)	67	18-78	0	+	0	0	+	+	0	0
Chiang et al. (2022)	32	50-87	0	0	0
Li et al. (2021)	44	M = 59	Δ	Δ	Δ	Δ	Δ	0	Δ
Gustavsson et al. (2022)	208	20-79	Δ	Δ
Garzon et al. (2018)	22	26-42
	18	65-77			+		+			+
Berman et al. (2022)	42	18-30
	25	61-70			0		0
Liu et al. (2015c)	30	< 45
	17	45-59
	20	60+						+
Zhang et al. (2018)	166	1-83						−
Jäschke et al. (2023)	154	17-78							+

Note. QSM = quantitative susceptibility mapping. Age = range or mean years of age of sample.

For studies with two or more age groups, age group differences are presented in the cells for the oldest group. For each region, symbols indicate observations of increases (+), decreases (−), or non-significant differences (0) in susceptibility as a function of increasing age. Duplicate symbols (++) indicate that the magnitude of the age effect was larger for that region relative to other regions examined. Blank cells indicate that the region was not examined in the corresponding study. Studies are sorted by age effects in the caudate and putamen.

CN = caudate nucleus; PT = putamen; SN = substantia nigra; GP = globus pallidus (combining internal and external limbs); RN = red nucleus; TH = thalamus; DN = dentate nucleus of the cerebellum; STN = subthalamic nucleus; Δ = longitudinal change.

Table 5.

Moderators of QSM in Aging

				QSM region
First author (year)	N	Age	Moderator	FC	TC	PC	OC	CN	PT	GP	RN	SN	TH	HC	AM
Li et al. (2023)	220	10-70	Female sex						+		0
Acosta-Cabronero et al. (2016)	116	2O-79	Female sex	0	0	0	0	0	0	0	0	0	0	0	0
Persson et al. (2015)	183	21-69	Female sex					0	0	0	0	−	0
Treit et al. (2021)	498	5-90	Female sex					0	0	0			0
Gong et al. (2015)	42	25-78	Female sex					0	0	0	−	0	−
Xu et al. (2008)	78	22-78	Female sex					0	0	0	0	0	0
Li et al. (2021)	44	M = 59	Female sex					0	0	0	0	0	0
			Smoking					0	0	0	0	0	+
			Diabetes					+	+	0	+	0	0
			Hypertension					0	0	0	−	0	0
Zachariou et al. (2021)	73	61-86	Healthy diet			−			−
Acosta-Cabronero et al. (2016)	116	20-79	Hypertension	0	0	0	0	0	0	0	0	0	0	0	0
Nir et al. (2022)	27,535	44-80	APOE ε4											+	+
Kalpouzos et al. (2021)	208	20-79	HFE	0	0	0	0	0	+	0
Elliott et al. (2018)	8,428	40-69	HFE					+	+	+
			TF					+	+	+
			SLC25A37					+	+	+
Gustavsson et al. (2022)	208	20-79	COMT	+				+	+

Note. Age = range or mean years of age of sample.

For each study, symbols indicate positive (+), negative (−), or nonsignificant (0) effects of the moderating variable on susceptibility values for a given region. Studies are sorted by the moderator that was examined (sex, lifestyle or health-related factor, genotype). Blank cells indicate that the region was not examined in the corresponding study.

FC = frontal cortex, TC = temporal cortex, PC = parietal cortex, OC = occipital cortex, CN = caudate nucleus, PT = putamen, GP = globus pallidus, RN = red nucleus, SN = substantia nigra, TH = thalamus, HC = hippocampus, AM = amygdala. APOE ε4, HFE, TF, SLC25A37, COMT = individual genes.

4. Deep Gray Matter Susceptibility Patterns in Healthy Aging

As noted previously (Section 1, Introduction), the histology data of Hallgren and Sourander (1958) demonstrated that the concentration of iron was higher for deep gray matter regions (4.76 – 21.30 mg/100 g tissue) than for cortical regions (2.92 – 5.03 mg/100 g tissue). These authors also observed, however, that within the deep gray matter regions, age-related differences in iron were independent of the overall level of iron. The globus pallidus (along with the substantia nigra and red nucleus) exhibited the highest concentration of iron overall, but iron in the globus pallidus did not increase markedly following 30 years of age. In contrast, iron content in the putamen and caudate nucleus, while lower overall, continued to increase beyond 50–60 years of age.

One theme of QSM studies (Table 1) is that susceptibility varies in relation to deep gray matter region, and age, in a manner consistent with the Hallgren and Sourander (1958) histological data. For example, Gong et al. (2015) reported that susceptibility values were highest for the globus pallidus, substantia nigra, and red nucleus, consistent with Hallgren and Sourander. Similarly, Gong et al. reported that the magnitude of the age-related increase in susceptibility was greater for the putamen than for other deep gray matter regions. These authors also distinguished left and right hemisphere components of gray matter nuclei and found that susceptibility was relatively higher for the left side of the caudate and substantia nigra. This hemispheric asymmetry may reflect dopamine levels associated with lateralized motor function (Xu et al., 2008), although this hemispheric effect appeared to be independent of the age-related effects in susceptibility.

In their QSM study, Li et al. (2023) investigated six deep gray matter nuclei, for 220 individuals 10–70 years of age. In addition to susceptibility, these authors analyzed estimated iron content, for each deep gray matter region, from the multiplicative product of susceptibility and regional volume (adjusted for total intracranial volume). For both susceptibility and iron content, the putamen exhibited the most pronounced increase with age. Li et al. noted that whereas susceptibility increased with age for all deep gray matter regions, individual age-related trends could be either linear (substantia nigra), quadratic (putamen, caudate, and globus pallidus), or exponential (red nucleus, dentate nucleus). However, Li et al. did not directly compare the deep gray matter regions in terms of the absolute value of either susceptibility or iron content.

Whereas QSM studies reliably confirm the age-related increase in iron in the putamen, findings for other gray matter regions, particularly the thalamus, are mixed. Zhou et al. (2020) reported a statistically significant increase in thalamic susceptibility with age, consistent what they observed for other deep gray matter regions, but the effect size was small, r = 0.164. Gong et al. (2015) reported that the thalamus did not show any significant age-related effect, in contrast to the age-related increase in susceptibility for other deep gray matter regions. Taege et al. (2019) and Treit et al. (2021) reported significant age-related declines in susceptibility for the thalamus.

Several functional and structural properties of the thalamus may contribute to this variation in age-related effects. Deep gray matter nuclei are primarily associated with motor functioning, but the thalamus comprises heterogeneous sub-nuclei (e.g., pulvinar) implicated in sensory and cognitive functions, especially visual attention (LaBerge, 2000; LaBerge and Buchsbaum, 1990). The thalamus is relatively high in myelin, within the internal and external inter-medullary lamina, and the combination of these different sources of paramagnetic and diamagnetic signals may contribute to variability (Betts et al., 2016). Finally, it is important to note that the majority of the studies to date are cross-sectional, and thus age-related differences are necessarily confounded with individual differences. Although longitudinal studies have confirmed age-related increase in deep gray matter susceptibility (Gustavsson et al., 2022; Li et al., 2021), additional exploration of regional longitudinal trends is needed.

5. Cortical Susceptibility Patterns in Healthy Aging

The majority of the studies in Table 1 used a region of interest approach that focused exclusively on susceptibility values that are averaged across voxels from anatomically defined deep gray matter regions. Although some studies compared deep gray matter and selected supratentorial cortical regions (Gustavsson et al., 2022; Li et al., 2014), the majority focused on the deep gray matter regions. That is a logical and necessary first step, given the relatively lower levels of iron observed in cortex relative to deep gray matter nuclei in the absence of disease, as noted previously. However, in view of the anatomical connections between deep gray matter and cortical regions, and the role of deep gray matter regions in the coordination of sensorimotor functions (Alexander et al., 1986; Cummings, 1993; Graybiel and Saka, 2004; LaBerge, 2000), the effects of iron deposition in cortical regions other than deep gray matter are also important. The articles included in Table 2 illustrate a second theme in QSM research, the age-related variation in susceptibility across cortical regions, for healthy adults.

Table 2:

Age-Related Differences in QSM for Cortical, Limbic, and White Matter Regions

			QSM Region
First author (year)	N	Age	FC	TC	PC	OC	HC	AM	WM
Acosta-Cabronero et al. (2016)	116	20-79	++	+	++	0	0	0	0
Betts et al. (2016)	42	18-30
	25	61-70	++	+	+	+	+	+	+
Howard et al. (2022)	67	18-78	+	+	+		0	0
Kalpouzos et al. (2021)	208	20-79	+	+	+	+
Zachariou et al. (2023)	95	60-86	+	+	+	+
Zachariou et al. (2021)	73	61-86	+	0	+	0
Li et al. (2014)	191	1-83	+		+		+
Chiang et al. (2022)	32	50-87	0	0	0		0	0
van Bergen et al. (2018a)	80	55-96	−	−	−		0	0
Gustavsson et al. (2022)	208	20-79	Δ
Bilgic et al. (2012)	11	21-29
	12	64-86					0
Xu et al. (2008)	78	22-78					+
Koskimäki et al. (2020)	7	18-49
	11	50-79							+

Note. QSM = quantitative susceptibility mapping. Age = range or mean years of age of sample.

For studies with two or more age groups, age group differences are presented in the cells for the oldest group. For each region, symbols indicate observations of increases (+), decreases (−), or non-significant differences (0) in susceptibility as a function of increasing age. Duplicate symbols (++) indicate that the magnitude of the age effect was larger for that region relative to other regions examined. Blank cells indicate that the region was not examined in the corresponding study. Studies are sorted by age effects in the frontal lobe.

FC = frontal cortex; TC = temporal cortex; PC = parietal cortex; OC = occipital cortex; HC = hippocampus; AM = amygdala; WM = white matter; Δ = longitudinal change.

The application of voxelwise, whole-brain QSM measures has been a significant methodological advance in research on cortical susceptibility and aging. Two articles in 2016 reported whole-brain QSM patterns. Betts et al. (2016) conducted a voxelwise QSM analysis of 20 younger adults and 20 older adults at 7T, with the additional feature of constructing separate maps for positive and negative susceptibility. These authors observed that increased susceptibility for the older adult group was evident in deep gray matter regions, as expected, but also in supratentorial cortical regions, particularly superior frontal regions surrounding primary motor cortex. In addition, the clusters of negative susceptibility, primarily in white matter tracts, tended to be more negative (i.e., increasingly diamagnetic) for older adults relative to younger adults. Similarly, Acosta-Cabronero et al. (2016) conducted a voxelwise QSM analysis (at 3T) of 116 individuals 20–79 years of age, and because they sampled age as a continuous variable, they could define clusters of interest from the age-susceptibility correlation, rather than from a group contrast as in Betts et al. (2016).

In addition to confirming the strong age-related trends for increased susceptibility of deep gray matter, Acosta-Cabronero et al. (2016) observed significant age-related increases in susceptibility in sensorimotor cortex and prefrontal, insular, and dorsomedial frontal cortex, consistent with the Betts et al. (2016) findings. Acosta-Cabronero et al. found that age-related susceptibility effects in white matter tended to be positive, in contrast to Betts et al., though Acosta-Cabronero et al. did not separate positive and negative QSM maps as Betts et al. had done. Acosta-Cabronero et al. found that cortex rostral to the central sulcus (motor, premotor, dorsal prefrontal, dorsomedial surface, and insula) was more prone to iron accumulation with age than more posterior cortical regions, leading them to propose that the motor system, broadly defined, has a tendency to accumulate iron with age.

Thus, in contrast to region of interest analyses, voxelwise analyses provide a more comprehensive view of iron deposition across the whole brain and have yielded novel findings. The voxelwise approach, however, has several features that should be considered when interpreting these results. The threshold for cluster significance in a voxelwise analysis is based on the contrast or correlation with an independent variable, for example, age, disease status, or a behavioral outcome, and thus the clusters of interest will vary in relation to threshold definition. In addition, unless gray and white matter tissue compartments are separated, the paramagnetic and diamagnetic components of the QSM outcome variable will combine in their contributions to cluster definition, which complicates interpretation. Finally, voxelwise analyses are inherently conservative by correcting for the multiple comparisons made across the population of voxels, and these analyses may miss more subtle effects of iron that are limited to particular cortical layers or depths (Deistung et al., 2013; Lee et al., 2023).

6. Relation of Susceptibility Measures to Neurocognitive Function

Whereas semantic knowledge and various forms of expertise (crystallized cognition) can remain constant or even improve with adult age, abilities that are dependent on perceptual-motor speed and working memory (fluid cognition) decline during adulthood (Craik and Bialystok, 2006; Horn, 1982; Park et al., 2002; Salthouse, 2004). A generalized, age-related slowing of central nervous system function appears to be a fundamental dimension of age-related decline in fluid cognitive abilities (Birren, 1965; Brinley, 1965; Madden, 2001; Salthouse, 1996, 2017; Salthouse and Madden, 2007). The vast majority of structural and functional neuroimaging studies of aging have focused on gray matter and white matter in the cerebral cortex, especially structural volume and functional activation in the case of gray matter, and the microstructural integrity of white matter as reflected in measures of the diffusivity of molecular water (Dennis and Cabeza, 2008; Fjell and Walhovd, 2010; Grady, 2012; Raz et al., 2010). As discussed in the previous sections of this article, excessive levels of brain iron contribute to neurodegenerative disease, and increases in brain iron occur during adulthood even in the absence of disease. A third theme, from recent QSM studies (Table 3), is that regional increases in brain iron contribute to age-related decline in neurocognitive function in healthy adults.

Table 3.

Relations between QSM and Cognitive Aging

First author (year)	N	Age	QSM region	Cognitive domain	Relation
Chen et al. (2018)	27	60-82	Putamen	A Language	−
			Globus pallidus	Fluid cognition	−
Chen et al. (2021b)	150	M = 69	Hippocampus	Episodic memory	−
			Hippocampus	Executive function	−
			Globus pallidus	Executive function	−
			Temporal cortex	Episodic memory	−
			Frontal cortex	Episodic memory	−
Gustavsson et al. (2022)	208	20-79	A Prefrontal cortex	A Working memory	+
Howard et al. (2022)	67	18-78	Temporal cortex	Fluid cognition	−
			Posterior cingulate	Fluid cognition	−
			Motor cortex	Fluid cognition	−
			Putamen	Fluid cognition	−
Kalpouzos et al. (2021)	208	20-79	Putamen	Working memory	−
			Putamen	Executive function	+
Li et al. (2015a)	132	43-80	Globus pallidum	Manual dexterity	−
			Red nucleus	Manual dexterity	−
Persson et al. (2020) *	21	M = 36
	15	M = 70	Striatum	Implicit learning	+
Treit et al. (2021)	498	5-90	Caudate	Episodic memory	+
				Executive function	+
van Bergen et al. (2018b)	116	50-95	Hippocampus	Global cognition	0
			Cortex	Global cognition	0
			Deep gray matter	Global cognition	0
Zachariou et al. (2020)	55	61-86	Parietal cortex	Working memory	−
Zachariou et al. (2023)	95	60-86	Precentral gyrus	Executive function	−

Note. Age = range or mean years of age of sample. For each study, symbols indicate positive (+), negative (−), or nonsignificant (0) associations between susceptibility in the listed region and performance on the listed cognitive domain. Studies are sorted alphabetically by first author name.

Δ = longitudinal change.

^*Data presented are averaged across all participants.

Initial studies of cognitive aging, based largely on neuropsychological assessment, proposed that decline in the structure and function of the frontal lobes was responsible for age-related decline in fluid cognitive abilities such as working memory and inhibitory function (Dempster, 1992; Moscovitch and Winocur, 1992; West, 2000; West, 1996). Subsequent research incorporating neuroimaging methods has led to a more nuanced view, in which age-related differences in behavioral measures reflect the connectivity of brain networks that vary in scale (Dennis and Cabeza, 2008; Madden et al., 2020a; Madden et al., 2017; Merenstein et al., 2023b; Monge et al., 2017). These networks, in turn, are comprised of deep gray matter and cerebral cortical regions that form anatomically and functionally distinct networks critical for behavior and cognition (Behrens et al., 2003; O’Muircheartaigh et al., 2015; Zhang et al., 2008; Zhang et al., 2010).

The deep gray matter nuclei are highly interconnected with virtually the entire cerebral cortex (Alexander et al., 1986; Fama and Sullivan, 2015; Haber and McFarland, 2001; Martin, 1996). Given the importance of deep gray matter regions to cortical network connections, it is likely that age-related increases in the deposition of iron in these regions would have consequences for cognitive function, especially the fluid abilities that are most vulnerable to cognitive decline. In their 1970 review of aging and psychomotor slowing, Hicks and Birren (1970) proposed that the basal ganglia and their associated cortical targets comprised a neural mechanism of age-related psychomotor slowing. Rubin (1999) pointed out that evidence linking the frontal lobes to age-related decline in specific cognitive functions (e.g., inhibition) was no stronger than the evidence linking deep gray matter regions, especially the caudate, to the same form of age-related cognitive decline. Similarly, Grahn et al. (2008) surveyed evidence across basic neurobiological and clinical studies and concluded that the caudate has a significant cognitive dimension. These authors proposed that the caudate nucleus contributes to behavior through the excitation of correct action schemas and the selection of appropriate sub-goals based on an evaluation of action-outcomes, both processes fundamental to successful goal-directed action. The putamen, in contrast, appears to coordinate cognitive functions related more closely to stimulus-response, or habit, learning.

Previous reviews of imaging studies of brain iron suggest that age-related increase in deep gray matter iron, particularly in the caudate and putamen, contributes to deficits in neurocognitive function (Daugherty and Raz, 2013; Daugherty and Raz, 2015; Ghadery et al., 2015), perhaps by leading to a decrease in volume of cognitively relevant brain structures (Rodrigue et al., 2013). The majority of the studies included in these previous reviews, however, were based on MRI relaxometry rather than QSM as method of estimation for iron. In addition, previous research has not often compared different forms of neurocognitive outcome, focusing instead on a single outcome, such as working memory (Daugherty et al., 2015; Rodrigue et al., 2013) or motor performance (Adamo et al., 2014), and is limited by single age-group designs. However, the initial studies from MRI relaxometry have provided evidence for a relation between brain iron and age-related decline in neurocognitive function. For example, Ghadery et al. (2015) examined R2*-based estimates of iron in six deep gray matter regions and three cognitive domains (psychomotor speed, executive function, memory, and a composite global measure), in a sample of 336 individuals 55–72 years of age. These authors found that estimated iron load in the putamen accounted for 18–24% of the age-related variance in executive function, global cognitive function, and psychomotor speed, whereas iron in the globus pallidus accounted for only 7–9% of the age-related variance in these measures.

In the first QSM study of the relation between brain iron and neurocognitive function (Table 3), Li et al. (2015a) reported a significant correlation between increasing susceptibility in the globus pallidus and red nuclei, and decreasing manual dexterity, for 132 healthy adults 40–83 years of age. These authors, however, focused entirely on deep gray matter regions, and thus the potential role of cortical iron was not assessed. In addition, although composite measures of manual dexterity and executive function were obtained, the behavioral measures were weighted more towards motor function (Purdue pegboard), and age-related differences in the susceptibility-behavioral relation were not tested specifically.

As illustrated by the pattern in Table 3, QSM studies focusing on the age-cognition relation more directly have fairly consistently indicated an association between age-related increases in brain iron and decline in measures of fluid cognition. For example, in a voxelwise analysis of 67 healthy, community-dwelling individuals 18–78 years of age, Howard et al. (2022) defined clusters from the correlation between susceptibility and fluid cognition. Consistent with the age-related effects reported by Acosta-Cabronero et al. (2016), Howard et al. found that susceptibility for pre- and post-central frontal gyri, among other regions, was related to fluid cognition and comparable in magnitude to those in the putamen (Figure 5 in Howard et al., 2022). In addition, increasing susceptibility in inferior temporal cortex, particularly in the right hemisphere, exhibited a mediating influence on the relation between age and fluid cognition. In a sample of 55 healthy older adults, Zachariou et al. (2020) found that high QSM-based iron concentration in the parietal lobe was associated with poorer working memory task performance. Thus, while motor and premotor cortical regions appear to be preferentially vulnerable to iron deposition, other cortical regions also appear to be involved when analyses focus on the age-cognition relation. However, some paradoxical effects have also been reported, in which higher levels of brain iron are associated with better neurocognitive performance in older adults (Kalpouzos et al., 2021; Persson et al., 2020; Treit et al., 2021). One potential explanation for these surprising results might be the use of net susceptibility measures, rather than separately analyzing the positive (paramagnetic) signal from the negative (diamagnetic) signal. Regardless, an important future direction for this line of work is to distinguish regions of age-related increase in iron deposition (both deep gray matter and cortical) from those regions contributing specifically to the age-related decline in fluid cognitive abilities.

7. Multimodal Imaging Studies of Susceptibility in Aging

As noted previously, variation in the level of brain iron can have either positive or negative consequences for overall brain functioning, because iron is a necessary nutrient for neural physiology and repair, and yet excessive iron also contributes to various forms of neurodegenerative disease. Several neuroimaging studies represent a fourth theme of QSM research, the combination of two or more imaging modalities to characterize age-related differences in susceptibility-based estimates of brain iron (Table 4). These multimodal studies are informative regarding both the underlying neurobiology and behavioral sequelae of increased brain iron. Although several multimodal QSM studies have focused specifically on AD, the biomarkers relevant for AD, primarily Aβ and tau, are not present exclusively in disease but occur in healthy adults as well, at sub-clinical levels. Our focus here is on susceptibility-estimated brain iron from QSM, as combined with information from other imaging modalities, to characterize neurocognitive function in healthy adults.

Table 4.

Relations between QSM and Other Imaging Measures

				QSM region
First author (year)	N	Age	Other measure	FC	TC	PC	OC	CN	PT	GP	RN	SN	TH	HC	AM
van Bergen et al. (2018b)	116	50-95	AP	+	+	+	+	+	+	+				+	+
Chen et al. (2021b)	150	M = 69	AP	+	+	+	+	−	0	0				+	+
Bauer et al. (2021)	80	60-86	Deep WMH	+	+	+	+	+	+
			PV WMH	0	0	0	0	0	0
van Bergen et al. (2018a)	80	55-96	Global WMH	0	−	−	0							0	0
Chiang et al. (2022)	32	50-87	Global WMH	0	0	0		0	0	0				0	0
Li et al. (2021)	44	M = 59	Global WMH					0	0	0	0	0	0
Gong et al. (2015)	42	25-78	FA/M Kurtosis					0	+	0	0	0	0
Yang et al. (2022)	25	M = 67	Axial kurtosis					+	0	0	0	0
Zhou et al. (2020)	213	43-80	Diffusivity index					+	+	+	+	+	+
Zachariou et al. (2023)	95	60-86	Neurite density	−		−
Garzon et al. (2017) *	22	M = 36	R2*					+	+	+	+	+		+
	18	M = 70
Taege et al. (2019)	33	M = 52	R2*					+	+	+	+		−
Li et al. (2014)	191	1-83	R2*	+		+		+	+	+	+	+
Quevenco et al. (2017)	37	62-89	Nodal strength	+	0	+	+	+	0	0				0	0
Persson et al. (2020) *	21	M = 36	BOLD signal	+	+
	15	M = 70
Zachariou et al. (2020)	55	61-86	Functional conn	−
Guan et al. (2019)	35	M = 58	Functional conn	−	−

Note. Age = range or mean years of age of sample.

For each study, symbols indicate positive (+), negative (−), or nonsignificant (0) associations between susceptibility and another MRI measure. Studies are sorted by the secondary MRI measure that was examined.

FC = frontal cortex; TC = temporal cortex; PC = parietal cortex; OC = occipital cortex; CN = caudate nucleus; PT = putamen; GP = globus pallidus; RN = red nucleus; SN = substantia nigra; TH = thalamus; HC = hippocampus; AM = amygdala; Aβ = amyloid beta; WMH = white matter hyperintensities; FA = fractional anisotropy; BOLD = blood oxygen level dependent; PV = periventricular; conn = connectivity.

^*Data presented are averaged across all participants.

Combining QSM with positron emission tomography (PET) indicates that increased brain iron tends to co-localize with Aβ, even in healthy adults (Cogswell and Fan, 2023). Van Bergen et al. (2018b) provided clear evidence for the co-localization of iron and Aβ in a study of 116 healthy older adults, combining QSM with [18F]-flutemetamol PET, which can localize Aβ. In a voxelwise analysis, these authors found that positive correlations between iron load and Aβ plaques were present in a bilateral pattern of clusters in basal ganglia but also several regions in the frontal, temporal, and parietal lobes. When these clusters were thresholded for the level of Aβ, individuals with higher levels of Aβ in frontal and temporal clusters exhibited lower scores on a composite measure of fluid cognition. Van Bergen et al. (2018a) added analyses of fluid attenuated inversion recovery (FLAIR) images, as an estimate of small vessel cerebrovascular disease (from the presence of white matter hyperintensities), to the Aβ PET and QSM imaging. Their findings indicated that in the oldest-old group (85–96 years), a relatively lower cortical iron load was associated with a lower vulnerability to loss of cognitive function, even when combined with other neuropathologies (e.g., decreased cortical volume and increased prevalence of cerebrovascular disease).

Only a handful of studies have combined diffusion-weighted imaging (DWI) with QSM. Two of these studies focused on diffusion properties of deep gray matter nuclei in relation to susceptibility in those nuclei, finding positive relations between diffusivity and susceptibility in the striatum (Gong et al., 2015; Yang et al., 2022). An additional study focused on diffusion properties of white matter found that lower neurite density in frontoparietal white matter was associated with higher susceptibility in adjacent frontoparietal gray matter regions (Zachariou et al., 2023). Higher susceptibility in deep gray matter nuclei has also been related to worse white matter microstructure, seen as higher diffusivity in association and projection tracts (Zhou et al., 2020). Together, these findings support the theoretical notion that the excessive accumulation of iron in gray matter regions should negatively interact with the function of oligodendrocytes (Todorich et al., 2009), seen as lower measures of white matter microstructure.

Relaxometry studies (e.g., Salami et al., 2018) suggest that deep gray matter iron may contribute to age-related disruption of resting-state functional connectivity in healthy aging, as assessed by functional magnetic resonance imaging (fMRI). Several QSM studies have also incorporated fMRI to similarly investigate the relation between susceptibility and measures of cortical activation or functional connectivity. Zachariou et al. (2020) combined QSM with task-related fMRI activation in an analysis of working memory performance, for 55 healthy older adults. These authors found that task performance was correlated positively with the strength of task-based functional connectivity between brain regions of a frontoparietal network associated with working memory. Higher cortical iron concentration in the parietal lobe, however, was associated with lower activation within this frontoparietal network and with poorer working memory performance, after controlling for both cerebral blood flow and brain volume. Zachariou et al. concluded that high cortical iron concentration disrupts communication within the frontoparietal networks supporting older adults’ working memory performance.

The Zachariou et al. (2020) findings are consistent with the role of iron in age-related decline in fluid cognition, reviewed in Section 6 of this article. Other findings, however, do not fit this pattern. Persson et al. (2020), for example, found that striatal iron concentration, for a combined sample of younger and older adults, was related positively to both task-related fMRI activation and behavioral performance, within an implicit sequence learning task. Thus, exactly how brain iron deposition and fMRI activation interact with age-related declines in fluid cognition may depend on the nature of the behavioral task and the sampled brain regions. In addition, whereas Zachariou et al. (2020) limited analyses only to measures of positive susceptibility, Persson et al. (2020) instead used an average measure of susceptibility that is sensitive to iron, among several other neurobiological properties (e.g., the degree of myelination).

8. Moderators of Susceptibility Measures in Aging

A handful of studies illustrate our fifth theme, the role of other neurobiological variables as modifiers of brain iron in healthy aging (Table 5). The potential moderating effect of biological sex has been most frequently assessed, but these studies overwhelmingly report no significant differences between males and females in susceptibility for deep gray matter nuclei (Acosta-Cabronero et al., 2016; Gong et al., 2015; Li et al., 2023; Li et al., 2021; Persson et al., 2015; Treit et al., 2021; Xu et al., 2008). There is some minimal evidence that females may have lower susceptibility than males, but these effects vary anatomically (e.g., red nucleus vs. substantia nigra; Gong et al., 2015; Li et al., 2021) and are not consistent across studies. Only one study has assessed sex-related differences within cortical regions, and similarly observed no significant group differences (Acosta-Cabronero et al., 2016). Menstruation-related loss of iron in females has been posited as a likely mechanism of sex-related differences in brain iron accumulation (Tishler et al., 2012). However, most prior analyses have been conducted across women of all ages, including both early pre-menopausal (age 20–30 years) and late post-menopausal (ages 65–70 years) women, and the interaction of these lifespan differences may contribute to a net minimal effect of biological sex. Future investigations of these sex-related differences may therefore benefit from comparisons between pre- and post-menopausal women in midlife.

The effects of several health-related variables on iron accumulation in aging have also been examined in a few studies. There is some preliminary evidence that a diagnosis of type II diabetes is associated with higher susceptibility in the dorsal striatum and red nucleus, potentially through the damaging effects of hyperglycemia on neuronal metabolic functions (Li et al., 2021). Smoking tobacco products has also been linked to elevated susceptibility in the thalamus (Li et al., 2021), which could theoretically be attributed to hypertension. However, that same study observed that being hypertensive was paradoxically linked to lower susceptibility in the red nucleus (Li et al., 2021), and a separate study reported no significant effect of hypertension in cortical or deep gray matter regions across the adult lifespan (Acosta-Cabronero et al., 2016). But relative to the less modifiable effect of disease status, one particularly promising modifiable lifestyle variable is the consumption of a diet that is high in nuts, fish, and healthy oils, which has been linked to lower susceptibility in parietal cortex and the putamen among older adults (Zachariou et al., 2021).

Beyond biological sex and health-related variables, particular genetic combinations have been identified as a third moderating variable of susceptibility in aging, with some genotypes being associated with lower susceptibility values. For example, adults across the lifespan with less favorable combinations on genes involved in iron transport and storage, particularly C282Y and H63D mutations on the HFE gene (but also TF and SLC25A37; Elliott et al., 2018), have higher susceptibility in basal ganglia nuclei (Elliott et al., 2018; Kalpouzos et al., 2021). Studies of middle-aged and older adults similarly report that individuals with an ε4 allele on the APOE gene (involved in the transport of cholesterol and phospholipids in the brain) had higher susceptibility in the hippocampus, amygdala, caudate, and temporal and parietal cortices than those who do have not the ε4 allele (Ayton et al., 2017; Nir et al., 2022). Similarly, for the COMT gene (involved in endogenous dopamine synthesis), one study reported that older adults with the less favorable Met allele combination had higher susceptibility values in the striatum and dorsolateral prefrontal cortex when compared to those with the more favorable Val combination (Gustavsson et al., 2022). Intriguingly, the difference in susceptibility values was not significant among younger adults in this study. Thus, the genetic influence on susceptibility patterns may become magnified as a function of increasing age, similar to previous reports between genotype combinations and diffusion-tensor based measures of white matter microstructure between younger-old (ages 65-80) and oldest-old (ages 80+ years) adults (Merenstein and Bennett, 2022).

9. Conclusions and Future Directions

QSM is a valuable tool for assessing the degree of cerebral iron accumulation in vivo and has shown great promise for contributing to our understanding of healthy neurocognitive aging. QSM has confirmed previous findings from ex vivo histology indicating that, across the adult lifespan, some deep gray matter nuclei (e.g., putamen) are more vulnerable to iron accumulation than others (e.g., thalamus, globus pallidus; Table 1). QSM has also confirmed and extended previous reports that, beyond deep gray matter nuclei, the frontal, temporal, and parietal cortical regions exhibit age-related increases in iron deposition (Table 2). Although the magnitude of cortical iron is lower than that of deep gray matter iron, both deep gray matter and cortical iron accumulation are associated with age-related decline in several domains of fluid cognition (Table 3).

The evidence to date, however, is not yet conclusive as to whether the relation of age to iron, and age to fluid cognition, are independent effects, or whether brain iron deposition influences the relation between adult age and cognition (but cf. Howard et al., 2022). This line of work therefore may benefit from more fine-grained analytical scales. For example, depth-wise analyses of cortical iron load can separately examine age-related differences in susceptibility at the most superficial (pial) depths versus deeper depths near the gray matter / white matter boundary (Lee et al., 2023), and these measures may, in turn, differentially explain age-related decline in cognitive performance. Multimodal neuroimaging studies can also provide a more detailed understanding about the interaction between brain iron and other neural substrates, with preliminary support for the notion that increased iron negatively interacts with neuroimaging measures of brain function, white matter microstructure, and AD-related pathologies, even in the absence of frank disease (Table 4).

Based on the methodological variability among QSM studies in aging, research in this field may benefit from the development of standardized toolboxes for QSM processing, such as IronSmith (Zachariou et al., 2022), and large-scale imaging consortia, to help guide the choice of acquisition parameters for QSM studies (e.g., the Human Connectome Project; Glasser et al., 2016). It is also imperative that future research determine the most appropriate reference region for studies of aging, by comparing susceptibility measured from samples of cerebrospinal fluid versus susceptibility in ex vivo white matter and gray matter tissue. Finally, we suggest that future research separately examine both positive and negative sources of susceptibility to help better characterize these distinct signals and how they vary in relation to age and neurodegenerative disease (Ahmed et al., 2023; Betts et al., 2016; Chen et al., 2021a; Shin et al., 2021).

On the behavioral side, additional analyses are needed that can separate different components of fluid cognition. Diffusion decision modeling of reaction time, for example, can distinguish nondecision time (sensory encoding and response initiation) from the rate of information extraction and evidence thresholds (Ratcliff et al., 2016; Voss et al., 2013). Application of this modeling to date suggests that the age-related decline in fluid cognition is dominated by increased nondecision time and cautiousness (Madden et al., 2020b; Merenstein et al., 2023a; Ratcliff, 2008). Because nondecision time relies on sensorimotor circuits comprising deep gray matter-cortical connections, the age-related increase in brain iron may have a specific relation to this component of reaction time. Future studies incorporating these new directions, and adopting longitudinal assessments where possible, should seek to determine how the detrimental effects of iron accumulation can be modified, and potentially mitigated, through neural and lifestyle interventions (Table 5).