Brain Iron Overload, Insulin Resistance, and Cognitive Performance in Obese Subjects: A Preliminary MRI Case-Control Study

Recent evidence supports a linkage among metabolic homeostasis, brain structure, and cognition in the obese population (1,2). In this context, several case-control studies have demonstrated that obesity and insulin resistance (IR) are associated with worse cognitive performance, such as executive function, memory, attention, or processing speed (3–5), and with increased risk of developing dementia (5,6). The better understanding of mechanisms by which obesity and IR contribute to brain damage and cognition performance is an important topic of research.

It is well known that there are bidirectional relationships among brain, obesity, and IR. Chronic fat intake can induce inflammatory changes in the hypothalamus contributing to leptin resistance, weight gain, and impaired central glucose sensing (7–9). Insulin contributes to synaptogenesis and synaptic remodeling, and it regulates cortical glucose metabolism and the expression of the neurotransmitters acetylcholine and norepinephrine, which have implications on cognition performance (2). Besides brain metabolic changes, structural and functional alterations, such as decreased white and gray matter volumes, hypothalamic gliosis, reduced functional connectivity, or decreased regional blood flow in the cortex, have been shown by MRI in patients with IR (1).

Recently, there has been an increasing interest in understanding the role of tissue iron stores in the pathogenesis of IR, nonalcoholic fatty liver disease (NAFLD), as well as neurodegenerative disorders (10–13). Iron, an essential element involved in oxygen transport, oxidative stress, and cellular aerobic metabolism, is mainly stored in ferritin and hemosiderin, which serves as a buffer against iron deficiency or iron overload. In the brain, iron plays an important role in oxygen transport, myelin production, neurotransmitter synthesis, as well as oxidative stress. However, these high requirements coupled with the high susceptibility of the brain to iron-generated peroxidative damage, caused by a very high content of polyunsaturated fatty acids imply a stringent regulation of the availability of iron (14). Ferritin-bound iron increases until the fourth decade of life, and the basal ganglia have been described as being among the regions with the highest brain iron overload (BIO) (15). The liver, the major site of iron storage, plays a central role in the maintenance of systemic iron homeostasis. A primary clinical-pathological syndrome characterized by an increased hepatic iron concentration (HIC) with normal transferrin saturation has been associated with the presence of one or more traits of the metabolic syndrome (16,17). Accordingly, increased HIC has been observed in subjects with altered glucose homeostasis and NAFLD (18,19).

Interestingly, House et al. (20) showed a parallel increase between HIC and BIO at some brain nuclei in elderly subjects. This is important because BIO has recently been associated with early cognitive impairment in animal and human models (11,12). Among these lines, Freeman et al. (21) have demonstrated that diet-induced obesity increases the levels of reactive oxygen species in the brain and promotes worse cognitive performance. In this study, we aimed to identify the factors that contribute to increased HIC and BIO as assessed by MRI and evaluate their impact on cognitive performance in obese and nonobese subjects.

From January to September 2012, we undertook a cross-sectional, case-control study based on BIO and HIC assessment by MRI in 23 obese subjects (BMI >30 kg/m²) in whom type 2 diabetes and transferrin saturation >45% were specifically excluded. Twenty healthy nonobese subjects (BMI <30 kg/m²) were matched for sex and age, and served as a control group. All subjects were consecutively recruited from the ongoing FLORINASH project, which was undertaken to evaluate the role of intestinal microflora in adults with NAFLD. Inclusion criteria were age 30–65 years, and exclusion criteria were any manifestation of cardiovascular disease, systemic disease, infection in the previous month, serious chronic illness, >20 g ethanol intake per day, or use of medications that might interfere with insulin action and inability to understand the study procedures. The institutional review board approved the study protocol, and all subjects provided informed written consent.

Patients underwent anthropometric measurements, and after 8 h of fasting blood was provided for the measurement of plasma lipids, glucose, and insulin. A standard 75-g oral glucose tolerance test (OGTT) was performed for the determination of glucose and insulin levels at 0, 30, 60, 90, and 120 min. Glucose and lipid profiles were determined by standard laboratory methods. Serum insulin level was measured by radioimmunoassay. IR was determined by HOMA-IR. Serum ferritin concentrations were determined by direct chemiluminescent two-site sandwich immunoassay using an ADVIA Centaur Immunoassay System (Siemens Healthcare Diagnostics, Deerfield, IL). Serum alanine aminotransferase (ALT) levels were determined using enzymatic methods. Serum lipopolysaccharide-binding protein (LBP) levels were measured by a human LBP ELISA kit (Hycult Biotech, Uden, the Netherlands). Body composition, including fat mass and android fat mass, was assessed by DEXA (GE Lunar, Madison, WI) using region-of-interest analysis from a total body scan.

Brain and hepatic iron load were assessed on a 1.5T MRI scanner (Intera; Philips Medical Systems). Brain R2* relaxation data were acquired with a multiecho gradient-echo sequence with eight equally spaced echoes (first echo 2.2 ms; interecho spacing 5 ms). T2* was calculated by fitting the single exponential terms to the signal decay curves of the respective multiecho data. R2* values were calculated as R2* = 1/T2* and are expressed in hertz. High R2* values indicate high iron deposition (15). Based on previous observational studies (15,20,22) showing increased iron load at the caudate, lenticular nucleus, thalamus, and white matter, mean R2* values in those regions were obtained by drawing regions of interest. Because of growing evidence suggesting hippocampal and hypothalamic changes in association with obesity and IR (1,2,7), R2* values at the hypothalamus, hippocampus, and amygdala were obtained. A fluid-attenuated inversion recovery sequence was used to exclude preexisting brain lesions. HIC was assessed using T2* (repetition time 140 ms, echo time 14 ms, flip angle 10°) and proton density (repetition time 140 ms, echo time 4 ms, flip angle 10°) techniques. The averages of the measurements of liver and muscle signal intensities were used to assess the HIC, as described previously (23). Additionally, the liver fat fraction was obtained using the three echoes Dixon method (24).

General cognitive performance was assessed with the Wechsler Adult Intelligence Scale—Third Edition (WAIS-III) subsets of vocabulary and digit span (25). Attention and executive functions were assessed with the Trail Making Test (TMT) (26). Memory and working memory were evaluated using the Rey-Osterrieth Complex Figure (ROCF) test (27). Processing speed and selective attention were assessed with the Stroop Neuropsychological Screening Test (28). Risk taking and impulsive behaviors were assessed using the Iowa Gambling Task (29).

Results are expressed as the mean ± SD for continuous variables and as frequencies for categorical variables. As a first step in statistical analyses, normal distribution and homogeneity of variances were tested. To determine differences between obese and nonobese subjects, we used the χ² test for categorical variables and the Student t test for quantitative variables. Nonparametric Mann-Whitney U tests were performed for non-normal distributed variables and are expressed as the median (interquartile range). Nonparametric Spearman analysis was used to determine the correlation between quantitative variables. To identify the independent associations of iron load, two multivariate linear regressions models (enter method) were used; first, a regression model adjusted for age and sex and second, a model controlling for age, sex, and BMI. Receiver operating characteristic analyses were performed to identify R2* and HIC cutoff values to discriminate obesity-associated iron overload. The Student t test was used to determine differences between subjects with and without obesity-associated BIO and cognitive scores. Multivariate linear regression analyses adjusted for sex, age, and years of education, as a cognitive-level measure, were used to determine independent associations of cognitive scores. A P value <0.05 was considered to indicate statistical significance. All statistical analyses were performed with SPSS, version 19 (SPSS, Inc., Chicago, IL).

Subjects’ characteristics are summarized in Table 1. Features of the metabolic syndrome, circulating inflammatory markers, and liver fat fraction were increased in obese subjects. Obese subjects had significantly higher iron load at the caudate, lenticular nucleus, hypothalamus, hippocampus, and liver when compared with nonobese subjects (Fig. 1).

Besides obesity indices, univariate analysis grouping obese and nonobese subjects showed that the area under the curve (AUC) of insulin (AUC_insulin) and HOMA-IR were positively associated with HIC (r = 0.54, P < 0.001 and r = 0.566, P < 0.001) and caudate R2* (r = 0.453, P = 0.003; and r = 0.51, P < 0.001, respectively). HIC increased in parallel with R2* at caudate (r = 0.517, P < 0.001), hypothalamus (r = 0.396, P = 0.009), and hippocampus (r = 0.347, P = 0.023), whereas age and LBP were mainly associated with lenticular nucleus R2* values (r = 0.44, P = 0.003; and r = 0.404, P = 0.007, respectively). Significant correlations between brain R2* at these nuclei and metabolic data are shown in Supplementary Table 1. Subcortical white matter R2* was associated with AUC_insulin (r = 0.373, P = 0.018), liver fat fraction (r = 0.43, P = 0.004), and ALT levels (r = 0.401, P = 0.007). No other associations were found.

In nonobese subjects, HIC was associated with hypothalamic R2* (r = 0.562, P = 0.01), liver fat fraction (r = 0.512, P = 0.021), and circulating iron makers such as serum ferritin (r = 0.483, P = 0.031), serum transferrin saturation (r = 0.675, P = 0.001), serum iron (r = 0.612, P = 0.004), and serum transferrin concentration (r = −0.531, P = 0.016). Hypothalamic R2* was also associated with serum iron levels (r = 0.453, P = 0.045) and transferrin saturation (r = 0.517, P = 0.02). Lenticular nucleus R2* was associated with age (r = 0.595, P < 0.006) and serum transferrin concentration (r = −0.458, P < 0.042).

When obese subjects were analyzed separately, AUC_insulin was associated with HIC (r = 0.579, P = 0.007) and R2* at the caudate (r = 0.466, P < 0.038). HIC was also associated with levels of liver fat (r = 0.712, P < 0.001), serum ferritin (r = 0.539, P = 0.008), and serum iron (r = 0.434, P < 0.049). As in the nonobese group, R2* at the lenticular nucleus was associated with age (r = 0.527, P < 0.01) and serum transferrin concentration (r = −0.614, P < 0.004). Scatterplots of the associations among AUC_insulin, HIC, and BIO at the caudate and hypothalamus are shown in Supplementary Fig. 1.

Table 2 shows significant regression models of the target variables associated with HIC and BIO. Adjusting for sex and age, BMI and obesity indices were independently associated with HIC and R2* at the caudate, lenticular nucleus, and hippocampus. AUC_glucose, AUC_insulin, and HOMA-IR were independently associated with HIC and caudate R2*. Lenticular nucleus R2* was mainly associated with age, obesity indices, serum ferritin, and serum transferrin. Other regression models controlling for age, sex, and BMI showed AUC_insulin independently associated with caudate and hippocampal R2* and serum ferritin and serum transferrin with BIO at the lenticular nucleus. AUC_insulin and liver fat, serum ferritin, and serum transferrin levels were independently associated with HIC.

R2* cutoff values that best discriminated obesity-associated iron overload were 15.68 Hz at the caudate, 23.22 Hz at the lenticular nucleus, 17.75 Hz at the hypothalamus, and 20.30 Hz at the hippocampus. The sensitivity, specificity, and positive and negative predictive values for discriminating obesity by means of R2* >15.68 Hz at the caudate nucleus were 85.7%, 77.3%, 78.3%, and 85%, respectively. The receiver operating characteristic AUC was 0.82 with an error of 0.07. The respective values at the lenticular nucleus were 77.3%, 71.4%, 73.9%, and 75% (AUC 0.758; error 0.077; 95% CI 0.606–0.909); respective values at the hypothalamus were 64%, 61.1%, 69.6%, and 55% (AUC 0.732; error 0.077; 95% CI 0.58–0.883); respective values at the hippocampus were 75%, 65.2%, 65.2%, and 75% (AUC 0.805; error 0.07; 95% CI 0.669–0.942). In the liver, the HIC cutoff value for discriminating obesity-associated iron overload was 13.87 μmol Fe/g (86.4%, 81%, 82.6%, and 85%; AUC 0.872; error 0.063; 95% CI 0.748–0.995).

Obese subjects obtained worse cognitive scores on the digit span test and part A of the TMT when compared with nonobese subjects (Table 1). Besides obesity indices, lower scores on these tests were also associated with levels of LBP (r = −0.560, P = 0.001 and r = 0.491, P = 0.007) and C-reactive protein (r = −0.426, P = 0.019 and r = 0.453, P < 0.014). Age was associated with scores on the TMT, part A (r = 0.547, P = 0.002) and part B (r = 0.611, P < 0.001).

Lenticular nucleus R2* values were associated with worse scores in the digit span test (P = 0.011), the ROCF test (P = 0.001), the TMT part A (P = 0.01), and the Iowa Gambling Task test (P = 0.025) (Supplementary Table 1). Worsening scores on part A of the TMT were also associated with R2* at the caudate (P < 0.001) and hypothalamus (P = 0.007). Hippocampal R2* was associated with worse scores on the ROCF copy test (P = 0.016).

As shown in Supplementary Fig. 2, the presence of obesity-associated iron overload by means of R2* cutoff values at the caudate (P < 0.001), lenticular nucleus (P = 0.005), hypothalamus (only in men, P = 0.010), and liver (P = 0.004) discriminated differences in the TMT. Hypothalamic and hippocampal R2* cutoff values discriminate score differences on the deferred memory test (P = 0.039) and the copy ROCF test (P = 0.023), respectively.

Table 3 shows multivariate regression analysis of the associations between obesity-associated BIO and cognitive scores. R2* values at the lenticular nucleus, hippocampus, hypothalamus, and caudate were associated with worse cognitive scores on the TMT independent of age, sex, or years of education.

Iron load at the caudate, lenticular nucleus, hypothalamus, and hippocampus were significantly increased in obese subjects when compared with nonobese control subjects. We found that the strongest predictors of HIC and BIO at these nuclei were obesity and AUC_insulin. A positive association was also found between HIC and BIO at the caudate, hypothalamus, and hippocampus. Importantly, obesity-associated BIO in these nuclei correlated with worse cognitive performance. Therefore, our preliminary results support the hypothesis that BIO is linked with obesity in a context of IR resulting in worse cognitive performance (1,2,10).

Circulating markers of iron stores and HIC are known to be associated with altered glucose homeostasis, fat mass, and NAFLD (10,18,19). Iron seems to play a direct and causal role in the pathophysiology of type 2 diabetes mediated both by β-cell failure and IR. In line with previous evidence, we found obesity, IR, and liver fat to be independently associated with HIC. Importantly, parallel increases in HIC and BIO at the caudate nucleus, hippocampus, and hypothalamus were also observed (20). It is plausible to hypothesize that in a context of IR, iron in excess, given its high reactivity and through the generation of hydroxyl radicals, may cause not only metabolic perturbations in the liver but also alterations in some target brain nuclei. The results suggest that iron in excess impacts on systemic metabolism across the entire spectrum of iron stores (10,30).

There is very little information about the influence of IR and obesity on BIO. In high-fat diet–fed rats, Morris et al. (31) have shown that progressive IR runs in parallel with iron deposition in the substantia nigra. Lin et al. (30) have reported BIO at the basal ganglia in liver failure patients. On the other hand, obesity and aging have been related with blood-brain barrier hypometabolic dysfunction (32). The hypometabolic state of the blood-brain barrier could be a predisposing factor for iron deposition (33). According to these findings, we found obesity interacting with IR and increased BIO at the caudate, lenticular nucleus, and hippocampus, whereas the increased BIO at the hypothalamus seemed more fat mass dependent. Although it remains unclear whether BIO at the hypothalamus is involved in the pathogenesis of obesity or is a biomarker, it seems reasonable to anticipate that damage to a brain area critical for body weight control might play a role in obesity (7–9).

It is well known that obesity and IR are linked with decreased cognitive performance (3–5). A high-fat diet has been shown to trigger increases in oxidative stress in the brain cortex, which is associated with functional impairment (21). In a context of iron dysregulation, BIO could be an important underlying factor triggering increased oxidative stress leading to neuronal death (33). In line with previous observations (13,34), our data showed that subjects with BIO obtained worse cognitive scores related to motor speed, attention, and cognitive flexibility, as well as to worsening memory scores. The hippocampus and caudate nucleus have a critical role in supporting the planning and execution of strategies, behavior required for achieving complex goals, and memory functions (35,36), whereas the hypothalamus is basically involved in emotional regulation and vital functions (37). These regions are connected with prefrontal cortex and frontal cortical areas, brain regions that usually are associated with cognitive and executive functions (35,36,38). The structural damage due to iron overload in these nuclei might produce a disruption of cortical projections that could explain part of the worse cognitive performance observed in obese subjects.

MRI is a reliable and cost-effective tool for noninvasive assessment of tissue iron stores. The MRI techniques used for iron assessment are based on the changes in relaxation times produced by local magnetic field inhomogeneities and intrinsic tissue properties. Microscopic field gradients induced by paramagnetic ferritin-loaded cells produce a random phase shift of the hydrogen protons, affecting its relaxation. Consequently, relaxation signals of tissues are affected by diffusion-mediated contributions of iron. The amount of ferritin and hemosiderin are considered to be the only forms of iron that have a paramagnetic effect strong enough to significantly affect the MRI signal. Therefore, in this study the iron load assessed by MRI is based mainly on the contribution of ferritin (15). Although in this study subjects with type 2 diabetes were excluded, our preliminary data suggest that brain iron stores determined by MRI should be used as a biomarker of iron metabolism and may be helpful to noninvasively assess structural and functional changes in the context of subjects at risk for impaired glucose metabolism.

Our study has several limitations. This was a single-center study, and the patient sample was too small to draw any definite conclusions. The potential of MRI as tool for detecting obesity-induced BIO should be confirmed. In this study, there was a relatively narrow age range; as a result, it is hard to generalize the findings of this study to a wider population, and this could be explored in a larger study. In addition, the high comorbidities in the presence of obesity make it difficult to discern the real contribution of obesity to structural brain damage. Finally, study subjects were not screened for Wilson disease, which can promote relaxation changes similar to the ones observed in this study. Future studies could consider including such cohorts to serve as disease control subjects.

In summary, our preliminary data show an obesity-related increase of tissue iron stores in a context of IR. Importantly, obesity-associated BIO at the hypothalamus, hippocampus, lenticular, and caudate nuclei correlates with worse cognitive performance in middle-aged subjects. This study highlights the potential utility of MRI as a surrogate marker of iron metabolism, a novel player in obesity-associated cognitive dysfunction.

Funding. This work was supported by European Project FLORINASH grant FP7-HEALTH-2009-2.4.5-1 and by Instituto de Salud Carlos III (ISCIII) grant PI 11/1532. Centro de Investigación Biomédica en Red Fisiopatología de la Obesidad y Nutrición is an initiative of ISCIII.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. G.B. researched the data, performed the statistical analysis, and wrote the manuscript. J.P. and J.M.F.-R. researched the data and contributed to the writing and editing of the manuscript. J.D.-i.-E. and M.P.-O. contributed to the statistical analyses. X.M. researched data and performed the MRI analysis. G.X. researched the data. F.F.-A., S.P., and W.R. contributed to the discussion and reviewed the manuscript. J.M.F.-R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.