The impact of brain iron accumulation on cognition: A systematic review

Iron is involved in many processes in the brain including, myelin generation, mitochondrial function, synthesis of ATP and DNA and the cycling of neurotransmitters. Disruption of normal iron homeostasis can result in iron accumulation in the brain, which in turn can partake in interactions which amplify oxidative damage. The development of MRI techniques for quantifying brain iron has allowed for the characterisation of the impact that brain iron has on cognition and neurodegeneration. This review uses a systematic approach to collate and evaluate the current literature which explores the relationship between brain iron and cognition. The following databases were searched in keeping with a predetermined inclusion criterion: Embase Ovid, PubMed and PsychInfo (from inception to 31st March 2020). The included studies were assessed for study characteristics and quality and their results were extracted and summarised. This review identified 41 human studies of varying design, which statistically assessed the relationship between brain iron and cognition. The most consistently reported interactions were in the Caudate nuclei, where increasing iron correlated poorer memory and general cognitive performance in adulthood. There were also consistent reports of a correlation between increased Hippocampal and Thalamic iron and poorer memory performance, as well as, between iron in the Putamen and Globus Pallidus and general cognition. We conclude that there is consistent evidence that brain iron is detrimental to cognitive health, however, more longitudinal studies will be required to fully understand this relationship and to determine whether iron occurs as a primary cause or secondary effect of cognitive decline.


Introduction
Iron has many biological roles including the cycling of neurotransmitters, enzyme and mitochondrial function, ATP and DNA synthesis and myelin generation [1][2][3][4]. In the healthy human adult brain, the total concentration of iron is around 0-200μg per gram of tissue, typically being lower in the White Matter (WM) and cortical Grey Matter (GM) (<60 μg per gram) [2]. 90% of brain iron is stored in ferritin with only 0.05% of brain iron being present in the labile iron pool [5]. In healthy aging, iron accumulates heterogeneously in specific regions of the brain, bound mainly to ferritin and neuromelanin [6] and largely located in the deep

Study selection
All studies found in the electronic search were assessed for their eligibility for inclusion in this review by Holly Spence. Studies were included if they met all the inclusion criteria and included studies had their referenced papers reviewed for eligibility for inclusion. The included studies ultimately consisted of published articles and theses only.

Synthesis of results
The following study characteristics were extracted from each study for assessment of study quality and study comparison: Number of participants; participant gender ratio; participant average age; type of study design; measures of cognition used; measures of brain iron used; statistical methods used. Results which were statistically significant (p<0.05) were extracted and summarised from each study.

Quality assessment
Each study which satisfied the inclusion and exclusion criteria was assessed for quality via a 10-point based system using the following 10 criteria: (1) Does the study have a clearly defined research objective? (2) Does the study adequately describe the inclusion/exclusion criteria? (3) Is the sample size adequate? (4) Does the study report on the population parameters/demographics? (5) Does the study report detail on appropriate assessment of Cognition? (6) Does the study report detail of the assessment of iron? (7) Does the study provide an appropriate control group? (8) Does the study apply the appropriate statistical analyses? (9) Does the study adequately report the strength of results? (10) Do the authors report on the limitations of their study?

Study selection
The electronic search of Embase Ovid, PubMed and PsychInfo yielded 643 citations in total. After duplicates were removed, 411 studies remained for screening. Once these studies were screened, 141 non-human studies were excluded, 4 papers were excluded due to being unavailable in English and 106 reviews, editorials and conference abstracts were removed. In total 6 studies were excluded due to being single case studies, 60 were excluded for not measuring brain iron and 6 were excluded for not measuring cognition. Finally, 24 studies did not assess statistically the relationship between iron and cognition and so were excluded, as well as 3 studies which measured only the effects of maternal iron on offspring cognition. A total of 28 studies remained for reference screening. After references were reviewed, a further 13 eligible studies were obtained. A total of 41 studies were therefore included in this review. The full details of the study selection process are outlined in Fig 1.

Study characteristics
Study characteristics were collated and are presented in Tables 2 and 3; with Table 2 presenting details on overall study design and Table 3 presenting details on participants and study groups

Quality assessment
All included studies were assessed for quality using a 10-point-based scoring system and each score was converted to a % Quality Score (QS). The quality scores for all studies can be seen in

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review      Table 4. 26 of the 41 assessed studies were of high quality (QS>90%), 14 were of very good quality (QS of 80%-90%) and 1 study was of good quality (QS of 70%-80%).

Summary of results
The key findings relating to the brain iron-cognition relationship were extracted from each study in this review and are summarised in Table 5. 11 of the reviewed papers showed a significant relationship between whole brain iron concentration and measures of cognition consisting of logical memory total, verbal paired associates (both at total, immediate and delayed recall) and spatial span total score, letter-number sequencing and DSB.
MRI (1.5T) T1/T2 � W Total and regional iron and WMH volumes were standardised and presented as % of intracranial volume, age was added as covariate of all models; volumes of iron and WMH were positively skewed and so were log transformed prior to analysis; Multivariate and Bivariate regression models were performed

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review Table 5. Summary of results.

Summary of Findings relating iron to cognition
Ayton et al., 2019 [27] Inferior temporal iron levels were increased only in people with clinical diagnosis of dementia who also had moderate (P = 0.0003) and high pathology (P = 0.0190) and fit the CERAD criteria for probable (P = 0.0066) and definite pathology (P = 0.0003), and Braak criteria IV (P = 0.0067) and V (P = 0.0031); In people with high AD pathology, iron was strongly associated (P<0.0001) with the rate of decline in Global Cognition composite; mediation analysis showed that iron levels mediated 17% of the effect of NFTs on Global Cognition; In subjects with low AD pathology, elevated inferior temporal iron burden was associated with decline in global cognitive score (P = 0.001), but not the individual cognitive domain scores Bartzokis et al., 2011 [28] Significant negative association between HP iron and episodic memory in men only (p = 0.003); Significant effect of iron genes on association between BG iron and working memory/attention score (p = 0.006); Significant correlation between BG iron and working memory/attention in those without H63D and TfC2 genes (r = -0.49, p = 0.005) [29] LN 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); Worse performance in the TMT-A were also associated with R2 � in CN (P = 0.001) and HS (P = 0.007); HP R2 � was associated with worse performance in the ROCF copy test (P = 0.016); HS and HP R2 � cut off values discriminate score differences on the deferred memory test (P = 0.039) and the copy ROCF test (P = 0.023), respectively  [32] Significant correlation between brain iron in GP and fluid composite score in control group (p<0.01); Baseline PU brain iron is negatively associated with changes in oral reading recognition test scores in the control group (p<0.01) [33] Significant correlation between CN iron working memory performance in children (r = 0.64, p = 0.004) and adults (r = 0.46, p = 0.04); mainly driven by the R-CN in children; Other subcortical nuclei were not significantly correlated to working memory performance after Bonferroni correction for multiple comparisons Daugherty, 2014 [34] Greater iron content at baseline was associated with slower iron accumulation in CN (p<0.05) and PU (p = 0.05); Higher metabolic syndrome score was associated with higher iron in the CN (p = 0.003) and LQ (p = 0.02); Inflammation score was unrelated to iron content; Non-verbal working memory didn't change with age (p = 0.76); Verbal working memory improved over two years (p<0.001); Virtual Morris water maze test score was unrelated to iron or volume in any region

Daugherty, Haacke and Raz, 2015 [35]
Cognitive switching ability was found to be inversely proportional to striatal iron (p<0.001)

Daugherty et al., 2019 [36]
Greater baseline CN iron was associated with lesser improvement in working memory over 2 years (p = 0.01); Change in verbal working memory was unrelated to iron in the PU (p>0.52) or HP (p>0.17); Episodic memory wasn't associated with baseline iron (p>0.31)

Ding et al., 2009 [37]
Mean MMSE score was significantly lower in AD patients than controls (p<0.001); AD group showed significantly lower phase value in all brain structures measured (p<0.05); Phase value in R-head of HP had positive correlation with MMSE score (r = 0.603, p = 0.000) and negative correlation with disease duration (r = -0.677, p = 0.013) (Continued )

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review   [42] Higher age associated with lower education level, higher frequency of risk factors, worse cognitive performance, greater extent of focal brain lesions and lower brain volume (p<0.05); Higher iron load in PA related inversely with all cognitive measures except memory; R2 � in PU was related to global cognitive function and psychomotor speed (p<0.05); No relationship between R2 � in neocortex or HP and cognition; Associations between R2 � iron and cognition were strongest in ages above 71; R2 � iron in the pallidum accounted for 9% of the age-related variance in executive function, 7% in global cognitive function, and 8% in psychomotor speed; R2 � iron in the PU accounted for 24% of the age-related variance in executive function, 18% in global cognitive function, and 21% in psychomotor speed Haller et al., 2010 [43] There was a significantly increased iron concentration in R-PA and R-SN in MCI groups compared to controls (p<0.01); There was significantly decreased iron concentration in the R-RN in MCI groups compared to controls (p<0.05); No difference in iron concentration was found in any regions between stable and progressive MCI Hect et al., 2018 [44] Brain iron in CN (p = 0.03), PU (p<0.01), GP (p = 0.04) and SN (p<0.01) correlated with general intelligence scores; Brain iron in the CN (p<0.001) and PU (p<0.01) correlated processing speed; HP (p>0.69) and RN (p>0.33) iron content were unrelated to cognition; Greater general brain iron content predicted faster processing speed (p = 0.02) and better general intelligence (p = 0.01)

House et al., 2006 [45]
Least cognitively impaired memory-complaint group (MC1) had significantly higher R2 in R-temporal cortex and significantly lower R2 in the L-internal capsule compared to controls; MC1 and MC2 groups showed significant correlation between R2 and immediate, short-delay and long-delay free recall scores in CVLT in TH and RN (r = -0.62 to -0.77, p<0.04); R2 in the RN was negatively correlated to MMSE scores (p<0.02); Negative correlation coefficients were more frequently associated with R2 in GM regions for the immediate free recall scores (p = 0.001), SDFR cognitive score (p = 0.0002) and LDFR test scores (p = 0.0002) [46] Higher striatal iron in the older group was associated with poorer recall in motor condition (p = 0.02); Striatal iron was not significantly associated with recall in the younger adults (p>0.7); Bootstrapping analysis indicated reliable association between striatal R2 � and memory performance in older group; Greater striatal iron was associated with less inferior frontal cortex activation when age and striatal volume were controlled for (p = 0.05); Higher iron in R-PU was associated with lower activity in the R-PU when controlling for age and R-PU volume (p = 0.04)

Larsen et al., 2020 [1]
Developmental trajectory of R2 � in PU significantly interacted with overall cognitive score (p = 0.006) with poorer performance becoming increasingly associated with lower R2 � levels; Developmental trajectories of R2 � were most strongly associated with complex cognitive performance (p = 0.004) with significant association between R2 � trajectory and social cognition (p = 0.031) and executive function (p = 0.032), No significant effect of R2 � on memory performance (p = 0.39)

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review  [47] Decrease in manual dexterity score was significantly associated with increase in magnetic susceptibility in the GP and RN; In younger participants the susceptibility-dexterity correlation was significant for GP (p<0.01) but not RN (p = 0.028); In older participants the susceptibility-dexterity correlation was significant for RN (p<0.05) but not for GP (p = 0.11); Only GP magnetic susceptibility was a significant predictor of variance in manual dexterity score (with higher GP magnetic susceptibility associated with lower manual dexterity score) Lu et al., 2015 [48] Compared to control group, cmTBI patients had significantly higher angle radian values in CN (p<0. 3 months after shunt surgery, brain-type Tf strongly correlated with MMSE scores (r = 0.697, p = 0.037) and FAB score (r = 0.727, p = 0.041); 12 months after shunt surgery, brain-type Tf moderately correlated MMSE scores (r = 0.549, p = 0.022) and FAB score (r = 0.373, p = 0.154); mRS scores were not associated with braintype Tf before or after surgery

Penke et al., 2012 [51]
Compared with the group without detectable Iron Deposits (IDs), those with IDs at age 72 had significantly lower general cognitive ability at age 70 (p = 0.043), and age 72 (p = 0.0004), but not at age 11 (p = 0.19); Censored correlations showed greater IQ at 11 was significantly associated with fewer iron deposits at age 72 (p = 0.0324, r = -0.19); Reading recognition tests showed significant negative association with iron deposits (r = -0.18, p = 0.0253); Iron deposits were significantly associated with lower general cognitive ability at age 70 (r = -0.27, p = 0.0015) and 72 (r = -0.31, p<0.0001) Pinter et al., 2015 [52] Magnetisation transfer ration for normal appearing brain tissue explained 26.7% variance in overall cognition; Overall iron deposition did not account for variance in overall cognition significantly; Basal ganglia R2 � explained 22.4% variance of cognitive efficiency; HP magnetic transfer ration of normal appearing brain tissue (22.4%) also accounted for memory variance; TH volume was the only predictor of memory function after multivariate modelling; The only predictor of cognitive efficiency after multivariate modelling was R2 � in the basal ganglia (explaining 22.4% variance) Qin et al., 2011 [53] R2 � in HP, PC, PU and CN of AD significantly higher than control group (p<0.05); R2 � in PC, HP and L-PU in mild AD group were significantly higher than in controls (p<0.05); R2 � in HP, PC, PU and DN in patients with severe AD were significantly higher than the control and mild AD groups; MMMSE was negatively correlated with R2 � and iron concentration in PC and HP in AD group (p<0.01)

Rodrigue et al., 2013 [54]
Increased HP iron and smaller HP volume accounted for age-related memory deficits (p = 0.05) whereas, CN did not have this effect; Younger participants with larger HP and lower HP iron had the highest memory composite scores; Single indirect path modelling showed a negative indirect association of age with HC volume through increased HP iron concentration (p<0.0001) and advanced age was indirectly related to poorer memory performance through a shorter HP T2 � and then smaller HP volume (p<0.0001) Rodrigue et al., 2020 [55] Significant decline in performance across all levels of n-back tests (p<0.05) with increasing age but no iron interaction in this model (p>0.174); No association found between iron and performance in executive function performance (Continued )

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review  [56] Significant negative association between striatal R2 � and coherence in connectivity of the CN (r = -0.41, p = 0.008) and PU (r = -0.32, p = 0.047); Significant association between striatal iron and coherence of connectivity in the CN restingstate network in the older group (r = -0.53, p = 0.04) but not in the younger group (r = -0.24, p = 0.27); Association between QSM and CN connectivity coherence confirmed significance (r = 0.398, p = 0.015) but the PU connectivity coherence was not significantly associated with QSM (p = 0.07); Significant positive association between coherence of PU networks and task performance with the dominant hand across age groups (p = 0.04); Significant association between striatal iron and motor performance with the dominant hand across the age groups (p = 0.047)

Schmalbrock et al., 2016 [57]
Flanker test for inhibitory control was significantly associated with QSM in CN (p = 0.01) and anterior PU (p = 0.045); Stroop test for inhibitory control was not significantly associated with brain iron measures; Disease duration was significantly associated with QSM in the CN (p = 0.02); Sqrt (Flanker) was significantly associated with age adjusted QSM in the CN (p = 0.0058) and anterior PU (p = 0.016); Duration adjusted Expanded disability status score was significantly associated with age adjusted QSM in the posterior PU (p = 0.032) and age adjusted R2 in the CN (p = 0.014), PU (p = 0.0059, Anterior p = 0.0054, Posterior p = 0.019)

Smith et al., 2010 [58]
Controls had significantly lower cortical redox iron than other groups (p<0.05); Controls had significantly less iron accumulation in the cerebellum but had high metal deposition in the purkinje cell layer; Iron accumulation did not occur not in purkinje cells for MCI brains but instead in spherical glial associated structures; MCI cases had significantly more iron accumulation than controls in the purkinje layer associated with glial cells  [61] svMCI group had significantly lower composite, attention-executive, memory and language z scores than controls; significantly higher susceptibility in svMCI group over controls in R-HP (p<0.01), L-HP (p<0.01), R-PU (p<0.05); svMCI group had significantly negative correlation between sus in R-HP and memory z sore (p = 0.012); susceptibility in R-HP of svMCI group was positively correlated to language z score (p = 0.026); susceptibility in R-PU in the svMCI group was significantly negatively correlated to attention-executive z score (p = 0.033); composite z score not related to susceptibility Thomas et al., 2020 [62] Increase in QSM in PD compared to controls in prefrontal cortex, R-PU and Rtemporal cortex (p<0.05); Increased QSM in SN in PD compared to controls (p = 0.004); In PD patients there was susceptibility increase with decreasing MoCA scores in HP, TH, CN, caudal regions of ventromedial prefrontal cortex, regions of basal forebrain, R-PU and R-insular cortex; Increased absolute susceptibility with increased dementia risk score in PD patients (p<0.05); widespread QSM increases in patients with poor visual performance (p<0.05); PD group showed significant increase in susceptibility (p<0.05) with UPDRS-III in right PU (Continued ) (including memory, general intelligence, visual performance, processing speed, social cognition and BOLD modulation). Every other study reviewed reported significant associations between iron levels in specific brain regions and individual measures of cognition, as presented in Fig 2.

Summary of evidence
This review analysed human studies in which brain iron and cognition were measured and their relationship assessed statistically. Many of studies assessed reported a significant relationship between total brain iron and general cognitive performance and many links between regional iron levels and specific measures of cognition were also reported. Memory function was the most frequently reported cognitive measure to be correlated with brain iron, however, this was the most frequently assessed cognitive outcome. Regions where iron was most frequently reported to correlate with memory performance were the Caudate nuclei, Hippocampus and Thalamus. All other regions were also associated with memory in at least one study except for the Globus Pallidus where regional iron had no reported associations with memory. The associations between the caudate, hippocampus and thalamus iron and memory are somewhat unsurprising as each of these regions are known to be involved in different facets of memory function [67][68][69] and so it is plausible that disruption of these circuits via iron accumulation would confer memory dysfunction. The efficacy of interactions between the caudate and hippocampus in memory function has been associated with availability of dopamine receptors [70,71], which has in turn been proposed as having a potential role in iron accumulation [72]. Studies have suggested that iron and dopamine can interact to induce oxidative stress and neurodegeneration by forming a toxic couple [72]. Animal studies have also demonstrated that iron deficient mice and rats show decreased dopamine transporter and receptor levels and general dopaminergic dysfunction [73,74]. This suggests that with an increase iron, there could be an increase in dopamine receptors and transporters, enhancing toxic coupling between iron and dopamine and thus increasing neurodegeneration in dopamine rich regions, however, this requires further investigation. Furthermore, higher iron levels in the caudate nuclei were also consistently reported to correlate to poorer general cognitive performance. However, the putamen had the most reported associations with general cognition, with the Globus Pallidus and the Substantia Nigra also being associated with general cognition in more than one study. The putamen has roles in many different neurological functions such as, sensory and motor information processing, learning and language [75][76][77]. This could explain the consistency of reports that iron accumulation here is associated with poorer general cognitive performance, further suggesting that iron accumulation causes atrophy which leads to a localised disruption of function.  Figure presents number of studies reporting significant association (p<0.05) between regional iron and cognition measures. � Pallidum had associations between regional iron and memory in one study [30] but had association in all cognitive measures except memory in a second study [42]. https://doi.org/10.1371/journal.pone.0240697.g002

PLOS ONE
The impact of brain iron accumulation on cognition: A systematic review Although assessed in less of the studies reviewed, there were associations between reduced motor function and increased striatal iron content, as well as, increased iron in the Putamen and increased disability scores, such as, Dementia rating scale, Extended disability status score and the UPDRS-III for rating of Parkinson's pathology. Due to its many neurological roles and connections, atrophy in the Putamen is known to be involved in pathology of several diseases such as, Parkinson's disease, Multiple Sclerosis and Dementia with Lewy Bodies [78][79][80][81]. The striatum consists of the caudate and putamen and is crucial for connections to the basal ganglia which is heavily involved in motor control [82]. These associations therefore suggest that iron accumulation is capable of either causing atrophy or is accompanied by atrophy, which in turn causes regional damage that can interfere with circuitry in the brain. This is in line with the findings of several of the included studies that increases in regional brain iron were strongly associated with regional volume decrease [34,35,49,54], suggesting that brain iron increase is correlated with atrophy.
Whilst not evaluated in the included studies, differences in iron status have been observed between sexes particularly during development. Larsen et al. observed these differences in their 2020 study, which determined that male brain iron levels plateau at an earlier age than in females in some brain regions. Due to this later plateau during development females generally begin adulthood with higher brain iron levels than men. However, at older age, females are shown to have generally lower iron stores in some brain regions than males, potentially due to menstruation [83][84][85]. Female brain iron deficit mediated by menstruation would however, be highly variable, dependent on the characteristics of an individuals' menstruation (i.e. menstruation pattern, heaviness of blood loss etc.) [83]. This may put females at a lower risk of brain iron-mediated cognitive impairment, however the effects of sex-mediated brain iron on cognition have not been extensively studied.
All studies included in this review controlled for sex during their analyses. Fifteen of the included studies assessed sex-mediated brain iron differences statistically; 1 study found that while temporal iron levels did not significantly differ between men and women, cerebellar iron was significantly higher in males compared to females [27]; 1 studies found significantly higher hippocampal iron in men compared to women [28]; 1 study found that regional brain iron in women plateaus later than in men during development [1] and the remaining 12 studies observed no significant difference in brain iron in any assessed region between males and females [29, 31, 34-36, 38, 42, 44, 47, 51, 54, 63].
Although primarily focusing on the associations between brain iron and cognition, some of the papers reviewed did provide insight into potential mechanisms for this relationship. It has been previously reported that iron in the brain tends to localise to protein aggregates and some studies have shown that iron plays a role in the toxicity of some of these aggregates [86][87][88][89]. In fact, when amyloid β (Aβ) is complexed with iron it can induce ROS via Fenton's reaction leading to oxidative stress and activation of the Bcl-2 apoptotic pathway [18,19]. Iron has also been shown to localise with protein aggregates such as tau and amyloid beta in animal models for AD and PD [88,90,91]. Several of the studies included in this review reported that iron was localised to Aβ plaques and neurofibrillary tangles [27,58]. A study by Ayton et al. [27], included in this review, found that brain iron level mediated 17% of the effect of Neurofibrillary tangles on cognitive performance. This, taken with the afore mentioned literature, suggests that iron could amplify neurodegenerative processes such as protein misfolding, rather than being a primary cause or effect of disease.

Limitations
Whilst this article was able to provide a comprehensive review of the literature investigating the relationship between brain iron and cognition, there were several limitations to this study.
Firstly, there was a wide variety of methods for measuring both brain iron and cognition and this must be considered when comparing the included studies. Secondly, although a thorough search of the literature was conducted, it is possible that relevant studies were missed and thus not included. Furthermore, all included studies were published articles or theses and thus there is an element of publication bias in this review that must be considered. Additionally, some of the studies included in this review had relatively small sample sizes which may reduce the power of some of the conclusions made. The participants all bar one of the studies in this review were adults and so the findings cannot be applied to children or adolescents. Finally, the potential mechanisms by which iron accumulation in the brain could cause cognitive dysfunction were not assessed in this review and remain unclear.

Conclusions
To conclude, this review has investigated the effects of brain iron on aspects of cognition. There is consistent evidence in the studies reviewed that in adulthood, an increase in brain iron had a detrimental effect on cognitive ability. However, it seems that iron accumulates heterogeneously across brain regions and when only some regions have high iron, their specific function can be impaired. In this way, increased iron in the Caudate nuclei, Hippocampus and Thalamus is consistently reported to correlate to poorer memory performance; whereas, increased iron in the putamen was more consistently reported to correlate to poorer general cognition. These findings strongly suggest an effect of brain iron on cognition and this relationship should therefore be investigated further. Going forward, it is important to determine whether iron is a primary cause of brain atrophy or whether brain iron accumulation is a secondary effect of brain atrophy. Regardless of the mechanisms underlying the relationship between brain iron and cognition, MRI techniques for quantifying brain iron therefore show promise as a potential non-invasive biomarker for age-related cognitive decline.