Introduction

Psychiatric disorders exhibit a high prevalence and disability rate, significantly affecting both public health and quality of life [1]. Currently, the clinical diagnosis of psychiatric disorders largely depends on behaviors and symptoms. It is often inadequate due to the subjective nature and variability of clinical features among these disorders and is further complicated by overlapping symptoms from different disorders [2]. This highlights the urgent need to examine the biological alterations that could map the neuropathological mechanisms in each disorder and extend our understanding of these illnesses [3]. Developing objective biomarkers is crucial for improving diagnostic accuracy and guiding/developing effective treatment [4].

Iron is a critical cofactor in physiological processes, including oxidative phosphorylation, myelin phospholipid production, and neurotransmitter synthesis and metabolism [5]. Brain iron deficiency could hinder myelin formation and lead to disruptions in neural transmission [6], while iron overload may result in ferroptosis and demyelination of neurons associated with cognition and emotional regulation, cellular respiration, and inflammation [7]. Based on the magnetic properties of brain iron, quantitative susceptibility mapping (QSM) has recently emerged as a valuable tool for assessing brain iron levels [8], which holds significance in revealing the pathological mechanisms underlying a wide range of illnesses, including neurodegenerative diseases and psychiatric disorders [9,10,11,12,13,14,15,16,17,18,19,20,21,22]. For example, observational evidence suggested that patients with schizophrenia (SCZ) tend to have higher iron levels in the bilateral putamen, and lower iron levels in the bilateral substantia nigra, left red nucleus, and left thalamus compared to healthy controls [12, 13]. Patients with depression showed increased iron in the bilateral putamen [14] and thalamus [15], while children with attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD) exhibited reduced iron levels in regions such as the striatum, substantia nigra, hippocampus, and caudate [16,17,18]. Higher iron levels were found in the hippocampus in patients with Parkinson’s disease and comorbid anxiety disorders (ANX) than those without ANX [19], in the pallidum in patients with obsessive-compulsive disorder (OCD) [20, 21], and in the thalamus, caudate, and putamen in patients with primary insomnia [22].

These findings highlight the distinct patterns of subcortical iron alterations across psychiatric disorders. However, the discrepancy between these findings is notable, and observational studies, though essential for identifying these findings, are susceptible to confounding biases and often limited by small sample sizes. Therefore, the nature of these associations, i.e., whether these changes in iron levels are a cause or a consequence of these psychiatric disorders, has not yet been established. Mendelian randomization (MR) analysis offers a robust solution to reveal causal relationships between iron levels in specific brain regions and psychiatric disorders by employing accessible genome-wide association studies (GWAS). Unlike traditional observational studies, MR leverages the principle of random allocation of genetic variants (SNPs), minimizing confounding factors and reducing the risk of reverse causality [23, 24]. Previous MR studies have established a causal link between brain iron levels and neurodegenerative disorders, such as Parkinson’s disease and Alzheimer’s disease [9, 10]. Given that iron-rich deep brain, nuclei are critical to motor, emotional, and cognitive functions that are significantly impaired in various psychiatric disorders [25], recent findings have also indicated the associations between higher brain iron levels in regions including the putamen, caudate, and thalamus and lower gray matter volumes [26]. Examining the bidirectional causal relationship between brain iron levels of these regions and psychiatric disorders is important and focusing on a single subcortical region through univariable MR analysis allows us to identify their specific contribution to psychiatric disorders, just as did in previous work in neuropsychiatric disorders [27,28,29,30,31,32].

With the aforementioned considerations, in this study, we conducted a two-sample MR analysis to investigate the causal association between iron levels in 16 subcortical regions and eight psychiatric disorders including SCZ, MDD, ASD, ADHD, bipolar disorder (BD), ANX, OCD, and insomnia. Multivariate Mendelian randomization (MVMR) was performed to assess whether structural alterations in subcortical regions play a mediating role between iron levels and psychiatric disorders. We hypothesized that changes in iron levels in specific subcortical brain regions (such as putamen, caudate, thalamus, hippocampus, pallidum, substantia nigra, etc.) might be associated with different psychiatric disorders and different psychiatric disorders may also cause specific iron changes.

Methods

Instrumental variables

QSM was employed to determine the magnetic susceptibility distribution of tissues from gradient-echo MRI phase images [33, 34]. It estimates the median susceptibility within a voxel, reflecting brain iron levels, a major factor in susceptibility [35, 36]. This study obtained the summary statistics of QSM from the largest GWAS from UK Biobank Brain Imaging, encompassing 35,273 individuals of European ancestry, as reported by Wang et al. [36]. The study analyzed 16 subcortical brain regions for QSM with valid GWAS data, including bilateral thalamus, caudate, putamen, pallidum, hippocampus, amygdala, accumbens, and substantia nigra. In addition, GWAS resources on the structure of 14 subcortical brain regions (except for the left and right substantia nigra), including gray matter volume, total volume, and density, were also obtained online via the IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/). The GWAS resources for SCZ [37], MDD [38], ASD [39], ADHD [40], BD [41], ANX [42], and OCD [43] were acquired from the Psychiatric Genomics Consortium, while the GWAS resource for insomnia was acquired from a recent meta-analysis [44] (Supplementary Table 1). The data utilized in this study were sourced from publicly accessible databases, and no additional ethical approval was required. All original studies related to databases obtained the necessary ethical approval and participant consent before the investigation. Populations with comorbid psychiatric disorders have been excluded from the investigation of their respective conditions in GWAS resources. Instrumental variables mainly included single nucleotide polymorphisms (SNPs) that reached the genome-wide significance (p < 5E-08). If the number of selected SNPs was insufficient for analysis (i.e., fewer than 3 SNPs), the significance threshold was adjusted to p < 5E-06 [45]. Subsequently, SNP clumping was performed using the linkage disequilibrium (LD) structure from the 1000 Genomes Project to reduce LD bias and false-positive causal associations by identifying independent SNPs [46], thereby enhancing the accuracy and reliability of our analysis. Only independent SNPs meeting the criterion of R2 < 0.001 with any other SNP within 10 Mb were retained in this study. The robustness of instrumental variables was further assessed by using F-statistics, keeping those above 10 [47]. The SNPs associated with confounding factors were manually excluded using PhenoScanner v2 [48]. This study, relying on publicly available GWAS data, required no additional ethical approval. The study design is displayed in Fig. 1.

Fig. 1: The workflow of two-sample MR analysis.
figure 1

SNPs single nucleotide polymorphisms, IVW inverse variance weighted, MR mendelian randomization.

Full size image

MR statistical analysis

The two-sample MR analysis was performed to examine the causal association between iron levels in 16 subcortical brain regions and SCZ, MDD, ASD, ADHD, BD, ANX, OCD, and insomnia based on the three following principles: (1) instrumental variables used for analysis should exhibit a close correlation with the exposure, (2) selected instrumental variables should not be correlated with any confounding factor, and (3) the relationship between instrumental variables and outcome factors should be only mediated by the exposure factors, not through other alternative pathways (Fig. 1). We excluded SNPs in LD to ensure independence. Before the two-sample MR analysis, we conducted harmonization to exclude strand mismatches and ensure alignment in SNP effect sizes.

As the most efficient and widely used statistical approach, inverse variance weighting (IVW) was employed as the primary MR analysis method in this study, which is an unbiased estimator under the validity assumption (i.e., no horizontal pleiotropy among the instrumental variables) [49, 50]. Moreover, MR-Egger, weighted median, simple mode, and weighted mode were employed as additional methods to assess the reliability of the main results [51]. The weighted median method exhibits greater resistance to outliers and can mitigate the influence of extreme values on the results to a certain extent, and the weighted mode could better control the effect of genotype frequency differences on the analysis results and improve the robustness and accuracy of the analysis [50]. However, neither method is completely unbiased. MR-Egger introduces an intercept term to consider pleiotropy, but its estimation efficiency is lower [50]. Simple mode is typically utilized to present frequency or proportion comparisons between control and experimental groups regarding genotypes or phenotypes. However, it is not directly employed for causal effect estimation. Furthermore, we performed Cochran’s Q-test along with MR-Egger and IVW methods to assess heterogeneity, and MR-Egger regression was used to assess potential horizontal pleiotropy in instrumental variables, thereby ensuring the sensitivity and reliability of MR results [52]. MR estimates were recalculated after excluding the outlier SNPs identified through MR-PRESSO [51]. In the presence of heterogeneity (p < 0.05), a multiplicative random-effects IVW approach was performed to report results [53]. Finally, a leave-one-out analysis was performed to detect potential outliers in instrumental variables, aiming to assess the impact of each data point on the overall results [54]. The Bonferroni calibration correction method was employed to address multiple comparisons, ensuring statistical rigor. The significance thresholds in the MR analysis were set at p < 3.125E-03 (p < 0.05/16 exposures) to account for the inclusion of brain iron levels from 16 distinct subcortical brain regions as exposure variables. The p-values between 0.05 and 3.125E-03 were considered trends, suggesting marginal causal association [55]. Two-sample MR analyses between subcortical iron levels and structure, subcortical structure, and psychiatric disorders were performed to determine potential mediating effects, corresponding to statistically significant results in two-sample MR of iron levels in subcortical brain regions and psychiatric disorders. MVMR was then performed to assess the associations and possible mediating effects of subcortical structure between iron levels and psychiatric disorders (Supplementary Fig. 1 and Supplementary Table 2) [56]. All statistical analyses were performed using R software (version 4.3.1) with the TwoSampleMR (version 0.5.7) and Mendelian randomization package [56, 57].

Results

Following a thorough selection process, the number of SNPs chosen for MR analysis ranged from 3 to 129. Notably, all the chosen SNPs exhibited F-statistics above 10, highlighting their robustness as instrumental variables in our analysis (Supplementary Tables 4 and 8).

Forward MR: changes in brain iron leading to psychiatric disorders

In forward MR analysis, the iron levels in the left substantia nigra (OR = 0.94, 95% CI = 0.91–0.97, p < 0.001) showed a negative association with the risk for MDD (Fig. 2A, B), while the iron levels in the left putamen (OR = 1.11, 95% CI = 1.04–1.19, p = 0.002) showed a positive association with the risk for ASD (Fig. 2D, E) in IVW method. Leave-one-out analysis suggested that the MR estimate was not driven by a single SNP (Fig. 2C, F). The symmetry of the funnel plot indicated that there was no significant reporting bias, suggesting the reliability and robustness of the results (Supplementary Figs. 2 and 3). Additionally, there were seven trends discovered in forward MR analysis (3.125E-03 < p < 0.05). Specifically, iron levels in the left substantia nigra, bilateral pallidum, and right caudate were negatively associated with a lower risk of SCZ, MDD, and insomnia, respectively. Conversely, iron levels in the right thalamus, right hippocampus, and left thalamus positively correlated with the risk of ASD, ADHD, and OCD, respectively. Consistent results were obtained from other supplementary MR methods (Fig. 3). MR-Egger regression revealed no horizontal pleiotropy (all p > 0.05), and Cochran’s Q-test indicated no heterogeneity except for the MR estimate in iron levels in the right pallidum and MDD (Table 1 and Supplementary Table 6).

Fig. 2: Changes in brain iron levels in the left substantia nigra and left putamen lead to MDD and ASD, respectively.
figure 2

A Scatter plot of the causal association between iron levels in the left substantia nigra and MDD. The slope represents the strength of causal association. B Forest plot of the causal associations between SNPs related to iron levels in the left substantia nigra and MDD. The red line represents the MR effect size. C Leave-one-out plot of the causal association between iron levels in the left substantia and MDD. D Scatter plot of the causal association between iron levels in the left putamen and ASD. The slope represents the strength of causal association. E Forest plot of the causal associations between SNPs related to iron levels in the left putamen and ASD. The red line represents the effect size. F Leave-one-out plot of the causal association between iron levels in the left putamen and ASD. MDD major depressive disorder, ASD autism spectrum disorder, IVW inverse variance weighted, MR Mendelian randomization, SNPs single nucleotide polymorphisms.

Full size image
Fig. 3: Causal effects of iron levels in specific brain regions on risk of psychiatric disorders.
figure 3

IVW inverse variance weighted, MR Mendelian randomization, nSNP number of single nucleotide polymorphism, OR odds ratio, CI confidence interval, SCZ schizophrenia, MDD major depressive disorder, ASD autism spectrum disorder, ADHD attention-deficit/hyperactivity disorder, OCD obsessive-compulsive disorder.

Full size image
Table 1 Sensitivity analysis of significant and marginally significant results in MR analyses.
Full size table

Reverse MR: changes in brain iron caused by psychiatric disorders

In reverse MR analysis, there were eight trends between psychiatric disorders and iron levels in specific subcortical brain regions. Specifically, there were negative causal associations between SCZ and the left hippocampus and bilateral substantia nigra, ASD and the left putamen, ADHD and the left accumbens, ANX and the right thalamus, BD and the right accumbens, as well as a positive causal association between ADHD and the left thalamus (Fig. 4). Cochran’s Q-test revealed no horizontal pleiotropy (all p > 0.05), and heterogeneity tests indicated no heterogeneities except for the MR estimate in SCZ (Table 1 and Supplementary Table 10).

Fig. 4: Causal effects of psychiatric disorders on iron levels in specific brain regions.
figure 4

IVW inverse variance weighted, MR Egger Mendelian randomization Egger, nSNP number of single nucleotide polymorphism, OR odds ratio, CI confidence interval, SCZ schizophrenia, MDD major depressive disorder, ASD autism spectrum disorder, ADHD attention-deficit/hyperactivity disorder, BD bipolar disorder, ANX anxiety disorders.

Full size image

MVMR: changes in brain iron caused by psychiatric disorders

MVMR analysis showed that increased gray matter volume in the right caudate was positively correlated with the risk of insomnia after adjusting for iron levels (p < 0.05), as shown in Supplementary Table 16. This causal association remained significant after correction for iron levels. However, no mediating effect of the subcortical structure was observed in the association of iron levels with psychiatric disorders (Supplementary Table 18).

Discussion

This study is the first to examine bidirectional causal associations between iron levels in subcortical regions and psychiatric disorders. The study showed that iron levels in the left substantia nigra were negatively associated with the risk of MDD, and suggestively with SCZ. The iron levels in the left putamen were positively associated with the risk of ASD. In addition, the study observed several trends between iron levels in various subcortical regions and psychiatric disorders. These findings provide new evidence about the brain iron changes in psychiatric disorders and provide potentially novel considerations of treatment development for early prevention and intervention through the modulation of brain iron levels.

This study found that the iron levels in the left substantia nigra were negatively associated with the risk of MDD, suggesting that iron deficiency, rather than iron overload, may play a significant role in the development of MDD. Iron deficiency in the substantia nigra could affect dopamine production because iron serves as a critical cofactor for tyrosine hydroxylase (TH), which is essential for dopamine synthesis in TH-positive neurons [58]. Moreover, iron in the substantia nigra pars reticulata has been shown to be involved in the formation of myelin sheath [59], potentially increasing the risk of MDD. Brain iron deficiency may also lead to depression through the downregulation of the hippocampal glucocorticoid receptor signaling pathway [60]. Prior observational studies reported increased iron levels in the bilateral putamen in MDD patients [14], and an independent factor between higher thalamic iron levels and depressive symptoms in older adults [15]. The inconsistencies with our findings could be attributed to several factors: the inherent limitations of observational studies, such as limited sample size and susceptibility to confounding biases, and the potential influence of MRI scanners with different field strengths or manufacturers used for the detection of abnormalities.

This current study indicated that iron levels in the left putamen were positively associated with the risk of ASD. Iron overload in the brain, particularly in the left putamen, is closely associated with neurodegenerative alterations, such as the generation of reactive oxygen species, heightened inflammatory responses [61], and ferroptosis [62], which could contribute to the pathogenesis of ASD [63, 64]. Individuals with ASD showed hypoactivation [65] and reduced gray matter volume [66] in the putamen which is crucial for motor response to environmental stimuli [67], potentially indicating the disruption in the processing of visual and motor stimuli. Consequently, the iron overload in the left putamen may increase the risk of ASD via triggered oxidative stress inextricably and ferroptosis [68], highlighting a potential pathogenic mechanism.

The forward MR analysis further revealed a marginal negative causal association between iron levels in the left substantia nigra and SCZ, suggesting that iron deficiency in the left substantia nigra may contribute to the development of SCZ [13]. Importantly, this finding, together with a similar relationship observed in MDD, indicated a shared pathogenic alteration between MDD and SCZ, possibly involving dopaminergic dysfunction within the nigrostriatal pathway [69]. The clinical features and other pathological alterations shared by SCZ and MDD further support this finding. Both disorders are characterized by emotional and cognitive dysfunction [70, 71], which might be related to dysregulated dopaminergic activity within the nigrostriatal pathway [72, 73]. Moreover, neuroimaging studies, including positron emission tomography, have revealed specific dopaminergic dysfunctions in the substantia nigra and related neural circuits in SCZ and MDD patients [74, 75]. Reduced iron levels in the left substantia nigra could impair the function of the D2 receptor [76], a critical presynaptic dopamine receptor rich in iron within the nigrostriatal pathways, suggesting that alterations in brain iron levels could affect dopaminergic neurotransmission, thereby contributing to the pathology of SCZ and MDD. These findings suggested iron deficits in the substantia nigra might not only represent a common pathological mechanism but also a potential therapeutic target for SCZ and MDD. Conversely, the reverse MR analysis revealed a marginal negative causal association between SCZ and iron levels in bilateral substantia nigra. The bidirectional associations between iron in the substantia nigra and SCZ suggest a potential feedback loop where SCZ affects brain iron levels, which in turn may exacerbate or influence the course of the disorder.

This study also revealed several other marginal bidirectional causal associations between iron levels in various subcortical brain regions and psychiatric disorders (such as ADHD, OCD, insomnia, ANX, and BD), suggesting a complex role of brain iron in psychiatric disorders. For instance, the association between ADHD and altered iron levels in the right hippocampus highlights the potential role of iron in the pathophysiology of ADHD. Concurrently, ADHD potentially influences iron regulation in the left accumbens and thalamus, indicating a complex, interconnected pathogenic cycle. Although no direct studies have yet reported the correlation between thalamic iron and ANX, the finding from Kuchcinski et al.’s study indicated that iron accumulation after thalamic infarction is independently associated with post-stroke anxiety supports this correlation [77]. Similarly, an observational study revealed that elevated brain iron levels in the caudate nucleus in patients with restless legs syndrome (RLS) contribute to symptomatic improvement, whereas RLS as a sleep disorder often manifests as insomnia [78]. Few studies have examined brain iron changes in patients with OCD and BD, highlighting the need for further investigation to reveal the mechanisms underlying these relationships.

Our MVMR analysis found no mediation effect of structural changes in the association of iron levels with psychiatric disorders in subcortical brain regions. Iron, as a critical cofactor in brain biochemical and metabolic processes, affects not only structural integrity but also neurotransmitter synthesis and signaling [5, 79]. Changes in subcortical iron may be associated with psychiatric disorders through other complex pathways, possibly not involving direct structural changes.

This study has several limitations. This study is based on the European population, limiting the generalizability of our findings to populations in other regions. Additionally, our analyses only used brain iron levels in 16 subcortical nuclei available from GWAS data of the UK Biobank Brain Imaging. Unfortunately, data on iron levels in other brain regions were not available. However, it’s noteworthy that this study primarily analyzed iron levels in subcortical regions due to their well-documented and significant links to psychiatric disorders [80], suggesting that these areas are more critical for our research focus than other brain regions. Finally, two marginally significant casual associations in the reverse MR analyses differed from the results obtained using alternative complementary MR methods compared to the IVW method. These associations between SCZ and the iron levels in the left hippocampus, as well as BD and the iron levels in the right accumbens, should be interpreted with caution.

Conclusion

In conclusion, this study demonstrated that lower iron levels in the left substantia nigra are associated with an increased risk of MDD, and marginally significantly associated with SCZ, indicating a potential shared pathogenic mechanism underlying these disorders. Higher iron levels in the left putamen may lead to the development of ASD. Additionally, several other marginally significant casual associations were identified between brain iron levels and psychiatric disorders, highlighting the complex and dynamic interaction between brain iron regulation and these disorders. There is no mediating effect of structural changes in the association between iron levels and psychiatric disorders in subcortical brain regions. Further studies are needed to investigate the direct causal associations between brain iron levels and psychiatric disorders and elucidate the specific physiological mechanisms involved.