Introduction

Humans are ubiquitously exposed to aluminum (Al) through various pathways1. Workers in Al-related industries face substantially higher exposure levels2, and Al may cross the blood-brain barrier, accumulate in neural tissues, and induce neurotoxic effects3. Consequently, occupational Al exposure is recognized as a significant occupational health hazard. Substantial evidence confirms that long-term Al exposure correlates strongly with cognitive decline, particularly in memory and processing speed4,5,6, and may contribute to neurodegenerative diseases like mild cognitive impairment (MCI) and Alzheimer’s disease (AD)7,8. Additionally, previous studies9 also identified a dose-response relationship between Al exposure and cognitive impairment. However, the precise neural mechanisms underlying cognitive impairment induced by occupational Al exposure remain unclear and warrant further investigation.

MCI represents an intermediate cognitive state between normal aging and dementia. Increasing evidence from neuroimaging studies has provided valuable insights into the underlying mechanisms of MCI10,11,12. Cognitive decline in MCI is often closely associated with atrophy of related gray matter (GM) structures. Research has shown that, compared to age-matched controls, MCI participants exhibit significantly reduced gray matter volume (GMV) in areas such as the hippocampus, temporal gyrus, thalamus, amygdala, caudate nucleus (CAU), and other brain regions12,13,14,15. MCI induced by occupational Al exposure may also be associated with GM atrophy. A recent study suggested that the reduced GMV of the default mode network may represent a neural mechanism of Al-induced MCI16. However, that study had two major limitations: it lacked a non-exposed control group, which made it impossible to disambiguate the effects of occupational Al exposure from other potential confounding factors; furthermore, non-negative matrix factorization is designed to extract patterns of co-varying networks across the brain, which inherently limits its ability to pinpoint regional atrophy. To address these limitations, our study introduces a non-exposed control group to isolate Al-specific effects and employs voxel-based morphometry for precise whole-brain localization of atrophy. This approach is particularly well-suited for identifying specific vulnerable structures and to directly correlate structural changes with exposure levels and cognitive metrics.

In this study, we hypothesized that specific patterns of GMV atrophy would be associated with cognitive impairment in Al-induced MCI. To test this hypothesis, we first obtained the GMV of each participant from structural MRI. Intergroup comparisons were then conducted to identify differences in GMV among the groups. Finally, partial correlation analyses were conducted to explore the neural substrates associated with cognitive performance. This study contributes to understanding the neuroimaging mechanisms of MCI induced by occupational Al exposure.

Methods

Participants

This study focuses on MCI participants with more than 10 years of occupational Al exposure in an Al factory (Al-MCI), using a cognitively normal group with comparable occupational Al exposure (Al-HC) and an age-matched healthy control group without any occupational Al exposure history (Non-Al-HC) as controls. All participants were native Mandarin Chinese speakers. We excluded individuals who met any of the following criteria: (1) left-handedness, (2) contraindications to MRI, (3) a history of psychiatric or neurological disorders, (4) a family history of neurodegenerative diseases, (5) current serious illnesses, or (6) long-term use of medications that could potentially enhance or impair cognitive function. Additionally, participants in the Non-Al-HC group were required to have no history of long-term occupational Al exposure. To minimize the risk of significant Al exposure, we meticulously investigated all participants’ work and residential histories through questionnaires and interviews. Individuals were also excluded from this group if they had: (1) a history of employment in any Al-related or other heavy metal industries; or (2) a residential history in immediate proximity to major industrial complexes or known sources of industrial Al emissions.

Clinical and neurocognitive assessments

Clinical data were collected from occupationally Al-exposed participants before the MRI scan, including age, sex, body mass index (BMI), education level, employment history, and general health status. In the early morning, 10mL of venous blood was extracted from each occupationally Al-exposed participants into a test tube containing EDTA at the blood collection laboratory of the Al factory employee hospital, and the plasma Al concentration was determined.

Cognitive function was evaluated using a battery of neuropsychological tests. Verbal memory was assessed using the Rey Auditory Verbal Learning Test (AVLT), which included immediate recall (AVLT-I), delayed recall (AVLT-D), and memory recognition (AVLT-R). Cognitive processing speed and executive function were assessed using the Trail Making Test (TMT)17. Overall cognitive function was evaluated using the Montreal Cognitive Assessment (MoCA), and based on the MoCA score, occupationally Al-exposed participants were classified into two groups: the Al-MCI (MoCA < 26) and the Al-HC (MoCA ≥ 26)18,19.

The Non-Al-HC group provided demographic data (age, sex, and BMI) and underwent structural MRI scanning only. Cognitive assessments and plasma Al concentration measurements were omitted because these participants were recruited from the community without any history of occupational Al exposure. This group served primarily as a reference to compare GMV differences across varying levels of Al exposure or cognitive status.

MRI imaging data acquisition and processing

MRI data of all participants were acquired utilizing a 3.0 T MRI scanner (Skyra, Siemens, Erlangen, Germany) featuring a 32-channel head coil. A 3D T1-weighted image of the sagittal plane was obtained using a magnetization-prepared rapid gradient echo sequence with the following setting: time repetitive (TR) = 2.53 s, time echo (TE) = 2.01 ms, flip angle = 7◦, matrix size = 256 × 256, field of view (FOV) = 256 mm×256 mm, slice thickness = 1 mm, number of slices = 192. A radiologist examined these images to exclude participants with organic brain injuries based on T1-weighted images.

For image analysis, we used the automated Computational Anatomy Toolbox (CAT12, http://dbm.neuro.uni-jena.de/cat12/), running within the Statistical Parametric Mapping (SPM12, Welcome Department of Imaging Neuroscience, London, UK) under MATLAB (r2013b). Initially, images were manually corrected for orientation in SPM12. High-resolution structural images were then segmented into GM, white matter (WM), cerebrospinal fluid (CSF), bone, non-brain soft tissue, and background using the ICBM Tissue Probabilistic Atlases from SPM12. The total intracranial volume (TIV) was computed for each subject at this stage by summing the native-space volumes of GM, WM, and CSF. After segmentation, Differential Anatomic Registration Through Exponentiated Lie Algebra was applied for anatomical registration, normalization, and modulation20. All images were resampled to 1.5 × 1.5 × 1.5 mm³ and normalized to the Montreal Neurological Institute (MNI152) space. To preserve the accuracy of volume measurements, the inverse Jacobian matrix of the local transformation was applied to modulate segmented GM. A 6 mm full width at half maximum Gaussian kernel was applied to smooth the data, compensating for residual inter-subject anatomical variability and improving the signal-to-noise ratio, while its moderate size preserved the anatomical details of subcortical structures21,22. The Automated Anatomical Labeling (AAL) template23 was loaded into CAT12, and MRI images of each participant were registered to the AAL template to ensure proper alignment of brain regions. Finally, the volumes of 90 brain regions were automatically extracted for further analysis.

Statistical analyses

Demographic and cognitive data were analyzed using SPSS software (Version 22.0). After confirming that the data met the assumptions of normality and homogeneity of variances, the analysis of variance (ANOVA) and two-sample t-tests were performed to compare the demographic and cognitive test results. Additionally, analysis of covariance (ANCOVA) was applied with age and TIV as covariates to investigate the differences in GMV among the Al-MCI, Al-HC, and Non-Al-HC groups for each component. For regions showing a significant main effect of group, post-hoc pairwise comparisons were conducted using the least significant difference (LSD) test. Partial correlation analyses were conducted between the GMV of regions showing significant group differences and all cognitive scores within the occupationally Al-exposed groups (Al-MCI and Al-HC groups), controlling for age, BMI, and education level. All results were corrected by the false discovery rate (FDR) method. p < 0.05 were considered statistically significant.

To further investigate the indirect effect of left CAU volume on the relationship between AVLT-I and MoCA scores, mediation analyses were performed using the SPSS macro PROCESS (with the bootstrapping method). AVLT-I was treated as the independent variable (X), the left CAU volume as the mediator variable (M), and the MoCA score as the dependent variable (Y). Age, BMI, and education level were included as control variables. The model consisted of four paths: path a represents the effect of X on M, path b represents the effect of M on Y, path c represents the total effect of X on Y, and path c’ represents the direct effect of X on Y. The indirect effect was calculated as c - c’ (or a × b). The indirect effect was considered statistically significant if the 95% confidence interval (CI) obtained through 5,000 iterations of the bootstrapping method did not include 0.

Results

Demographic characteristics and cognitive test results

The demographic characteristics, plasma Al concentrations, and cognitive performance are summarized in Table 1. A total of 86 individuals were included in the study: 30 in the Al-MCI group, 26 in the Al-HC group, and 30 in the Non-Al-HC group. All participants across the three groups were male. There was no significant difference in age (F = 0.003, p = 0.997) and TIV (F = 2.251, p = 0.112) among the groups, although a significant group difference was found in BMI (F = 3.774, p = 0.027). Additionally, the Al-MCI group had significantly lower education levels than the Al-HC group (t = −3.462, p = 0.001).

Table 1 Demographic and cognitive test results of Al-MCI, Al-HC, and Non-Al-HC Groups.
Full size table

Plasma Al concentrations were measured only in the Al-MCI and Al-HC groups. The Al-MCI group (42.74 ± 26.00 µg/L) showed no significant difference compared to the Al-HC group (32.52 ± 30.86 µg/L; t = 1.345, p = 0.184). Plasma Al concentrations were not assessed in the Non-Al-HC group due to the absence of occupational Al exposure.

Compared to the Al-HC group, the Al-MCI group exhibited significantly poorer cognitive performance. Specifically, they had lower total MoCA scores (t = −15.693, p < 0.001) and reduced sub-scores in visuospatial/executive (t = −8.799, p < 0.001), naming (t = −5.291, p < 0.001), attention (t = −2.395, p = 0.020), abstraction (t = −6.574, p < 0.001), orientation (t = −2.946, p = 0.005), and language (t = −7.200, p < 0.001). In terms of memory performance, the Al-MCI group scored lower on AVLT-I (t = −4.222, p < 0.001), AVLT-D (t = −3.955, p < 0.001), and AVLT-R (t = −2.730, p = 0.009). Additionally, the Al-MCI group also required more time to complete both TMT-A (t = 3.454, p = 0.001) and TMT-B (t = 3.490, p < 0.001), indicating slower cognitive processing speed and impaired executive function.

Differences in GMV among Al-MCI, Al-HC, and Non-Al-HC groups

After controlling for age and TIV, the ANCOVA revealed significant group differences in GMV only in the bilateral CAU (CAU_L: F = 13.699, p < 0.001; CAU_R: F = 10.173, p < 0.001, FDR-corrected). No other regions showed significant group differences. Post-hoc LSD tests indicated that GMV in the bilateral CAU differed significantly between all three pairs of groups (p < 0.05 for all pairwise comparisons, FDR-corrected). Moreover, a clear gradient of volume reduction was observed: the Al-MCI group showed the smallest CAU volumes, followed by the Al-HC group, while the Non-Al-HC group exhibited the largest volumes (Al-MCI < Al-HC < Non-Al-HC). Detailed results are provided in Table 2, and Fig. 1 illustrates the locations of the bilateral CAU.

Table 2 Gray matter volume differences among Al-MCI, Al-HC, and Non-Al-HC groups.
Full size table
Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Locations of the bilateral caudate nuclei. The 3D images of the bilateral caudate nuclei were generated using BrainNet Viewer42 (Version 1.7, available at: https://www.nitrc.org/projects/bnv/). Comparative analysis of gray matter volumes among the Al-MCI, Al-HC, and Non-Al-HC groups revealed significant differences in the volumes of the bilateral caudate nuclei. The left CAU is highlighted in red, and the right CAU is highlighted in yellow. L: left; R: right.

Correlation between bilateral CAU volumes and cognitive performance

Based on the significant group differences identified in the bilateral CAU, we focused subsequent partial correlation analyses on these regions. After controlling for age, BMI, and education level, partial correlation analyses revealed positive correlations between bilateral CAU volumes and MoCA scores within the Al-exposed groups (CAU_L: r = 0.378, p = 0.004; CAU_R: r = 0.315, p = 0.018; FDR-corrected, Fig. 2a and b). Furthermore, the left CAU volume was positively correlated with abstract and language scores (abstract score: r = 0.294, p = 0.028; language score: r = 0.321, p = 0.016; FDR-corrected). In contrast, the positive correlation between right CAU volume and abstract score (r = 0.274, p = 0.041) did not survive FDR correction. The correlations between bilateral CAU volumes and MoCA sub-scores are visually presented in Fig. 3.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Correlation and mediation analysis of bilateral caudate nuclei volumes and cognitive scores. (a, b) The gray matter volumes of the bilateral caudate nuclei are positively correlated with MoCA scores. (c) The gray matter volume of the left CAU shows a positive correlation with AVLT-I. (d) AVLT-I can influence MoCA scores by altering the gray matter volume of the CAU_L. Age, BMI, and education level were controlled in the model. MoCA, Montreal cognitive assessment; CAU_L, left caudate nucleus; CAU_R, right caudate nucleus; AVLT-I, Rey Auditory Verbal Learning Test immediate recall.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Correlation between bilateral caudate nuclei volumes and MoCA sub-scores. (a) The left CAU is positively correlated with abstraction score and language scores. (b) The right CAU is positively correlated with abstraction score. CAU_L, left caudate nucleus; MoCA, Montreal cognitive assessment; CAU_R, right caudate nucleus.

In terms of memory and executive function, the left CAU volume was positively correlated with AVLT-I (r = 0.289, p = 0.030, FDR-corrected, Fig. 2c), while the right CAU volume showed a negative correlation with TMT-A error responses (r = −0.306, p = 0.022, FDR-corrected), indicating that smaller volume was associated with more errors. Additionally, AVLT-I was positively correlated with MoCA scores (r = 0.580, p < 0.001, FDR-corrected), while the negative correlation between TMT-A error responses and MoCA scores (r = −0.268, p = 0.046) did not survive FDR correction. All correlation results are summarized in Table 3.

Table 3 Correlation between gray matter volumes and cognitive scores.
Full size table

Mediation analysis in our study revealed that AVLT-I contributed to cognitive impairment by affecting left CAU volume (indirect effect = 0.0563, 95% CI = [0.0092, 0.1452], p < 0.05, Fig. 2d).

Discussion

In this study, we investigated the intergroup GMV differences and the relationship between GMV and cognitive function. Our primary finding was that bilateral CAU atrophy may represent a key neuroanatomical characteristic of MCI in the context of occupational Al exposure. The degree of CAU atrophy was associated with the severity of cognitive impairment among exposed individuals. Notably, the left CAU is critically involved in language processing and episodic memory, whereas the right CAU primarily supports executive cognitive functions.

Our analysis revealed a graded pattern of CAU atrophy that closely corresponded to both Al exposure and cognitive status: the most severe atrophy was observed in the Al-MCI group, intermediate but significant atrophy in the Al-HC group, and the largest volumes in the Non-Al-HC group. The significant atrophy observed in the Al-HC group suggests that structural brain changes can occur before the onset of overt cognitive symptoms, indicating that CAU atrophy is not merely a consequence of MCI but may represent an early signature of Al-induced neurotoxic processes. Furthermore, despite comparable plasma Al concentrations between the two exposed groups, the degree of CAU atrophy differed, highlighting the role of individual susceptibility in determining whether Al exposure leads to cognitive decline. Long-term Al exposure has been shown to exert neurotoxic effects on the brain via mechanisms such as oxidative stress, inflammation, and apoptosis24,25. Al has also been reported to impair iron metabolism and promote neurotoxicity by activating astrocytes and pro-inflammatory markers, leading to neuronal death26. The CAU, a crucial component of the striatum, regulates cognitive, motor, and emotional regulation through its extensive connections within the corticostriatal circuit27,28. Atrophy of the CAU may therefore disrupt these neural pathways, contributing to cognitive impairment. Previous studies have demonstrated that reduced CAU volume, or altered cortex-to-caudate ratios, are closely associated with cognitive decline in MCI and AD29,30,31 and serves as important markers for early AD diagnosis32. Our findings extend this evidence by showing that CAU atrophy not only occurs in individuals with Al-related MCI but also appears in exposed yet cognitively normal individuals, underscoring its potential as an early neuroimaging signature of Al-related neurotoxicity.

Our study also revealed a positive correlation between bilateral CAU volume and overall MoCA scores, suggesting that a reduction in CAU volume may serve as an indicator of cognitive decline, underscoring its critical role in cognitive function32,33. Specifically, left CAU volume was associated with MoCA abstract scores. Abstract thinking is a higher-order cognitive function closely linked to the prefrontal cortex34. The CAU may contribute to the regulation of abstract thinking by integrating and processing information through its neural network connectivity with the prefrontal cortex. Additionally, left CAU volume was positively correlated with language scores, consistent with existing research, and highlights the important role of the CAU in language tasks35,36,37. Our findings indicate that only left CAU volume is associated with language scores, which may be linked to the dominant role of the left hemisphere in language processing38.

This study also identified a positive correlation between left CAU volume and AVLT-I. Similarly, other studies have reported significant associations between changes in CAU volume and both episodic memory and semantic processing39, potentially linked to the presence of D2 receptors within the CAU40. In contrast, right CAU volume was negatively correlated with TMT-A error responses, consistent with previous research indicating that a higher number of errors corresponds to slower processing speed and greater atrophy of the right CAU41. In summary, these findings further emphasize the importance of the CAU in cognitive processes, suggesting that assessing CAU volume could provide valuable insights into cognitive decline in individuals with occupational Al exposure.

Interestingly, mediation analysis revealed that left CAU volume significantly mediated the relationship between AVLT-I and MoCA scores, suggesting that the decline in immediate memory function may contribute to overall cognitive impairment by altering the left CAU volume. This finding underscores the critical role of the CAU in cognitive function, particularly in the connection between memory performance and overall cognitive ability39. In contrast, the lack of a significant correlation between TMT-A error responses and MoCA scores precluded a mediation analysis for the right CAU, indicating that the right CAU may be more specifically related to processing speed rather than serving as a mediator for overall cognition. Future longitudinal studies with larger samples are warranted to confirm left CAU atrophy as a mediating pathway in cognitive decline and to further elucidate the distinct neurobehavioral correlates of right CAU integrity.

Limitations

Despite these significant findings, several limitations should be acknowledged. Firstly, the relatively small sample size may limit statistical power. Secondly, all participants in our study were male, reflecting the demographic reality of the Al factory where most Al-exposure workers are male. While this controls for sex-based confounders, it undoubtedly restricts the generalizability of our findings to female populations. We recognize this as an important caveat, and validating these results in female workers remains a crucial objective for future research. Thirdly, the absence of cognitive data and objective biomarker confirmation for the Non-Al-HC group is a limitation. While this group was primarily included as a neuroanatomical reference to establish normative brain volumes, the lack of cognitive data prevents a direct comparison of cognitive performance between exposed and non-exposed individuals. Furthermore, although we applied stringent criteria to define non-exposure, the inability to biochemically verify the non-exposed status remains a constraint. We acknowledge these shortcomings, and future studies will incorporate comprehensive cognitive assessments and a wider range of clinical and biological indicators for all participant groups to address these gaps. Finally, the cross-sectional design of this study precludes conclusions about the causal or long-term effects of Al exposure on cognitive function. Longitudinal research is needed to track structural and functional brain changes associated with prolonged exposure over time.

Conclusions

In conclusion, long-term occupational Al exposure may contribute to the development of MCI. Bilateral CAU atrophy was significantly associated with cognitive decline in Al-induced MCI and may represent a specific neuroimaging signature for detecting cognitive impairment in occupationally Al-exposed workers.