, Introduction
Chronic kidney disease (CKD) is a progressive condition affecting over 10% of the global population, with the number of patients with CKD steadily rising [1]. A wide range of disorders can develop as a consequence of the loss of kidney function. Patients with CKD typically face a significantly higher burden of cardiovascular disease, hypertension, anemia, heart failure, stroke, and altered bone mineral metabolism than those without CKD [2–7].
Moreover, there is growing interest in the association between CKD and mental health challenges, including depression, bipolar disorder, and anxiety [8,9]. Depression is particularly common in patients with kidney impairment, with prevalence rates of approximately 20% to 25% [10]. Identifying the causal relationship between kidney function and mental illnesses is essential, as preventive measures targeting kidney function may help reduce mental disorders, though this area remains underexplored.
A recent Mendelian randomization (MR) study demonstrated that psychological well-being is causally linked to improved kidney function, while depression increases the risk of kidney impairment [11]. This evidence highlights the need for mental health surveillance to manage kidney function decline. However, evidence for reverse causal relationships between kidney function and mental illnesses remains limited, requiring further investigation into the causal link with brain structure changes. Furthermore, these observations contribute to the emerging concept of the “kidney-brain axis,” suggesting that renal dysfunction may directly influence brain structure and function, thereby playing a pivotal role in the development of CKD-associated neuropsychiatric disorders [12,13].
Previous studies have shown that brain volume abnormalities, including those in gray and white matter, as well as hippocampal and amygdala volumes, are associated with depression and anxiety [14–18]. Additionally, in the context of the kidney–brain axis, kidney dysfunction has been independently associated with white matter intensity and has been shown to influence cortical structure [12,13]. These results suggest that brain structure abnormalities may mediate the relationship between kidney impairment and mental disorders.
In this study, we evaluated the causal effects of kidney impairment, measured by creatinine- and cystatin-C-based estimated glomerular filtration rate (eGFR), on specific regions of brain volume traits using two-sample MR analysis in the discovery phase. We also investigated whether these brain volume changes are associated with mental health outcomes using two-sample MR in the discovery phase. Finally, in the validation phase, we confirmed causal links between kidney impairment and specific brain volumes, as well as between brain volumes and mental disorders, using one-sample MR analysis in the UK Biobank (UKB).
Methods
Ethics approval and consent to participate
This study was exempted from the Chonnam National University Hospital Institutional Review Board review (No. CNUH-EXP-2024-361) as it used publicly available cohort information.
Overview
This study aims to investigate the following using MR: (i) associations between eGFR and brain volume traits; (ii) associations between significant brain volume traits identified in (i) and mental disorder traits; (iii) the associations between CKD and brain volume traits; and (iv) associations between significant brain volume traits identified in (iii) and mental disorder traits. We identified these four associations (i through iv) in the discovery phase and validated them in a subsequent validation phase (Fig. 1).
In the discovery phase, instrumental variables (IVs) were selected from cohorts including the UKB, Chronic Kidney Disease Genetics (CKDGen), and a combined cohort of CKDGen and UKB. The outcomes in the discovery phase included 507 genome-wide association studies (GWAS) summary datasets on various brain volume traits, such as the amygdala, obtained from the UKB. Additionally, 242 GWAS summary datasets on mental disorder traits were collected from various cohorts, such as the UKB (Fig. 1A; Supplementary Table 1, available online). The two-sample MR, specifically using the inverse-variance weighted (IVW) method, was applied in this phase.
In the validation analysis, we assessed the reproducibility of the associations identified in the discovery two-sample MR analysis using one-sample MR with the UKB cohort. We obtained data on creatinine-based eGFR (eGFRcrea), cystatin C-based eGFR (eGFRcys), the CKD trait (case/control), amygdala volume traits, and psychiatric disorder traits in the UKB (Fig. 1G). For one-sample MR, two-stage least square (TSLS) regression was used for continuous outcome variables, and two-stage residual inclusion (TSRI) regression was used for categorical outcome variables.
Data sources of the discovery phase
The exposure and outcome cohorts for the discovery phase of two-sample MR analysis were sourced from GWAS summary data, including eGFRcrea, eGFRcys, brain volumes, and psychiatric disorders, from the Integrative Epidemiology Unit OpenGWAS Project (data sources in Fig. 1A and Supplementary Table 1 [available online]) [19]. The eGFR GWAS summary data were based on natural logarithmically transformed eGFR values [20]. For brain volume traits data, detailed processes for handling subgroups and missing data for each brain image-derived phenotype (IDP) were not identified because the number of brain volume GWAS samples varied across IDPs [21]. Additionally, GWAS summary data for CKD were obtained from the CKDGen consortium (Fig. 1A) [22,23].
Instrumental variable selection in the discovery phase
In the discovery phase, we selected single-nucleotide polymorphisms (SNPs) as IVs from summary data from exposure GWAS cohorts (Fig. 1A). Details of the exposure cohorts are described in the following section. For each exposure cohort, a GWAS p-value threshold of 1 × 10–5 [24,25] was applied with criteria in previous studies [19,26], including linkage disequilibrium (LD) clumping harmonization, and removal of palindromic SNPs, using the “TwoSampleMR” package (version 0.5.8) in R (R Foundation for Statistical Computing) [19]. Instrumental strength was assessed by calculating the F-statistics of selected SNPs as IVs using an established formula [27,28]. SNPs with F-statistic >10 were considered strong, reducing bias in MR analysis (Supplementary Tables 2–12, available online) [29].
To support the exclusion restriction assumption of MR, we implemented a rigorous, multi-step filtering protocol. Due to the persistent technical unavailability of PhenoScanner [30] since late 2024, we performed expert-curated manual screening by clinicians instead of a simplified automated search. Specifically, candidate SNPs were cross-referenced against high-confidence datasets from recent landmark studies [20,31,32] to identify and remove instruments associated with non-GFR-related metabolic phenotypes (e.g., creatinine and cystatin C metabolism) at a genome-wide significance level (p < 5 × 10–8). The list of SNPs excluded at this step is provided in Supplementary Table 13 (available online). Subsequently, we performed LD clumping to obtain the final IVs, and additional MR analyses were conducted to confirm the robustness of the causal estimates.
Exposure and outcome variables in the discovery phase
We established various exposure–outcome pairs across multiple cohorts to explore diverse causal relationships through two-sample MR analysis (Table 1). Two-sample MR was conducted with the IVW method in the “TwoSampleMR” R package (version 0.5.8) [19]. The cohorts included CKDGen and UKB, comprising participants predominantly of European ancestry.
Two-sample MR analysis of estimated glomerular filtration rate and brain volume traits
In the initial analysis, we examined the causal relationship between eGFR and brain volume traits. eGFRcrea and eGFRcys served as exposures, while 507 brain structural traits from the UKB constituted the outcomes (Fig. 1B). Statistically significant associations with brain regions, such as amygdala volume traits, were identified for further analysis.
Two-sample MR analysis of chronic kidney disease and amygdala volume traits
The relationship between CKD and amygdala volume traits was evaluated by using CKD as an exposure, while 24 amygdala traits served as outcomes (Fig. 1D). Patients with CKD were defined as cases if they had an eGFR of <60 mL•min–1 per 1.73 m2, while those with an eGFR of ≥60 mL•min–1 per 1.73 m2 were considered controls [23]. This two-sample MR aimed to identify specific amygdala subregions that might be linked to CKD, facilitating the exploration of possible brain-renal health connections.
Data sources for the validation phase
We employed one-sample MR analyses for the validation phase to replicate significant associations identified during the discovery phase. This included verifying the relationship between eGFR and brain volume traits, the association between amygdala volume traits and psychiatric disorders, and the link between CKD and amygdala volume traits (Fig. 1B–E). IV selection and one-sample MR methodologies were applied to ensure reliable replication of results. Data for these analyses were drawn from the UKB, comprising individuals of European ancestry (Fig. 1G).
The eGFRcrea and eGFRcys were calculated using serum creatinine and cystatin C levels (UKB field IDs: 23478, 30720), while amygdala subtype volumes were derived from subcortical volumetric segmentation and T1 images for approximately 46,000 participants. Mental health outcomes were sourced from the International Classification of Diseases (10th revision)-coded diagnoses, including depressive episodes, anxiety disorders, and dementia. Supplementary Table 14 (available online) provides details on traits and their UKB field IDs. Please refer to Supplementary Method 1 (available online) for details.
Instrumental variable selection in the validation phase
We performed GWAS analyses on 6,214,907 SNPs for 444,105 participants with available genotypes in the UKB imputation dataset, focusing on traits including eGFRcrea, eGFRcys, seven amygdala volume traits (i.e., volume of the amygdala in the right hemisphere [RH], lateral nucleus in the RH, cortical nucleus in the RH, corticoamygdaloid transition in the RH, medial nucleus in the RH, medial nucleus in the left hemisphere [LH], and whole amygdala), four mental disorder phenotypes (i.e., depressive episode, anxiety disorder, specific phobia, and post-traumatic stress), and the CKD phenotype (Supplementary Fig. 1, available online). In the validation stage, the criteria for CKD in participants were set to be the same as those used for CKD in the discovery stage (Supplementary Method 1, Supplementary Tables 15–25; available online) [19].
One-sample MR in the validation phase
We conducted one-sample MR in R (version 4.3.2) using the “ivreg” R package (version 0.6-2). We used the TSLS and TSRI regression methods to validate the discovery finding.
One-sample MR analysis of estimated glomerular filtration rate and brain volume traits
The validation analysis examined the causal relationship between eGFR and six brain volume traits identified in the discovery phase (Fig. 1H).
Secondary one-sample MR analysis of amygdala volume traits and psychiatric disorders
Amygdala volume traits associated with eGFR in the discovery phase served as exposures in a secondary one-sample MR analysis. The relationship between specific amygdala volume subregions and six psychiatric disorders was investigated (Fig. 1I).
One-sample MR analysis of chronic kidney disease and amygdala volume traits
The relationship between CKD and amygdala volume traits was validated by treating CKD as the exposure, while right and left medial nucleus volumes served as outcomes (Fig. 1J).
Secondary one-sample MR analysis of amygdala volume traits and psychiatric disorders
Amygdala volume traits significantly associated with CKD were further evaluated as exposures, exploring their connection to six psychiatric disorders, including acute stress (F430), anxiety disorder (F419), depressive episode (F329), isolated phobia (F402), post-traumatic stress (F431), and dementia (F030) (Fig. 1K).
Robustness in the validation phase
The power of the MR analysis was calculated using the mRnd tool (https://shiny.cnsgenomics.com/mRnd/) [33] (Fig. 1L) with a type-I error rate of 0.05 to assess the robustness of our findings in one-sample MR.
Results
Discovery phase: two-sample MR of kidney function and chronic kidney disease on amygdala volume and depression
We conducted a two-sample MR analysis to investigate the causal relationship between kidney function and brain volume and examine the causal effect of eGFR on brain volume traits (Fig. 2A). We set eGFRcrea and eGFRcys as exposures, with brain volume traits as outcomes. Among the 507 brain volume traits analyzed, amygdala-related traits were consistently identified as significant across both CKD and eGFR models. Due to this robust consistency, we focused on the amygdala as a primary brain structure affected by impaired kidney function (Supplementary Table 1, available online). The results revealed a statistically significant negative causal association between eGFRcrea and eGFRcys and amygdala volume traits, indicating that impaired kidney function is associated with increased amygdala volume. Specifically, six amygdala volume traits showed negative causal associations with both eGFRcrea and eGFRcys (Fig. 2B; Supplementary Table 26, available online). The associated amygdala regions were located in the RH, with four significant subregions identified: the accessory basal nucleus, cortical nucleus, cortico-amygdaloid transition area, and lateral nucleus. These findings from the two-sample MR analysis suggest that reduced kidney function is significantly associated with an increase in amygdala volumes. To further validate our findings (Fig. 2B) against potential bias from sample overlap, we conducted a ‘true’ two-sample MR analysis using the older version of the CKDGen dataset (CKDGen 2016 [32]), which does not include UKB participants. Despite the reduced sample size of the earlier GWAS, the results were consistent with our primary findings. The IVW estimates reaffirmed significant negative causal associations between eGFR and several amygdala subregion volumes (Supplementary Fig. 2, available online). This consistency across independent, non-overlapping cohorts provides strong evidence that the observed causal link is reliable and not an artifact of sample overlap.
Since the findings might be affected by some SNPs that were pleiotropically linked to creatinine/cystatin C metabolism rather than true kidney function, we removed SNPs associated with serum creatinine and cystatin C levels from the IVs and performed additional MR analyses. While the association between eGFRcrea and the volume of the lateral nucleus in the RH did not reach statistical significance after this adjustment, the remaining eleven associations remained robustly significant (Supplementary Fig. 3, available online). This high degree of reproducibility across different eGFR markers confirms the overall consistency of our findings, suggesting that the causal link is not driven by pleiotropic metabolic pathways.
We identified a statistical causal relationship between eGFR and amygdala volume traits in the previous section. We then investigated whether amygdala volumes associated with eGFR had a causal relationship with psychiatric disorders (Fig. 2C). Amygdala volume traits associated with eGFR were set as the exposures, and psychiatric disorders were set as the outcomes. We investigated whether amygdala volume traits consistently associated with various psychiatric disorders also exhibited causal associations in a consistent direction. As a result, we found a statistically significant positive causal relationship between amygdala volume and depression. Specifically, three amygdala volume traits (i.e., volume of the lateral nucleus, cortical nucleus, and whole amygdala in the RH) were positively associated with depressive symptoms (Fig. 2D; Supplementary Table 27, available online). Thus, MR analysis suggests that increases in amygdala volume associated with reduced kidney function present a significant risk for depression.
In this study, we have identified a statistically significant causal relationship between eGFR and amygdala volume traits, as well as between amygdala volume traits and psychiatric disorders. A decline in eGFR below the normal range indicates kidney dysfunction and levels below a certain threshold are diagnostic of CKD. Therefore, it is essential to investigate whether CKD, closely associated with eGFR, has a causal relationship with amygdala volume traits. The IVW method in two-sample MR was conducted to investigate causal relationships between CKD, eGFR, and amygdala volumes. Two-sample MR analysis revealed the statistically significant positive causality between CKD incidence and the two-volume amygdala traits (volume of the medial nucleus of amygdala nuclei in RH and volume of the medial nucleus of amygdala nuclei in LH) (Fig. 3A; Supplementary Table 28, available online).
We conducted two-sample MR using the two amygdala volume traits as exposures and depression symptom traits as outcomes to investigate whether the two-volume traits of the medial nucleus of the amygdala with the statistically significant causal relationships to CKD in the previous section are also causally associated with psychiatric disorders. The two-sample MR results revealed statistically significant positive causal relationships between the volume of the medial nucleus of the amygdala and depression symptoms (Fig. 3B; Supplementary Table 29, available online). Notably, the volume of the medial nucleus of the amygdala in RH was more significantly associated with depressive symptoms compared to that in LH.
Validation phase: one-sample MR of kidney function and chronic kidney disease on amygdala volume and depression
In the validation analysis using the UKB, we employed TSLS regression in the one-sample MR framework to validate the significant negative associations identified in the discovery analysis between eGFR and amygdala volume traits. Using TSLS regression, we examined the relationships between eGFRcrea, eGFRcys, and the volumes of six amygdala subregions (i.e., total amygdala volume, volume of the right lateral nucleus, volume of the right corticoamygdaloid transition, volume of the right cortical nucleus, volume of the right amygdala [Freesurfer automatic segmentation], and volume of the right accessory basal nucleus). The results confirmed significant negative associations between eGFRcrea and volumes of amygdala subtypes and the entire amygdala (T1 image-derived), as well as between eGFRcys and the volume of the cortical nucleus (Fig. 4A; Supplementary Table 30, available online).
The one-sample MR analysis, using the TSRI method, confirmed the two-sample MR findings by observing a negative correlation between eGFRcrea and the volumes of amygdala subtypes, as well as a positive correlation with the risk of depression-related mental disorders (Fig. 4B; Supplementary Table 31, available online). These results also align with those from the association analysis conducted via regression analysis (Fig. 4C, D).
The one-sample MR in the validation stage identified a significant association between CKD causality and the volume of the medial nucleus of the amygdala in RH (beta = 4.65, p = 0.043), confirming the two-sample MR results (Fig. 5A; Supplementary Table 32, available online).
The one-sample MR identified the significant associations between the volume of the medial nucleus of the amygdala in RH and dementia (beta = 0.252, p = 0.036), between the volume of the medial nucleus of amygdala in LH and dementia (beta = 0.379, p = 0.040) or anxiety disorder (beta = 0.07, p = 0.037), confirming the two-sample MR results (Fig. 5B; Supplementary Table 33, available online).
Robustness of our MR results
The leave-one-out analysis and pleiotropy tests for two-sample MR in the discovery phase were performed to evaluate whether the MR results were robust, while power calculations for one-sample MR in the validation were conducted.
In the discovery phase, pleiotropy tests and the leave-one-out sensitivity analysis for the two-sample MR results (Figs. 2B, D, and Fig. 3) supported the absence of pleiotropy (p > 0.05). They demonstrated the robustness of the causal relationships, respectively. Specifically, the following causal relationships were supported: (i) between eGFR and amygdala volume (Supplementary Table 34, Supplementary Figs. 4, 5; available online), (ii) between amygdala volume and depression (Supplementary Table 35, Supplementary Fig. 6; available online), (iii) between CKD and the volume of the medial nucleus of the amygdala (Supplementary Table 36, Supplementary Fig. 7; available online), and (iv) between the volume of the medial nucleus of the amygdala and depression (Supplementary Table 37, Supplementary Fig. 8; available online).
Although we used IVW as the primary estimator, we additionally utilized the MR-Egger and weighted median methods to assess the robustness of the estimated associations. We applied two MR methods to the causal associations that were statistically significant by the IVW method (Supplementary Figs. 9–12, available online). Among the associations between eGFR-related amygdala subregions and depression traits, the relationship between the cortical amygdala nucleus and sensitivity and hurt feelings showed statistically significant and consistent positive causal association across the two additional MR methods (Supplementary Fig. 9, available online). In addition, most associations between CKD-related medial amygdala nucleus and depression traits showed statistically significant and consistent positive causal association across the two additional MR methods (Supplementary Figs. 11, 12; available online).
In the validation phase, the power analysis for the one-sample MR results (Figs. 4, 5) covered the following associations: (i) between eGFR and the amygdala volume (Supplementary Table 38, available online), (ii) between CKD and the amygdala volume (Supplementary Table 39, available online), (iii) between amygdala volume and depression (Supplementary Table 40, available online), and (iv) between amygdala volume traits and depression (Supplementary Table 41, available online). The MR power estimations for validation of one-sample MR analyses revealed substantial power for key causal relationships. Notably, studies examining the relationships between eGFR and amygdala volume (power up to 0.97 [Supplementary Table 38, available online]) and CKD and amygdala volume (0.72 [Supplementary Table 39, available online]) provided strong evidence for the robustness of these findings. These findings underscore the utility of one-sample MR in exploring the kidney–brain axis and provide key insights into the systemic interplay affecting brain structures.
Discussion
We identified that CKD or eGFR was a significant causative factor for increased amygdala volume traits, and these traits associated with CKD or eGFR showed significant causal estimates for depression disorders in both the discovery and validation phases. Our results have important implications, systematically demonstrating that kidney dysfunction is causally linked to amygdala volume and depressive mood. These findings reinforce the concept of the kidney–brain axis, suggesting that renal impairment may trigger neuroanatomical alterations that elevate the risk of psychiatric disorders.
Similar to previous studies highlighting the association between kidney dysfunction and structural brain changes [13,34–36], our findings support the kidney–brain crosstalk hypothesis. Systemic inflammation caused by CKD progression can disrupt the blood–brain barrier, promote brain injury via inflammatory cytokines, and induce water influx, leading to brain edema [25,37,38]. Functional connectivity studies in hemodialysis patients have also implicated amygdala structural changes in depressive symptoms and cognitive control deficits [35,39]. Our MR analysis further confirmed these trends, showing increased amygdala volumes due to kidney dysfunction.
Decreased kidney function was also shown to increase depressive traits via amygdala volume alterations causally. The amygdala, essential for processing emotions, social behavior, and fear conditioning, has been linked to behavioral disorders, with enlarged volumes correlating with fearfulness [40,41]. Leveraging MR in large cohorts, we demonstrated robust causal links between kidney dysfunction, amygdala volume changes, and depression-related disorders. These findings reveal a strong link between renal impairment and brain structure alterations. Our study highlights the clinical importance of the kidney–brain axis for understanding and potentially mitigating neuropsychiatric complications in CKD patients.
Although the exact pathophysiologic mechanism for this CKD-associated amygdala volume change followed by depressive disorder is still unclear, a previous animal study showed that uremic toxins and the shift in tryptophan metabolism in the amygdala mediate anxiety behavior in a 5/6 nephrectomy rat model [42]. Moreover, in adenine-induced CKD and UT-B–/– mouse models, single-cell transcriptome analysis revealed that demyelination was associated with abnormal proliferation of oligodendrocyte lineage cells in the urea accumulation amygdala [43]. These findings indicated that alteration in the metabolic pathway or cell-type specificity in the amygdala may account for mental disorders in patients with CKD [42,43]. Our MR analyses indicate genetic causality rather than direct mechanistic pathways. Therefore, any experimental or in vivo findings related to our MR results should be interpreted only as potential evidence of causality, not as confirmatory proof.
Our MR analysis met three MR assumptions. Regarding the relevance assumption, it was supported because only instruments with an F-statistic >10 were selected [44]. Regarding the independence assumption, because genetic variants are randomly allocated at conception, they can be considered instruments that are independent of confounding variables, and thus this assumption is supported [45]. Regarding the exclusion restriction assumption, we demonstrated that the assumption was supported by showing no horizontal pleiotropy through MR-Egger intercept tests. There may be sample overlap between the exposure and outcome GWASs used in the two-sample MR analysis. We acknowledged that sample overlap can lead to weak instrument bias in the direction of the confounded observational estimate. To confirm the reproducibility of the two-sample MR analysis based on summary statistics in the population, we additionally performed one-sample MR analysis at the individual level and corroborated reproducibility. Finally, we performed a pleiotropy test using the MR-Egger intercept; however, due to the low statistical power of this method, it has limited ability to detect such pleiotropy [46]. The GWAS datasets used in two-sample MR analysis are predominantly composed of individuals of European ancestry. To compensate for potential ethnic bias, we additionally performed a one-sample MR analysis using a dataset consisting exclusively of individuals of European descent. We acknowledge methodological limitations associated with using CKD as a binary exposure in an MR framework.
There is a discrepancy between our study and a previous report that kidney dysfunction is associated with a decrease in amygdala volume [47]. We suggest four possible explanations for this discrepancy. First, eGFR instrumental variables may be associated with creatinine metabolism or muscle mass rather than true renal dysfunction. To address this limitation, we additionally used eGFRcys instruments to exclude the association with creatinine metabolism or muscle mass, as eGFRcrea may be associated with muscle mass. Second, reverse causation between neural regulation and kidney function may be possible. Third, sample overlap or allele misalignment between the GWAS datasets could contribute to bias. Finally, there may be heterogeneity across amygdala subregions.
To ensure a robust causal inference along the kidney-amygdala-mental disorder pathway, we employed a complementary analytical framework. We utilized two-sample MR to leverage large-scale genomic data for enhanced statistical power, alongside one-sample MR to ensure internal consistency and minimize population stratification bias.
The validity of our causal estimates hinges on the three core instrumental variable assumptions: relevance, independence, and exclusion restriction. By performing pleiotropy tests (e.g., MR-Egger intercept tests), we confirmed the absence of significant horizontal pleiotropy, which directly supports the independence and exclusion restriction assumptions. This rigorous filtering ensures that the IVW method provides unbiased estimates of the causal effect [48]. Therefore, our statistical approach is appropriate for estimating the precise causality along the kidney-amygdala-mental disorder pathway.
While significant, several limitations should be considered. The low power for psychiatric outcomes (0.05–0.10 [Supplementary Tables 40, 41; available online]) warrants cautious interpretation. Additionally, reliance on European ancestry data may affect generalizability. Lastly, our findings cannot confirm whether improving kidney function reduces depression risk through changes in amygdala volume. Clinical trials are necessary to clarify this connection and its impact on depressive behaviors.
In conclusion, we conducted MR analysis in two large cohorts to identify and validate the associations between reduced kidney function and increased amygdala volume, as well as between increased amygdala volume and kidney dysfunction associated with depressive disorder. These findings significantly contribute to the understanding of the kidney–brain axis by establishing a novel causal link between kidney impairment and psychological disorders.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement 1
Print
