Effects of electroconvulsive therapy on cerebellar–cerebral functional connectivity in patients with treatment‐resistant depression and their correlations with gene expression profiles

Abstract

Aims

Electroconvulsive therapy (ECT) is an effective intervention for treatment‐resistant depression (TRD). Although prior studies have identified cerebellar functional alterations following ECT, cerebellar–cerebral functional connectivity (FC) changes remain insufficiently explored.

Methods

Cerebellar seeds were defined based on prior connectivity studies, targeting cerebellar regions functionally coupled with executive, default‐mode, and affective‐limbic networks. Seed‐based cerebellar–cerebral FC analyses were conducted using neuroimaging data from 72 patients with TRD (52 of whom completed post‐ECT assessment) and 63 healthy controls (HCs). Correlation analyses assessed clinical associations, and support vector regression was applied to predict treatment response using baseline FC. Neuroimaging–transcriptomic association analyses were conducted to identify gene expression patterns correlated with post‐ECT FC alterations.

Results

Compared to HCs, patients with TRD exhibited decreased cerebellar connectivity with the sensorimotor network (SMN) and the default‐mode network (DMN). Following ECT, TRD patients showed significant alleviation of depressive symptoms, which was accompanied by a marked increase in cerebellar–DMN connectivity. Enhanced FCs between the left Crus II_Exec2 and the bilateral medial superior frontal gyrus were positively correlated with Hamilton Depression Rating Scale (HAMD) score reduction (r = 0.343, P = 0.026). Additionally, baseline FC of the left Crus II_Exec2 significantly predicted HAMD reduction (r = 0.412, P < 0.001). Neuroimaging–transcription association analysis revealed these FC alterations were associated with normative gene patterns enriched in synaptic signaling, ion channel, and chromatin remodeling.

Conclusions

TRD is associated with reduced cerebellar–SMN and cerebellar–DMN connectivity. ECT may enhance cerebellar–DMN connectivity, potentially contributing to antidepressant effects.

Major depressive disorder (MDD) is one of the most prevalent mental disorders and remains a leading cause of morbidity and mortality. ¹ , ² The standard of care for MDD primarily involves pharmacotherapy, particularly selective serotonin reuptake inhibitors. Notably, approximately 50% of patients fail to achieve an adequate response to initial first‐line antidepressants, and up to 30% remain symptomatic despite multiple trials of diverse pharmacological interventions. ³ Such individuals are clinically classified as having treatment‐resistant depression (TRD), a condition typically defined by an insufficient response to at least two antidepressant trials of adequate dose and duration. ³ , ⁴ For this challenging population, electroconvulsive therapy (ECT) is recognized as a premier therapeutic option. ⁵ ECT is distinguished by its rapid‐acting antidepressant properties, achieving remission rates as high as 80% in TRD. ⁶ Nevertheless, the neurobiological mechanisms underlying the efficacy of ECT remain to be fully elucidated.

Recent functional magnetic resonance imaging (fMRI) studies have provided critical pathological insights into TRD, suggesting that impaired functional integration across multiple resting‐state networks may underpin its clinical characteristics. These studies mainly reported decreased functional connectivity (FC) in the default‐mode network (DMN), executive network (ECN), and affective‐limbic network (ALN) in patients with TRD compared to both healthy controls (HCs) and non‐TRD patients. ⁷ , ⁸ , ⁹ , ¹⁰ Intriguingly, ECT has been shown to modulate the FC within these specific circuits. For instance, Mo et al. reported increased FC between the left angular gyrus (a key node of the DMN) and regions within the ECN and DMN following ECT in patients with MDD. ¹¹ Similar longitudinal enhancements within the DMN and ECN have been documented post‐ECT, with these changes significantly correlating with the alleviation of depressive symptoms. ¹² , ¹³ Furthermore, cross‐diagnostic investigations revealed increased DMN–ECN FC in both MDD and schizophrenia after treatment. ¹⁴ Collectively, these findings underscore that ECT may exert antidepressant effects by enhancing FCs across large‐scale brain networks. Importantly, baseline and longitudinal changes in these networks appear to serve as viable biomarkers for treatment response. Moreno‐Ortega et al. demonstrated that post‐ECT symptom improvement significantly correlates with FC alterations both between the ECN and DMN and within the DMN. ¹⁵ Similarly, Van Waarde identified two predictive networks centered on the dorsomedial prefrontal cortex and anterior cingulate cortex (ACC), key hubs of the ECN and ALN, that could forecast clinical recovery following ECT. ¹⁶

Despite these important findings, existing ECT studies have predominantly focused on cortical large‐scale networks, while the role of the cerebellum, particularly its functional integration with cerebral networks, has not been systematically investigated. The cerebellum has traditionally been associated with motor coordination and balance. However, a growing body of research has revealed that it also plays a critical role in higher‐order functions, including the modulation of cognition and emotion. ¹⁷ , ¹⁸ , ¹⁹ As supporting evidence, previous studies using FC have reported that distinct cerebellar subregions have connections with specific cerebral regions, forming complex cerebellar–cerebral circuits that mediate executive control, default‐mode, and affective processes. ²⁰ , ²¹ These findings suggest that the diverse functions of the cerebellum rely primarily on its functionally distinct, topographically organized subsystems that correspond to the ECN, DMN, and ALN. Previous studies have reported abnormal cerebellar–cerebral FCs in patients with MDD, which may potentially account for the cognitive and emotional deficits observed in this disorder. ²² , ²³ , ²⁴ , ²⁵ These results, despite some mixed findings, primarily suggest decreased connectivity between the cerebellum and cerebral ECN, DMN, and ALN compared to HCs. ²² , ²³ , ²⁴ , ²⁵ Notably, the work of Guo et al. revealed that patients with TRD showed more pronounced decreased FC than those with treatment‐sensitive depression. ²² This suggests that cerebellar–cerebral dysconnectivity may be a distinct pathological hallmark of TRD, rather than a general feature of MDD (Table 1).

Table 1.

The changes of cerebellar–cerebral functional connectivity in treatment‐resistant depression patients at baseline compared to healthy controls

Cerebellar seeds	Brain regions	Side	MNI (x, y, z)	T values	Cluster size
Executive network
R Crus IIExec2	Paracentral lobule/supplementary motor area	L	−6 −15 63	−4.66	72
	Precentral/postcentral gyrus	L	−48 −3 36	−4.37	49
L Lobule VIExec3	Superior parietal lobule/precuneus	R	21 −69 51	−4.63	143
	Cerebellum VII	R	30 −51 −45	−4.38	69
R Lobule VIExec3	Cerebellum VII	R	33 −57 −45	−4.94	238
	Cerebellum IV/V/III	R	18 −39 −24	−4.69	51
Affective‐limbic network
L Lobule VIAff	Middle frontal gyrus	L	−27 48 36	−4.53	53
	Superior frontal gyrus	R	21 0 72	−4.38	53
VermisLimbic	Cerebellum VII	R	18 −57 −45	−4.72	100

L, left; MNI, Montreal Neurologic Institute; R, right.

Notably, emerging evidence indicates ECT‐related effects on the cerebellum. For instance, Depping et al. revealed increased gray matter volume in the left cerebellar area VIIa crus I, attributed to the cerebellar component of the ALN, in patients with MDD following ECT. ²⁶ Functional changes following ECT, including alterations of (fractional) amplitude of low‐frequency fluctuations, ¹² , ²⁷ , ²⁸ , ²⁹ homotopic connectivity, ³⁰ and FC with the posterior cingulate cortex ¹³ and the insula, ⁹ have also been observed in patients with MDD. These ECT‐related changes may contribute to the alleviation of cognitive and emotional deficits. However, most previous ECT studies have primarily focused on cortical networks while largely neglecting cerebellar–cerebral circuits, leaving it unclear how ECT modulates cerebellar–cerebral FC and related large‐scale network reorganization.

Recently, neuroimaging–transcriptomic spatial association analysis has emerged as a novel approach that integrates neuroimaging phenotypes with gene expression profiles to explore the molecular basis of brain functional alterations. Gene expression data were acquired from the Allen Human Brain Atlas (AHBA), which provides high‐resolution transcriptomic maps generated from microarray samples of six postmortem donors. ³¹ Although the precise mechanisms underlying the therapeutic effects of ECT remain unclear, previous studies have suggested that genetic variation may contribute to individual differences in ECT response. ³² Leveraging AHBA data, a recent study revealed that ECT‐related functional brain changes in patients with MDD might be associated with pathways such as oxidative phosphorylation and endocytosis, providing potential molecular bases of ECT. ³³ Another study indicated that alterations in brain entropy and FC in MDD following ECT might be related to the MAPK signaling pathway, synapse activities. ³⁴ Despite these findings, the spatial correlations between cerebellar–cerebral FC alterations and normative gene expression profiles from AHBA in the context of ECT remain to be elucidated (Table 2).

Table 2.

The changes of cerebellar–cerebral functional connectivity in treatment‐resistant depression patients after electroconvulsive therapy compared to baseline

Cerebellar seeds	Brain regions	Side	MNI (x, y, z)	T values	Cluster size
Executive network
L Crus IExec1	Medial superior frontal gyrus/anterior cingulate cortex	R	18 48 24	5.38	186
L Crus IIExec2	Medial superior frontal gyrus	Bilateral	3 54 30	4.52	120
	Superior frontal gyrus	Bilateral	0 51 51	4.44	92
R Crus IIExec2	Medial superior frontal gyrus/anterior cingulate cortex	L	−6 51 3	4.18	103

L, left; MNI, Montreal Neurologic Institute; R, right.

As a severe subtype of MDD, TRD is characterized by elevated relapse rates and a substantial disease burden. ³⁵ , ³⁶ While ECT remains one of the most effective interventions for TRD, its underlying neurobiological mechanisms are not yet fully understood. Although evidence suggests that ECT induces functional alterations within the cerebellum, the specific changes in cerebellar–cerebral FC following treatment remain poorly defined. To fill these research gaps, we conducted seed‐based FC analyses to investigate cerebellar–cerebral changes in patients with TRD before and after ECT. Eleven cerebellar seeds contributing to the ECN, DMN, and ALN were selected. Furthermore, we performed correlation analyses between the baseline FC and ECT‐induced FC alterations and clinical symptom reduction. Additionally, support vector regression (SVR) was conducted to assess the predictive capacity of baseline FC in forecasting treatment response. To identify potential molecular mechanisms of ECT, correlations between ECT‐induced cerebellar–cerebral changes and transcriptional data from the AHBA were analyzed. We hypothesized that (i) as previous study suggested, patients with TRD would exhibit decreased cerebellar–cerebral FCs involving the ECN, DMN, and ALN compared to HCs; (ii) ECT would normalize or significantly enhance these baseline connectivity deficits, and these baseline FC patterns would effectively predict subsequent clinical improvement; (iii) longitudinal alterations in cerebellar–cerebral FC following ECT would correlate with the alleviation of depressive symptoms; (iv) potential candidate genes and biological pathways spatially associated with these post‐ECT FC changes would be investigated through neuroimaging–transcriptomic association analysis using normative gene expression profiles.

Methods

Participants

A total of 74 patients with TRD were enrolled from the Inner Mongolia Autonomous Region Mental Health Center, located in Hohhot, China. Eligible patients were aged 18–65 years, right‐handed, had completed at least 9 years of education, and were diagnosed with MDD according to DSM‐IV criteria, confirmed by an experienced psychiatrist. All patients exhibited depressive symptoms, with a Hamilton Depression Rating Scale (HAMD) score ≥ 21. TRD was defined as an inadequate response to stable antidepressant therapy before enrollment, indicated by a reduction of less than 20% in HAMD scores. Exclusion criteria included: (i) severe physical conditions or organic brain disorders; (ii) current or past substance abuse or dependence; (iii) pregnancy or breastfeeding; and (iv) contraindications to MRI or ECT.

Additionally, 66 HCs were recruited from the local population and matched to patients by age, gender, handedness, and years of education. Participants with either a personal or familial history of psychiatric illness were excluded. The remaining exclusion standards were identical to those applied in the TRD cohort.

The study protocol was reviewed and approved by the Ethics Committee of the Inner Mongolia Autonomous Region Mental Health Center. All subjects gave written informed consent before enrollment.

Assessments

Depressive, anxiety, and psychiatric symptoms were evaluated at both baseline and following ECT using standardized instruments, including the HAMD, Hamilton Anxiety Rating Scale (HAMA), and the Brief Psychiatric Rating Scale (BPRS). All assessments were conducted by two experienced psychiatrists to ensure consistency and reliability. In addition, self‐reported scales, including the Self‐Rating Depression Scale (SDS), Self‐Rating Anxiety Scale (SAS), and the Automatic Thoughts Questionnaire (ATQ), were used to capture patients' subjective symptoms.

Study design

In this study, patients with TRD received modified bitemporal ECT using the spECTrum 5000Q device. The first three sessions were administered on consecutive days, followed by treatments every other day, for a total of 6–12 sessions. All patients were anesthetized with propofol and received muscle relaxation with succinylcholine and atropine before each procedure. If clinical improvement was considered insufficient after the initial six sessions, the treatment course was extended at the treating physician's discretion to a maximum of 12 sessions (mean ± SD: 10.57 ± 1.76). The mean stimulus duration was 5.89 ± 1.16 s, the mean stimulus charge was 152.32 ± 57.09 mC, the mean seizure duration on electroencephalography was 31.85 ± 4.76 s, and the mean stimulus energy was 26.75 ± 10.01 J. Detailed information on medication is available in Table S1.

HCs only underwent once resting‐state fMRI scanning at baseline. For patients with TRD, both clinical assessments and fMRI scans were conducted at baseline and after completing the ECT course. Scans were performed within 24 h before the first ECT session and within 24 h following the final session.

Imaging data acquisition and preprocessing

Functional images were obtained using a Siemens Skyra 3.0 T scanner with a gradient‐recalled echo‐planar imaging (EPI) sequence (slice = 37, TR = 2,000 ms, TE = 30 ms, FA = 90°, acquisition matrix = 64 × 64, FOV = 240 × 240 mm², slice thickness = 4 mm with no gap, total scan time = 480 s). High‐resolution T1‐weighted anatomical images were acquired (slice = 192, TR = 8.2 ms, TE = 3.2 ms, FA = 12°, acquisition matrix = 256 × 256, FOV = 240 × 240 mm², slice thickness = 1 mm with no gap). All the participants were instructed to stay motionless, keep their eyes closed, and maintain wakefulness. Preprocessing processes of the imaging data were presented in Appendix S1.

Cerebellar regions of interest selection and FC analysis

Seed‐based FC analyses were performed using RESTplus software to investigate cerebellar–cerebral networks. ³⁷ Cerebellar regions of interest (ROIs), previously shown to be functionally connected with cerebral networks, were selected to represent the ECN, ALN, and the DMN. These seeds were categorized by network and detailed in Table S2. Each ROI was defined as a 6‐mm radius sphere, and the mean time series of voxels within each ROI was extracted to reflect seed activity. Pearson correlation coefficients were computed between each ROI's time series and all other brain voxels to generate whole‐brain connectivity maps. Spatial smoothing with a 6‐mm full width at half‐maximum Gaussian kernel was applied, followed by Fisher's r‐to‐z transformation to improve the normality of the correlation distribution.

Statistical analysis

Differences in demographic and clinical characteristics between patients with TRD and HCs were assessed using the Mann–Whitney U‐test or Chi‐square test, based on data distribution. For the TRD group, baseline and posttreatment clinical measures were compared using paired t‐tests or Wilcoxon signed‐rank tests. Moreover, comparisons between follow‐up completers and non‐completers, as well as between completers and the total baseline sample, were conducted to examine potential selection bias.

Two‐sample t‐tests were then performed to examine FCs between cerebellar ROIs and the whole brain, controlling for age, sex, and education. Paired t‐tests were performed within the TRD group to examine FC changes before and after ECT, with medication as covariates (0 = antidepressants only; 1 = antidepressants + anxiolytics; 2 = antidepressants + antipsychotics; 3 = antidepressants + anxiolytics + antipsychotics). ³⁸ Multiple comparisons were corrected using Gaussian Random Field (GRF) theory, with a voxel‐wise threshold of P < 0.001 and cluster‐level significance at P < 0.05.

Partial correlation analyses were performed in SPSS to examine the associations between baseline FCs and clinical variables (e.g. HAMD, HAMA, SDS, SAS, ATQ, and BPRS), as well as between FC changes after treatment and the reduction rates of these scores. Moreover, we conducted additional analyses examining the relationships between ECT parameters (e.g. number of sessions, stimulus duration, stimulus charge, seizure duration, and mean stimulus energy) and both symptom remission and FC changes. All analyses were adjusted for age, sex, and years of education. Statistical significance was set at P < 0.05.

Validation analysis

To mitigate potential a priori bias associated with predefined cerebellar seeds, a data‐driven independent component analysis (ICA) was performed as a supplementary validation. ICA was used to decompose the resting‐state fMRI data into spatially independent functional networks without prior anatomical constraints. Components corresponding to resting‐state networks (e.g. ECN, DMN, and ALN) were identified based on their spatial patterns. For networks exhibiting significant baseline group differences and longitudinal changes following ECT, the cerebellar regions within these corresponding components were extracted as data‐driven seeds. These seeds were then utilized in subsequent cerebellar–cerebral FC analyses, following the same statistical procedures as the primary seed‐based analyses. The statistical procedures, as in the primary seed‐based analyses, were applied. Detailed ICA procedures are provided in Appendix S1.

SVR analysis

Voxel‐wise SVR was conducted using the MVPANI toolbox to examine whether baseline FC patterns could predict symptom reduction after ECT. ³⁹ Specifically, for each ROI, the baseline FC map served as the input feature, while the corresponding post‐ECT FC change map was utilized as a mask to constrain the analysis to voxels exhibiting significant longitudinal alterations. All features were normalized before model construction. The leave‐one‐out validation method was applied to evaluate predictive performance and model robustness. Given the restricted feature space defined by the longitudinal mask, no additional feature selection was performed. Predictive performance was quantified using the Pearson correlation coefficient between the predicted and observed changes in HAMD scores. Statistical significance was assessed by 5,000 permutation tests, with P values reflecting the probability that the observed prediction accuracy exceeded chance level under random label assignment.

Neuroimaging–transcription association analysis

Gene expression data were sourced from the publicly available Allen Human Brain Atlas (AHBA), providing detailed transcriptomic profiles from six adult brain donors. ³¹ Data preprocessing was conducted using the Abagen toolbox (version 0.1.1; https://github.com/rmarkello/abagen), which reannotates probe information and maps expression data to the AAL‐116 brain atlas. This process yielded a 15,633 × 112 matrix (genes × brain regions), with four regions excluded due to the absence of tissue samples in the AHBA.

Partial least squares (PLS) correlation analysis was applied to assess associations between gene expression and post‐ECT FC changes in TRD patients. The first component (PLS1) captured the spatial pattern explaining the largest variance in gene expression across FC alterations. Permutation testing was performed to determine whether the covariance explained by PLS1 exceeded chance levels (P < 0.05). Gene weights were derived using 1,000 bootstrap iterations, and corresponding Z‐scores were computed and evaluated using false discovery rate (FDR) correction at P < 0.05. In order to obtain more representative feature genes, common PLS+ and PLS− genes present in at least 80% of ROIs (i.e., 9 of 11) showing post‐ECT FC changes were selected as candidate genes for subsequent functional enrichment analysis.

Gene enrichment analyses

Enrichment analysis of Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways was performed separately on PLS+ and PLS− gene sets using the Metascape platform (https://metascape.org/). A protein–protein interaction (PPI) network analysis was constructed to investigate potential interactions among proteins encoded by the candidate genes using the STRING database (https://cn.stringdb.org/).

Furthermore, specific expression analysis (http://genetics.wustl.edu/jdlab/csea-tool-2/) was conducted to examine the expression of PLS+/PLS− gene sets across distinct brain cell types and developmental stages in different brain regions. The enrichment of these gene sets was evaluated using the specificity index probability, applying a significance threshold of 0.05. To determine overlap between PLS+/PLS− genes and transcripts preferentially expressed in specific cell types or developmental stages, Fisher's exact test was performed, with multiple comparisons corrected by the Bonferroni method (P < 0.05).

Results

Demographic and clinical characteristics

Two patients with TRD at baseline, five patients post‐ECT, and three HCs were excluded due to excessive head movement. Consequently, the final analysis included 72 patients with TRD (n = 52 of whom completed the post‐ECT follow‐up) and 63 HCs. As shown in Table S3 and Fig. S1, the TRD and HC groups were comparable in terms of age, gender, and years of education (P > 0.05). Furthermore, no significant demographic or clinical differences were observed between follow‐up completers (n = 52) and non‐completers (n = 20), nor between the completers (n = 52) and the total baseline sample (n = 72) (Tables S4,S5).

After ECT, patients with TRD showed significant improvements in the scores of HAMD, HAMA, SDS, SAS, ATQ, and BPRS (P < 0.001), with scores that were significantly lower than baseline (Table S6).

Group differences in cerebellar–cerebral FC analyses

At baseline, patients with TRD showed decreased cerebellar–cerebral FC in the ECN and ALN relative to HCs (Fig. 1 and Table 1). For the ECN, patients with TRD exhibited reduced FCs between (i) the right Crus II_Exec2 and the left paracentral lobule/supplementary motor area and left precentral/postcentral gyrus, (ii) the left Lobule VI_Exec3 and the right superior parietal lobule/precuneus and right cerebellum VIII; and (iii) the right Lobule VI_Exec3 and the right cerebellum VIII and right cerebellum IV/V/III. For the ALN, patients with TRD exhibited decreased FC between (i) the left Lobule VI_Aff and the left middle frontal gyrus (MFG) and right superior frontal gyrus (SFG) and (ii) the right Vermis_Limbic and right cerebellum VIII.

Fig. 1.

Abnormalities of cerebellar–cerebral functional connectivity based on cerebellar ROIs of the ECN (a) and ALN (b) in patients with treatment‐resistant depression at baseline compared to healthy controls. Orange and green balls indicate regions of interest in the cerebellum, corresponding to the ECN and ALN, whereas black balls indicate the peak regions connected to them. The brain maps show the spatial distribution of these clusters. The color bar represents T values from two‐sample t‐tests. ALN, affective‐limbic network; ECN, executive control network; HCs, healthy controls; ROI, region of interest; SFG, superior frontal gyrus; SMA, supplementary motor area; SPL, superior parietal lobule; TRD, treatment‐resistant depression.

Following ECT, there were increased cerebellar–cerebral FCs in the ECN in patients with TRD. Specifically, there were increased FCs between (i) the left Crus I and the right medial SFG (MSFG)/ACC; (ii) the left Crus II and the bilateral MSFG and SFG; and (iii) the right Crus II and the left MSFG/ACC (Fig. 2 and Table 2).

Fig. 2.

Cerebellar–cerebral functional connectivity alterations in patients with treatment‐resistant depression after ECT, with medication as covariates. Orange balls indicate regions of interest in the cerebellum corresponding to the ECN, whereas black balls indicate the peak regions connected to them. The brain maps show the spatial distribution of these clusters. Color bar represents the T values obtained from the paired t‐tests. ACC, anterior cingulate cortex; ECN, executive control network; ECT, electroconvulsive therapy; L, left; R, right; ROI, region of interest; SFG, superior frontal gyrus; SMFG, superior medial frontal gyrus.

Correlation analyses

At baseline, FC between the right Lobule VI_Exec3 and the right cerebellum IV/V/III was negatively correlated with the BPRS scores (r = −0.299, P = 0.012). FC between right vermis and right cerebellum VIII was positively correlated with ATQ scores (r = 0.255, P = 0.035, Fig. S2).

Following ECT, two significant positive correlations were observed between FC changes and symptom improvement. Specifically, increases in FC between the left Crus II_Exec2 and the bilateral MSFG were positively correlated with the HAMD reduction rate. Additionally, FC changes between the right Crus II_Exec2 and the MSFG/ACC also showed a significant positive correlation with the ATQ reduction rate (Fig. 3). No significant correlations were observed between ECT parameters and either symptom remission or FC changes.

Fig. 3.

(a) Following electroconvulsive therapy, clinical symptoms of patients with treatment‐resistant depression present a significant reduction. (b) The cerebellar–cerebral FC alterations following electroconvulsive therapy were significantly correlated with the symptom reduction rates. (c) Baseline FC of the left Crus II_Exec2, restricted to regions showing post‐ECT FC changes, successfully predicted HAMD reduction using a support vector regression model (r = 0.412, P < 0.001). ACC, anterior cingulate cortex; ATQ, Automated Thinking Questionnaire; BPRS, Brief Psychiatric Rating Scale; FC, functional connectivity; HAMA, Hamilton Anxiety Scale; HAMD, Hamilton's Depression Scale; L, left; MCC, middle cingulate cortex; R, right; SAS, Self‐Rating Anxiety Scale; SDS, Self‐Rating Depression Scale; SMFG, superior medial frontal gyrus.

Validation analysis

To validate the primary findings, ICA‐based analyses were specifically focused on ECN‐related components, as only the cerebellar FC within the ECN exhibited significant alterations at both baseline and following ECT in the seed‐based approach. Two independent components (IC6 and IC14) representing the ECN were identified, and their cerebellar subregions were extracted as data‐driven seeds for subsequent FC analyses. Using the IC6‐derived cerebellar seed, patients with TRD exhibited significantly decreased FC with the left precentral gyrus compared to HCs, although no significant longitudinal changes were observed following ECT. Using the IC14‐derived cerebellar seed, patients showed reduced FC with the MFG and middle temporal gyrus at baseline, which was followed by a significant increase in FC with the angular gyrus post‐ECT (Figs S3,S4).

Support vector regression

An SVR model using baseline FC maps of the left Crus II_Exec2, constrained by the longitudinal post‐ECT FC change mask, significantly predicted HAMD reduction rate, with a positive correlation between predicted and observed changes (Fig. 3, r = 0.412, P < 0.001 by permutation test). No significant predictive models were established for the left Crus I_Exec1 and the right Crus II_Exec2.

Gene expression associated with FC changes following ECT

PLS1 explained variance ranging from 10% to 40% for each ROI (Fig. S5), with significance confirmed by permutation testing (P < 0.05). For each ROI, the gene number for the PLS+ and PLS− gene is shown in Table S7. A total of 1,221 PLS+ and 814 PLS− common genes correlated with at least 80% of ROIs were identified. Enrichment analyses suggested PLS+ genes were functionally enriched in terms such as ‘regulation of transsynaptic signaling’, ‘post synapse’, and ‘transporter regulator activity’, PLS− genes were associated with pathways including ‘chromatin remodeling’, ‘regulation of developmental growth’, and ‘structural constituent of chromatin’ (Fig. 4a). In PPI network analyses, ACTB and TNF were key hub proteins among PLS+ and PLS− genes, highly interconnected with other proteins (Fig. 4b, Figs S6,S7).

Fig. 4.

Enrichment analyses of the common genes that positively (red) and negatively (purple) associated with cerebellar–cerebral functional connectivity alterations following electroconvulsive therapy. (a) GO, KEGG analysis PLS1+/PLS1− gene. Bar plots present the top 10 significantly enriched pathways within GO (BP, CC, MF) and KEGG. The y‐axis indicates the specific terms across major categories, while the x‐axis shows the number of genes associated with each term. (b) PPI analysis illustrates hub genes associated with functional connectivity alterations. (c) The lifespan windows across brain regions that PLS1+/PLS1− gene sets were expressed derived from specific expression analysis. BP, biological processes; CC, cellular components; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular functions; PPI, protein–protein interaction networks.

Cell‐type‐specific expression analysis indicated PLS+ gene sets were primarily expressed in spiny neurons in the striatum and layer 5 and 6 neurons, while PLS− genes were expressed in the granule neurons in the cerebellum (Fig. S8). PLS1+ genes were predominantly significantly enriched between early middle fetal and young adulthood, extensively involved in the cortex; PLS− genes were predominantly significantly enriched in young adulthood, involved in the cerebellum (Fig. 4c).

Discussion

In the present study, we investigated cerebellar–cerebral FCs in patients with TRD before and after ECT. The main findings are summarized as follows. First, compared to HCs, patients with TRD exhibited decreased cerebellar–cerebral FCs with regions in the sensorimotor network (SMN; including the left paracentral lobule/supplementary motor area and left precentral/postcentral gyrus), the DMN (including the right superior parietal lobule/precuneus, the left MFG, and right SFG). Second, following ECT, patients with TRD showed significant symptom improvement, accompanied by increased cerebellar–cerebral FCs between the cerebellum and DMN (MSFG, SFG, and MSFG/ACC). Consistently, our ICA‐based validation confirmed these findings, revealing decreased cerebellar–cerebral connectivity with the SMN (precentral gyrus) and DMN (MFG) at baseline, alongside increased FC with the DMN (angular gyrus) following ECT. Third, longitudinal alterations in FC of the Crus II_Exec2 were positively correlated with depressive symptom reduction. Notably, baseline FC of this region – when constrained by post‐ECT change masks – successfully predicted individual treatment responses. Finally, by integrating these imaging findings with normative transcriptomic data, we revealed specific functional pathways, hub genes, and cell‐type‐specific signatures whose spatial distributions were similar to post‐ECT cerebellar–cerebral changes.

Decreased connectivity of the cerebellum‐cerebral FCs in TRD

Compared with HCs, patients with TRD showed reduced FC between the cerebellum (Crus II_Exec2) and the SMN (left paracentral lobule/supplementary motor area and left precentral/postcentral gyrus). The SMN is fundamentally involved in processing exteroceptive information, including sensory, visual, auditory, and perceptual stimuli, while the ECN governs higher‐order cognitive functions such as goal‐directed behavior, task execution, and attentional regulation. The abnormal connectivity of the SMN has been reported in patients with MDD, such as increased DMN‐SMN FC. ⁴⁰ Similarly, decreased FCs between the left Lobule VI_Exec and the precentral cortex were reported in patients with geriatric depression. ⁴¹ Regarding TRD, Lee et al. suggested that structural data from the SMN could serve as a strong predictor of TRD. ⁴² Patients with TRD also presented reduced FCs between SMN and temporal regions compared to both HCs and non‐TRD. ¹⁰ All these findings suggested SMN might play a role in TRD. We speculated that the decreased FCs between the cerebellar ECN and SMN might indicate reduced functional integration of cognitive and sensory function. In this context, patients with TRD may exhibit reduced sensitivity to environmental stimuli, which could manifest as sluggish responses to external inputs. ⁴³ Moreover, sensory modalities such as olfactory, visual, and auditory inputs are closely involved in the recall and reexperiencing of emotional events. ⁴⁴ , ⁴⁵ Reduced FC may impair patients' ability to perceive physically pleasurable experiences (e.g. touch, movement) and may be associated with the common symptom of anhedonia in MDD. ⁴⁶

Patients with TRD showed reduced FC between the cerebellum (Lobule VI_Exec3) and the DMN (the right superior parietal lobule/precuneus) relative to HCs. The DMN is primarily involved in self‐referential processes, including introspection and self‐awareness. ⁴⁷ A recent meta‐analysis has suggested that patients with TRD exhibit decreased FCs both within the DMN and between the DMN and other regions. ⁴⁸ Decreased FCs between the ECN and the DMN have been observed in patients with TRD compared to both HCs and non‐TRD patients. While Lui et al. reported that decreased FC between the dorsolateral prefrontal cortex and the precuneus/parietal cortex was only evident when comparing TRD to HCs – but not to non‐TRD ⁸ – other studies have identified similar reductions in FC between the dorsolateral prefrontal cortex and the inferior parietal lobule (a key DMN node) specifically in the TRD population. ⁹ Collectively, these findings suggest that DMN hypoconnectivity may serve as a critical neurobiological feature that differentiates TRD from non‐TRD. ⁴⁹ However, limited research has focused on cerebellar FCs in patients with TRD. Notably, Guo et al. reported decreased FCs between the cerebellar ECN (bilateral Crus I and right Crus II) and the precuneus in TRD compared to non‐TRD, indicating that impaired functional integration of the cerebellum and the DMN might be a specific change in TRD. ²² These reductions in FCs may underlie treatment resistance, highlighting the role of disrupted cerebellar–cerebral networks in TRD.

Decreased FCs between the cerebellum (Lobule VI_Aff) and the prefrontal regions (the left MFG and right SFG) were identified in patients with TRD. This finding is consistent with previous reports of decreased FC between the Lobule VI_Aff and the ventral medial prefrontal cortex. ⁴¹ Similarly, Zhu et al. documented decreased dynamic FC between cerebellar limbic components and the ventral medial prefrontal cortex in patients with MDD, noting that these alterations were positively correlated with the severity of depressive and anxiety symptoms. ²⁵ Given that prefrontal regions serve as critical hubs for the dual regulation of cognition and emotion, we speculate that impaired cerebellum–prefrontal integration may underpin deficits in emotional regulation, thereby contributing to the persistent symptomatology of TRD. Moreover, according to the work of Alves et al., both the MFG and SFG are also components of the DMN. ⁵⁰ Compared to non‐TRD, patients with TRD presented decreased FCs between the right caudate and the right MFG and SFG. ⁵¹ Similarly, Guo et al. reported decreased FCs between the Lobule VI_Aff and the DMN (precuneus, angular gyrus, and inferior parietal lobule) in patients with TRD compared to non‐TRD. ²² Taken together, these findings support our hypothesis that reduced DMN connectivity may underlie the neurobiological mechanisms of TRD.

ECT improved the cerebellum‐cerebral FCs in TRD

Following ECT, patients with TRD exhibited significant symptom reduction, which was accompanied by increased FC between the cerebellar ECN and the DMN (MSFG, SFG, and MSFG/ACC). Moreover, these FC alterations were significantly positively correlated with the reduction rates of depressive symptoms and negative automatic thoughts. These findings suggest that alterations in cerebellar–DMN FC may be associated with the clinical response to ECT. Notably, ECT‐induced increases in FC involving the MFG and SFG were the most consistent results in previous investigations. ⁵² ECT has been shown to modulate FC between the SFG/MFG and multiple regions, including the DMN (ACC, posterior cingulate cortex, as well as the right inferior parietal lobule), emotion‐related areas (hippocampus, bilateral orbitofrontal cortex, amygdala, and insula), as well as the bilateral temporoparietal junction, left inferior/middle temporal gyrus in patients with MDD. ¹² , ¹³ , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ These findings suggest that ECT may improve functional integration between the SFG/MFG and widespread cortical and subcortical regions. However, the role of the cerebellum in these processes has largely been neglected. Our study highlights the critical involvement of the cerebellum in the therapeutic mechanisms of ECT.

On the other hand, increased cerebral blood flow in the cerebellum has also been reported during the postictal phase of seizures. ⁵⁷ Based on this, some studies have argued that cerebellar alterations observed after ECT might reflect seizure‐related effects rather than specific modulation of affective processes. ⁵² , ⁵⁸ However, emerging evidence provides a strong counter‐argument. Notably, previous studies reported that increased gray matter volume in the left VIIa crus I, a region related to the affective/limbic cerebellum, was correlated with depressive symptom relief following ECT. ²⁶ Consistent with this, our study identified two significant positive correlations between the cerebellum‐cerebral FCs, including MSFG and ACC, and symptom reduction. Moreover, the baseline FC of the left Crus II_Exec2, constrained by post‐ECT FC change masks, could predict the depressive symptom reduction. The fact that pretreatment connectivity can forecast clinical outcomes suggests that these cerebellar changes are more likely associated with the therapeutic process, rather than merely reflecting a generalized, nonspecific response to seizure activity. Collectively, these findings underscore the specific and pivotal contribution of the cerebellum to the antidepressant efficacy of ECT. Nevertheless, in the absence of a longitudinal control group, these findings should be interpreted as preliminary evidence, as nonspecific test–retest or time‐related effects cannot be fully excluded, thereby limiting causal inference. At baseline, decreased cerebellar–DMN connectivity was observed, which subsequently showed significant enhancement following ECT. Notably, although the specific DMN subregions showing posttreatment enhancement did not entirely overlap with those demonstrating baseline deficits, both sets of regions are integral components of the DMN. This spatial divergence within a common system suggests that ECT may not merely reverse localized deficits but rather be accompanied by a broad functional reorganization across the cerebellar–DMN network. Collectively, these findings suggest that cerebellar–DMN circuits may be implicated in the pathophysiology of TRD and may be relevant to ECT‐related effects. These circuits may therefore represent potential candidate biomarkers, pending further validation.

Potential genetic mechanisms of ECT

By leveraging transcriptomic data from the AHBA, we explored the spatial relationship between cerebellum‐cerebral changes following ECT and normative gene expression patterns, thereby offering insight into molecular signatures that may be associated with these changes. In the PLS1+ gene set, enriched functional pathways were mainly associated with synaptic function and signaling, as well as ion channel activity and regulation. These results align with previous studies that leveraged normative AHBA data to characterize the transcriptomic patterns spatially associated with ECT‐related changes in MDD, which also highlighted pathways such as synaptic plasticity. ³³ Similarly, Zhu et al. reported spatial correspondence between MDD‐related brain alterations and the normative expression of genes involved in transsynaptic signaling and ion transport regulation. ⁵⁹ Furthermore, a recent meta‐analysis utilizing similar transcriptomic‐imaging frameworks reported consistent spatial associations with genes related to synaptic signaling, calcium ion binding, and neurodevelopment. ⁶⁰ Dysfunction of synaptic plasticity in the prefrontal‐limbic regions may lead to impaired cognitive control over negative emotions in MDD. ⁶¹ The spatial coupling between FCs and the normative expression of genes related to synaptic functioning suggests that these cerebellar circuits might be markers of recovery and neural reorganization. Regarding hub genes identified from the PPI network, CALM3 and CAMK2A are linked to calcium signaling, while GRIA1, GRIA2, GRIN2A, and GRIN2B are associated with NMDA receptor function. Together, the enrichment of these molecular profiles suggests that these cerebellar–cerebral FCs spatially overlap with gene expression profiles related to synaptic remodeling, which may be relevant to the neuromodulatory effects of ECT. However, these findings require further validation in future studies integrating multimodal data from the same cohort.

In the PLS1− gene set, enriched functional pathways were primarily associated with growth regulation, chromatin remodeling, and transcriptional control. Furthermore, hub genes such as H3‐4, H3‐3B, H3C12, and H2BC21 are all members of the histone family and are involved in maintaining chromatin structure and regulating gene expression. H3 acetylation in the prefrontal cortex and hippocampus has been demonstrated as a therapeutic mechanism mediating antidepressant‐like effects in mice. ⁶² , ⁶³ , ⁶⁴ Replication‐independent histone variants contribute to MDD pathogenesis by dynamically regulating chromatin structure and gene expression in the nucleus accumbens, serving as a key epigenetic mechanism that links environmental stressors to depressive behaviors. ⁶⁵ Based on these findings, our results suggest that the spatial patterns of post‐ECT reorganization may align with gene sets involved in epigenetic modulation, potentially reflecting molecular contexts relevant to neural plasticity. Clinically, these normative transcriptomic associations may offer a preliminary context for understanding biological variability in treatment response to ECT.

Limitations

First, our study was limited to a specific cohort of right‐handed Chinese patients with TRD. Potential cultural, genetic, and clinical heterogeneity across different populations may limit the generalization of the results. Second, the HCs underwent only a single baseline scan, lacking a longitudinal follow‐up. This design limitation makes it difficult to fully account for nonspecific factors, such as test–retest effects or time‐related fluctuations in FC. Consequently, while the observed changes correlate with clinical improvement, the absence of a longitudinal control group precludes direct causal attribution, and these findings should be considered as putative neural correlates rather than therapeutic pathways of ECT. Finally, our neuroimaging–transcriptomic association analysis relies on the AHBA database, which is derived from a limited number of normative postmortem brains from healthy donors. Because these data are obtained from healthy individuals, they may not reflect the specific molecular state of patients or the neuroplastic changes induced by ECT in a clinical population. Consequently, our findings should be interpreted with caution, as they represent indirect spatial associations rather than patient‐specific molecular processes.

Conclusions

Patients with TRD presented reduced functional integration between the cerebellum and both the SMN and DMN relative to HCs. ECT might be associated with enhanced cerebellar–DMN connectivity, which may contribute to antidepressant effects. The spatial distribution of post‐ECT FC alterations may be associated with the normative expression of genes involved in synaptic signaling, ion channel activity, and chromatin remodeling pathways. These findings provide exploratory insights into the molecular context of ECT‐related functional changes.

Author contributions

L.Z. contributed to writing – review and editing, writing – original draft, methodology, formal analysis. Y.H. contributed to methodology and formal analysis. D.L., P.Y., L.C., Y.Y., J.Z., and W.G. contributed to writing – review and editing, visualization and validation. Z.G. and Y.O. contributed to writing – review and editing, methodology and formal analysis.

Funding statement

This study was supported by grants from the National Natural Science Foundation of China (grant nos. 82171508) and the Natural Science Foundation of Hunan (grant no. 2025JJ60706).

Disclosure statement

The authors have no conflicts of interest to declare.

Supporting information

Figure S1. Characteristics of patients between groups. ATQ, Automated Thinking Questionnaire; BPRS, Brief Psychiatric Rating Scale; HAMA, Hamilton Anxiety Scale; HAMD, Hamilton's Depression Scale; SAS, Self‐Rating Anxiety Scale; SDS, Self‐Rating Depression Scale; TRD, data from all treatment‐resistant depression patients at baseline (n = 72); T1:TRD, baseline data of follow‐up treatment‐resistant depression patients (n = 52); T2:TRD, endpoint data of follow‐up treatment‐resistant depression patients (n = 52).

Figure S2. Correlations between baseline abnormal functional connectivity (FC) changes in the treatment‐resistant depression patients (n = 72) and clinical symptoms, controlled for age, gender, and educational level. ATQ, Automated Thinking Questionnaire; BPRS, Brief Psychiatric Rating Scale; FC, functional connectivity; HAMA, Hamilton Anxiety Scale; ROI, region of interest.

Figure S3. Independent component (IC) 6 identified by independent component analysis (ICA), corresponding to the central executive network (CEN), and its cerebellar portion used as a seed for subsequent functional connectivity (FC) analysis. Using this cerebellar IC‐derived seed, patients with TRD exhibited decreased FC with the left precentral gyrus compared with healthy controls (HCs), which was consistent with the main seed‐based results.

Figure S4. Independent component (IC) 14 identified by independent component analysis (ICA), corresponding to the central executive network (CEN), and its cerebellar portion used as a seed for subsequent functional connectivity (FC) analysis. Using this cerebellar IC‐derived seed, patients with TRD exhibited decreased FC with the middle frontal gyrus and middle temporal gyrus at baseline compared with healthy controls (HCs), and increased FC with the angular gyrus following ECT.

Figure S5. PLS1 (the first component of partial least squares) explained variance for each ROI. PLS1 explained variance ranging from 10% to 40% for each regions of interest (ROIs).

Figure S6. A protein–protein interaction network of PLS+ gene sets. Nodes represent proteins, scaled and colored according to their degree. Edges represent protein–protein interactions. Node size and color intensity increase with the degree, highlighting potential hub proteins with central roles in the network.

Figure S7. A protein–protein interaction network of PLS− gene sets. Nodes represent proteins, scaled and colored according to their degree. Edges represent protein–protein interactions. Node size and color intensity increase with the degree, highlighting potential hub proteins with central roles in the network.

Figure S8. Cell‐type‐specific expression analysis for PLS+ and PLS− genes. Cell‐type‐specific expression analysis indicated PLS+ gene sets were primarily expressed in spiny neurons in the striatum and layer 5 and 6 neurons, while PLS− genes were expressed in the granule neurons in the cerebellum. PLS = partial least squares.

Table S2. Cerebellar regions of interest (ROIs) and coordinates grouped by networks.

Table S3. Demographic and clinical characteristics of patients and healthy controls.

Table S4. Comparison of baseline characteristics of follow‐up completers and non‐completers.

Table S5. Comparison of baseline characteristics between the total TRD cohort and follow‐up completers.

Table S6. Characteristics of patients who completed the follow‐up assessments.

Table S7. Number of genes that correlated with different cerebellar regions of interests (ROIs).

Table S1. Treatments applied to patients with treatment‐resistant depression who completed follow‐up.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.