Bipolar disorder is a chronic psychiatric illness characterised by recurrent episodes of mania, hypomania and depression, with a lifetime prevalence of approximately 2.4% worldwide. Reference Grande, Berk, Birmaher and Vieta1 It is associated with poor psychosocial functioning and elevated risks of psychiatric and medical comorbidities, all-cause mortality and death by suicide. Reference Grande, Berk, Birmaher and Vieta1 In most patients, the first episode of bipolar disorder occurs during late adolescence, and this disorder is therefore a leading cause of disability in early adulthood. Reference Grande, Berk, Birmaher and Vieta1 Furthermore, early onset during adolescence is associated with severe illness, treatment delay and poor prognosis. Reference Bolton, Warner, Harriss, Geddes and Saunders2
The limbic network, which comprises the amygdala, hippocampus, thalamus, basal ganglia, parahippocampal gyrus (PHG), cingulate gyrus and hypothalamus, plays a central role in memory and emotional processing. Reference Kamali, Milosavljevic, Gandhi, Lano, Shobeiri and Sherbaf3 The interplay between this network and frontal cortices mediates the integration of motivational, affective and inhibitory processes. Reference Kamali, Milosavljevic, Gandhi, Lano, Shobeiri and Sherbaf3 Limbic network dysfunction was proposed to be a core feature of bipolar disorder, characterised by hyperactivation of limbic structures, disrupted intra-limbic connectivity and impaired fronto-limbic integration. Reference Mesbah, Koenders, van der Wee, Giltay, van Hemert and de Leeuw4 Consistent with this framework, meta-analytic evidence from adults with bipolar disorder has demonstrated increased limbic activity during emotion processing as well as hyperconnectivity within limbic and between limbic and large-scale brain networks. Reference Mesbah, Koenders, van der Wee, Giltay, van Hemert and de Leeuw4 These abnormalities appear to extend to younger populations. A resting-state functional connectivity study reported similar patterns of cortico-limbic hyperconnectivity in paediatric patients with bipolar disorder, particularly involving the amygdala, even during euthymic states. Reference Guo, Wang, Jiao, Cao, Cui and Gao5 However, to the best of our knowledge, no study has directly examined resting-state functional connectivity within the limbic network in paediatric bipolar disorder.
Within the limbic circuitry involved in appetite regulation, the amygdala contributes to the integration of emotional aspects of feeding behaviour, whereas the hypothalamus serves as a critical hub linking limbic neural inputs with peripheral metabolic signals. Reference Jais and Brüning6 Appetite-related hormones, including insulin, leptin, ghrelin and adiponectin, influence hypothalamic circuits, including the arcuate nucleus, through distinct and partly opposing signalling pathways to regulate feeding behaviour, energy balance and metabolic homeostasis. Reference Jais and Brüning6 Supporting this neurobiological framework, alterations in appetite hormone signalling have been associated with changes in intra-limbic and cortico-limbic functional connectivity in healthy individuals, although the specific role of hypothalamic connectivity remains largely unexplored. Reference Al-Zubaidi, Iglesias, Stephan, Buades-Rotger, Heldmann and Nolde7 Given the established role of appetite hormones in modulating hypothalamic and limbic function, the appetite hormone dysregulation observed in bipolar disorder may be associated with the limbic network abnormalities implicated in the disorder. Reference Misiak, Kowalski, Stańczykiewicz, Bartoli, Carrà and Samochowiec8
Beyond emotional dysregulation, abnormalities within limbic circuitry may also contribute to impaired inhibitory control, a core feature of bipolar disorder. A meta-analysis of 318 studies reported a medium effect size (Hedges’ g = 0.52) for inhibitory control deficits, commonly assessed using the go/no-go (GNG) task, in individuals with bipolar disorder. Reference Wright, Lipszyc, Dupuis, Thayapararajah and Schachar9 Neuroimaging studies included adults with bipolar disorder suggest altered limbic and fronto-limbic activity during a neutral GNG task. Reference Kaladjian, Jeanningros, Azorin, Nazarian, Roth and Anton10 However, whether appetite hormone dysregulation and limbic functional connectivity abnormalities are associated with inhibitory control deficits especially in younger populations remains unknown.
In the present study, we evaluated appetite hormone levels (including adiponectin, ghrelin, insulin and leptin), inhibitory control function and limbic network connectivity changes by using resting-state functional magnetic resonance imaging (MRI) in youths with bipolar disorder. We hypothesised that youths with bipolar disorder would exhibit altered appetite hormone levels (i.e. increased insulin levels), impaired inhibitory control function and altered functional connectivity between limbic regions of interest (ROIs). In addition, we hypothesised that limbic network connectivity changes would be associated with appetite hormone dysregulation and impaired inhibitory control in youths with bipolar disorder.
Method
Participants
In all, 68 youths aged 12–22 years with bipolar disorder, and 57 age- and gender-matched controls, were enrolled. The diagnosis of bipolar disorder was made by senior child and adolescent psychiatrists according to the criteria outlined in the DSM-5. Exclusion criteria included intellectual disability, substance or alcohol use disorders, epilepsy, pregnancy or breastfeeding, unstable physical conditions and contraindications for MRI. Normal participants had no DSM-5 diagnoses or major physical illnesses. Demographic characteristics, including age, gender and body mass index (BMI), were recorded and clinical assessments, including the Young Mania Rating Scale (YMRS) Reference Young, Biggs, Ziegler and Meyer11 and the Montgomery–Åsberg Depression Rating Scale (MADRS), Reference Montgomery and Åsberg12 were examined. The GNG task was used to measure inhibitory control function. Participants were instructed to react promptly by pressing a key when the × symbol appeared and to refrain from responding when the + symbol was shown. Following a pretest where all answers were correct, participants undertook the formal test, during which their correct answers, errors, and reaction times (mean±s.d.) were recorded. The GNG task has been commonly used in our previous studies. Reference Chen, Bai, Tsai, Hsu and Chen13
Ethical statement
The study protocols were approved by the Institutional Review Boards of Taipei Veterans General Hospital (approval number: 2021-02-006A) and conducted following the Declaration of Helsinki of 1975, as revised in 2013. Written informed consent was obtained from all participants. If participants were aged <18 years, written informed consent was also obtained from their legal guardians.
Measurement of appetite hormones
Participants were asked to obtain at least 8 h of sleep and nothing per mouth (NPO) before the blood test. Adherence was verified orally and in writing with each participant and their parents, if aged under 18 years. All measured hormones (leptin, ghrelin, insulin and adiponectin) reach diurnal trough levels in the morning, with meal-induced fluctuations thereafter. Reference Templeman, Smith, Walhin, Middleton, Gonzalez and Karagounis14 Therefore, fasting serum samples were collected between 09.00 and 12.00 h in serum separator tubes, clotted for 30 min and stored at –80°C until use. The appetite hormones, including leptin, ghrelin, insulin and adiponectin, were examined. Ghrelin was measured by radioimmunoassay kit (Peninsula Laboratories, San Carlos, California, USA), with intra- and inter-assay coefficients of variation of 5 and 8%, respectively. Reference Vila, Grimm, Resl, Heinisch, Einwallner and Esterbauer15 Insulin concentrations were analysed by radioimmunoassay kit (Coat-A Count Insulin; Diagnostic Product Corporation, Los Angeles, California, USA), with intra- and inter-assay coefficients of variation of 5 and 6%, respectively. Reference Casazza, Goran and Gower16 Serum adiponectin level was measured with a quantitative human adiponectin enzyme-linked immunosorbent assay kit (B-Bridge International, Mountain View, California, USA), with intra- and inter-assay coefficients of variation of <4 and <8%, respectively. Reference Yu, Yang, Zhang, Lu, Zhang and Wang17 All assays were performed according to the vendor’s instructions. The final absorbance of each sample of the mixture was measured and analysed at 450 nm, using an enzyme-linked immunosorbent assay plate reader with Bio-Tek Power Wave Xs and Bio-Tek’s KC junior software (Winooski, Vermont, USA). The standard range was considered as specified in the vendor’s instructions. A linear regression R 2-value of approximately 0.95 was considered a reliable standard curve.
MRI image acquisition
Images were acquired at Taipei Veterans General Hospital, using a 3-Tesla scanner (GE Healthcare Life Sciences, Little Chalfont, UK) equipped with a quadrature head coil. Anatomical whole-brain T1-weighted images were obtained by using a magnetisation-prepared rapid acquisition gradient echo three-dimensional T1-weighted sequence with the following parameters: repetition time 12.2 ms, echo time 5.2 ms, 168 axial slices, flip angle 12°, field of view 256 × 256 mm, matrix size 256 × 256 and slice thickness 1 mm. The resting-state MRI was acquired by using a T2-weighted gradient-echo echo-planar sequence with the following parameters: repetition time 2500 ms, echo time 30 ms, 43 axial slices, flip angle 90° and voxel size 3.5 × 3.5 × 3.5 mm. Each participant underwent the acquisition of 200 MRI volumes while keeping their eyes closed, maintaining a neutral mental state and avoiding movement or falling asleep.
Preprocessing
Neuroimaging data were analysed using Statistical Parametric Mapping 12 for Windows (SPM12; Wellcome Centre for Human Neuroimaging, University College London, London, UK; https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) implemented in MATLAB R2024a for Windows (The MathWorks, Inc., Natick, Massachusetts, USA https://www.mathworks.com/help/install/ug/install-products-with-internet-connection.html). The preprocessing of functional and anatomical data was performed with the CONN toolbox pipeline. Reference Whitfield-Gabrieli and Nieto-Castanon18 Functional scans were realigned to the first volume by using SPM12’s realign and unwarp procedure with a six-parameter rigid body transformation and B-spline interpolation. Slice-timing correction was performed with sinc interpolation to resample each slice to the mid-acquisition time. Potential outlier scans were identified using ART Reference Whitfield-Gabrieli, Nieto-Castanon and Ghosh19 as acquisitions with framewise displacement above 0.5 mm or global blood oxygenation level-dependent (BOLD) signal changes above 3 s.d. A reference BOLD image was generated by averaging non-outlier scans. Functional and anatomical data were normalised into standard Montreal Neurological Institute space, segmented into grey matter, white matter and cerebrospinal fluid tissue classes, and resampled to 2 mm isotropic voxels via direct normalisation procedure using the SPM unified segmentation and normalisation algorithm with the default IXI-549 tissue probability map template. Reference Calhoun, Wager, Krishnan, Rosch, Seymour and Nebel20 Functional data were smoothed using spatial convolution with a Gaussian kernel of 6 mm full width at half maximum. Participants were excluded if their motion or outlier metrics exceeded the 1.5 times interquartile range above the third quartile of the group distribution.
Denoising
Functional data were denoised with a standard denoising pipeline, including the regression of white matter and cerebrospinal fluid CompCor components (5 each), 12 motion parameters (6 realignment and their first-order derivatives) and identified outlier scans. Additionally, session effects and linear trends were regressed, followed by bandpass filtering (0.008–0.09 Hz). Reference Neito-Castanon21 CompCor components were estimated from eroded masks to ensure signal purity. Reference Behzadi, Restom, Liau and Liu22
First-level analysis
Limbic network is defined as 17 ROIs, including bilateral hippocampus, bilateral amygdala, bilateral anterior PHG and posterior PHG, bilateral nucleus accumbens and bilateral thalamus as defined by the Harvard–Oxford Atlas; Reference Makris, Goldstein, Kennedy, Hodge, Caviness and Faraone23 bilateral subgenual and pregenual anterior cingulate gyrus as defined by the Automated Anatomical Labelling Atlas 3; Reference Rolls, Huang, Lin, Feng and Joliot24 and hypothalamus as defined by the CIT168 Reinforcement Learning Atlas. Reference Pauli, Nili and Tyszka25 ROI-to-ROI connectivity matrices were estimated characterising the functional connectivity between each pair of regions among 17 ROIs. Functional connectivity strength was represented by Fisher-transformed bivariate correlation coefficients from a general linear model, characterising the association between their BOLD signal time-series.
Statistical analysis
Statistical analyses were performed with SPSS version 25 for Windows (2017 release; IBM Corporation, Armonk, New York, USA; https://www.ibm.com/products/spss). Group differences in demographic and clinical characteristics were assessed with independent t-test for continuous variables and χ 2-tests for nominal variables. A p-value of <0.05 (two-tailed) was considered statistically significant.
Appetite hormones and adiposity
Group differences in appetite hormone levels were assessed with multiple linear regression, adjusting for BMI and antipsychotic usage. To evaluate the relationship between BMI and hormones, partial Spearman’s correlations (adjusting for age, gender, medications and symptom severity) were performed for the total sample and each group separately. Additionally, analysis of covariance was used to test for group × BMI interaction effects.
Intra-limbic functional connectivity
A generalised linear model examined group differences in Z-values of ROI-to-ROI connectivity analysis, controlling for age, gender, medication use (antidepressants, antipsychotics and mood stabilisers), BMI and symptom severity as rated by the MADRS and YMRS. For each individual connection, the first-level connectivity measure served as the dependent variable, whereas group and other individual-level identifiers were entered as independent variables. Cluster-level inferences were based on nonparametric statistics using threshold free cluster enhancement (TFCE), with 1000 residual-randomisation iterations. Reference Smith and Nichols26 Results were thresholded by family-wise error (FWE) corrected p < 0.05.
Correlation analyses
Partial Spearman’s correlations examined associations between appetite hormones (log-transformed) and functional connectivity Z-values. Similarly, partial Pearson’s correlations were used to assess the relationship between GNG task performance and functional connectivity Z-values. Both models adjusted for group, age, gender, medications, symptom severity and BMI additionally for appetite hormones. False discovery rate correction was applied to control for multiple comparisons (q < 0.05).
Sensitivity analysis
To assess the robustness of our major findings, we conducted several sensitivity analyses. First, we repeated the primary analyses after excluding participants with bipolar disorder who had psychotic features, using the same preprocessing pipeline, statistical models, covariates and significance thresholds as in the primary analyses. Psychotic features were assessed using the Positive and Negative Syndrome Scale and operationally defined as a score of ≥4, indicating at least moderate severity, on at least two of the following three items: delusions (P1), hallucinatory behaviour (P3) and unusual thought content (G9). This sensitivity analysis excluded nine participants with bipolar disorder, representing 13.24% of the original bipolar disorder cohort. Second, to evaluate whether the hypothalamic connectivity finding was influenced by spatial signal mixing from adjacent regions, we repeated the ROI-to-ROI analysis using the same preprocessing pipeline except that spatial smoothing was omitted. The same ROI definitions, denoising procedure, second-level model and covariates, group contrast and TFCE-based FWE correction were retained from the primary analysis. Third, to assess potential nonlinear age-related effects across the 12–22 year age range, we conducted additional regression analyses of the significant functional connectivity findings. The models included diagnostic group and gender as covariates. Linear models included centred age, whereas quadratic models additionally included the squared centred-age term (age_c 2). Linear and quadratic nested models were compared with F-tests in the pooled sample and within each diagnostic group, and model fit was additionally evaluated with the Akaike information criterion (AIC). We further tested group × linear age and group × quadratic age interaction terms to examine whether age-related associations differed between diagnostic groups.
Results
Table 1 summarises the demographic and clinical characteristics of youths with bipolar disorder and normal controls. No significant differences were observed among groups in age (p = 0.055) or gender distribution (p = 0.297). Youths with bipolar disorder showed significantly higher BMI (p = 0.015). For symptom severity, significant group differences were observed on both the MADRS and YMRS (both p < 0.001). The majority of the bipolar disorder cohort was either euthymic (n = 34, 50.0%) or depressed (n = 31, 45.6%), with few participants exhibiting manic or hypomanic mood states (n = 3, 4.4%).
Demographic and clinical characteristics between patients and controls
Table 1 Long description
The table compares demographic and clinical characteristics between youths with bipolar disorder and healthy controls. It has 10 rows and 5 columns. The columns are labeled as Youths with bipolar disorder (n = 68), Healthy controls (n = 57), and p-value (95% CI). The rows include Age, Gender, BMI, Age at onset, Bipolar disorder subtype, Presence of psychotic feature, Prior mood episodes, Psychotropic medications, Mood state, Symptom severity, Appetite hormones, and Go/no-go task. Each row provides specific data points for both groups and the corresponding p-values indicating statistical significance. Notable trends include higher BMI in youths with bipolar disorder and significant differences in symptom severity scores on the MADRS and YMRS.
BMI, body mass index; MADRS, Montgomery–Åsberg Depression Rating Scale; YMRS, Young Mania Rating Scale. Bold type indicates the statistical significance p < 0.05.
a. Log-transformed.
Significantly higher leptin (p = 0.005, Cohen’s d = 0.540) and insulin (p < 0.001, Cohen’s d = 0.669) levels can be observed in youths with bipolar disorder than in healthy controls (Supplementary Table 1 available at https://doi.org/10.1192/bjo.2026.12144). These differences remained robust after adjusting for BMI and antipsychotic usage; notably, ghrelin levels also emerged as significantly different between groups in the adjusted model (Supplementary Table 2). Regarding the relationship between adiposity and hormone levels, BMI showed consistent associations across groups (Supplementary Table 3). Leptin showed strong positive correlations in all three groups (whole cohort: p = 0.002, r = 0.368; bipolar disorder: p < 0.001, r = 0.67; normal control: p < 0.001, r = 0.58). Adiponectin exhibited consistent moderate negative correlations across groups (whole cohort: p = 0.042, r = − 0.242; bipolar disorder: p = 0.003, r = − 0.391; normal control: p = 0.031, r = 0.308). Insulin displayed a modest positive correlation in normal control (p = 0.031, r = 0.308) and a trend-level association in youths with bipolar disorder (p = 0.064, r = 0.256). Ghrelin showed no significant correlations with BMI in any group. The analysis of covariance revealed non-significant group × BMI interaction effects for all four appetite hormones (all p > 0.05). The results remained robust after excluding participants with psychosis (Supplementary Table 4). In the GNG task, youths with bipolar disorder showed significantly worse performance than healthy controls, characterised by lower accuracy across all trials (p = 0.027, Cohen’s d = 0.411), higher rate of omission errors (p = 0.015, Cohen’s d = 0.411) and prolonged mean reaction time (p = 0.036, Cohen’s d = 0.388) (Table 1). Notably, the bipolar disorder group demonstrated significantly greater reaction time variability (p < 0.001, Cohen’s d = 0.640).
Figure 1 and Table 2 illustrated that the ROI-to-ROI analyses found significant hyperconnectivity in the bipolar disorder group compared with controls within the limbic network, after adjusting for gender, age, medication use, symptom severity and BMI (TFCE, FWE-corrected p < 0.05). One cluster comprising two edges was identified: the right anterior PHG–right thalamus connection and the right anterior PHG–hypothalamus connection (Fig. 1 and Table 2). Representative edges are illustrated in Fig. 2, with bar plots showing group differences in Fisher z-transformed connectivity strength. The direction of the right anterior PHG–thalamus and anterior PHG–hypothalamus effects was preserved after excluding participants with psychotic features (Supplementary Table 5), whereas no cluster survived TFCE-based FWE correction in the unsmoothed analysis, although both edges retained the bipolar disorder > control direction (Supplementary Table 6). Thus, the corrected cluster-level result was not robust to omission of spatial smoothing. The correlation analyses, adjusting for group, age, gender, BMI, medication use and symptom severity, showed no significant associations between functional connectivity, appetite hormones or GNG task performances.
ROI-to-ROI functional connectivity circular connectivity diagram between youths with bipolar disorder and healthy controls among limbic-associated ROIs, adjusted for gender, age, BMI, medication usage (antidepressants, antipsychotics and mood stabilisers), symptom severity (by MADRS and YMRS scales). Significant functional connectivity changes were observed between right aPHG, right thalamus and hypothalamus. Cluster-level inferences were based on nonparametric statistics using threshold free cluster enhancement (TFCE). Results were corrected for multiple comparisons using the family-wise error rate p < 0.05. The colour bar on the right side corresponds to the TFCE values. aPHG, anterior parahippocampal gyrus; BMI, body mass index; MADRS, Montgomery–Åsberg Depression Rating Scale; ROI, region of interest; YMRS, Young Mania Rating Scale; r/R, right; l/L, left; ACC, anterior cingulate cortex; aPaHC, anterior parahippocampal gyrus; pPaHC, posterior parahippocampal gyrus; aal3.ACC_pre_R, right pregenual anterior cingulate gyrus; aal3.ACC_pre_L, left pregenual anterior cingulate gyrus; aal3.ACC_sub_R, right subgenual anterior cingulate gyrus; aal3.ACC_sub_L, left subgenual anterior cingulate gyrus; aal3/AAL3, automated anatomical labelling atlas 3.
ROI-to-ROI functional connectivity analysis between youths with bipolar disorder and healthy controls among limbic-associated ROIs, adjusted for gender, age, medication use (antidepressants, antipsychotics and mood stabilisers), symptom severity (by MADRS and YMRS scales), and BMI-adjusted ((a), (b), (c)). Brain maps show the anatomical location of significant connections across axial and sagittal views. (d) Bar plots illustrate group differences in Fisher z-transformed connectivity strength (mean±s.d.) for the mean functional connectivity of cluster 1 and representative edges. Significant between-group differences were observed in one cluster with two significant edges. Statistical details are provided in Table 2. aPHG, anterior parahippocampal gyrus; BMI, body mass index; MADRS, Montgomery–Åsberg Depression Rating Scale; ROI, region of interest; YMRS, Young Mania Rating Scale.
ROI-to-ROI functional connectivity analysis between youths with bipolar disorder and healthy controls among limbic-associated ROIs, adjusted for gender, age, body mass index, medication usage (antidepressants, antipsychotics and mood stabilisers) and symptom severity (by MADRS and YMRS scales)
Table 2 Long description
A table with four rows and four columns comparing group differences in functional connectivity among limbic-associated regions in bipolar disorder and control groups. The columns are labeled Analysis unit, Group difference, Statistic (TFCE/T), and p-value. Row 1: Analysis unit, Cluster 1; Group difference, Bipolar disorder > normal controls; Statistic (TFCE/T), 25.94; p-value, 0.045 (FWE). Row 2: Analysis unit, Right aPHG-right thalamus; Group difference, Bipolar disorder > normal controls; Statistic (TFCE/T), 3.33; p-value, 0.002 (uncorrected). Row 3: Analysis unit, Right aPHG-hypothalamus; Group difference, Bipolar disorder > normal controls; Statistic (TFCE/T), 2.83; p-value, 0.006 (uncorrected).
Cluster-level inferences were based on nonparametric statistics using threshold free cluster enhancement (TFCE). Results were corrected for multiple comparisons using the family-wise error rate (FWE) p < 0.05. T, conventional t statistic; ROI, region of interest; MADRS, Montgomery–Åsberg Depression Rating Scale; YMRS, Young Mania Rating Scale; aPHG, anterior parahippocampal gyrus.
Additional analyses were conducted to assess potential nonlinear age-related effects on the significant functional connectivity findings. In group-stratified analyses, inclusion of a quadratic age term significantly improved model fit in the normal control group (F(1, 47) = 5.66, p = 0.02; AIC: linear, −38.82; quadratic, −42.62), but not in the bipolar disorder group (F(1, 58) = 0.17, p = 0.69; AIC: linear, −41.93; quadratic, −40.10). In the pooled sample, the quadratic model showed a modestly lower AIC than the linear model (−83.24 v. −81.99), although the improvement did not reach statistical significance in the nested-model comparison (F(1, 108) = 3.15, p = 0.08). Neither the group × linear age nor group × quadratic age interaction was statistically significant (95% CI −0.002 to 0.022; p = 0.107), providing no evidence that the cross-sectional age–functional connectivity associations differed between diagnostic groups (Supplementary Table 7).
Discussion
In this study, youths with bipolar disorder exhibited concurrent abnormalities across neural, metabolic and behavioural domains, including elevated leptin and insulin levels, impaired inhibitory control on the GNG task and increased resting-state functional connectivity between the right anterior PHG and right thalamus. Rather than diffuse connectivity abnormalities across the examined limbic regions, our findings indicated relatively focal hyperconnectivity within limbic–subcortical circuitry.
Bipolar disorder is characterised by emotion dysregulation, which has been associated with abnormalities in neural systems supporting the processing and regulation of emotionally salient information. Reference Phillips, Ladouceur and Drevets27 Within this framework, the thalamus supports the orienting, selection and integration of behaviourally relevant sensory and affective information, Reference Zhou, Zhu, Hou, Chen, Chen and Yang28 whereas the anterior PHG participates in medial temporal memory circuits in which amygdala-derived affective signals modulate item- and context-related mnemonic representations. Reference Roesler and McGaugh29 Increased anterior PHG–thalamus connectivity may therefore partially reflect altered coupling between affectively modulated mnemonic processing and the selection and integration of perceived information. The right anterior PHG–thalamus hyperconnectivity observed in the present cohort converges with resting-state functional MRI evidence in adults with bipolar disorder demonstrating increased functional coupling between the pulvinar thalamus and parahippocampal regions. Reference Teng, Lu, Wang, Li, Tu and Hung30 At a broader network level, a large-scale meta-analysis involving 1526 individuals with bipolar disorder similarly reported hyperconnectivity between the thalamus and an affective network encompassing the PHG. Reference Gong, Wang, Chen, Qi, Luo and Wang31 Together with reports of salience-related and limbic abnormalities in paediatric bipolar disorder and youths at familial risk, these findings suggest that disturbances involving interacting limbic and thalamic systems may be detectable relatively early in the development of bipolar disorder. Reference Saccaro, Delavari, Van De Ville and Piguet32 Accordingly, limbic hyperconnectivity may not represent a uniform or directionally static signature across all phases of bipolar disorder. Reference Saccaro, Delavari, Van De Ville and Piguet32 Although the current cross-sectional design cannot definitively separate state-related mood fluctuations from an enduring trait vulnerability, the observed anterior PHG–thalamus hyperconnectivity in this predominantly euthymic or depressed youth cohort may be relevant to characterising limbic–thalamic connectivity alterations in young people with bipolar disorder. Longitudinal studies are needed to determine whether anterior PHG–thalamus connectivity varies across mood states and developmental trajectories, and whether it is associated with the clinical course of bipolar disorder.
Hypothalamic functional connectivity has been comparatively underexamined in youths with bipolar disorder, whereas in major depressive disorder with psychotic features, the reduced functional connectivity between the hypothalamus and subgenual cortex has been linked to depressive severity and cortisol dysregulation. Reference Sudheimer, Keller, Gomez, Tennakoon, Reiss and Garrett33 Beyond hypothalamic–pituitary–adrenal axis regulation, the hypothalamus is a crucial node of the central homeostatic network, functioning to integrate interoceptive metabolic states with limbic affective processing. Specifically, circulating hormones such as leptin and insulin act upon the hypothalamic arcuate nucleus to regulate energy homeostasis. Reference Obradovic, Sudar-Milovanovic, Soskic, Essack, Arya and Stewart34 These functions are relevant to bipolar disorder, in which neuroendocrine dysregulation and cumulative allostatic load have been implicated in illness progression and systemic comorbidity. Reference Belvederi Murri, Prestia, Mondelli, Pariante, Patti and Olivieri35
However, both of our functional connectivity findings did not remain statistically significant in the unsmoothed sensitivity analysis. Spatial smoothing in small subcortical nuclei may both increase sensitivity by improving signal/noise ratio and increase susceptibility to signal mixing from adjacent tissues. Given these competing considerations, the findings should be regarded as preliminary and warrant replication using higher-resolution acquisitions.
Our findings of elevated leptin and insulin levels in youths with bipolar disorder are consistent with a systematic review and meta-analysis of adult bipolar disorder samples. Reference Misiak, Kowalski, Stańczykiewicz, Bartoli, Carrà and Samochowiec8 These metabolic alterations may be understood within the broader context of the low-grade inflammatory state implicated in the pathophysiology of bipolar disorder. Reference Grande, Berk, Birmaher and Vieta1 Misiak et al included individuals across different mood states and found that leptin and insulin levels were also elevated in euthymic samples, suggesting that these alterations may represent persistent or trait-like metabolic abnormalities rather than solely acute mood-state effects. Reference Misiak, Kowalski, Stańczykiewicz, Bartoli, Carrà and Samochowiec8 However, mood state and medication exposure were only partially accounted for through covariate adjustment, the present cross-sectional finding cannot distinguish illness-related metabolic differences from state- or treatment-related effects. Prospective studies with detailed medication exposure data and repeated assessments across mood states are needed to clarify the temporal stability and clinical correlates of these hormone alterations.
Behaviourally, the GNG task findings suggest a profile of attentional instability rather than a simple increase in impulsive responding. Youths with bipolar disorder showed lower response accuracy, more omission errors, longer mean reaction time and greater reaction-time variability – with reaction-time variability yielding the largest effect size. This dissociation is inconsistent with classic motor impulsivity, in which commission errors typically predominate. Reference Wright, Lipszyc, Dupuis, Thayapararajah and Schachar9 Instead, the co-occurrence of elevated omission errors and increased intra-individual reaction-time variability may reflect inconsistent attentional engagement rather than a pure deficit in motor inhibition. Reference Dykiert, Der, Starr and Deary36 Prior work has implicated thalamic systems in attentional regulation and cortical arousal during task performance, Reference Schiff, Shah, Hudson, Nauvel, Kalik and Purpura37 and altered anterior PHG–thalamic connectivity may therefore be broadly relevant to the increased reaction-time variability observed in youths with bipolar disorder. The present data do not establish a direct association between anterior PHG–thalamic connectivity and GNG task performance. Future studies combining task-based functional MRI with response-inhibition paradigms are needed to determine whether anterior PHG–thalamic coupling is associated with behavioural performance in youths with bipolar disorder.
A central interpretive challenge is that despite significant between-group differences in all three domains, multivariable-adjusted and false discovery rate-corrected analyses revealed no robust linear coupling among limbic functional connectivity, appetite hormones and GNG task performance. Growing evidence suggests that distinct biological processes may co-exist relatively independently within a single psychiatric diagnosis without exhibiting strong linear coupling, Reference Vanes and Dolan38 which is broadly consistent with our findings of multidimensional abnormalities in the absence of direct associations. Another possible explanation is that such relationships may be less detectable early in illness course; longitudinal work in recently diagnosed bipolar disorder suggests that brain–behaviour associations may become clearer over follow-up intervals of 6–27 months. Reference Macoveanu, Damgaard, Ysbæk-Nielsen, Frangou, Yatham and Chakrabarty39 This raises the possibility that neural, metabolic and behavioural abnormalities in youth bipolar disorder may become more tightly coupled over the course of illness progression and neurodevelopment.
The sensitivity analyses provided mixed evidence regarding the robustness of the neuroimaging findings. After excluding participants with psychotic features, right anterior PHG–thalamus connectivity remained greater in youths with bipolar disorder than in healthy controls, and group differences in leptin and insulin levels were retained. In contrast, both edges (right anterior PHG–hypothalamus, right anterior PHG–right thalamus) did not reach statistical significance when spatial smoothing was omitted. Given the small size of the subcortical ROI and the competing concerns of spatial signal mixing in smoothed data and reduced spatial signal/noise ratio in unsmoothed data, the findings should be regarded as preliminary. Accordingly, the present results identify increased right anterior PHG–thalamus and right anterior PHG–hypothalamus connectivity in youths with bipolar disorder within the specified analytic framework, but require replication using acquisition and preprocessing approaches optimised for small subcortical structures.
Several limitations merit acknowledgment. First, the cross-sectional design precludes causal inference regarding the temporal ordering of limbic hyperconnectivity, metabolic abnormalities and inhibitory deficits. Second, the bipolar disorder sample was heterogeneous in mood state and no subgroup analyses by mood state or bipolar disorder subtype were conducted; controlling for MADRS and YMRS scores captures symptom severity linearly, but cannot account for qualitative differences between mood states. Third, pubertal status was not formally assessed. Although additional analyses examined linear and quadratic age associations, chronological age remains an imperfect proxy for pubertal and neurodevelopmental stage, and the cross-sectional design precludes characterisation of within-individual developmental trajectories. Fourth, medication exposure was coded as class-level binary variables without information on dose, duration, cumulative exposure and medication-specific metabolic risk. Lifestyle and physiological factors affecting endocrine measurements were not systematically assessed. Both represent potential confounders of fasting hormone measurements. In addition, without these data, we were unable to assess their potential impact on limbic-seeded functional connectivity. Prospective studies incorporating detailed medication exposure data and repeated assessments across mood states are needed to clarify these potential effects. Fifth, the lack of a comprehensive neuropsychological battery limits our ability to determine the extent to which broader cognitive impairment may have influenced task performance and functional connectivity findings. Sixth, the study did not include stress-related endocrine markers such as cortisol, nor more comprehensive metabolic indices such as the Homeostatic Model Assessment for Insulin Resistance, which may have limited sensitivity to more nuanced endocrine–neural interactions in bipolar disorder. Seventh, the neutral GNG task paradigm primarily captures response inhibition and motor inhibitory control, and may therefore have been less sensitive to affective or reward-related impulsivity more directly linked to limbic circuitry. Eighth, in the absence of an active psychiatric comparison group and significant associations across these domains, the findings cannot be considered specific to bipolar disorder or interpreted as evidence of a shared underlying mechanism.
Future longitudinal work should examine whether the three domains of abnormality identified here remain dissociated across illness progression or converge over time with repeated mood episodes. Incorporating task-based functional MRI during emotionally valenced paradigms, comprehensive metabolic phenotyping including the Homeostatic Model Assessment for Insulin Resistance, hypothalamic–pituitary–adrenal axis biomarkers and formal causal mediation analyses would be particularly valuable for distinguishing parallel dimensions of abnormality from potentially shared neurometabolic processes in youths with bipolar disorder.
Supplementary material
The supplementary material is available online at https://doi.org/10.1192/bjo.2026.12144
Data availability
The data that support the findings of this study are available on request from the corresponding author, M.-H.C. The data are not publicly available due to Taiwan’s clinical trial ethical regulations.
Acknowledgements
The authors thank Mr I-Fan Hu (Courtauld Institute of Art, University of London; National Taiwan University) for his friendship and support. Mr Hu declares no conflicts of interest. Generative AI was used to assist with language polishing of the manuscript. The authors reviewed and edited the final content and take full responsibility for the manuscript.
Author contributions
M.-H.C., Y.-M.B. and J.-W.H. designed the study. L.-K.C. and H.-J.C. analysed the data. H.-J.C. drafted the paper. L.-C.C., P.-C.T. and S.-J.T. performed the literature review and critically reviewed the manuscript and interpreted the data. All authors contributed substantially to the manuscript, and approved the final manuscript for submission. All authors are responsible for the integrity, accuracy and presentation of the data.
Funding
The study was supported by grants to M.-H.C. from Taipei Veterans General Hospital (grant numbers V113C-039, V113C-011, V113C-010, V114C-089, V114C-064, V114C-217, V115C-169, V115C-172 and V115C-019); Yen Tjing Ling Medical Foundation (grant numbers CI-113-32, CI-113-30, CI-114-35 and CI-115-35); Ministry of Science and Technology, Taiwan (grant numbers NSTC113-2314-B-075-042, NSTC114-2628-B-A49-008-MY3, NSTC114-2314-B-075-022, NSTC114-2314-B-075-023, NSTC 115-2410-H-075-002 and NSTC 115-2314-B-075-027); Taipei, Taichung, Kaohsiung Veterans General Hospital, Tri-Service General Hospital, Academia Sinica Joint Research Program (grant numbers VTA114-V1-4-1 and VTA115-V1-5-1); Veterans General Hospitals and University System of Taiwan Joint Research Program (grant numbers VGHUST114-G1-9-1 and VGHUST115-G1-7-1); and Cheng Hsin General Hospital (grant numbers CY11402-1, CY11402-2, CY115-11402-1 and CY115-11402-2). The funding sources had no role in any process of our study.
Declaration of interest
None.
Open access
eLetters
No eLetters have been published for this article.