Introduction
Major depressive disorder (MDD) is a highly heterogeneous syndrome comprising a mixture of affective, cognitive, and psychomotor symptoms with unknown causes (Njenga, Ramanuj, de Magalhães, & Pincus, Reference Njenga, Ramanuj, de Magalhães and Pincus2024; Sumedrea, Sumedrea, & Săvulescu, Reference Sumedrea, Sumedrea and Săvulescu2024; Tunc-Ozcan et al., Reference Tunc-Ozcan, Peng, Zhu, Dunlop, Contractor and Kessler2019). Age of onset is considered a key factor for the heterogeneity in MDD and could serve as a useful clinical indicator for subtyping MDD (Cui et al., Reference Cui, Wang, Cao, Wu, Peng, Chen, Yang, Rong, Liu and Fang2023; Park et al., Reference Park, Hahn, Hwang, Kim, Jun, Lee, Kim, Yim and Park2014). Youth-onset MDD is more likely to be associated with family history, genetic factors, and early-life stressful events, typically presenting more severe depressive symptoms, higher suicide risk, and more frequent comorbidities (Parker et al., Reference Parker, Roy, Hadzi-Pavlovic, Mitchell and Wilhelm2003; Zisook et al., Reference Zisook, Rush, Albala, Alpert, Balasubramani, Fava, Husain, Sackeim, Trivedi and Wisniewski2004, Reference Zisook, Lesser, Stewart, Wisniewski, Balasubramani, Fava, Gilmer, Dresselhaus, Thase, Nierenberg, Trivedi and Rush2007). By contrast, adult-onset MDD is more strongly linked to environmental risk factors, low socioeconomic status, low educational level, and chronic diseases (Bukh et al., Reference Bukh, Bock, Vinberg, Gether and Kessing2011; Sayed, Mahmoud, & Hadad, Reference Sayed, Mahmoud and Hadad2022). From a neuroimaging perspective, previous studies have shown differential cortical morphological change patterns between youth-onset and adult-onset MDD. For instance, cortical thickness (CT) reductions in the prefrontal cortex, temporal cortex, anterior cingulate gyrus, pre- and postcentral gyrus, and lingual gyrus are often reported in youth-onset MDD (Jaworska et al., Reference Jaworska, MacMaster, Gaxiola, Cortese, Goodyear and Ramasubbu2014; Marrus et al., Reference Marrus, Belden, Nishino, Handler, Ratnanather, Miller, Barch, Luby and Botteron2015; Shen et al., Reference Shen, Jiang, Cheng, Ye, Lu, Zhou, Li, Dai, He and Xu2019; Truong et al., Reference Truong, Minuzzi, Soares, Frey, Evans, MacQueen and Hall2013), whereas the orbitofrontal cortex, anterior and posterior cingulate gyrus, prefrontal cortex, temporal cortex, insula, and fusiform gyrus are found to be preferentially affected in adult-onset MDD (Disabato et al., Reference Disabato, Morris, Hranilovich, D’Angelo, Zhou, Wu, Doraiswamy and Sheline2014; Lim et al., Reference Lim, Jung, Ahn, Won, Hahn, Lee, Kim and Lee2012; Schmaal et al., Reference Schmaal, Hibar, Sämann, Hall, Baune, Jahanshad, Cheung, van Erp, Bos, Ikram, Vernooij, Niessen, Tiemeier, Hofman, Wittfeld, Grabe, Janowitz, Bülow, Selonke, Völzke and Veltman2017; Shen et al., Reference Shen, Jiang, Cheng, Ye, Lu, Zhou, Li, Dai, He and Xu2019). Nevertheless, the neurobiological factors underlying the divergent change patterns remain largely unknown.
The human brain is a complex network involving a broadly distributed set of regions interconnected by white matter tracts, known as the structural connectome (Sporns, Reference Sporns2011; Sporns, Tononi, & Kötter, Reference Sporns, Tononi and Kötter2005). Fiber tracts facilitate functional organization and communication between brain regions (Avena-Koenigsberger, Misic, & Sporns, Reference Avena-Koenigsberger, Misic and Sporns2017; Fries, Reference Fries2015). Crucially, the structural connectome can also act as conduits of pathological spread, supporting an expansion of neural pathologies along existing white matter pathways (Fornito, Zalesky, & Breakspear, Reference Fornito, Zalesky and Breakspear2015; Jucker & Walker, Reference Jucker and Walker2018; Warren et al., Reference Warren, Rohrer, Schott, Fox, Hardy and Rossor2013; Zuo et al., Reference Zuo, Liu, Lan, Chen, Li, Yan, Li, Luo, Kemp, Lui, Suo and Gong2026). In this context, there is ample neuroimaging evidence demonstrating that the spatial topographies of cortical morphological changes in neuropsychiatric disorders are constrained by the brain’s connectome architecture (Chopra et al., Reference Chopra, Segal, Oldham, Holmes, Sabaroedin, Orchard, Francey, O’Donoghue, Cropley, Nelson, Graham, Baldwin, Tiego, Yuen, Allott, Alvarez-Jimenez, Harrigan, Fulcher, Aquino and Fornito2023; Georgiadis et al., Reference Georgiadis, Larivière, Glahn, Hong, Kochunov, Mowry, Loughland, Pantelis, Henskens, Green, Cairns, Michie, Rasser, Catts, Tooney, Scott, Schall, Carr, Quidé and Kirschner2024; Shen et al., Reference Shen, Wang, Chen, Lu, Li, Wang, Cao, Chen, Chen, Chen, Chen, Cheng, Chu, Cui, Cui, Deng, Gong, Guo, He and Yu2026; Yau et al., Reference Yau, Zeighami, Baker, Larcher, Vainik, Dadar, Fonov, Hagmann, Griffa, Mišić, Collins and Dagher2018; Zeighami et al., Reference Zeighami, Ulla, Iturria-Medina, Dadar, Zhang, Larcher, Fonov, Evans, Collins and Dagher2015), frequently followed by performance of an epicentre mapping analysis to further identify the likely regional epicentres (Chopra et al., Reference Chopra, Segal, Oldham, Holmes, Sabaroedin, Orchard, Francey, O’Donoghue, Cropley, Nelson, Graham, Baldwin, Tiego, Yuen, Allott, Alvarez-Jimenez, Harrigan, Fulcher, Aquino and Fornito2023; Georgiadis et al., Reference Georgiadis, Larivière, Glahn, Hong, Kochunov, Mowry, Loughland, Pantelis, Henskens, Green, Cairns, Michie, Rasser, Catts, Tooney, Scott, Schall, Carr, Quidé and Kirschner2024; Jiang et al., Reference Jiang, Palaniyappan, Luo, Chang, Zhang, Tang, Zhang, Li, Zhou, Yu, Li, An, Zhou, Huang, Tsai, Lin, Cheng, Wang, Yao, Cheng and Feng2024; Yau et al., Reference Yau, Zeighami, Baker, Larcher, Vainik, Dadar, Fonov, Hagmann, Griffa, Mišić, Collins and Dagher2018). In addition, there is now widespread agreement that highly connected hub regions in the structural connectome are more vulnerable to diseases and thereby tend to exhibit more prominent abnormalities, known as the hub vulnerability hypothesis (Fornito et al., Reference Fornito, Zalesky and Breakspear2015; Zhou et al., Reference Zhou, Gennatas, Kramer, Miller and Seeley2012), that has been corroborated in many neuropsychiatric disorders (de Haan et al., Reference de Haan, Mott, van Straaten, Scheltens and Stam2012; Georgiadis et al., Reference Georgiadis, Larivière, Glahn, Hong, Kochunov, Mowry, Loughland, Pantelis, Henskens, Green, Cairns, Michie, Rasser, Catts, Tooney, Scott, Schall, Carr, Quidé and Kirschner2024; Hansen et al., Reference Hansen, Shafiei, Vogel, Smart, Bearden, Hoogman, Franke, van Rooij, Buitelaar, McDonald, Sisodiya, Schmaal, Veltman, van den Heuvel, Stein, van Erp, Ching, Andreassen, Hajek and Misic2022b; Larivière et al., Reference Larivière, Rodríguez-Cruces, Royer, Caligiuri, Gambardella, Concha, Keller, Cendes, Yasuda, Bonilha, Gleichgerrcht, Focke, Domin, von Podewills, Langner, Rummel, Wiest, Martin, Kotikalapudi and Bernhardt2020). Combined, examining the associations with the structural connectome may help understand the innate biological programs that shape the spatial topographies of CT alterations in MDD.
CT alterations in MDD are not uniform across the cortex; instead, certain regions appear to be more susceptible to disease pathology. This observation aligns with the regional susceptibility hypothesis, which posits that local features such as neurotransmitter profile, gene expression, and energy metabolism may play a critical role in the neuropathology of MDD. Continued advancements in nuclear imaging techniques and tracers have enabled reliable and accurate quantification of diverse neurotransmitter receptors and transporters across the whole brain (Hansen et al., Reference Hansen, Shafiei, Markello, Smart, Cox, Nørgaard, Beliveau, Wu, Gallezot, Aumont, Servaes, Scala, DuBois, Wainstein, Bezgin, Funck, Schmitz, Spreng, Galovic and Misic2022a), allowing researchers to investigate the neurochemical basis of healthy and diseased brains (Lan et al., Reference Lan, Liu, Zuo, Chen, Wang, Luo, Kuang, Kemp, Lui, Suo and Gong2025; Sun et al., Reference Sun, Liu, Liu, Guo, Ma, Zhu and Qian2024). In parallel, the current availability of modern brain-wide gene expression atlases, such as the Allen Human Brain Atlas (AHBA) (Hawrylycz et al., Reference Hawrylycz, Lein, Guillozet-Bongaarts, Shen, Ng, Miller, van de Lagemaat, Smith, Ebbert, Riley, Abajian, Beckmann, Bernard, Bertagnolli, Boe, Cartagena, Chakravarty, Chapin, Chong and Jones2012; Shen, Overly, & Jones, Reference Shen, Overly and Jones2012), offers investigators the ability to explore the genetic architecture of neuroimaging findings (Shen et al., Reference Shen, Zhang, Cui, Wang, Cai, Zhao, Zhu and Yu2022; Zhao et al., Reference Zhao, Cai, Mo, Lu, Yao, Yu and Zhu2023), including structural and functional brain abnormalities in MDD (Zhao et al., Reference Zhao, Zhu, Zhang, Zhang, Shen, Yu and Zhu2025; Zhu et al., Reference Zhu, Chen, Lu, Li, Wang, Cao, Chen, Chen, Chen, Chen, Cheng, Chu, Cui, Cui, Deng, Gong, Guo, He, Hu and Yu2024). It is generally accepted that mitochondria power brain activity and mitochondrial defects are linked to brain disorders (Buck, Maarman, Dube, & Bardien, Reference Buck, Maarman, Dube and Bardien2025; Clemente-Suárez et al., Reference Clemente-Suárez, Redondo-Flórez, Beltrán-Velasco, Ramos-Campo, Belinchón-deMiguel, Martinez-Guardado, Dalamitros, Yáñez-Sepúlveda, Martín-Rodríguez and Tornero-Aguilera2023; Jiang, Wang, & Sheng, Reference Jiang, Wang and Sheng2024). Recently, a comprehensive brain atlas of mitochondrial features has been developed, which provides a powerful tool for examining the molecular energetic landscape related to normal and abnormal neural processes (Mosharov et al., Reference Mosharov, Rosenberg, Monzel, Osto, Stiles, Rosoklija, Dwork, Bindra, Junker, Zhang, Fujita, Mariani, Bakalian, Sulzer, de Jager, Menon, Shirihai, Mann, Underwood and Picard2025). Collectively, investigating the relationships with these cross-modal brain atlases can facilitate improved understanding of the susceptibilities that lead to the spatial patterning of CT alterations across the cortex in MDD.
In the present study, we aimed to characterize the neurobiological factors related to the differential CT alteration patterns in youth-onset and adult-onset MDD. We first derived CT alterations in large samples of youth-onset and adult-onset MDD patients from the Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA) consortium. Next, by leveraging a normative structural connectome constructed based on diffusion magnetic resonance imaging (MRI) data from the Human Connectome Project (HCP), we interrogated whether the structural connectome constrains MDD-related CT alteration patterns, pursued by identifying likely epicentre regions and examining the hub vulnerability hypothesis. Finally, in combination with the neurotransmitter, transcriptome, and mitochondria atlases, we further explored the neurochemical basis, genetic architecture, and molecular energetic landscape associated with the CT alterations. A schematic overview of the study design and analysis pipeline is shown in Figure 1.
A schematic overview of the study design and analysis pipeline. (a) We first derived CT alterations in large samples of youth-onset and adult-onset MDD patients from the ENIGMA consortium. (b) By leveraging a normative structural connectome constructed based on diffusion MRI data from the HCP, we interrogated whether the structural connectome constrains MDD-related CT alteration patterns, pursued by identifying likely epicentre regions and examining the hub vulnerability hypothesis. (c) In combination with the neurotransmitter, transcriptome, and mitochondria atlases, we further explored the neurochemical basis, genetic architecture, and molecular energetic landscape related to the CT alterations. Abbreviations: CT, ‘cortical thickness’; ENIGMA, ‘enhancing neuroimaging genetics through meta-analysis’; HCP, ‘human connectome project’; MDD, ‘major depressive disorder’; MRI, ‘magnetic resonance imaging’.
Figure 1. Long description
Panel A, Cortical thickness alterations from E N I G M A cohort (patient-control). It displays brain maps for M D D-youth (age of onset less than or equal to 21 years) with N equals 472 patients and 2885 controls, and M D D-adult (age of onset greater than 21 years) with N equals 1214 patients and 3329 controls. A color scale for Cohen’s d ranges from negative 0.2 (blue) to 0.2 (red).
Panel B, Connectome constraint model and epicenter mapping. On the left, Empirical C T alteration brain maps and an H C P structural connectome matrix (Node 1, 2, to Node 68) feed into two pathways. The top pathway shows Node i within a network leading to an Estimated C T alteration brain map, supported by the equation d-hat sub i equals 1 over N sub i summation from j equals 1, j not equal to i to N sub i of A sub i j d sub j. This results in a scatter plot of Empirical C T alteration versus Estimated C T alteration showing a positive linear correlation. The bottom pathway illustrates epicenter mapping using the equation f(t) equals e to the power of negative a H t times f sub 0. It shows a time-series of brain networks (Step 1, 2, t) with a seed region, leading to a line graph of r versus t where the Epicenter curve reaches a Max r higher than Nonepicenter curves.
Panel C, Further analyses. Four icons represent different study areas from left to right. Hub susceptibility shows a brain with a lightning bolt and network nodes. Neurochemical basis shows a synaptic cleft with neurotransmitters. Genetic architecture shows two D N A double helices. Energetic landscape shows a cross-section of a mitochondrion.
Materials and methods
ENIGMA neuroimaging summary statistics
For our analyses, we used multi-site neuroimaging summary statistics published by the ENIGMA Consortium and publicly available within the ENIGMA Toolbox (https://github.com/MICA-MNI/ENIGMA) (Larivière et al., Reference Larivière, Paquola, Park, Royer, Wang, Benkarim, Vos de Wael, Valk, Thomopoulos, Kirschner, Lewis, Evans, Sisodiya, McDonald, Thompson and Bernhardt2021). The ENIGMA MDD Working Group currently includes 20 international groups with neuroimaging and clinical data from MDD patients and healthy controls. MDD patients were stratified into youth-onset (≤21 years) and adult-onset (>21 years) groups based on their age of onset (Schmaal et al., Reference Schmaal, Hibar, Sämann, Hall, Baune, Jahanshad, Cheung, van Erp, Bos, Ikram, Vernooij, Niessen, Tiemeier, Hofman, Wittfeld, Grabe, Janowitz, Bülow, Selonke, Völzke and Veltman2017). We focused on case-control differences in CT (472 youth-onset patients versus 2,885 controls and 1,214 adult-onset patients versus 3,329 controls) by performing cross-site random-effects meta-analyses of Cohen’s d values with adjustment for covariates including age, sex, and site. In the resultant Cohen’s d maps, positive/negative effects indicated increases/decreases in CT in patients relative to controls. Prior to the computation of summary statistics, structural MRI data were preprocessed, segmented, and parcellated based on the Desikan–Killiany atlas (Desikan et al., Reference Desikan, Ségonne, Fischl, Quinn, Dickerson, Blacker, Buckner, Dale, Maguire, Hyman, Albert and Killiany2006) in FreeSurfer (Dale, Fischl, & Sereno, Reference Dale, Fischl and Sereno1999) at each site and according to standard ENIGMA quality control protocols (http://enigma.ini.usc.edu/protocols/imaging-protocols). Individual cohort researchers obtained approval from local institutional ethics boards, and informed consent was obtained from study participants or their guardians.
Normative structural connectome
To assess the connectome constraints on MDD-related CT alterations and to identify likely epicentre regions, a normative structural connectome was constructed based on diffusion MRI data of 207 unrelated healthy adults (83 males, mean age ± standard deviation = 28.73 ± 3.73 years, age range = 22–36 years) from the HCP (Van Essen et al., Reference Van Essen, Ugurbil, Auerbach, Barch, Behrens, Bucholz, Chang, Chen, Corbetta, Curtiss, Penna, Feinberg, Glasser, Harel, Heath, Larson-Prior, Marcus, Michalareas, Moeller and Yacoub2012). Details on diffusion MRI data processing and structural connectome construction are provided in Supplementary Method. This procedure yielded a group-level 68 × 68 structural connectivity matrix based on the Desikan–Killiany atlas.
Connectome constraint model
A connectome constraint model was used to examine the connectome constraint effects on CT alterations (Cohen’s d maps) in youth-onset and adult-onset MDD, separately. To achieve this goal, we initially derived the estimated CT alteration of node i (
$ {\hat{\;d}}_i $
) based on its connected neighbour nodes using the connectome constraint model (Chopra et al., Reference Chopra, Segal, Oldham, Holmes, Sabaroedin, Orchard, Francey, O’Donoghue, Cropley, Nelson, Graham, Baldwin, Tiego, Yuen, Allott, Alvarez-Jimenez, Harrigan, Fulcher, Aquino and Fornito2023; Hansen et al., Reference Hansen, Shafiei, Vogel, Smart, Bearden, Hoogman, Franke, van Rooij, Buitelaar, McDonald, Sisodiya, Schmaal, Veltman, van den Heuvel, Stein, van Erp, Ching, Andreassen, Hajek and Misic2022b; Liang et al., Reference Liang, Sun, Liao, Lei, Xia, Duan, Zeng, Li, Xu, Men, Wang, Tan, Gao, Qin, Tao, Dong, Zhao and He2024; Shafiei et al., Reference Shafiei, Bazinet, Dadar, Manera, Collins, Dagher, Borroni, Sanchez-Valle, Moreno, Laforce, Graff, Synofzik, Galimberti, Rowe, Masellis, Tartaglia, Finger, Vandenberghe, de Mendonça and Ducharme2023) defined as:
$ {\hat{d}}_i=\frac{1}{N_i}{\sum}_{j=1,j\ne i}^{N_i}{d}_j{A}_{ij} $
, where Ni is the number of structurally connected neighbours of node i, dj is the CT alteration observed in the j-th neighbour of node i, and Aij defines the structural connectivity between nodes i and j. The group-level structural connectome was used to define the connected neighbours of each node. Then, we calculated the spatial correlation between the empirical and estimated CT alterations across 68 cortical nodes. We tested the spatial correlation against two null models to examine the significance. In the first null model, we used the spin permutation test to investigate whether the observed correlation was specific to the actual CT alteration pattern rather than due to the spatial autocorrelation (Alexander-Bloch et al., Reference Alexander-Bloch, Shou, Liu, Satterthwaite, Glahn, Shinohara, Vandekar and Raznahan2018). Specifically, we first recorded the spherical coordinates of centroids for each node in the Desikan–Killiany atlas. Then, we randomly rotated the nodes while maintaining spatial autocorrelation and reassigned node values to the nearest nodes. This procedure was repeated 1,000 times to create surrogate brain maps. The second rewired null model was used to evaluate whether the observed correlation is determined by the empirical structural connectome topology rather than the basic spatial embedding of the structural connectome (such as the distribution of node degree and edge length) (Betzel & Bassett, Reference Betzel and Bassett2018). Specifically, we first divided edges into different bins according to their Euclidean distance. To preserve the degree sequence and approximate edge length distribution of the empirical structural connectome, edge pairs were randomly swapped within each bin. Finally, 1,000 rewired connectomes were generated by repeating this procedure using a combination of 10 distance bins and 50,000 edge swaps (Chopra et al., Reference Chopra, Segal, Oldham, Holmes, Sabaroedin, Orchard, Francey, O’Donoghue, Cropley, Nelson, Graham, Baldwin, Tiego, Yuen, Allott, Alvarez-Jimenez, Harrigan, Fulcher, Aquino and Fornito2023). For the above two null models, P-values were calculated as the fractions of correlations in null models that exceeded the observed correlation.
Epicentre mapping
The connectome constraint model evaluates the degree to which the structural connectome shapes the spatial pattern of MDD-related CT alterations, but it offers limited insight into the dynamics of the spreading process, nor is it able to identify regions from which the spreading may initiate. Therefore, we used a network diffusion model to directly test whether certain brain nodes act as likely epicentre regions. The network diffusion model simulates the dynamic spread of pathology between the nodes of the structural connectome via a process of diffusion, defined as
$ f(t)={e}^{- aHt}{f}_0 $
, where t is the model diffusion time, which has arbitrary units, and f(t) is a vector characterizing the amount of diffusion in each node at time t. The strength of the diffusion process is controlled by a constant a, H is the Laplacian of the structural connectome matrix, and f0 represents the initial distribution of pathology. We repeatedly initialized the model using each of the 1,000 nodes as the starting seed, such that the initial state was set to 1 for the seed node and 0 for all other nodes. At each initialization, using a constant of a = 1, the network diffusion model was used to estimate the diffusion at all other nodes at time t = 0–50 (Chopra et al., Reference Chopra, Segal, Oldham, Holmes, Sabaroedin, Orchard, Francey, O’Donoghue, Cropley, Nelson, Graham, Baldwin, Tiego, Yuen, Allott, Alvarez-Jimenez, Harrigan, Fulcher, Aquino and Fornito2023; Raj, Kuceyeski, & Weiner, Reference Raj, Kuceyeski and Weiner2012). In this way, we were able to determine whether a diffusion process seeded from each node resulted in a spatial distribution of CT alterations that matched the empirically observed pattern. Model performance was evaluated as the Pearson correlation between the empirical CT alterations and the predicted diffusion at each time step, with the maximum correlation (r
max) across all time steps being retained for each node. The nodes with greater r
max values were considered as the epicentres. Statistical significance of each node’s r
max was evaluated using the spin permutation test, which generated 1,000 spatially constrained null maps of MDD-related CT alterations. For these null maps, we repeated the network diffusion model to yield a null distribution of 1,000 r
max values for each node. P-value was calculated as the fraction of r
max values in the null distribution that exceeded the observed r
max.
Relation to connectome degree centrality
To test the hub vulnerability hypothesis that hub regions are more vulnerable to disease and tend to show more pronounced structural abnormalities, we performed spatial correlations between MDD-related CT alteration patterns (Cohen’s d maps) and the weighted degree centrality derived from the normative structural connectome. Weighted degree centrality of a given region was defined by the sum of all its weighted structural connections. Brain regions with higher degree centrality are considered hub regions. Similarly, to evaluate statistical significance while eliminating spatial autocorrelation, the spin permutation test was performed.
Relation to neurotransmitters
To explore the neurochemical basis of MDD-related CT alterations, we examined their spatial correlations with 19 maps of neurotransmitter receptors and transporters (Hansen et al., Reference Hansen, Shafiei, Markello, Smart, Cox, Nørgaard, Beliveau, Wu, Gallezot, Aumont, Servaes, Scala, DuBois, Wainstein, Bezgin, Funck, Schmitz, Spreng, Galovic and Misic2022a) involving serotonin, acetylcholine, cannabinoid, dopamine, GABA, histamine, glutamate, opioid, and norepinephrine (Supplementary Table S1). Specifically, Pearson correlation coefficients between the Cohen’s d maps and these neurotransmitter maps were calculated across 68 cortical regions. Likewise, we used the spin permutation test to assess the statistical significance, and P-values were adjusted using the Bonferroni method.
Relation to transcriptome
We adopted neuroimaging-transcriptome association analysis to investigate the genetic architecture associated with MDD-related CT alterations, using brain-wide gene expression data provided by the AHBA dataset (http://www.brainmap.org) (Hawrylycz et al., Reference Hawrylycz, Miller, Menon, Feng, Dolbeare, Guillozet-Bongaarts, Jegga, Aronow, Lee, Bernard, Glasser, Dierker, Menche, Szafer, Collman, Grange, Berman, Mihalas, Yao and Lein2015). Microarray expression data were processed in abagen (Arnatkeviciute, Fulcher, & Fornito, Reference Arnatkeviciute, Fulcher and Fornito2019), including intensity-based filtering, normalization, and aggregation within Desikan–Killiany parcels and across donors. Moreover, genes whose similarity across donors fell below a threshold (r < 0.2) were removed, leaving a total of 12,668 genes for analysis (Larivière et al., Reference Larivière, Paquola, Park, Royer, Wang, Benkarim, Vos de Wael, Valk, Thomopoulos, Kirschner, Lewis, Evans, Sisodiya, McDonald, Thompson and Bernhardt2021). We tested the spatial association between gene expression and MDD-related CT alteration patterns using partial least squares (PLS) regression, a multivariate statistical technique that can predict a set of dependent variables from a set of independent variables (Krishnan, Williams, McIntosh, & Abdi, Reference Krishnan, Williams, McIntosh and Abdi2011). PLS regression is technically well suited to the high collinearity of gene expression data and has been extensively applied to transcriptome-neuroimaging association research (Hansen et al., Reference Hansen, Markello, Vogel, Seidlitz, Bzdok and Misic2021; Thomas et al., Reference Thomas, Zarkali, Ryten, Shmueli, Gil-Martinez, Leyland, McColgan, Acosta-Cabronero, Lees and Weil2021). In our analysis, the gene expression data were defined as the independent variables, and the Cohen’s d values of CT alterations were designated as the dependent variables. PLS regression identified the first PLS component where the weighted sum of gene expression (gene score) was most strongly correlated with the CT alteration maps. Pearson correlation coefficient between the gene score and the Cohen’s d values was calculated, and its statistical significance was assessed using the spin permutation test. Briefly, we created 1,000 surrogate CT alteration maps that maintained spatial autocorrelation and then used these surrogate maps to repeat the above PLS regression analysis to produce a null distribution. P-value was determined by calculating the frequency with which the true correlation coefficient exceeded those in the null distribution.
Gene contribution was estimated with gene loading, that is, the Pearson correlation coefficient between an individual gene’s expression level and the PLS-derived gene score across brain regions. Genes were then segregated according to the sign of their loadings, where positive genes refer to genes with positive loadings and negative genes refer to those with negative loadings. Strongly contributing genes were defined as those among the top 10% of the positive and negative genes. To better understand the biological significance of all strongly contributing genes, we conducted functional enrichment analyses with the ToppGene portal (https://toppgene.cchmc.org/) (Chen, Bardes, Aronow, & Jegga, Reference Chen, Bardes, Aronow and Jegga2009). Gene Ontology (GO) was used to determine the biological functions, including molecular functions (MFs), biological processes (BPs), and cellular components (CCs). For the enrichment analyses, Fisher’s exact tests were used to assess their statistical significance. Multiple testing was corrected using the false discovery rate (FDR) method with a corrected P-value of 0.05.
Relation to brain mitochondria
To understand the molecular energy landscape that is associated with MDD-related CT alterations, we examined their spatial correlations with a comprehensive brain atlas of mitochondrial features, including oxidative phosphorylation (OXPHOS) enzyme activities (CI, CII, and CIV), mitochondrial density (MitoD), mitochondrial respiratory capacity (MRC), and tissue respiratory capacity (TRC). Statistical significance of the spatial correlations was determined by the spin permutation test.
Results
Structural connectome constraints
Whole-brain CT alteration patterns (Cohen’s d maps) in youth-onset and adult-onset MDD from the ENIGMA cohorts are shown in Figure 1a. To determine whether structural connectome constrains the CT alteration patterns, we constructed a normative structural connectome based on diffusion-weighted MRI data from an independent cohort of 207 healthy adults collected from the HCP dataset. On the basis of the normative structural connectome, we obtained the estimated CT alteration of each node according to its connected neighbour nodes using the connectome constraint model. Significant spatial correlations were found between the empirical and estimated CT alterations across 68 cortical nodes for youth-onset MDD (r = 0.24, P spin = 0.006, P rewire = 0.001) and adult-onset MDD (r = 0.49, P spin = 0.004, P rewire = 0.005) (Figure 2a). This result suggests structural connectome constraints on CT alteration patterns in both youth-onset and adult-onset MDD. Moreover, we transformed the spatial correlation coefficients into Fisher’s z values for comparisons, which revealed significant differences between the groups (z = −1.66, P = 0.0485), indicating a greater connectome constraint effect in adult-onset MDD than in youth-onset MDD.
Structural connectome constraints, epicentre mapping, and hub vulnerability. (a) Scatterplots of the spatial correlations between the empirical and estimated CT alterations across 68 cortical nodes. Significant correlations suggest structural connectome constraints on MDD-related CT alteration patterns. (b) Brain maps showing the regional epicentres whose structural connectivity profiles significantly correlated with MDD-related CT alteration patterns. (c) Scatterplots of the spatial correlations between MDD-related CT alteration patterns and nodal degree centrality of the structural connectome. Significant correlations indicate that hub regions with higher degree centrality tend to present more pronounced structural abnormalities. Abbreviations: CT, ‘cortical thickness’; MDD, ‘major depressive disorder’; L, ‘left’; R, ‘right’.
Figure 2. Long description
The figure is divided into three horizontal panels labeled A, B, and C.
Panel A, Connectome constraint model, contains two scatterplots. The x-axis is Estimated C T alteration and the y-axis is Empirical C T alteration.
* The left plot (Youth, purple) shows a positive linear trend with r = 0.24, P sub spin = 0.006, and P sub rewire = 0.001.
* The right plot (Adult, orange) shows a stronger positive linear trend with r = 0.49, P sub spin = 0.004, and P sub rewire = 0.005.
Panel B, Epicenter mapping, displays eight cortical surface maps.
* On the left (L), four purple-shaded brain views show epicenters in the bilateral lateral prefrontal cortex. A scale bar for Max r ranges from 0 to 0.5.
* On the right (R), four red-shaded brain views show epicenters concentrated in the bilateral orbitofrontal cortex, anterior and middle cingulate gyrus. A scale bar for Max r ranges from 0 to 0.5.
Panel C, Hub susceptibility, contains two scatterplots. The x-axis is Degree centrality and the y-axis is C T difference.
* The left plot (Youth, purple) shows a negative correlation with r = minus 0.21 and P sub spin = 0.048.
* The right plot (Adult, orange) shows a weak negative correlation with r = minus 0.09 and P sub spin = 0.28.
Epicentre regions
We used a network diffusion model to determine whether certain brain nodes act as likely epicentre regions. Results showed that the epicentres of youth-onset MDD were mainly located in the bilateral lateral prefrontal cortex, while those of adult-onset MDD were primarily in the bilateral orbitofrontal cortex, anterior and middle cingulate gyrus (P spin < 0.05) (Figure 2b).
Hub vulnerability
To test the hub vulnerability hypothesis, spatial correlations between MDD-related CT alteration patterns and nodal degree centrality of the structural connectome were conducted. We found a significant negative association of degree centrality with CT alterations in youth-onset MDD (r = −0.21, P spin = 0.048) (Figure 2c), indicating that hub regions with higher degree centrality tend to present more pronounced structural abnormalities. However, no significant association was observed for adult-onset MDD (r = −0.09, P spin = 0.28).
Neurochemical basis
We performed spatial correlations between 19 neurotransmitter maps and MDD-related CT alteration patterns to explore their neurochemical basis. The significant associations are shown in Figure 3 and Supplementary File S1 (P < 0.05, Bonferroni corrected). Specifically, CT alterations in youth-onset MDD showed a significant positive correlation with 5-HTT (r = 0.418, P spin = 0.010), as well as negative correlations with CB1 (r = −0.337, P spin = 0.029) and 5-HT4 (r = −0.349, P spin = 0.048). CT alterations in adult-onset MDD showed significant negative correlations with CB1 (r = −0.515, P spin = 0.048), D1 (r = −0.512, P spin = 0.038), H3 (r = −0.458, P spin = 0.019), MOR (r = −0.709, P spin < 0.001), and 5-HT4 (r = −0.485, P spin = 0.038).
Neurochemical basis. Spatial Pearson correlations between 19 neurotransmitter maps and MDD-related CT alteration patterns were examined to explore their neurochemical basis. The resultant Pearson correlation coefficients were assessed against a spatial autocorrelation preserving null model (spin permutation test). Points represent the empirical Pearson correlation coefficients, with the red ones indicating P spin < 0.05 after Bonferroni correction and the blue ones indicating P spin > 0.05 after Bonferroni correction. In the boxplots of the null distributions derived from 1,000 permutations, the ends of the boxes represent the first (25%) and third (75%) quartiles, the center line represents the median, the whiskers represent the non-outlier end-points, and the small gray points represent outliers. Abbreviations: 5HT, ‘5-hydroxytryptamine’; 5HTT, ‘5-hydroxytryptamine transporter’; CB, ‘cannabinoid type’; CT, ‘cortical thickness’; D, ‘dopamine’; DAT, ‘dopamine transporter’; GABAa, ‘gamma-aminobutyric acid a’; H, ‘histamine’; M, ‘muscarinic’; MDD, ‘major depressive disorder’; mGluR, ‘metabotropic glutamate type’; MOR, ‘mu opioid receptor’; NET, ‘norepinephrine transporter’; NMDA, ‘N-methyl-D-aspartate receptor’; VAChT, ‘vesicular acetylcholine transporter’.
Figure 3. Long description
A multi-panel plot with two horizontal rows of data. Each row features brain heatmaps at the top, followed by a graph where the y-axis is Pearson’s r ranging from negative 0.8 to 0.6, and the x-axis lists neurotransmitter maps.
Top Panel (Left to Right):
* a 4 beta 2: Two blue dots near negative 0.2.
* C B 1: One red dot at negative 0.35 and one red dot at negative 0.5.
* D 1: One blue dot at negative 0.1 and one red dot at negative 0.5.
* D 2: One blue dot at negative 0.15 and one blue dot at negative 0.45.
* D A T: One blue dot at 0.1 and one blue dot at negative 0.45.
* G A B A a: One blue dot at 0.1 and one blue dot at 0.05.
* H 3: One blue dot at negative 0.05 and one red dot at negative 0.45.
* M 1: One blue dot at negative 0.15 and one blue dot at 0.1.
* m Glu R 5: One blue dot at negative 0.1 and one blue dot at negative 0.3.
* M O R: One blue dot at negative 0.3 and one red dot at negative 0.7.
Bottom Panel (Left to Right):
* N E T: One blue dot at 0.0 and one blue dot at 0.15.
* N M D A: One blue dot at 0.05 and one blue dot at negative 0.15.
* 5-H T 1 a: One blue dot at negative 0.1 and one blue dot at negative 0.45.
* 5-H T 1 b: One blue dot at 0.05 and one blue dot at 0.2.
* 5-H T 2 a: One blue dot at negative 0.35 and one blue dot at negative 0.1.
* 5-H T 6: One blue dot at negative 0.1 and one blue dot at 0.0.
* 5-H T T: One red dot at 0.4 and one blue dot at negative 0.05.
* 5-H T 4: One red dot at negative 0.35 and one red dot at negative 0.5.
* V A Ch T: One blue dot at 0.05 and one blue dot at negative 0.45.
Legend:
* Red dots: Empirical P Bonferroni less than 0.05.
* Blue dots: Empirical P Bonferroni greater than 0.05.
* Gray boxes: Spatial null (Youth).
* Peach boxes: Spatial null (Adult).
Genetic architecture
With use of the AHBA transcriptomic dataset, we investigated the associations between gene expression and MDD-related CT alteration patterns to elucidate their genetic architecture. PLS regression identified the first PLS component where the weighted sum of gene expression (gene score) was most strongly correlated with CT alterations in adult-onset MDD (r = 0.634, P spin < 0.001) (Figure 4a), which was significantly greater than expected by chance (Figure 4b). All the 12,668 genes were ranked by their contributions to the correlation (i.e., gene loadings) (Figure 4c). A total of 1267 genes, including PLS+ genes (the top 10% of genes with positive loadings) and PLS- genes (the top 10% of genes with negative loadings), were defined as strongly contributing genes. To better understand the biological significance of the strongly contributing genes, we conducted functional enrichment analyses. The gene list and the results of functional enrichment are listed in Supplementary File S2 and are depicted in Figure 4d (P < 0.05, FDR-corrected). Specifically, CT alterations in adult-onset MDD were linked to gene categories primarily involving synapse, neuron, synaptic transmission, and ion channel activity. However, no significant associations were found between gene expression and CT alterations in youth-onset MDD.
Genetic architecture. (a) Scatterplot of the overall correlation between gene expression and CT alteration pattern in adult-onset MDD identified by PLS regression. (b) Histogram of the permutation distribution showing that the overall correlation (dotted line) was significantly greater than expected by chance. (c) Illustrative example of the loadings (gene contributions) assigned to representative genes. Genes with the highest positive loadings are colored in pink and those with the lowest negative loadings are colored in blue. (d) Functional enrichment of the strongly contributing genes (the top 10% of genes with positive and negative loadings). For the bubble chart, the x-axis denotes the rich factor and the y-axis denotes items from the GO. The rich factor refers to the ratio of the number of strongly contributing genes annotated to the item to the number of all genes annotated to the item. The bubble size represents the number of genes overlapping with those belonging to each item, and the bubble color represents the -Log10(P) with the P-value corrected by the FDR method. Abbreviations: BP, ‘biological process’; CC, ‘cellular component’; CT, ‘cortical thickness’; FDR, ‘false discovery rate’; GO, ‘gene ontology’; MDD, ‘major depressive disorder’; PLS, ‘partial least squares’.
Figure 4. Long description
Panel A is a scatterplot with Gene score on the x-axis ranging from minus 1000 to 500 and C T difference on the y-axis from minus 0.05 to 0.05. Orange data points show a positive linear correlation with r equals 0.634 and P sub spin less than 0.001.
Panel B is a histogram of relative frequency percent against r values from 0.0 to 0.7. A blue bell-shaped curve fits the distribution of light blue bars centered around r equals 0.2. A vertical dashed purple line at r equals 0.634 indicates the observed correlation is significantly higher than the permutation distribution.
Panel C is a table titled Gene name and Gene loading. The top section in pink lists genes with high positive loadings: E P B 4 1 (0.9567), N C O A 3 (0.9563), E N T P D 4 (0.9562), A T P 2 B 2 (0.9523), S L C 3 8 A 2 (0.9469), T T C 2 1 B (0.9456), I F N L R 1 (0.9449), and S Y C P 2 (0.9434). The bottom section in blue lists genes with low negative loadings: W D R 8 6 (minus 0.9578), P E A 1 5 (minus 0.9586), P K I A (minus 0.9587), D K F Z p 7 7 9 M 0 6 5 2 (minus 0.9587), D Y D C 2 (minus 0.9590), N D U F A F 8 (minus 0.9591), T I M M 8 B (minus 0.9611), and I F T 2 2 (minus 0.9644).
Panel D is a bubble chart for G O enrichment. The x-axis is Rich factor from 0.08 to 0.20 and the y-axis lists B P and C C terms. Bubble size represents Gene number (40, 80, 120) and color represents minus Log 10 (P) from 2 to 4. Key terms include B P Synaptic signaling and C C Synapse, which show high gene numbers and significance. C C Organelle membrane contact site shows the highest rich factor near 0.20.
Molecular energetic landscape
Spatial correlations between mitochondrial features (CI, CII, CIV, MitoD, MRC, and TRC) and MDD-related CT alteration patterns were examined to understand the molecular energy landscape related to the CT alterations. We found that CT alterations in adult-onset MDD were positively correlated with CI (r = 0.286, P spin = 0.027) and CII (r = 0.397, P spin = 0.004) (Figure 5). However, no significant spatial correlations were observed between mitochondrial features and CT alterations in youth-onset MDD.
Molecular energetic landscape. Spatial correlations between mitochondrial features and MDD-related CT alteration patterns were examined to understand the molecular energy landscape related to the CT alterations. The resultant Pearson correlation coefficients were assessed against a spatial autocorrelation preserving null model (spin permutation test). Points represent the empirical Pearson correlation coefficients, with the red ones indicating P spin < 0.05 after Bonferroni correction and the blue ones indicating P spin > 0.05 after Bonferroni correction. In the boxplots of the null distributions derived from 1,000 permutations, the ends of the boxes represent the first (25%) and third (75%) quartiles, the center line represents the median, the whiskers represent the non-outlier end-points, and the small gray points represent outliers. Abbreviations: CI, NADH-ubiquinone oxidoreductase; CII; succinate dehydrogenase; CIV cytochrome c oxidase; CT, ‘cortical thickness’; MDD, ‘major depressive disorder’; MitoD, ‘mitochondria tissue density’; MRC, ‘mitochondrial respiratory capacity’; TRC, ‘tissue respiratory capacity’.
Figure 5. Long description
The y-axis represents Pearson’s r, ranging from minus 0.6 to 0.6. The x-axis lists six categories: C I, C I I, C I V, Mito D, M R C, and T R C. Below each label is a small brain map showing cortical distribution patterns.
A legend at the top identifies four data types: red dots for Empirical P less than 0.05, blue dots for Empirical P greater than 0.05, gray boxes for Spatial null Youth, and light orange boxes for Spatial null Adult.
For each category, two boxplots are shown side-by-side with vertical whiskers and outlier dots.
* C I: The first empirical dot is blue near 0.0; the second is red at approximately 0.3.
* C I I: The first dot is blue near 0.1; the second is red at approximately 0.4.
* C I V: Both dots are blue, one near minus 0.15 and one near 0.1.
* Mito D: Both dots are blue, one near 0.05 and one near minus 0.1.
* M R C: Both dots are blue, one near minus 0.05 and one near 0.2.
* T R C: Both dots are blue, one near minus 0.1 and one near 0.15.
In all categories, the empirical red dots for C I and C I I sit significantly above the whiskers of the null distribution boxplots, while blue dots generally fall within or near the range of the null distributions.
Discussion
Based on large-scale ENIGMA neuroimaging summary statistics, the current study sought to characterize the neurobiological factors related to the differential CT alteration patterns in youth-onset and adult-onset MDD. First, our data demonstrated that the diffusion MRI-derived structural connectome constrained CT alteration patterns in both youth-onset and adult-onset MDD, but the constraint effect was greater in adult-onset MDD than in youth-onset MDD. Subsequent analysis revealed that the epicentres were largely divergent for youth-onset (lateral prefrontal cortex) and adult-onset (orbitofrontal cortex, anterior and middle cingulate gyrus) MDD. The observation of CT alterations in youth-onset MDD as a function of connectome degree centrality confirmed the hub vulnerability hypothesis in this condition only. Moreover, we found common and distinct neurotransmitter systems linked to CT alterations in youth-onset and adult-onset MDD, indicating shared and unique neurochemical bases related to these conditions. Transcriptomic contextualization demonstrated that CT alterations in adult-onset MDD were related to gene categories primarily involving synapse, neuron, synaptic transmission and ion channel activity, yet those in youth-onset MDD were not related to gene expression. In addition, associations between mitochondrial features and CT alterations were found in adult-onset MDD but not in its youth-onset counterpart.
In accordance with previous studies reporting the constraint effect of the brain connectome on cortical morphological alterations in neurodegenerative and psychiatric disorders (Chopra et al., Reference Chopra, Segal, Oldham, Holmes, Sabaroedin, Orchard, Francey, O’Donoghue, Cropley, Nelson, Graham, Baldwin, Tiego, Yuen, Allott, Alvarez-Jimenez, Harrigan, Fulcher, Aquino and Fornito2023; Georgiadis et al., Reference Georgiadis, Larivière, Glahn, Hong, Kochunov, Mowry, Loughland, Pantelis, Henskens, Green, Cairns, Michie, Rasser, Catts, Tooney, Scott, Schall, Carr, Quidé and Kirschner2024; Shafiei et al., Reference Shafiei, Bazinet, Dadar, Manera, Collins, Dagher, Borroni, Sanchez-Valle, Moreno, Laforce, Graff, Synofzik, Galimberti, Rowe, Masellis, Tartaglia, Finger, Vandenberghe, de Mendonça and Ducharme2023; Yau et al., Reference Yau, Zeighami, Baker, Larcher, Vainik, Dadar, Fonov, Hagmann, Griffa, Mišić, Collins and Dagher2018; Zeighami et al., Reference Zeighami, Ulla, Iturria-Medina, Dadar, Zhang, Larcher, Fonov, Evans, Collins and Dagher2015), our findings confirm a connectome-constrained spatial patterning of CT alterations. Although the exact mechanism remains unclear, two possible explanations have been proposed. First, there are shared characteristics of connected regions: cortical regions connected via axonal projections tend to share similar cytoarchitectural features (Arnatkeviciute et al., Reference Arnatkeviciute, Fulcher, Oldham, Tiego, Paquola, Gerring, Aquino, Hawi, Johnson, Ball, Klein, Deco, Franke, Bellgrove and Fornito2021; Beul, Barbas, & Hilgetag, Reference Beul, Barbas and Hilgetag2017), exhibit similar neurotransmitter receptor distributions (Froudist-Walsh et al., Reference Froudist-Walsh, Xu, Niu, Rapan, Zhao, Margulies, Zilles, Wang and Palomero-Gallagher2023; Hansen et al., Reference Hansen, Shafiei, Markello, Smart, Cox, Nørgaard, Beliveau, Wu, Gallezot, Aumont, Servaes, Scala, DuBois, Wainstein, Bezgin, Funck, Schmitz, Spreng, Galovic and Misic2022a), and exhibit coordinated spontaneous activity (van den Heuvel, Mandl, Kahn, & Hulshoff Pol, Reference van den Heuvel, Mandl, Kahn and Hulshoff Pol2009). This similarity and coordination may render connected regions jointly susceptible to disease-related changes. Second, axonal transport acts as a diffusion pathway: the structural connectome is capable of transporting nutrients, oxygen, and neurotransmitters (Di Maio, Reference Di Maio2021); in neurodegenerative diseases, pathological proteins have been shown to propagate along axons (Raj et al., Reference Raj, Kuceyeski and Weiner2012; Shafiei et al., Reference Shafiei, Bazinet, Dadar, Manera, Collins, Dagher, Borroni, Sanchez-Valle, Moreno, Laforce, Graff, Synofzik, Galimberti, Rowe, Masellis, Tartaglia, Finger, Vandenberghe, de Mendonça and Ducharme2023; Suo et al., Reference Suo, Chen, Chen, Luo, Kemp, Lui and Sun2025), and even in psychiatric disorders without visible protein aggregation, subtle alterations in protein homeostasis may likewise propagate via axonal pathways (Bradshaw & Korth, Reference Bradshaw and Korth2019).
Further analysis revealed that the potential epicentres of CT alterations in youth-onset MDD were predominantly located in the bilateral lateral prefrontal cortex, while those in adult-onset MDD were in the bilateral orbitofrontal cortex, anterior and middle cingulate gyrus. Our results suggest these regions as likely epicentre regions of CT changes in MDD patients, subsequently affecting structurally connected brain regions and driving the progression of this disease. These findings are consistent with previous studies indicating that the prefrontal cortex may constitute a “hub” in MDD and its anomaly might be linked to disease severity (Zhang et al., Reference Zhang, Peng, Sweeney, Jia and Gong2018). There is also strong evidence supporting the importance of the orbitofrontal cortex in emotional regulation, reward processing, and decision-making, and its functional or structural abnormalities are assumed to be associated with the onset of MDD (Rolls, Reference Rolls2019; Rolls, Cheng, & Feng, Reference Rolls, Cheng and Feng2020; Zhang et al., Reference Zhang, Rolls, Wang, Xie, Cheng and Feng2024). In addition, prior research has shown that changes in white matter connectivity of the anterior cingulate gyrus may impair its role in emotional regulation and increase the risk of developing depression (Bae et al., Reference Bae, MacFall, Krishnan, Payne, Steffens and Taylor2006). Activation of the middle cingulate gyrus indicates symptom remission following pharmacological treatment in MDD patients, implying that this region represents a functionally active brain area in MDD (Karim et al., Reference Karim, Wang, Andreescu, Tudorascu, Butters, Karp, Reynolds and Aizenstein2018). Apart from a deeper understanding of the neuropathological underpinnings of MDD, our results may have potential clinical implications: the distinct epicentre regions suggest that age-of-onset-dependent targeting strategies may be warranted for early intervention in MDD patients, potentially guiding the development of individualized neuromodulation therapies (e.g., TMS).
Our data revealed CT alterations in youth-onset MDD as a function of connectome degree centrality, in favour of the hub vulnerability hypothesis in this condition. Several potential mechanisms might account for the great vulnerability of densely connected hub regions. Based on the stress-threshold model, centrally located neuronal populations tend to be particularly susceptible to pathophysiological perturbations due to their higher neuronal activity and concomitant metabolic demands (Crossley et al., Reference Crossley, Mechelli, Scott, Carletti, Fox, McGuire and Bullmore2014; Saxena & Caroni, Reference Saxena and Caroni2011). Additionally, hub regions are more likely to be part of the transmodal associative cortices that have a prolonged window of developmental plasticity, such that early genetic or environmental neural insults may derail or interact with normal brain development in hub regions, leading to aberrant maturation of later-developing systems (Sydnor et al., Reference Sydnor, Larsen, Bassett, Alexander-Bloch, Fair, Liston, Mackey, Milham, Pines, Roalf, Seidlitz, Xu, Raznahan and Satterthwaite2021). The lack of an association between connectome degree centrality and CT alterations in adult-onset MDD may partly support the neurodevelopmental origin.
Spatial correlation analyses demonstrated common and distinct neurotransmitter systems linked to CT alterations in youth-onset and adult-onset MDD, indicating shared and unique neurochemical bases associated with these conditions. In youth-onset MDD, CT alterations were correlated with CB1, 5-HTT, and 5-HT4, while those in adult-onset depression were correlated with CB1, D1, H3, MOR, and 5-HT4. Increased expression of CB1 receptors in astrocytes protects blood-brain barrier integrity, thereby reducing anxiety and depression-like behaviours (Dudek et al., Reference Dudek, Paton, Binder, Collignon, Dion-Albert, Cadoret, Lebel, Lavoie, Bouchard, Kaufmann, Clavet-Fournier, Manca, Guzmán, Campbell, Turecki, Mechawar, Flamand, Lavoie-Cardinal, Silvestri and Menard2025; Salort, Hernández-Hernández, García-Fuster, & García-Sevilla, Reference Salort, Hernández-Hernández, García-Fuster and García-Sevilla2020). 5-HT4 receptors play an essential role in regulating mood, anxiety, and cognition, and 5-HT4 receptor agonists have been shown to have rapid antidepressant effects (Karayol et al., Reference Karayol, Medrihan, Warner-Schmidt, Fait, Rao, Holzner, Greengard, Heintz and Schmidt2021). It is also evident that the interaction between the frontal and limbic lobes in MDD is regulated by 5-HTT, which may be related to emotional information processing (Friedel et al., Reference Friedel, Schlagenhauf, Sterzer, Park, Bermpohl, Ströhle, Stoy, Puls, Hägele, Wrase, Büchel and Heinz2009). 5-HTT gene polymorphisms (e.g., 5-HTTLPR) make individuals susceptible to environmental adversity, especially when exposed to stress in early life (Ignácio, Réus, Abelaira, & Quevedo, Reference Ignácio, Réus, Abelaira and Quevedo2014). The binding potential of dopamine D1 receptors in the brain is significantly reduced in MDD patients, exhibiting a negative correlation with disease severity (Cannon et al., Reference Cannon, Klaver, Peck, Rallis-Voak, Erickson and Drevets2009). Histamine and its receptors can improve depressive symptoms by regulating immune responses, and H3 receptors can also interact with other neurotransmitters (e.g., 5-HT and DA) (Qian, Shu, Xiao, & Wang, Reference Qian, Shu, Xiao and Wang2022). This highlights the complex role of histamine in depression and indicates that histamine receptors may serve as new targets for antidepressant treatment. There is evidence that MDD patients show altered activation of the MOR system during social rejection and acceptance, and altered endogenous opioid activity in MDD may impede emotional recovery from negative social interactions and decrease pleasure derived from positive interactions (Hsu et al., Reference Hsu, Sanford, Meyers, Love, Hazlett, Walker, Mickey, Koeppe, Langenecker and Zubieta2015).
Transcriptomic contextualization demonstrated differential genetic influences on CT alterations in youth-onset and adult-onset MDD, characterized by the presence of a significant correlation with gene expression in adult-onset MDD, but not in youth-onset MDD. Childhood adversity (e.g., abuse or neglect) is a central risk factor for youth-onset MDD and can lead to persistent epigenetic modifications (e.g., glucocorticoid receptor methylation) and dysregulation of stress systems (e.g., HPA axis) (Green et al., Reference Green, Stolicyn, Harris, Shen, Romaniuk, Barbu, Hawkins, Wardlaw, Steele, Waiter, Sandu, Campbell, Porteous, Seckl, Lawrie, Reynolds, Cavanagh, McIntosh and Whalley2021; Tozzi et al., Reference Tozzi, Farrell, Booij, Doolin, Nemoda, Szyf, Pomares, Chiarella, O’Keane and Frodl2018). Such youth-onset stress may lead to fixed structural abnormalities rather than progressive changes by affecting brain development during critical periods of synaptogenesis (Malave, van Dijk, & Anacker, Reference Malave, van Dijk and Anacker2022). Adult-onset MDD is driven more by cumulative damage (e.g., neuroinflammation or oxidative stress). For instance, DNA methylation of the NLRP3 gene may potentiate the gene expression effect through increasing the related neuroinflammatory process (Han et al., Reference Han, Choi, Kim, Kang, Kang, Tae, Han and Ham2022). These previous findings suggest different etiological and pathophysiological mechanisms for youth-onset and adult-onset MDD. Our data showed that CT alterations in adult-onset MDD were spatially correlated with gene expression of GO categories mainly involving synapse, neuron, synaptic transmission, and ion channel activity. With respect to signal transduction, synapse is the site of information transfer from the presynaptic neuron to the postsynaptic neuron, a process mediated primarily by neurotransmitters released from presynaptic axon terminals, which subsequently bind to receptors on postsynaptic dendritic spines (Raven, Van der Zee, Meerlo, & Havekes, Reference Raven, Van der Zee, Meerlo and Havekes2018). Many interconnected molecular pathways underlie depression, ultimately leading to disturbed synaptic activity through influencing a variety of mechanisms such as neurotransmitter production and release, membrane excitability, and dendritic spine elimination (Fries, Saldana, Finnstein, & Rein, Reference Fries, Saldana, Finnstein and Rein2023). Synaptic transmission is considered to play a crucial role in the information exchange between neurons, which depends on neurotransmitter systems (Hyman, Reference Hyman2005). Ion channels are important for generating membrane potential, neuronal growth and differentiation, signal transduction, and neurotransmitter release (Kumar et al., Reference Kumar, Kumar, Jha, Jha and Ambasta2016; Moody & Bosma, Reference Moody and Bosma2005; Smith & Walsh, Reference Smith and Walsh2020), which may hold promise as therapeutic targets for MDD (Hashimoto, Reference Hashimoto2019). Notably, our transcriptomic contextualization was performed across bilateral cortical regions. Given that most AHBA data are limited to the left hemisphere, we cannot entirely rule out possible interhemispheric differences in gene expression measurement. Future studies using independent transcriptomic datasets with bilateral sampling are warranted to validate our findings.
MDD patients typically exhibit mitochondrial energy metabolism dysfunction and cerebral energy imbalance, with reduced ATP levels in the brain (Martins-de-Souza et al., Reference Martins-de-Souza, Guest, Harris, Vanattou-Saifoudine, Webster, Rahmoune and Bahn2012). Research shows that all stages of aerobic glucose oxidation are inhibited to varying degrees in MDD, and the inhibition of mitochondrial oxidative phosphorylation is most closely related to the onset of depression (Arzola, Xiong, & Mei, Reference Arzola, Xiong and Mei2022). The normal expression of mitochondrial respiratory chain complexes I, II, III, and ATP synthase is the basis for achieving mitochondrial oxidative phosphorylation. Ben-Shachar et al. found that the expression of subunits of respiratory chain complex I in the cerebellum of MDD patients was significantly reduced (Ben-Shachar & Karry, Reference Ben-Shachar and Karry2008). The positive correlation between CT and mitochondrial CI/CII activity in adult-onset MDD can be understood through two complementary explanatory frameworks: the mitochondrial reserve hypothesis and the compensatory upregulation mechanism. High mitochondrial reserve provides a ‘safety cushion’ against ageing-related metabolic stress (Swerdlow, Burns, & Khan, Reference Swerdlow, Burns and Khan2014), whilst compensatory upregulation reflects the adaptive response of surviving neurons to local stress (Du et al., Reference Du, Wang, Hunter, Wei, Blumenthal, Falke, Khairova, Zhou, Yuan, Machado-Vieira, McEwen, Manji and Manji2009). These two mechanisms may act in concert, enabling cortical regions with stronger mitochondrial function to better maintain structural integrity. These studies indicate that impaired oxidative phosphorylation caused by reduced mitochondrial respiratory chain activity is involved in the pathophysiology of depression, which may provide a new perspective for understanding the pathogenesis of MDD to some extent.
Our study has a few limitations. First, we obtained neuroimaging summary statistics from the ENIGMA cohorts, such that we were unable to determine whether the use of controls shared among studies influenced our results. Second, the CT alterations in youth-onset and adult-onset MDD are only available for cortical regions from the coarse Desikan–Killiany parcellation. Further analysis based on a more fine-grained parcellation with higher resolution and subcortical structures is warranted to validate our findings. Third, since the ENIGMA aggregate statistics do not provide data at the individual level, we could not test alternative age-of-onset cutoffs (e.g., 18 or 25 years) and control for potential clinical confounders such as medication, illness duration, and comorbidity. Future studies should utilize individual-level neuroimaging data to further validate the robustness of our findings. Fourth, the data on disease-related CT alterations, structural connectome, neurotransmitters, transcriptome, and mitochondrial features were derived from different samples. Such heterogeneity may introduce potential biases and should be considered when interpreting our results. Future studies using sample-matched datasets are needed to validate our findings. Finally, the absence of transcriptomic and mitochondrial associations in youth-onset MDD likely reflects two factors. On the one hand, compared with adult-onset MDD, the effect sizes of CT alterations in youth-onset MDD are smaller; thus, the lack of transcriptomic and mitochondrial associations is to some extent a predictable null result. On the other hand, for the ENIGMA datasets, the sample sizes of the youth-onset cohorts are smaller than the adult-onset cohorts, which could lead to reduced accuracy of the CT alteration maps for youth-onset MDD and may ultimately result in biased findings of the spatial correlation analyses.
In summary, using a combination of ENIGMA neuroimaging summary statistics and HCP diffusion MRI data, we interrogated the structural connectome architecture shaping the divergent CT alteration patterns in youth-onset and adult-onset MDD. In conjunction with the publicly available neurotransmitter, transcriptome, and mitochondria atlases, we furthermore explored the neurochemical basis, genetic architecture, and molecular energetic landscape underlying such divergence. These findings may not only enrich our understanding of the neurobiological factors related to the differential cortical abnormalities between youth-onset and adult-onset MDD, but also have clinical implications for inspiring new therapeutic options. More broadly, our findings suggest that the cross-modal analytical framework is useful in the domain of psychiatric neuroimaging and has the potential to be applied to other mental disorders.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726105777.
Data availability statement
All data analyzed in this paper were obtained from open-access sources. MDD-related Cohen’s d maps derived from the ENIGMA meta-analyses were accessed via the ENIGMA Toolbox (https://enigma-toolbox.readthedocs.io/en/latest/). Through the toolbox, we also accessed normative structural connectome data from a Human Connectome Project young adult sample (HCP; http://www.humanconnectome.org/) and gene expression data from the Allen Human Brain Atlas (https://human.brain-map.org/) as accessible through abagen.
Acknowledgements
The study was supported by the National Natural Science Foundation of China (grant numbers: 82471952 and 82371928), the Anhui Provincial Natural Science Foundation (grant number: 2308085MH277), and the Scientific Research Foundation of Anhui Medical University (grant number: 2022xkj143).
Competing interests
The authors declare no conflict of interests.
Open access
