Skip to main content
NCBI home page
Psychoradiology logoLink to Psychoradiology
. 2026 Jan 24;6:kkag005. doi: 10.1093/psyrad/kkag005

Brain-predicted age difference is associated with the progression of subthreshold depression: evidence from the UK Biobank

Haowei Dai 1, Lijing Niu 2, Qingzi Zhu 3, Yuanyuan Zeng 4, Yutong Ying 5, Xueping Yin 6, Xiangyi Liang 7, Xingqin Wang 8, Bihua Zhou 9, Qing Ma 10,✉, Ruibin Zhang 11,12,13,✉
PMCID: PMC12947158  PMID: 41767426

Abstract

Background

Previous studies have reported accelerated brain aging in individuals with major depressive disorder (MDD) compared to healthy controls. However, these findings are based primarily on cross-sectional data, limiting dynamic association between brain aging and MDD. Here, we examined the relationship between brain aging and MDD progression by focusing on subthreshold depression, a prodromal stage of MDD, and aimed to determine whether quantitative markers of brain aging exhibit a stable association with disease progression.

Methods

Using neuroimaging data from the UK Biobank and a support vector regression (SVR) model, we predicted brain age in individuals who exhibited subthreshold depressive symptoms at baseline but showed divergent mental status at follow-up, and then conducted between-group comparisons. Logistic regression was then applied to assess whether brain-predicted age difference (Brain-PAD) stably associates with the progression of subthreshold depression after adjusting for covariates.

Results

Individuals with subthreshold depression showed a higher risk of progression to MDD compared to healthy controls. Those whose condition worsened from subthreshold depression to MDD exhibited greater brain aging than those who remained subthreshold or recovered. Importantly, Brain-PAD remained significantly and stably associated with this progression after controlling for sex, ethnicity, lifestyle, and socioeconomic factors.

Conclusions

This study supports an association between brain aging and MDD progression and demonstrates a robust association between an increased Brain-PAD and the conversion from subthreshold depression to MDD. These findings enhance our understanding of MDD’s developmental trajectory and suggest that Brain-PAD may facilitate early detection and intervention targeting brain aging.

Keywords: major depressive disorder, subthreshold depression, brain aging, brain-predicted age difference, machine learning

Introduction

Major depressive disorder (MDD) is highly prevalent and frequently recurrent, imposing a considerable burden on individuals and society (Røssberg et al., 2021; Burani et al., 2023). Beyond affective and motivational disturbances, its clinical phenotype frequently encompasses cognitive slowing, fatigue, and dysregulation of sleep and metabolism (Villanueva, 2013; Shi et al., 2024), features that partially overlap with age-related decline (Wylie et al., 2022; Arleo et al., 2024). These observations underpin the hypothesis that a subset of individuals with MDD may exhibit accelerated or excess aging (Koutsouleris et al., 2014; Han et al., 2021; Luo et al., 2022). It remains unresolved whether such aging-related features constitute antecedents, correlates, or sequelae of MDD, and elucidating this relationship could refine pathophysiological understanding and inform preventive and therapeutic strategies.

Previous research has demonstrated that individuals with MDD may exhibit brain aging patterns that differ from those of healthy individuals matched on chronological age (Cole et al., 2018; Kaufmann et al., 2019; Leonardsen et al., 2022). To quantify such deviations in brain aging, the brain-predicted age difference (Brain-PAD), derived from neuroimaging data, has been introduced as a biomarker of accelerated or decelerated brain aging (Liem et al., 2017; Elliott et al., 2021). Brain-PAD is typically defined as the difference between a magentic resonance imaging (MRI)-derived estimate of an individual’s brain age and their chronological age, indexing macroscale deviation from normative aging (Wrigglesworth et al., 2022). In population cohorts, larger brain-age gaps have been associated with adverse cognitive and mental-health outcomes, suggesting potential clinical relevance (Koutsouleris et al., 2014; Han et al., 2021; Luo et al., 2022; Dai et al., 2025). However, most evidence is cross-sectional. Longitudinal designs with repeated neuroimaging and symptom assessments are needed to clarify how Brain-PAD relates to symptom trajectories and the temporal dynamics between brain aging and MDD, particularly during the period prior to meeting diagnostic criteria for MDD.

Conceptually and clinically, MDD is viewed as spanning a continuum, with subthreshold depression often representing a prodromal stage. Subthreshold depression refers to the presence of clinically significant depressive symptoms that do not meet the full diagnostic criteria for MDD (Kroenke, 2017). Prior research has indicated that individuals experiencing subthreshold depression may be at an elevated risk of developing MDD (Lee et al., 2019). This risk appears to be particularly pronounced among older adults, with studies suggesting that approximately 10–50% of this population may be in a prodromal phase of depression, which predisposes them to the onset of full-syndrome MDD (Meeks et al., 2011; Ludvigsson et al., 2018; Bosman et al., 2019). Moreover, subthreshold depression has been reported to be approximately three times more prevalent than MDD among older adults (Vaccaro et al., 2017; Xiang et al., 2018), and individuals experiencing subthreshold depressive symptoms appear to be at greater risk of progressing to full-syndrome MDD (Lee et al., 2019). From a public-health perspective, scalable markers that capture clinically meaningful variation across this continuum—beyond demographic and lifestyle factors—could facilitate risk stratification and monitoring and help inform the timing of interventions without presupposing causal direction (Gao et al., 2025).

Evidence linking depression to accelerated brain aging is suggestive but mixed. Several studies report larger brain-age gaps in MDD (Koutsouleris et al., 2014; Han et al., 2021; Luo et al., 2022; Jiang et al., 2024; Dai et al., 2025), whereas others find small or null effects (Besteher et al., 2019; Kaufmann et al., 2019). In many of these studies, reliance on clinic-recruited case-control samples may limit the generalizability of the results. Moreover, group differences are often reported without testing whether brain age provides incremental value after accounting for socioeconomic context, health behaviors, and symptom burden. Evidence on subthreshold depression is particularly sparse, with few large community-based studies systematically examining associations between brain aging and symptom severity.

In the present study, we investigated the association between Brain-PAD and the longitudinal progression of subthreshold depression within the large-scale UK Biobank cohort. The analytical procedure was conducted in two main steps. First, we examined whether Brain-PAD differed between individuals who shared the same baseline status (i.e. healthy controls or those with subthreshold depression) but exhibited divergent clinical trajectories at follow-up. Brain age was estimated using structural MRI (sMRI) features and a support vector regression (SVR) model, which has previously been validated for its robust and accurate prediction of brain age. Second, among individuals with subthreshold depression at baseline, we assessed whether increases in Brain-PAD were consistently and significantly associated with symptom worsening after adjusting for relevant covariates.

Methods

Study design and samples

The UK Biobank is a large-scale, detailed prospective study that recruited over 500,000 participants aged 40–69 years between 2006 and 2010. At baseline, it gathered extensive information on participants' mental health, lifestyles, socioeconomic factors, and cognitive function. During follow-up visits, in addition to collecting this type of information, the study also conducted MRI scans to capture brain structural details (such as cortical thickness and subcortical volumes) (Sudlow et al., 2015). The present study comprised two analyses. The first analysis focused on estimating brain age from structural imaging data and group comparisons. Following data cleaning procedures, a total of 37,936 participants were included in this phase. These individuals had complete data on chronological age, mental health status at both baseline and follow-up, as well as neuroimaging features required for brain age prediction derived from MRI scans. The second analysis involved sensitivity analysis. Consistent with the procedures used in the first part, data cleaning resulted in a final sample of 19,703 participants. In addition to the variables used in the first analysis, this subset also included all available data on potential covariates. After repeating the initial analyses to ensure consistency, we then conducted sensitivity analysis on the 2,011 individuals who presented with subthreshold depression at baseline. During the data cleaning process, we excluded individuals with missing values, as well as those who selected “prefer not to answer” or “do not know” for any items across the relevant questionnaires (Fig. 1).

Figure 1.

Figure 1

Open in a new tab

Overview of the analysis pipeline. Initially, baseline data were combined with follow-up data. Participants with missing data in any of these two datasets were excluded from further analyses. In Analysis 1, to accurately assess individuals’ mental health status, participants who responded with “prefer not to answer” or “do not know” on the PHQ-4 and PHQ-9 questionnaires were excluded. Subsequently, brain age prediction was performed, and between-group comparisons of Brain-PAD were conducted. In Analysis 2, additional potential covariates were incorporated, and participants who provided “prefer not to answer” or “do not know” responses in these covariate assessments were similarly excluded. The analyses performed in Analysis 1 were repeated on this reduced sample to verify the consistency of the results. Finally, individuals classified as having StD at baseline were included in subsequent covariate analyses. StD, subthreshold depression; PHQ, Patient Health Questionnaire; sMRI, structural magnetic resonance imaging.

Assessment of subthreshold depression and MDD

Depressive symptoms were assessed at both baseline and follow-up using validated mental health questionnaires. At baseline, subthreshold depression and MDD were evaluated with the Patient Health Questionnaire (PHQ)-4 (Stanhope, 2016). Participants rated four items on a four-point Likert scale (0 = not at all to 3 = nearly every day): 1, frequency of depressed mood (Data-Field: 2050); 2, frequency of unenthusiasm/disinterest (Data-Field: 2060); 3, frequency of tenseness/restlessness (Data-Field: 2070); and 4, frequency of tiredness/lethargy (Data-Field: 2080). Since items 3 and 4 primarily assess anxiety symptoms, only items 1 and 2 were considered for depression scoring. Total scores (range 0–6) were interpreted as follows: 2–3 indicated subthreshold depression, while >3 indicated MDD. During the 2016–2017 follow-up, mental health was assessed using the PHQ-9 (Kroenke et al., 2001) with the same Likert scale. The nine items included: 1, recent feelings of inadequacy (Feelings of inadequacy; Data-Field: 20507); 2, recent trouble concentrating on things (Cognitive problems; Data-Field: 20508); 3, recent feelings of depression (Depressed mood; Data-Field: 20510); 4, recent poor appetite or overeating (Appetite changes; Data-Field: 20511); 5, recent thoughts of suicide or self-harm (Suicidal ideation; Data-Field: 20513); 6, recent lack of interest or pleasure in doing things (Anhedonia; Data-Field: 20514); 7, trouble falling or staying asleep, or sleeping too much (Sleeping problems; Data-Field: 20517); 8, recent changes in speed/amount of moving or speaking (Psychomotor changes; Data-Field: 20518); and 9, recent feelings of tiredness or low energy (Fatigue; Data-Field: 20519). Total scores (range 0–27) were categorized as: 6–9 for subthreshold depression and ≥10 for MDD. We then calculated longitudinal transitions in depression status between baseline and follow-up.

Feature selection and brain age prediction

Based on established feature selection approaches from prior brain-age studies of MDD (Han et al., 2021; Dai et al., 2025), we included 162 brain imaging features for brain-age prediction. These features consisted of subcortical volumes from 14 brain regions and average cortical thickness from 148 regions defined by the Destrieux atlas(a2009s) (Destrieux et al., 2010). We employed SVR to develop our brain-age prediction model. The training dataset comprised 10,000 individuals randomly selected from participants who remained healthy (non-depressed) at both baseline and follow-up (n = 29,921). Model performance was evaluated using 10-fold cross-validation. The remaining participants were reserved as a validation sample for between-group comparisons. These validation subjects were stratified into six groups according to mental health status at baseline [healthy controls (HC) subthreshold depression] and follow-up (HC, subthreshold depression, MDD), forming 2 (baseline) × 3 (follow-up) groups for subsequent brain age prediction and Brain-PAD group analyses. First, we sought to determine whether, among individuals with subthreshold depression at baseline, those who later converted to MDD would exhibit higher brain age than those who recovered to a healthy state or remained subthreshold. Second, we aimed to examine, within participants who were healthy at baseline, whether brain age at follow-up varied in proportion to the degree of subsequent worsening in mental status.

Furthermore, previous research has suggested that cognitive aging or cognitive impairment in middle-aged and older adults may often be confounded with depressive symptoms (Yin et al., 2024), Therefore, in analysis 2, we conducted a covariate-adjusted group comparison with Brain-PAD as the dependent variable and follow-up mental health group as the fixed factor, including Global Cognitive Function (Data-Field: 20197; follow-up) as a covariate to assess the robustness of between-group Brain-PAD differences. We further included a Group × Global Cognitive Function interaction term to test whether cognitive status moderated the group effect.

Inclusion of covariates

All covariates were derived from baseline assessments, as we aimed to examine, in addition to Brain-PAD, which baseline factors are associated with subsequent transitions among individuals with subthreshold depression. We included sex (Data-Field: 31), race (Data-Field: 21000), body mass index (BMI, Data-Field: 21001), smoking status (Data-Field: 20116), alcohol consumption status (Data-Field: 20117), healthy physical activity status (Data-Field: 22038–22040), Townsend deprivation index (Data-Field: 22189), and childhood adversity as potential covariates. Following the approach of a previous study utilizing the UK Biobank dataset on depression and anxiety (Gao et al., 2023), these covariates were carefully categorized and operationalized. Specifically, body mass index (BMI) was classified into three categories: underweight or normal weight (<25 kg/m2), overweight (25–30 kg/m2), and obese (≥30 kg/m2). Healthy physical activity status was treated as a binary variable, defined according to criteria based on the short International Physical Activity Questionnaire (IPAQ) and expressed in Metabolic Equivalent Task (MET) minutes (Craig et al., 2003): moderate activity ≥150 min/week, vigorous activity ≥75 min/week, or a combination of moderate and vigorous activity totaling ≥150 min/week. The Townsend deprivation index was included as a continuous variable, constructed from four indicators: unemployment, household overcrowding, non-car ownership, and non-home ownership; higher scores on this index indicate greater levels of socioeconomic deprivation. Childhood adversity was assessed using a shortened version of the Childhood Trauma Questionnaire administered during the follow-up survey. This measure includes five items that evaluate different domains of childhood adversity: emotional abuse, physical abuse, emotional neglect, sexual abuse, and physical neglect. Specifically, the items are as follows: (i) feeling hated by a family member (emotional abuse; Data-Field: 20487), (ii) being physically abused by family members during childhood (physical abuse; Data-Field: 20488), (iii) feeling loved as a child (emotional neglect; Data-Field: 20489), (iv) experiencing sexual molestation during childhood (sexual abuse; Data-Field: 20490), and (v) having someone to take the participant to the doctor when needed as a child (physical neglect; Data-Field: 20491). For each item, responses were recorded on a five-point scale ranging from “never true” to “very often true.” Scoring was conducted as follows: for the physical neglect item, responses of “never true,” “rarely true,” “sometimes true,” or “often true” were assigned a score of 1; for emotional neglect, responses of “never true,” “rarely true,” or “sometimes true” were scored as 1; and for the items assessing sexual abuse, physical abuse, and emotional abuse, responses of “rarely true,” “sometimes true,” “often true,” or “very often true” were scored as 1. The total score ranged from 0 to 5, with higher scores indicating greater severity of childhood adversity.

Statistical analysis

We employed logistic regression analyses to examine whether Brain-PAD was associated with the progression from subthreshold depression at baseline to MDD at follow-up, and to investigate whether an increase in Brain-PAD is associated with this progression. A series of models with incremental adjustment for covariates were fitted: Model 1 included sex and race; Model 2 (lifestyle) adjusted for BMI, smoking status, alcohol consumption status, and healthy physical activity status; and Model 3 (socioeconomic) controlled for Townsend deprivation index and childhood adversity.

Results

Conversion of participants

At baseline, there were 33,191 individuals classified as healthy, among whom 2,365 met the criteria for subthreshold depression and 905 transitioned to MDD at follow-up. Among the 4,036 individuals with subthreshold depression at baseline, 2,484 had recovered to a healthy status at follow-up, while 650 had progressed to MDD. As expected, only a small proportion of initially healthy individuals (2.73%) developed MDD during the follow-up period, whereas 16.11% of those with subthreshold depression experienced a worsening to MDD. The likelihood of progression from subthreshold depression to MDD was approximately 5.90 times greater than that observed in healthy individuals. Furthermore, among individuals diagnosed with MDD at baseline (N = 709), just 38.36% recovered to a healthy state, compared to a higher recovery rate of 61.55% observed in those with subthreshold depression. The probability of recovering to a healthy status was therefore approximately 1.60 times greater for individuals with subthreshold depression than for those with MDD (Fig. 2A).

Figure 2.

Figure 2

Open in a new tab

(A) This analysis included a total of 37,936 individuals, and their mental health status transitions between baseline and follow-up were examined. (B) The 10 brain regions with the largest contributions to brain-age prediction; β denotes the model coefficient (weight) for each region (red: positive weights; blue: negative weights). (C) Among participants classified as having subthreshold depression at baseline, a between-group comparison of Brain-PAD across different follow-up outcomes revealed significant differences. Specifically, individuals who converted to MDD at follow-up exhibited the highest Brain-PAD. (D) Among participants classified as HC at baseline, a between-group comparison of Brain-PAD across different follow-up outcomes revealed significant differences. Specifically, individuals who converted to MDD at follow-up exhibited the highest Brain-PAD. HC, healthy control; StD, subthreshold depression; MDD, major depressive disorder. Transition groups are denoted accordingly, e.g. StD2HC refers to individuals with subthreshold depression at baseline who transitioned to healthy control status at follow-up. *P < 0.05; **P < 0.01; ****P < 0.0001.

Brain prediction and group comparison

Using a brain-age prediction model trained on healthy individuals, we obtained a mean absolute error (MAE) of 4.75 in a test set comprising participants who exhibited subthreshold depression at baseline and divergent mental status at follow-up. This performance is broadly consistent with previously reported results from brain-age prediction models in patients with MDD (Han et al., 2021; Dai et al., 2025). Subsequently, we estimated Brain-PAD for individuals with subthreshold depression and healthy controls at baseline, stratified by their mental health status at follow-up. Among individuals with subthreshold depression at baseline, between-group comparisons of Brain-PAD revealed a significant effect (F = 16.66, P < 0.0001, Fig. 2C). Tukey’s multiple comparisons test indicated that those who progressed to MDD exhibited the highest Brain-PAD at follow-up. The difference in Brain-PAD between individuals who deteriorated to MDD and those who remained with subthreshold depression was significant (mean difference = 0.87, q = 4.18, P = 0.0089), whereas a significant difference was observed between those who progressed to MDD and those who remitted to a healthy state (mean difference = 1.42, q = 7.99, P < 0.0001). Additionally, individuals who maintained subthreshold depression showed significantly higher Brain-PAD compared to those who remained healthy (mean difference = 0.55, q = 3.52, P = 0.0341). Similarly, among individuals classified as healthy at baseline, group comparisons also revealed significant differences (F = 168.40, P < 0.0001, Fig. 2D). Tukey’s test showed that those who converted to MDD had the highest Brain-PAD at follow-up, which was significantly greater than both those who transitioned to subthreshold depression (mean difference = 1.17, q = 7.53, P < 0.0001) and those who remained healthy (mean difference = 2.71, q = 20.05, P < 0.0001). Furthermore, participants who developed subthreshold depression exhibited significantly higher Brain-PAD compared to those who remained healthy (mean difference = 1.54, q = 17.80, P < 0.0001). Subsequently, the weights (i.e. model coefficients) assigned to each brain region within the SVR model were ranked in descending order based on their absolute values. This procedure identified the top 10 brain regions contributing most significantly to brain age prediction: differences in cortical thickness in several regions—including the bilateral superior frontal gyrus and bilateral anterior transverse temporal gyrus; the left-hemispheric central sulcus, superior parietal lobule, and anterior part of the cingulate gyrus and sulcus (ACC); and the right-hemispheric inferior segment of the circular sulcus of the insula, angular gyrus, and lingual gyrus—were identified as the brain regions contributing most strongly to brain-age prediction and we performed between-group comparisons of cortical thickness in these regions (Fig. 2B; Supplementary Tables 1, 2, 3). Finally, we calculated the Brain-PAD across the different participant groups at follow-up. Our computations found that individuals with MDD continued to exhibit a significantly higher Brain-PAD compared to the other two groups. Furthermore, a moderately elevated Brain-PAD was also observed in the subthreshold depression group compared to the healthy controls (Supplementary Fig. 2).

Sensitivity analysis including covariates

In the sensitivity analysis, the inclusion of covariates resulted in a reduced sample size; therefore, we re-analyzed this subset of data using the approach applied in the first analysis. The findings were consistent with those obtained in the initial analysis (Supplementary results 1, 2). To further investigate potential confounding effects, global cognitive function was included as a covariate. In the subthreshold depression group at baseline, both follow-up group status (F = 8.23, P = 0.0003) and global cognitive function (F = 11.68, P = 0.0006) showed significant effects on Brain-PAD, while their interaction was not significant (F = 0.36, P = 0.6956). These findings suggest that although cognitive function influences Brain-PAD, it does not compromise the consistency of our primary results.

Among the 2011 individuals with subthreshold depression at baseline, 457 subsequently progressed to MDD. Logistic regression analyses across all models consistently indicated that Brain-PAD at follow-up was significantly associated with the transition from subthreshold depression to MDD, even after adjusting for various covariates [Model 1: odds ratio (OR) = 1.04, 95% confidence interval (CI) = 1.02–1.06, P < 0.001; Model 2: OR = 1.03, 95% CI = 1.01–1.06, P < 0.05; Model 3: OR = 1.03, 95% CI = 1.01–1.06, P < 0.05]. In Model 1, males were less likely than females to progress from subthreshold depression to MDD: OR = 0.76, 95% CI = 0.59–0.97, P < 0.05. In Model 2, which included lifestyle-related covariates, obesity as measured by BMI and current smoking emerged as significant risk factors for progression to MDD [BMI (Obese): OR = 1.75, 95% CI = 1.27–2.42, P < 0.001; Smoking (current): OR = 1.53, 95% CI = 1.01–2.26, P < 0.05]. Furthermore, Model 3, which accounted for socioeconomic variables, revealed that both higher Townsend deprivation index (OR = 1.05, 95% CI = 1.00–1.09, P< 0.05) and greater childhood adversity (OR = 1.41, 95% CI = 1.29–1.54, P < 0.001, Fig. 3) significantly increased the risk of deterioration from subthreshold depression to MDD. However, the associated between transition status and Brain-PAD at follow-up remained significant.

Figure 3.

Figure 3

Open in a new tab

A logistic regression analysis was conducted among 2,011 participants who were identified as subthreshold depression at baseline, with the outcome variable defined as conversion to MDD at follow-up. Across all three models tested, Brain-PAD consistently associated with the progression from subthreshold depression to MDD. In addition to Brain-PAD, other variables associated with increased risk included sex, obesity, smoking, higher levels of socioeconomic deprivation, and exposure to childhood adversity.

Discussion

This study represents one of the first efforts to investigate brain aging in relation to subthreshold depression within an elderly population. As expected, individuals with subthreshold depression had a 5.90-fold higher likelihood of progressing to MDD than their non-depressed counterparts. Moreover, those who converted from subthreshold depression to MDD exhibited greater brain aging than individuals who remained subthreshold or recovered. Finally, among participants with subthreshold depression at baseline, Brain-PAD remained significantly associated with progression to MDD even after adjusting for multiple covariates.

Our findings indicate that among individuals with subthreshold depression at baseline, both those who progressed to MDD and those who remained in the subthreshold depression state exhibited significantly higher Brain-PAD at follow-up compared to individuals who recovered to a healthy state. These results contribute to establishing a link between the degree of brain aging and the severity of depressive symptoms. Subthreshold depression, often characterized by atypical somatic symptoms and subtle emotional manifestations, is considered a latent state preceding MDD, and is challenging to detect through behavioral assessments alone (Tuithof et al., 2018). In this context, Brain-PAD, as a biomarker of brain aging, appears to sensitively reflect morphological and functional changes associated with MDD. Consistent with these findings, Gao et al. (2023), utilizing data from the UK Biobank, demonstrated that accelerated biological aging—measured using biological age metrics that index whole-body physiological aging via integrated clinical biomarkers, in contrast to the structural MRI-based brain aging metric used in the present study—is associated with an increased risk of depression, independent of genetic predisposition. By employing Brain-PAD as a measure of brain aging, our study provides complementary evidence at the neurobiological level. It is plausible that brain aging and biological aging assessed through blood markers share underlying mechanisms, such as inflammation or oxidative stress, which may partly explain their association with depression progression (Miller and Raison, 2016; Picard et al., 2019; Beurel et al., 2020).

Our analysis identified several specific brain regions that contribute significantly to the brain age prediction model, including the bilateral superior frontal gyrus and bilateral anterior transverse temporal gyrus; the left-hemispheric central sulcus, superior parietal lobule, and anterior part of the cingulate gyrus and sulcus; and the right-hemispheric inferior segment of the circular sulcus of the insula, angular gyrus, and lingual gyrus. In our model, the bilateral superior frontal gyrus (SFG) ranked among the regions contributing most strongly to brain-age prediction. This is consistent with evidence that normative aging is accompanied by robust cortical thinning in frontal cortex, including the superior frontal gyrus (Lemaitre et al., 2012). Beyond aging, convergent evidence in MDD also indicates cortical thickness alterations in prefrontal regions; notably, a meta-analysis of medication-free MDD reported reduced cortical thickness in the orbital segment of the SFG relative to healthy controls (Li et al., 2020). Taken together, these findings suggest that SFG morphology may represent an anatomical intersection between brain aging-related cortical change and MDD. The left central sulcus, which anatomically demarcates the primary motor and somatosensory cortices, was among the features contributing most strongly to brain-age prediction in our model. This finding is plausible given prior evidence that central sulcus morphology and peri-central cortical architecture show age-related changes in healthy aging cohorts (e.g. reductions in sulcal surface area and related metrics) (Li et al., 2011). Clinically, depression—particularly in later life—often involves psychomotor and somatic symptoms, and recent work has increasingly implicated motor/sensorimotor circuits in these symptom dimensions (Liang et al., 2025). Accordingly, the prominence of the central sulcus in our brain-age model may reflect the sensitivity of sensorimotor cortical structure to normative aging processes, with potential relevance to MDD; however, we interpret this association cautiously and do not infer disorder-specific mechanisms from feature importance alone.

A particularly notable aspect of this study is the robust association observed between high Brain-PAD and progression from subthreshold depression to MDD, even after adjustment for multiple potential confounding factors. These covariates included sex, ethnicity, various lifestyle factors (such as smoking status, alcohol, and physical activity), and socioeconomic variables—specifically, the Townsend deprivation index and childhood adversity—with socioeconomic factors emerging as key risk contributors to the worsening of subthreshold depression. Supporting this, prior research in adolescent populations has identified trauma exposure as a critical risk factor for the transition from subthreshold depression to MDD; adolescents who have experienced multiple traumatic events exhibit a higher prevalence of depressive disorders (Chen et al., 2025). These findings underscore the heterogeneity inherent in the progression of subthreshold depression and highlight the importance of tailored intervention strategies targeting specific subgroups. For instance, early intervention and treatment may be particularly warranted for individuals with a history of trauma. Importantly, Brain-PAD appears to be associated with progression to depression relatively independently from these demographic, social, and behavioral variables, suggesting that it does not merely serve as a proxy for other known risk factors, but rather functions as an independent biological marker perhaps reflecting intrinsic brain health and resilience.

Limitations

Our study provides evidence that follow-up Brain-PAD is associated with subsequent clinical outcomes among individuals with subthreshold depression. However, several limitations of our research should be acknowledged. First, although we utilized longitudinal data, our analyses were largely limited to two time points—baseline and follow-up—because most UK Biobank neuroimaging was acquired at a single follow-up visit. As a result, we could estimate brain age only at follow-up, and consequently reverse causation cannot be ruled out as it is not possible to disentangle cause and effect in our study design. Future work should pair more frequent mental health assessments with repeated imaging to examine the dynamics between brain aging and psychiatric status. In addition to database constraint, different questionnaires were used to assess depressive status at baseline and follow-up. Although we chose measures with similar constructs to enhance comparability, future longitudinal studies should administer the same instrument across time points to further improve the accuracy of symptom change estimates. Second, the primary aim of our study was to investigate whether increased Brain-PAD constitutes a stable biomarker for subthreshold depression progressing to MDD. Consequently, lifestyle variables were not exhaustively incorporated as covariates, and additional risk and protective factors may remain unaccounted for. Third, we did not adjust for antidepressant treatment. Although the UK Biobank provides self-reported medication use at the assessment visit (Data-Field 20003), it lacks key information on treatment timing and dosage, which may influence brain structural measures; future longitudinal studies with detailed medication histories are needed to address this potential confounding.

Conclusions

In summary, among older adults, individuals with subthreshold depression are more likely to progress to MDD than healthy peers. Those who deteriorated to MDD showed greater brain aging, reflected by higher Brain-PAD values at follow-up, which was robust to inclusion of demographic, lifestyle, and socioeconomic factors. Overall, subthreshold depression may represent a transitional state toward MDD. Future studies should assess the potential for Brain-PAD to identify at-risk individuals and guide early intervention to enable timely prevention and mitigate clinical deterioration.

Supplementary Material

kkag005_Supplemental_File
kkag005_supplemental_file.docx (379.2KB, docx)

Acknowledgements

We thank Jonathan P. Rosier for suggestions for the manuscript preparation. This study was supported by National Key R & D Program of China [grant number SIT2030-Major Projects 2022ZD0214300], Nature Science Foundation of China [grant number 32271139], and Natural Science Foundation of Guangdong Province of China [grant number 2023A1515011331].

Contributor Information

Haowei Dai, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Lijing Niu, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Qingzi Zhu, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Yuanyuan Zeng, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Yutong Ying, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Xueping Yin, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Xiangyi Liang, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China.

Xingqin Wang, Department of Neurosurgery, Institute of Brain Diseases, Nanfang Hospital of Southern Medical University, Guangzhou 510515, China.

Bihua Zhou, School of Laboratory Medicine and Biotechnology, Southern Medical University, Guangzhou 510515, China.

Qing Ma, Shanghai Key Laboratory of Mental Health and Psychological Crisis Intervention, School of Psychology and Cognitive Science, East China Normal University, Shanghai 200062, China.

Ruibin Zhang, Cognitive Control and Brain Healthy Laboratory, Department of Psychology, School of Public Health, Southern Medical University, Guangzhou 510515, China; Department of Psychiatry, Zhujiang Hospital, Southern Medical University, Guangzhou 510515, China; Guangdong-Hong Kong-Macao Greater Bay Area Center for Brain Science and Brain-Inspired Intelligence, Guangdong-Hong Kong Joint Laboratory for Psychiatric Disorders, Guangdong Basic Research Center of Excellence for Integrated Traditional and Western Medicine for Qingzhi Diseases, Southern Medical University, Guangzhou 510515, China.

Author contributions

Haowei Dai (Writing–original draft, Investigation, Methodology, Formal analysis, Visualization), Lijing Niu (Investigation), Qingzi Zhu (Investigation), Yuanyuan Zeng (Investigation), Yutong Ying (Investigation), Xueping Yin (Investigation), Xiangyi Liang (Investigation), Xingqin Wang (Investigation), Bihua Zhou (Investigation), Qing Ma (Conceptualization, Resources, Supervision, Writing–review & editing), Ruibin Zhang (Conceptualization, Resources, Funding acquisition, Supervision, Writing–review & editing).

Conflicts of interest

The authors have no conflicts of interest.

References

  1. Arleo  A, Bareš  M, Bernard  JA  et al. (2024) Consensus Paper: cerebellum and ageing. Cerebellum. 23(2):802–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Besteher  B, Gaser  C, Nenadić  I (2019) Machine-learning based brain age estimation in major depression showing no evidence of accelerated aging. Psychiatry Res Neuroimaging. 290, 1–4. [DOI] [PubMed] [Google Scholar]
  3. Beurel  E, Toups  M, Nemeroff  CB (2020) The bidirectional relationship of depression and inflammation: double trouble. Neuron. 107(2):234–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bosman  RC, ten Have  M, de Graaf  R  et al. (2019) Prevalence and course of subthreshold anxiety disorder in the general population: a three-year follow-up study. J Affect Disord. 247: 105–13. [DOI] [PubMed] [Google Scholar]
  5. Burani  K, Brush  CJ, Shields  GS  et al. (2023) Cumulative lifetime acute stressor exposure interacts with reward responsiveness to predict longitudinal increases in depression severity in adolescence. Psychol Med. 53(10):4507–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Chen  K, Li  Q, Zhu  Z  et al. (2025) Risk and protective factors for the evolution of subthreshold depression during early adolescence. J Adolesc Health. 76(3):385–95. [DOI] [PubMed] [Google Scholar]
  7. Cole  JH, Ritchie  SJ, Bastin  ME  et al. (2018) Brain age predicts mortality. Mol Psychiatry. 23(5):1385–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Craig  CL, Marshall  AL, Sjöström  M  et al. (2003) International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exercise. 35(8):1381–95. [DOI] [PubMed] [Google Scholar]
  9. Dai  H, Niu  L, Peng  L  et al. (2025) Accelerated brain aging in patients with major depressive disorder and its neurogenetic basis: evidence from neurotransmitters and gene expression profiles. Psychol Med. 55, e71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Destrieux  C, Fischl  B, Dale  A  et al. (2010) Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature. Neuroimage. 53(1):1–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Elliott  ML, Belsky  DW, Knodt  AR  et al. (2021) Brain-age in midlife is associated with accelerated biological aging and cognitive decline in a longitudinal birth cohort. Mol Psychiatry. 26(8):3829–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Gao  Q-L, Chen  X, Castellanos  FX  et al. (2025) Towards closed-loop precision psychiatry: integrating MRI biomarkers for individualized care of major depressive disorder. Psychoradiology. 5, kkaf024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Gao  X, Geng  T, Jiang  M  et al. (2023) Accelerated biological aging and risk of depression and anxiety: evidence from 424,299 UK Biobank participants. Nat Commun. 14(1):2277. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Han  LKM, Dinga  R, Hahn  T  et al. (2021) Brain aging in major depressive disorder: results from the ENIGMA major depressive disorder working group. Mol Psychiatry. 26(9):5124–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Jiang  J, Ferraro  S, Zhao  Y  et al. (2024) Common and divergent neuroimaging features in major depression, posttraumatic stress disorder, and their comorbidity. Psychoradiology. 4, kkae022. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kaufmann  T, van der Meer  D, Doan  NT  et al. (2019) Common brain disorders are associated with heritable patterns of apparent aging of the brain. Nat Neurosci. 22(10):1617–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Koutsouleris  N, Davatzikos  C, Borgwardt  S  et al. (2014) Accelerated brain aging in schizophrenia and beyond: a neuroanatomical marker of psychiatric disorders. Schizophr Bull. 40(5):1140–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kroenke  K (2017) When and how to treat subthreshold depression. JAMA. 317(7):702–4. [DOI] [PubMed] [Google Scholar]
  19. Kroenke  K, Spitzer  RL and Williams  JB (2001) The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med. 16:606–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Lee  YY, Stockings  EA, Harris  MG  et al. (2019) The risk of developing major depression among individuals with subthreshold depression: a systematic review and meta-analysis of longitudinal cohort studies. Psychol Med. 49(1):92–102. [DOI] [PubMed] [Google Scholar]
  21. Lemaitre  H, Goldman  AL, Sambataro  F  et al. (2012) Normal age-related brain morphometric changes: nonuniformity across cortical thickness, surface area and grey matter volume?. Neurobiol Aging. 33(3):617.e1–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Leonardsen  EH, Peng  H, Kaufmann  T  et al. (2022) Deep neural networks learn general and clinically relevant representations of the ageing brain. Neuroimage. 256, 119210. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Li  Q, Zhao  Y, Chen  Z  et al. (2020) Meta-analysis of cortical thickness abnormalities in medication-free patients with major depressive disorder. Neuropsychopharmacol. 45(4):703–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Li  S, Xia  M, Pu  F  et al. (2011) Age-related changes in the surface morphology of the central sulcus. Neuroimage. 58(2):381–90. [DOI] [PubMed] [Google Scholar]
  25. Liang  Q, Zhou  Z, Li  Y  et al. (2025) Motor circuits and beyond: functional connectivity related to psychomotor syndromes in depression. Psychol Med. 55, e285. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Liem  F, Varoquaux  G, Kynast  J  et al. (2017) Predicting brain-age from multimodal imaging data captures cognitive impairment. Neuroimage. 148, 179–88. [DOI] [PubMed] [Google Scholar]
  27. Ludvigsson  M, Bernfort  L, Marcusson  J  et al. (2018) Direct costs of very old persons with subsyndromal depression: a 5-year prospective study. Am J Geriatr Psychiatry. 26(7):741–51. [DOI] [PubMed] [Google Scholar]
  28. Luo  Y, Chen  W, Qiu  J  et al. (2022) Accelerated functional brain aging in major depressive disorder: evidence from a large scale fMRI analysis of Chinese participants. Transl Psychiatry. 12(1):397. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Meeks  TW, Vahia  IV, Lavretsky  H  et al. (2011) A tune in “a minor” can “b major”: a review of epidemiology, illness course, and public health implications of subthreshold depression in older adults. J Affect Disord. 129(1):126–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Miller  AH, Raison  CL (2016) The role of inflammation in depression: from evolutionary imperative to modern treatment target. Nat Rev Immunol. 16(1):22–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Picard  M, Trumpff  C, Burelle  Y (2019) Mitochondrial psychobiology: foundations and applications. Curr Opin Behav Sci. 28, 142–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Røssberg  JI, Evensen  J, Dammen  T  et al. (2021) Mechanisms of change and heterogeneous treatment effects in psychodynamic and cognitive behavioural therapy for patients with depressive disorder: a randomized controlled trial. BMC Psychol. 9, 11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Shi  Y, Zhu  J, Hou  C  et al. (2024) Mining key circadian biomarkers for major depressive disorder by integrating bioinformatics and machine learning. Aging. 16(12):10299–320. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Stanhope  J (2016) Patient Health Questionnaire-4. Occup Med (Lond). 66:760–761. [DOI] [PubMed] [Google Scholar]
  35. Sudlow  C, Gallacher  J, Allen  N  et al. (2015) UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 12(3):e1001779. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Tuithof  M, ten Have  M, van Dorsselaer  S  et al. (2018) Course of subthreshold depression into a depressive disorder and its risk factors. J Affect Disord. 241, 206–15. [DOI] [PubMed] [Google Scholar]
  37. Vaccaro  R, Borrelli  P, Abbondanza  S  et al. (2017) Subthreshold depression and clinically significant depression in an italian population of 70-74-year-olds: prevalence and association with perceptions of self. Biomed Res Int. 2017, 1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Villanueva  R (2013) Neurobiology of major depressive disorder. Neural Plast. 2013, 873278. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Wrigglesworth  J, Yaacob  N, Ward  P  et al. & ASPREE investigator group. (2022) Brain-predicted age difference is associated with cognitive processing in later-life. Neurobiol Aging. 109, 195–203. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Wylie  GR, Pra Sisto  AJ, Genova  HM  et al. (2022) Fatigue across the lifespan in men and women: state vs. Trait. Front Hum Neurosci. 16, 790006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Xiang  X, Leggett  A, Himle  JA  et al. (2018) Major depression and subthreshold depression among older adults receiving home care. Am J Geriatr Psychiatry. 26(9):939–49. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Yin  J, John  A, Cadar  D (2024) Bidirectional associations of depressive symptoms and cognitive function over time. JAMA Netw Open. 7(6):e2416305. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

kkag005_Supplemental_File
kkag005_supplemental_file.docx (379.2KB, docx)

Articles from Psychoradiology are provided here courtesy of Oxford University Press

RESOURCES