Brainstem volume mediates seasonal variation in depressive symptoms: A cross sectional study in the UK Biobank cohort

Seasonal differences in mood and depressive symptoms affect a large percentage of the general population, with seasonal affective disorder (SAD) representing the most common presentation. SAD affects up to 3% of the world’s population, and it tends to be more predominant in females than males. The brainstem has been shown to be affected by photoperiodic changes, and that longer photoperiods are associated with higher neuronal density and decreased depressive-like behaviours. We predict that longer photoperiod days are associated with larger brainstem volumes and lower depressive scores, and that brainstem volume mediates the seasonality of depressive symptoms. Participants (N = 9289, 51.8% females and 48.1% males) ranging in age from 44 to 79 years were scanned by MRI at a single location. Photoperiod was found to be negatively correlated with low mood and anhedonia in females while photoperiod was found to be positively correlated with brainstem volumes. In females, whole brainstem, pons and medulla volumes individually mediated the relationship between photoperiod and both anhedonia and low mood, while midbrain volume mediated the relationship between photoperiod and anhedonia. No mediation effects were seen in males. Our study extends the understanding of the neurobiological factors that contribute to the pathophysiology of seasonal mood variations.

Seasonal fluctuations in mood and depressive symptoms affect a large number of the general population, and these depressive symptoms such as depressed mood and fatigue have been found to be greater in winter compared to summer seasons in higher latitude countries [1][2][3][4] . Populations with seasonal affective disorder (SAD), a type of recurring major depression with a seasonal pattern, represent the most common form of seasonal fluctuations in mood 5,6 . SAD is often characterized by depression and fatigue occurring in winter with full remission taking place in summer. It has been reported that SAD affects up to 3% of the world's population, and it tends to be more predominant in females than males with a reported female-to-male ratio of 4:1 [7][8][9] . Females have been found to suffer from mood changes and depressive symptoms related to dark and cloudy weather at a greater rate compared to males 4,9 . Although seasonal variations in mood have been studied widely among sexes, little is known about the neurobiological factors linking light exposure and mood.
It has been suggested that changes in photoperiod (duration of sunlight) may be associated with seasonal mood variations 10,11 by shifting the circadian phase with its associated disruptions in sleep and other health outcomes. However, photoperiodic changes have also been suggested to affect specific brain regions that might be implicated in mood disorders. For example, the hippocampus and hypothalamus have been shown to be affected by seasonal changes in photoperiod. In particular, a smaller volume of the hippocampus was associated with shorter photoperiods in the winter months compared to summer [12][13][14] , and higher gene expression and hormonal activity of the hypothalamus were associated with longer photoperiod days in summer compared to winter 15 . In addition, the brainstem has been shown to be associated with seasonal changes. In particular, in Rana temporaria L., the size of the nuclei of the medulla oblongata cells controlling lipofuscin in pigment was significantly associated with changes in photoperiod during the annual cycle, and that higher volume of the nuclei of the medulla oblongata was detected in July and lower volume was detected in March 16 . Moreover, photoperiodic changes have been shown to influence the midbrain dorsal raphe serotonin neurons. For example, mice exhibit increased firing Association of photoperiod with depressive symptoms. Negative binomial regression was conducted to investigate the association between photoperiod and depressive symptoms including low mood, anhedonia, tenseness, tiredness and total depressive score. For all participants there was a significant negative correlation between photoperiod and low mood (p < 0.05). When corrected for age, ethnicity, living area (urban or rural) and Townsend deprivation index the correlation in low mood did not remain significant (See Table 3). No significant correlations between photoperiod and anhedonia, tenseness, tiredness and total depressive score for all participants were seen. In females, photoperiod was negatively correlated only with low mood and anhedonia (p = 0.03) when corrected for age, ethnicity, living area (urban or rural) and Townsend deprivation index. When the p-value was Bonferroni corrected (0.05/15 = 0.003), there were no significant correlations between photoperiod and low mood and anhedonia in females. In males, no significant correlations between photoperiod and any of these depressive symptoms (low mood, anhedonia, tenseness, tiredness and total depressive score) were seen either before or after correction for the above confounders.
Association of brainstem substructure volumes with photoperiod. There were significant correlations between brainstem and substructure volumes and photoperiod in all participants and both females and males separately when the p-value was corrected for multiple comparisons (p < 0.003). To further explore the observed association between brainstem volume and photoperiod, a multiple linear regression model was applied to better understand the variance in the brainstem volume when accounting for known confounds. Age and total brain volume (TBV) covariates were found to associate with brainstem and substructure volumes, and that age was negatively correlated with all brainstem volumes (p < 0.001), while TBV was positively correlated with all brainstem volumes in all participants and both females and males. Thus, these covariates were entered in separate blocks in a hierarchical regression model to account for their confounding effects. When photoperiod was corrected for these two covariates, there were significant correlations between brainstem and substructure volumes and photoperiod in all participants and both females and males separately with the p-value corrected for multiple comparisons (p < 0.003). Photoperiod was positively correlated with whole brainstem r 2 (9289) = 0.021, medulla r 2 (9289) = 0.031, pons r 2 (9289) = 0.016, SCP r 2 (9289) = 0.004 and midbrain r 2 (9289) = 0.015 volumes in all participants, and with whole brainstem r 2 (4817) = 0.024, medulla r 2 (4817) = 0.033, pons r 2 (4817) = 0.019, SCP r 2 (4817) = 0.004 and midbrain r 2 (4817) = 0.014 volumes in females and with whole brainstem r 2 (4472) = 0.021, medulla r 2 (4472) = 0.031, pons r 2 (4472) = 0.014, SCP r 2 (4472) = 0.005 and midbrain r 2 (4472) = 0.019 volumes, p < 0.001 in males ( Fig. 1 and Table 4).
Association of brainstem substructure volumes with depressive symptoms. There were significant correlations between brainstem and substructure volumes and depressive symptoms including low mood, anhedonia and total depressive symptoms in all participants but only in females when the group was categorised by sex (corrected for age, TBV, ethnicity, living area and Townsend deprivation index). For all participants, low mood, anhedonia and total depressive score were negatively correlated with all brainstem volumes except the midbrain and SCP, p < 0.05 for all (See Table 5). When the p-value was Bonferroni corrected (0.05/75 = www.nature.com/scientificreports www.nature.com/scientificreports/ 0.0006), only the correlation between medulla and total depressive symptom remained significant (p < 0.0006). In females, whole brainstem and pons were associated with low mood, anhedonia and total depressive score, and the medulla volume was associated with low mood, anhedonia, tenseness and total depressive score, whereas midbrain volume was significantly associated with only anhedonia, p < 0.05 for all (See Table 5). When the p-value was Bonferroni corrected (0.05/75 = 0.0006), the correlations between anhedonia and whole brainstem and pons remained significant (p < 0.0006). No significant correlations between SCP volume and all depressive symptoms including low mood, anhedonia, tenseness, tiredness and total depressive score were seen in females. No significant correlations between all brainstem substructure volumes and low mood, anhedonia, tenseness, tiredness and total depressive score were seen in males.

Mediation analysis.
A mediation analysis was performed to examine whether brainstem volumes mediate the relationship between photoperiod and depressive symptoms including low mood and anhedonia in females (Fig. 2). Negative binomial regression analysis (corrected for age, ethnicity, living area, Townsend deprivation index and TBV) was used to test path-correlations. Photoperiod and the hypothesised mediator(s) (whole brainstem, midbrain, pons and medulla volumes) were significantly associated (Table 4). In addition, whole brainstem, pons and medulla volumes were significantly related to both low mood and anhedonia, while midbrain volume was significantly related to anhedonia (Table 5). To test whether volume reduced the associations of photoperiod and anhedonia or photoperiod and low mood, whole brainstem, midbrain, pons and medulla volumes were added separately as predictors to negative binomial regression models. We found that: (1) the association between photoperiod and anhedonia was reduced and no longer significant when whole brainstem, midbrain, pons and medulla were included; β = −0. .005] and p = 0.097 for medulla, while the association between photoperiod and low mood remained significant and did not reduce when midbrain was included, β = −0.030, CI [−0.063 to −0.002] and p = 0.040. Because these results satisfy the requirements of the mediation analysis, we examined whether whole brainstem, midbrain, pons and medulla significantly mediate the relationship between photoperiod and anhedonia, and whether whole brainstem, pons and medulla significantly mediate the relationship between photoperiod and low mood.
To formally test the mediation, we used a bias corrected and accelerated bootstrap method (PROCESS macro in SPSS). The indirect effects were significant (See Table 6) meaning that longer photoperiod days were associated with (1)    .010] and p = 0.112, suggesting that it was not associated with reporting reduced low mood. When the p-value was Bonferroni corrected (p = 0.006), the indirect effects of the mediators (midbrain, pons, medulla and whole brainstem) on the relationship between photoperiod and anhedonia and low mood (except midbrain) remained significant. To conclude, in females whole brainstem, midbrain, pons and medulla volumes mediate the relationship between photoperiod and both anhedonia and low mood, while midbrain volume mediates the relationship between photoperiod and anhedonia.

Discussion
We have shown that brainstem volumes are associated with photoperiod in humans. Interestingly, we found that in females whole brainstem, pons, and medulla volumes mediated the relationship between photoperiod and both anhedonia and low mood, while midbrain volume mediated the relationship between photoperiod and anhedonia only. No mediation effects for the other depressive symptoms were found in females. No mediation effects were found in males. These findings are the first to demonstrate the mediating effects of brainstem volumes on the seasonal variability of mood and anhedonia.
We also found that photoperiod was associated with depressive symptoms including low mood and anhedonia in females but not in males, where longer photoperiod days were associated with reporting reduced low mood and anhedonia. This however did not remain after Bonferroni correction. This association has been previously reported 4 . Lyall et al., study had significantly greater statistical power (n = up to 80,000) which may explain why the association no longer remains significant after correction.
To our knowledge no previous animal or human studies have reported seasonal variations in brainstem volumes with only a small number focusing on the association between photoperiod and the density of serotonergic and dopaminergic neurons and binding transporters in the brainstem, especially midbrain or raphe nuclei as described above 20,21,17,18,30 . However, circadian rhythms are generated and maintained by a neural clock that is regulated by midbrain raphe nuclei in the suprachiasmatic nucleus (SCN) 31 . Therefore, any circadian clock disruptions caused by changes in photoperiod may alter midbrain raphe nuclei which in turn may lead to morphology changes of the brainstem. Our findings support the notion that changes in photoperiod change the brainstem substructure volumes.
The biological mechanisms behind changes in brainstem volumes in mood disorders are still unclear. Previous studies [22][23][24] that have shown that individuals with major depressive disorder (MDD) showed increased whole brainstem and midbrain volumes compared to healthy controls. In addition, previous studies 32 have shown that the echogenicity of the brainstem raphe nuclei is altered in patients with unipolar depressive disorders (UDD)  www.nature.com/scientificreports www.nature.com/scientificreports/   www.nature.com/scientificreports www.nature.com/scientificreports/ compared to healthy controls. Our finding of a negative association between brainstem volumes and depressive symptoms (low mood, anhedonia and total depressive score) in a large population cohort adds to this evidence.
Further, previous studies 30,33,34 have suggested that seasonal changes in serotonin (5-hydroxytryptamine; 5-HT) expression, which is mainly synthesized by several nuclei of the midbrain and pons such as dorsal raphe nucleus and locus coeruleus 31 could be the molecular mechanism that drives this correlation. They found that individuals with seasonal affective disorder showed higher cerebral serotonin transporter binding in winter, but not in summer, compared to healthy controls, and this change in serotonin transporter binding was positively associated with severity of depressive symptoms. Together, these results support the notion of the role of the brainstem in regulating related-mood processing.
In addition, the result of the association of mood or depressive symptoms including low mood and anhedonia with season in only females before correction for multiple comparisons is consistent and supported by several previous studies 4,5,35 in which females were found to have more hospital admissions due to winter depression and also to report higher prevalence of depression or depressive symptoms during shorter photoperiod days in the winter months compared to males. Little is known about the mechanisms underlining sex-related differences in   Table 5. Associations between brainstem substructure volumes and depressive symptoms in all participants, females and males. Negative binomial regression coefficients (b), robust standard error (SE) for association between brainstem substructure volumes (corrected for age, ethnicity, TBV, living area and Townsend deprivation index) and low mood, anhedonia, tenseness, tiredness and total depressive score in (a) all participants, (b) females and (c) males. Significant associations (p < 0.05) are shown in bold. www.nature.com/scientificreports www.nature.com/scientificreports/ seasonality of mood though this phenomenon has been widely investigated. One possibility is that the sex-related differences in seasonal variation in mood could be due to the differences in cortico-limbic mood regulation network, which includes the hippocampus, amygdala, prefrontal and anterior cingulate cortices and anterior thalamic nuclei, between males and females 36,37 . Interestingly, the features of subgenual anterior cingulate cortex (sgACC), which have been shown to elevate the metabolic activity in the presence of depression, dysfunction in mood disorders are different between males and females, and females exhibit higher levels of reactivity compared to males 37 . Therefore, it is possible that the corticolimbic network of mood regulation in females is more affected by photoperiod than males, leading to impact the mood status during the year between sexes. Another possibility could be explained by the difference in cortisol hormone and inflammatory stress responses, which have been linked to the prevalence of depression 8,38 . It has been shown that females have greater cortisol and inflammatory stress responses are more sensitive to depressed mood when inflammation is present 39,40 . Together, these studies suggest neuroanatomical and/or hormonal sex-related differences that could be the mechanisms underlining seasonal variation in mood between sexes.
The current study design has three limitations. First, our study was cross sectional in which participants were measured only once rather than at different times over the year, therefore the brainstem volumes measured represent inter-individual variance not change. Making a causal statement about seasonal change of an individual brainstem volume would require a longitudinal study. Second, depressive symptom scores were taken from questions about feelings over the previous two weeks and may be subject to recall bias and also gender biases in reporting mood. Third, we included all data available in the January 2017 brain imaging data release. This means that we included participants who may have medical or psychiatric issues related to their brain such as stroke, Alzheimer's disease, congenital or acquired structural brain defects.
To conclude, our study is the first to demonstrate that brainstem volumes fully mediate the seasonal variability of depressive symptoms. We further showed that this mediating effect is only present in females. This finding advances our understanding of brainstem morphology and suggests it may be an important neural substrate in the pathophysiology of seasonal mood disorders. This finding adds to the evidence supporting the role of photoperiod on brain structural plasticity which will have implications for future investigations of changes in mood associated with human exposure to variations in natural and artificial light. Methods participants. From 2006From -2010,655 participants aged 37-73 years were recruited into the UK Biobank cohort. Participants attended one of 22 assessment centres across the UK and completed a range of lifestyle, demographic, health and mood questionnaires, cognitive assessments and physical measures 41 , and subsequently brain imaging at a single centre between 2014 and 2016. More details can be found on the UK Biobank online data showcase (http://biobank.ctsu.ox.ac.uk/crystal/label.cgi). The 10,103 participants aged between 45 and 79 years (mean = 62.4, SD = 7.4) in the January 2017 brain imaging data release were included in this cross-sectional study. Sixty-nine participants were excluded from the study because of issues with their T1 weighted MRI structural images. Out of 10,034 participants, 745 participants were excluded because they did not complete their mood measures in the two weeks prior to the scanning.
All UK Biobank participants gave written, informed consent. UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee (11/NW/03820). This research was conducted using the UK Biobank Resource under Application Number 24089 (PI Waiter). All UK Biobank methods were performed in accordance with the UK regulations (https://www.ukbiobank.ac.uk/gdpr/). This work makes use of an open access MRI database of images. As a UK based multicentre trial each of the participating sites are compliant with MHRA guidelines for clinical MRI and participants imaged accordingly.
Environmental variable (Photoperiod). Photoperiod in hours of daylight on the day of scan was derived from the latitude and longitude information of the location of residence for each participant. Photoperiod in hours was calculated by subtracting sunset from sunrise on the day of scan.

Mood variable.
Mood outcomes composed of scores reflecting the frequency of low mood, anhedonia, tenseness and tiredness over the two weeks before the assessment. Participants were asked to indicate how often they experience these depressive symptoms including low mood, anhedonia, tenseness and tiredness over the   Table 6. Mediation analyses examining the relationship between photoperiod and both low mood and anhedonia in females via whole brainstem, midbrain, pons and medulla volumes. Negative binomial regression coefficients (b), robust standard error (SE), bootstrapping confidence interval (CI), lower level and upper level confidence interval (LL and UL) for the mediation analysis of brainstem substructure volumes on the relationship between photoperiod and depressive symptoms. Significant associations (p < 0.05) are shown in bold.
Volumetric analysis and segmentation. Volumetric processing and segmentation were performed using a development version of the FreeSurfer v6.0 software package (http://surfer.nmr.mgh.harvard.edu), with brainstem segmentation 44 . We chose FreeSurfer for segmentation due to its good reproducibility in brainstem segmentations compared to other methods 45 . FreeSurfer was used to process the data including averaging volumetric T1 weighted images, motion correction, transformation to Talairach image space, nonuniform intensity normalization for intensity inhomogeneity correction, removal of non-brain tissues using hybrid watershed, and segmentation of subcortical volumetric structures; white matter and deep grey matter [46][47][48] . FreeSurfer was used to segment the brainstem subfield volumes (medulla oblongata, pons, superior cerebellar peduncle and midbrain). Briefly, 39 MRI scans were manually delineated to highlight the whole brainstem, together with manual labelling of brainstem structures in 10 MRI scans from in vivo T1 weighted images (1 mm resolution) 44 . The manual delineation and labelling from in vivo scans were combined together to build an atlas of brainstem structures with a new robust Bayesian inference algorithm to detect local variations in MRI contrast. For each subject, volumetric data for these four brainstem structure volumes was calculated using the software's automatic Bayesian segmentation technique 47,49 . Brain volumes including total brain volume (TBV) and intracranial volume (ICV) were also calculated by FreeSurfer using the Talairach transformation matrix created from the registration of normalisation and MNI atlas 50 . All segmentations for the brainstem were visually checked for errors. No manual interventions were performed on the data.
Statistical analysis. Statistical analyses were conducted using SPSS version 24, with an alpha for all analyses of p = 0.05. We tested the seasonal pattern of photoperiod (as a continuous measure of day length) and we found that it follows a sinusoidal pattern. No further transforms were applied. To investigate the association of photoperiod with depressive symptoms including low mood, anhedonia, tenseness, tiredness and total depressive symptoms score, a negative binomial regression model was used. Likelihood ratio tests for these depressive scores showed that over-dispersion was greater than 1, i.e. their variance was greater than their mean 51 . We investigated the seasonality of brainstem volumes using a cosinor generalized regression analysis with both sine and cosine functions and month as the time variable 52 . Sine and cosine transformations of the month of scan were calculated using the formulas: where M = month of scan (integer number from: 1 to 12). We assessed whether the seasonal pattern of the brainstem is sinusoidal, by comparing a model including sine and cosine month transformations and the covariates of age and TBV with models excluding sine and cosine month transformations. We determined two specific criteria for indicating the significance of seasonality or improved model fit and these were (1) significance of sine and/or cosine (cosinor) terms (p < 0.025), with amplitude significantly greater than zero and (2) lower Akaike Information Criterion (AIC) for the model including the cosinor terms 4,52 . The amplitude of the cosinor model (or curve) was calculated as: where β and γ are cosine and sine generalized regression coefficients respectively. Finally, the Acrophase (φ; peak of cosinor model) in month of scan was calculated from: 12 tan ( / ) 2 1 1 φ = * γ β * π + − Pearson (bivariate) correlations between photoperiod and brainstem substructure volumes as well as age, and total brain volume (TBV) were performed, with significance levels Bonferroni-corrected for multiple comparisons and set at p = 0.003. To investigate the predictability for each of these independent variables for the brainstem subfield volumes, single linear regression models for each were created. There were significant correlations between brainstem subfield volumes and age, total brain volume, ethnicity, living area (urban or rural)