Serum BDNF levels correlate with regional cortical thickness in minor depression: a pilot study

Serum brain-derived neurotrophic factor (BDNF) reflects state changes in mood disorders. But its relation to brain changes in depression has rarely been investigated in humans. We assessed the association between serum BDNF, cortical thickness, or gray matter volume in 20 subjects with a minor depressive episode and 40 matched healthy subjects. Serum BDNF positively correlated with cortical thickness and volume in multiple brain regions in the minor depression group: the bilateral medial orbitofrontal cortex and rostral anterior cingulate cortex, left insula, and cingulum, right superior frontal gyrus, and other regions—regions typically affected by major depression. Interestingly, these correlations were driven by subjects with first episode depression. There was no significant association between these imaging parameters and serum BDNF in the healthy control group. Interaction analyses supported this finding. Our findings point to a specific association between serum BDNF and magnetic resonance imaging parameters in first-episode minor depression in a region- and condition-dependent manner. A positive correlation between serum BDNF and structural gray matter estimates was most consistently observed for cortical thickness. We discuss why cortical thickness should be preferred to volumetric estimates for such analyses in future studies. Results of our pilot study have to be proven in future larger-scale studies yielding higher statistical power.

Scientific RepoRtS | (2020) 10:14524 | https://doi.org/10.1038/s41598-020-71317-y www.nature.com/scientificreports/ attempted to relate sBDNF to brain imaging parameters. Some region-of-interest-based analyses revealed a positive correlation between sBDNF and the volume of the hippocampus in healthy subjects 10,11 . Others found no correlation of hippocampal and amygdala volumes with sBDNF, neither in healthy subject 12 nor in subjects with mood disorders 13 or schizophrenia 12 . One study did not find any relation of cortical thickness across the brain to sBDNF in healthy subjects and subjects with recurrent MDD 14 , and another reported a negative correlation in patients with schizophrenia 12 .
Histologically, parameters such as gray matter volume and cortical thickness measured by magnetic resonance imaging (MRI) in vivo represent distinct brain features 15 . Gray matter volume is, mathematically, a product of thickness and area, where area has more weight 16,17 . In ontogenesis, cortical surface area is defined by the number of neuronal columns and cortical thickness is defined by the number of neurons within the columns. Moreover, these brain features are related to distinct sets of genes 17 . In neuroimaging studies, the histological underpinnings of imaging parameters are rarely taken into account.
In this perspective, studies on the correlation between sBDNF and MRI parameters lack a systematic approach, investigating different diseases using different analysis methods, with potentially improper parameters. Since cortical thickness and volume are distinct measures of the brain 16 , we performed a systematic whole-brain structural MRI study correlating sBDNF levels to these imaging parameters estimated with FreeSurfer. Due to the neurotrophic effects of BDNF we generally hypothesized a positive correlation between sBDNF and cortical estimates, modified due to the reduction of sBDNF and regional gray matter volume/cortical thickness in depressive disorders. Differences between subjects with or without a history of depression were assessed in an explorative analysis.

Methods
Subjects. Twenty subjects satisfying DSM-IV criteria 5 for minor depressive episode were selected from the database of the population-based LIFE-Adult study 18 . In accordance with Structured Psychiatric Interview for DSM-IV Disorders (SKID), every subject had one to four depressive symptoms for at least two weeks, with depressed mood or loss of interest being one of them. Forty healthy volunteers from the same study were free from depressive symptoms or cognitive impairment and were matched at a 1:2 ratio by sex and age to the subjects with minor depression. The study was carried out in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the University of Leipzig. All participants gave written informed consent.
Mild and major neurocognitive disorders 2-4 (formerly known as mild cognitive impairment and dementia) were excluded according to DSM-5 diagnostic criteria for mild Neurocognitive Disorder (NCD). These criteria require: (A) presence of subjective cognitive disturbance; (B) objective cognitive decline 1-2 standard deviations (SD) below sex-and age-adjusted norms in at least one of five cognitive domains; (C) preserved activities of daily living according to the Activities of Daily Living scale (ADL); (D) absence of delirium and major psychiatric illness (E).
Cognitive testing was performed using the German version of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD)-plus test battery and a Stroop test. Specific tests or subtests were assigned to each DSM-5 cognitive domain. With Trail Making Test (TMT)-A and Stroop neutral we evaluated attention, with TMT-B/A and Stroop incongruent/neutral executive function. The word list subtest from the CERAD-plus test battery was used for assessment of learning and memory, figure drawing test was used for the visuo-construction/ perception domains. Participants' scores were compared to normative values adjusted for sex, age, and education, obtained from the Basel memory clinic (www.memor yclin ic.ch). A mean deviation from the norms was calculated for each cognitive domain if this domain was assessed with more than one test. " BDnf measurement. Blood samples were withdrawn from subjects by venipuncture, between 7:25 and 10:45 in the morning, after an overnight fasting. Serum was prepared using the standard operating procedures. In brief, samples were left for 45 min for clotting, followed by a centrifugation step (10 min, 2,750 g, 15 °C). Samples were then filled in straws (CryoBioSytems IMV, France) by an automatic aliquoting system (DIVA, CryoBioSytems IMV, France). To minimize freeze-thaw cycles, samples were sorted in a cryogenic work bench (temperatures below − 100 °C) and automatically stored in tanks with a coolable top frame in the gas phase of liquid nitrogen (Askion, Germany) and stored for analysis 18 . Serum BDNF was assessed using an ELISA kit manufactured by R&D Systems (Wiesbaden, Germany) as previously described 4 .
MR images were preprocessed using the standard pipeline recon-all. After normalization and skull-stripping of the T1-weighted images, cortical tissue boundaries were reconstructed and transformed to a subject-specific surface mesh. The distance between pial and gray/white matter surfaces at each vertex location of the mesh was calculated in order to obtain cortical thickness measurements 20 . Based on Desikan-Killiany's cortical parcellation, regional cortical thickness and gray matter volume was extracted separately for the several brain regions in each hemisphere and averaged for the analysis. All images were visually checked for misplaced tissue boundaries and manually corrected if necessary.

participants' characteristics.
Subjects with minor depressive episode were not significantly different from control subjects in terms of age, sex, body mass index (BMI), and amount of white matter hyperintensities as rated using the Fazekas scale. Levels of sBDNF were also comparable, i.e. not significantly different, between both groups (see Table 1).
cortical thickness. Cortical thickness, and gray matter volume, were not statistically different between both groups (Supplementary Table 1 and 2), whereas sBDNF correlated with imaging parameters. At p < 0.05, we observed a positive correlation between sBDNF and cortical thickness only in the minor depression group as illustrated in Fig. 1 and Table 2. On the uncorrected level, sBDNF positively correlated with cortical thickness in the left medial orbitofrontal, the rostral and caudal anterior cingulate cortex, posterior and isthmus cingulate cortex, and the insula and precuneus. In the right hemisphere we observed positive correlations between sBDNF and cortical thickness in the medial orbitofrontal, superior frontal, rostral anterior cingulate cortex, superior parietal cortex, temporal pole and transverse temporal, as well as with the supramarginal, postcentral and pericalcarine gyrus (Fig. 1). No regions remained significant after the FDR correction for multiple comparisons p FDR < 0.05 (see Table 2). In healthy subjects, contrary to our hypothesis, correlations tended to be negative (Table 2). On the uncorrected level (p < 0.05), we observed significant negative correlations between sBDNF and cortical thickness of the bilateral cuneus, right lingual gyrus, and insula. Positive correlations were observed only for the left caudal anterior cingulate cortex and right entorhinal region. Interestingly, a negative correlation between sBDNF and thickness in the right cuneus was significant at the p FDR < 0.05 threshold.
Between-group interaction effects were significant for correlations between sBDNF and cortical thickness in the bilateral cuneus and insula, left medial orbitofrontal cortex, precuneus, isthmus and posterior cingulate cortex, as well as the right pericalcarine and lingual gyrus, pars opercularis and superior parietal lobule. In all these cases, we observed positive correlations in subjects with minor depression and near-zero or negative correlations in healthy participants ( Table 2). cortical volume. Correlations between sBDNF and volumetric data are illustrated in Fig. 2 and Table 3. The regional correlation pattern was similar between the volumetric and cortical thickness data (see Figs. 1 and 2). In subjects with minor depression at p < 0.05, sBDNF correlated positively with bilateral medial orbitofrontal and pericalcarine cortical volume. Additionally, in the left hemisphere, we observed positive correlations between sBDNF and volumes of the left rostral, caudal, and anterior cingulate, as well as the posterior cingulate cortex, precuneus, fusiform, entorhinal, and postcentral gyrus. In the right hemisphere, sBDNF positively correlated with volumes of the isthmus cingulate, lateral orbitofrontal, precentral cortex, pars orbitalis of the inferior frontal gyrus, superior parietal and superior temporal gyrus, as well as with the temporal pole and supramarginal gyrus.
In healthy subjects, negative correlations at p < 0.05 were found between sBDNF and volumes of the right superior parietal cortex, right cuneus, lingual and fusiform, as well as with the left postcentral, and lingual gyrus. None of these correlations remained significant after FDR correction. www.nature.com/scientificreports/ Interaction effects were significant for correlations of sBDNF with volumes of the left posterior and rostral anterior cingulate cortex, precuneus, postcentral, lingual gyrus, as well as for correlations with right medial orbitofrontal, middle temporal, lingual, superior parietal, superior temporal and supramarginal volumes. Similar to cortical thickness, positive correlations characterized the minor depression group, and negative ones the healthy control group.
Subgroup analysis. Finally, we performed a post hoc subgroup analysis to investigate potential differences between persons with and without a history of depression (n = 8 vs n = 12). The results are depicted in Supplementary Tables 3-6 and Fig. 2. Interestingly, cortical thickness was larger in subjects without history of depression (Supplementary Table 3).
Further analysis showed that correlation between cortical thickness and sBDNF in the minor depression group was driven by subjects without a history of depression. Correlation between sBDNF and right medial orbitofrontal cortical thickness in this subgroup remained significant after FDR correction. Interaction effects between both subgroups were significant for the left lateral orbitofrontal gyrus, right medial orbitofrontal gyrus, right pars triangularis of the inferior frontal gyrus, the rostral anterior cingulate cortex, and superior frontal gyrus. In all regions, correlations in subjects with first-episode minor depression were significantly higher than in subjects with recurrent depression.
Gray matter volume correlated both positively and negatively with sBDNF in both subgroups. However, none of these correlations remained significant after FDR correction. Interaction effects were significant for correlation between sBDNF and left middle temporal, right pericalcarine, and right posterior cingulate volumes. In all these cases, negative correlations were observed in subjects with first-episode minor depression and positive correlations in subjects with recurrent episode.

Discussion
To our knowledge, this is the first structural MRI study investigating the correlation between sBDNF and gray matter parameters in minor depression. At the uncorrected level (p < 0.05) positive correlation was detected in multiple depression-related regions in subjects with minor depressive episode, but not in the control group. The respective interaction effects were significant. The post hoc analysis revealed that correlations with cortical thickness were driven by subjects with first-episode minor depression, while volumetric data showed mixed effects.  Imaging phenotype matters-cortical thickness should be preferred to cortical volume in depression. Following a recent publication from the field of imaging genetics 17 , it is reasonable to argue that thickness and volume estimates are not interchangeable also in clinical investigations. In the FreeSurfer estimations gray matter volume is a product of cortical area by cortical thickness 16,17 . Since cortical area has larger inter- Table 2. Correlation between cortical thickness and serum BDNF in subjects with minor depression and healthy controls. BDNF brain derived neurotrophic factor, 1-tailed p-values are reported, FDR p value is derived using the Benjamini-Hochberg procedure, Fisher's z-test for interaction analysis was performed only for significant correlations. Regions significantly correlating with sBDNF at p < 0.05 are marked as bold.  16 . Moreover, the FreeSurfer algorithm has shown a tendency to misestimate cortical volume 22 . Histologically, cortical area is defined by the number of neuronal columns, while cortical thickness by the number of neurons and their connections within the column 23 . The change of clinical state from euthymic to depressed is unlikely to alter the number of neuronal columns, and, therefore, cortical area and volume. Furthermore, sBDNF is a dynamic measure 24,25 . In light of the neurotrophic hypothesis, a number of neuronal connections is thought to decrease due to deficiency of neurotrophic factors in depression 6 . Therefore, we suggest that cortical thickness is much more useful for clinical studies compared to cortical volume to examine state changes in depression. Accordingly, we will further discuss results for this parameter only. correlation between serum BDnf and regional cortical thickness seems to be relevant in early minor depressive states. In this study, sBDNF correlated positively with cortical thickness of numerous brain regions in minor depression. Though none of these correlations remained significant after the rigorous FDR correction, the total number of correlations was substantially higher than the expected at 5% false-positive rate (3.4 significant results are expected out of 68). Moreover, note that correlation coefficients reached relatively high values, explaining a high amount of variability in the data. sBDNF correlated positively with the thickness of the bilateral medial orbitofrontal cortex and rostral anterior cingulate, left cingulate cortex, insula, and right superior frontal gyrus. These regions are typically activated in functional MRI paradigms that assess emotion regulation in healthy subjects 26,27 and in major depression 22,27,28 , and show changes in structure and glucose metabolism in MDD as revealed by systematic and quantitative meta-analyses 29 and histopathological studies with glial and later neuronal alterations [30][31][32] .

Region of interest
Whether this correlation is specific to minor depression as compared to major depression remains to be investigated. Some considerations may be drawn from other studies of cortical thickness and sBDNF. Cortical thinning was robustly detected in patients with first episode major depression in a large scale study of ENIGMA consortium 33 , as well as smaller studies [34][35][36] . In minor depression we did not observe these effects 37 . Serum BDNF has been unchanged in first episode major depression 38 and in minor depression 4 . One study has reported a positive correlation between sBDNF and hippocampal volume in first episode major depression in a region-ofinterest analysis 39 .  www.nature.com/scientificreports/ An earlier study, investigating the relation of sBDNF to cortical thickness in patients with recurrent major depression, did not show such a correlation 14 . These patients had a recurrent severe (major) depressive disorder, which likely exhausted BDNF resources. Our previous meta-analysis investigating the effects of electro-convulsive therapy on BDNF in such patients showed no response of sBDNF to therapy 40 . In patients with less severe depressive disorder sBDNF responds much better to anti-depressive treatment 9 . In line with this argument, our minor depression subtype analyses revealed that the correlation between sBDNF and cortical thickness was driven by subjects without a history of depression. In summary, a significant positive correlation between sBDNF and cortical thickness might be only relevant in early depressive states and might indicate a compensatory mechanism, because it was neither detected in healthy controls nor in minor depressive states with a history of depression. With respect to healthy controls, we replicated previous findings 14 .
Regions correlating with sBDNF in our study substantially overlapped with regional cortical thinning in MDD shown in a recent very powerful meta-analysis 33 . Here, the bilateral medial orbitofrontal cortex, fusiform gyrus, insula, rostral anterior and posterior cingulate cortex and, unilaterally, the left middle temporal gyrus, right inferior temporal gyrus, and right caudal anterior cingulate cortex were significantly thinner in the MDD group than in healthy controls. The obtained effect sizes for cortical thinning were relatively small in this metaanalysis (Cohen's d − 0.13 to 0.49) 33 . This suggests that large sample sizes are required for such studies.
We have shown here for the first time that in minor depression the correlation of sBDNF with cortical thickness is significantly different from controls. The subtype analysis suggested that this correlation was mainly driven by subjects with first episode depression. These data provide insight into the early mechanisms of depression with a focus on neuroendocrine mechanisms, possibly indicating an early compensatory mechanism, similar to other diseases 41,42 . Furthermore, it also shows that no universal positive correlation between brain measures and BDNF exists. Similarly, animal studies have shown that correlations between brain BDNF and sBDNF is very much region-and strain-specific 40 .
Whilst cortical thickness is a relatively straightforward measure, biological processes, reflected by sBDNF, are less understood. It has been long supposed that sBDNF reflects cortical and hippocampal secretion of BDNF 43,44 . However, a recent study has shown that sBDNF is instead derived from megakaryocytes 45 and not from the brain. Therefore, mechanisms linking brain and serum BDNF are yet to be further examined.
Both cortical thickness 46 and sBDNF 9,47 are reduced in MDD. In our previous reports comparing subjects with minor depression, we found neither sBDNF differences 4 nor differences in cortical measures 37 . The evidence we provide here is correlational and by no means causative. However, we might have observed an early sign of Table 3. Correlation between normalized gray matter volume and serum BDNF in subjects with minor depression and healthy controls. BDNF brain derived neurotrophic factor, 1-tailed p values are reported, FDR p value is derived using the Benjamini-Hochberg procedure, Fisher's z-test for interaction analysis was performed only for significant correlations. Regions significantly correlating with sBDNF at p = 0.05 are marked as bold. www.nature.com/scientificreports/ neurotrophic function in early subclinical depression, not yet visible on the biomarker or whole-brain level. This observation should be confirmed by future studies.

Limitations
Our study has a number of limitations. Firstly, due to the unexpectedly low prevalence of minor depression in our sample, which originated from a large population-based study with approximately 2,500 participants, our sample size was relatively small and we had to include subjects with and without a history of depression. Because our study is the first one in minor depression, an a priori power analysis was not feasible. A previous study investigating the correlation between sBDNF and hippocampal volumes in early major depression used a comparable sample size n = 25 39 , suggesting we had enough statistical power. To increase power, we matched our sample on a 1:2 basis to healthy controls. Secondly, we did not have precise information on the duration of minor depressive state burden, which might be an additional parameter of interest for further analyses. Although the ELISA kits used for sBDNF quantification were not optimal according to a recent publication 48 , these kits were purchased prior to this publication. We used a whole-brain approach guaranteeing data-driven statistics in both cortical thickness and volume. Although only a minority of results survived correction for multiple comparisons using the FDR procedure, we underlined validity of our findings by interaction analyses demonstrating specificity compared to healthy subjects. Future studies are necessary to prove our pilot findings in larger and preferably multi-centric cohorts. Finally, we did not use the voxel-wise estimation, because we wanted to make our data comparable to the recent meta-analysis by the ENIGMA consortium.

Summary
In this study, we observed a positive correlation between serum BDNF measurements and structural gray matter estimates in minor depression. The correlation between sBDNF and imaging parameters was region-and condition-dependent. These findings require verification in larger samples considering a-priori power estimations and controlling for the duration of depression burden. Furthermore, our analysis suggests that cortical thickness is a more suitable structural parameter for biomarker studies than gray matter volume, at least in studies of depression.