Brain Structure and Subclinical Symptoms: A Dimensional Perspective of Psychopathology in the Depression and Anxiety Spectrum

Human psychopathology, emotions, cognitions, and behaviours are just as complex as the human brain itself and their underlying neurobiology.

Yet still, clinical psychiatry and research depend on categorical diagnoses defined in ICD-10 or DSM-5. These categories rarely capture each individual’s symptom profile and are thus not able to distinguish between neurobiologically distinct disease entities [1, 2]. For example, the “major depression” diagnostic category is composed of several different, sometimes even opposing symptoms [3]. On the other hand, minor symptoms in otherwise healthy persons cannot be categorised and therefore not be accounted for in healthy study participants, in at-risk subjects or in the psychiatric history of a patient, because the cut-off is set to separate clinical pathology from “sanity.” One example is the “soft bipolar spectrum,” which is mostly disregarded in ICD-10, and DSM-5, but has contributed greatly to individual case history comprehension [4-6]. Of course, the categorical approach seems practical considering training, clinical and scientific application, epidemiologic enquiries, health insurance issues, and communication between professionals.

But there is a major downside for clinical practice, because if the current psychiatric nosology does not sufficiently capture the individual symptom constellation and pathogenesis, information is lost and treatment response may be suboptimal [7, 8], which impedes “precision psychiatry.” In typical psychiatric case-control studies, cohorts are mostly chosen according to presence or absence of DSM-5 or ICD-10 diagnoses and therefore do not specifically consider the neurobiological substrate and aetiology of the particular psychiatric illness. This is also an obstacle for research on neurobiological markers in at-risk subjects with subclinical symptoms. As a consequence of this lack of neurobiologically driven research, pharmacological and other research on mental disorders has seen a recent decline due to the lack of identified biomarkers and specific therapeutic targets [9, 10], emphasising the need for a paradigm shift towards a fully dimensional scope of psychiatric disorders.

The idea of dimensional psychiatry is old though. One of the major dimensional models introduced to the scientific community was the endophenotype concept of schizophrenia by Gottesmann and Shields [11, 12], covering the genetic foundation, endophenotype and reaction surface with intra-individual shifting by age. But the idea only slowly translated into research or clinical practice. The current versions of ICD-11 and DSM-5 have adopted some dimensional aspects but have refrained from making radical changes [13]. More ground-breaking efforts are the Hierarchical Taxonomy Of Psychopathology (HiTOP) for clinical practice [14] and the Research Domain Criteria (RDoC) by the National Institute of Mental Health [15] and the European “Roadmap for Mental Health Research” (ROAMER) [16] for scientific application.

There is considerable support for a dimensional approach with evidence for a neurobiological overlap of nominally distinct psychiatric disorders on several levels, starting with genes [17, 18] and spanning molecules [19], cells [20], brain structure [21, 22] and function [23, 24], as well as cognitive psychology [25]. And there is also evidence for alterations of brain structure and function associated with psychopathological symptoms in clinically healthy or at-risk patients not meeting the ICD or DSM cut-off [26]. Such findings open up new perspectives for understanding pathophysiology and early diagnosis of mental illnesses.

The aim of this review is to summarise the current evidence for this ongoing paradigm shift towards a neurobiologically founded, dimensional psychiatric nosology based on structural imaging findings (voxel-based morphometry [VBM], cortical thickness, and gyrification) in psychiatrically healthy subjects with minor depressive and anxiety symptoms. Studies covering this non-clinical spectrum are expected to expand our understanding of a dimensional continuum onto the neurobiological level.

Analysis of brain structure from MRI scans, especially using VBM, is a highly standardised and automated process applied in many studies across psychopathological dimensions [27]. These analyses are mostly analyses across the whole-brain and structural alterations and are considered to be relatively stable (although modifiable) over time [28]. In addition, multiple surface-based methods have emerged, including cortical thickness and gyrification, with the latter presumably being temporally stable from a very early age onwards [29, 30]. We aim to: (a) emphasise the great value of a dimensional scope for investigation of basic neurobiological links to common psychopathological symptoms; (b) draw conclusions about the pathogenesis of psychopathological symptoms from the findings in both the depressive and anxiety spectrum; and (c) identify problems and perspectives for further investigation of the underlying neurobiology of psychopathological symptoms in healthy participants.

We searched PubMed for the literature using structural imaging findings of associations of grey matter volume (GMV), cortical thickness, and gyrification with subclinical psycho(patho)logical symptoms in healthy participants in the spectra of (1) depressive symptoms and (2) symptoms of anxiety. We did not apply a filter for the year of publication and used a couple of search terms in “all fields” to cover the three imaging modalities of cortical volume, cortical thickness and gyrification: “subclin* depr* AND cortical volume,” “subclin* depr* AND cortical thickness,” “subclin* depr* AND gyrification,” “trait anx* AND cortical volume,” “trait anx* AND cortical thickness,” “trait anx* AND gyrification.”

Sixteen original research papers have been published over the last decade applying different measures of GMV or cortical thickness and subclinical depressive symptoms, sometimes focusing on subgroups (e.g., elderly subjects or females only). These studies are summarised in Table 1. The results are as heterogeneous as the chosen study protocols and raise several concerns that will be addressed later on. One very frequently discussed finding in the most recent studies is an association of hippocampal grey matter with subclinical depressive symptoms [31-34], mostly a negative correlation. Older studies specifically state no associations between subclinical depressive symptoms and hippocampal volume, though [35-37]. Another area of interest with reported structural alterations in some of these studies is the anterior cingulate cortex (ACC) [38], where an effect is shown mostly in women [39, 40], as well as less frequently in the posterior cingulate cortex and isthmus of the cingulate [36, 41, 42]. Less frequently and less consistently, other prefrontal areas have been associated with subclinical depressive symptoms, mostly with negative correlations: the superior and middle frontal gyrus [35, 37, 38], dorsomedial prefrontal cortex (dmPFC) in men [43] and the orbitofrontal cortex (OFC) [36, 38]. Superior and inferior temporal gyrus volume was positively correlated with subclinical depressive symptoms in 2 studies [44, 45] and negatively correlated in 2 other studies [36, 38].

Only 2 of the studies calculated cortical thickness and found inconsistent results of a negative correlation of symptoms with cortical thickness in precuneus [46] and positive correlation with cortical thickness in precuneus [42] in smaller groups (n = 73 and n = 43, respectively) of elderly participants (mean age 71, 69 years, respectively). Associations of gyrification with subclinical depressive traits in non-clinical subjects have not yet been explored.

Twenty studies have reported brain structural alterations associated with subclinical anxiety symptoms, summarised in Table 2.

Correlations with amygdala volume have been investigated in multiple studies, some showing positive [47-50], some negative [51-53] correlations of its (total) volume with subclinical anxiety. Other reported areas are the ACC [53, 54], insula and OFC [47, 52, 55, 56], hippocampus [48, 57, 58], parahippocampal cortex (positive GMV correlation [49, 59] and negative GMV correlation [53]), and precuneus (negative GMV correlation [60], positive GMV correlation with state measures [45] and [55]). One study showed a negative correlation of gyrification of the precuneus with trait anxiety [61].

Sporadically, studies imply that areas like the thalamus or hypothalamus have negative and positive GMV correlations [62] or lingual gyrus and temporal cortex have positive GMV correlations [63] and positive cortical thickness associations [47], inferior frontal cortex have negative GMV correlations [64], or left postcentral gyrus have negative GMV correlations [65].

Summarising and comparing these studies highlights several methodological challenges irrespective of the examined spectrum. At first it is crucial to consider age and gender of the chosen cohort, which seem to have an impact on measured correlations. It seems plausible to expect structural variations relevant to the pathogenesis of MDD or an anxiety disorder in adolescents or young adults, because the peak onset for both spectra is at an age between 30 and 40 years. Studies in geriatric cohorts often analysed the subclinical depressive spectrum (Table 1) and are more relevant for investigation of geriatric psychiatric hypotheses. Aging effects on the brain structure mediated by altered connectivity have to be taken into account though [66].

Some analyses in both spectra describe a structural effect of gender on areas associated with subclinical symptoms. Since there is plenty of evidence for genetically driven differential neurodevelopment in men and women, especially with regard to psychopathology [67], there should be more analyses addressing the effect of gender in larger samples. For example, healthy hippocampus function as well as functional decline in illnesses like Alz-heimer’s disease or major depression is considerably modulated by gender, which can be demonstrated clinically, functionally, and structurally, thus hinting at gender-specific neurogenesis [68].

Another important issue is the direction of correlations in these studies. There is limited basic research to mechanistically explain how higher or lower GMV in a certain area affects its function. Under pathological conditions including atrophy, lower GMV is commonly assumed to lead to decreased function, because it is mostly found in functionally impaired subjects, i.e. in patients or in the aging brain [69]. A positive association of subclinical symptoms with GMV as in some of the mentioned studies appears less intuitive. However, these findings merit particular attention. Higher GMV could be mediated by multiple structural or functional effects, including volume of neuropil, glial cells, or tissue water as well as differences in synaptic pruning [70-72], lack of inhibition, or even rapid changes in T1 scans, probably reflecting regional cerebral blood flow [73]. Some studies report worse task performance is associated with larger cortical volume [74, 75]. It is unclear whether there is a linear decreasing GMV curve for a certain brain area across a putative spectrum (i.e., from low depressive healthy over high depressive healthy to an MDD patient). Indeed, several recent studies using a correlational approach between subclinical or personality measures and brain structure have identified positive correlations [76-81]. There are several possible explanations to this, including a non-linear or “inverted U-shape curve” relationship, that would imply a positive correlation in the non-clinical part of the spectrum, but negative correlations in the pathological part of the spectrum. Another largely unexplored explanation might be related to resilience or compensation: relatively larger grey matter in non-clinical subjects might thus be an indicator of resilience to transition into a full clinical phenotype. This aspect highlights both a current problem in interpretation as well as a significant future potential of these studies in the non-clinical spectrum.

Moreover, the choice of the anatomical measure is crucial for the analysis of subclinical symptoms in healthy adults. Most of the studies performed morphometry of volume images using VBM, conventional volumetry or region-of-interest (ROI) atlas-based morphometry with FreeSurfer software (14 studies in the depressive spectrum, 18 in the anxiety spectrum). Only 2 studies investigated cortical thickness based on cortical surface estimations, using a vertex-wise measure of the distance between the cortical surface and the border between grey and white matter. Cortical volume on the other hand amalgamates aspects of cortical thickness, cortical surface, and folding. Grey matter changes can affect both measures in different ways, and it is recommended to use both techniques [82]. Yet, the interpretation of divergent results is not trivial, because the underlying neuroanatomical factors are incompletely understood. For example, a study in older adults indicated that cortical thickness was independent of the cortical neuron count, emphasising the role of glial components of the cortex [83]. Cortical thickness and surface area measurements were found to be genetically and phenotypically independent. While both thickness and area influenced volume measurements of cortical grey matter, volume was more closely related to surface area than cortical thickness [84].

One study in the subclinical anxiety spectrum measured the association of symptoms with gyrification [61]. This is an interesting structural feature for analysis, because cortical gyrification is assumed to take place prenatally and in the first months or years postnatally, remaining stable in adult life [29, 30]. However, more recent studies show age-related neurodevelopmental changes in gyrification depending on the method used for analysis [85-87], possibly mediated by changes in cortical connectivity throughout the lifespan [88-90]. Thus, to understand the connection between changes of gyrification and a liability for anxiety or depression, longitudinal studies in adolescents or young adults in relation to their age are needed. Innovative measures like cortical complexity or sulcal depth [91] extend the spectrum of analysis for gyrification and should be assessed in future studies.

Most of the cited studies used a multiple regression model to correlate subclinical symptoms with a structural feature accounting for several confounding factors, which has been a common statistical approach for a long time [92]. Another technique in some of the studies was to artificially divide the healthy cohort into two groups without or with very light symptoms and with more severe symptoms (but still subclinical). Creating a case-control setting like this seems reasonable when there is a meaningful difference of symptom severity within the cohort but might decrease statistical power due to dividing and matching of the sample.

Another major issue is the choice of self-report instruments. For the depressive spectrum, studies mostly used the BDI-II, which typically captures symptoms during the preceding 2 weeks, while CES-D, DASS, GDS, and SCL-90-R focus on 1 week prior to assessment. One might argue that a liability for or predisposition towards depressive mood as a trait also increases the probability for higher depressive state scores like these. But correlating state measures with brain structure might show volatile associations mediated by brain plasticity during stressful life events or even more transient changes like regional blood flow or regional blood volume. Trait measures appear more suitable to investigate predisposition towards certain symptoms.

In contrast, anxiety was mostly measured with the STAI as a trait anxiety measure, less frequently with other trait measures like the BIS, SPSRQ, and TCI or state measures like SCL-90-R and SAS accounting for symptoms of the preceding week.

The validity of the chosen questionnaire is of similar importance. The RDoC initiative (https://www.nimh.nih.gov/research-priorities/rdoc/index.shtml) makes recommendations about self-report measures reporting the neurobiologically distinct domain of negative valence containing reaction to loss and potential threat, which are the core concepts of the depressive/anxiety spectra. However, a current review argues that none of these suggested measures are specific to either loss or sustained threat [93], which necessitates a clear rationale for each included self-report instrument in a study, since there is little guidance from RDoC so far.

An important step in interpreting the findings from the subclinical spectrum is to place them in the context of dimensional continuum models, i.e. including phenotype expression across the full spectrum. Discussing the neurobiological plausibility of these subclinical findings is potentially conflicting, because our models of depression and anxiety are derived from findings in case-control studies, which eliminate associations with subclinical symptoms in the healthy control group by design. Still, there are some similarities in the frequently emerging brain areas in patients and healthy participants, which will be discussed (separately for each spectrum).

Major depression affects the structure, connectivity, and biochemistry of several, especially frontal brain areas implicated in cognitive control, emotion regulation, and memory processing [94]. Several studies have robustly shown evidence of grey matter reduction in the ACC [95-97]. The ACC is involved in multiple cognitive and affective functions, such as decision-making [98], empathy [99], conflict-monitoring [100], working memory, attention, and information-processing [101]. In only 5 of the above-cited studies in subclinical cohorts, did the ACC show positive or negative correlations with subclinical depressive symptoms, with the largest study (n = 810 participants) demonstrating a negative correlation of GMV in bilateral ACC with symptoms only in females [39]. This might point towards a gender-specific development of non-clinical depressive symptoms or even insignificant structural change and involvement of ACC in subclinical or transient depressive symptoms.

Further structural abnormalities in depressive patients occur in the insula, pallidum, thalamus, and hippocampus [102-104]. The anterior insula is connected to the inferior frontal cortex and ACC and is involved in social-emotional and cognitive networks [105, 106]. The thalamus also has multiple cortical connections and is linked to negative emotion-generating limbic structures, e.g. the amygdala [104]. These connections could underlie deficits in the top-down regulation of negative affects in MDD. The hippocampus is not only involved in memory processing [107, 108] but shows reduced volume in patients probably due to stress-related or/and repeated neurotoxic processes associated with cumulative exposure to stress and depressive symptoms [109, 110]. Of these areas, the hippocampus is also the most investigated structure with negative volume correlations with subclinical depressive symptoms, which is comprehensible considering the short temporal perspective of the applied self-report measures and the suggested stress dependence of hippocampal volume. However, some other studies explicitly state no associations with hippocampal volume, similarly to the observations regarding the amygdala.

Associations of subclinical symptoms of several prefrontal areas (dmPFC, OFC, superior and middle frontal gyri) and the pallidum are reported similarly to the findings in MDD patients. Divergent to the observations in clinical populations, structural associations in the thalamus and insula are not reported and therefore probably not relevant in subclinical cohorts.

Our review also identified areas associated with subclinical symptoms (but not in the MDD literature) like the precuneus and superior temporal cortex GMV; these could play a singular role in early development of depressive symptoms. One might speculate a relation between these areas and the default mode network (DMN), which has been identified in functional MRI studies and is also associated with the posterior cingulate cortex and medial PFC [111, 112]. The DMN is thought to mediate the psychological process of introspection and mental movement away from externally concentrated thoughts [113]. A majority of studies have demonstrated that the DMN is hyperactive in MDD, possibly leading to rumination as an important clinical feature of depressive patients [114, 115]. This observation is an interesting intersection with subclinical symptoms and might point out an early origin of depressive thoughts, probably even before emotion regulation is involved. However, since the structural substrate of the DMN as a functional construct is inconsistent in different studies, this idea needs further investigation.

There is also a study in untreated, first-episode, mid-life MDD patients showing increased cortical thickness in the right medial orbitofrontal gyrus, pars opercularis, rostral middle frontal gyrus, and supramarginal gyrus. Increased thickness of rostral middle frontal gyrus was negatively correlated with depression severity on the Hamilton Depression Rating Scale (HDRS) [116], indicating that increased thickness might be present in milder cases or that it might represent a compensation for inflammatory factors or other aspects of the pathophysiology of MD. This could explain the positive correlations observed in the subclinical context.

Anxiety-related behaviours and thoughts in patients and healthy persons are suggested to correspond to morphological differences and functional alterations in a left-lateralised circuit with higher GMV in the amygdala and anterior parahippocampal gyrus (PHG) and lower GMV in the OFC extending into the perigenual ACC [26, 117].

The amygdala as the key structure of this circuit mediates fear experience and conditioning, as well as saliency detection [118-120] and emotional memory [121] via its connections with the memory structures in the medial temporal lobe. This includes the hippocampus and anterior PHG, which encodes emotional stimuli [122] and places them in a context [123]. PHG and the ventrolateral PFC (vlPFC) are modulated by amygdala input during emotionally engaged learning contexts [124]. The connection of OFC and ACC with the amygdala has been implicated in fear conditioning and suppression [125-127]. Efficient emotion regulation depends on the balance between the PFC and amygdala activity; greater PFC activation reduces amygdala activation [128]; and deficits in emotion regulation reverse this hierarchy [129]. These functional observations are supported by structural findings of increased GMV in the left amygdala and anterior PHG and reduced GMV in OFC and ACC as well as altered WM microstructure in the uncinate fasciculus, cingulum, and other cortico-cortical tracts associated with poor emotion regulation and fear extinction [26]. Structural findings in amygdala, PHG, ACC, OFC, and other prefrontal areas in the majority of studies in this current review support this hypothesis indicating it to be a relevant mechanism in subclinical trait anxiety as well. But the direction of associations is not always as expected from defective fear extinction.

Apart from this mechanism, the studies in subclinical anxiety report alterations in sensory cortices like the somatosensory cortex, the auditory cortex and the visual cortex implicating changes in the processing of external stimuli. This corresponds with fMRI findings of reduced functional connectivity in these perceptual systems associated with high trait anxiety [130]. These findings are mostly not replicated in the studies of brain structure and trait anxiety and therefore seem to lack a structural correlate in most cases.

Again, precuneus and temporal areas are associated with subclinical anxiety symptoms similar to the observations in the depressive spectrum, hinting at possible neural overlaps with the DMN, although the structural overlap appears to be highly inconsistent. There are several functional and EEG studies stating a failure to synchronise DMN in healthy participants with high trait anxiety during resting state, probably reflecting a top-down cognitive control deficit [131]. A recent large meta-analysis showed a hypo-connectivity of the affective network with executive control network and the DMN and a decoupling of both networks in patients with anxiety disorders [132], which makes the structural finding in the subclinical cohorts relevant for the early development of a clinical anxiety disorder.

Summing up the neuro-mechanistic discussion, there is a lot of overlap between altered cortical structure associated with subclinical depressive or anxiety symptoms and findings in patients. The structural variation in the precuneus and temporal areas in subclinical symptoms in both spectra suggest functional and structural alterations of the DMN to be relevant in the subclinical symptom range. The significance of all these findings for the development of a clinical relevant mental illness has yet to be shown in longitudinal studies though. Regarding the methodical considerations, especially the types of structural and self-report measurement have to be chosen carefully and ideally consistent with a neurobiologically driven, dimensional scope of psychiatric nosology.

Taken together, these findings and progress in related future studies might pave the way for a fully dimensional understanding of depression and anxiety irrespectively of the changing categorical definitions of disorders. These efforts are in synch, but not dependent, on approaches like RDoC. Yet, they provide substantially larger opportunities for early intervention, adaptation of treatment programs, and thus moves towards individualised medicine or “precision psychiatry,” which takes into account also those subclinical phenotypes that so far have received insufficient attention in clinical studies.

The authors have no ethical conflicts to disclose.

The authors have no conflicts of interest to declare.

None of the authors received funding to create this paper.

All authors planned the review. B.B. created the first draft of the manuscript, C.G. and I.N. contributed to the finalisation of the manuscript.