Introduction
Affective temperament describes the behavior, attitudes, and emotional orientation of a person based on the interaction of genetic and environmental factors. Akiskal proposed five major affective temperament dimensions: depressive temperament is characterized by sensitivity to suffering, self-denying, and striving to devote themselves to others and confirm to social norms; hyperthymic temperament shows upbeat and excessively energetic traits; cyclothymic temperament tends to shift between intense high and low emotions; anxious temperament shows exaggerated worries while irritable temperament can be expressed as anger, frustration, and skeptical traits (Akiskal & Akiskal, Reference Akiskal and Akiskal2005a, Reference Akiskal and Akiskal2005b; Akiskal, Akiskal, Haykal, Manning, & Connor, Reference Akiskal, Akiskal, Haykal, Manning and Connor2005; Rovai et al., Reference Rovai, Maremmani, Rugani, Bacciardi, Pacini, Dell'Osso and Maremmani2013). The affective temperaments have been considered as subsyndromal (trait-related) manifestations and commonly the antecedents of affective disorders (Kesebir, Gundogar, Kucuksubasi, & Tatlidil Yaylaci, Reference Kesebir, Gundogar, Kucuksubasi and Tatlidil Yaylaci2013; Rihmer, Akiskal, Rihmer, & Akiskal, Reference Rihmer, Akiskal, Rihmer and Akiskal2010; Serafini et al., Reference Serafini, Pompili, Innamorati, Fusar-Poli, Akiskal, Rihmer and Tatarelli2011; Solmi et al., Reference Solmi, Zaninotto, Toffanin, Veronese, Lin, Stubbs and Correll2016). For example, depressive and anxious temperaments are linked to a lack of antidepressant response in patients with mood disorder (De Aguiar Ferreira, Vasconcelos, Neves, & Correa, Reference De Aguiar Ferreira, Vasconcelos, Neves and Correa2014) and composite affective temperament score (cyclothymic + depressive + irritable − hyperthymic) are high-risk factors for suicidal acts or ideation (Pompili et al., Reference Pompili, Baldessarini, Innamorati, Vazquez, Rihmer, Gonda and Girardi2018; Rihmer et al., Reference Rihmer, Rozsa, Rihmer, Gonda, Akiskal and Akiskal2009; Tondo, Vázquez, Sani, Pinna, & Baldessarini, Reference Tondo, Vázquez, Sani, Pinna and Baldessarini2018; Vazquez, Gonda, Lolich, Tondo, & Baldessarini, Reference Vazquez, Gonda, Lolich, Tondo and Baldessarini2018).
It has been implicated that prefrontal cortex and limbic structures are associated with fundamental dimensions of affective temperaments (Hatano et al., Reference Hatano, Terao, Hayashi, Hirakawa, Makino, Mizokami and Shimomura2019; Whittle, Allen, Lubman, & Yucel, Reference Whittle, Allen, Lubman and Yucel2006). At present, in major depressive disorder (MDD) patients, neural correlates of the affective temperaments and their relationship with depressive symptom dimensions are yet unclear. A possible neural key correlate might be the insula, a brain hub that anatomically has connections with the prefrontal cortex, the limbic system, and the thalamus (Namkung, Kim, & Sawa, Reference Namkung, Kim and Sawa2017). Multiple lines of evidence support the role of the insula in not only subjective feeling states (Critchley, Wiens, Rotshtein, Ohman, & Dolan, Reference Critchley, Wiens, Rotshtein, Ohman and Dolan2004; Namkung et al., Reference Namkung, Kim and Sawa2017) and cognitive control (Diener et al., Reference Diener, Kuehner, Brusniak, Ubl, Wessa and Flor2012; Uddin, Reference Uddin2014), but also certain temperament constructs (Michalska et al., Reference Michalska, Feldman, Ivie, Shechner, Sequeira, Averbeck and Pine2018). The insula has also been considered as a pivot for internal and external information exchange and disruptions of the interoceptive–exteroceptive integration have been suggested as core symptoms of depression (Harshaw, Reference Harshaw2015).
Previous neuroimaging studies have indicated that the insula is related to the key symptoms in MDD (Manoliu et al., Reference Manoliu, Meng, Brandl, Doll, Tahmasian, Scherr and Sorg2013; Serafini et al., Reference Serafini, Pompili, Romano, Erbuto, Lamis, Moraschi and Bozzao2017; Sprengelmeyer et al., Reference Sprengelmeyer, Steele, Mwangi, Kumar, Christmas, Milders and Matthews2011). For example, altered functional connectivity (FC) of the insular cortex has been associated with the severity of depressive symptoms (Manoliu et al., Reference Manoliu, Meng, Brandl, Doll, Tahmasian, Scherr and Sorg2013; Wang et al., Reference Wang, Wu, Chen, Xu, Li, Li and Wang2018) and altered blood oxygenation level-dependent (BOLD) signal in the insula has been found to be correlated with a higher level of hopelessness (Serafini et al., Reference Serafini, Pompili, Romano, Erbuto, Lamis, Moraschi and Bozzao2017; Wiebking et al., Reference Wiebking, Duncan, Tiret, Hayes, Marjanska, Doyon and Northoff2014). It should be noted that different parts of the insular cortex exhibit variable cytoarchitecture features, connectivity, input–output relations, and therefore different divisions of labor (Craig, Reference Craig2009, Reference Craig2010; Namkung et al., Reference Namkung, Kim and Sawa2017; Tian & Zalesky, Reference Tian and Zalesky2018). The posterior regions (granular region: an isocortical region with six differentiated layers) receive thalamocortical inputs and have a role in somatosensory and motor integration; whereas the anterior regions (agranular region: an isocortical region with a relatively undifferentiated layer II and layer III, and lack of layer IV that contains stellate granule cells), which have reciprocal connections to limbic regions (e.g. the anterior cingulate cortex, the amygdala, and the ventral striatum) and the dorsolateral and ventromedial prefrontal cortex (for a review see Namkung et al., Reference Namkung, Kim and Sawa2017), have been implicated in the integration of interoception (Craig, Reference Craig2009), subjective feelings (Craig, Reference Craig2009; Damasio, Reference Damasio2003; Kesebir et al., Reference Kesebir, Gundogar, Kucuksubasi and Tatlidil Yaylaci2013), and motivations (Naqvi & Bechara, Reference Naqvi and Bechara2009). For example, the mid-posterior insula is activated when subjects receive painful stimulation and plays a greater role in regulating homeostatic states (Craig, Reference Craig2009, Reference Craig2010; Menon & Uddin, Reference Menon and Uddin2010; Uddin, Nomi, Hebert-Seropian, Ghaziri, & Boucher, Reference Uddin, Nomi, Hebert-Seropian, Ghaziri and Boucher2017). The anterior insula plays a role in the salience network, which is important for monitoring the saliency of external inputs and internal brain events. Thus far, whether different insular subregions contribute to specific depressive symptom dimensions is not clear.
Because of its key anatomical position, compared with the amygdala, which is responsible for impulsive and automatic emotional processing, the insula tends to introduce subjective feelings state into cognition and motivation (Bechara, Reference Bechara2005; Dolan, Reference Dolan2002; Namkung et al., Reference Namkung, Kim and Sawa2017). It seems that the insula or its FC could be more related to the affective temperaments, which identify the stable biological or inherited ‘core’ of an individual's emotional response tendencies and personality of integration of emotional and cognitional activity (Jabbi et al., Reference Jabbi, Kippenhan, Kohn, Marenco, Mervis, Morris and Berman2012; Rihmer et al., Reference Rihmer, Akiskal, Rihmer and Akiskal2010, Reference Rihmer, Gonda, Torzsa, Kalabay, Akiskal and Eory2013). Previous studies have revealed that a combination of certain affective temperaments is related to suicide risk in MDD (Pompili et al., Reference Pompili, Innamorati, Gonda, Serafini, Sarno, Erbuto and Girardi2013, Reference Pompili, Innamorati, Gonda, Erbuto, Forte, Ricci and Girardi2014, Reference Pompili, Baldessarini, Innamorati, Vazquez, Rihmer, Gonda and Girardi2018). Moreover, insula hyperactivity has been found to be associated with anxiety-related temperamental traits (Liu, Taber-Thomas, Fu, & Perez-Edgar, Reference Liu, Taber-Thomas, Fu and Perez-Edgar2018b; Stein, Simmons, Feinstein, & Paulus, Reference Stein, Simmons, Feinstein and Paulus2007). Thus, the insula may affect certain affective temperaments, and the dysregulated insula-temperament covariation, which represents an alteration in a synergistic effect of emotion and personality, might contribute to certain emotional–cognitive integration-related depressive symptoms in MDD patients. Thus far, whether different insular subregions contribute to specific affective temperaments, leading to specific depressive symptoms, has not been investigated. Answering to the question may be of value in exploring the etiology of MDD and provide a target for neurostimulation therapy of depression.
In this study, we aimed to examine whether spontaneous activity or FC of the insula is related to certain affective temperaments and which insular subregion contributes to the association. Also, we explored whether the covariation of insula-temperament is impaired in patients with MDD and related to depressive symptoms. It has been suggested that there is a multidimensional-to-multidimensional relationship between psychopathological subtypes, clinical manifestation subtypes, and biological data (Kang, Bowman, Mayberg, & Liu, Reference Kang, Bowman, Mayberg and Liu2016; Moser et al., Reference Moser, Doucet, Lee, Rasgon, Krinsky, Leibu and Frangou2018; Perlis et al., Reference Perlis, Giles, Buysse, Thase, Tu and Kupfer1997). Unlike previous studies focusing on variations of univariate analyses wherein either a single variable (subscale score) is treated as computationally independent or a global approach that treats the set of variables as a whole (total score of a scale or questionnaire) is used, we adopted a multivariate method, a partial least square regression (PLSR), to identify associations between a set of predictor variables (the resting-state activity of insular subregions) and a set of response variables (the dimensions of affective temperaments). The PLS method is especially appropriate when the predictor variables are highly interdependent or multicollinear (Abdi, Reference Abdi2010; Krishnan, Williams, Mcintosh, & Abdi, Reference Krishnan, Williams, Mcintosh and Abdi2011). We hypothesized that some distinctive affective-temperament profiles would be associated with certain patterns of insular activity in both medication-free MDD patients and healthy controls. Then, in the MDD patients, we examined whether the altered covariation between the insula-activity profile and the affective-temperament profile was associated with depressive symptoms, and if so, which symptom dimension it would contribute to. To avoid the effect of pharmacological treatments and comorbidity on the current research, we only included patients who had no comorbidity and took no medication within 3 months prior to participating in this study.
Materials and methods
Participant
Twenty-nine patients with MDD (mean age = 26.7 ± 6.0), diagnosed according to the Structured Clinical Interview for DSM-IV (SCID-DSM-IV) Patient Edition, were recruited from the Guangzhou Huiai Hospital, Guangzhou, China. Each of the patients had a score of at least 21 on the 24-item Hamilton Depression Rating Scale (HDRS) (Hamilton, Reference Hamilton1960). None of them had any other comorbid psychiatric or neurological diseases. Twenty-five of the MDD patients never received anti-depressant treatment (i.e. medication-naïve) and the rest four MDD patients took no medication within 3 months prior to participating in this study (i.e. medication-free).
Fifty-eight healthy control participants (mean age = 27.9 ± 5.9) were screened with the SCID-DSM-IV Non-Patient Edition to confirm the lifetime absence of Axis I illnesses. The healthy control participants had no history of psychiatric illness in any two lines of first- to third-degree biological relatives.
All participants gave their written informed consent to participate in this study. The procedures of this study were approved by the Independent Ethics Committee (IEC) of the Guangzhou Huiai Hospital. The investigation was carried out in accordance with the World Medical Association (2013) Declaration of Helsiniki-Ethical Principles for medical research involving human subjects.
Measures
Hamilton Depression Rating Scale
A 24-item HDRS was used to rate the severity of patients’ depression by probing seven factors of depressive symptoms: anxiety/somatization (physical and somatic anxiety, general and gastrointestinal somatic symptoms, hypochondriasis and insight); weight loss; cognitive disturbance (guilt, suicide, agitation, depersonalization and derealization, paranoid symptoms, and obsessional symptoms); diurnal variation; retardation (depressed mood, work and interests, slowness of thought, speech and activity, apathy, and loss of libido); sleep disturbance (difficulty in falling asleep, waking during the night, and waking in early hours of the morning and unable to fall asleep again); and hopelessness (helplessness, hopelessness, and worthlessness) (Hamilton, Reference Hamilton1960; Rabkin & Klein, Reference Rabkin and Klein1987). The Chinese version 24-item HDRS has been widely used and has demonstrated adequate validity in studies of clinical and non-clinical populations in China (Leung, Wing, Kwong, Lo, & Shum, Reference Leung, Wing, Kwong, Lo and Shum1999; Shi et al., Reference Shi, Hu, Dong, Gao, An, Liu and Sun2008; Tu et al., Reference Tu, Fang, Cao, Wang, Park, Jorgenson and Kong2018).
Temperament Evaluation of Memphis, Pisa, Paris, and San Diego
The Temperament Evaluation of Memphis, Pisa, Paris, and San Diego (TEMPS-A) is a 110-item ‘yes’ or ‘no’ answer self-rated questionnaire that assesses the five temperaments on five scales: depressive, cyclothymic, hyperthymic, irritable, and anxious temperaments (Akiskal & Akiskal, Reference Akiskal and Akiskal2005a, Reference Akiskal and Akiskal2005b; Akiskal et al., Reference Akiskal, Akiskal, Haykal, Manning and Connor2005). The TEMPS have been widely used in populations of different countries and has shown adequate internal consistency (Akiskal & Akiskal, Reference Akiskal and Akiskal2005b; Akiskal et al., Reference Akiskal, Akiskal, Haykal, Manning and Connor2005; Elias et al., Reference Elias, Kohler, Stubbs, Maciel, Cavalcante, Vale and Carvalho2017; Kawamura et al., Reference Kawamura, Akiyama, Shimada, Minato, Umekage, Noda and Akiskal2010; Lin et al., Reference Lin, Xu, Miao, Ning, Ouyang, Chen and Akiskal2013; Pompili et al., Reference Pompili, Baldessarini, Innamorati, Vazquez, Rihmer, Gonda and Girardi2018).
Magnetic resonance imaging data acquisition
Magnetic resonance imaging (MRI) was conducted at the Department of Radiology, Guangzhou Huiai Hospital, China, on a 3.0-Tesla MRI system (Achieva X-series Scanner; Philips, Medical Systems, Best, The Netherlands) with an eight-channel SENSE head coil. The resting-state functional MRI (rs-fMRI) was acquired using a gradient-echo echo-planar-imaging sequence (TR = 2000 ms, TE = 30 ms, flip angle = 90°, matrix = 64 × 64, field of view = 220 mm × 220 mm, number of slices = 33, slice thickness = 4 mm with interslice gap = 0.6 mm, voxel size = 3.4 mm × 3.4 mm × 4.6 mm). The rs-fMRI scanning lasted 8 min and consisted of 240 time points. Participants were instructed to relax and to remain still with their eyes closed during the rs-fMRI scanning. A high-resolution structural image was acquired with a three-dimensional T1-weighted turbo field-echo sequence (TR/TE = 8.2/3.7 ms, flip angle = 7°, acquisition matrix = 256 × 256, slice thickness = 1 mm, voxel size = 1 mm × 1 mm × 1 mm, and number of slices = 188). Earplugs were used to attenuate scanner noise and a foam pillow and extendable padded head clamps were used to restrain the head motions of participants. No participant was excluded due to excessive head motion (1.5 mm in translation or 1.5° in rotation) during rs-fMRI scanning.
MRI data preprocessing
The rs-fMRI data were processed and analyzed using the SPM 12 software (http://www.fil.ion.ucl.ac.uk/spm/) with FC toolbox v17 (CONN, www.nitrc.projects/conn) (Whitfield-Gabrieli & Nieto-Castanon, Reference Whitfield-Gabrieli and Nieto-Castanon2012). The preprocessing pipeline included participant motion estimation and correction, structural segmentation and normalization [resampling to a voxel size of 2 mm × 2 mm × 2 mm in the standard Montreal Neurological Institute (MNI) space], ART-based functional outlier detection and scrubbing, and functional spatial smoothing with an 8 mm Gaussian kernel. Before the first-level analysis, a denoising step (linear regression and bandpass filtering) was conducted to remove possible confounds including the BOLD signal from the white matter and cerebrospinal fluid, realignment parameters (six motion parameters and six first-order temporal derivatives), and scrubbing parameters (maximum interscan movement and identified invalid scans; the framewise-displacement values for all subjects were below 0.3) (Power, Barnes, Snyder, Schlaggar, & Petersen, Reference Power, Barnes, Snyder, Schlaggar and Petersen2012).
Fractional amplitude of low-frequency fluctuations (fALFF) in subregions of the insula
The fALFF, which is defined as the fractional sum of the amplitudes within the low-frequency range (0.01–0.08 Hz) divided by the sum of amplitude across the entire frequency range (0–0.25 Hz) (Zou et al., Reference Zou, Zhu, Yang, Zuo, Long, Cao and Zang2008), was calculated first for each GM voxel. The participant-level voxel-wise fALFF maps were then further standardized by subtracting the mean whole-brain voxel-wise fALFF from the participant-level voxel-wise fALFF and dividing it by the standard deviation. The fALFF has been considered as an index for detecting spontaneous brain activities (Zou et al., Reference Zou, Zhu, Yang, Zuo, Long, Cao and Zang2008; Zuo et al., Reference Zuo, Di, Kelly, Shehzad, Gee, Klein and Milham2010).
The regions of interest for the insular subregions were obtained directly from the Human Brainnetome Atlas (http://atlas.brainnetome.org/bnatlas.html). In this atlas, the insula was first extracted from the Desikan–Killiany (DK) atlas, and the connectional architecture was then mapped with probabilistic tractography using diffusion MRI. The insula was symmetrically divided into six subregions in each hemisphere with anatomical connectivity patterns by calculating the similarity/dissimilarity between the connectivity architecture (Fan et al., Reference Fan, Li, Zhuo, Zhang, Wang, Chen and Jiang2016). The insular subregions (Fig. 1) contain the hypergranular insula (L1 and R1), the ventral agranular insula (L2 and R2), the dorsal agranular insula (L3 and R3), the ventral dysgranular and granular insula (L4 and R4), the dorsal granular insula (L5 and R5), and the dorsal dysgranular insula (L6 and R6) (Fan et al., Reference Fan, Li, Zhuo, Zhang, Wang, Chen and Jiang2016; Wang et al., Reference Wang, Wu, Chen, Xu, Li, Li and Wang2018). The fALFF values were extracted from each of the 12 insular subregions using the Marsbar toolbox (http://marsbar.sourceforge.net/).
Fig. 1. PLS association analyses in all the participants. (a) The second pair of latent variables (LVs) from the partial least squares regression (PLSR) analysis between spontaneous neural activity in the insular subregions measured by the fALFF and affective temperaments measured by the TEMPS (Dep: depressive; Cyc: cyclothymic; Hyp: hyperthymic; Irr: irritable and Anx: anxious). Left: Bootstrap ratios (BSR) of saliences (weights) for spontaneous neural activity in each insular subregion; right: Bootstrap ratios (BSR) of saliences for each subscale of TEMPS. (b) Illustration of the insular subregions from the Human Brainnectome Atlas (http://atlas.brainnetome.org/bnatlas.html). L1 and R1: hypergranular insula; L2 and R2: the ventral agranular insula; L3 and R3: the dorsal agranular insula; L4 and R4: the ventral dysgranular and granular insula; L5 and R5: the dorsal granular insula; L6 and R6: the dorsal dysgranular insula. L, left; R, right. (c) The associations between the insular-fALFF profile and the TEMPS profile were significant for not only all the participants (left panel), but also for either the major depressive disorder (MDD) patients or the healthy control participants (right panel).
FC of the insular subregions
We used a Global Brain Connectivity (GBC) method to characterize each insular subregion's full range FC with voxel-wise resolution (Cole, Yarkoni, Repovs, Anticevic, & Braver, Reference Cole, Yarkoni, Repovs, Anticevic and Braver2012; Wang et al., Reference Wang, Zhen, Song, Huang, Kong and Liu2016). The GBC of a voxel was generally defined as the averaged FC of the voxel to the rest of the voxels in the whole brain or within a predefined mask (Wang et al., Reference Wang, Zhen, Song, Huang, Kong and Liu2016). First, the GBC of each insular subregion was computed as the averaged FC (Pearson correlation) of each insular-subregion voxel to all the voxels in a whole brain gray matter mask (the bilateral insula was excluded from the mask). Then, participant-level GBC maps were transformed to z-score maps using Fisher's z-transformation to yield normally distributed values. Finally, the GBC values of each participant were entered into a two-sample t test for group comparison and a PLS regression for exploring the association with the TEMPS dimensions.
Statistical analysis
Statistical analyses were performed using R 3.4.0. With univariate analyses, we tested differences in scale scores, insula-subregional fALFF values, and insula-subregional GBC values between groups. With partial least squares regression analyses, we investigated patterns of associations between insular-subregional spontaneous activity or GBC and affective temperaments. With partial correlation analyses, we investigate the association of insula-temperament covariation with the depressive symptoms.
The PLS analyses were conducted using the R Package PLS (https://cran.r-project.org/web/packages/pls/index.html) and Morpho (https://cran.r-project.org/web/packages/Morpho/index.html). PLS combines a principal component analysis-style dimensionality reduction with linear regression and finds the components from predictor variables (e.g. fALFF values or GBC values in each insular subregion: matrix X) that have maximum covariance with the response variables (e.g. the TEMPS subscale scores of participants: matrix Y). The PLS components are ranked by covariance between predictor and response variables, and the first few PLS components (PLS1, PLS2, PLS3, etc.) provide the optimal low-dimensional representation of the covariance between the higher dimensional data matrices (Abdi, Reference Abdi2010; Krishnan et al., Reference Krishnan, Williams, Mcintosh and Abdi2011; Vertes et al., Reference Vertes, Rittman, Whitaker, Romero-Garcia, Vasa, Kitzbichler and Bullmore2016). To express the saliences relative to the X measures and the Y measures, the original matrices X and Y are projected onto their respective saliences. This creates pairs of latent variables (LVs) – which are linear combinations of the original variables – that are called X and Y scores. A pair of LVs (weighted scores) reflects a relationship between the predictor and the response variable. The significance of the covariance of the components was tested by comparing it with the distribution of covariance arising from random permutation tests (1000 times). The weight (salience) of each variable indicates and ranks the contribution of this variable to a PLS component. A bootstrap procedure (5000 times) was used to test the reliability (bootstrap ratio > 2) of each variable in the significant PLS component (Krishnan et al., Reference Krishnan, Williams, Mcintosh and Abdi2011; McIntosh & Lobaugh, Reference McIntosh and Lobaugh2004).
To identify the association between the resting-state activity of the insular cortex and the affective temperaments, a 12 (six insular subregions in each hemisphere) by 87 (number of all participants) matrix of the insula-activity measures and a 5 (TEMPS subscales) by 87 matrix of the affective-temperament measures were entered into a PLSR analysis. Then, the PLS-insula scores and the PLS-TEMPS scores from the insula-TEMPS PLSR were obtained for each participant. Finally, partial correlations were conducted to examine whether the insula-temperament covariation (the mean score averaged across the PLS-insula scores and the PLS-TEMPS scores) contributed to the depressive symptom dimensions.
Results
Comparisons in scale scores, fALFF, and FC between MDD participants and healthy controls
There was no significant difference in age, sex, or educational level between the MDD group and the healthy control group. Compared to the healthy controls, the MDD patients had higher TEMPS depressive scores, higher TEMPS cyclothymic scores, higher TEMPS irritable scores, higher TEMPS anxious scores, and lower TEMPS hyperthymic scores (all p values <0.002) (Table 1).
Table 1. Demographic and clinical characteristics in patients with major depressive disorder and healthy controls
s.d., standard deviation; CI, confidence interval; HC, healthy controls; MDD, major depressive disorder; TEMPS, Temperament Evaluation of Memphis; HDRS, Hamilton Depression Rating Scale. The HDRS was not rated for healthy controls.
a Analyses were conducted between the two patient groups by t tests for normally distributed variables and χ2-tests for categorical variables (two-tailed).
The fALFF value in the L1 subregion (left hypergranular insula), the L5 subregion (left dorsal granular insula), the R2 subregion (right ventral agranular insula), and the R4 subregions (right ventral dysgranular and granular insula) were lower in the MDD patients than those in the healthy controls (uncorrected p values were 0.030, 0.030, 0.006, and 0.040; two-tailed). All the p values did not survive correction for multiple comparisons with the Benjamini–Hochberg standard false discovery rate (FDR) method (see online Supplementary Fig. S1).
The GBC value of the L3 subregion (left dorsal agranular insula) was higher in the MDD patients than that in the healthy controls (uncorrected p = 0.038), and the group difference in the GBC value of the R1 subregion (right hypergranular insula) was marginally significant (uncorrected p = 0.055). Both the p values did not survive correction for multiple comparisons with the Benjamini–Hochberg standard FDR method (see online Supplementary Fig. S2).
Multivariate associations between insular fALFF/GBC values and TEMPS scores
The PLSR between the fALFF values in the insular subregions and the TEMPS subscale scores yielded five sets of LVs capturing the insula-temperament associations, ordered by the size of the explained variance (73.80, 22.36, 3.26, 0.33, and 0.25%). Each component (LV pair) was the linear combination of the weighted insula-fALFF scores that most strongly predicted weighted affective-temperament scores. Only the second PLS component (PLS2) was significant (permutation test, p = 0.018) (Fig. 1a). The bootstrapping test showed that the fALFF value in the L3 subregion, L6 subregion, and R2 subregion of the insula reliably contributed to the PLS2-insula LV (Fig. 1b); the depressive temperament (T1), hyperthymic temperament (T3), irritable temperament (T4), and anxious temperament (T5) reliably contributed to the PLS2-temperament LV. Thus, for all the participants, the insula-TEMPS PLS2 identified a profile of spontaneous activity of the left anterior dorsal insula and right anterior ventral insula that positively predicted the depressive, anxious, and irritable temperaments and negatively predicted the hyperthymic temperament (age, sex, and educational years were adjusted) (left panel of Fig. 1c). Moreover, the association was also significant for either the MDD patients (age, sex, educational years, age of onset, and illness duration were controlled as nuisance covariates in the MDD group) or the healthy participants (age, sex, and educational years were controlled in the healthy group) (right panel of Fig. 1c), supporting our hypothesis that there may be a common pattern of insular-TEMPS covariation between MDD patients and healthy controls.
The PLSR between the GBC values in the insular subregions and the TEMPS subscale scores yielded five sets of LVs representing the insula-temperament associations, ordered by the size of the explained variance (77.46, 16.27, 5.32, 0.86, and 0.09%). However, none of the PLS components was significant (permutation test, all p values >0.05).
Compared with the healthy participants, the MDD patients showed higher PLS2 insula-fALFF scores, significantly higher PLS2 affective-temperament scores, and significantly enhanced covariation between the insula-fALFF profile and the TEMPS profile (Fig. 2).
Fig. 2. Comparisons in the PLS2 insula-fALFF scores, PLS2 TEMPS scores, and PLS2 Insula-TEMPS covariation scores between the MDD patients and the healthy participants. ***p < 0.001 (two-tailed).
To examine whether the pair of the insular LV and the TEMPS LV was related to depressive symptoms, in the MDD patients, with the covariates (including age, sex, educational years, age of onset, and illness duration) controlled, we correlated the PLS2 insula-fALFF scores, the affective-temperament (TEMPS) scores, and their covariation index scores with each depressive symptom dimension. The results showed that only the sleep disturbance dimension was significantly correlated with the PLS2-insula scores and the insula-temperament covariation (Fig. 3; FDR-corrected p values <0.05).
Fig. 3. Associations of the sleep disturbance scores with the PLS2 insula-fALFF scores, PLS2 TEMPS scores, and PLS2 insula-TEMPS covariation scores in MDD patients. Age, sex, educational years, age of depression onset, and illness duration were controlled as covariates.
Further inspection of each item of the sleep-disturbance dimension showed that both the middle-insomnia (waking during the night) score and the late-insomnia (waking in the early hours of the morning and unable to fall asleep again) score, but not the early-insomnia (difficulty in falling asleep) score, were associated with both the insula profile and the affective-temperament profile (Fig. 4; all FDR-corrected p values <0.05). Moreover, both the middle-insomnia score (r = 0.521, p = 0.009, FDR-corrected p = 0.028) and the late-insomnia score (r = 0.527, p = 0.008, FDR-corrected p = 0.029), but not the early-insomnia score (r = 0.387, p = 0.062, FDR-corrected p = 0.070), contributed to the positive correlation between the sleep-disturbance symptoms and the insula-temperament covariation (Fig. 4). Thus, for the medication-free MDD patients, the higher the covariation between the spontaneous insular activity and the affective temperaments, the more severe the sleep disturbances (especially for middle insomnia and late insomnia) were.
Fig. 4. Associations of the early insomnia, middle insomnia, and late insomnia scores with the PLS2 insula-fALFF scores, PLS2 TEMPS scores, and PLS2 insula-TEMPS covariation scores in MDD patients. Age, sex, educational years, age of depression onset, and illness duration were controlled as covariates.
Discussion
This study, for the first time, investigated the relationship between the resting-state activity of the insular subregions and the affective temperaments. Additionally, in a sample of medication-free MDD patients, we explored whether the insula-temperament covariation was associated with the depressive symptom dimensions. Using a multivariate analysis technique, we identified a profile of spontaneous activity of the insular cortex (left anterior dorsal agranular–dysgranular insula and right anterior ventral agranular insula) that was associated with an affective-temperament (depressive, irritable, anxious, and less hyperthymic) profile. In the MDD patients, the insula-temperament covariation was significantly correlated with sleep disturbance symptoms, especially middle insomnia and late insomnia.
As far as we know, this is the first study investigating the relationship between insular subregional activity and distinct affective-temperament dimensions. In both the MDD patients and healthy control participants, we found that a profile of spontaneous activity of the anterior insular cortex (increased fALFF in the left dorsal agranular–dysgranular insula and decreased fALFF in the right ventral agranular insula) was associated with a profile of affective temperaments (more depressive, irritable, anxious, and less hyperthymic), indicating a role of the anterior insular subregions in the affective temperament profile, a composite cognitive-affective personality characterized by being pessimistic, sensitive to suffering, self-denying, easy to exaggerate anxiety, skeptical and anger/frustration traits. Moreover, the negative results of PLSR analysis between the insular-subregional GBC values and the affective temperament dimensions seem to suggest that it is the functional covariation within the anterior insular subregions, not the insular-subregional GBC, contributes to a specific affective temperament profile. The anterior dorsal insula is considered as a cognitive-control-related region and the anterior ventral insula is considered as a socio-emotion-related region (Kurth et al., Reference Kurth, Eickhoff, Schleicher, Hoemke, Zilles and Amunts2010). Greater diversity within the anterior insula has been associated with higher scores on measures of positive effect, self-efficacy, emotion recognition, and motor dexterity (Tian & Zalesky, Reference Tian and Zalesky2018). Moreover, it has been implicated that objective sensory information that is passed from the posterior insula is re-presented in the anterior insula by integration of the primary emotional, cognitive, and motivational information from the limbic system and prefrontal cortex, forming the subjective or interoceptive feeling states (Craig, Reference Craig2010; Namkung et al., Reference Namkung, Kim and Sawa2017). Thus, it seems that the anterior insula underlies the subjective emotional response that is manifested in emotional personality or affective temperaments. Note that PLS2 of the insular fALFF-temperament association explained only 22.3% of the total variance, suggesting that there may be other unknown neural activities that modulate affective temperaments in addition to the spontaneous activity in the insular subregions.
The role of the insula in depression has been highlighted in several meta-analyses and reviews (Harshaw, Reference Harshaw2015; Menon & Uddin, Reference Menon and Uddin2010; Otte et al., Reference Otte, Gold, Penninx, Pariante, Etkin, Fava and Schatzberg2016; Su et al., Reference Su, Cai, Xu, Dutt, Shi and Bramon2014). Some research reported increased ALFF or fALFF in the insular cortex of patients with MDD (Liu et al., Reference Liu, Ma, Song, Fan, Wang, Lv and Wang2015; Yu et al., Reference Yu, Liu, Wang, Huang, Jie, Sun and Zhang2017) but others reported reduced fALFF in the insula (Fitzgerald, Laird, Maller, & Daskalakis, Reference Fitzgerald, Laird, Maller and Daskalakis2008; Liu et al., Reference Liu, Ma, Yuan, Song, Jing, Lu and Wang2017; Su et al., Reference Su, Cai, Xu, Dutt, Shi and Bramon2014). The inconsistent results may be explained by several factors including whether patients were comorbid with other disorders, medicated, suffered from a first episode, or were chronic patients (Su et al., Reference Su, Cai, Xu, Dutt, Shi and Bramon2014). In contrast to previous studies focusing on variations of the univariate analysis, our study extends previous findings by identifying an increased covariation between the spontaneous neural activity of three anterior insular subregions and four affective temperament dimensions in the medication-free MDD patients, and the association was related to only the sleep disturbance symptom dimension when the potential covariates (including age, sex, educational years, illness duration, and the age of onset) were controlled, suggesting that the insula-temperament covariation, which represents a biological-personality interaction, may have a synergistic effect on sleep disturbances in MDD patients.
Human sleep neuroimaging studies support the role for the insular cortex in pathologic sleep associated with both depression and insomnia (Cheng, Rolls, Ruan, & Feng, Reference Cheng, Rolls, Ruan and Feng2018; Liu et al., Reference Liu, Guo, Lu, Tang, Fan, Wang and Liu2018a; Perico et al., Reference Perico, Skaf, Yamada, Duran, Buchpiguel, Castro and Busatto2005; Wang et al., Reference Wang, Han, Zhao, Du, Zhou, Liu and Li2019). For example, it has been found that increased ALFF in the right anterior insula is related to sleep disturbance scores in MDD patients with insomnia complaints (Liu et al., Reference Liu, Guo, Lu, Tang, Fan, Wang and Liu2018a). In primary insomnia people, more negative sleep-onset discrepancy has been associated with the higher relative regional cerebral metabolic rate for glucose in the right anterior insula during non-rapid eye movement sleep (Kay et al., Reference Kay, Karim, Soehner, Hasler, James, Germain and Buysse2017). The results of this study are consistent with the conclusion that the anterior insula is involved in sleep disturbances in MDD patients. Furthermore, the above-mentioned insula-temperament covariation made contributions to the middle and late insomnia in the MDD patients, suggesting that the left anterior dorsal insula and right anterior ventral insula might be involved in regulating the sleep–wake circle, which is more likely through dysregulation of physiological reactivity covering the sympathetic and parasympathetic action (Chouchou et al., Reference Chouchou, Mauguiere, Vallayer, Catenoix, Isnard, Montavont and Mazzola2019; Park et al., Reference Park, Palomares, Woo, Kang, Macey, Yan-Go and Kumar2016). The sympathetic–parasympathetic control spatially distributes from the posterior to anterior insula and regulates cardiovascular and neural activity during sleep and waking (Chouchou et al., Reference Chouchou, Mauguiere, Vallayer, Catenoix, Isnard, Montavont and Mazzola2019; Park et al., Reference Park, Palomares, Woo, Kang, Macey, Yan-Go and Kumar2016). Moreover, an objective–subjective signal processing also distributes spatially from the posterior to anterior insula, which regulates integration of interoceptive–exteroceptive information (Craig, Reference Craig2009, Reference Craig2010; Namkung et al., Reference Namkung, Kim and Sawa2017). Thus, alterations in the profile of anterior insular activity might cause disturbances in the sleep–wake circle via the dysregulation of sympathetic–parasympathetic related emotional–cognitive components, which contributes to neural vulnerability to sleep loss in the medication-free MDD patients. The insula-temperament covariation identified in this study may be a potential biopsychological marker underlying the middle and late insomnia in MDD.
In summary, our study indicated that covariation between spontaneous activity in the anterior insula (especially the left dorsal insula and the right ventral insula) and the composite depressive–irritable–anxious temperament (characterized by being pessimistic and self-denying, skeptical and easy to get angry, and prone to exaggerate worries) is associated with sleep disturbance symptoms (middle and late insomnia) in medication-free patients with MDDs. Because of the cross-sectional design, we cannot confirm whether the alterations in insular activity are neurodevelopmental or a consequence of depression onset. This study shed light on the etiology of depression and provides a practical method for exploring the relationship among multivariate measures in mental disorders.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291719003647
Acknowledgments
This work was supported by the National Natural Science Foundation of China (81601168), the Medical Scientific Research Foundation of Guangdong Province of China (B2018063), the Guangdong Natural Science Foundation of China (2015A030313800), and the Shandong Provincial Natural Science Foundation, China (ZR2017LH036).
Author contributions
HW, SS, and CW designed the study and wrote the first draft of the manuscript. CW carried out data analyses. HW, YZ, QZ, SS, JD, HP, JGZ, and JPZ conducted data collection. All authors contributed to and approved the final version of the manuscript.
Conflict of interest
All authors declare no conflicts of interest.
