Differentiating unipolar and bipolar depression by alterations in large‐scale brain networks

Abstract

Background

Misdiagnosing bipolar depression can lead to very deleterious consequences of mistreatment. Although depressive symptoms may be similarly expressed in unipolar and bipolar disorder, changes in specific brain networks could be very distinct, being therefore informative markers for the differential diagnosis. We aimed to characterize specific alterations in candidate large‐scale networks (frontoparietal, cingulo‐opercular, and default mode) in symptomatic unipolar and bipolar patients using resting state fMRI, a cognitively low demanding paradigm ideal to investigate patients.

Methods

Networks were selected after independent component analysis, compared across 40 patients acutely depressed (20 unipolar, 20 bipolar), and 20 controls well‐matched for age, gender, and education levels, and alterations were correlated to clinical parameters.

Results

Despite comparable symptoms, patient groups were robustly differentiated by large‐scale network alterations. Differences were driven in bipolar patients by increased functional connectivity in the frontoparietal network, a central executive and externally‐oriented network. Conversely, unipolar patients presented increased functional connectivity in the default mode network, an introspective and self‐referential network, as much as reduced connectivity of the cingulo‐opercular network to default mode regions, a network involved in detecting the need to switch between internally and externally oriented demands. These findings were mostly unaffected by current medication, comorbidity, and structural changes. Moreover, network alterations in unipolar patients were significantly correlated to the number of depressive episodes. Conclusion: Unipolar and bipolar groups displaying similar symptomatology could be clearly distinguished by characteristic changes in large‐scale networks, encouraging further investigation of network fingerprints for clinical use. Hum Brain Mapp 37:808–818, 2016. © 2015 Wiley Periodicals, Inc.

INTRODUCTION

Distinguishing bipolar and unipolar depression as early as possible is fundamental for correct clinical choices and consequently better short and long‐term outcomes. Up to 60% of bipolar patients are misdiagnosed as unipolar patients and only 20% receive the correct diagnosis within the first year what may considerably increase risks of switching to mania, suicide and poorer treatment responses [Hirschfeld et al., 2003]. In order to improve the actual situation, the development of new markers that support the differential diagnosis between these two disorders is mandatory [Cardoso de Almeida and Phillips, 2013; Phillips and Swartz, 2014].

Resting‐state functional connectivity (FC) analysis has extended our comprehension of large‐scale brain networks [Biswal, 2012; Biswal et al., 1995]. The representation of a major network involved in internally‐oriented processes, known as the default mode (DM), has opened up a new perspective to understand network intercorrelations [Fox et al., 2005; Raichle et al., 2001]. Since than many studies have highlighted the importance of the interaction between the DM network and the frontoparietal (FP), an externally‐oriented network involved in the effective accomplishment of cognitive tasks [Chadick and Gazzaley, 2011; Spreng et al., 2010]. The accuracy in performing tasks, such as the antisaccade task, was shown to involve the deactivation of DM regions and activation of the cingulo‐opercular (CO) regions [Polli et al., 2005]. Later, it was seen that the CO network was distinguished from the FP network by being sensitive to the need of adapting executive control and therefore necessary for cognitive flexibility [Dosenbach et al., 2007].

During a depressive episode, in which important cognitive limitations are generally present, aberrant deactivation in DM regions was shown in unipolar patients [Greicius et al., 2007; Pizzagalli, 2011; Sheline et al., 2009]. Also, in an important attempt to make sense of the connectivity alterations in patients with unipolar depression, a model was suggested in which an aberrant CO network weakly engage the DM by salient events [Menon, 2011]. Conversely, patients with bipolar disorder presented abnormal activation in regions typically involved in the FP network, according to a meta‐analysis [Chen et al., 2011]. A study investigating multiple networks in euthymic bipolar I patients also identified an aberrantly increased connectivity in the FP network [Lois et al., 2014]. If such network alterations could be trait‐related and sustained even in depressive phases, they would comprise important value in the differential diagnosis between bipolar and unipolar patients.

To our best knowledge the direct comparison of these candidate networks (namely FP, CO, and DM) between similarly depressed unipolar and bipolar I patients in contrast to well‐matched controls has not been presented in the literature. Using resting‐state fMRI, we intended to characterize specific alterations, which could perhaps contribute in the future for developing network fingerprints as imaging markers for the differential diagnosis of depression. Based on the literature, we expected specific alterations in bipolar patients to involve the FP network, whereas alterations in unipolar patients would comprise the CO and the DM networks. Lastly, we expected the network alterations to correlate with the history of depressive episodes (trait‐) and not to actual symptom scores (phase‐dependency).

METHODS

Study Participants

We recruited in‐ and out‐patients in treatment of an actual depressive episode at the psychiatric clinic of the university hospital. The inclusion criteria were patients between 18 and 60 years old, clinically well‐diagnosed with major depressive disorder or type I bipolar disorder according to ICD‐10 criteria. The clinical groups were carefully matched for age, gender, and education level. According to the goal of the study, we recruited well‐characterized patients with longer clinical assessments to reduce chances of including new and misdiagnosed bipolar patients. Therefore, all but six patients presented medication use for medical and ethical reasons. A higher proportion of bipolar patients used mood stabilizers and neuroleptics, while a higher proportion of unipolar patients used antidepressants. Due to these differences, medication use was tested as a covariate in between‐group analyses in accordance to the expected standards to explore the effects of medication variety and dose [total medication load composite in Hassel et al., 2008; Grotegerd et al., 2014; Magioncalda et al., 2015; Phillips et al., 2008]. The exclusion criteria comprised past or actual presence of other psychiatric or neurological diagnoses, organic causes of affective symptomatology, or the existence of MRI contraindications. Due to the high prevalence of comorbid anxiety disorders among patients with mood disorders, patients were not excluded if they presented anxiety disorder as a secondary diagnosis or comorbid personality disorders likewise. Also, based on the high prevalence of substance abuse among these patients, patients in abstinent over the last 5 years, who were never diagnosed with substance dependency according to ICD‐10, were not excluded. Potential effects of these comorbidities over the main results were explored with covariates‐of‐no‐interest in between‐group analysis. Healthy controls matching the age, gender, and the education levels of the patient sample were included in the analysis, so that demographic factors were controlled in the final results. The exclusion criteria for controls involved actual or previous medical conditions, including neurological and psychiatric diagnosis, actual or previous use of psychotropic substances, or the existence of MRI contraindications. All participants provided their written consent to the research protocol, which was approved by the local ethics committee.

Clinical Assessment

Patients were systematically investigated by an experienced psychiatrist blinded to the research question with standardized psychometric scales (Wechsler Adult Intelligence Scale Vocabulary Score, WAIS; Montgomery‐Åsberg Depression Rating Scale, MADRS; Beck Depression Inventory, BDI; Young Mania Rating Scale, YMRS) and with systematic registration of clinical parameters such as number of previous depressive episodes, age at illness onset, and duration of disorder. Clinical parameters were compared across groups (t or χ², two‐tailed P < 0.05, SPSS version 22, Chicago, IL).

Imaging Acquisition and Preprocessing

MR images were acquired with a 3T system (Magnetom TRIO, Siemens Healthcare, Erlangen, Germany). Whole‐brain resting‐state fMRI was performed using an 8‐channel head coil (eyes open, 5.3 min,33 slices, 64 × 64 matrix, 3 mm thickness, 0.6 mm spacing, TR 2 s, flip angle 70°, FOV 192 mm, TE 30 ms, 160 volumes). High‐resolution 1 mm isotropic T1‐weighted structural scans were acquired to exclude subjects with potential anomalies, to serve as individual templates, and to investigate potential structural gray matter (GM) differences across groups. Inhouse scripts were used for the systematic standard preprocessing of functional images. After four initial volumes were discarded, functional images were slice‐time corrected, realigned to the fifth volume using rigid body transformation, unwarped, co‐registered, and normalized to the EPI template in Montreal Neurological Institute (MNI) stereotactic space with a voxel resolution 3 × 3 × 3 mm³ with SPM8 (http://www.fil.ion.ucl.ac.uk/spm8) and MATLAB (The MathWorks, Natick, MA). We carefully checked for motion parameters as it is known that these parameters might induce false‐positive results [Van Dijk et al., 2012]. We set a cutoff for motion quality of the images of < 2 mm for the three translation planes over the entire experiment, which is smaller than the voxel size, and 2° for the three rotation planes. If participants had exceeded this motion threshold, they would have been excluded from the final analysis due to excessive movement. Moreover, in order to evaluate whether potential differences in frame‐by‐frame displacements across groups could contribute as a bias to the differential FC seen in our results, we computed individual root‐mean‐square of the relative displacements between adjacent volumes in mm and compared the groups as done by Van Dijk et al. [2012]. Additionally, all preprocessed functional images were extensively “cleaned up” by regressing out 16 nuisance factors, namely the six movement parameters, WM and CSF signals, and their first temporal derivatives.

Identification of Independent Components and Comparison of Networks

Resulting images were given to FSL MELODIC [Chen et al., 2008; Smith et al., 2009] for independent component analysis (ICA) after temporal concatenation, default high pass filter cutoff of 0.01 Hz was preserved, and the FP, CO, and DM components were selected (Supporting Information Fig. 1) in each group for posterior group comparison [as performed in Celone et al., 2006; Chen et al., 2008]. Back transformation was implemented for each component and for each subject with band‐filtered (Butterworth 0.01–0.1 Hz) first Eigenvariate time courses, followed by whole‐brain GLM and Fisher's r‐to‐z transformation of 8 × 8 × 8 FWHM smoothed images (called beta images). Individual networks were then used for second‐level analysis and comparison was implemented with between‐group one‐way ANOVA (http://www.fil.ion.ucl.ac.uk/~wpenny/publications/rik_anova.pdf). Differences driving the results were post‐hoc described with pairwise comparisons. An initial statistical threshold P < 0.05 cluster corrected (voxel threshold P < 0.005, cluster size k > 48) for multiple comparisons was set based on the Monte Carlo simulation using the AlphaSim implementation in REST [Song et al., 2011], but note that most findings survive a more rigorous P < 0.05 whole‐brain FWE correction for type I error. Differences between networks across groups were visualized with BrainNet Viewer (http://www.nitrc.org/projects/bnv). Individual values were extracted from beta images using a mask based on the alterations for each network and these values were correlated with individual clinical parameters to investigate their association with network alterations (r Pearson P < 0.05, SPSS version 22, SPSS, Chicago, IL). Secondary diagnoses included anxiety disorders, substance abuse, and personality disorders and were computed (Supporting Information Table). Considering that comorbidities along with other factors such as medication use and brain volume could potentially influence the network findings, we carefully explored whether our main results would be sustained with the inclusion of these parameters as covariates‐of‐no‐interest in the statistical model. We examined the impact of total medication load as a composite of the sum [Hassel et al., 2008; Philipps et al., 2008] or of the first Eigenvariate (probably a more parsimonious composite) of each agent coded as 0, 1, or 2 for no medication, equivalents dose equal or below, or above the median effective daily dose (ED₅₀), respectively [Aktories et al., 2013; Davis and Chen, 2004]. And finally, regarding potential biases of structural changes resulting in functional differences, we used the structural images to perform a voxel brain morphometry analysis (http://www.neuro.uni-jena.de/vbm) with SPM8 and investigate whether main network results could be simply reflecting local structural GM changes.

RESULTS

After the exclusion of one patient due to interruption of functional image acquisition, we included 40 patients (n = 20 unipolar and n = 20 bipolar) and 20 age, gender, and education matched controls in the analysis. None of our participants presented excessive movements to be excluded from the analysis. Besides, the analysis of the mean displacement showed no significant differences across groups (U:M = 0.15 SD = 0.11; B:M = 0.17 SD = 0.12; C:M = 0.14 SD = 0.06; F = 0.42, P = 0.657) that could potentially bias the FC results. Patient groups presented similar levels of depressive symptomatology by the time of scanning (Tables 1). They also presented comparable demographic and clinical parameters, except by the presence of previous manic episodes in bipolar patients.

Table 1.

Comparison of demographic, psychometric, and clinical parameters across groups

	Unipolar n = 20	Bipolar n = 20	Control n = 20	ANOVA		B vs. U
Parameter	M (SD)	M (SD)	M (SD)	F/χ2	P,2‐tail	t/χ2	P,2‐tail
Age (years)	35.6 (10.4)	35.8 (10.2)	36.2 (11.3)	0.02	0.981	−0.06	0.951
Education (years)	15.0 (3.2)	14.6 (3.0)	14.7 (2.6)	0.09	0.913	0.40	0.692
Gender (female/total)	14/20	13/20	13/20	0.15	0.928	0.11	0.736
WAIS vocabulary score	31.0 (3.4)	30.3 (3.1)	N/A	–	–	0.67	0.506
MADRS total score	13.5 (6.4)	9.1 (8.6)	N/A	–	–	1.64	0.108
BDI total score	18.5 (10.4)	N/A	N/A	–	–	–	–
YMRS total score	–	3.6 (5.9)	N/A	–	–	–	–
Age first episode (years)	26.2 (9.9)	24.6 (8.3)	N/A	–	–	0.55	0.585
Duration of disorder (years)	9.5 (9.9)	11.4 (6.5)	N/A	–	–	−0.69	0.491
Depressive episodes (n)	4.2 (2.7)	5.6 (3.8)	N/A	–	–	1.30	0.202
Manic episodes (n)	0 (0)	2.6 (1.6)	N/A	–	–	7.25	<0.01**
Mood stabilizers (n of users)	1/20	15/20	N/A	–	–	20.41	<0.01**
Antidepressants (n of users)	16/20	10/20	N/A	–	–	3.96	0.096
Neuroleptics (n of users)	4/20	12/20	N/A	–	–	6.67	<0.05*
Hypnotics (n of users)	3/20	1/20	N/A	–	–	1.11	0.605
Unmedicated (n of nonusers)	4/20	2/20	N/A	–	–	0.78	0.331
Comorbidities	12/20	4/20	N/A	–	–	6.67	<0.05*
Substance abuse	4/20	4/20	N/A	–	–	0.00	1.00

WAIS Wechsler Adult Intelligence Scale Vocabulary Score (37 items), MADRS Montgomery‐Åsberg Depression Rating Scale (10 items), BDI Beck Depression Inventory (21 items), YMRS Young Mania Rating Scale (11 items).

Although the candidate networks FP, CO, and DM were well represented in all groups, robust group‐specific differences were identified (Table 2, Figs. 1, 2). Bipolar patients displayed significantly increased FC in the FP network mainly in the left ventrolateral and dorsolateral prefrontal cortex (VLPFC and DLPFC) in contrast to unipolar patients and controls. Conversely, unipolar patients presented significantly reduced FC in the CO network in the anterior cingulate cortex (ACC), pregenual, and subgenual ACC, posterior cingulate cortex (PCC), and inferior parietal cortex bilaterally in contrast to bipolar patients and controls. Unipolar patients also exhibited increased FC in the DM network in the precuneus (PCN) and hippocampus (HPC) bilaterally in contrast to bipolar patients and controls.

Table 2.

Differences in the MNI peak voxels in unipolar, bipolar, and control groups with between‐group one‐way ANOVA for each network (FP, frontoparietal; CO, cingulo‐opercular; DM, default mode)

Network	Anatomical structure	BA	x	y	z	F value	Z value	k
FP**	L Mid frontal gyrus	9	−42	12	39	31.17	6.05	116
			−48	30	24	23.70	5.41
			−42	12	54	21.24	5.15
	L Mid frontal orbital gyrus		−39	45	−6	28.60	5.85	56
	L Inf parietal gyrus	7	−33	−66	45	25.45	5.57	52
CO**	L Sup frontal gyrus		−15	54	27	67.02	Inf	2044
			−3	33	−6	55.01	7.44
			3	45	18	54.01	7.40
	R Postcingulate cort	31	9	−51	30	41.26	6.73	511
			−6	−45	36	38.70	6.58
			−6	−57	27	29.07	5.89
	R Inf temporal gyrus	21	48	9	−39	40.85	6.71	115
			60	−3	−27	33.82	6.25
			63	−12	−21	24.77	5.51
	L Angular gyrus		−45	−63	36	30.93	6.03	105
	R Supramarginal gyrus		63	−36	33	25.43	5.57	80
	R Angular gyrus		57	−60	27	25.37	5.56	74
			45	−54	27	22.56	5.29
DM*	R Intraparietal sulcus		27	−36	45	14.43	4.30	139
			18	−33	42	10.74	3.70
	L Precuneus	7	−9	−57	66	14.00	4.24	578
			3	−48	69	11.57	3.84
			9	−66	60	11.49	3.83
	R Parah/hipocampus		12	−33	−3	10.15	3.58	110
			27	−33	0	9.74	3.50
	L Parah/hipocampus	35	−21	−24	−15	9.80	3.51	69
			−24	−33	6	8.42	3.23
			−33	−39	9	7.94	3.12
	L Thal/pulvinar		0	−21	24	8.68	3.28	82
			−15	−24	15	7.40	2.99
			−6	−21	18	7.14	2.93

BA Brodmann area, MNI Montreal Neurological Insitute, k cluster size, * P _clus < 0.05, ** P _FWE < 0.05, L left, R right.

Figure 1.

Differences for each network (FP, frontoparietal; CO, cingulo‐opercular; DM, default mode) in unipolar (U), bipolar (B), and control (C) groups with between‐group one‐way ANOVA (increased with warm colors; decreased with cold colors; P < 0.05 corrected). The histograms represent the direction of the differences seen.

Figure 2.

Post hoc description of differences in networks with pairwise comparisons. In the top rows, color overlays represent the network (FP, frontoparietal; CO, cingulo‐opercular; DM, default mode) in each group (U, unipolar; B, bipolar; C, control); in the bottom rows, the color overlays represent the alterations in the functional connectivities (increased with warm colors; decreased with cold colors; P < 0.05 corrected); note that the alterations in the CO network include a collection of blobs over typical nodes of the DM network.

These connectivity alterations in the CO and DM were significantly correlated to the number of depressive episodes in unipolar patients (CO: Pearson r = 0.51, P = 0.025, DM: r = 0.54, P = 0.017, Fig. 3) but not to actual symptom scores (all P > 0.7). As bipolar patients did not present significant correlations to the number of depressive episodes (FP: Pearson's r = −0.02, P = 0.92) nor to actual symptom scores (FP: Pearson r = 0.28, P = 0.23), we hypothesized it could be due to the existence of previous manic episodes. However, manic episodes were also not correlated (FP: Pearson r = −0.19, P = 0.43) to the FP alterations.

Figure 3.

Individual beta‐weights extracted from altered regions of the cingulo‐opercular (CO) and default mode (DM) networks correlate to the number of depressive episodes in unipolar patients (CO: Pearson r = 0.51, P = 0.025, DM: Pearson r = 0.54, P = 0.017). [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Figure 4.

A model representing the specific network alterations in patients is shown; note the bipolar patients presented increased frontoparietal (FP) functional connectivity in the dorsolateral and ventrolateral prefrontal cortex (DLPFC, VLPFC); unipolar patients presented increase default mode (DM) functional connectivity in the precuneus and hippocampus (PCN, HPC) as well as decreased cingulo‐opercular (CO) functional connectivity in the rostral and subgenual anterior cingulate cortex (rACC, sgACC).

Main FC results were sustained after including one or the other total medication load composite versions together with total gray matter volume scores and number of comorbidities (see Supporting Information Table) as covariates‐of‐no‐interest in the statistical model, which indicates that our network findings are not driven by these factors. Lastly, regions presenting potential structural differences across groups (with or without age as a covariate‐of‐no‐interest) did not overlap with the FC results (Supporting Information Fig. 2), showing that the main functional differences are not resulting from structural differences in GM.

DISCUSSION

Our study investigated connectivity changes in three major large‐scale networks across acutely depressed unipolar and bipolar patients under resting‐state condition to determine characteristic alterations of one or of the other disorder. Groups were well‐matched for age, gender, and education level. As we expected, despite similar depressive symptom scores, robust group‐specific alterations in the FP, CO, and DM networks were seen. Bipolar patients were characterized by increased FC within regions of the FP network. Conversely, unipolar patients presented increased FC within regions of the DM network, as much as reduced connectivity of the CO network to DM regions. These connectivity changes were significantly correlated to the history of depressive episodes and not to present symptom scores but only in unipolar patients. The contribution of other factors, such as medication use, comorbidities, and potential structural GM differences to the main results was not relevant.

All three networks were well represented in our patient and control samples. However, the connectivity in the DM network was seen exceptionally increased in unipolar patients, extending to parts of the PCN and the posterior HPC bilaterally. Our findings are in line with findings from previous studies showing increased DM network connectivity in unipolar patients [Greicius et al., 2007]. DM dysconnectivity has been related to negative and ruminative thoughts expressed by unipolar patients [Hamilton et al., 2011; Zhu et al., 2012], therefore being considered as correlates of altered self‐referential and introspective mental activity. Such an increase in this network has not only been reported during resting‐state but also during the performance of goal‐oriented tasks [Hamilton et al., 2011; Sheline et al., 2009], especially regarding the anterior portions of the DM network, namely the pregenual and subgenual ACC. Remarkably, our findings show strong dysconnectivity of these regions in unipolar patients to the CO network, which extends to the right operculum and dorsal ACC. The CO network, also known as salience network, signalizes the need to adapt control [Dosenbach et al., 2007] and plays an important role in switching from DM‐related demands to executive‐related demands and vice versa. In this context, the CO network would be particularly impaired in unipolar patients in assisting the physiological interplay between the internally‐oriented DM and the externally‐oriented FP network [for review, see Menon, 2011]. In addition, other studies have reported impairments in the salience network in major depression [Connolly et al., 2013] as well as the lack of coordinated deactivation in the pregenual and subgenual ACC from rest to goal‐oriented mental activity [see Pizzagalli, 2011]. Based on the reduced FC in the CO network and increased FC in the DM network, our results reinforce that a deficient salience network might be associated to the inability to reallocate resources to externally‐oriented mental activity in unipolar patients [Menon, 2011], which would predispose and/or sustain internally‐oriented ruminative thoughts associated to the negative biases in information processing. A platform for cognitive and psychological reconsolidation [Harmer et al., 2009] might dependent on the restoration of the FC between the hippocampus and the medial prefrontal cortex [Genzel et al., 2015]. Although we could not evaluate it prospectively, the retrospective collection of the clinical features has given us an important insight into the implication of these dysconnectivities. Our data show that unipolar patients who presented more dysconnectivity in the CO network to the DM regions have experienced more depressive episodes in their lifespan. This information is in line with a recent study, showing that dysconnectivities involving the operculum bilaterally has significantly correlated with the amount of depressive episodes in unipolar patients [Meng et al., 2014]. Therefore, new light is shed on the vulnerability to depressive episodes by these network alterations. Above all, our results suggest the concept of dysconnectivities in the PCN, HPC, and pregenual and subgenual ACC as a fingerprint for unipolar depression in contrast to bipolar patients and controls (Fig. 4).

The FP or central‐executive network includes portions of lateral PCN, DLPFC, and VLPFC, and has been associated with externally‐oriented and cognitively demanding mental activity [Dosenbach et al., 2007]. The dysconnectivity of the FP network particularly in the VLPFC and DLPFC was the main characteristic of bipolar patients in our findings. Considering that these patients were not in manic phases, during which increased attentional and externally oriented activity would be expected, these findings are noteworthy. Altered connectivity in specific prefrontal areas at baseline is consistent with dysfunctional implementation of top‐down control and impaired executive functioning as described in the literature of bipolar disorder [Brooks et al., 2006; Chai et al., 2011; Chen et al., 2011; Fernandez‐Corcuera et al., 2013; Kronhaus et al., 2006; Reinke et al., 2013]. As part of the FP network these prefrontal regions have been consistently reported regarding abnormal activity and connectivity. According to a quantitative meta‐analysis [Chen et al., 2011], bipolar patients have shown alterations in the activation of the VLPFC during cognitive and emotional processing regardless of mood state. In resting state, DLPFC dysconnectivity has been identified in bipolar patients [Chai et al., 2011; Khadka et al., 2013], eventually signalizing psychotic features of the disorder [Anticevic et al., 2013]. Additionally, the dysconnectivity of this region has been already reported in pediatric patients [Dickstein et al., 2010], which speaks in favor of a very early characteristic of the disorder. Such prefrontal dysconnectivities were not only seen in bipolar patients, but also in their unaffected first‐degree relatives [Khadka et al., 2013; Meda et al., 2012; Singh et al., 2014], suggesting them to represent a vulnerability heritable factor for developing the disorder. Notably, VLPFC and DLPFC dysconnectivites strongly suggest the contribution of an inefficient executive control system within the FP network in the development of depressive symptoms. Unfortunately, this question could not be further elucidated by our results as neither the actual degree of symptoms nor the history of depressive episodes has correlated with the connectivity changes seen in these patients. In any case, patterns of dysconnectivity in the prefrontal cortex have robustly differentiated bipolar from unipolar patients and controls, which encourage further investigation of such neuroimaging tools towards supporting the differential diagnosis of depression between two highly prevalent affective disorders.

Although unipolar and bipolar patient groups have been more commonly studied independently than in conjunction, some functional studies have compared them under the same research protocol, ranging from reward, cognitive and emotional task‐based to resting‐state paradigms [Anand et al., 2009; Bertocci et al., 2012; Chase et al., 2013; Grotegerd et al., 2012; Lawrence et al., 2004; Liang et al., 2013; Liu et al., 2012, 2013; Mourao‐Miranda et al., 2012; Versace et al., 2010]. Reward expectancy was evaluated in both types of depression with a card‐guessing game and reduced activation in the dorsal ACC was seen in depressed patients from both groups [Chase et al., 2013]. Among other regions, the authors report increased activation in the VLPFC and DLPFC in bipolar I patients. In another study, the dorsal ACC was reported to differentiate unipolar from bipolar depressed females using an emotional n‐back task [Bertocci et al., 2012]. Increased activation was seen in patients with unipolar disorder in contrast to healthy controls and bipolar patients during 2‐back condition with neutral face distracters. Even though a direct translation to resting‐state connectivity findings can be difficult, it is worth mentioning that our study has also identified the ACC in unipolar patients and the VLPFC and DLPFC in bipolar patients as key regions altered in these two disorders. Also, increased activation in the VLPFC has been shown during the presentation of emotional faces when bipolar patients were compared to unipolar patients and controls [Lawrence et al., 2004]. When applying a classifier to investigate the discriminability between controls, unipolar and bipolar patients, according to their patterns of brain activation during presentation of emotional facial expressions, groups seem reasonably differentiable [Grotegerd et al., 2012; Mourao‐Miranda et al., 2012]. Our findings are also in line with regional homogeneity (ReHo) data presenting the DLPFC and VLPFC in bipolar patients and the parahippocampal region in unipolar patients as disorder‐specific altered regions [Liang et al., 2013]. Collectively, differences in the anterior insula seen between unipolar and bipolar groups with amplitude of low frequency fluctuation (ALFF) and ReHo [Liu et al., 2012, 2013] analyses support the relevance of the networks alterations seen in our results. The region is typically part of the CO network, which was specifically altered in unipolar patients in contrast to bipolar patients and controls.

Overall, a triple network model of aberrant salience and cognitive dysfunction was proposed [Menon, 2011], in which FP, CO, and DM networks play major roles [Hamilton et al., 2013]. We have compared these networks across groups after decomposing whole‐brain spatiotemporal components and we confirm that disorder‐specific networks can be found with this method. As described by Cole et al. (2013), the use of ICA to guide network selection is advantageous, because differently from seed‐based approaches, it is less susceptible to spatial selection problems derived from the location of seeds. Another important aspect is that through ICA, one can access the individual first Eigenvariate representing each network, which has been suggested as a “cleaner” representation of the time series in comparison to an average across voxels [Friston et al., 1996; O'Reilly et al., 2010]. And above all, the networks identified during rest have been shown to be likewise represented during the performance of task‐based goal‐oriented paradigms [Smith et al., 2009]. Therefore, the main advantage of our study is to report a “cleaner” comparison between the main large‐scale networks across patient groups, extending the knowledge of disorder‐specific dysconnectivities in depression, which was performed in resting‐state but may as well represent dysconnectivities involved in cognitive impairments.

Potential limitations include the study sample and cross‐sectional design and should be taken into account. According to the goal of this study, we have rather included well‐diagnosed patients with mood disorders due to much lower chances of changing groups (i.e., patient with depression diagnosed unipolar but in reality a bipolar patient). Consequently, an important point is noteworthy. Our results might depict rather a chronic subsample of medicated patients. As expected, more bipolar than unipolar patients were in use of neuroleptics and mood stabilizers. Although we carefully addressed this point by additionally regressing out medication use together with comorbidities and brain volumes scores and it did not seem to influence the main results, data should be interpreted carefully. Additional replication studies with bigger independent samples from multicentric cohorts or data sharing initiatives could correctly address the accuracy of this method (this should not be performed with our sample as the use of dependent samples generally incurs in overfitting results due to double‐dipping). Moreover, our cross‐sectional design limits the evaluation of causative effects. To further understand the underpinnings of affective disorders and its recovery, which was not designed to be evaluated here but remains nevertheless imperative, it would be ideal to have patients prospectively followed‐up during treatment onset and to have their clinical responses correlated to possible changes in large‐scale networks.

CONCLUSION

To summarize, despite similar depressive symptoms, connectivity changes in large‐scale networks have been very distinct across unipolar and bipolar groups. Our candidate networks, the FP, CO, and DM network, displayed robust disorder‐specific changes. Bipolar patients were characterized by increased FC in the FP network, a central executive and external‐oriented network. Unipolar patients presented increased FC in the DM network, an introspective and self‐referential network. The latter also presented reduced connectivity in the CO network to DM regions, a network involved in dynamic switching between DM and executive‐related demands. These connectivity changes significantly correlated to the history of depressive episodes in unipolar patients but not to actual symptom scores.

Disclosure

All authors report no external funding and no potential conflicts of interest.

Supporting information

Supporting Information

Supporting Information