Altered effective connectivity model in the default mode network between bipolar and unipolar depression based on resting-state fMRI

Altered effective connectivity model in the default mode network between bipolar and unipolar depression based on resting-state fMRI

Author's Accepted Manuscript Altered effective connectivity model in the default mode network between bipolar and unipolar depression based on restin...

1MB Sizes 0 Downloads 79 Views

Author's Accepted Manuscript

Altered effective connectivity model in the default mode network between bipolar and unipolar depression based on resting-state fMRI Yunting Liu, Xia Wu, Jiacai Zhang, Xiaojuan Guo, Zhiying Long, Li Yao

www.elsevier.com/locate/jad

PII: DOI: Reference:

S0165-0327(15)00224-4 http://dx.doi.org/10.1016/j.jad.2015.04.009 JAD7386

To appear in:

Journal of Affective Disorders

Received date: 3 December 2014 Revised date: 27 March 2015 Accepted date: 2 April 2015 Cite this article as: Yunting Liu, Xia Wu, Jiacai Zhang, Xiaojuan Guo, Zhiying Long, Li Yao, Altered effective connectivity model in the default mode network between bipolar and unipolar depression based on resting-state fMRI, Journal of Affective Disorders, http://dx.doi.org/10.1016/j.jad.2015.04.009 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Altered effective connectivity model in the default mode network between bipolar and unipolar depression based on resting-state fMRI

Yunting Liu1, Xia Wu1,2,∗, Jiacai Zhang1, Xiaojuan Guo1, Zhiying Long2, Li Yao1, 2 1

College of Information Science and Technology, Beijing Normal University, Beijing

100875, China 2

State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal

University, Beijing 100875, China



   ∗

 Correspondence: Xia Wu, College of Information Science and Technology, Beijing

Normal University, Xin Jie Kou Wai Street 19#, Haidian District, Beijing 100875, China. Tel.: +86 10 58800427; fax: +86 10 58800056; Email: [email protected] 

Abstract Background: Bipolar depression (BD) is characterized by alternating episodes of depression and mania. Patients who spend the majority of their time in episodes of depression rather than mania are often misdiagnosed with unipolar depression (UD) that only exhibits depressive episodes. It would be important to explore the construction of more objective biomarkers which can be used to more accurately differentiate BD and UD. Methods: The effective connectivity model of BD and UD in the default mode network (DMN) was constructed based on resting-state fMRI data of 17 BD (32.128.57 years old) and 17 UD (32.599.77 years old) patients using a linear nonGaussian acyclic model (LiNGAM). The effective connectivity differences were obtained by conducting a permutation test. Results: The following connections were stronger in the BD group than in the UD group: medial prefrontal cortex (MPFC) posterior cingulate cortex (PCC), right inferior parietal cortex (rIPC)  left hippocampus (lHC) and rIPC  right insula (rInsula). In contrast, the following connections were weak or unapparent in the BD group: MPFClHC, rHCMPFC, rHCrInsula and rInsulalHC. Limitations: First, the medication effect is a confounding factor. Second, as with most fMRI studies, the subjects’ thoughts during imaging are difficult to control. Conclusions: The brain regions in these altered connections, such as the HC, insula, MPFC and IPC, all play important roles in emotional processing, suggesting that these altered connections may be conducive to better distinguish between BD and UD.

Keywords: bipolar depression, unipolar depression, resting-state fMRI, effective connectivity, LiNGAM

1. Introduction Bipolar depression (BD) is one of the most common and severe mental diseases (de Almeida and Phillips, 2013; Liu et al., 2012; Murray and Lopez, 1996), which can lead to serious damage to the family, occupation and society (Hirschfeld and Vornik, 2005; Yatham et al., 2009). It is characterized by alternating episodes of depression and mania, but patients spend more time in the depressive phase than in the manic phase (Chen et al., 2011; Judd et al., 2003). Thus, BD is often misdiagnosed with unipolar depression (UD) that is characterized by depressive episodes alone (Bowden, 2001; Versace et al., 2010). Frequent misdiagnoses of one of these two conditions to another will lead to inappropriate treatment, high medical costs and poor outcomes (Bowden, 2010; Hirschfeld et al., 2003). Thus, it would be important to explore the construction of more objective biomarkers which can be used to more accurately differentiate BD and UD. Neuroimaging technology such as functional magnetic resonance imaging (fMRI) has been a promising method to investigate neurophysiological and neuroanatomical correlates of affective disease (Hamilton et al., 2011; Strakowski et al., 2014). Applications of fMRI and other imaging technologies may help to reveal the characteristics of BD and UD, or to provide auxiliary for the identification between BD and UD. Resting-state functional MRI has become a valuable tool for the investigation of brain network function (Biswal et al., 1995; Greicius et al., 2003), and it holds great

promise for examining abnormalities in mental disease patients (Hasler and Northoff, 2011). For example, the resting-state functional connectivity of BD and UD was compared, indicating a more severe decreased connectivity in the BD patients than in the UD patients (Anand et al., 2009). Differences in resting-state brain activity between BD and UD patients were investigated by measuring the amplitude of the low-frequency fluctuations (ALFF) of fMRI signals, and the results supported the belief that insular sub-regions contribute to the precise differences between BD and UD (Liu et al., 2012). Effective connectivity analysis, focusing on the causal relationship between brain regions, is widely used in the studies of resting state brain networks. It can reflect how one brain region influences another and can quantitatively depict the strength of such influence (Friston, 1994). Studies have shown that the altered effective connectivity may serve as a potential biomarker to reveal characteristics of neurological disease such as Alzheimer’s disease (Wu et al., 2011) and primary progressive aphasia (Sonty et al., 2007). Similarly, the application of

effective

connectivity

in

the

depression

studies

may

reflect

more

pathophysiological characteristics of depression patients from a new perspective and potential use as a biomarker. The default mode network (DMN) is a group of areas in the human brain, and some of its regions are most active when people are in a resting state (Raichle et al., 2001; Raichle and Snyder, 2007; Sheline et al., 2009). Studies have found that the DMN is related to reviewing past knowledge for preparing future action (Binder et al., 1999), episodic memory processing (Greicius et al., 2003) and so on. Although the

exact function of the DMN is still not clear, a large number of studies have shown that some regions in the DMN are associated with mental diseases, such as BD and UD (Adler et al., 2006; Altshuler et al., 2005; Killgore et al., 2008; Malhi et al., 2007; Peng et al., 2011; Sprengelmeyer et al., 2011; Yurgelun

Todd et al., 2000). However,

most of these studies are based on the level of brain activity. Effective connectivity differences between BD and UD in the DMN are yet to be explored. The current study aimed to identify DMN effective connectivity differences between BD and UD patients. The DMN effective connectivity pattern of these two groups of patients was constructed using resting-state fMRI data by a linear nonGaussian acyclic model (LiNGAM) (Shimizu et al., 2006). In Shimizu’s study, the linear non-Gaussian acyclic model (LiNGAM) was proposed, and a non-Gaussian method closely related to independent component analysis (ICA) called ICALiNGAM algorithm was also developed to estimate the new model. Recently, an improved method called pooling-LiNGAM (pLiNAM) was proposed to make the results more stable (Xu et al., 2014). The effective connectivity differences estimated from the pLiNGAM algorithm in the present study may become possible indices for distinguishing BD patients from UD patients.

2. Methods 2.1 LiNGAM theory First, the LiNGAM theory based on ICA algorithm (Shimizu et al., 2006) was introduced.

1) Suppose that there exists a causal order  among variables  

 , in that no later variable determines any earlier variable. The variables  are generated recursively (Shimizu and Kano, 2008), and can be represented by a directed acyclic graph (DAG) according to their causal order (Pearl, 2000; Spirtes et al., 2000). Each variable  can be represented by a linear combination of its parent variables: 2)

 





       

where  represents the connection strength from variable  to  ,  indicates an optional term and  denotes a non-Gaussian noise and are independent of each other. The model (1) can be transformed and rewritten in a matrix form:       where x is a vector constituted by all the variables  , B is the connection strength matrix for the variables, and can be permuted to a strict lower triangular matrix if there exists a causal order of the variables. The equation (2) can be further written as:    where A= (I – B)-1. For B can be transformed to a strict lower triangular matrix, A can be permuted to lower triangularity (but not strict lower triangularity) with all diagonal elements non-zero. Due to the independence and non-Gaussianity of e, equation (3) defines an independent component analysis (ICA) model. A model with the characteristics above can be called a Linear, Non-Gaussian, Acyclic Model (LiNGAM). ICA is essentially able to estimate A (and W=A-1=  ! ) if given enough

observed vectors x, but it has indeterminacies of permutation, scaling and sign. Actually, ICA gives W =PDW, where P is an unknown permutation matrix and D is an unknown diagonal matrix. In LiNGAM algorithm, since matrix B can be permuted to a strict lower triangular matrix and W = I – B, the correct permutation matrix P is the only one that gives no zeros in the diagonal of DW (Shimizu et al., 2006). Then, the correct scaling and signs of the independent components can be found by using the unity on the diagonal of W = I – B. The rows of DW have to be divided by its corresponding diagonal elements to obtain W. Finally, the matrix B = I – W can be computed (Shimizu et al., 2006; Shimizu et al., 2011). Previous study on simulated data showed that the ICA-LiNGAM algorithm need more data points (e.g., more than 1000 data points) to perform more stably (Smith et al., 2011). However, the number of data points in most fMRI is usually no more than 300. To solve this problem, an improved method called pooling-LiNGAM (pLiNGAM) was proposed by Xu et al. (2014), and its feasibility and efficiency was demonstrated by the estimation of effective connectivity on both simulated and real fMRI data. The pLiNGAM algorithm obtains long data points by pooling data points across multiple subjects and the pooling subject is called as virtual subject (Smith et al., 2011; Xu et al., 2014). Then the traditional ICA-LiNGAM method is used to estimate the casual relationship models for the virtual subjects. Details about the pLiNGAM algorithm are not included in this research because of concerns about limitations on the paper length. See Xu et al. (2014) for more detailed theory of the pLiNGAM. Considering the stability of the results, the pLiNGAM method was

chosen to construct the effective connectivity models of BD and UD. 2.2 Subjects Patients involved in this study were recruited from the outpatient clinic at Anding Hospital, Capital Medical University and included 17 currently depressed individuals with bipolar depression [6 males and 11 females, ages between 20 and 53 years (meanSD: 32.128.57 years)] and 17 currently depressed individuals with unipolar depression [5 males and 12 females, ages between 21 and 57 years (meanSD: 32.599.77 years)]. There were no significant differences between the two groups in age, gender, or educational level, and all patients were right-handed. Diagnoses of BD and UD were made based on DSM-IV (The Diagnostic and Statistical Manual of Mental Disorders). Demographic statistics and clinical features of the subjects are shown in Table 1. The study was approved by the Imaging Center for Brain Research, Beijing Normal University and the Institutional Review Board of Anding Hospital, Capital Medical University. In addition, all subjects provided written informed consent prior to entering the study. 2.3 Data acquisition MRI scanning was performed on a 3.0 T Siemens whole-body MRI system in the State Key Laboratory for Cognitive Neuroscience and Learning, Beijing Normal University. Functional images were acquired using a gradient echo planar imaging sequence (33 axial slices, TR=2000 ms, TE=30 ms, FA=90º, thickness/gap=3.5/0.6 mm, FOV=220×220 mm, matrix=64×64 mm, volume=240). The subjects were required to hold still and keep their eyes closed but not fall asleep during the scan.

2.4 Data preprocessing For each participant, the first 10 scans of the fMRI time series were discarded to allow for equilibration of the magnetic field. The data preprocessing steps included slice timing correction, motion correction, spatial normalization and smoothing using SPM8 (http://www.fil.ion.ucl.ac.uk/spm). Participants would be excluded for excessive head motion (>1 mm of displacement or >1ºof rotation in any direction). All images were spatial normalized to Montreal Neurological Institute (MNI) space using the EPI template, and the voxel size of the resample was 3 mm×3 mm×4 mm. To eliminate the high-frequency noise, the normalized images were spatially smoothed with an 8 mm full-width-at-half-maximum (FWHM) Gaussian kernel. 2.5 Generating the spatial pattern of DMN by Group ICA The preprocessed data for all subjects were entered into the Group ICA program in the fMRI Toolbox GIFT (http://icatb.sourceforge.net), which performed data reduction by two rounds of principle component analysis (PCA), ICA separation and back-reconstruction (Calhoun et al., 2001b). First, the number of independent components of each group was estimated based on the principle of minimum description length (MDL), and the data were dimension-reduced by two rounds of PCA. In the first round of PCA, the data for each individual subject were dimensionreduced to the optimal number temporally. After concatenation across subjects within groups, the dimensions were again reduced to the optimal numbers via the second round of PCA. Then, the data were separated by ICA using the Extended Infomax algorithm (Lee et al., 1999). Finally, the mean independent components (ICs) and the

corresponding mean time courses over all of the subjects were used for the backreconstruction of the ICs and the time courses for each individual subject (Calhoun et al., 2001b). The independent components that best matched the DMN, as previously reported, were selected in each group. A DMN template was developed based on a dataset of regions reported previously (Greicius et al., 2004). Each region in the template was a sphere with a radius of 5 mm (variations in the size of the sphere had no effect on the component identification). To determine the DMN among a number of independent components for a subject, the average intensity over voxels within each of the spheres minus that over the voxels outside all of the spheres was calculated for each component. The component that had the best fit was designated as the DMN for that subject. After the conversion of the intensity values to Z-scores in each IC spatial map, one sample t-test (false discovery rate, FDR; p=0.005) was performed to determine the DMN for each of the two groups (Calhoun et al., 2001a). Differences in DMN functional connectivity between the two groups were determined by two sample ttests (p=0.001, uncorrected). 2.6 Defining the regions of interest and extracting the time series On the basis of the functional connectivity discoveries, several core brain regions were identified to construct effective connectivity models separately for BD and UD patients. First, eight key DMN regions, as regions of interest (ROIs) for BD and UD, were identified separately. Then, two regions that had significant differences in functional

connectivity between the two groups were selected. Each ROI was defined as a 6 mm sphere centered on the local maximum cluster in the functional connectivity maps. For each ROI, resting-state time series were extracted by averaging the intensities over all voxels within the ROI at each time point for each individual. 2.7 Constructing the effective connectivity model of DMN using pLiNGAM algorithm In the current study, the pLiNGAM algorithm was applied to the resting-state fMRI data of 17 BD and 17 UD patients. Every subject had 10 variables (ROIs), and the length of time series of the variables was 230. Thus each subject can be shown using a 10"230 matrix. The time series of the BD and UD data were separately concatenating across subjects to generate one virtual BD subject and one virtual UD subject according to the pLiNGAM algorithm, thus the length of each variable of every virtual data was 230*17=3910 points. The BD virtual subject and UD virtual subject can be shown using a 10"3910 matrix. Then, the Kolmogorov-Smirnov (K-S) test was used to investigate the distribution of the virtual subjects’ time series. The KS test can determine whether an empirical distribution is in line with a theoretical distribution, such as a Gaussian distribution. The results of the test showed that all of the data series were non-Gaussian (p<0.05). Finally, the LiNGAM algorithm was applied to the two virtual subjects to get the causal relationship models, that is, effective connectivity models. Here, each ROI variable corresponded to the variable  in the LiNGAM algorithm, and the connection strength between ROIs were equivalent to  in LiNGAM.

2.8 Calculating effective connectivity differences of DMN between BD and UD To examine the effective connectivity differences between BD and UD patients, a randomized permutation test was used (Hesterberg et al., 2005). A permutation test is a statistical inference method put forward by Fisher in the 1930s. This method is based on a large number of calculations, using full permutation or random permutation of the sample data for statistical inference. It recalculates the statistical measure and constructs an empirical distribution by replacing the order of the samples, and then, it calculates the p value for the inference. In the current study, 17 BD subjects and 17 UD subjects participated. We combined the data of the 34 subjects, extracted 17 subjects randomly and concatenated them as a new BD virtual subject. Then, the remaining 17 subjects were concatenated as a new UD virtual subject. A LiNGAM model for each group was constructed, and the differences in the weight coefficients between the BD and UD group were taken as the statistical measure. The statistics for the original two groups of 17 UD and 17 BD subjects were also calculated. We ran a total of 3000 permutations and assessed the sample distributions. Finally, the probabilities of the connections in the LiNGAM effective connectivity model of the BD group that were different from those of the UD group were examined. We ultimately obtained 8 connections that were stronger in the BD group than in the UD and 8 connections that were stronger in the UD group than in the BD group (p=0.05).

3. Results 3.1 The spatial patterns of DMN

Figure 1 demonstrates the spatial maps of the DMN results of the BD patients detected by Group ICA followed by one sample t-test (FDR, p=0.005), and Figure 2 shows the corresponding results for the UD patients. The DMN in both BD and UD include the posterior cingulate cortex (PCC), medial prefrontal cortex (MPFC), left/right inferior parietal cortex (lIPC/rIPC), left/right inferior temporal cortex, (lITC/rITC), and left/right hippocampus cortex (lHC/rHC). ‘aDMN’ and ‘pDMN’ represent the anterior and posterior portions of the DMN separately. 3.2 Differences in DMN functional connectivity between the BD and UD group Before constructing the effective connectivity network of the two groups, we first explored their differences in functional connectivity. A two sample t-test (p=0.001, uncorrected) was used for the two groups’ DMN maps by Group ICA, and the results are shown in Figure 3, Figure 4 and Table 2. In the left middle temporal cortex (lMTC), medial prefrontal cortex (MPFC), posterior cingulate cortex (PCC), left inferior parietal cortex (lIPC), and right inferior parietal cortex (rIPC), BD patients had a stronger functional connectivity than UD patients. Furthermore, in the right insula (rInsula) and lMTC, UD patients had a stronger functional connectivity than BD patients. On the basis of the functional connectivity discoveries, 10 ROIs (8 classical DMN regions, rInsula and lMTC) were selected to estimate the DMN effective connectivity models. The information regarding these 10 ROIs is shown in Table 3(a) and Table 3(b). 3.3 Effective connectivity models of DMN in BD and UD

The results of the effective connectivity analysis of the DMN obtained by pLiNGAM for the BD and UD patients are summarized in Table 4 (a) and Table 4 (b). The data for both the BD and UD patients are presented in 10×10 matrices. Each row/column in the table corresponds to a ROI. If the (i, j)th element is not zero, there exists a connection from the ith region to the jth region, and the value of the element represents the strength of the connection. The “positive” connection strengths (value more than 0) imply promotion in that the increased activity of one brain region could lead to the increasing of another brain region activity. Conversely, the “negative” connection strength (value less than 0) imply inhibition in that the increased activity of one brain region could cause the reduced activity of another brain region (Smith et al., 2011).The effective connectivity models of DMN for BD and UD are shown in Figure 5. From the results, the connections that existed in both the BD and UD groups included PCClHC, MPFCPCC, MPFClIPC, MPFC lHC, MPFCrInsula, rIPC  PCC, rIPC  lIPC, rIPC  lHC, rITC  PCC, rITC  lIPC, rITC  rIPC, rITC  lHC, rITC  rHC, rITC  rInsula, lITC  PCC, lITC  lIPC, lITC  MPFC, lITC  rIPC, lITC  lHC, lITC  rHC, lITC  rInsula, rHC  lHC, rInsula  PCC, rInsulalIPC, rInsulalHC and lMTClIPC. The weight coefficients for the BD group were all larger than those for the UD group. Connections only in the BD group included PCC  lIPC, PCC  lMTC, MPFC  lMTC, rIPC  MPFC, rIPC  rInsula, rIPC  lMTC, rITC  MPFC, rITC  lITC, rITC  lMTC, lITC  lMTC, lHC  lIPC, lHC  lMTC, rHC  PCC,

rHC  lIPC, rHC  MPFC, rHC  rIPC, rHC  rInsula, rHC  lMTC, and rInsulalMTC. Connections only in the UD group included PCCrHC, lIPCPCC, lIPC  lHC, lIPC  rHC, MPFC  rIPC, MPFC  rITC, MPFC  rHC, rIPC  rHC, lITCrITC, rInsularIPC, rInsularHC, lMTCPCC, lMTCMPFC, lMTCrIPC, lMTCrITC, lMTClITC, lMTClHC, lMTCrHC, and lMTCrInsula. 3.4 Effective connectivity differences between BD and UD To investigate the effective connectivity differences in the DMN between the two groups, a random permutation test was used (p=0.05). The between-group effective connectivity difference is shown in Table 5, and the effective connectivity difference model is shown in Figure 6.We found that the following connections were stronger in the BD group than in the UD group: MPFC  PCC, rIPC  MPFC, rIPC  lHC, rIPCrInsula, rITClITC, lITCrIPC, lHClIPC and rHCPCC. The following connections were weaker in the BD group than in the UD group: PCC  rHC, MPFClHC, rIPCPCC, lITCMPFC, rHCMPFC, rHClHC, rHCrInsula and rInsulalHC (p=0.05).

4. Discussion In this study, to explore the construction of more objective biomarkers that reflect pathophysiologic differences between BD and UD, the DMN effective connectivity model was constructed using the pLiNGAM algorithm. Compared to the UD groups, the following connections were weaker or unapparent in the BD group: PCC  rHC, MPFC  lHC, rIPC  PCC, lITC  MPFC, rHC  MPFC, rHC  lHC, rHCrInsula and rInsulalHC. Conversely, the following connections were stronger

in the BD group than in the UD group: MPFCPCC, rIPCMPFC, rIPClHC, rIPCrInsula, rITClITC, lITCrIPC, lHClIPC and rHCPCC. The hippocampus is part of the brain’s limbic system, and it plays an important role in emotional processing. In our study, the effective connectivity from rHC to PCC was greater in the BD group than in the UD group. Interestingly, the connectivity from PCC to rHC was greater in the UD group than in the BD group. This connection change may be an important index for differentiating BD patients from UD patients. In addition, among the other connections that were greater in the UD group than in the BD group, MPFClHC, rHCMPFC, rHClHC, and rHCrInsula, were all related to the hippocampus. These connections were reduced in the BD group, suggesting that there is hippocampal activity abnormalities in BD patients compared to UD patients. Studies regarding functional connectivity have also found abnormalities in hippocampal activity in bipolar patients. For example, the brain activity in the right parahippocampus of bipolar depression patients was found to be obviously reduced compared to that of healthy adults (Malhi et al., 2007). Whereas, brain activity in the parahippocampus was found to be increased in a study of attention tasks in bipolar depression patients (Adler et al., 2006). A study of the brain structure of bipolar depression patients, showed that there were structural abnormalities in the hippocampus of the BD patients (Strakowski et al., 2004). The results of these previous studies and our current results suggest that effective connectivity alterations related to the hippocampus and abnormalities in hippocampal activity can be used as important indices for distinguishing BD from UD.

MPFC, together with the amygdala, insula, anterior cingulate and ventral lateral prefrontal cortex, are considered to be important nerve components for emotional processing and regulation (Phillips et al., 2003). The results of the current study showed that the MPFClHC and rHCMPFC connections were weaker in the BD group than in the UD group. Changes in the MPFC connected with the left and right hippocampus are asymmetric, which may provide new insight for the study of depression. There were also effective connectivities that were greater in the BD group than in the UD group, such as MPFCPCC and rIPCMPFC. Other investigations also found anomalies in the medial prefrontal cortex based on the research of functional connections in BD and UD patients. Yurgelun

Todd et al. (2000) and

Killgore et al. (2008) found that the activity of MPFC was reduced in BD patients, while Altshuler et al. (2008) found that it was increased in BD patients during face stimulus tasks. By analyzing resting-state data from BD and UD patients, Liu et al. (2013) found that the activity value of MPFC was higher in BD patients than in UD patients. Although a large number of studies have explored differences in MPFC between BD and UD patients, their conclusions have been inconsistent (Baxter et al., 1985; Buchsbaum, 1986; Hosokawa et al., 2009). The present study aimed to find connection differences based on the effective connectivity from the aspect of causation, and these new findings provide a new perspective for research on BD and UD.  Three altered connections, rIPCrInsula, rHCrInsula and rInsulalHC, were associated with the insula. Among them, the latter two connections were lower in the

BD patients than in the UD patients. Similar to the differences in DMN functional connectivity, the functional connectivity in the right insula was lower in BD patients compared to UD patients. Hence, either from the aspect of functional connectivity or effective connectivity, the insula played an important role in the differentiation of BD and UD. Although previous studies have not reached the same conclusions, the results of the present study may aid in distinguishing BD from UD. Enhanced connections in the BD, including rIPC  MPFC, rIPC  lHC, rIPCrInsula, and lITCrIPC, were found in the current research. They were all related to IPC. The parietal cortex can integrate multiple life processes, such as somatic sensation, motion and visual space. Damage to the left parietal lobe affects space information processing, while damage to the right parietal lobe leads to time disorder (Hosokawa et al., 2009). Studies of BD and UD using positron emission tomography found that there are differences between BD and UD patients in the prefrontal cortex and limbic system, including the parietal cortex (Hosokawa et al., 2009). In the current study, rIPCMPFC, rIPClHC, rIPCrInsula and lITCrIPC, were stronger in the BD group than in the UD group, reflecting a causal relationship between increased connectivities of rIPC with other brain regions and BD. Although the relationship between the parietal lobe and depression is not clear, damage to the areas of parietal lobe can cause neurological disorders, thus, these altered indices may also be used to differentiate BD and UD. As for the effective connectivity method, there are a variety of methods used to evaluate effective connectivity such as structural equation modeling (SEM) (Mclntosh

and Gonzalez Lima, 1994; Schlösser et al., 2003), dynamic causal modeling (DCM) (Friston et al., 2003; Friston et al., 2014), granger causality mapping (GCM) (Goebel et al., 2003; Liao et al., 2010) and Bayesian network (BN) (Wu et al., 2011; Zheng and Rajapakse, 2006). Among these methods, both SEM and DCM are model-driven algorithms which need assumption of priori model, hence, may not suited for resting state fMRI (Heckerman, 2008). GCM is a data-driven method that obtains the causal relation model by multiple regression method and has high degree of dependence on time information. It is sensitive to down sampling and noise that may generate false causal relationship in some cases (Chen et al., 2006). BN is also a data-driven method and the data set to be analyzed should obey the Gaussian distribution (Zheng and Rajapakse, 2006). Linear non-Gaussian acyclic model (LiNGAM) algorithm, proposed by Shimizu et al. (2006), is a non-Gaussian method to estimate the casual model. The Gaussian assumption is usually made in effective connectivity researches, but most of the fMRI data don’t obey Gaussian distribution (Xu et al., 2014). Our own empirical data examinations (see Methods, section 2.7) confirmed the nonGaussian assumption based on which we decided its use for our current analysis. There are limitations to the current study. First, the diversity of medications the participants were taking may have affected the results, and the medication effect is a confounding factor. Second, as with most fMRI studies, the subjects’ thoughts during imaging are difficult to control. Therefore, our results should be considered preliminary. If allowed, future studies of a larger sample of non-medicated subjects should be designed to obtain more reliable results.

In conclusion, the DMN effective connectivity models in BD and UD patients were constructed using the pLiNGAM algorithm based on resting-state fMRI data, and some altered connections were found between these two groups of patients. These altered connections between BD and UD are expected to reflect the pathological differences of BD and UD, to differentiate BD and UD more accurately, and further to be used for the diagnosis and treatment of these diseases. Of course, future studies are needed to confirm these results, and the study approach should be improved to obtain more reliable results.

References Adler, C.M., DelBello, M.P., Strakowski, S.M., 2006. Brain network dysfunction in bipolar disorder. Cns Spectrums 11, 312. Altshuler, L., Bookheimer, S., Proenza, M.A., Townsend, J., Sabb, F., Firestine, A., Bartzokis, G., Mintz, J., Mazziotta, J., Cohen, M.S., 2005. Increased amygdala activation during mania: a functional magnetic resonance imaging study. American Journal of Psychiatry 162, 1211-1213. Altshuler, L., Bookheimer, S., Townsend, J., Proenza, M.A., Sabb, F., Mintz, J., Cohen, M.S., 2008. Regional brain changes in bipolar I depression: a functional magnetic resonance imaging study. Bipolar Disorders 10, 708-717. Anand, A., Li, Y., Wang, Y., Lowe, M.J., Dzemidzic, M., 2009. Resting state corticolimbic connectivity abnormalities in unmedicated bipolar disorder and unipolar depression. Psychiatry Research: Neuroimaging 171, 189-198. Baxter, L.R., Phelps, M.E., Mazziotta, J.C., Schwartz, J.M., Gerner, R.H., Selin, C.E., Sumida, R.M., 1985. Cerebral metabolic rates for glucose in mood disorders: studies with positron emission tomography and fluorodeoxyglucose F 18. Archives of general psychiatry 42, 441-447. Binder, J.R., Frost, J.A., Hammeke, T.A., Bellgowan, P., Rao, S.M., Cox, R.W., 1999. Conceptual processing during the conscious resting state: a functional MRI study. Journal of Cognitive Neuroscience 11, 80-93. Biswal, B., Zerrin Yetkin, F., Haughton, V.M., Hyde, J.S., 1995. Functional connectivity in the motor cortex of resting human brain using echo

planar mri. Magnetic resonance in medicine 34, 537-541.

Bowden, C., 2010. Diagnosis, treatment, and recovery maintenance in bipolar depression. The Journal of clinical psychiatry 71, e01-e01. Bowden, C.L., 2001. Strategies to reduce misdiagnosis of bipolar depression. Psychiatric Services 52, 51-55. Buchsbaum, M.S., 1986. Brain imaging in the search for biological markers in affective disorder. Journal of Clinical Psychiatry.

Calhoun, V., Adali, T., McGinty, V., Pekar, J., Watson, T., Pearlson, G., 2001a. fMRI activation in a visual-perception task: network of areas detected using the general linear model and independent components analysis. Neuroimage 14, 1080-1088. Calhoun, V., Adali, T., Pearlson, G., Pekar, J., 2001b. A method for making group inferences from functional MRI data using independent component analysis. Human brain mapping 14, 140-151. Chen, C.H., Suckling, J., Lennox, B.R., Ooi, C., Bullmore, E.T., 2011. A quantitative meta

analysis of

fMRI studies in bipolar disorder. Bipolar Disorders 13, 1-15. Chen, Y., Bressler, S.L., Knuth, K.H., Truccolo, W.A., Ding, M., 2006. Stochastic modeling of neurobiological time series: power, coherence, Granger causality, and separation of evoked responses from ongoing activity. Chaos: An Interdisciplinary Journal of Nonlinear Science 16, 026113. de Almeida, J.R.C., Phillips, M.L., 2013. Distinguishing between unipolar depression and bipolar depression: current and future clinical and neuroimaging perspectives. Biological psychiatry 73, 111118. Friston, K.J., 1994. Functional and effective connectivity in neuroimaging: a synthesis. Human brain mapping 2, 56-78. Friston, K.J., Harrison, L., Penny, W., 2003. Dynamic causal modelling. Neuroimage 19, 1273-1302. Friston, K.J., Kahan, J., Biswal, B., Razi, A., 2014. A DCM for resting state fMRI. Neuroimage 94, 396-407. Goebel, R., Roebroeck, A., Kim, D.-S., Formisano, E., 2003. Investigating directed cortical interactions in time-resolved fMRI data using vector autoregressive modeling and Granger causality mapping. Magnetic resonance imaging 21, 1251-1261. Greicius, M.D., Krasnow, B., Reiss, A.L., Menon, V., 2003. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proceedings of the National Academy of Sciences 100, 253-258. Greicius, M.D., Srivastava, G., Reiss, A.L., Menon, V., 2004. Default-mode network activity distinguishes Alzheimer's disease from healthy aging: evidence from functional MRI. Proceedings of the National Academy of Sciences of the United States of America 101, 4637-4642. Hamilton, J.P., Chen, G., Thomason, M.E., Schwartz, M.E., Gotlib, I.H., 2011. Investigating neural primacy in major depressive disorder: multivariate Granger causality analysis of resting-state fMRI time-series data. Molecular psychiatry 16, 763-772. Hasler, G., Northoff, G., 2011. Discovering imaging endophenotypes for major depression. Molecular psychiatry 16, 604-619. Heckerman, D., 2008. A tutorial on learning with Bayesian networks. Innovations in Bayesian Networks. Springer, pp. 33-82. Hesterberg, T., Moore, D.S., Monaghan, S., Clipson, A., Epstein, R., 2005. Bootstrap methods and permutation tests. Introduction to the Practice of Statistics 5, 1-70. Hirschfeld, R., Vornik, L.A., 2005. Bipolar disorder--costs and comorbidity. The American journal of managed care 11, S85-90. Hirschfeld, R.M., Lewis, L., Vornik, L.A., 2003. Perceptions and impact of bipolar disorder: how far have we really come? Results of the national depressive and manic-depressive association 2000 survey of individuals with bipolar disorder. The Journal of clinical psychiatry, 161-174. Hosokawa, T., Momose, T., Kasai, K., 2009. Brain glucose metabolism difference between bipolar and unipolar mood disorders in depressed and euthymic states. Progress in Neuro-Psychopharmacology and Biological Psychiatry 33, 243-250.

Judd, L.L., Schettler, P.J., Akiskal, H.S., Maser, J., Coryell, W., Solomon, D., Endicott, J., Keller, M., 2003. Long-term symptomatic status of bipolar I vs. bipolar II disorders. The International Journal of Neuropsychopharmacology 6, 127-137. Killgore, W.D., Gruber, S.A., Yurgelun-Todd, D.A., 2008. Abnormal cortico-striatal activity during fear perception in bipolar disorder. Neuroreport 19, 1523. Lee, T.-W., Girolami, M., Sejnowski, T.J., 1999. Independent component analysis using an extended infomax algorithm for mixed subgaussian and supergaussian sources. Neural computation 11, 417-441. Liao, W., Mantini, D., Zhang, Z., Pan, Z., Ding, J., Gong, Q., Yang, Y., Chen, H., 2010. Evaluating the effective connectivity of resting state networks using conditional Granger causality. Biological cybernetics 102, 57-69. Liu, C.-H., Ma, X., Wu, X., Li, F., Zhang, Y., Zhou, F.-C., Wang, Y.-J., Tie, C.-L., Zhou, Z., Zhang, D., 2012. Resting-state abnormal baseline brain activity in unipolar and bipolar depression. Neuroscience letters 516, 202-206. Liu, C.-H., Ma, X., Wu, X., Zhang, Y., Zhou, F.-C., Li, F., Tie, C.-L., Dong, J., Wang, Y.-J., Yang, Z., 2013. Regional homogeneity of resting-state brain abnormalities in bipolar and unipolar depression. Progress in Neuro-Psychopharmacology and Biological Psychiatry 41, 52-59. Malhi, G.S., Lagopoulos, J., Owen, A.M., Ivanovski, B., Shnier, R., Sachdev, P., 2007. Reduced activation to implicit affect induction in euthymic bipolar patients: an fMRI study. Journal of Affective Disorders 97, 109-122. Mclntosh, A., Gonzalez

Lima, F., 1994. Structural equation modeling and its application to network

analysis in functional brain imaging. Human brain mapping 2, 2-22. Murray, C.J., Lopez, A.D., 1996. Evidence-Based Health Policy---Lessons from the Global Burden of Disease Study. Science 274, 740-743. Pearl, J., 2000. Causality: models, reasoning and inference. Cambridge Univ Press. Peng, J., Liu, J., Nie, B., Li, Y., Shan, B., Wang, G., Li, K., 2011. Cerebral and cerebellar gray matter reduction in first-episode patients with major depressive disorder: a voxel-based morphometry study. European journal of radiology 80, 395-399. Phillips, M.L., Drevets, W.C., Rauch, S.L., Lane, R., 2003. Neurobiology of emotion perception I: The neural basis of normal emotion perception. Biological psychiatry 54, 504-514. Raichle, M.E., MacLeod, A.M., Snyder, A.Z., Powers, W.J., Gusnard, D.A., Shulman, G.L., 2001. A default mode of brain function. Proceedings of the National Academy of Sciences 98, 676-682. Raichle, M.E., Snyder, A.Z., 2007. A default mode of brain function: a brief history of an evolving idea. Neuroimage 37, 1083-1090. Schlösser, R., Gesierich, T., Kaufmann, B., Vucurevic, G., Hunsche, S., Gawehn, J., Stoeter, P., 2003. Altered effective connectivity during working memory performance in schizophrenia: a study with fMRI and structural equation modeling. Neuroimage 19, 751-763. Sheline, Y.I., Barch, D.M., Price, J.L., Rundle, M.M., Vaishnavi, S.N., Snyder, A.Z., Mintun, M.A., Wang, S., Coalson, R.S., Raichle, M.E., 2009. The default mode network and self-referential processes in depression. Proceedings of the National Academy of Sciences 106, 1942-1947. Shimizu, S., Hoyer, P.O., Hyvärinen, A., Kerminen, A., 2006. A linear non-Gaussian acyclic model for causal discovery. The Journal of Machine Learning Research 7, 2003-2030. Shimizu, S., Inazumi, T., Sogawa, Y., Hyvärinen, A., Kawahara, Y., Washio, T., Hoyer, P.O., Bollen, K., 2011. DirectLiNGAM: A direct method for learning a linear non-Gaussian structural equation model. The Journal of Machine Learning Research 12, 1225-1248.

Shimizu, S., Kano, Y., 2008. Use of non-normality in structural equation modeling: Application to direction of causation. Journal of Statistical Planning and Inference 138, 3483-3491. Smith, S.M., Miller, K.L., Salimi-Khorshidi, G., Webster, M., Beckmann, C.F., Nichols, T.E., Ramsey, J.D., Woolrich, M.W., 2011. Network modelling methods for FMRI. Neuroimage 54, 875-891. Sonty, S.P., Mesulam, M.-M., Weintraub, S., Johnson, N.A., Parrish, T.B., Gitelman, D.R., 2007. Altered effective connectivity within the language network in primary progressive aphasia. The Journal of Neuroscience 27, 1334-1345. Spirtes, P., Glymour, C.N., Scheines, R., 2000. Causation, prediction, and search. MIT press. Sprengelmeyer, R., Steele, J.D., Mwangi, B., Kumar, P., Christmas, D., Milders, M., Matthews, K., 2011. The insular cortex and the neuroanatomy of major depression. Journal of Affective Disorders 133, 120-127. Strakowski, S., Delbello, M., Adler, C., 2004. The functional neuroanatomy of bipolar disorder: a review of neuroimaging findings. Molecular psychiatry 10, 105-116. Strakowski, S.M., Adler, C.M., Holland, S.K., Mills, N.P., DelBello, M.P., Eliassen, J.C., 2014. Abnormal FMRI brain activation in euthymic bipolar disorder patients during a counting Stroop interference task. Versace, A., Almeida, J.R., Quevedo, K., Thompson, W.K., Terwilliger, R.A., Hassel, S., Kupfer, D.J., Phillips, M.L., 2010. Right orbitofrontal corticolimbic and left corticocortical white matter connectivity differentiate bipolar and unipolar depression. Biological psychiatry 68, 560-567. Wu, X., Li, R., Fleisher, A.S., Reiman, E.M., Guan, X., Zhang, Y., Chen, K., Yao, L., 2011. Altered default mode network connectivity in Alzheimer's disease—a resting functional MRI and Bayesian network study. Human brain mapping 32, 1868-1881. Xu, L., Fan, T., Wu, X., Chen, K., Guo, X., Zhang, J., Yao, L., 2014. A pooling-LiNGAM algorithm for effective connectivity analysis of fMRI data. Frontiers in computational neuroscience 8. Yatham, L.N., Kennedy, S.H., Schaffer, A., Parikh, S.V., Beaulieu, S., O'Donovan, C., MacQueen, G., McIntyre, R.S., Sharma, V., Ravindran, A., 2009. Canadian Network for Mood and Anxiety Treatments (CANMAT) and International Society for Bipolar Disorders (ISBD) collaborative update of CANMAT guidelines for the management of patients with bipolar disorder: update 2009. Bipolar Disorders 11, 225-255. Yurgelun

Todd, D.A., Gruber, S.A., Kanayama, G., Killgore, W.D., Baird, A.A., Young, A.D., 2000.

fMRI during affect discrimination in bipolar affective disorder. Bipolar Disorders 2, 237-248. Zheng, X., Rajapakse, J.C., 2006. Learning functional structure from fMR images. Neuroimage 31, 1601-1613.

Figure 1. Axial images show the DMN of the BD group. T score bar is shown on the right (FDR, p=0.005). ‘aDMN’ and ‘pDMN’ represent the anterior and posterior portions of the DMN separately. Figure 2. Axial images show the DMN of the UD group. T score bar is shown on the right (FDR, p=0.005). ‘aDMN’ and ‘pDMN’ represent the anterior and posterior portions of the DMN separately. Figure 3. Difference in the functional connectivity of DMN between BD and UD (BD>UD), two sample t-test (p=0.001, uncorrected) Color bar represents T value. Figure 4. Difference in the functional connectivity of DMN between BD and UD (BDUD (left), BD
Table 1 Demographic statistics and clinical features of the patients Variables (Mean±SD)

BD (n=17)

UD (n=17)

P value

Gender(M:F) 6:11 5:12 0.7101 Age(years) 32.12±8.57 32.59±9.77 0.3032 Age range(years) 20-53 21-57 Education level(years) 4.59±1.58 4.47±1.50 0.421b Number of depressive episodes 3.59±3.89 2.53±1.28 0.294 Duration of illness(years) 9.35±8.31 7.26±8.20 0.464 HAMD 22.35±3.10 21.41±3.84 HAMA 9.59±6.21 15.59±9.46 Monotherapy 7 21 SSRI 3 16 SNRI 3 3 Mirtazapin 0 1 Trazodone 1 1 combined therapy 1 2 SSRI+AAP 1 2 Medication Free 1 6 SD: Standard deviation, BD: bipolar depression, UD: unipolar depression, HAMD: Hamilton Depression Rating Scale, HAMA: Hamilton Anxiety Scale, SSRI: Selective serotonin reuptake inhibitor, SNRI: serotonin-norepinephrine reuptake inhibitor, APP: atypical antipsychotics.  1

p

2

p



value of the chi-square test value of the F-test

Table 2 Regions of functional connectivity differences between BD and UD group (Two sample t-test, p=0.001, uncorrected) MNI coordinate Brain region Side BD>UD lMTC MPFC PCC lIPC rIPC BD
BA

T value x

y

z

Number of voxels

left right left left right

21 9 23 39 39

4.6 6.54 6.1 5.34 4.35

-63 18 -6 -51 48

-27 51 -54 -60 -75

-15 6 15 30 39

61 267 76 56 30

right left

13 21

4.67 5.65

39 -45

-30 -54

18 3

34 76

BA: Brodmann area, MNI coordinate: Montreal Neurological Institute coordinate, BD: bipolar depression, UD: unipolar depression, lMTC: left middle temporal cortex, MPFG: medial prefrontal cortex, PCC: posterior cingulate cortex, lIPC: left inferior parietal cortex, rIPC: right inferior parietal cortex, rInsula: right insula.

Table 3 (a) The information of ROIs in the BD group (one sample t test, FDR, p=0.005) Brain region PCC lIPC MPFC rIPC rITC lITC lHC rHC rInsula lMTC

BA 31 39 10 40 21 20 28 35 13 21

MNI coordinate

T value 10.47 13.55 22.1 8.8 12.33 9.5 5.59 4.85 3.44 5.65

x 9 -45 3 51 61 -61 -9 12 36 -58

y -51 -66 51 -51 -13 -12 -27 -27 -21 -28

z 17 39 3 30 -23 -24 0 3 18 -15

Number of voxels 1514 180 7234 14 111 82 177 27 27 100

Table 3 (b) The information of ROIs in the UD group (one sample t test, FDR, p=0.005) MNI coordinate Brain region PCC lIPC MPFC rIPC rITC lITC lHC rHC rInsula lMTC

BA 30 39 10 40 21 40 28 35 13 21

x

y

z

Number of voxels

0 -48 -3 54 54 -59 -9 12 36 -58

-55 -69 48 -51 -60 -15 -27 -27 -21 -28

19 39 -6 23 27 -24 0 3 18 -15

519 154 373 163 118 33 17 7 27 112

T value 10.7 6.92 14.8 6.9 7.71 5.38 5.59 4.85 3.44 5.65

BA: Brodmann area, MNI coordinate: Montreal Neurological Institute coordinate, PCC: Posterior cingulate cortex, MPFC: Medial prefrontal cortex, lIPC/rIPC: Left/right inferior parietal cortex, lITC/rITC: Left/right inferior temporal cortex, lHC/rHC: Left/right hippocampus cortex, rInsula: right insula, lMTC: left middle temporal cortex.

Table 4 (a) Weight coefficients matrix of BD in LiNGAM effective connectivity model BD

PCC

lIPC

MPFC

rIPC

rITC

lITC

lHC

rHC

rInsula

lMTC

PCC

0

-1.279

0

0

0

0

1.085

0

0

1.087

lIPC

0

0

0

0

0

0

0

0

0

0

MPFC

3.420

4.375

0

0

0

0

-3.713

0

-1.182

-3.719

rIPC

-4.028

-5.152

1.177

0

0

0

4.372

0

1.392

4.379

rITC

-13.338

-17.069

3.900

-3.312

0

1.344

14.478

3.274

4.615

14.509

lITC

9.914

12.681

-2.898

2.461

0

0

-10.760

-2.432

-3.428

-10.779

lHC

0

1.179

0

0

0

0

0

0

0

-1.002

rHC

4.078

5.215

-1.192

1.012

0

0

-4.426

0

-1.409

-4.433

rInsula

2.893

3.700

0

0

0

0

-3.140

0

0

-3.145

lMTC

0

1.176

0

0

0

0

0

0

0

0

Table 4 (b) Weight coefficients matrix of UD in LiNGAM effective connectivity model UD

PCC

lIPC

MPFC

rIPC

rITC

lITC

lHC

rHC

rInsula

lMTC

PCC

0

0

0

0

0

0

0.476

0.812

0

0

lIPC

1.393

0

0

0

0

0

-0.594

0.399

0

0

MPFC

-1.630

-0.057

0

0.981

0.812

0

-0.181

0.280

-0.024

0

rIPC

0.738

0.151

0

0

0

0

0.358

-0.132

0

0

rITC

-1.422

-0.080

0

0.713

0

0

2.493

-0.523

0.880

0

lITC

-1.280

0.934

-0.141

0.713

-0.702

0

1.562

-1.118

1.384

0

lHC

0

0

0

0

0

0

0

0

0

0

rHC

0

0

0

0

0

0

0.293

0

0

0

rInsula

1.841

0.630

0

-0.839

0

0

1.187

-0.775

0

0

lMTC

0.234

0.115

0.486

-0.701

-0.038

0.245

0.827

-0.045

0.276

0

Table 5 Results of different connections between BD and UD group by permutation test (p=0.05). BD>UD MPFCPCC rIPCMPFC rIPClHC rIPCrInsula rITClITC lITCrIPC lHClIPC rHCPCC

probabilities 0.0047 0.0147 0.0083 0.0207 0.0143 0.0303 0.0003 0.0033

BD
probabilities 0.0143 0.0390 0.0006 0.0007 0.0073 0.0003 0.0120 0.0057



Highlights z

Brain regions e.g. hippocampus and insula have difference in EC between BD and UD.

z

The altered connections might be special markers for distinguishing BD from UD.

z

The effective connectivity models were constructed using ICA-LiNGAM.

Acknowledgement The authors gratefully acknowledge Beijing Normal University Imaging Center for Brain Research for the contributions in MRI data acquisition.

Conflict of Interest The authors declared that they have no conflicts of interest to this work.

Contributors 1. Dr. Xia Wu, Li Yao, Zhiying Long designed the study. 2. Dr. Jiacai Zhang and Xiaojuan Guo collected the original imaging data. 3. Dr. Xia Wu, Yunting Liu and Xiaojuan Guo managed and analyzed the imaging data. 4. Yunting Liu and Xia Wu wrote the manuscript. 5. All authors contributed to and have approved the final manuscript.

Role of the Funding Source This work was supported by the Funds for International Cooperation and Exchange of the National Natural Science Foundation of China (61210001), the general Program of National Natural Science Foundation of China (61222113), Program for New Century Excellent Talents in University (NCET-12-0056) and the Fundamental Research Funds for the Central Universities.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6