The findings of this study indicated decreased corticolimbic connectivity in BD patients compared with healthy subjects, similar to results previously reported for MDD; however, the abnormalities seemed to be more severe in the BD group. This is not a surprising finding as BD is a more severe illness of mood regulation than MDD. The BD subgroup also had a longer duration of illness and had had more mood episodes than the MDD group, and it was slightly older than the MDD and healthy control groups. The greater severity of mood disorder in the BD group could also explain the greater decrease in connectivity in this group. Motion was slightly greater in the bipolar group, but when used as a covariate in the analysis, it did not change the findings of the study.
An exploratory analysis was done (keeping in mind the small number of patients in each subgroup of BD) and showed that the BDD and the BDM subgroups had similar decreases in corticolimbic connectivity compared with healthy subjects. Some differences were noted, e.g. the decreased pgACC-left AMYG connectivity only in BDM and not in BDD, and the decreased pgACC-right AMYG connectivity in BDD, which will need to investigated in future studies with a larger number of subjects.
The decreased corticolimbic LFBF correlations results indicate possible decreased phase coherence between LFBF sampled in the ACC and the limbic regions in BD and MDD patients. Phase synchrony has been related to the integrity of the circuits between two brain regions (Spencer et al., 2004
). Single neuron studies with intraneuronal electrodes and, to some extent, electroencephalograhic studies have shown that if two brain regions are locked in phase with each other, their functioning is closely connected (Varela et al., 2001
). Hence, decreased phase coherence could be associated with a decreased regulatory effect of the ACC over the limbic areas leading to mood dysregulation in bipolar and unipolar depression as well as mania.
The decreased corticolimbic connectivity seen in mood disorders across diagnosis and phase of illness suggests that the decreased connectivity may be a trait abnormality. However, in a previous study (Anand et al., 2005b
), we found that antidepressant treatment leads to an increase in corticolimbic connectivity in MDD patients, and therefore the connectivity abnormality may be state-dependent. To investigate whether the decreased connectivity is state- or trait-dependent, these findings will need to be investigated in BD before and after treatment and also in unmedicated euthymic BD and MDD patients.
The ROIs were placed within the corticolimbic system based on a priori
identified and agreed upon anatomical landmarks as discussed in Section 2.4. The EPI slices that were selected on matching high-resolution T1 images were not contiguous but included only slices chosen by the radiologist to cover the ROIs. Therefore, the location of the EPI slices, the distance between the four EPI slices, and the placement of the ROIs did vary slightly from subject to subject due to differences in subjects’ anatomy and head position. To place ROIs in exactly the same location for all subjects, it would have been necessary to normalize the data into a standardized space and therefore have considerably larger slice coverage. This was not possible without a very significant increase of the TR. As discussed in Section 2, the data were acquired with a short TR to avoid aliasing effects of fluctuations in the BOLD signals due to cardiac and respiratory cycles. In future studies, to acquire data from the whole brain, methods such as recording of cardiac and respiratory cycles along with fMRI acquisition with subsequent retrospective correction for effects of these physiological variables on the BOLD signal could be used (Glover et al., 2000
The correlation of LFBF between two areas is a measure of functional connectivity, i.e. that the two are in synchrony (Friston et al., 1993
). However, this could also occur due to the influence of a third factor that may be simultaneously affecting both the areas. In the future, to measure the direct effect of one area over another, i.e. to measure effective connectivity, techniques such as structural equation modeling (SEM) (Seminowicz et al., 2004
) or newer techniques such as dynamic causal modeling (DCM) (Friston et al., 2003
) could be used. An investigation of structural connectivity using diffusion tensor imaging (DTI) could also shed light on the relationship between functional and structural connectivity.
The analysis performed here is a straightforward hypothesis-driven analysis based on an a priori
expectation of involved regions of the brain. The a priori
defined ROI approach has the advantage of reducing the magnitude of correction needed for a large number of voxels; one can correct only for a small number of ROIs, thereby considerably increasing statistical power (Poldrack, 2007
). The same analysis was performed on controls and patients, and statistically meaningful conclusions were drawn. Connectivities with other regions were not investigated, and no conclusions were drawn regarding regions that were not examined.
Out of the three a priori
identified limbic structures whose connectivity with the pgACC was investigated in this study, the connectivity of the pgACC-DMTHAL was present in all mood disorders (). This is not surprising as the thalamus is an integral part of the cingulate–pallidostriatal–thalamic–amygdala mood-regulating circuit (Taber et al., 2004
). The DMTHAL has major connections with the ACC, the ventral PST, and the AMYG, and therefore it is central to the circuit (Taber et al., 2004
). Decreases in pgACC connectivity were also seen for the AMYG. The AMYG is located in the more ventral part of the brain, and the BOLD signal from the AMYG has a lower signal-to-noise ratio due to susceptibility artifacts. Therefore, the variance for the data was greater in this region. In future studies, more sophisticated techniques using advanced hardware and techniques such as z-shimming to reduce susceptibility artifacts could be used to image the ventral areas of the brain (Glover, 1999
Another limitation of this study was that we did not measure differences in gray matter density within ROIs between groups due to the small number of subjects studied. It is possible that the ROIs may have contained more or less gray matter in the different groups, leading to partial-volume effects that could have affected the results. In future studies, measurement of gray matter density using techniques such as voxel-based morphometry (Ashburner and Friston, 2000
) could be used to investigate differences in gray matter density between groups.
The duration of medication-free period for psychiatric studies is always a compromise between what is ideal and what is clinically feasible. We chose a minimum period of 2 weeks for patients to be off medication (except for fluoxetine, for which we required a 4-week drug-free period) and inclusion criteria for no substance dependence in the past year and a negative urine drug screen at the time of screening for the study. However, long-term effects of psychotropic agents may still be present. Future studies will need to be conducted to address this issue with a larger number of subjects with longer medication-free and substance-free periods before the study.
The findings of this study are consistent with a common abnormality of corticolimbic functional connectivity in bipolar disorder and depression, and they need to be confirmed with a larger number of patients and with more sophisticated techniques to measure functional connectivity within the brain.