3.2.1 Lateral Ventricles

Patients with manic depressive illness were first reported to display increased ventricular area compared with controls in a computerized axial tomography study in 1985 (2). Since then, there has been consistent evidence from numerous MRI studies that enlarged volume of the lateral ventricles characterizes bipolar disorder, emphatically confirmed by systematic reviews and by combining datasets through meta- and mega-analyses (3–8). In contrast to schizophrenia (9), there is evidence for more prominent ventriculomegaly on the right side (4, 5, 7), the reason for which is unclear, but it echoes reports of cortical and subcortical right hemisphere pathologies being more frequently associated with bipolar disorder (10).

3.2.2 Subcortical Structures

Many imaging studies of bipolar disorder have focused on subcortical structures, especially anterior limbic system structures, given their key role in emotional regulation. Recent meta- and mega-analyses have highlighted the markedly heterogeneous nature of studies into volume deviation of the hippocampus and amygdala, reported variously in individual studies as being increased, decreased, or unchanging in volume compared with controls, with no overall change when these studies were combined (4, 7, 11). There is also evidence from meta-analytical studies for increased striatal volumes in bipolar disorder, including right putamen (7), left putamen (12), and globus pallidus (6). As well as clinical heterogeneity, methodological heterogeneity may be contributing to the mixed results for small gray matter structures such as the hippocampus, which are difficult to segment precisely using fully automated techniques (13).

The largest international collaborative meta-analytic combination of such data to date is through the Bipolar Disorder Working Group of ENIGMA (Enhancing Neuroimaging Genetics Through Meta-analysis) by Hibar et al. (8), which incorporated MRI data from 1,710 bipolar disorder patients and 2,594 healthy controls and a consistent image segmentation process. This study identified a small but significant volume reduction of the hippocampus, amygdala, and thalamus in bipolar disorder. This small bilateral hippocampal volume reduction is consistent with the most recent systematic review of twenty-one published studies, which reported that the deficit was more pronounced in early-onset cases (14). The weight of this current evidence toward medial temporal lobe and thalamic deficits in bipolar disorder is in keeping with the functional neuroanatomy of these limbic structures and their role in the neurocircuitry of emotional processing and declarative memory, which are characteristically impaired in the illness.

3.2.3 Cortical Regions

Whereas global cerebral volume is generally preserved in bipolar disorder in most – although not all (11) – meta-analyses (4, 5, 7), there is evidence of regional gray matter deficit in the frontal cortex (6). Several meta-analyses of whole brain voxel-based morphometry studies in bipolar disorder compared with controls using differing methodologies have now been conducted. These have reported further gray matter deficits in the bilateral insula and anterior cingulate (15), the fronto-insular cortex (16, 17), bilateral ventrolateral and right dorsolateral prefrontal gray matter (18), right-sided frontotemporal gray matter incorporating prefrontal cortex, anterior temporal cortex, claustrum, and insula (19), and in the left medial frontal gyrus and right inferior/precentral gyri incorporating the insula (12).

Other large-scale studies have focused on regional parcellated cortical volume and estimates of cortical thickness and surface area. A systematic review of seventeen studies of cortical thickness in bipolar disorder identified illness-related decreased cortical thickness in bilateral prefrontal regions, left anterior cingulate, paracingulate, and superior temporal gyrus (20). The ENIGMA consortium completed a highly powered analysis of individual-level data on a cohort of 1,837 participants with bipolar disorder and 2,582 healthy controls, conducted with a harmonized software processing pipeline (21). Bipolar patients displayed a widespread pattern of bilaterally reduced thickness in frontal, temporal, and parietal regions, with the largest effect in the left pars opercularis, left fusiform gyrus, and left rostral middle frontal cortex. Cortical surface area differences were not found between adult patients with bipolar disorder and healthy controls (21), indicating that the cortical volume loss associated with bipolar disorder is consequential to the reduced thickness and preserved surface area. Taken together, these cortical studies identify illness-related gray matter deficits in paralimbic, frontotemporal, and prefrontal cortex anatomical regions subserving emotional processing, attentional, and executive functions known to be abnormal in bipolar disorder.

3.2.4 White Matter

With the advent of MRI technology in the early 1990s, studies began reporting the excessive presence of qualitatively assessed white matter hyperintensities in bipolar disorder (22, 23). Subsequent meta-analyses demonstrated that bipolar disorder was associated with a threefold increase in the rates of deep white matter hyperintensities compared with controls (5, 24), more marked in the right hemisphere and frontoparietal regions (5), and suggestive of white matter damage disrupting brain connectivity in bipolar illness.

Meta-analyses have reported that bipolar disorder is associated with a reduced area of the corpus callosum, the largest white matter inter-hemispheric pathway responsible for the integration of inter-hemispherical information (5, 11). A recent international multicenter study indicated that callosal area reductions in bipolar disorder are most prominent in the posterior sections of the corpus callosum (25).

As with gray matter volume, meta-analyses indicate that global white matter volume appears to be preserved in bipolar disorder (4–6), but there is evidence for regional white matter deficit. A meta-analysis of voxel-based morphometry studies (12) reported a reduction in white matter density in the left corona radiata, inferior longitudinal fasciculus, and posterior cingulum. A recent meta-analysis of voxel-based studies using seed-based mapping analysis of white matter (26), including 765 patients with bipolar disorder and 1,055 healthy controls, reported a large region of decreased white matter volume in the posterior corpus callosum and posterior cingulate gyrus in bipolar disorder.

3.2.5 Diffusion Tensor Imaging

Diffusion tensor imaging (DTI) can be used as an indirect measure of white matter microstructural organization by measuring directional constraint of the diffusivity of water molecules due to the local cellular environment (27). The most commonly employed metric to quantify white matter organization is fractional anisotropy (FA), which represents the level of regional organization within fiber bundles (28). Studies in bipolar disorder have repeatedly identified reduced regional FA in patients compared with healthy volunteers. The first meta-analysis conducted by Vederine and colleagues (29) of ten whole brain DTI studies, employing an activation likelihood estimation technique, demonstrated two significant clusters of decreased FA in the right hemisphere in bipolar disorder. One of these regions, close to the parahippocampal gyrus, has a role in subprocesses associated with automatic emotion regulation (30), and the other, close to the right anterior cingulate cortex and subgenual cortex, is important in the identification of emotionally salient stimuli and automatic emotion regulation (30, 31). A further systematic review and meta-analysis using effect size-signed differential mapping of fifteen whole brain DTI studies reported widespread FA reductions in bipolar disorder across commissural, association, and projection tracts, with the meta-analysis identifying FA reductions in the right parieto-occipital, left mid-posterior cingulate, and left anterior cingulate white matter (32). The specific white matter tracts involved in these regions of FA deficit include the long association tracts of the inferior longitudinal fasciculus and inferior frontal-occipital fasciculus, as well as anterior limbic system tracts, and indicate that white matter microstructural abnormalities might underpin the cognitive deficits as well as affective deficits linked to bipolar disorder (32). A further meta-analysis of voxel-based DTI studies, employing anisotropic effect size-signed differential mapping and including eighteen studies, identified FA deficits in bipolar patients that incorporated the left cingulum, right anterior superior longitudinal fasciculus, and genu, extending to the frontolimbic tracts, including the uncinate fasciculus (33).

DTI studies also confirm abnormal white matter integrity in all divisions of the corpus callosum in patients with bipolar disorder compared to healthy controls (29, 32–36), extending callosal area studies to further implicate disrupted interhemispheric communication in the illness. Taken together, these DTI studies indicate that bipolar disorder is associated with widespread microstructural disorganization of white matter tracts consistent with structural dysconnectivity in anatomical regions underpinning the emotional dysregulation and cognitive dysfunction associated with the disorder (37, 38).

3.2.6 Structural Network Findings

The abnormalities described thus far in gray and white matter regions are focal; however, the brain functions via a series of interconnected neuroanatomical networks. The “dysconnectivity” theory postulates that major psychotic illnesses can be explained by impaired integration between brain regions, rather than specific focal brain abnormalities (39). Through graph theory, it is now possible to investigate topology within the brain’s global structural connectivity network in vivo using data derived from structural MRI to define cortical and subcortical gray matter regions (“nodes”) and from diffusion MRI to define the white matter tracts interconnecting these regions (“edges”)(40). Such structural connectivity investigations comparing patients with bipolar disorder and controls report evidence of impaired integration (34, 41–43) and segregation (34, 42–44) in bipolar disorder. Specific brain networks found to have abnormal anatomical connectivity include those incorporating left orbitofrontal cortex, left hippocampus, bilateral isthmus cingulate (34), left cuneus, right cerebellum, inferior frontal gyrus, right calcarine gyrus (43), superior and middle frontal gyri (42, 44), and superior and middle occipital gyri (42). Furthermore, there is evidence from these network analyses of impaired inter-hemispheric integration in bipolar disorder (34, 41, 45, 46), with interhemispheric dysconnectivity especially prominent in the frontal lobes (34). Rich club connectivity, which plays an important role in integrating information across functionally specialized neural circuits (47), is also reported to be reduced in bipolar disorder (42, 43), although there is conflicting evidence for this (41). Taken together, these findings provide network-level evidence for altered anatomical brain connectivity in BD that disrupts global integration and local segregation and extends across anterior, posterior, and interhemispheric regional networks.

3.4.1 Psychotropic Medication

Considerable attention has been given to the effect of medication use on brain structure in bipolar disorder. The effect of lithium, in particular, has been investigated extensively in post hoc analyses, in part because of the preclinical evidence that lithium activates neurotrophic and neuroprotective pathways and associated signaling mechanisms (61), which may be detectable macroscopically using MRI. An early meta-analysis by Kempton et al. (5) included a meta-regression which demonstrated that those studies with a higher proportion of patients taking lithium reported higher cerebral gray matter for patients. Hallahan and colleagues’ mega-analysis (7) confirmed an increase of global cerebral volume in bipolar patients taking lithium at the time of scanning. Besides, they demonstrated that such patients had larger bilateral hippocampal and amygdala volume than controls, whereas patients not taking lithium had volume deficit of these structures. Another recent meta-analysis (62) based on fifteen studies reported that global gray matter was significantly larger in patients treated with lithium compared with patients who were not treated with lithium. A large case-control study based on 266 patients and 171 healthy volunteers reported that patients on lithium had a significantly larger total brain, thalamus, putamen, pallidum, hippocampus, and accumbens volumes compared to lithium-free patients (63). The ENIGMA meta-analyses reported larger thalamic volume in patients treated with lithium compared with patients not taking lithium (8), as well as significantly increased cortical thickness in patients taking lithium, most prominently in the left paracentral gyrus and left and right superior parietal gyrus, and increased surface area of the left paracentral lobule (21). The small number of longitudinal studies performed in bipolar disorder also indicate that treatment with lithium is associated with volume increases in gray matter, prefrontal gray matter, and hippocampal volume (64–66).

In contrast, the ENIGMA meta-analyses reported reduced hippocampal volume (8) and reduced cortical thickness in the left and right lateral occipital gyrus and right paracentral gyrus (21) in patients taking antiepileptic mood stabilizers compared with patients not taking antiepileptic medications. However, other individual studies report no differences or even gray matter increases in the prefrontal cortex and anterior cingulate in bipolar patients treated with antiepileptic mood stabilizers [67, 68]. Studies report antipsychotic medication use is associated with reduced gray matter volume in schizophrenia (69, 70). Similarly, there is evidence for reduced cortical surface area with atypical antipsychotic treatment in bipolar patients, compared with those patients not taking atypical antipsychotics, in the right rostral middle frontal gyrus and right superior frontal gyrus in the ENIGMA meta-analysis (21). Arnone and colleagues (6) reported a significant association between antipsychotic use in bipolar disorder and reduced volume of gray matter and right amygdala. A reduction of volume in the right amygdala was also associated with antidepressant exposure in this meta-analysis (6).

The effect of psychotropic medications on white matter metrics has been less widely investigated than gray matter structures. In the largest single-site DTI study to date of patients with bipolar disorder and healthy volunteers that explored the impact of pharmacotherapy, Abramovic and colleagues (71) reported that patients off lithium showed a significant lower fractional anisotropy values than patients on lithium in the corpus callosum, fornix, and the major and minor forceps. Although evidence is mixed to date, some other individual DTI studies have also reported a normalizing effect of lithium on fractional anisotropy reductions in bipolar disorder (68). No differences have been reported in network-level connectivity measures between patients on and off lithium (41, 42), nor in patients exposed or not exposed to antipsychotic medications (41).

Overall, evidence from both large-scale cross-sectional and longitudinal structural neuroimaging studies in bipolar disorder are that medication use is a significant source of heterogeneity, and that treatment with lithium has an ameliorating effect on gray matter and subcortical structures, possibly through the neuroprotective properties of this medication, and might also attenuate white matter aberrations. Whereas it is difficult to separate medication use from clinical course characteristics in observational cross-sectional studies (e.g., patients with more severe or persistent symptoms may be more likely to be prescribed antipsychotic medications long term and also have more cortical thinning), the evidence to date from neuroimaging studies indicates that antiepileptic mood stabilizers have a less ameliorating effect than lithium, and that antipsychotic medications are associated with gray matter deficits.

3.4.2 Demographic and Clinical Variables

Increasing age has been linked in the ENIGMA meta-analyses to proportionately greater hippo-campal volume deficits in bipolar disorder (8) and reduced cortical thickness more prominently in the left rostral middle frontal gyrus (21). These effects are subtle, however, and indicate that the accelerated aging of the brain reported in schizophrenia is not mirrored in bipolar disorder (72, 73). An association with gender was also identified for subcortical structures, with increased thalamic volume in female patients with bipolar disorder (8). However, there was no impact of gender on cortical volume or thickness (17, 21). This is in contrast to schizophrenia where male dominated samples with poorer prognosis, and more neurodevelopmental compromise may be partially driving the more substantial gray matter deficits in this syndrome than a bipolar disorder (17). Analyses of white matter metrics, such as through volume or network analyses, have mostly not reported links with age or gender (26, 34, 41, 46). However, there are individual studies that do report associations between increasing age and reduced fractional anisotropy in frontal tracts (74).

Age of onset and duration of illness are often assessed as sources of variation when analyzing changes in cortical and subcortical regions in patients with bipolar disorder. Hallahan et al.’s mega-analysis (7) reported a significant association between earlier age of onset in patients with bipolar disorder and reduced cerebral volume and left thalamic volume, as well as increased left amygdala volume. There was no association between age of onset and subcortical volume in the ENIGMA meta-analysis (8); however, reduced cortical thickness was associated with longer illness duration, with the strongest effect present bilaterally in the pericalcarine gyrus, left rostral anterior cingulate gyrus, and right cuneus, while a significant association with increased thickness was found in the right entorhinal gyrus (21). Longer duration of illness has been associated with more cerebral gray matter volume loss (6), but with increased gray matter in limbic system structures such as the amygdala, thalamus, and anterior cingulate (6,16). Of course, other variables correlated with duration of illness (such as the amount of time on lithium or other mood stabilizers) may be driving the associations reported between increased duration of illness and increased volume of anterior limbic system structures. Pezzoli et al.’s meta-analysis (26) reported no significant association between age of onset or duration of illness and regional white matter volume, and individual DTI studies largely do not report such associations either (35, 42).

Although some individual studies report bipolar subtype differences, the largest and most statistically powerful studies failed to detect any significant difference in volume, cortical thickness, surface area, or white matter regional volume when comparing bipolar I disorder and bipolar II disorder (5, 6, 8, 21, 26) A large multicenter DTI study reported greater microstructural impairment of the corpus callosum body (36) in bipolar patients with a history of psychotic symptoms than in those patients without psychotic features.

Ultimately cross-sectional studies assessing the impact of demographics or clinical variables on brain structure in bipolar disorder are limited by their methodology, given the nonlinear trajectory of brain development, and that illness of varying severity can emerge across the age range. For example, studies on clinical subgroups suggest that childhood onset bipolar disorder is associated with more prominent and progressive gray matter deficits in emotional regulation regions and that late-onset bipolar disorder is more likely to be associated with white matter abnormalities (75). The interaction between risk factors (and ameliorative factors such as psychotropic medication) and neuroanatomy in the context of dynamic processes underpinning brain development are difficult to decipher post hoc, even with large-scale observational studies. Longitudinal neuroimaging studies with rich phenotyping are required to dissect the likely complex interplay between the progression of illness with age and with other modulating factors such as medication use, age of onset, bipolar subtype, genotypic variation, and environmental risk factors.