Advertisement

Functional Dysconnection of the Inferior Frontal Gyrus in Young People With Bipolar Disorder or at Genetic High Risk

      Abstract

      Background

      Bipolar disorder (BD) is characterized by a dysregulation of affect and impaired integration of emotion with cognition. These traits are also expressed in probands at high genetic risk of BD. The inferior frontal gyrus (IFG) is a key cortical hub in the circuits of emotion and cognitive control, and it has been frequently associated with BD. Here, we studied resting-state functional connectivity of the left IFG in participants with BD and in those at increased genetic risk.

      Methods

      Using resting-state functional magnetic resonance imaging we compared 49 young BD participants, 71 individuals with at least one first-degree relative with BD (at-risk), and 80 control subjects. We performed between-group analyses of the functional connectivity of the left IFG and used graph theory to study its local functional network topology. We also used machine learning to study classification based solely on the functional connectivity of the IFG.

      Results

      In BD, the left IFG was functionally dysconnected from a network of regions, including bilateral insulae, ventrolateral prefrontal gyri, superior temporal gyri, and the putamen (p < .001). A small network incorporating neighboring insular regions and the anterior cingulate cortex showed weaker functional connectivity in at-risk than control participants (p < .006). These constellations of regions overlapped with frontolimbic regions that a machine learning classifier selected as predicting group membership with an accuracy significantly greater than chance.

      Conclusions

      Functional dysconnectivity of the IFG from regions involved in emotional regulation may represent a trait abnormality for BD and could potentially aid clinical diagnosis.

      Keywords

      Bipolar disorder (BD) is a relatively common and highly disabling condition (
      • Whiteford H.A.
      • Degenhardt L.
      • Rehm J.
      • Baxter A.J.
      • Ferrari A.J.
      • Erskine H.E.
      • et al.
      Global burden of disease attributable to mental and substance use disorders: Findings from the Global Burden of Disease Study 2010.
      ,
      • Mitchell P.
      • Johnston A.
      • Frankland A.
      • Slade T.
      • Green M.
      • Roberts G.
      • et al.
      Bipolar disorder in a national survey using the World Mental Health Version of the Composite International Diagnostic Interview: The impact of differing diagnostic algorithms.
      ) in which causative and pathophysiological underpinnings are poorly understood. The characteristic profile of depressive and manic episodes in BD likely reflects impaired functioning of the neural substrates of emotion regulation and cognitive control. These have been hypothesized to include a constellation of regions in the medial prefrontal cortex (mPFC), the inferior frontal gyrus (IFG), anterior insula, and the dorsolateral PFC (
      • Phillips M.L.
      • Swartz H.A.
      A critical appraisal of neuroimaging studies of bipolar disorder: Toward a new conceptualization of underlying neural circuitry and a road map for future research.
      ,
      • Phillips M.
      • Ladouceur C.
      • Drevets W.
      A neural model of voluntary and automatic emotion regulation: Implications for understanding the pathophysiology and neurodevelopment of bipolar disorder.
      ). These regions appear to be crucial for the integration of emotional processing with executive function (
      • Seeley W.W.
      • Menon V.
      • Schatzberg A.F.
      • Keller J.
      • Glover G.H.
      • Kenna H.
      • et al.
      Dissociable intrinsic connectivity networks for salience processing and executive control.
      ). Dysregulation of the coordinated activity in these regions may thus underlie the phenotypic expression of BD.
      Imaging studies of BD have highlighted a particularly key role of the IFG (
      • Chen C.H.
      • Suckling J.
      • Lennox B.R.
      • Ooi C.
      • Bullmore E.T.
      A quantitative meta-analysis of fMRI studies in bipolar disorder.
      ). The IFG is involved in a number of cognitive processes of potential relevance to BD, including response inhibition, set switching, socioemotional learning, and sustained attention (
      • Liakakis G.
      • Nickel J.
      • Seitz R.
      Diversity of the inferior frontal gyrus—A meta-analysis of neuroimaging studies.
      ). A meta-analysis of functional magnetic resonance imaging (fMRI) findings in BD found attenuated activation of the IFG across a range of emotional and cognitive tasks (
      • Chen C.H.
      • Suckling J.
      • Lennox B.R.
      • Ooi C.
      • Bullmore E.T.
      A quantitative meta-analysis of fMRI studies in bipolar disorder.
      ). Similar findings implicating the IFG are now also emerging in subjects at high genetic risk of BD (
      • Brotman M.A.
      • Deveney C.M.
      • Thomas L.A.
      • Hinton K.E.
      • Yi J.Y.
      • Pine D.S.
      • et al.
      Parametric modulation of neural activity during face emotion processing in unaffected youth at familial risk for bipolar disorder.
      ,
      • Hajek T.
      • Cullis J.
      • Novak T.
      • Kopecek M.
      • Blagdon R.
      • Propper L.
      • et al.
      Brain structural signature of familial predisposition for bipolar disorder: Replicable evidence for involvement of the right inferior frontal gyrus.
      ). Our group recently reported reduced IFG activation in high-genetic-risk individuals during response inhibition to fearful stimuli in an emotional go-no go task (
      • Roberts G.
      • Green M.J.
      • Breakspear M.
      • McCormack C.
      • Frankland A.
      • Wright A.
      • et al.
      Reduced inferior frontal gyrus activation during response inhibition to emotional stimuli in youth at high risk of bipolar disorder.
      ). These studies suggest the possibility that loss of the functional integrity of the IFG may underlie trait dysfunction in BD.
      A functional disturbance may reflect local, incipient pathology or the compromised ability of that brain region to functionally integrate into larger neuronal circuits (
      • Valenzuela M.J.
      • Breakspear M.
      • Sachdev P.
      Complex mental activity and the aging brain: Molecular, cellular and cortical network mechanisms.
      ,
      • Seeley W.W.
      • Crawford R.K.
      • Zhou J.
      • Miller B.L.
      • Greicius M.D.
      Neurodegenerative diseases target large-scale human brain networks.
      ,
      • Fornito A.
      • Zalesky A.
      • Breakspear M.
      The connectomics of brain disorders.
      ). Resting-state fMRI (rs-fMRI) has proven to be an effective way of assessing the integrity of brain circuits in psychiatric disorders and eschews the need for complex cognitive tasks (
      • Avery J.A.
      • Drevets W.C.
      • Moseman S.E.
      • Bodurka J.
      • Barcalow J.C.
      • Simmons W.K.
      Major depressive disorder is associated with abnormal interoceptive activity and functional connectivity in the insula.
      ,
      • Biswal B.B.
      Resting state fMRI: A personal history.
      ). Prior analyses of rs-fMRI data in BD have demonstrated disrupted connections between the PFC and limbic-related structures (
      • Torrisi S.
      • Moody T.D.
      • Vizueta N.
      • Thomason M.E.
      • Monti M.M.
      • Townsend J.D.
      • et al.
      Differences in resting corticolimbic functional connectivity in bipolar I euthymia.
      ,
      • Mamah D.
      • Barch D.M.
      • Repovš G.
      Resting state functional connectivity of five neural networks in bipolar disorder and schizophrenia.
      ,
      • Liu H.
      • Tang Y.
      • Womer F.
      • Fan G.
      • Lu T.
      • Driesen N.
      • et al.
      Differentiating patterns of amygdala-frontal functional connectivity in schizophrenia and bipolar disorder.
      ,
      • Dickstein D.P.
      • Gorrostieta C.
      • Ombao H.
      • Goldberg L.D.
      • Brazel A.C.
      • Gable C.J.
      • et al.
      Fronto-temporal spontaneous resting state functional connectivity in pediatric bipolar disorder.
      ,
      • Chepenik L.G.
      • Raffo M.
      • Hampson M.
      • Lacadie C.
      • Wang F.
      • Jones M.M.
      • et al.
      Functional connectivity between ventral prefrontal cortex and amygdala at low frequency in the resting state in bipolar disorder.
      ,
      • Anticevic A.
      • Brumbaugh M.S.
      • Winkler A.M.
      • Lombardo L.E.
      • Barrett J.
      • Corlett P.R.
      • et al.
      Global prefrontal and fronto-amygdala dysconnectivity in bipolar I disorder with psychosis history.
      ,
      • Anand A.
      • Li Y.
      • Wang Y.
      • Lowe M.J.
      • Dzemidzic M.
      Resting state corticolimbic connectivity abnormalities in unmedicated bipolar disorder and unipolar depression.
      ,
      • Chai X.J.
      • Whitfield-Gabrieli S.
      • Shinn A.K.
      • Gabrieli J.D.
      • Nieto Castañón A.
      • McCarthy J.M.
      • et al.
      Abnormal medial prefrontal cortex resting-state connectivity in bipolar disorder and schizophrenia.
      ). Other studies have documented altered patterns of functional connectivity within the so-called default mode (
      • Öngür D.
      • Lundy M.
      • Greenhouse I.
      • Shinn A.K.
      • Menon V.
      • Cohen B.M.
      • et al.
      Default mode network abnormalities in bipolar disorder and schizophrenia.
      ,
      • Calhoun V.D.
      • Maciejewski P.K.
      • Pearlson G.D.
      • Kiehl K.A.
      Temporal lobe and “default” hemodynamic brain modes discriminate between schizophrenia and bipolar disorder.
      ,
      • Calhoun V.D.
      • Sui J.
      • Kiehl K.
      • Turner J.A.
      • Allen E.A.
      • Pearlson G.
      Exploring the psychosis functional connectome: Aberrant intrinsic networks in schizophrenia and bipolar disorder.
      ,
      • Ford K.A.
      • Théberge J.
      • Neufeld R.J.
      • Williamson P.C.
      • Osuch E.A.
      Correlation of brain default mode network activation with bipolarity index in youth with mood disorders.
      ), a network of posterior and midline regions that become more active during internally generated cognition (
      • Raichle M.E.
      • MacLeod A.M.
      • Snyder A.Z.
      • Powers W.J.
      • Gusnard D.A.
      • Shulman G.L.
      A default mode of brain function.
      ). To date, there have been three studies of rs-fMRI in psychotic BD patients and their unaffected relatives (
      • Khadka S.
      • Meda S.A.
      • Stevens M.C.
      • Glahn D.C.
      • Calhoun V.D.
      • Sweeney J.A.
      • et al.
      Is aberrant functional connectivity a psychosis endophenotype? A resting state functional magnetic resonance imaging study.
      ,
      • Meda S.A.
      • Gill A.
      • Stevens M.C.
      • Lorenzoni R.P.
      • Glahn D.C.
      • Calhoun V.D.
      • et al.
      Differences in resting-state functional magnetic resonance imaging functional network connectivity between schizophrenia and psychotic bipolar probands and their unaffected first-degree relatives.
      ,
      • Lui S.
      • Yao L.
      • Xiao Y.
      • Keedy S.
      • Reilly J.
      • Keefe R.
      • et al.
      Resting-state brain function in schizophrenia and psychotic bipolar probands and their first-degree relatives.
      ). These studies found both shared and unique resting-state network connectivity in probands with psychotic BD or schizophrenia and their unaffected relatives (
      • Khadka S.
      • Meda S.A.
      • Stevens M.C.
      • Glahn D.C.
      • Calhoun V.D.
      • Sweeney J.A.
      • et al.
      Is aberrant functional connectivity a psychosis endophenotype? A resting state functional magnetic resonance imaging study.
      ,
      • Meda S.A.
      • Gill A.
      • Stevens M.C.
      • Lorenzoni R.P.
      • Glahn D.C.
      • Calhoun V.D.
      • et al.
      Differences in resting-state functional magnetic resonance imaging functional network connectivity between schizophrenia and psychotic bipolar probands and their unaffected first-degree relatives.
      ,
      • Lui S.
      • Yao L.
      • Xiao Y.
      • Keedy S.
      • Reilly J.
      • Keefe R.
      • et al.
      Resting-state brain function in schizophrenia and psychotic bipolar probands and their first-degree relatives.
      ). Atypical patterns of prefrontal and subcortical intrinsic resting-state connectivity have also been identified in offspring of patients with BD compared with control subjects (
      • Singh M.K.
      • Chang K.D.
      • Kelley R.G.
      • Saggar M.
      • L Reiss A.
      • Gotlib I.H.
      Early signs of anomalous neural functional connectivity in healthy offspring of parents with bipolar disorder.
      ). However, the specific role of the left IFG in mediating risk or expression of BD remains unknown.
      We characterized resting-state functional connectivity in BD and those at genetic risk. We focused on a functional cluster within the left IFG for which we recently observed hypoactivation during response inhibition to fearful stimuli in those at genetic risk (
      • Roberts G.
      • Green M.J.
      • Breakspear M.
      • McCormack C.
      • Frankland A.
      • Wright A.
      • et al.
      Reduced inferior frontal gyrus activation during response inhibition to emotional stimuli in youth at high risk of bipolar disorder.
      ). We undertook three complementary analyses of our rs-fMRI data: network-based statistics (NBS) were used to study groupwise differences in the functional connections between the left IFG and all other gray matter regions (
      • Zalesky A.
      • Fornito A.
      • Bullmore E.T.
      Network-based statistic: Identifying differences in brain networks.
      ). We hypothesized that functional connectivity would be selectively diminished between the IFG and regions crucial to emotion and cognitive control in a dose-dependent manner (i.e., the effect will be stronger in those with the established disorder than in those at genetic risk). We also used graph theory to study the complex system-level interactions between the left IFG and the rest of the brain. Whereas NBS captures selective pairwise effects, graph theoretical measures reveal distributed, network-level changes in functional integration and segregation (
      • Rubinov M.
      • Sporns O.
      Complex network measures of brain connectivity: Uses and interpretations.
      ). This approach has revealed subtle disturbances in a number of disorders, including functional connectivity in depression (
      • Lord A.
      • Horn D.
      • Breakspear M.
      • Walter M.
      Changes in community structure of resting state functional connectivity in unipolar depression.
      ) and structural connectivity in BD (
      • Leow A.
      • Ajilore O.
      • Zhan L.
      • Arienzo D.
      • GadElkarim J.
      • Zhang A.
      • et al.
      Impaired inter-hemispheric integration in bipolar disorder revealed with brain network analyses.
      ,
      • Leow A.
      • Zhan L.
      • Ajilore O.
      • GadElkarim J.
      • Zhang A.
      • Arienzo D.
      • et al.
      Measuring inter-hemispheric integration in bipolar affective disorder using brain network analyses and HARDI.
      ). We also moved beyond group differences toward diagnostic classification using machine learning. This approach has shown promise in predicting neuropsychiatric classification from neuroimaging data (
      • Lord A.
      • Horn D.
      • Breakspear M.
      • Walter M.
      Changes in community structure of resting state functional connectivity in unipolar depression.
      ,
      • Koutsouleris N.
      • Meisenzahl E.M.
      • Davatzikos C.
      • Bottlender R.
      • Frodl T.
      • Scheuerecker J.
      • et al.
      Use of neuroanatomical pattern classification to identify subjects in at-risk mental states of psychosis and predict disease transition.
      ). In parallel with our between-group contrasts, we thus used a machine learning classifier to investigate whether the functional connectivity of the left IFG could yield diagnostically informative classification.

      Methods and Materials

      Participants

      Comprehensively phenotyped participants (n = 200) aged 16–30 years comprising three groups were drawn from an ongoing longitudinal study of at-risk individuals: 1) 71 participants genetically at-risk (AR) for BD, 2) 80 matched control (CON) subjects, and 3) 49 BD participants. Details of sample ascertainment, current psychotropic medication, demographic characteristics, and the clinical assessments for younger (16–21 years) and older (22–30 years) age categories are provided in the Supplement.

      Data Acquisition and Preprocessing

      Participants were asked to lie quietly in a 3T Philips Achieva (Amsterdam, The Netherlands) scanner with their eyes closed, while 188 functional images were acquired at a repetition time of 2 seconds. Participants were requested to clear their mind to the best of their ability without falling asleep. Preprocessing of data used included realignment, unwarping, anatomical coregistration, and spatial normalization. The functional data were corrected for white matter and cerebrospinal fluid signal. Global signal regression was not performed unless otherwise stated. Further details of image acquisition and preprocessing are provided in the Supplement.

      IFG Region of Interest

      We constructed a region of interest (ROI) mask for the left IFG by using a contrast associated with inhibiting a motor response to the perception of a fearful face, as reported in our previous analysis (
      • Roberts G.
      • Green M.J.
      • Breakspear M.
      • McCormack C.
      • Frankland A.
      • Wright A.
      • et al.
      Reduced inferior frontal gyrus activation during response inhibition to emotional stimuli in youth at high risk of bipolar disorder.
      ). Although centered on the IFG, this cluster also extends into adjacent regions, principally the insula, orbitofrontal cortex, and putamen (Supplemental Figure S1, Supplemental Table S1). We then parcellated the remaining gray matter voxels into 512 contiguous regions of approximately the same volume as this ROI mask (Supplemental Table S2). Mean time courses were extracted from these ROIs, and a functional connectivity matrix of 513 × 513 pairwise Pearson correlation coefficients was calculated within each subject.

      Network-Based Statistics

      To identify between-group differences in functional connectivity we used NBS, a permutation-based method to control familywise error (FWE). We tested for group differences in the strength of functional connectivity between the left IFG and each of the other 512 gray matter parcels. An omnibus F test was first conducted to test for the influence of group. One-tailed two-sample t tests were then calculated to test pairwise differences between the participant groups. All reported subnetworks survive FWE correction (p < .05) using a conservative search threshold of t = 3.75 (
      • Zalesky A.
      • Fornito A.
      • Bullmore E.T.
      Network-based statistic: Identifying differences in brain networks.
      ).

      Graph Theoretical Analysis: Network Metrics

      We estimated three network properties of the connections from the IFG to the rest of the brain: 1) the path length (PL); 2) the participation index (PI); and 3) the clustering coefficient (CC). These three metrics were chosen because they capture global (PL), intermediate (PI), and local (CC) aspects of network structure (Figure 3, Supplemental Table S3).

      Classification Using Machine Learning

      In parallel with these group-based contrasts, we also sought to study whether the functional connectivity of the IFG could be used to classify participants into their respective groups. We applied support vector classifiers, which are widely used and perform well in many different settings (
      • LaConte S.
      • Strother S.
      • Cherkassky V.
      • Anderson J.
      • Hu X.
      Support vector machines for temporal classification of block design fMRI data.
      ), to the functional connectivity of the left IFG. Reduction of this high-dimensional functional connectivity data was performed by recursively removing the least informative functional edges; surviving edges hence represent the most informative features for disambiguating the groups.

      Results

      IFG Functional Connectivity: NBS

      The omnibus F test for the left IFG functional connectivity revealed a strong group effect (p < .0001 FWE-corrected). A subsequent one-tailed t test revealed a significant pairwise difference between BD and CON participants (p < .001 FWE-corrected). In BD, the left IFG was functionally dysconnected from a constellation of frontotemporal regions (Figure 1A), including proximal ipsilateral cortex (left insula, left putamen, left superior temporal gyrus, left ventrolateral PFC [vlPFC], and mPFC), as well as contralateral regions of a similar anatomical distribution (including right superior temporal gyrus, vlPFC, and mPFC). In total, some 17 nodes comprised this subnetwork, although several of these were confined within the same anatomical region of cortex (Table 2, Figure 1A).
      Figure 1
      Figure 1Functional subnetworks identified by network-based statistics. (A, B) Functional subnetworks identified by inferior frontal gyrus region of interest to whole-brain analysis. Shown are (A) the constellation of connections between the IFG and frontotemporal regions that significantly differed between the control (CON) and bipolar disorder (BD) groups and (B) those connections that significantly differed between the CON and at-risk (AR) groups. The bar graphs shows mean ± SEM of the average of these 17 and 3 functional connections in the CON, AR, and BD groups, respectively. (C, D) Subnetworks identified by whole-brain analysis that include connections with the inferior frontal gyrus. Shown are (C) the subnetwork that significantly differed between the CON and BD groups and (D) the subnetwork that significantly differed between the CON and AR groups. The bar graph shows mean ± SEM of the average of these 10 and 7 functional connections in the CON, AR, and BD groups, respectively. ***p < .001, **p < .01, *p < .05 whole-brain familywise error corrected cluster threshold.
      We also identified a small network that showed weaker functional connectivity in AR than CON participants (p < .006 FWE-corrected; Figure 1B). This left lateralized network included neighboring insular regions and a node in the anterior cingulate cortex. No other pairwise contrast reached significance.
      Regression of the global fMRI signal is often performed as a preprocessing step to mitigate physiological noise. Because the contribution of the global signal to between-group differences in functional connectivity is an unresolved topic (
      • Murphy K.
      • Birn R.M.
      • Handwerker D.A.
      • Jones T.B.
      • Bandettini P.A.
      The impact of global signal regression on resting state correlations: Are anti-correlated networks introduced?.
      ), we performed further analyses to explore potential confounds of the global signal. First, we observed that the average global signal did not significantly differ between our groups (CON vs. BD: p < .129, CON vs. AR: p < .091, AR vs. BD: p < .998). Second, following prior recommendations (
      • Saad Z.S.
      • Gotts S.J.
      • Murphy K.
      • Chen G.
      • Jo H.J.
      • Martin A.
      • et al.
      Trouble at rest: How correlation patterns and group differences become distorted after global signal regression.
      ), we repeated analysis of our principal effect, including the global signal as a covariate; our between-group difference remained significant (Wald χ22 = 29.51, CON > BD: p < .001). Third, we repeated the NBS analysis using an alternative to global signal regression, namely median angle regression (
      • He H.
      • Liu T.T.
      A geometric view of global signal confounds in resting-state functional MRI.
      ). This revealed a slightly weakened, albeit significant, effect in a network (Supplemental Figure S2B) that overlaps substantially with the 17 edges presented in Figure 1A and Table 2. Further details on preprocessing comparisons are provided in Supplemental Results and Supplemental Figure S2C.
      There were no significant group differences in total head motion or maximum or average frame-to-frame head motion (Supplemental Table S4). Nine of our participants showed >3 mm of total head motion; exclusion of these participants did not alter the significance of our findings (Supplemental Figure S3B). Finally, all data were also scrubbed, removing frames with >0.5 mm of frame-to-frame motion (
      • Power J.D.
      • Barnes K.A.
      • Snyder A.Z.
      • Schlaggar B.L.
      • Petersen S.E.
      Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion.
      ). We observed a highly robust subnetwork that almost perfectly overlapped (15 of 17 edges) with the results in Figure 1A (Supplemental Figure S3A).
      Although the focus of the present study was on functional connectivity of the left IFG, we also explored whole-brain NBS, using conservative whole-brain–corrected thresholds. Again, an omnibus F test yielded a highly significant effect of group. Post hoc one-tailed t tests revealed significant effects for the contrasts between CON and BD participants and between CON and AR participants. Intriguingly, these whole-brain contrasts included dysconnection between the left IFG and bilateral homologous areas in reasonably extensive bilateral networks (CON > BD [Figure 1C], CON > AR [Figure 1D, Supplemental Table S5]). There was no significant effect for any other paired one-tailed contrast. There was no significant functional dysconnection of the contralateral (right) IFG at the NBS search threshold used in this study. At more lenient thresholds, a small right subnetwork was evident. Contrasting this to the extent of functional dysconnection of the left IFG at that threshold reinforces the specificity of our principal finding (Supplemental Figure S4).
      Although the focus of the present study was on BD as either a trait or a genetic risk, both the BD and AR groups had higher ratings of depressed mood than the CON group (Table 1). However, current mood state as assessed by different instruments for younger (16–21 years) and older (22–30 years) age categories was not associated with the strength of functional connections presented in Figure 1A–C in either age group (p > .16). Hence, these network disturbances suggest a weakening of the functional connections of the IFG in the BD and AR groups that reflect an underlying trait disturbance. The (whole brain CON > AR) network connectivity in Figure 1D was correlated with mood in the younger (p < .02) but not in the older group. Nonetheless, when mood was included as a covariate, the between-group effect remained significant in both the younger (Wald χ22 = 18.38, p = .0001, CON > AR: p < .001) and older (Wald χ22 = 25.60, p < .01, CON > AR: p < .001) groups. This suggests an influence of mood on this network in the young group that is superimposed on a strong underlying AR trait effect.
      Table 1Demographic and Clinical Data for the AR, CON, and BD Groups
      All participants had an IQ >80. None had a current major depressive or hypo/manic episode.
      VariableCON (n = 80)AR (n = 71)BD (n = 49)Difference StatisticPairwise Comparisons
      Confidence rating ranges using the best estimate methodology vary from 1 to 4, where 1 represents criteria not met for a diagnosis and 4 represents a definite diagnosis. All diagnoses listed here have a confidence rating of 3 or higher.
      Fχ2
      Demographic Data
       Age, years, mean ± SD23.7 ± 3.223.6 ± 4.125.0 ± 3.72.6
       IQ, mean ± SD117.8 ± 11.0117.8 ± 9.2117.8 ± 11.10.00
       Sex, female, n (%)48 (60.0)44 (62.0)33 (67.3)0.713
       Sex, male, n (%)32 (40.0)27 (38.0)16 (32.7)0.713
       Any lifetime diagnosis, n (%)22 (27.5)47 (66.2)49 (100)68.4
      p < .001.
      BD > CON
      p < .001.
      ; BD > AR
      p < .001.
      ; AR > CON
      p < .001.
       Lifetime major depressive episode, n (%)
      A lifetime major depressive episode is defined as meeting DSM-IV criteria for at least one major depressive episode.
      ,
      The average age of onset of recurrent or single episode major depressive disorder in our sample was 21.1 ± 5.5 in the AR group, 23.0 ± 3.4 in the CON group, and 16.0 ± 3.4 in the BD group. For anxiety disorders, the figures were 21.0 ± 6.5, 22.3 ± 5.5, and 17.7 ± 8.6, respectively.
      8 (10.0)23 (32.4)47 (95.9)96.3
      p < .001.
      BD > CON
      p < .001.
      ; BD > AR
      p < .001.
      ; AR > CON
      p < .001.
       Lifetime anxiety disorder, n (%)
      p < .001.
      8 (10.0)19 (26.8)5 (51.0)26.6
      p < .001.
      BD > CON
      p < .001.
      ; BD > AR
      p < .001.
      ; AR > CON
      p < .001.
       Lifetime behavioral disorder, n (%)
      Among the six participants with a behavioral disorder, five had attention-deficit/hyperactivity disorder (one current), one had oppositional defiant disorder (current), and one had conduct disorder (current).
      0 (0)3 (4.2)3 (6.1)4.5
      p < .01.
      BD > CON
      p < .05.
       Lifetime substance disorder, n (%)5 (6.2)10 (14.1)14 (28.6)12.2BD > CON
      p < .01.
      Symptom Severity Scales
       MADRS,
      The age range was 22–30 years for the CON (n = 61), AR (n = 48), and BD (n = 40) groups.
      mean ± SD
      1.9 ± 3.13.0 ± 4.011.4 ± 10.927.2
      p < .001.
      BD > CON
      p < .001.
      ; BD > AR
      p < .001.
       CDI,
      The age range was 16–21 years for the CON (n = 19), AR (n = 23), and BD (n = 9) groups.
      mean ± SD
      6.5 ± 4.611.7 ± 7.921.6 ± 9.911.1
      p < .001.
      BD > CON
      p < .001.
      ; BD > AR
      p < .001.
      AR, at-risk; BD, bipolar disorder; CDI, Children’s Depression Inventory; CON, control; MADRS, Montgomery–Åsberg Depression Rating Scale.
      a All participants had an IQ >80. None had a current major depressive or hypo/manic episode.
      b Confidence rating ranges using the best estimate methodology vary from 1 to 4, where 1 represents criteria not met for a diagnosis and 4 represents a definite diagnosis. All diagnoses listed here have a confidence rating of 3 or higher.
      c p < .001.
      d A lifetime major depressive episode is defined as meeting DSM-IV criteria for at least one major depressive episode.
      e The average age of onset of recurrent or single episode major depressive disorder in our sample was 21.1 ± 5.5 in the AR group, 23.0 ± 3.4 in the CON group, and 16.0 ± 3.4 in the BD group. For anxiety disorders, the figures were 21.0 ± 6.5, 22.3 ± 5.5, and 17.7 ± 8.6, respectively.
      f Among the six participants with a behavioral disorder, five had attention-deficit/hyperactivity disorder (one current), one had oppositional defiant disorder (current), and one had conduct disorder (current).
      g p < .01.
      h p < .05.
      i The age range was 22–30 years for the CON (n = 61), AR (n = 48), and BD (n = 40) groups.
      j The age range was 16–21 years for the CON (n = 19), AR (n = 23), and BD (n = 9) groups.
      We further subdivided our AR group into those with and without a prior episode of major depression. No subgroup differences were found in any of the four networks identified in Figure 1A–D (p > .51). There were no subgroup differences based on the current use of antidepressants (all p > .13). Within the BD group, current use of lithium, mood stabilizers, antipsychotics, antidepressants, or a combination were not associated with the strength of any of these four identified networks (p > .09). In addition, none of networks presented in Figure 1 A–D revealed a main effect of age or an age × group interaction (p > .10).
      All four of the reported subnetworks reported in Figure 1 remained highly significant (p < .0001) after excluding all subjects with any lifetime diagnosis of anxiety or depression from the CON and AR groups. In addition, there were no significant subgroup differences within the CON group in the strength of any of the dysconnected networks in those with versus without a lifetime diagnosis of a major depressive or anxiety episode (p > .47). The between-group contrasts also remain highly significant when including lifetime diagnosis of a major depressive or anxiety episode as an additional regressor in our generalized estimating equation model (p < .0007).
      Is the subnetwork shown in Figure 1A a broad peak above a general change in functional connectivity in BD, or does it reflect a specific functional disconnection of the IFG? To investigate this, we plotted the between-group t statistic of this network of 17 edges (Figure 2, red arrow, CON-BD) against 10,000 subnetworks composed of 17 edges chosen at random (Figure 2, blue histogram). There is indeed a subtle albeit nonsignificant (p > .10) decrease in functional connectivity in BD. However, the subnetwork discovered through the NBS analysis of the left IFG is categorically stronger than this typical background effect.
      Figure 2
      Figure 2Specificity of reduced functional connectivity of the 17-node subnetwork in bipolar disorder. Red arrow shows the t statistic for the network shown in . Blue bars show the results of a Monte Carlo analysis, where a comparable 17-node subnetwork is chosen at random. Although there is a bias of functional connectivity to be weaker in the bipolar disorder (BD) group (t statistics generally >1), none of these surpass a corrected p value. In stark contrast, the original (reported) subnetwork is strongly disconnected (t = 5.4). CON, control.

      IFG Functional Connectivity: Topological Graph Metrics

      We analyzed the network characteristics of the IFG functional connectivity, focusing on measures of local (CC), intermediate (PI) and global (PL) network topology (Figure 3). CC differed significantly across the three groups (Wald χ22 = 10.76, p < .005). Post hoc comparison showed that this effect was primarily driven by lower clustering in the BD group than the CON group (CON > BD: p < .005). The current use of lithium, mood stabilizers, antipsychotics, and antidepressants were not associated with CC in the BD group (p > .54). The effect remained significant in the younger but not the older group (Wald χ22 = 4.01, p = .52) when mood was included as a covariate, and in this younger group it was also accompanied by differences between AR and BD participants (Wald χ22 = 10.48, p < .005, CON > BD: p < .005, CON > AR: p < .02) (Supplemental Figure S4). No differences in CC in the younger group were associated with current use of antidepressants, the previous diagnosis of at least one depressive episode, or anxiety (all p > .20).
      Figure 3
      Figure 3Graph of theoretic analysis of the functional connectivity of the left inferior frontal gyrus. (Top row) Schematic illustration of the three topological measures studied: path length, participation index, and clustering coefficient. These capture network organization from large integrative scales (path length) to intermediate (participation index) and local (clustering coefficient) scales related to segregation (clustering coefficient). (Bottom row) Groupwise estimates of these measures in the control (CON), at-risk (AR), and bipolar disorder (BD) groups. Bar graphs represent mean ± SEM. [Adapted from Rubinov and Sporns (
      • Rubinov M.
      • Sporns O.
      Complex network measures of brain connectivity: Uses and interpretations.
      ), with permission from Elsevier.]
      Analysis of the PI revealed a weak effect of group accompanied by nonsignificant but trend-level post hoc two-group tests (Wald χ22 = 7.24, p < .03, CON < AR: p = .07, AR > BD: p = .06, CON > BD: p = .93). PL did not significantly differ between groups (Wald χ22 = 3.20, p = .20).

      IFG Functional Connectivity: Machine Learning Classification

      Application of machine learning to our data yielded classification rates that were well above chance (Supplemental Table S6, Figure 4). The mean (balanced) accuracy for the three-group classifier was 64.3% (due to the unequal group size, the chance rate was 40.8%). In general, the negative predictive values (72%–79%) were higher than the positive predictive values (49%–54%). Hence, the classifiers were better at excluding incorrect group assignment than identifying correct assignments, although both rates were above the chance rate. BD subjects were more likely to be wrongly assigned to the AR group than the CON group. Likewise, the CON subjects were more likely to be erroneously categorized as AR than BD. This accords with the greater phenotypic and genotypic distinction between CON and BD than the intermediate AR group. False classification of AR subjects was reasonably evenly split between both groups, although misclassification to the CON group was slightly higher.
      Figure 4
      Figure 4The constellation of frontotemporal brain regions identified by the three-group machine learning algorithm trained to classify subjects into the control (CON), at-risk (AR), and bipolar disorder (BD) groups using left inferior frontal gyrus functional connections. The bar graph shows the positive predictive value (PPV) and negative predictive value (NPV) for the support vector classifier (). The dashed line denotes the background (chance) accuracy of 40.8%. Note that accuracy for AR subjects is lower than CON and BD, particularly for the PPV.
      The edges of the IFG functional connectivity matrix that yielded classification in our data (Figure 4) showed a strong overlap with those observed in the traditional between-group (NBS) analysis (Table 2). More specifically, 11 nodes revealed by the NBS contrast between BD and CON participants, 2 nodes from the AR and CON contrast, and one node identified with both contrasts directly overlapped with the 13 classifier nodes.
      Table 2Functional Connections of the Left IFG as Revealed by NBS and Machine Learning Algorithm
      RegionPresent inEdge WeightsMNI CoordinatesBrodmann Areas
      NBSMachine LearningCON, Mean ± SDAR, Mean ± SDBD, Mean ± SD
      Left IFG/Pars TriangularisCON > BD00.45 ± 0.140.39 ± 0.180.34 ± 0.14−4032646, 47
      Left InsulaCON > BD00.58 ± 0.140.52 ± 0.150.46 ± 0.18−4312−213, 22, 38, 47
      Left Middle Orbitofrontal GyrusCON > BDX0.53 ± 0.150.45 ± 0.170.40 ± 0.16−2235−1711
      Left Orbital IFGCON > BDX0.58 ± 0.150.55 ± 0.150.46 ± 0.18−4723−711, 22, 38, 47
      Left Orbital IFGCON > BDX0.54 ± 0.140.46 ± 0.150.39 ± 0.18−2634−83, 47
      Left PutamenCON > BDX0.67 ± 0.120.60 ± 0.130.58 ± 0.17−2518−713, 47
      Left Rolandic OperculumCON > BDX0.53 ± 0.160.43 ± 0.190.41 ± 0.19−45156, 13, 22, 38, 43
      Left Superior Frontal GyrusCON > BDX0.31 ± 0.170.24 ± 0.170.18 ± 0.17−19603210, 9
      Left Rolandic OperculumCON > BD00.54 ± 0.150.48 ± 0.150.43 ± 0.17−581621, 22, 6, 44
      Left Superior Temporal GyrusCON > BDX0.52 ± 0.150.43 ± 0.180.39 ± 0.17−60−0.7−219, 5
      Left Superior Temporal GyrusCON > BDX0.45 ± 0.180.40 ± 0.180.32 ± 0.20−62−26742, 22, 21
      Right Orbital IFGCON > BD00.50 ± 0.160.46 ± 0.130.39 ± 0.183234−811, 47
      Right Orbital Inferior Frontal GyrusCON > BD00.36 ± 0.170.31 ± 0.160.24 ± 0.175250−410, 46, 47
      Right Superior Frontal GyrusCON > BDX0.41 ± 0.180.35 ± 0.200.27 ± 0.1816−12766
      Right Superior Temporal GyrusCON > BDX0.53 ± 0.160.45 ± 0.160.39 ± 0.174912−213, 22, 38, 47
      Right Superior Temporal GyrusCON > BD00.53 ± 0.170.47 ± 0.170.41 ± 0.1966−0.7−221, 22
      Left InsulaCON > BD CON > ARX0.59 ± 0.130.47 ± 0.200.46 ± 0.19−35−1313
      Left Medial Cingulate GyrusCON > ARX0.42 ± 0.170.31 ± 0.210.32 ± 0.15−7−184324, 6, 31, 32
      Lentiform NucleusCON > ARX0.62 ± 0.130.53 ± 0.170.54 ± 0.12−25−12−513, 35
      AR, at-risk; BD, bipolar disorder; CON, control; IFG, inferior frontal gyrus; NBS, network-based statistics; X, present; 0, absent.

      Discussion

      We report a functional dysconnection of the IFG from frontolimbic regions that appears to underlie trait-related risk of affect dysregulation in BD and, to a lesser extent, participants at genetic risk of BD. An overlapping constellation of functionally dysconnected regions carries potential to classify individuals into the BD, AR, and CON groups. We also report a loss of local network clustering in the resting-state functional connectivity of the IFG in BD. Our findings extend prior research that suggests a potential role of the IFG in the genetic risk and phenotypic expression of BD by highlighting the specific pattern of resting-state functional connections from the IFG in those with BD and those at high risk.
      Several of these functionally dysconnected regions warrant further consideration. Prior functional neuroimaging studies in BD have found evidence of superior temporal gyrus dysfunction during emotional processing (
      • Mitchell R.L.
      • Elliott R.
      • Barry M.
      • Cruttenden A.
      • Woodruff P.W.
      Neural response to emotional prosody in schizophrenia and in bipolar affective disorder.
      ,
      • Pavuluri M.N.
      • O’Connor M.M.
      • Harral E.
      • Sweeney J.A.
      Affective neural circuitry during facial emotion processing in pediatric bipolar disorder.
      ,
      • Malhi G.S.
      • Lagopoulos J.
      • Sachdev P.S.
      • Ivanovski B.
      • Shnier R.
      • Ketter T.
      Is a lack of disgust something to fear? A functional magnetic resonance imaging facial emotion recognition study in euthymic bipolar disorder patients.
      ,
      • Chen C.-H.
      • Lennox B.
      • Jacob R.
      • Calder A.
      • Lupson V.
      • Bisbrown-Chippendale R.
      • et al.
      Explicit and implicit facial affect recognition in manic and depressed states of bipolar disorder: A functional magnetic resonance imaging study.
      ) as well as at rest (
      • Dickstein D.P.
      • Gorrostieta C.
      • Ombao H.
      • Goldberg L.D.
      • Brazel A.C.
      • Gable C.J.
      • et al.
      Fronto-temporal spontaneous resting state functional connectivity in pediatric bipolar disorder.
      ,
      • Xiao Q.
      • Zhong Y.
      • Lu D.
      • Gao W.
      • Jiao Q.
      • Lu G.
      • et al.
      Altered regional homogeneity in pediatric bipolar disorder during manic state: a resting-state fMRI study.
      ). The vlPFC has been previously implicated in the structural pathology of BD patients and unaffected relatives [although such findings have been inconsistent (
      • Schneider M.R.
      • DelBello M.P.
      • McNamara R.K.
      • Strakowski S.M.
      • Adler C.M.
      Neuroprogression in bipolar disorder.
      )]. Reduced functional connectivity between the IFG and the default network may reflect some of the distinct cognitive attributes and symptomatic descriptions of BD (
      • Ricciardiello L.
      • Fornaro P.
      Beyond the cliff of creativity: A novel key to bipolar disorder and creativity.
      ) such as intrusive emotional ruminations that worsen in the absence of environmental distraction. Ventral and orbital components of mPFC, Brodmann areas 24 and 32 (contributors to our CON > AR functional subnetwork), have been suggested to contribute to emotional dysregulation in BD (
      • Phillips M.
      • Ladouceur C.
      • Drevets W.
      A neural model of voluntary and automatic emotion regulation: Implications for understanding the pathophysiology and neurodevelopment of bipolar disorder.
      ). Reduced functional integration between mPFC and the rest of the PFC has been reported in BD subjects compared with control subjects during rs-fMRI (
      • Anticevic A.
      • Brumbaugh M.S.
      • Winkler A.M.
      • Lombardo L.E.
      • Barrett J.
      • Corlett P.R.
      • et al.
      Global prefrontal and fronto-amygdala dysconnectivity in bipolar I disorder with psychosis history.
      ). In addition, a recent meta-analysis of diffusion tensor imaging studies identified decreased fractional anisotropy in the cingulate cortex (Brodmann area 32) in BD (
      • Vederine F.-E.
      • Wessa M.
      • Leboyer M.
      • Houenou J.
      A meta-analysis of whole-brain diffusion tensor imaging studies in bipolar disorder.
      ). Although these findings converge with those presently reported, we additionally implicate functional connectivity of these regions to the IFG. Although our seed region was derived from group differences during task execution, caution nonetheless is required when interpreting the complex relationships between structure, function, and connectivity.
      For clinical translation, imaging research must progress beyond group differences toward classification of individuals. A significant effect on a contrast cannot, on its own, inform the potential accuracy of a classifier. Machine learning, performed in parallel with the traditional between-group contrasts, yielded a constellation of functionally dysconnected regions that closely resemble the network identified by our classic (frequentist) analysis, and which had significantly greater than chance classification accuracy. Machine learning has been previously applied in task-related fMRI to disambiguate BD from schizophrenia during a verbal fluency task (
      • Costafreda S.G.
      • Fu C.H.
      • Picchioni M.
      • Toulopoulou T.
      • McDonald C.
      • Kravariti E.
      • et al.
      Pattern of neural responses to verbal fluency shows diagnostic specificity for schizophrenia and bipolar disorder.
      ) and to classify AR from CON participants during facial emotion processing (
      • Mourão-Miranda J.
      • Oliveira L.
      • Ladouceur C.D.
      • Marquand A.
      • Brammer M.
      • Birmaher B.
      • et al.
      Pattern recognition and functional neuroimaging help to discriminate healthy adolescents at risk for mood disorders from low risk adolescents.
      ). To our knowledge, our study is the first to disambiguate BD, AR, and CON, a more challenging (3-group) classification task. The notion that high genetic risk is associated with an intermediate endophenotype is reflected in the higher misclassification of the AR participants than the CON or BD groups. Of interest, the negative predictive value (the rate of accurately excluding an effect) was higher than the positive predictive value. Although both are below the accuracy that would support direct clinical translation, this classification rests solely on one feature of our data, namely the functional connectivity of the left IFG. For clinical translation, a classifier could draw on other data, including other features of imaging data in addition to phenotypic information (
      • Huys Q.J.
      • Maia T.V.
      • Frank M.J.
      Computational psychiatry as a bridge from neuroscience to clinical applications.
      ). The present finding points to the potential utility of resting-state functional connectivity as an imaging-based screening test to complement such other information.
      We explored three graph metrics that focus on local (CC), intermediate (PI), and global (PL) aspects of network topology. It has been suggested that a high CC promotes segregation of cortical regions into local functional circuits (
      • Bassett D.S.
      • Bullmore E.
      Small-world brain networks.
      ,
      • Liu Y.
      • Liang M.
      • Zhou Y.
      • He Y.
      • Hao Y.
      • Song M.
      • et al.
      Disrupted small-world networks in schizophrenia.
      ). We found lower CC in the BD than CON subjects, suggesting a subtle randomization of the functional circuits associated with the IFG. This is consistent with the direction of our whole-brain findings, which also suggest compromised integration of local IFG activity in large-scale networks. In the younger group of BD subjects, mood state was associated with CC. AR participants also had lower CC than BD participants in this younger group, suggesting that transient mood-related disruptions, which are characteristic of mood dysregulation in BD, may superimpose on possible trait-related disruptions.
      Although previous research has examined rs-fMRI in BD relatives (
      • Khadka S.
      • Meda S.A.
      • Stevens M.C.
      • Glahn D.C.
      • Calhoun V.D.
      • Sweeney J.A.
      • et al.
      Is aberrant functional connectivity a psychosis endophenotype? A resting state functional magnetic resonance imaging study.
      ,
      • Meda S.A.
      • Gill A.
      • Stevens M.C.
      • Lorenzoni R.P.
      • Glahn D.C.
      • Calhoun V.D.
      • et al.
      Differences in resting-state functional magnetic resonance imaging functional network connectivity between schizophrenia and psychotic bipolar probands and their unaffected first-degree relatives.
      ,
      • Lui S.
      • Yao L.
      • Xiao Y.
      • Keedy S.
      • Reilly J.
      • Keefe R.
      • et al.
      Resting-state brain function in schizophrenia and psychotic bipolar probands and their first-degree relatives.
      ,
      • Singh M.K.
      • Chang K.D.
      • Kelley R.G.
      • Saggar M.
      • L Reiss A.
      • Gotlib I.H.
      Early signs of anomalous neural functional connectivity in healthy offspring of parents with bipolar disorder.
      ), to the best of our knowledge, ours is the first to focus specifically on the left IFG. In particular, we specifically seeded our analysis with a region that we previously observed to show a task-related deficit in this same cohort. Our work hence bridges resting-state and task-related domains. In addition, the study samples of these prior reports were either younger (
      • Singh M.K.
      • Chang K.D.
      • Kelley R.G.
      • Saggar M.
      • L Reiss A.
      • Gotlib I.H.
      Early signs of anomalous neural functional connectivity in healthy offspring of parents with bipolar disorder.
      ) or older (
      • Khadka S.
      • Meda S.A.
      • Stevens M.C.
      • Glahn D.C.
      • Calhoun V.D.
      • Sweeney J.A.
      • et al.
      Is aberrant functional connectivity a psychosis endophenotype? A resting state functional magnetic resonance imaging study.
      ,
      • Meda S.A.
      • Gill A.
      • Stevens M.C.
      • Lorenzoni R.P.
      • Glahn D.C.
      • Calhoun V.D.
      • et al.
      Differences in resting-state functional magnetic resonance imaging functional network connectivity between schizophrenia and psychotic bipolar probands and their unaffected first-degree relatives.
      ,
      • Lui S.
      • Yao L.
      • Xiao Y.
      • Keedy S.
      • Reilly J.
      • Keefe R.
      • et al.
      Resting-state brain function in schizophrenia and psychotic bipolar probands and their first-degree relatives.
      ) than ours. Our cohort is of an age whereby transition to BD is approaching its peak incidence (
      • Goodwin F.K.
      • Jamison K.R.
      Manic-Depressive Illness: Bipolar Disorders and Recurrent Depression.
      ,
      • Loranger A.W.
      • Levine P.M.
      Age at onset of bipolar affective illness.
      ). These other studies have either explored connectivity among a small number of a priori chosen regions, or used independent components analysis to derive large spatial maps to study correlations between these functional networks, prohibiting a direct comparison with our results. Although the high heritability of graph metrics (
      • Smit D.J.
      • Stam C.J.
      • Posthuma D.
      • Boomsma D.I.
      • de Geus E.J.
      Heritability of “small‐world” networks in the brain: A graph theoretical analysis of resting‐state EEG functional connectivity.
      ) motivates their use as a marker to predict genetic risk of neuropsychiatric conditions (
      • Bullmore E.
      • Sporns O.
      Complex brain networks: Graph theoretical analysis of structural and functional systems.
      ), considerably more work is required before a consensus integration of rs-fMRI findings can be achieved.
      The BD group had higher state depression than the AR group whom in turn had higher depression severity than the CON group. We thus performed several analyses which suggest that our findings do not reflect current mood, past episodes of depression, or medication effects. We also observed several effects for which CON > BD > AR, hence defying the gradient of low mood. Previous seed-based connectivity analyses have revealed that adolescents with major depressive disorder have decreased functional connectivity in a subgenual anterior cingulate cortex (sgACC)-based functional network that includes the IFG compared with control subjects (
      • Connolly C.G.
      • Wu J.
      • Ho T.C.
      • Hoeft F.
      • Wolkowitz O.
      • Eisendrath S.
      • et al.
      Resting-state functional connectivity of subgenual anterior cingulate cortex in depressed adolescents.
      ,
      • Cullen K.R.
      • Gee D.G.
      • Klimes-Dougan B.
      • Gabbay V.
      • Hulvershorn L.
      • Mueller B.A.
      • et al.
      A preliminary study of functional connectivity in comorbid adolescent depression.
      ). However, the sgACC did not appear within our group at different subnetworks (even at very liberal thresholds). Moreover, there were no group effects in our data when seeding the sgACC. Taken together, these observations suggest that the integrity of the crucial IFG-sgACC functional connection may be diminished in unipolar depression but relatively preserved in BD.
      Several methodological issues require consideration. In our cross-sectional study, it is unclear whether changes in connectivity in the BD group reflect the cause or the consequence of the illness or its treatment. The presence of intermediate effects in the AR group, who are not currently prescribed mood stabilizers, mitigates this concern and suggests that our results are at least partly attributable to prior trait-related dysfunction. Our AR group is inherently heterogeneous; some will ultimately develop BD and others will not. The intermediate effects in the AR group may be a mixture of stronger effects in those whom will develop BD and weaker effects in those whom are unlikely to convert. All the present participants are in an ongoing longitudinal study; follow-up data will further elucidate the underlying mechanisms that contribute to conversion from AR to BD.
      The analysis of rs-fMRI data holds both promise as well as disadvantages. Task instructions are minimal; hence, participation is straightforward. However, the lack of an explicit task and behavioural outcome challenges inference on emotional and cognitive correlates. On this latter point, it is particularly interesting to recall that the functional mask for our core IFG region was previously reported by our own group to show a lack of engagement during motor inhibition to fearful faces in AR participants compared with control subjects (
      • Roberts G.
      • Green M.J.
      • Breakspear M.
      • McCormack C.
      • Frankland A.
      • Wright A.
      • et al.
      Reduced inferior frontal gyrus activation during response inhibition to emotional stimuli in youth at high risk of bipolar disorder.
      ) arising through hierarchical interactions between cognitive control and emotional networks (
      • Breakspear M.
      • Roberts G.
      • Green M.J.
      • Nguyen V.T.
      • Frankland A.
      • Levy F.
      • et al.
      Network dysfunction of emotional and cognitive processes in those at genetic risk of bipolar disorder.
      ). The present study suggests that in both BD participants and those at genetic risk of BD, a disturbance of the functional integration of this region persists in the absence of a task. Indeed, the lack of a task or other dynamic distractor may increase the dysphoria associated with affective disorders, allowing attention to drift toward unpleasant memories, intrusive ruminations, and somatic symptoms.
      In summary, resting-state functional connectivity of the IFG may offer diagnostic utility as well as point toward novel pathophysiological correlates of BD and may identify more subtle effects seen in those at high risk of the disorder. Some of our AR participants will likely experience future manic episodes and hence transition to BD. The present findings provide a priori predictions to guide future hypotheses, eschewing reliance on exploratory analyses.

      Acknowledgments and Disclosures

      This study was funded by the Australian National Medical and Health Research Council Program Grant No. 1037196 (PM, MB), the Lansdowne Foundation (PM), and the Queensland Government Office of Health and Medical Research (MB).
      The authors report no biomedical financial interests or potential conflicts of interest.

      Appendix A. Supplementary material

      References

        • Whiteford H.A.
        • Degenhardt L.
        • Rehm J.
        • Baxter A.J.
        • Ferrari A.J.
        • Erskine H.E.
        • et al.
        Global burden of disease attributable to mental and substance use disorders: Findings from the Global Burden of Disease Study 2010.
        Lancet. 2013; 382: 1575-1586
        • Mitchell P.
        • Johnston A.
        • Frankland A.
        • Slade T.
        • Green M.
        • Roberts G.
        • et al.
        Bipolar disorder in a national survey using the World Mental Health Version of the Composite International Diagnostic Interview: The impact of differing diagnostic algorithms.
        Acta Psychiatr Scand. 2013; 127: 381-393
        • Phillips M.L.
        • Swartz H.A.
        A critical appraisal of neuroimaging studies of bipolar disorder: Toward a new conceptualization of underlying neural circuitry and a road map for future research.
        Am J Psychiatry. 2014; 171: 829-843
        • Phillips M.
        • Ladouceur C.
        • Drevets W.
        A neural model of voluntary and automatic emotion regulation: Implications for understanding the pathophysiology and neurodevelopment of bipolar disorder.
        Mol Psychiatry. 2008; 13: 833-857
        • Seeley W.W.
        • Menon V.
        • Schatzberg A.F.
        • Keller J.
        • Glover G.H.
        • Kenna H.
        • et al.
        Dissociable intrinsic connectivity networks for salience processing and executive control.
        J Neurosci. 2007; 27: 2349-2356
        • Chen C.H.
        • Suckling J.
        • Lennox B.R.
        • Ooi C.
        • Bullmore E.T.
        A quantitative meta-analysis of fMRI studies in bipolar disorder.
        Bipolar Disord. 2011; 13: 1-15
        • Liakakis G.
        • Nickel J.
        • Seitz R.
        Diversity of the inferior frontal gyrus—A meta-analysis of neuroimaging studies.
        Behav Brain Res. 2011; 225: 341-347
        • Brotman M.A.
        • Deveney C.M.
        • Thomas L.A.
        • Hinton K.E.
        • Yi J.Y.
        • Pine D.S.
        • et al.
        Parametric modulation of neural activity during face emotion processing in unaffected youth at familial risk for bipolar disorder.
        Bipolar Disord. 2014; 16: 756-763
        • Hajek T.
        • Cullis J.
        • Novak T.
        • Kopecek M.
        • Blagdon R.
        • Propper L.
        • et al.
        Brain structural signature of familial predisposition for bipolar disorder: Replicable evidence for involvement of the right inferior frontal gyrus.
        Biol Psychiatry. 2012; 73: 144-152
        • Roberts G.
        • Green M.J.
        • Breakspear M.
        • McCormack C.
        • Frankland A.
        • Wright A.
        • et al.
        Reduced inferior frontal gyrus activation during response inhibition to emotional stimuli in youth at high risk of bipolar disorder.
        Biol Psychiatry. 2013; 74: 55-61
        • Valenzuela M.J.
        • Breakspear M.
        • Sachdev P.
        Complex mental activity and the aging brain: Molecular, cellular and cortical network mechanisms.
        Brain Res Rev. 2007; 56: 198-213
        • Seeley W.W.
        • Crawford R.K.
        • Zhou J.
        • Miller B.L.
        • Greicius M.D.
        Neurodegenerative diseases target large-scale human brain networks.
        Neuron. 2009; 62: 42-52
        • Fornito A.
        • Zalesky A.
        • Breakspear M.
        The connectomics of brain disorders.
        Nat Rev Neurosci. 2015; 16: 159-172
        • Avery J.A.
        • Drevets W.C.
        • Moseman S.E.
        • Bodurka J.
        • Barcalow J.C.
        • Simmons W.K.
        Major depressive disorder is associated with abnormal interoceptive activity and functional connectivity in the insula.
        Biol Psychiatry. 2013; 76: 258-266
        • Biswal B.B.
        Resting state fMRI: A personal history.
        Neuroimage. 2012; 62: 938-944
        • Torrisi S.
        • Moody T.D.
        • Vizueta N.
        • Thomason M.E.
        • Monti M.M.
        • Townsend J.D.
        • et al.
        Differences in resting corticolimbic functional connectivity in bipolar I euthymia.
        Bipolar Disord. 2013; 15: 156-166
        • Mamah D.
        • Barch D.M.
        • Repovš G.
        Resting state functional connectivity of five neural networks in bipolar disorder and schizophrenia.
        J Aff Disord. 2013; 150: 601-609
        • Liu H.
        • Tang Y.
        • Womer F.
        • Fan G.
        • Lu T.
        • Driesen N.
        • et al.
        Differentiating patterns of amygdala-frontal functional connectivity in schizophrenia and bipolar disorder.
        Schizophr Bull. 2013; 40: 469-477
        • Dickstein D.P.
        • Gorrostieta C.
        • Ombao H.
        • Goldberg L.D.
        • Brazel A.C.
        • Gable C.J.
        • et al.
        Fronto-temporal spontaneous resting state functional connectivity in pediatric bipolar disorder.
        Biol Psychiatry. 2010; 68: 839-846
        • Chepenik L.G.
        • Raffo M.
        • Hampson M.
        • Lacadie C.
        • Wang F.
        • Jones M.M.
        • et al.
        Functional connectivity between ventral prefrontal cortex and amygdala at low frequency in the resting state in bipolar disorder.
        Psychiatry Res. 2010; 182: 207-210
        • Anticevic A.
        • Brumbaugh M.S.
        • Winkler A.M.
        • Lombardo L.E.
        • Barrett J.
        • Corlett P.R.
        • et al.
        Global prefrontal and fronto-amygdala dysconnectivity in bipolar I disorder with psychosis history.
        Biol Psychiatry. 2012; 73: 565-573
        • Anand A.
        • Li Y.
        • Wang Y.
        • Lowe M.J.
        • Dzemidzic M.
        Resting state corticolimbic connectivity abnormalities in unmedicated bipolar disorder and unipolar depression.
        Psychiatry Res. 2009; 171: 189-198
        • Chai X.J.
        • Whitfield-Gabrieli S.
        • Shinn A.K.
        • Gabrieli J.D.
        • Nieto Castañón A.
        • McCarthy J.M.
        • et al.
        Abnormal medial prefrontal cortex resting-state connectivity in bipolar disorder and schizophrenia.
        Neuropsychopharmacology. 2011; 36: 2009-2017
        • Öngür D.
        • Lundy M.
        • Greenhouse I.
        • Shinn A.K.
        • Menon V.
        • Cohen B.M.
        • et al.
        Default mode network abnormalities in bipolar disorder and schizophrenia.
        Psychiatry Res. 2010; 183: 59-68
        • Calhoun V.D.
        • Maciejewski P.K.
        • Pearlson G.D.
        • Kiehl K.A.
        Temporal lobe and “default” hemodynamic brain modes discriminate between schizophrenia and bipolar disorder.
        Hum Brain Mapp. 2008; 29: 1265-1275
        • Calhoun V.D.
        • Sui J.
        • Kiehl K.
        • Turner J.A.
        • Allen E.A.
        • Pearlson G.
        Exploring the psychosis functional connectome: Aberrant intrinsic networks in schizophrenia and bipolar disorder.
        Front Psychiatry. 2012; 2: 75
        • Ford K.A.
        • Théberge J.
        • Neufeld R.J.
        • Williamson P.C.
        • Osuch E.A.
        Correlation of brain default mode network activation with bipolarity index in youth with mood disorders.
        J Aff Disord. 2013; 150: 1174-1178
        • Raichle M.E.
        • MacLeod A.M.
        • Snyder A.Z.
        • Powers W.J.
        • Gusnard D.A.
        • Shulman G.L.
        A default mode of brain function.
        Proc Natl Acad Sci U S A. 2001; 98: 676-682
        • Khadka S.
        • Meda S.A.
        • Stevens M.C.
        • Glahn D.C.
        • Calhoun V.D.
        • Sweeney J.A.
        • et al.
        Is aberrant functional connectivity a psychosis endophenotype? A resting state functional magnetic resonance imaging study.
        Biol Psychiatry. 2013; 74: 458-466
        • Meda S.A.
        • Gill A.
        • Stevens M.C.
        • Lorenzoni R.P.
        • Glahn D.C.
        • Calhoun V.D.
        • et al.
        Differences in resting-state functional magnetic resonance imaging functional network connectivity between schizophrenia and psychotic bipolar probands and their unaffected first-degree relatives.
        Biol Psychiatry. 2012; 71: 881-889
        • Lui S.
        • Yao L.
        • Xiao Y.
        • Keedy S.
        • Reilly J.
        • Keefe R.
        • et al.
        Resting-state brain function in schizophrenia and psychotic bipolar probands and their first-degree relatives.
        Psychol Med. 2015; 45: 97-108
        • Singh M.K.
        • Chang K.D.
        • Kelley R.G.
        • Saggar M.
        • L Reiss A.
        • Gotlib I.H.
        Early signs of anomalous neural functional connectivity in healthy offspring of parents with bipolar disorder.
        Bipolar Disord. 2014; 16: 678-689
        • Zalesky A.
        • Fornito A.
        • Bullmore E.T.
        Network-based statistic: Identifying differences in brain networks.
        Neuroimage. 2010; 53: 1197-1207
        • Rubinov M.
        • Sporns O.
        Complex network measures of brain connectivity: Uses and interpretations.
        Neuroimage. 2010; 52: 1059-1069
        • Lord A.
        • Horn D.
        • Breakspear M.
        • Walter M.
        Changes in community structure of resting state functional connectivity in unipolar depression.
        PloS One. 2012; 7: e41282
        • Leow A.
        • Ajilore O.
        • Zhan L.
        • Arienzo D.
        • GadElkarim J.
        • Zhang A.
        • et al.
        Impaired inter-hemispheric integration in bipolar disorder revealed with brain network analyses.
        Biol Psychiatry. 2013; 73: 183-193
        • Leow A.
        • Zhan L.
        • Ajilore O.
        • GadElkarim J.
        • Zhang A.
        • Arienzo D.
        • et al.
        Measuring inter-hemispheric integration in bipolar affective disorder using brain network analyses and HARDI.
        Proc IEEE Int Symp Biomed Imaging. 2012; : 5-8
        • Koutsouleris N.
        • Meisenzahl E.M.
        • Davatzikos C.
        • Bottlender R.
        • Frodl T.
        • Scheuerecker J.
        • et al.
        Use of neuroanatomical pattern classification to identify subjects in at-risk mental states of psychosis and predict disease transition.
        Arch Gen Psychiatry. 2009; 66: 700-712
        • LaConte S.
        • Strother S.
        • Cherkassky V.
        • Anderson J.
        • Hu X.
        Support vector machines for temporal classification of block design fMRI data.
        Neuroimage. 2005; 26: 317-329
        • Murphy K.
        • Birn R.M.
        • Handwerker D.A.
        • Jones T.B.
        • Bandettini P.A.
        The impact of global signal regression on resting state correlations: Are anti-correlated networks introduced?.
        Neuroimage. 2009; 44: 893-905
        • Saad Z.S.
        • Gotts S.J.
        • Murphy K.
        • Chen G.
        • Jo H.J.
        • Martin A.
        • et al.
        Trouble at rest: How correlation patterns and group differences become distorted after global signal regression.
        Brain Connectivity. 2012; 2: 25-32
        • He H.
        • Liu T.T.
        A geometric view of global signal confounds in resting-state functional MRI.
        Neuroimage. 2012; 59: 2339-2348
        • Power J.D.
        • Barnes K.A.
        • Snyder A.Z.
        • Schlaggar B.L.
        • Petersen S.E.
        Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion.
        NeuroImage. 2012; 59: 2142-2154
        • Mitchell R.L.
        • Elliott R.
        • Barry M.
        • Cruttenden A.
        • Woodruff P.W.
        Neural response to emotional prosody in schizophrenia and in bipolar affective disorder.
        Br J Psychiatry. 2004; 184: 223-230
        • Pavuluri M.N.
        • O’Connor M.M.
        • Harral E.
        • Sweeney J.A.
        Affective neural circuitry during facial emotion processing in pediatric bipolar disorder.
        Biol Psychiatry. 2007; 62: 158-167
        • Malhi G.S.
        • Lagopoulos J.
        • Sachdev P.S.
        • Ivanovski B.
        • Shnier R.
        • Ketter T.
        Is a lack of disgust something to fear? A functional magnetic resonance imaging facial emotion recognition study in euthymic bipolar disorder patients.
        Bipolar Disord. 2007; 9: 345-357
        • Chen C.-H.
        • Lennox B.
        • Jacob R.
        • Calder A.
        • Lupson V.
        • Bisbrown-Chippendale R.
        • et al.
        Explicit and implicit facial affect recognition in manic and depressed states of bipolar disorder: A functional magnetic resonance imaging study.
        Biol Psychiatry. 2006; 59: 31-39
        • Xiao Q.
        • Zhong Y.
        • Lu D.
        • Gao W.
        • Jiao Q.
        • Lu G.
        • et al.
        Altered regional homogeneity in pediatric bipolar disorder during manic state: a resting-state fMRI study.
        PloS One. 2013; 8: e57978
        • Schneider M.R.
        • DelBello M.P.
        • McNamara R.K.
        • Strakowski S.M.
        • Adler C.M.
        Neuroprogression in bipolar disorder.
        Bipolar Disord. 2012; 14: 356-374
        • Ricciardiello L.
        • Fornaro P.
        Beyond the cliff of creativity: A novel key to bipolar disorder and creativity.
        Med Hypotheses. 2013; 80: 534-543
        • Vederine F.-E.
        • Wessa M.
        • Leboyer M.
        • Houenou J.
        A meta-analysis of whole-brain diffusion tensor imaging studies in bipolar disorder.
        Prog Neuropsychopharmacol Biol Psychiatry. 2011; 35: 1820-1826
        • Costafreda S.G.
        • Fu C.H.
        • Picchioni M.
        • Toulopoulou T.
        • McDonald C.
        • Kravariti E.
        • et al.
        Pattern of neural responses to verbal fluency shows diagnostic specificity for schizophrenia and bipolar disorder.
        BMC Psychiatry. 2011; 11: 18
        • Mourão-Miranda J.
        • Oliveira L.
        • Ladouceur C.D.
        • Marquand A.
        • Brammer M.
        • Birmaher B.
        • et al.
        Pattern recognition and functional neuroimaging help to discriminate healthy adolescents at risk for mood disorders from low risk adolescents.
        PloS One. 2012; 7: e29482
        • Huys Q.J.
        • Maia T.V.
        • Frank M.J.
        Computational psychiatry as a bridge from neuroscience to clinical applications.
        Nat Neurosci. 2016; 19: 404-413
        • Bassett D.S.
        • Bullmore E.
        Small-world brain networks.
        Neuroscientist. 2006; 12: 512-523
        • Liu Y.
        • Liang M.
        • Zhou Y.
        • He Y.
        • Hao Y.
        • Song M.
        • et al.
        Disrupted small-world networks in schizophrenia.
        Brain. 2008; 131: 945-961
        • Goodwin F.K.
        • Jamison K.R.
        Manic-Depressive Illness: Bipolar Disorders and Recurrent Depression.
        Oxford University Press, Oxford, United Kingdom2007
        • Loranger A.W.
        • Levine P.M.
        Age at onset of bipolar affective illness.
        Arch Gen Psychiatry. 1978; 35: 1345-1348
        • Smit D.J.
        • Stam C.J.
        • Posthuma D.
        • Boomsma D.I.
        • de Geus E.J.
        Heritability of “small‐world” networks in the brain: A graph theoretical analysis of resting‐state EEG functional connectivity.
        Hum Brain Mapp. 2008; 29: 1368-1378
        • Bullmore E.
        • Sporns O.
        Complex brain networks: Graph theoretical analysis of structural and functional systems.
        Nat Rev Neurosci. 2009; 10: 186-198
        • Connolly C.G.
        • Wu J.
        • Ho T.C.
        • Hoeft F.
        • Wolkowitz O.
        • Eisendrath S.
        • et al.
        Resting-state functional connectivity of subgenual anterior cingulate cortex in depressed adolescents.
        Biol Psychiatry. 2013; 74: 898-907
        • Cullen K.R.
        • Gee D.G.
        • Klimes-Dougan B.
        • Gabbay V.
        • Hulvershorn L.
        • Mueller B.A.
        • et al.
        A preliminary study of functional connectivity in comorbid adolescent depression.
        Neurosci Lett. 2009; 460: 227-231
        • Breakspear M.
        • Roberts G.
        • Green M.J.
        • Nguyen V.T.
        • Frankland A.
        • Levy F.
        • et al.
        Network dysfunction of emotional and cognitive processes in those at genetic risk of bipolar disorder.
        Brain. 2015; 138: 3427-3439