Altered Functional Subnetwork During Emotional Face Processing: A Potential Intermediate Phenotype for Schizophrenia

Importance  Although deficits in emotional processing are prominent in schizophrenia, it has been difficult to identify neural mechanisms related to the genetic risk for this highly heritable illness. Prior studies have not found consistent regional activation or connectivity alterations in first-degree relatives compared with healthy controls, suggesting that a more comprehensive search for connectomic biomarkers is warranted.

Objectives  To identify a potential systems-level intermediate phenotype linked to emotion processing in schizophrenia and to examine the psychological association, task specificity, test-retest reliability, and clinical validity of the identified phenotype.

Design, Setting, and Participations  The study was performed in university research hospitals from June 1, 2008, through December 31, 2013. We examined 58 unaffected first-degree relatives of patients with schizophrenia and 94 healthy controls with an emotional face-matching functional magnetic resonance imaging paradigm. Test-retest reliability was analyzed with an independent sample of 26 healthy participants. A clinical association study was performed in 31 patients with schizophrenia and 45 healthy controls. Data analysis was performed from January 1 to September 30, 2014.

Main Outcomes and Measures  Conventional amygdala activity and seeded connectivity measures, graph-based global and local network connectivity measures, Spearman rank correlation, intraclass correlation, and gray matter volumes.

Results  Among the 152 volunteers included in the relative-control sample, 58 were unaffected first-degree relatives of patients with schizophrenia (mean [SD] age, 33.29 [12.56]; 38 were women), and 94 were healthy controls without a first-degree relative with mental illness (mean [SD] age, 32.69 [10.09] years; 55 were women). A graph-theoretical connectivity approach identified significantly decreased connectivity in a subnetwork that primarily included the limbic cortex, visual cortex, and subcortex during emotional face processing (cluster-level P corrected for familywise error = .006) in relatives compared with controls. The connectivity of the same subnetwork was significantly decreased in patients with schizophrenia (F = 6.29, P = .01). Furthermore, we found that this subnetwork connectivity measure was negatively correlated with trait anxiety scores (P = .04), test-retest reliable (intraclass correlation coefficient = 0.57), specific to emotional face processing (F = 17.97, P < .001), and independent of gray matter volumes of the identified brain areas (F = 1.84, P = .18). Replicating previous results, no significant group differences were found in face-related amygdala activation and amygdala–anterior cingulate cortex connectivity (P corrected for familywise error =.37 and .11, respectively).

Conclusions and Relevance  Our results indicate that altered connectivity in a visual-limbic subnetwork during emotional face processing may be a functional connectomic intermediate phenotype for schizophrenia. The phenotype is reliable, task specific, related to trait anxiety, and associated with manifest illness. These data encourage the further investigation of this phenotype in clinical and pharmacologic studies.

Schizophrenia is a highly heritable mental disorder characterized by severe deficits in emotion processing.^1,2 Meta-analyses of patient data have pointed to a strong association between emotional deficits and dysfunctions of the limbic system.^3-5 However, whether and how these alterations relate to the genetic risk of schizophrenia remain unclear. A useful strategy to identify genetic mechanisms is the search for intermediate phenotypes, which are heritable traits related to the genetic predisposition to the disorder.^6,7 The study of unaffected first-degree relatives, who share an enriched set of schizophrenia risk genes but do not manifest clinical symptoms, provides important evidence for the establishment of a putative intermediate phenotype.^7-10 Although this strategy has been successful in cognitive domains, such as working memory,¹¹ declarative memory,^12,13 and reward processing,¹⁴ findings in emotional face processing have been inconclusive, with decreased,^15,16 increased,¹⁷ or unchanged¹⁸ amygdala reactivity and/or connectivity in relatives compared with controls.

This incongruence suggests that the search for an emotion-related intermediate phenotype may have to look beyond regional activation and connectivity analyses, which focus on a restricted set of regions and possibly overlook potential multidimensional network changes. Previous work^19-22 using graph theory–based methods has found significant alterations in the brain connectome of patients with schizophrenia during active tasks and resting state. The connectomic measures are heritable,^23,24 implying their utility for identifying intermediate phenotypes related to emotional face processing.

Our study aimed to identify potential connectomic intermediate phenotypes related to emotional face processing in schizophrenia. First, we used graph theory–based characterization of the connectome to identify potential functional network changes in unaffected first-degree relatives (58 relatives, 94 controls) using functional magnetic resonance imaging (fMRI) and a well-established emotional face-matching task.²⁵ In accordance with prior findings,²⁶ we posited alterations in connectivity between the visual cortex and the limbic system in relatives compared with controls. We further performed several follow-up analyses to investigate the utility of this potential intermediate phenotype by testing for psychological association, task specificity, test-retest reliability, and potential structural confounds. Second, we tested in an independent sample for the presence of the identified phenotype in schizophrenia (31 patients, 45 controls) for clinical validation.

Box Section Ref ID

One hundred fifty-two healthy volunteers from 3 sites in Germany (Mannheim, Bonn, and Berlin) were included in our relative-control sample from June 1, 2008, through December 31, 2013. Among these, 58 were unaffected first-degree relatives of patients with schizophrenia (mean [SD] age, 33.29 [12.56] years; 38 women), and 94 were healthy controls without a first-degree relative with mental illness (mean [SD] age, 32.69 [10.09] years; 55 women). The groups were balanced for a broad range of demographic, psychological, task performance, and image quality parameters (P > .10 for all) (Table 1 and eMethods in the Supplement). All participants provided written informed consent for the protocols approved by the institutional review boards of the University of Heidelberg, University of Bonn, and Universitätsmedizin Charité, Berlin.

Participants completed a well-established emotional face-matching fMRI task.^25,27,28 The face-matching task is an implicit emotional processing paradigm designed to challenge the amygdala.^25,27,28 The block-designed task consists of 2 conditions: an emotional condition (matching faces) and a control condition (matching forms). In the emotional condition, participants are presented with trios of faces from a standard set of pictures of facial affect²⁹ that depict fearful or angry expressions. Participants are instructed to match the 2 corresponding stimuli that illustrate the same individual. In the control condition, participants are presented with trios of simple geometric shapes (circles, vertical and horizontal ellipses) and are asked to match the 2 corresponding geometric shapes. In addition, we acquired resting-state fMRI images and high-resolution structural images for the participants (eMethods in the Supplement).

The MRI data were acquired from 3 Siemens 3-T scanners (Siemens Trio) with identical protocols. For fMRI data quality assurance, we quantified several head motion parameters and signal-to-noise ratio. Head motion parameters were quantified as previously detailed^14,30-32 and included the sum of volume to volume translational excursions across the time series, the sum of volume to volume rotational excursions across the time series, and the mean voxel-based framewise displacement. The signal to noise ratio of images was calculated using the New York University Center for Brain Imaging dataQuality toolbox (http://cbi.nyu.edu/software/dataQuality.php).^33-35 Statistical comparisons of data quality parameters between relatives and controls were performed with SPSS statistical software, version 20 (SPSS Inc), using independent t tests. As detailed in Table 1, groups were balanced for data quality assurance measures.

We first aimed to replicate the negative amygdala activation and seed-based connectivity results reported by Rasetti et al.¹⁸ We followed the methods outlined in that study and used standard procedures implemented in Statistical Parametric Mapping, version 8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). Briefly, for the activity analysis, functional images were realigned, slice time corrected, normalized to the Montreal Neurological Institute brain template, spatially smoothed, and subjected to first-level general linear model estimation. For the analysis of group differences, first-level contrast images (face matching greater than form matching) were entered into a second-level random-effects model. Results were reported after familywise error (FWE) correction across an a priori defined anatomical mask of the bilateral amygdala from the Automated Anatomical Labeling (AAL) atlas.³⁶ For the connectivity analysis, the mean time series of the bilateral amygdala from the AAL atlas were extracted after noise correction and entered into the first-level model. These first-level connectivity maps were then entered into a second-level random-effects model. Results for between-group effects were reported after FWE correction across a bilateral anterior cingulate cortex mask from the AAL atlas (eMethods in the Supplement).

Brain network analyses followed our previously published procedures.³¹ The mean time series during the face-processing task were extracted from each of the 90 anatomical nodes defined by the AAL atlas and corrected for noise. Whole-brain connectivity matrices were then calculated by pairwise Pearson correlations between nodes. The derived connectivity matrices were analyzed at 2 levels. At the global level, we further thresholded the matrices with multiple densities and computed 4 commonly used global network properties (smallworldness, modularity, global efficiency, and characteristic path length) for each density. Repeated-measures analysis of covariance (ANCOVA) models were used for group comparison with density as the within-subject factor and group (relatives vs controls) as the between-subject factor, covarying for age, sex, site, and mean framewise displacement. At the local level, the network-based statistic^37,38 was used for linkwise comparisons of the connectivity matrices between groups. This was done as previously described^37,38 by 2 steps. First, we computed initial t test statistics for all pairwise connections (contrasting relatives and controls and controlling for age, sex, site, and mean framewise displacement), which generated a set of suprathreshold links. Second, permutation testing was used to derive corrected P values for the generated link clusters (eMethods in the Supplement).

We performed several follow-up analyses to substantiate our connectomic finding as a potential intermediate phenotype for schizophrenia. In follow-up analyses 1 to 4, we reduced data dimensionality by averaging the connectivity estimates of all links in the identified subnetwork for each participant. See eMethods in the Supplement for details on the rationale and methods of the follow-up analyses.

Trait anxiety is an important assessment of emotion stability and may plausibly relate to the identified subnetwork connectivity phenotype. We probed for the presence of this association by calculating the Spearman rank correlation between the mean subnetwork connectivity estimates and trait anxiety scores derived from the State-Trait Anxiety Inventory across the participants in the control group. Significance was set at P < .05.

We further aimed to verify the presence of alterations in the identified subnetwork connectivity phenotype in schizophrenia. For this, we collected fMRI data from independent samples of 31 patients with schizophrenia (mean [SD] age, 30.58 [7.05] years; 7 women) and 45 healthy controls (mean [SD] age, 32.00 [9.50] years; 19 women). Participants were recruited from psychiatric hospitals and communities in and around Mannheim, Germany, and Bari, Italy. All patients were diagnosed as having schizophrenia by psychiatrists according to the Structured Clinical Interview for DSM-IV and were taking antipsychotic medication. Participants from Mannheim completed the same face-matching fMRI task used in the relatives study, and participants from Bari underwent scanning with a slightly modified version of the task³⁹ and a different scanner (Signa 3-T scanner, GE Healthcare) (Table 2 and eMethods in the Supplement). Data processing followed the same procedures described in the first study. Mean subnetwork connectivity was calculated for each participant and group difference in the derived metric was tested using an ANCOVA model, covarying for age, sex, site and mean framewise displacement.

Test-retest reliability of the identified subnetwork connectivity measure was analyzed using an independent sample of 26 healthy controls scanned twice with the same face-matching task (mean [SD] age, 24.4 [2.8] years; 15 women; mean [SD] scan interval, 14.6 [2.1] days; data previously reported by Cao et al³¹). Intraclass correlation coefficient (ICC) was used as an index of robustness. Consistent with established criteria,^30,31,40 an ICC greater than 0.40 was interpreted to be indicative of fair to good reliability.

To investigate whether the identified subnetwork finding is specific to emotional face processing or represents a task-independent functional connectomic alteration, we compared the subnetwork connectivity measure during face processing with that during the resting state in the same individuals. Group differences during 2 fMRI task conditions (ie, face matching, resting state) and group by task interaction effects were examined using a repeated-measures ANCOVA model.

We analyzed high-resolution structural data of relatives and controls to test whether the identified functional alterations may be influenced by structural differences. We compared mean gray matter volumes of the identified subnetwork nodes derived from high-resolution structural MRI and voxel-based morphometry. We also computed structural covariance matrices of cross-participant correlations between gray matter volumes and investigated linkwise group differences in the structural matrices. Significance was set at P < .05 after FWE correction.

Consistent with a previous study,¹⁸ significant differences in amygdala activity (small-volume P corrected for FWE = .37) and amygdala–anterior cingulate cortex seeded connectivity (small-volume P corrected for FWE = .11) between healthy first-degree relatives and controls during the face-matching task were not detected.

At the global level, no significant group differences were found for the measured network properties (P = .82 for smallworldness, P = .59 for modularity, P = .46 for global efficiency, P = .43 for characteristic path length) (eFigure 1 in the Supplement). At the subnetwork level, the network-based statistic identified a cluster that showed significantly decreased coupling in relatives compared with controls (P corrected for FWE = .006), which consisted of 49 links between pairs of 33 nodes chiefly in the limbic system (amygdala, hippocampus, parahippocampal gyrus, insula, orbitofrontal cortex), the visual cortex (calcarine sulcus, fusiform gyrus, superior and middle occipital gyrus, cuneus, lingual gyrus), and the subcortex (pallidum and thalamus) (Figure 1 and eTable 1 in the Supplement). The mean connectivity across all subnetwork links was significantly decreased in relatives of patients with schizophrenia.

The Spearman correlation analysis revealed a significantly negative correlation between mean subnetwork connectivity and State-Trait Anxiety Inventory trait scores (ρ = −0.22, P = .04) (Figure 2), suggesting that lower subnetwork connectivity is associated with higher anxiety in healthy individuals.

We observed a significant decrease of mean subnetwork connectivity in patients with schizophrenia compared with healthy controls (P = .01, partial η² = 0.08, indicating a medium effect size⁴¹) (Figure 3). This finding indicates that the identified subnetwork alteration is also present in patients with the same directionality detected in first-degree relatives.

The ICC values for mean subnetwork connectivity were ICC_2,1 = 0.57 and ICC_3,1 = 0.57. These values indicate a fair to good reliability of the identified intermediate phenotype.

In contrast to a significant group difference detected for the emotional task (P = 6.13 × 10⁻¹⁰), analysis of participants’ resting state data revealed no significant between-group differences in mean subnetwork connectivity (P = .29). In addition, a significant group by task interaction was found for the subnetwork measure (P = 4.0 × 10⁻⁵) (Figure 2).

There were no significant differences in mean gray matter volumes of the identified subnetwork nodes (P = .18) (eFigure 2 in the Supplement) and in the constructed structural covariance matrices (P corrected for FWE > .99) between relatives and controls.

In this study, we identified decreased functional connectivity in a visual-limbic subnetwork in unaffected first-degree relatives during emotional face processing, verified comparable abnormalities in patients with schizophrenia, and provided evidence of the utility of this potential intermediate phenotype by analyses of its psychological association, reliability, task dependence, and potential structural confounds. In humans, emotion processing depends on at least 2 multinodal cooperative brain systems^42-44: a ventral system for the identification of emotional significance and the initiation of corresponding responses, including fusiform gyrus, amygdala, insula, basal ganglia, thalamus, ventral anterior cingulate cortex, and orbitofrontal and ventrolateral prefrontal cortex, and a dorsal system for the regulation of affective states, including dorsolateral and dorsomedial prefrontal cortex, dorsal anterior cingulate cortex, and hippocampus. In schizophrenia, meta-analytical evidence points toward functional alterations in multiple brain nodes during emotional face processing, including fusiform gyrus,^3,5 amygdala,^3-5 hippocampus,⁵ medial and dorsolateral prefrontal cortex,^3,5 and subcortex.^3,5 Because these deficits range across more than 1 aforementioned system, they may plausibly reflect a failure in the coordination of interacting functional subunits of the emotional network. A previous study²⁶ further found decreased effective connectivity among several of these brain regions in unaffected siblings of patients with schizophrenia. This finding may imply that the proposed multisystem coordination failure relates to the genetic risk of the illness. In line with this proposal, our study identified a connectivity deficit in a large-scale emotional face-processing network in unaffected first-degree relatives of patients. The deficit spanned the ventral (eg, fusiform gyrus, amygdala, insula, basal ganglia, thalamus, orbitofrontal cortex) and dorsal (eg, hippocampus) emotion-processing systems. Because we verified consistent abnormalities in patients with schizophrenia, these findings highlight a potential connectomic intermediate phenotype for schizophrenia.

We performed several follow-up analyses to sustain our subnetwork finding as a potential intermediate phenotype related to emotional face processing. First, we found that the subnetwork connectivity was significantly correlated with trait anxiety scores, suggesting a link between the identified subnetwork and emotion-related personality. Second, we found that the subnetwork connectivity alteration in first-degree relatives was evident in the face-matching task but not in resting state, suggesting a degree of functional specificity of this potential intermediate phenotype. Third, consistent with our prior work,³¹ we found that the subnetwork measure was test-retest reliable, suggesting that the proposed connectivity feature is a promising target intermediate phenotype for longitudinal studies, such as drug interventions.

We also addressed several important confounding factors in this study. First, in accordance with common practice,^6,8,14 we controlled for demographic differences by balancing our sample for age, sex, site, educational level, and handedness. Second, similar to a prior study,¹⁴ we controlled for group differences in psychological characteristics and fMRI task performances, thereby ensuring that the detected subnetwork anomalies are not confounded by these behavioral measures. Third, considering that head motion may cause spurious effects on connectivity measures,^45,46 we balanced our sample for several head motion and signal-quality parameters and adjusted for head motion in our analyses. Fourth, given that morphometric abnormalities have been reported for some subnetwork nodes in patients with schizophrenia,^47-49 we probed for potential group differences in gray matter volumes and structural covariance patterns of the nodes. The results suggest that the functional subnetwork alterations are not explained by preexisting structural anomalies.

Our data have provided empirical evidence of several conceptual requirements outlined for putative intermediate phenotypes.^6-10 Specifically, this subnetwork is altered in unaffected individuals with increased genetic risk for schizophrenia, is associated with the illness, and is quantitatively reliable. Although establishing a neuroimaging intermediate phenotype that satisfies all requirements is extremely challenging and rarely accomplished in practice,¹⁰ and some other measures for this subnetwork finding require further investigation (eg, heritability), our study points to a promising intermediate phenotype for schizophrenia during emotional face processing.

There were some negative findings of this study. First, consistent with the findings of Rasetti et al,¹⁸ we did not find significant differences in face-related amygdala activation and seed-based connectivity measures, suggesting a greater sensitivity of the connectomic approach in search for intermediate phenotypes during emotional face processing. Second, there were no significant group differences in measured global network properties. Global properties are, by nature, less sensitive to focal changes at the subnetwork level yet require stringent correction for multiple comparisons.⁵⁰ Furthermore, global network properties derived from the face-matching task have been found less reliable than connectivity properties on which the network-based statistic method is based.³¹ These points may have plausibly contributed to the negative results in global properties.

Our study has several limitations. First, despite the relatively large sample size, first-degree relatives in our study were from several generations, with a small number of individuals younger than 20 years. However, most previous studies^15,17,18 only focus on healthy siblings older than 20 years to minimize potential confounding effects, such as cohort-specific life experiences and prodromal effects.^7,8 Second, our findings are based on an emotional face-matching task that allows the assessment of perceptual functions related to emotional face processing but not the differentiation of specific emotional functions in the amygdala and other limbic regions. Third, the exclusion of psychiatric disorders in relatives and controls in our study were based on self-report assessments rather than standardized interviews. Fourth, future work should replicate and extend these results by examining test-retest properties of the identified subnetwork pattern and investigating the ability of the subnetwork to distinguish relatives and controls in independent samples.

Our study provides initial evidence of a visual-limbic subnetwork alteration during emotional face processing that may be a promising intermediate phenotype for schizophrenia. The data presented in the study support the further exploration of this potential connectomic intermediate phenotype in clinical and pharmacologic studies.

Submitted for Publication: December 7, 2015; final revision received January 17, 2016; accepted January 24, 2016.

Corresponding Author: Andreas Meyer-Lindenberg, MD, PhD, Department of Psychiatry and Psychotherapy, Central Institute of Mental Health, Medical Faculty Mannheim, University of Heidelberg, Square J5, 68159 Mannheim, Germany (a.meyer-lindenberg@zi-mannheim.de).

Published Online: May 4, 2016. doi:10.1001/jamapsychiatry.2016.0161.

Author Contributions: Drs Cao, Bertolino, Walter, Tost, and Meyer-Lindenberg contributed equally to this study. Drs Cao and Tost have full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Cao, Walter, Tost, Meyer-Lindenberg.

Acquisition, analysis, or interpretation of data: Cao, Bertolino, Walter, Schneider, Schäfer, Taurisano, Blasi, Haddad, Grimm, Otto, Dixson, Erk, Mohnke, Heinz, Romanczuk-Seiferth, Mühleisen, Mattheisen, Witt, Cichon, Noethen, Rietschel, Tost, Meyer-Lindenberg.

Drafting of the manuscript: Cao, Dixson, Schneider, Tost.

Critical revision of the manuscript for important intellectual content: Cao, Bertolino, Walter, Schneider, Schäfer, Taurisano, Blasi, Haddad, Grimm, Otto, Dixson, Erk, Mohnke, Heinz, Romanczuk-Seiferth, Mühleisen, Mattheisen, Witt, Cichon, Noethen, Rietschel, Tost, Meyer-Lindenberg.

Statistical analysis: Cao, Schäfer, Taurisano, Grimm, Cichon, Meyer-Lindenberg.

Obtained funding: Walter, Heinz, Noethen, Rietschel, Tost, Meyer-Lindenberg.

Administrative, technical, or material support: Walter, Schäfer, Grimm, Heinz, Romanczuk-Seiferth, Cichon, Rietschel.

Study supervision: Tost, Meyer-Lindenberg.

Conflict of Interest Disclosures: Dr Bertolino reported working as a full-time employee of Hoffmann-La Roche, Basel, Switzerland. Dr Walter reported receiving a speaker honorarium from Servier. Dr Meyer-Lindenberg reported receiving consultant fees from AstraZeneca, Elsevier, F. Hoffmann-La Roche, Gerson Lehrman Group, Lundbeck, Outcome Europe Sárl, Outcome Sciences, Roche Pharma, Servier International, and Thieme Verlag and lecture fees, including travel expenses, from Abbott, AstraZeneca, Aula Médica Congresos, BASF, Boehringer Ingelheim, Groupo Ferrer International, Janssen-Cilag, Lilly Deutschland, LVR Klinikum Düsseldorf, Otsuka Pharmaceuticals, and Servier Deutschland. No other disclosures were reported.

Funding/Support: This study was supported by IntegraMent grants 01ZX1314G and 01ZX1314B (Drs Meyer-Lindenberg, Walter, and Heinz) and NGFNplus MooDS grants 01GS08144, 01GS08147, and 01GS08148 (Drs Noethen, Cichon, Walter, Rietschel, Meyer-Lindenberg, and Heinz) from the German Federal Ministry of Education and Research and grants 115300 (Project EU-AIMS) (Dr Meyer-Lindenberg), 115008 (Project EU-NEWMEDS) (Dr Meyer-Lindenberg), 602805 (Project EU-AGGRESSOTYPE) (Dr Meyer-Lindenberg), and 602450 (Project EU-IMAGEMEND) (Dr Meyer-Lindenberg) from the European Community’s Seventh Framework Programme. Dr Tost acknowledges support from grant 01GQ1102 from the German Federal Ministry of Education and Research. Dr Cao is a PhD scholarship awardee of the Chinese Scholarship Council.

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication.

Additional Contributions: Christine Esslinger, MD, Dagmar Gass, Daniela Mier, PhD, and Carina Sauer, PhD (Central Institute of Mental Health, Manheim Germany), provided research assistance. No compensation was provided.