Default Mode Functional Connectivity is Associated with Social Functioning in Schizophrenia

Abstract

Individuals with schizophrenia display notable deficits in social functioning. Research indicates that neural connectivity within the default mode network (DMN) is related to social cognition and social functioning in healthy and clinical populations. However, the association between DMN connectivity, social cognition, and social functioning has not been studied in schizophrenia. For the present study, we used resting-state neuroimaging data to evaluate connectivity between the main DMN hubs (i.e., the medial prefrontal cortex (mPFC) and the posterior cingulate cortex-anterior precuneus (PPC)) in individuals with schizophrenia (n=28) and controls (n=32). We also examined whether DMN connectivity was associated with social functioning via social attainment (measured by the Specific Levels of Functioning Scale) and social competence (measured by the Social Skills Performance Assessment), and if social cognition mediates the association between DMN connectivity and these measures of social functioning. Results revealed that DMN connectivity did not differ between individuals with schizophrenia and controls. However, connectivity between the mPFC and PCC hubs was significantly associated with social competence and social attainment in individuals with schizophrenia but not in controls as reflected by a significant group-by-connectivity interaction. Social cognition did not mediate the association between social functioning and DMN connectivity in individuals with schizophrenia. Our findings suggest that fronto-parietal DMN connectivity in particular may be differentially associated with social functioning in schizophrenia and controls. As a result, DMN connectivity may be used as a neuroimaging marker to monitor treatment response or as a potential target for interventions that aim to enhance social functioning in schizophrenia.

Social functioning measures the way we engage with family, friends, co-workers, and service providers in conversation, shared decision-making, compromise, and social activities (Kalin et al., 2015). Social functioning is notably impaired in schizophrenia and is considered by many to be a fundamental characteristic of the disorder (Hafner, Nowotny, Loffler, & an der Heiden, 1995). Restoring an individual with schizophrenia’s social functioning via their social attainment (i.e., one’s social relationships and participation in community activities) and social competence (i.e., one’s capacity to effectively socialize with others) is a cornerstone of functional recovery. To date, research has evaluated several interventions aimed at improving social functioning (Horan et al., 2009; Horan et al., 2011; Kurtz & Mueser, 2008; Liberman, Mueser, & Wallace, 1986; Penn, Roberts, Combs, & Sterne, 2007). However, the effects have been modest, and as such, interventions targeting the social dysfunction associated with schizophrenia could benefit from additional optimization (Kurtz & Mueser, 2008; Kurtz & Richardson, 2012). One approach to optimize these interventions is to identify the mechanisms that underlie social functioning impairments, such as the neural systems that support key social behaviors in schizophrenia. Therefore, research is needed to understand the associations between neural networks and social functioning among individuals with schizophrenia.

The “dysconnectivity hypothesis” is a promising theory from which to investigate the neural basis of social functioning. This hypothesis suggests that abnormal communication between neural networks is responsible for the cognitive and clinical symptoms of schizophrenia, including social functioning deficits (Bullmore, Frangou, & Murray, 1997; Friston, 1994; Friston & Frith, 1995; Stephan, Baldeweg, & Friston, 2006). For instance, several studies observed aberrant connectivity within the default mode network (DMN) among individuals with schizophrenia (Alonso-Solis et al., 2012; Bluhm et al., 2007; Camchong, MacDonald, Bell, Mueller, & Lim, 2011; Chai et al., 2011; Liemburg et al., 2012; Liu et al., 2012; Ongur et al., 2010; Whitfield-Gabrieli et al., 2009; Zhou et al., 2007). Moreover, this aberrant DMN connectivity has been associated with deficits in social cognition (e.g., social perception, mentalizing) among individuals with schizophrenia (Brunet, Sarfati, Hardy-Bayle, & Decety, 2003; Delaveau et al., 2010; Fett et al., 2011; Mars et al., 2012; Mitchell, Banaji, & Macrae, 2005; Pelletier-Baldelli, Bernard, & Mittal, 2015; Shi et al., 2015; Spreng & Grady, 2010; Uddin, Lacoboni, Lange, & Keenan, 2007; Walter et al., 2009). Specifically, social perception is the ability to perceive social cues such as facial affect, tone, or gestures, while mentalizing is the ability to ascertain others’ emotions, beliefs, and intentions (Green, Horan, & Lee, 2015). In turn, a meta-analysis suggests that social cognition is the most proximal factor to social functioning among individuals with schizophrenia (Fett et al., 2011). Thus, social cognition may represent an underlying mechanism that could explain the association between DMN connectivity and social functioning.

Alternatively, prior studies suggest that functional connectivity of the default mode network (DMN) is directly related to social functioning in various clinical and healthy populations (Che et al., 2014; Dodell-Feder, DeLisi, & Hooker, 2014; Jung et al., 2014; Schreiner et al., 2013; Washington & VanMeter, 2015; Yerys et al., 2015). Specifically, increased functional connectivity within the DMN has been associated with better social functioning in healthy controls (Che et al., 2014; Jung et al., 2014; Washington & VanMeter, 2015), and decreased DMN functional connectivity has been associated with social impairment in individuals with autism and 22q11.2 deletion syndrome (Jung et al., 2014; Schreiner et al., 2013; Yerys et al., 2015). Most recently, Dodell-Feder and colleagues found that reduced resting-state DMN connectivity was associated with poorer social functioning in healthy controls and first-degree relatives of individuals with schizophrenia (Dodell-Feder et al., 2014). Thus, the DMN’s aberrant connectivity in individuals with schizophrenia, its relation to social cognition, and its relation to social functioning in various populations suggest that the DMN may be directly associated with social functioning in individuals with schizophrenia. To our knowledge, prior research has not investigated the associations between DMN functional connectivity and social functioning or the underlying mechanisms for such associations in schizophrenia. Thus, a direct investigation could elucidate whether DMN connectivity is a potential treatment target for functional recovery.

The DMN consists of a frontal hub (i.e., highly connected area within the brain) located in the medial prefrontal cortex and a posterior hub located in the posterior cingulate cortex and precuneus that are highly connected with each other and other areas of the brain (Andrews-Hanna, 2012; Laird et al., 2011). In the current paper, we define DMN functional connectivity as inter-network connectivity (i.e., correlations between the timeseries of data-derived neural networks) between the medial prefrontal cortex (mPFC) and posterior cingulate cortex and anterior precuneus (PPC) hubs. Prior research suggests inter-network connectivity of the mPFC and PPC hubs is related to social functioning among healthy individuals and neuropsychiatric populations (Che et al., 2014; Dodell-Feder et al., 2014; Jung et al., 2014; Schreiner et al., 2013; Washington & VanMeter, 2015; Yerys et al., 2015). Specifically, inter-network connectivity between the mPFC and PPC hubs activates during self-reflection (Amodio & Frith, 2006; Andrews-Hanna, 2012; Ochsner et al., 2004; Schmitz & Johnson, 2007) and when a person is presented with information about people who are important to them (e.g., friends, relatives, etc.) (Andrews-Hanna, 2012; Bartels & Zeki, 2004; Ochsner et al., 2005). This inter-network connectivity also activates when a person is anticipating a social reward or threat, such as positive or negative affect (Andrews-Hanna, 2012; Kober et al., 2008; Maddock, 1999).

Additionally, DMN connectivity can be observed during both resting and task-activated neural states. Meta-analytic results suggest that the spatial topographies of intrinsic connectivity networks (ICNS) at rest are similar to networks derived during task (Fox & Raichle, 2007; Laird et al., 2011; Smith et al., 2009). However, researchers can use resting-state methods to overcome certain limitations that are associated with task-based approaches (e.g., individual differences in attention, effort, or comprehension) (Callicott et al., 2000; Callicott et al., 2003; Hooker, Bruce, Lincoln, Fisher, & Vinogradov, 2011). Additionally, resting-state methods may be useful for capturing brain activity related to broader constructs that include multiple processes (e.g., social functioning) as this method does not depend on specific task demands (Abram et al., 2015). Thus, resting-state methods may be particularly useful for examining associations between connectivity and social functioning.

A common approach for evaluating patterns of resting-state connectivity is probabilistic independent component analysis (pICA) (McKeown et al., 1998), which is a data-driven procedure that separates multivariate signals into statistically independent sources, or group-level spatial maps. More specifically, the group-level spatial maps reflect temporally synchronized fluctuations in fMRI signals (Beckmann & Smith, 2004). As a data-driven procedure, pICA generates components that represent functionally integrated brain regions as opposed to structurally defined brain regions. In comparison to other connectivity approach (e.g., seed-based methods), pICA reduces bias by parsing variance related to artifacts (i.e., head motion, cardiac function) from neural activity of interest (Beckmann, DeLuca, Devlin, & Smith, 2005). We therefore used pICA to derive group-level spatial maps for the current study.

pICA can also help establish whether individuals with schizophrenia are characterized by either hypo- or hyper-DMN connectivity when compared to healthy controls as current findings are mixed (Alonso-Solís et al., 2015; Chang et al., 2014; Jafri, Pearlson, Stevens, & Calhoun, 2008; Li et al., 2015; Liang et al., 2006; Liemburg et al., 2012; Mingoia et al., 2012; Ongur et al., 2010; Schilbach et al., 2016; Whitfield-Gabrieli & Ford, 2012; Zhou et al., 2007). The inconsistent findings may be attributed to methodological differences, such as seed versus pICA analyses, or due to differences in the specific networks that were examined (Jafri et al., 2008; Whitfield-Gabrieli & Ford, 2012; Zhou et al., 2007). Many studies that evaluated resting-state DMN connectivity in schizophrenia using pICA reported on the connectivity between the main DMN hubs and other brain networks (Alonso-Solís et al., 2015; Jafri et al., 2008; Liang et al., 2006; Ongur et al., 2010; Zhou et al., 2007). Of the studies examining DMN connectivity (Alonso-Solís et al., 2015; Chang et al., 2014; Jafri et al., 2008; Li et al., 2015; Liang et al., 2006; Liemburg et al., 2012; Mingoia et al., 2012; Ongur et al., 2010; Schilbach et al., 2016; Zhou et al., 2007), only two studies directly examined connectivity between the DMN hubs (Chang et al., 2014; Liemburg et al., 2012). One of these studies found that individuals with schizophrenia exhibited decreased DMN inter-network connectivity as compared to controls (Liemburg et al., 2012), and one study did not observe group differences in DMN inter-network connectivity (Chang et al., 2014). Due to conflicting findings in the literature, further research comparing DMN inter-network connectivity of individuals with schizophrenia to controls is needed.

The aims of the current study were threefold. First, we compared resting-state DMN inter-network connectivity between individuals with schizophrenia and controls. We hypothesized that individuals with schizophrenia, as compared to controls, would either show decreased or similar connectivity between the DMN hubs. Second, we assessed between-group differences in DMN connectivity as well as the relationship between DMN connectivity and social functioning via measures of social attainment and social competence. We hypothesized that DMN connectivity would be positively associated with social attainment and social competence in healthy controls and individuals with schizophrenia. Third, we investigated whether social cognition (i.e., social perception and mentalizing) mediated the association between DMN connectivity and social functioning among individuals with schizophrenia.

Method

Participants and Procedure

Individuals with schizophrenia (n=28) and controls (n=32) ages 18–45 were group-matched for age, gender, and race. Participants completed a clinical interview, neuropsychological battery, and resting-state functional magnetic resonance imaging scan. Participants were recruited from the Northwestern University Schizophrenia Research Group. Details on recruitment can be found in Smith et al. (2015).¹ Individuals with schizophrenia were included in the study if they 1) met the Diagnostic and Statistical Manual of Mental Disorders, 4^th edition (DSM-IV) criteria for a diagnosis of schizophrenia, 2) were clinically stable (i.e., their symptoms remained unchanged for at least 2 weeks) (Rastogi-Cruz & Csernansky, 1997), and 3) were currently on antipsychotic medication. Individuals were excluded from the study if they 1) met DSM-IV criteria for intellectual disability, 2) met substance abuse or dependence DSM-IV criteria in the past 6 months, 3) had a documented neurological injury or disorder, or 4) had a severe medical disorder. Controls were also excluded from the study if they 1) had a lifetime history of a DSM-IV Axis I psychiatric disorder or 2) had a first-degree relative with a psychotic disorder. Additionally, we excluded nine participants from the analysis (five individuals with schizophrenia and four controls), for excessive in-scanner motion (mean absolute displacement above 1.5 mm, or any absolute displacement (translations or rotations) above 3 mm/degrees). The institutional review board at Northwestern University Feinberg School of Medicine approved the study procedures, and all participants provided informed consent (IRB approval number: STU00013034).

Measures

Demographic and clinical measures

DSM-IV diagnosis was determined through the Structured Clinical Interview for DSM Disorders (SCID-IV) (First, Spitzer, Gibbon, & Williams). The SCID-IV was conducted by masters- or PhD-level research staff and validated through a semi-structured interview by a study psychiatrist. All antipsychotic medication dosages were converted to chlorpromazine equivalents (CPZeq) (Andreasen, Pressler, Nopoulos, Miller, & Ho, 2010). Clinical symptoms of schizophrenia were assessed using the Scale for the Assessment of Positive Symptoms (SAPS) (Andreasen, 1983) and the Scale for the Assessment of Negative Symptoms (SANS) (Andreasen, 1983). Global ratings for hallucinations, delusions, bizarre behavior, positive formal thought disorder, affective flattening, alogia, avolition, and anhedonia were provided by masters- or PhD-level research staff. We generated global scores for positive, negative, and disorganized symptoms by calculating the mean of positive SAPS items, negative SANS items, and disorganized SANS items. Mean parental socioeconomic status (SES) was assessed using the Barrett Simplified Measure of Social Status (Barratt, 2005).

Neurocognitive assessments

Participants completed a battery of tests used to generate four neurocognitive domains (Nuechterlein et al., 2004), including: 1) Crystallized IQ from scores on the Vocabulary subtest of the Wechsler Adult Intelligence Scale – Third Edition (WAIS-III) (Wechsler, 1997a), 2) Working Memory from scores on the Continuous Performance Task (CPT) (Barch et al., 2004) and the Digit-Span, Spatial-Span, and Letter-Number Sequencing subtests of the Wechsler Memory Scale-Third Edition (WMS-III) (Wechsler, 1997b), 3) Episodic Memory from scores on the Logical Memory I and Family Pictures I subtests of the WMS-III and the free recall score from the first five trials of the California Verbal Learning Test (CVLT) (Delis, Kramer, Kaplan, & Ober, 1983), 4) Executive Functioning from scores on the Wisconsin Card Sort Task (WCST) (Heaton, 2003), Letter and Animal Fluency (Benton, Hamsher, & Sivan, 1994), the Trail Making Test Part B (Reitan & Wolfson, 1985), and the Matrix Reasoning subtest of the WAIS-III. Domain scores were calculated by first standardizing individual subtests (using control group data) and then averaging across these standardized scores within each domain (Smith et al., 2012). A global neurocognitive score was computed using the average of these four domains.

Social Cognition

Social Perception

We assessed social perception using a facial affect perception task and a social perception task. The facial affect perception task presented participants with 30 faces displaying happiness, sadness, fear, disgust, anger, or neutrality, and participants selected the correct emotional expression from two emotion labels (Smith et al., 2014). We used the half-profile of nonverbal sensitivity (PONS) to measure social perception. Specifically, the task presented participants with 110 video scenes containing facial expressions, voice intonations, and/or gestures. Following the scene, participants were presented with two labels and chose which label best described the social cue (Ambady, Hallahan, & Rosenthal, 1995; Rosenthal et al., 1979). We mean-centered accuracy rates to the control group, and averaged transformed scores from both tasks to obtain a social perception domain score.

Mentalizing

We assessed mentalizing using cognitive empathy tasks, specifically the Awareness of Social Inference Test (TASIT) Part III (McDonald, 2002) and an emotional perspective-taking task (Smith et al., 2014). The TASIT-III presented participants with 16 video vignettes of common social interactions. Each vignette contained an untrue comment presented as either a lie or sarcasm. Participants answered yes/no questions about the thoughts, intentions, beliefs, and feelings of the people in the vignette. The emotional perspective-taking task presented participants with 60 pictures of two person interactions depicting happiness, sadness, fear, disgust, anger, or neutrality. In each picture, one person’s face was masked and participants chose which of two faces depicted the emotion of the masked actor. We mean-centered accuracy rates to the control group, and then averaged the transformed accuracy scores to obtain the mentalizing domain score.

Social functioning

Social attainment

We assessed social attainment using the interview version of the Specific Levels of Functioning Scale (SLOF), which is a 30 item interview-based measure that assesses the following domains of social attainment: interpersonal relationships, social acceptability, activities of daily living, and work skills (Schneider & Struening, 1983). Higher scores on the SLOF indicate better social attainment (range = 84–150). Variance in SLOF scores did not significantly differ between groups (F_27,31 = 1.38, P = .39). The SLOF is considered the gold standard for assessing current functioning based on findings from the VALERO Study (Harvey et al., 2011; Leifker, Patterson, Heaton, & Harvey, 2011).

Social competence

We assessed social competence using the Social Skills Performance Assessment (SSPA) (Patterson, Moscona, McKibbin, Davidson, & Jeste, 2001), which is comprised of 2 role-play scenes between a trained actor and the participant that are video-recorded. The scenes involve meeting a new neighbor and making a maintenance request to a landlord. Two trained research assistants rated the performance on a 5-point scale across 8 criteria for the first scene and 9 criteria for the second scene. Additional details on the use of this measure can be found here: Smith et al., 2014. We calculated a final score by averaging the scores for each scene (ICC = .97 for 2 blinded raters on 25% of the videos). Higher scores on the SSPA indicate better social competence (range = 1.66–5.00). Variance in SSPA scores did not significantly differ between groups (F_22,26 = 1.89, P = .12). The SSPA is considered the gold standard for assessing social competence in schizophrenia (Harvey, Velligan, & Bellack, 2007; Kalin et al., 2015).

fMRI data acquisition and pICA

Resting-state scans were acquired on a 3T TIM Trio system (Siemens Medical Systems) scanner at Northwestern University’s Feinberg School of Medicine for the study sample. The scanning parameters included: gradient-echo echo-planar imaging of 164 volumes; repetition time (TR) = 2.5 s; echo time (TE) = 20 ms; flip angle = 80°; voxel size = 1.7 × 1.7 × 3 mm. The group-level spatial maps were derived from resting-state scans of an independent community sample of 218 volunteers collected at the University of Minnesota (mean age = 26 years [range 20 to 39], 49% male) (Abram et al., 2015). Data were preprocessed using the following preprocessing steps in the MELODIC (Multivariate Exploratory Linear Optimized Decomposition into Independent Components) toolkit in FSL (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/MELODIC): brain extraction, motion correction, and high-pass temporal filtering (threshold of 0.1 Hz) (Wisner, Patzelt, Lim, & MacDonald, 2013). As a final step, the data underwent motion regression outside of FSL using participant-specific movement parameters generated during registration (Moodie, Wisner, & MacDonald, 2014; Wisner, Atluri, Lim, & MacDonald, 2013).

As noted previously, the group-level spatial maps were generated from a larger dataset of 218 community volunteers using a meta-melodic spatial pICA pipeline (Abram et al., 2015). More specifically, the MELODIC function in FSL was used to run 25 temporal concatenation (model-free and multi-subject) group-level pICAs. Each ICA included a random order of 80 participants as inputs to decrease the likelihood of over-fitting, with a dimensionality constraint of 60 based on reliability research (Poppe et al., 2013) and findings that large-scale networks, such as the DMN, fractionate at higher dimensionalities (Ray et al., 2013; Wisner, Atluri, et al., 2013; Wisner, Patzelt, et al., 2013). The 60 components from each ICA were merged into one file which was used as the input to a meta-level MELODIC (meta-ICA). The meta-ICA generated the 60 most consistent group-level components. Visual inspection was done to identify artifactual components, which included components likely to show homeostatic fluctuations, white matter tracts, or movement (Kelly et al., 2010). 27 non-artifactual components were identified following these procedures.

As recommended by prior research, we used maps from this external sample because maps generated from larger datasets are likely to be more robust and these maps were unbiased to the schizophrenia or control groups used in the present study (Griffanti et al., 2016). We also used ICA maps from this larger dataset rather than mPFC and PCC atlas masks because we did not want to make assumptions regarding exactly which voxels may contribute to a network (Esposito et al., 2005; Ramnani, Behrens, Penny, & Matthews, 2004). By using maps from an external sample, we were able to assess functional connectivity within and between functionally-derived networks as opposed to structurally-defined networks. We applied the group-level spatial maps from the external dataset to our current dataset through FSL’s dual-regression function. This enabled us to derive individual timeseries and corresponding spatial maps for the current participants. We used the participant-specific timeseries and spatial maps to compute connectivity metrics for each participant.

Inter-network connectivity metric computations

We used dual regression to create connectivity maps and time courses for each subject based on the group-level spatial maps from the meta-ICA (Filippini et al., 2009; Wisner, Patzelt, et al., 2013; Zuo et al., 2010). We calculated inter-network connectivity metrics between all group-level spatial maps using the participant-level timeseries data from the dual regression (Wisner, Patzelt, et al., 2013). These values were Pearson correlations between all pairs of participant-level timeseries (i.e., for all combinations of the 60 networks), and reflect the temporal association between network pairs. Group-level means were calculated using Fischer-Z values, and were transformed back to r-values for reporting.

In the current study, we were interested in connectivity between the mPFC and the PPC (see Supplemental Figure 1). Of the 60 group-level spatial maps (Abram et al., 2015), three non-artifactual components included mPFC and twelve non-artifactual components included PPC (see Supplemental Table 1). We examined the 4 components that included an mPFC or PPC-overlap of at least 750 voxels in our analyses to ensure appropriate mPFC or PPC contribution. These components included the following compositions: the PPC, the mPFC, the precuneus (P), and the orbitofrontal cortex (OFC). Other areas such as the OFC and angular gyrus (included in the PPC component) likely emerged through this analysis because they are highly functionally connected to the main DMN hubs and are related to social functioning (Andrews, Wang, Csernansky, Gado, & Barch, 2014; Andrews-Hanna, 2012). The inter-network connectivity metrics were then as follows: 1) PPC-to-P metric, 2) mPFC-to-PPC metric, 3) OFC-to-PPC metric, 4) mPFC-to-P metric, 5) OFC-to-P metric, and 6) mPFC-to-OFC metric (see Table 1 for a full description of these metrics).

Table 1.

Inter-Network Connectivity Metrics

Hubs	Inter-network Connectivity Metric
Posterior Cingulate Cortex/Anterior Precuneus with additional Precuneus areas	PPC-to-P metric
Medial Prefrontal Cortex with Posterior Cingulate Cortex/Anterior Precuneus	mPFC-to-PPC metric
Orbitofrontal Cortex with Posterior Cingulate Cortex/Anterior Precuneus	OFC-to-PPC metric
Medial Prefrontal Cortex with Precuneus	mPFC-to-P metric
Orbitofrontal Cortex with Precuneus	OFC-to-P metric
Medial Prefrontal Cortex with Orbitofrontal Cortex	mPFC-to-OFC metric

Potential movement confounds

Movement was calculated as the root mean square (RMS) absolute and incremental movement for each group (Power, Barnes, Snyder, Schlaggar, & Petersen, 2012). We did not detect group differences in absolute (x̄_absCON = 0.36, x̄_absSCZ = 0.31, t₄₅= −0.90, P = 0.37) or incremental (x̄_incCON = 0.05, x̄_incSCZ = 0.05, t₄₅= −0.73, P = 0.47) movement. RMS incremental movement was significantly correlated with the PPC-to-P metric (r = 0.31, P < .05). Neither RMS absolute nor incremental movement was correlated with the other inter-network connectivity metrics, social attainment, or social competence (all P > .10). Therefore, we only included RMS incremental movement as a covariate in models that contained the PPC-to-P metric.

Statistical Analyses

Demographic and behavioral analyses

Group differences for demographic variables, neurocognitive performance, social perception, mentalizing, social attainment, and social competence were evaluated using t-tests for continuous variables and chi-square (χ²) tests for categorical variables.

Connectivity analyses

We used multivariate analysis of variance (MANOVA) to evaluate whether individuals with schizophrenia differed from controls with respect to the DMN connectivity metrics; in this model the PPC-to-P, mPFC-to-PPC, OFC-to-PPC, mPFC-to-P, OFC-to-P, and mPFC-to-OFC metrics served as the dependent variables.

Connectivity and social functioning analyses

Main Effects Model

We used robust linear regression to assess the associations between our DMN connectivity variables and social attainment and social competence. More specifically, we created two models with social attainment as the dependent variable in the first model, social competence as the dependent variable in the second model, and the PPC-to-P, mPFC-to-PPC, OFC-to-PPC, mPFC-to-P, OFC-to-P, and mPFC-to-OFC metrics and group status as the independent variables in both models. We used the modeling package ‘robust’ in R (Wang et al., 2008) to adjust for the presence of multivariate outliers (Field, 2009). To rule out RMS incremental movement as confounds, we re-assessed the association between DMN connectivity and social attainment and social competence while including the aforementioned variable as a covariate.

Follow-up Interaction Models

We used robust linear regression to examine between-group differences in the associations between DMN connectivity and social attainment and social competence. To conserve power, we only included DMN connectivity metrics that were significantly associated with social attainment or social competence in the main effects models. We again created two models with social attainment as the dependent variable in the first model, social competence as the dependent variable in the second model, and the mPFC-to-PPC metric, group status, and a group-by-connectivity interaction term as the independent variables in both models. Based on prior research suggesting that social status may be associated with neural activity in the main DMN hubs (Muscatell et al., 2012), we included parental SES as a covariate in both models. To account for the association between neurocognitive and social deficits observed in schizophrenia (Fett et al., 2011; Ventura et al., 2015), we also included mean global neurocognition as a covariate in both models. Finally, we included gender and age as covariates in both models given research that shows gender and age-related resting-state connectivity differences (Damoiseaux et al., 2008; Satterthwaite et al., 2014; Tian, Wang, Yan, & He, 2011).

Mediation analyses

In order for a variable to be a mediator, it must be related to both the independent and dependent variables (Baron & Kenny, 1986). Therefore, to determine if mediation analyses were appropriate for our proposed mediators (i.e., social perception and mentalizing), we first examined Pearson correlations to see whether the proposed mediators were significantly correlated with the social functioning variables and DMN connectivity in individuals with schizophrenia. For any significant associations, we performed mediation analyses to see if social perception or mentalizing mediated these associations.

Connectivity and activities of daily living analysis

As a final step, we assessed the specificity of any DMN connectivity and social functioning associations. Specifically, we tested for associations between DMN connectivity and a performance-based daily living skills measure (UPSA-B).

Results

Participant Characteristics

As shown in Table 2, individuals with schizophrenia and controls did not differ according to age, gender, or race. However, the groups differed with regard to parental SES (P < .01), global neurocognition (P < .001), social perception (P < .01), and mentalizing (P < .001). Individuals with schizophrenia also scored lower than controls on the social attainment and social competence measures (both P < .001). Descriptive data for duration of illness, CPZeq, and clinical symptoms are also presented in Table 2.

Table 2.

Study Sample Characteristics

	CON (n = 32)	SCZ (n = 28)	Χ2/t Statistics
Demographics
Age, mean (SD)	31.46 (8.06)	33.17 (6.63)	0.90
Gender (% male)	53.13	64.29	0.38
Parental SES, mean (SD)a	30.08 (9.21)	23.15 (9.87)	−2.75**
Duration of illness, mean years (SD)	---	14.57 (6.34)	---
CPZeq, mean (SD)	---	329.79 (207.31)	---
Race
% Caucasian	50.00	42.90	0.31
% African American	34.40	39.30
% Other	15.63	17.86
Neurocognitive Function
Global neurocognition, mean (SD)	0.00 (0.64)	−0.90 (0.58)	−5.64***
Clinical Symptomsb
Positive symptoms, mean (SD)	---	2.57 (1.86)	---
Negative symptoms, mean (SD)	---	2.86 (1.08)	---
Disorganized symptoms, mean (SD)	---	1.80 (1.29)	---
Social Cognitive Measures
Social Perception, mean (SD)	0.00 (0.86)	−0.77 (1.05)	−3.08**
Mentalizing, mean (SD)	0.00 (0.87)	−1.34 (1.09)	−5.18***
Functioning Measures
Social Attainment, mean (SD)	141.34 (12.72)	125.43 (14.93)	−4.41***
Social Competence, mean (SD)c	4.49 (0.63)	3.19 (0.87)	−5.96***

Note: SCZ, schizophrenia participants; CON, controls; SES, socioeconomic status; CPZeq, chlorpromazine equivalent.

^aCompleted by n = 31 CON and n = 27 SCZ

^bBased off of SAPS SANS Global Ratings

^cCompleted by n = 27 CON and n = 23 SCZ

^*P < .05,

^**P < .01,

^***P < .001

Between-Group Connectivity Analysis

Table 3 reports the between-group connectivity results. MANOVA revealed that individuals with schizophrenia and controls did not differ with respect to any of the connectivity variables: PPC-to-P metric, mPFC-to-PPC metric, OFC-to-PPC metric, mPFC-to-P metric, OFC-to-P metric, or the mPFC-to-OFC metric (F_6,53 = 0.49, P = .81).

Table 3.

Between-group comparisons of inter-network connectivity metrics

	CON (n = 32)	SCZ (n = 28)	F Statistic	P-value
Mean PPC-to-P metric (SD)	0.69 (0.0.21)	0.66 (0.22)	0.49	0.81
Mean mPFC-to-PPC metric (SD)	0.26 (0.27)	0.25 (0.30)
Mean OFC-to-PPC metric (SD)	−0.03 (0.22)	−0.05 (0.20)
Mean mPFC-to-P metric (SD)	0.31 (0.21)	0.24 (0.26)
Mean OFC-to-P metric (SD)	−0.06 (0.17)	−0.11 (0.21)
Mean mPFC-to-OFC metric (SD)	0.39 (0.20)	0.38 (0.24)

Note: SCZ, schizophrenia participants; CON, controls; PPC-to-P metric, inter-network connectivity between the posterior cingulate cortex-anterior precuneus and the precuneus; mPFC-to-PPC metric, inter-network connectivity between the medial prefrontal cortex and the posterior cingulate cortex-anterior precuneus; OFC-to-PPC metric, inter-network connectivity between the orbital frontal cortex and the posterior cingulate cortex-anterior precuneus; mPFC-to-P metric, inter-network connectivity between the medial prefrontal cortex and the precuneus; OFC-to-P metric, inter-network connectivity between the orbital frontal cortex and the precuneus; mPFC-to-OFC metric, inter-network connectivity between the medial prefrontal cortex and the orbital frontal cortex

Between-Group Robust Linear Regression Analyses

Main Effects Models

The overall model evaluating associations between all inter-network connectivity metrics and social attainment was significant (F_7,52 = 7.56, P < .001) (See Table 4), with main effects of the mPFC-to-PPC metric and group status (both P < .001). None of the other connectivity metrics were associated with social attainment (all P > .10) and, therefore, were not used in subsequent analyses. The exploratory covariate (i.e., RMS incremental movement) was non-significant within the model (P > .10). Thus, we did not include it in the final model to optimize statistical power.

Table 4.

Inter-network connectivity metrics associated with social attainment and social competence (between-group model)

	β (SE)	t-statistic	P-value
Social Attainment Robust Linear Regression Model
Inter-network Connectivity
PPC-to-P metric	−0.07 (0.10)	−0.71	0.48
mPFC-to-PPC metric	0.50 (0.16)	3.19	< 0.001**
OFC-to-PPC metric	−0.23 (0.14)	−1.66	0.10
mPFC-to-P metric	−0.09 (0.14)	−0.60	0.55
OFC-to-P metric	−0.03 (0.12)	−0.29	0.77
mPFC-to-OFC metric	0.10 (0.09)	1.13	0.26
Group	1.11 (0.17)	6.43	< 0.001***
	β (SE)	t-statistic	P-value
Social Competence Robust Linear Regression Modela
Inter-network Connectivity
PPC-to-P metric	−0.01 (0.12)	−0.86	0.40
mPFC-to-PPC metric	0.44 (0.17)	2.53	0.02*
OFC-to-PPC metric	−0.03 (0.15)	−0.17	0.87
mPFC-to-P metric	−0.18 (0.17)	−1.12	0.27
OFC-to-P metric	−0.04 (0.14)	−0.26	0.79
mPFC-to-OFC metric	0.21 (0.11)	1.85	0.07+
Group	1.39 (0.19)	7.16	< 0.001***

Note: PPC-to-P metric, inter-network connectivity between the posterior cingulate cortex-anterior precuneus and the precuneus; mPFC-to-PPC metric, inter-network connectivity between the medial prefrontal cortex and the posterior cingulate cortex-anterior precuneus; OFC-to-PPC metric, inter-network connectivity between the orbital frontal cortex and the posterior cingulate cortex-anterior precuneus; mPFC-to-P metric, inter-network connectivity between the medial prefrontal cortex and the precuneus; OFC-to-P metric, inter-network connectivity between the orbital frontal cortex and the precuneus; mPFC-to-OFC metric, inter-network connectivity between the medial prefrontal cortex and the orbital frontal cortex

^an = 27 CON and n = 23 SCZ

⁺P < .10,

^*P < .05,

^**P < .01,

^***P < .001

The overall model evaluating associations between all inter-network connectivity metrics and social competence was also significant (F_7,42 = 9.32, P < .001) (See Table 4), with main effects of the mPFC-to-PPC metric and group status (both P < .001). Although the OFC-to-P metric had a trend level association with social competence (P = .07), it was not used in subsequent analyses. None of the other connectivity metrics were significantly associated with social attainment (all P > .10) and, were also discarded from further analysis. The exploratory covariate (i.e., RMS incremental movement) was non-significant within the model (P > .10). Thus, we did not include it in the final model to optimize statistical power.

Follow-up Interaction Models

The overall model evaluating associations between connectivity and social attainment was significant (F_7,50 = 11.68, P < .001). There were significant main effects of the mPFC-to-PPC metric (P < .001), group status (P < .001), parental SES (P < .05), and global neurocognition (P < .05) (see Table 5). We observed an interaction between group and DMN connectivity (P < .01). A plot of the group-by-DMN connectivity interaction is presented in Figure 1. The other a priori covariates (i.e., age and gender) were not significantly associated with social attainment (both P > .10). Follow-up within group analysis indicated that DMN connectivity was positively correlated with social attainment in individuals with schizophrenia (partial r = 0.44, P < .05) but was not associated in controls (partial r = −0.08, P > .10). See Supplemental Table 2 for zero-order correlations.

Table 5.

Inter-network connectivity associated with social attainment and social competence (between-group model)

	β (SE)	t-statistic	P-value
Social Attainment Robust Linear Regression Modela
Inter-network Connectivity
mPFC-to-PPC metric	0.46 (0.10)	4.38	< 0.001***
Group	0.92 (0.19)	4.77	< 0.001***
Age	0.08 (0.08)	0.98	0.33
Gender	< 0.01(0.08)	0.05	0.96
Parental SES	−0.20 (0.09)	−2.24	0.03*
Global neurocognition	0.23 (0.11)	2.15	0.04*
Interaction Term
mPFC-to-PCC x Group	−0.52 (0.15)	−3.45	0.001**
	β (SE)	t-statistic	P-value
Social Competence Robust Linear Regression Modelb
Inter-network Connectivity
mPFC-to-PPC metric	0.46 (0.11)	4.16	< 0.001***
Group	1.01 (0.21)	4.87	< 0.001***
Age	0.02 (0.08)	0.26	0.80
Gender	−0.12 (0.09)	−1.34	0.19
Parental SES	0.07 (0.10)	0.71	0.48
Global neurocognition	0.23 (0.11)	2.07	0.05*
Interaction Term
mPFC-to-PCC x Group	−0.50 (0.17)	−2.97	< 0.01**

Note: mPFC-to-PPC metric, inter-network connectivity between the medial prefrontal cortex and the posterior cingulate cortex-anterior precuneus

^an = 31 CON and n = 27 SCZ

^bn = 26 CON and n = 22 SCZ

^*P < .05,

^**P < .01,

^***P < .001

Fig. 1.

Functional connectivity is associated with social attainment in schizophrenia

*p < 0.05

Social Attainment, Scores on the Specific Levels of Functioning Scale with higher scores indicating better social attainment; mPFC-to-PPC Metric, Inter-network connectivity of medial prefrontal cortex and posterior cingulate cortex-anterior precuneus with higher scores indicating more connectivity between networks

The overall model evaluating associations between connectivity and social competence was significant (F_7,42 = 14.07, P < .001). There were significant main effects of the mPFC-to-PPC metric (P < .001), group status (P < .001), and global neurocognition (P < .05) (see Table 4). There was a significant interaction between group and DMN connectivity (P < .01). A plot of the group-by-DMN connectivity interaction is presented in Figure 2. The other a priori covariates (i.e., age, gender, and parental SES) were not significantly associated with social competence (all P > .10). Follow-up within group analysis indicated that DMN connectivity was positively correlated with social competence in individuals with schizophrenia (partial r = 0.45, P < .05) but not correlated in controls (partial r = −0.03, P > .10). See Supplemental Table 2 for zero-order correlations.

Fig. 2.

Functional connectivity is associated with social competence in schizophrenia

*p < 0.05

Social Competence, Scores on the Social Skills Performance Assessment with higher scores indicating better social competence; mPFC-to-PPC Metric, Inter-network connectivity of medial prefrontal cortex and posterior cingulate cortex-anterior precuneus with higher scores indicating more connectivity between networks

Mediation Analyses

Social Attainment

Pearson correlations indicated that social perception and mentalizing were not significantly correlated with social attainment (all P > .10). Since neither of our possible mediators were significantly correlated with both social attainment and DMN connectivity, we did not conduct a mediation analysis.

Social Competence

Pearson correlations indicated that mentalizing was significantly correlated with social competence (r = 0.45, P < .05) in individuals with schizophrenia, but social perception was not significantly correlated with social competence (P > .10). Mentalizing was not significantly correlated with the mPFC-to-PCC metric in individuals with schizophrenia (r = 0.09, P > .10). Since neither of our possible mediators were significantly correlated with both social competence and DMN connectivity, we did not run a mediation analysis.

Within-Group Specificity Analysis

Follow-up analyses indicated specificity for the associations between mPFC-to-PPC inter-network connectivity with social attainment and social competence. In particular, this metric was not correlated with the UPSA-B in individuals with schizophrenia (r = 0.02, P > .10). Additionally, the correlation between social attainment and the mPFC-to-PPC metric was significantly stronger than the correlation between UPSA-B performance and the mPFC-to-PPC metric (Meng’s z = 2.06, P = .04). The correlation between social competence and the mPFC-to-PPC metric was stronger than the correlation between UPSA-B performance and the mPFC-to-PPC metric at trend level (Meng’s z = 1.66, P = .09).

Discussion

In the current study, we were interested in the association between DMN connectivity and social functioning in controls and individuals with schizophrenia and whether this association was mediated by social cognition. Specifically, we evaluated whether connectivity between DMN components differed between individuals with schizophrenia and controls, and whether specific DMN connectivity metrics were associated with two measures of social functioning (i.e., social attainment or social competence). We controlled for global neurocognition, age, gender, and parental SES. Our results suggested that the groups did not differ with respect to average DMN connectivity magnitude. However, connectivity between the mPFC and PPC hubs was differentially related to social attainment and social competence in individuals with schizophrenia and controls as indicated by a significant group-by-connectivity interaction. Follow-up tests revealed that DMN inter-network connectivity was positively associated with measures of social attainment and social competence among individuals with schizophrenia but was not associated with measures of social attainment or social competence in controls. Given non-significant associations between connectivity metrics with social perception or mentalizing, we did not conduct subsequent mediation analyses.

These findings build on our prior research from this dataset that examined associations between social cognition and social functioning (Abram et al., 2014; Karpouzian, Alden, Reilly, & Smith, 2016; Smith et al., 2014; Smith et al., 2012), and associations between neural activation during a mentalizing task and social functioning (Smith et al., 2015). In particular, the current findings indicate that resting state DMN connectivity is associated social functioning in individuals with schizophrenia, and that social cognition does not mediate this particular association. Additionally, these findings add to the emerging literature examining the behavioral correlates of DMN connectivity (Andrews-Hanna, Reidler, Sepulcre, Poulin, & Buckner, 2010; Brunet et al., 2003; Das, Lagopoulos, Coulston, Henderson, & Malhi, 2012; Dodell-Feder et al., 2014; Laird et al., 2009; Mitchell, Neil Macrae, & Banaji, 2005; Spreng & Grady, 2010; Walter et al., 2009; Welborn & Lieberman, 2015). Furthermore, to our knowledge, our study is the first to examine the associations between DMN inter-network connectivity and social attainment and social competence among individuals with schizophrenia.

Connectivity Predictor of Social Attainment and Social Competence

We observed that stronger temporal synchrony between the mPFC and PPC was associated with better social attainment and social competence in individuals with schizophrenia, but this connectivity was not associated with social functioning in controls. Moreover, these findings remained significant after controlling for age, gender, global neurocognition, and parental SES. Therefore, these neural-behavior associations were not simply due to neurocognitive impairment or demographic characteristics, and neural data may provide additional information on social attainment and social competence beyond what is captured with background characteristics and neuropsychological tests (MacDonald, 2013; Sprong, Schothorst, Vos, Hox, & van Engeland, 2007).

Our findings are consistent with a recent study that found stronger DMN subsystem connectivity was related to better social functioning in first-degree relatives of individuals with schizophrenia (Dodell-Feder et al., 2014). However, the null correlations between our measure of DMN connectivity and the two measures of social cognition were not consistent with prior studies showing stronger connectivity between the mPFC and PPC was related to better social cognition (Andrews-Hanna et al., 2010; Mitchell et al., 2005; Spreng & Grady, 2010; Welborn & Lieberman, 2015). Our inability to identify a mechanism underlying the association between DMN connectivity and social functioning is not unexpected given that current understanding of mechanisms underlying the association between neural connectivity and functioning is limited.

One possible explanation for the null correlation findings between DMN connectivity and social cognition is that there may be an indirect association between DMN connectivity and social functioning through social cognitive functions other than mentalizing and social perception. For example, stronger connectivity between the mPFC and PPC may facilitate enhanced self-reflection (Andrews-Hanna et al., 2010), and self-reflection has been linked with social functioning (Brune, Dimaggiob, & Lysaker, 2011). Therefore, self-reflection may mediate the association between DMN connectivity and social functioning in schizophrenia. Cognitive insight is another possible mediator for the association between DMN connectivity and social functioning in schizophrenia. Research suggests that cognitive insight may be positively associated with DMN connectivity in individuals with schizophrenia (Liemburg et al., 2012), and a treatment study demonstrated that improvements in cognitive insight through metacognitive training resulted in increased social functioning in individuals in the early stages of schizophrenia (Ussorio et al., 2016). Alternatively, there may be no underlying mechanism for the association between DMN connectivity and social functioning, because DMN connectivity may be a direct neural signature for social functioning.

Connectivity Alterations between Groups

The lack of a difference in connectivity between the main DMN hubs in individuals with schizophrenia compared to controls is consistent with work by Chang et al. (Chang et al., 2014), but differs from one study that reported connectivity differences between individuals with schizophrenia and controls (Liemburg et al., 2012). There are various explanations for the inconsistent findings. All three of these studies (including the current study) had relatively small sample sizes (individuals with schizophrenia, n < 35), so power issues may contribute to the differences in findings. Additionally, Liemburg and colleagues suggest that cognitive insight may affect DMN connectivity such that patients with poorer insight exhibit decreased DMN connectivity compared to patients with higher insight (Liemburg et al., 2012). However, only one study assessed insight so future studies may benefit from evaluating insight in this context. Based on these results, future research using similar methods (e.g., pICA) with larger sample sizes is needed to verify DMN connectivity patterns in schizophrenia.

Finally, although individuals with schizophrenia and controls did not differ with respect to DMN connectivity, individuals with schizophrenia had distinctly poorer social attainment and social competence than controls. Given that DMN connectivity was only associated with social functioning for individuals with schizophrenia, it appears that similar communication between neural hubs may have differential outcomes across controls and individuals with schizophrenia. Future research could evaluate whether strengthening DMN connectivity beyond the level of connectivity observed in controls yields improvement in social attainment and social competence in individuals with schizophrenia.

Study Implications

Our results suggest that DMN connectivity was associated with social attainment and social competence in schizophrenia. Therefore, DMN connectivity could be assessed as a treatment target for interventions focused on improving social attainment and social competence in schizophrenia. Such interventions are emerging and are much needed given the poor quality of life experienced by individuals with schizophrenia (Bradshaw, 2000; Gibson et al., 2014; Granholm et al., 2005; Ntoutsia, Katsamagkos, & Economou, 2013). Based on our results, interventions that strengthen connectivity between the mPFC and PPC may be especially efficacious. To support the feasibility of testing such a hypothesis, cognitive remediation and certain types of pharmacotherapy have been shown to alter DMN connectivity in individuals with schizophrenia (Bor et al., 2011; Eack, Newhill, & Keshavan, 2016; Penades et al., 2014; Sambataro et al., 2010). Additionally, rTMS and EEG neurofeedback methods have altered DMN connectivity among healthy controls and among individuals with major depressive disorder (Kluetsch et al., 2014; Liston et al., 2014; Ros et al., 2013; van der Werf, Sanz-Arigita, Menning, & van den Heuvel, 2010). Thus, future research should examine whether these and other interventions targeting social attainment and social competence are strengthening DMN connectivity. Additionally, DMN connectivity metrics could be used as neuroimaging markers to monitor the success of treatments that target social functioning. For example, the observation of early changes in DMN resting-state connectivity could predict the effectiveness of cognitive remediation or social skills training targeting schizophrenia (Subramaniam & Vinogradov, 2013).

Limitations

The current study had several limitations. First, cross-sectional studies cannot infer causality. Thus, longitudinal research is needed to determine whether DMN connectivity is a stable neuroimaging marker of social attainment and social competence. Second the sample size was relatively small. Thus, replication and further exploration with a larger sample is needed. Third, our schizophrenia sample consisted of chronically ill individuals. Therefore, future research on the association between DMN connectivity and social attainment and social competence among first episode patients could help address the generalizability of the findings. Lastly, we did not evaluate occupational functioning as an outcome, and future studies could examine the potential association between DMN connectivity and occupational functioning.

Conclusions

In conclusion, we observed that greater connectivity between structures of the DMN was associated with better social attainment and social competence in individuals with schizophrenia, despite the lack of group differences in average DMN connectivity. Additionally, we found no evidence to support social cognition as a mediator for the association between DMN connectivity and social functioning in individuals with schizophrenia. Our findings support the general hypothesis that DMN connectivity could potentially be a novel treatment target and a neuroimaging marker for monitoring treatments aimed to enhance social attainment and social competence in schizophrenia.

General Scientific Summary.

This study suggests that individuals with schizophrenia and healthy controls do not differ in default mode network (DMN) connectivity. However, default mode network connectivity is differentially associated with social functioning in individuals with schizophrenia and healthy controls. Social cognition does not underlie the relationship between DMN connectivity and social functioning in individuals with schizophrenia.