Impaired Visual Object Processing Across an Occipital- Frontal-Hippocampal Brain Network in Schizophrenia: An integrated neuroimaging study

Abstract

Background

Perceptual closure refers to the ability to identify objects with partial information. Deficits in schizophrenia are indexed by impaired generation of the closure-related negativity (N_CL) from ventral stream visual cortex (lateral occipital complex, LOC), as part of a network of brain regions that also includes dorsal stream visual regions, prefrontal cortex (PFC) and hippocampus. This study evaluates network-level interactions during perceptual closure in schizophrenia using parallel ERP, fMRI and neuropsychological assessment.

Methods

ERP were obtained from 24 patients and 20 healthy volunteers in response to fragmented (closeable) and control scrambled (noncloseable) line drawings. fMRI were obtained from 11 patients and 12 controls. Patterns of between group differences for predefined ERP components and fMRI regions of interest were determined using both analysis of variance and structural equation modeling. Global neuropsychological performance was assessed using elements of the WAIS-III, WMS-III and MATRICS batteries.

Results

Patients showed impaired visual P1 generation, reflecting dorsal stream dysfunction, along with impaired generation of N_CL components over PFC and LOC. In fMRI, patients showed impaired activation of dorsal and ventral visual regions, PFC and hippocampus. Impaired activation of dorsal stream visual regions contributed significantly to impaired PFC activation. Impaired PFC activation contributed significantly to impaired activation of hippocampus and LOC. Impaired LOC and hippocampal activation contributed significantly to deficits on WAIS-III Perceptual Organization Index (POI) and other tests of impaired perceptual processing in schizophrenia.

Conclusion

Schizophrenia is associated with severe activation deficits across a distributed network of sensory and higher order cognitive regions. Deficit in early visual processing within the dorsal visual stream contributes significantly to impaired frontal activation which, in turn, leads to dysregulation of hippocampus and ventral visual stream. Dysfunction within this network underlies impairment in more traditional measures of neurocognitive dysfunction such as POI, supporting distributed models of brain dysfunction in schizophrenia.

INTRODUCTION

Cognitive dysfunction is a critical component of schizophrenia and a major predictor of impaired long-term outcome ¹. Although traditional models of schizophrenia focused on dysfunction within a limited number of cognitive domains, more recent batteries incorporate a wider range of domains, including perceptual-level testing ².

Neurocognitive domains of interest ^{3, 4} have been expanded still further in cognitive neuroscience-based initiatives, such as the ongoing CNTRICS (Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia) initiative, which incorporates processes such as sensory gain and integration alongside more neuropsychologically driven domains ⁵. Nevertheless, the relationship between lower and higher levels of dysfunction, as well as the neural substrates underlying these domains, remains an area of active research.

The present study investigates mechanisms underlying cognitive dysfunction in schizophrenia using the perceptual closure task ⁶, which requires integration between lower and higher level brain regions. Deficits in perceptual closure have previously been demonstrated in schizophrenia ^7-9, along with deficits in earlier stages of perceptual processing ^10-12. The present study utilizes event-related potentials (ERP) and functional magnetic resonance imaging (fMRI), along with behavior¹³, to investigate mechanisms underlying perceptual closure impairments in schizophrenia.

Perceptual closure refers to the ability of the brain to recognize an object even when presented with only fragmentary information. Final common processes underlying closure involve activation of neuronal structures located within ventral stream visual areas such as lateral occipital complex (LOC) ^{14, 15}. Nevertheless, closure processes depend upon both bottom-up and top-down processes with convergence of these functions even at early sensory and perceptual processing stages ^{16, 17,18, 19,20}. The early visual system is divided into separable magno- and parvocellular pathways, which project respectively to the dorsal and ventral visual streams. One mode of visual form integration ^{16, 21} posits that the magnocellular visual system is specialized for rapid conduction of low resolution “framing” information via the dorsal stream to prefrontal regions, with feedback projection to ventral regions, such as LOC. The parvocellular system, in contrast, provides slower, higher resolution visual information directly to the LOC ²². Because of the more rapid transit of information through the magnocellular/dorsal stream pathway than through the parvocellular/ventral stream, information from the two pathways converges in LOC within the temporal window of object identification. Most recently, we have demonstrated significant synchrony between hippocampus and LOC underlying the closure process ²³. Given the widespread network involved in perceptual closure, this process thus provides an ideal model system in which to investigate potential impairment of distributed large-scale cortical networks^24-28 in schizophrenia.

This study uses parallel ERP and fMRI to investigate stages of impaired perceptual processing in schizophrenia. Prior ERP studies in schizophrenia have demonstrated impaired magnocellular/dorsal stream activation, manifested as attenuation of the visual P1 component to magnocellularly biased stimuli ^{10, 29,30}, as well as impaired perceptual closure as manifest by impaired generation of “closure negativity” (N_CL) ^{8, 9}. As opposed to P1, which occurs with a latency of approximately 100 ms over dorsal visual regions and is similar in amplitude to closeable and non-closeable stimuli, N_CL occurs with a latency of 230-400 ms over LOC and is discontinuously large to closeable vs. non-closeable images, suggesting that it represents the outcome of the perceptual closure process ¹⁴.

In normal volunteers, networks underlying perceptual closure have been further clarified using fMRI and intracranial recording studies. fMRI studies, for example, have confirmed the localization of N_CL generators to LOC and have, in addition, demonstrated activation of dorsal stream and prefrontal regions during the closure process ²⁰, consistent with prior fMRI literature ^{31, 32}. Intracranial studies demonstrated additional involvement of hippocampal formation ²³. Because of generator geometry, activity in lateral inferior PFC and hippocampus project poorly to the scalp and thus can only be assessed using fMRI or intracranial recordings. This, is the first study to utilize combined ERP/fMRI to investigate network function underlying impaired perceptual closure processing, in order to elucidate the contribution of sensory as well as higher cognitive processes, in a convergent model of object recognition in schizophrenia.

Perceptual closure is a cognitive neuroscience task developed to address particular activation of neural substrates of object recognition. However concepts underlying closure overlap with those incorporated into more traditional neuropsychological measures. In particular, the Wechsler Adult Intelligence Scale (WAIS-III) includes the Picture Arrangement (PA) test, as well as subtests comprising the Perceptual Organization Index (POI), which is included in the calculation of performance IQ ³³. In the present study, PA and POI were collected, along with other subtests from WAIS-III, the Brief Visual Memory Task from the MATRICS² (Measurement and Treatment Research to Improve Cognition in Schizophrenia) and the Wechsler Memory Scale-III (WMS-III) ³⁴, in order to characterize the dysregulation observed in various nodes of the neural network engaged in closure processing within the larger context of clinical neuropsychological dysfunctions in schizophrenia.

METHODS

Participants

Data were collected in two separate experiments. In experiment 1 (ERP), 24 male patients meeting DSM-IV criteria for schizophrenia (n = 22) or schizoaffective (n = 2) disorder and 20 healthy volunteers (17 male) of similar age participated. In experiment 2 (fMRI), 11 (10 male) patients meeting criteria for schizophrenia (n = 9) or schizoaffective (n = 2) disorder and 12 healthy volunteers (11 male) of similar age participated. Eight male patients and 9 controls (8 male) participated in both experiments, so that the total sample included 27 patients (26 male, 1 female) and 23 controls (20 male, 3 female). Experiment one consisted of a single ERP session and experiment two was part of a larger fMRI study.

Patients were recruited from inpatient and outpatient facilities associated with the Nathan Kline Institute for Psychiatric Research. Diagnoses were obtained using the Structured Clinical Interview for DSM-IV (SCID) ³⁵. Healthy volunteers with a history of SCID-defined Axis I psychiatric disorder were excluded. Subjects were excluded if they had any neurological or ophthalmologic disorders that might affect performance or met criteria for alcohol or substance dependence within the last six months or abuse within the last month. Informed consent was obtained after full explanation of procedures.

Patient and control groups did not differ significantly in age (patients, 38.9 ± 10.5 years; controls, 33.3 ± 10.0 years). Brief Psychiatric Rating Scale (BPRS) ³⁶ total score was 37.4 ± 9.4 (n=26) and Scale for the Assessment of Negative Symptoms (SANS) ³⁷ total score (including global scores) was 34.0 ± 16.0 (n=26). All patients were receiving antipsychotics with 20 patients receiving atypical antipsychotics, 3 patients receiving typical antipsychotics, and 4 patients receiving a combination of atypical and typical antipsychotics. Chlorpromazine equivalents were 1225 ± 589.5 mg/day. Duration of illness was 18.1 ± 9.9 years.

Stimuli and task

Methods were as described previously for ERP and fMRI studies ^{20, 23}. Briefly, fragmented line drawings of natural and man-made objects were drawn from the normed Snodgrass and Vanderwart picture set^{38, 39}. From these images, segments containing black pixels were randomly and cumulatively deleted to produce seven incrementally fragmented versions of each picture⁴⁰. From these images only pictures belonging to the third level of fragmentation were utilized, hereafter simply referred to as “nonscrambled pictures.” Previous studies have shown that at this level of fragmentation healthy participants correctly identify objects 73% of the time ⁴¹. A set of “scrambled pictures”, serving as control stimuli, was generated by dividing the images into 16 × 16 segments, which were then scrambled (Figure 1).

Figure 1.

Examples of the stimuli: Closeable (nonscrambled) line-drawings are shown in the left panel and noncloseable (scrambled) line-drawings are shown in the right panel. The figures from top to bottom are: Anchor, Ball, and Banana.

Stimulus presentation

For ERP, stimuli were presented on an Iiyama Vision Master Pro 502 monitor located 143 cm from the subject. Images subtended an average of 4.8° (±1.4°) of visual angle in the vertical plane and 4.4° (±1.2°) in the horizontal plane. For fMRI, stimuli were delivered through MR-compatible liquid crystal display goggles (Resonance Technology Inc., Northridge, CA).

Timing of stimulus presentation for the ERP study

For ERP each image appeared for 750 ms, followed by a blank screen for 800 ms. Then a “Y|N” response prompt appeared for 200 ms, followed by a blank screen for 2200 ms. Subjects were instructed to press one button when they recognized the image as an object and another when they did not, following presentation of the “Y|N” prompt. Subjects’ response window extended for 2300 ms from the onset of the “Y|N” prompt. A total of 400 unique images, 200 nonscrambled and 200 scrambled, were presented in eight runs of 3.25 min each. No images were repeated.

Timing of stimulus presentation for the fMRI study

For fMRI, each image appeared for 500 ms, followed by a blank screen for 500 ms resulting in a stimulus onset asynchrony of 1 sec. Nonscrambled, scrambled, and complete (not used in the analyses of this study) stimulus conditions were presented in a block design. The stimuli were presented in four 6.2-min runs of twelve blocks each. Each run consisted of twelve blocks i.e. 4 scrambled, 4 nonscrambled and 4 complete stimulus blocks, pseudo randomly distributed across the run such that a complete version of a stimulus was not presented before the presentation of scrambled and incomplete versions. Each block consisted of 24 unique images, and none was ever repeated. In order to insure central fixation, subjects were asked to respond with their right index finger each time a target image (a dog) was presented.,. Targets were presented with equal probability of 4.2% occurrence across blocks. Only standard stimuli were included in the analysis.

EEG data acquisition and analysis

Continuous EEG was acquired using a BioSemi ActiveTwo electrode system with 168 scalp electrodes, digitized at 512 Hz, re-referenced to the nose. Data were analyzed using BESA version 5.1 (Brain Electric Source Analysis, MEGIS Software GmbH). Electrode channels were subjected to an artifact criterion of ±120μV from −100 to 500 milliseconds. The vertical and horizontal electro-oculograms (HEOG and VEOG) were, in addition, visually inspected for blinks and large eye movements. For each subject, epochs were calculated for a time window from −100 to 500 ms post-stimulus and baseline-corrected relative to the prestimulus baseline. Accepted trials were then averaged separately for both nonscrambled and scrambled stimulus conditions to compute the VEP. Only identified stimuli were included in averages for nonscrambled stimuli.

A priori analysis ^{9, 20} tested between-group differences in amplitude of the ERP components P1, N1, N_CL, and frontal closure related activity (N^f_CL) within predetermined spatial and temporal windows (see legends to Figures 2,3).

Figure 2.

ERP responses to the nonscrambled and scrambled stimuli over posterior scalp, and scalp voltage maps of the response to the nonscrambled stimuli. Voltage maps at 100ms show the observed positivity (P₁) in controls (top) and in patients (bottom). The graphs show scalp recordings from two representative occipital electrodes (P05/P06) in controls (top) and in patients (bottom) indicating no significant differences between the conditions in either group. The bar-charts show significant differences between the groups in the amplitude of responses to the nonscrambled stimuli.

Figure 3.

Voltage maps at 330ms (peak N_CL activity) illustrate the relative negativity over lateral occipital scalp for nonscrambled versus scrambled stimuli. The graphs show scalp recording from two representative lateral occipital electrodes (PO7/PO8) and two representative lateral frontal electrodes (FC5/FC6) in controls (left) and in patients (right). The blue ribbon in the graphs show the tested window of time when the responses to the nonscrambled stimulus condition (in red) produced significantly larger negativity when compared to the scrambled pictures (in green). The bar-charts show significant differences between the groups in the amplitude of responses to the nonscrambled versus scrambled stimuli at the lateral occipital (N_CL) but not at the frontal scalp (N^f_CL). The bar-charts also show no significant differences between the groups in the amplitude of responses for N₁.

fMRI data acquisition and analysis

Images were acquired at the Center for Advanced Brain Imaging (CABI) at the Nathan Kline Institute on a 3 Tesla MRRS (formerly SMIS, Guildford, U.K.) MRI system. This system uses a 38 cm inner diameter (i.d.) gradient coil with gradient strength 40 mT/m, rise time of 280 us, and a 30 cm i.d. transmission line radio frequency coil (Morris Instruments). In each run, T2*-weighed echoplanar functional images (EPIs) (TR=2000, FA=90, matrix size= 64×64, FOV=224 mm, voxel size = 3.5mm³) emphasizing the blood oxygenation level dependent (BOLD) response were acquired while the participant attended to the visual stimuli. High resolution (1mm³) T1-weighed anatomical images of the whole brain were acquired from each subject using a standard three dimensional magnetization prepared rapid gradient echo (MPRAGE) pulse sequence in order to allow volume statistical analyses of signal changes. Head movement was minimized with the use of a custom-made head holder. In all subjects, motion never exceeded 0.75 mm along any axis.

The BrainVoyager QX software package ⁴² was used to process the fMRI data. Each subject’s data were analyzed separately. Preprocessing of functional scans included slice scan time correction, head movement measurements, removal of linear trends and temporal high pass filtering. Each subject’s functional data were co-registered with the anatomical data. The functional data were then transformed into Talairach space ⁴³ for the multisubject analysis. Group statistical maps were obtained using a random-effect analysis of the BOLD signal time series. To estimate the BOLD response associated with each condition, regressors representing the timing of each of the stimulation epochs were convolved with a canonical function adjusted with a double-gamma hemodynamic response delay ⁴⁴ and used in a multiple regression analysis. The resulting statistical maps were then grouped to obtain activation maps (Figures 4, 5). The P values of the statistical maps were corrected for multiple comparisons^{45, 46} and z-normalized. Different conditions were then contrasted using a general linear model ^{47, 48}.

Figure 4.

fMRI group activation maps for controls (left) and patients (right) in response to the nonscrambled (top panel) and scrambled (bottom panel) conditions. The functional data are presented on the Talairach normalized inflated brain of a single subject. All data are p ≤ 0.05, corrected for multiple comparisons across the entire image volume. Areas of significant BOLD signal in the visual stream and prefrontal cortical regions are observed.

Figure 5.

fMRI responses for the nonscrambled versus scrambled comparison in controls (left), patients (middle) and the difference between the two groups (right). The fMRI activation maps showing the difference between the two groups illustrate areas of significant BOLD signal increase in the four regions of interest (ROI) namely, dorsal visual stream, ventral visual stream, lateral frontal cortex, and hippocampal formation in controls compared to patients. The bar-chart shows the magnitude of the BOLD signal increase in response to the nonscrambled versus scrambled stimuli in controls and in patients at each ROI. Significant differences between controls and patients are observed at the four ROIs bilaterally. Error bars represent SEM.

Based on a priori hypothesis, four regions of interest (ROIs) were selected at dorsal visual stream, LOC, PFC and hippocampal formation and the parameter estimates (beta values) derived from the ROIs.

Clinical Variables

In addition to ERP and fMRI, several relevant neuropsychological measures were administered. These included the WAIS-III Picture Arrangement (PA) test ³³; the Perceptual Organization Index (POI) battery, which consists of Picture completion (PC), Matrix Reasoning (MR) and Block Design (BD) tests and the Processing Speed Index (PSI) from the WAIS-III; the Working Memory Index (WMI) from the WMS-III ³⁴; and the Brief Visuospatial Memory Test (BVMT-R) ⁴⁹ which assesses retention of visual memory over time.

Statistical analyses

Between-group analyses for ERP and fMRI measures were performed using separate repeated-measures multivariate analysis of variance (rmMANOVA) for each identified ERP component (P1, N1, N_CL, N^f_CL) and for each of the four ROIs in fMRI studies (dorsal stream, LOC, PFC, hippocampus). Analyses were conducted using SPSS software (SPSS Inc, Chicago, Il) with a between-subjects factor of group (patients and controls) and within-subjects factors of condition (nonscrambled and scrambled) and hemisphere. All tests were 2-tailed with a preset α level of P<.05.

Interrelationship among measures was determined by linear regression, supplemented with structural equation modeling (path analysis). Path analysis was implemented using AMOS 7.0 (SPSS Inc, Chicago, Il) ⁵⁰. Selection among alternative models was determined by minimizing χ2 variance, with paths entered according to the criterion χ2 to include = (χ2_without − χ2_with), (df_without−df_with). Residual error and goodness of fit measures including Cmin/df, RMSEA and NFI were used to assess model integrity.

All significance levels are two-tailed with preset alpha level for significance of p<.05. Values in text represent mean ± std. dev.

RESULTS

Behavioral Results

Patients recognized the nonscrambled objects 50.6 % of the time vs. 67.5% for controls (t = −3.7, df = 42, p < .002). The groups did not differ in their reaction times (patients, 1600 ms ± 500 ms; controls, 1900 ms ± 400 ms; t = −1.6, df = 42, P = 0.1). Control values are similar to those obtained previously ²⁰.

Electrophysiological Results

ERP measures were assessed at predefined electrodes within predefined intervals based upon prior studies with these stimuli in normal volunteers ²⁰. Separate analyses were conducted for the sensory components P1 and N1, and for closure-related components over LOC (N_CL) and PFC (N^f_CL).

Sensory potentials

P1 was maximal over dorsal stream electrodes (P05/P06) within the 90-110 ms interval (Figure 2). P1 amplitudes were significantly reduced in patients relative to controls (F_1,42 = 7.0, P < 0.01). The group × hemisphere (F_1,42 = 4.3; P < 0.05) interaction was also significant, reflecting differential asymmetry between groups. Group × condition (F_1,42 = 1.3; P = 0.3) and group × condition × hemisphere (F_1,42 = 0.02; P = 0.9) interactions were not significant, reflecting similar P1 amplitude to nonscrambled and scrambled stimuli.

N1 was maximal over ventral stream electrodes (P07/PO8) during the 160-180 ms latency range. N1 amplitude (Figure 3) was not significantly different between groups (F_1,42 = 1.8, P = .19). Group × hemisphere (F_1,42 = 0.96; P = 0.33), group × condition (F_1,42 = 0.05; P = 0.82), and group × condition × hemisphere (F_1,42 = 0.62; P = 0.44) interactions were also non-significant.

Closure related potentials

Closure related activity was observed primarily over LOC, with a smaller contribution over frontal regions (Figures 3). As in prior studies, the N_CL was defined as mean amplitude within a 300- to 330-millisecond-latency range over left and right ventral stream electrodes (PO7/PO8). The expected significant main effect of condition was observed (F_1,42 = 22.4; P < 0.001), indicating differential activity across groups to nonscrambled versus scrambled stimuli. A significant group × condition was also found (F_1,42 = 4.8; P < 0.04), indicating reduced differential activity to nonscrambled vs. scrambled in patients (F_1,23 = 3.9, P = 0.062) vs. controls (F_1,19 = 20.0, P<.001). No significant main or interaction effects involving hemisphere were observed. Group × hemisphere (F_1,42 = 0.15; P = 0.7), condition × hemisphere (F_1,42 = 0.94; P = 0.3), and group × condition × hemisphere (F_1,42 = 1.18; P = 0.3) interactions involving LOC were not significant.

Bilateral differential activity was also observed over lateral-frontal scalp regions (FC5/FC6) within the N_CL time frame (F_1,42 = 9.08; P < 0.004). The group × condition interaction was not significant (F_1,42 = 0.345; P < 0.56). Nevertheless, when analyses were conducted by group a significant effect of condition was observed in the control group (F_1,19 = 7.02, P = 0.016) but not in the patient group (F_1,23 = 2.9, P = 0.1) (Figure 3).

fMRI Results

Both experimental conditions (nonscrambled and scrambled), activated widespread and substantially overlapping cortical networks (Figure 4). BOLD activation patterns were assessed within predefined ROIs based upon prior studies with these stimuli in normal volunteers ²⁰. Four brain regions - dorsal visual stream, ventral visual stream, lateral PFC, and hippocampal formation - were interrogated by a MANOVA which revealed significant main effects of region (F_3,19 = 7.9; P = 0.001) and hemisphere (F_1,21 = 11.8; P = 0.003). No significant main effect of group was observed (F_1,21 = 2.23; P = 0.15) indicating absence of significant differences in the general activation of these brain regions across the two groups. However, significant effects for group × condition (F_1,21 = 37.4; P < 0.001), group × hemisphere (F_1,21 = 5; P = 0.03) and region × hemisphere (F_3,19 = 4.5; P = 0.016) were observed. No other significant interactions were found. Following a main effect of region, a set of preplanned ANOVAs were carried out to unpack the results observed at each region. Significant group × condition interactions were observed at all four regions bilaterally, indicating reduced closure-related activity in patients relative to controls (Table 1, Figure 5).

Table 1.

Talairach coordinates for four predefined regions of interest (ROIs) and results of individual ANOVAs showing main effect of group (control/patient) and group × condition (scrambled/nonscrambled) interaction.

ROI	Talairach: X, Y, Z	Group F(1, 21); p	Group × Condition F(1, 21); p
Dorsal stream	±26, −79, 23	0.007; 0.93	27.04*; 0.0001
Ventral stream (LOC)	±27, −59, −8	1.782; 0.2	14.97*; 0.001
Frontal cortex	±33, 4, 40	5.5*; 0.03	14.26*; 0.001
Hippocampus	±24, −22, −23	0.442; 0.51	18.53*; 0.0001

^*p< 0.05

Correlation within ERP and fMRI measures

In order to determine contributions of early stage measures to subsequent closure-related activity, separate sets of analyses were conducted for ERP and fMRI data. For ERP data, correlational analyses using the differential activity to nonscrambled vs. scrambled stimuli were employed. However, because the hippocampal node was not represented in scalp-related activity, formal path analysis was not employed. For fMRI data, a path analysis explored the interrelationship between ROIs, as well as the effect of cohort on pattern of interrelationships. Because only 8 subjects participated in both ERP and fMRI studies, correlation across modalities was not possible.

ERP

For both patients and controls, P1 amplitude correlated significantly with amplitude of N^f_CL (patients: r=.64, p=.001; controls: r=.55, p=.012). Also, in both groups, strength of N^f_CL correlated significantly with amplitude of N_CL (patients: r=.75, p<.001; controls: r=.76, p<.001). For patients (r=.64, p=.001), but not controls (r=.28, p=.24), deficits in P1 generation correlated significantly with deficits in N_CL generation over LOC.

fMRI

Interrelationship among regions was was assessed using path analysis. An iterative model was employed with paths added to the model only to the extent that they statistically reduced free variance. Significant paths were observed from dorsal visual stream to PFC, PFC to LOC, and PFC to hippocampal formation. In addition, a bidirectional interaction was observed between hippocampal formation and ventral visual stream (Figure 6). When group was entered into the model, significant independent effects of diagnosis were observed on dorsal stream and frontal nodes, but not with hippocampal formation or ventral visual stream, suggesting significant effects of pathological processing primarily on processing within dorsal stream and frontal regions.

Figure 6.

Differential activation indicated brain regions to closeable (nonscrambled) vs. noncloseable (scrambled) images. Standard regression coefficients between regions are determined by path analysis, based upon iterative path fitting. fMRI activations at dorsal stream and frontal regions significantly predicted group affiliation. Goodness of fit measures vs. suggested rule of thumb (Ro) values ⁵⁰ for the model are as follows: CMIN=1.036 (Ro<2.0), NFI .949 (Ro>.9); RMSEA .027 (Ro<.05); Hoelter .05=113

Correlation between ERP/fMRI and clinical measures

Mean Performance IQ in schizophrenia patients was 86.7 ± 17.3, which is significantly lower than normative mean of 100 (t=3.68, df=22, p=.001). Patients showed significant reductions in Picture Arrangement (PA) (t=2.25, p=.03), POI (t=2.62, p=.015) and PSI (PSI, t=10.8, p<.001), and Working Memory Index (WMI) (t=3.46, p=.002) relative to published norms, with greater deficit on PSI relative to POI (t=3.31, p=.003) (Table 2).

Table 2.

Mean ±sd scores on indicated neuropsychological tests and their probability levels (p) relative to normative means for schizophrenia patients (n=23), along with correlation coefficients (r) and probability levels (p) for relationship with fMRI activation measures in indicated ROIs (n=11).

Measure	Mean score (M ±sd; p)	Normative mean	Ventral stream (LOC) (r; p)	Hippocampus (r; p)
Picture arrangement (PA)	8.4* ±3.7; 0.03	10	0.73*; 0.01	0.65*; 0.03
Perceptual Organization Index (POI)	89.1* ±16.9; 0.003	100	0.67*; 0.02	0.6; 0.052
- Picture completion (PC)	7.1* ±3.1; 0.001	10	0.5; 0.12	0.75*; 0.008
-Matrix Reasoning (MR)	9.2 ±3.4; 0.22	10	0.74*; 0.009	0.62*; 0.04
-Block Design (BD)	8.3* ±3.2; 0.01	10	0.41; 0.21	0.39; 0.23
Processing Speed Index (PSI)	80.5* ±8.6; 0.001	100	0.3; 0.4	0.3; 0.4
Working Memory Index (WMI)	88.2* ±14.7; 0.001	100	−0.16; 0.65	0.25; 0.46
Brief Visual Memory Test (BVMT-R)	16.3* ±8.6; 0.001	25	0.35; 0.28	0.67*; 0.02

^*p< 0.05

Deficits in fMRI activation of LOC correlated significantly with PA scores, as well as the overall POI and MR subtest. In contrast, no significant correlation was observed between LOC activity and the PSI, WMI or the BVMT-R (Table 2). Significant correlations were also observed between hippocampal activation and several indices of visual processing, as well as with the BVMT-R. No significant correlations were observed between frontal or dorsal ROI activations and indices of perceptual closure (all p>.2). Similar correlations were observed with differential N1 activation to nonscrambled vs. scrambled stimuli vs. both PA (r=−.48, n=24, p=.017) and POI (r=−.41, p=.048), but not with PSI (r=−.11, p=.61), WMI (r=−.24, p=.29) or BVMT-R (r=−.24, p=.29).

Reduced N_CL generation correlated significantly with higher scores on PANSS total (r=.46, n=24, p=.025), positive (r=.44, p=.03) and general (r=.45, p=.03) symptoms.

DISCUSSION

We have previously demonstrated reduced perceptual closure ability in schizophrenia using both behavioral ⁹ and ERP ⁹ measures, along with impaired generation of the dorsal stream P1 potential ¹¹, suggesting significant contribution of early visual impairments⁵¹ to more complex forms of visual processing. Since that time, additional evidence has accumulated regarding the functional anatomy of the perceptual closure process using scalp ERP, fMRI and direct intracranial recordings in humans, permitting more detailed hypothesis-driven analysis of underlying physiological disturbances in schizophrenia.

This study confirms and extends previous findings in four ways. First, it replicates the prior reports of impaired P1 and N_CL generation using a more efficient paradigm recently employed by us in healthy individuals ^{20, 23}. Second, it combines ERP findings with results of parallel fMRI investigation, permitting an improved characterization of the deficit in schizophrenia, while third, providing a direct between-group comparison of fMRI activation patterns in patients and controls. Finally, neuropsychological data were collected to enable the characterization of the functional neuroanatomy of closure processes more fully within the context of neuropsychological dysfunction in schizophrenia.

Here, as in previous studies, the earliest deficits were found in generation of the dorsal stream P1 potential, suggesting failure of magnocellular and/or early stage cortical processing, followed by impaired generation of N_CL^{10, 29, 52}. In addition here specific measures were obtained for other brain regions that have been found to be involved in the closure process, including PFC and hippocampus ²³. Deficits in dorsal activation significantly predicted deficits in frontal processing as assessed using both ERP and fMRI measures. In both the ERP and fMRI experiments, the degree of correlation between dorsal and prefrontal activation was similar in patients and in controls suggesting relatively normal functional connectivity between these two regions in patients. However, the absolute level of activation was dramatically reduced over both brain regions in patients, suggesting that failure in processing at the level of dorsal stream visual cortex contributes directly to failures in processing at subsequent nodes in the closure network.

The study also permits the first assessment of potential hippocampal involvement in perceptual closure deficits in schizophrenia. Although hippocampus does not generate scalp-recordable activity because of its location and orientation within brain, we have recently demonstrated the involvement of hippocampal formation in the process of closure using recordings from intracranial electrodes in epilepsy patients ²³. In those recordings, a significant, sustained β-synchrony interaction was observed between the hippocampal formation and LOC during the N_CL timeframe, suggesting that closure may depend upon a matching process between sensory inputs and mnemonic representations activated within hippocampus. In support, Bar and coworkers ^{32, 53} have postulated that the magnocellular provides rapid low-resolution input to the frontal cortex, which then helps trigger top-down object recognition. The failure of frontal and hippocampal activation observed here suggests that this top-down mechanism is lost in schizophrenia, due primarily to loss of ascending magnocellular/dorsal stream input. However, intrinsic abnormalities in hippocampal function and structure have also been reported ⁵⁴, leaving unresolved the degree to which intrinsic hippocampal pathology may contribute to perceptual closure ability as well.

As in our earlier studies, prominent closure-related activity was observed to nonscrambled vs. scrambled objects in LOC, consistent with the concept that LOC represents the locus of conscious object identification ^{31, 32}. However, several lines of evidence suggest that activation deficits were not due to intrinsic structural dysfunction within LOC. First, generation of the N1 potential, which reflects initial activation of LOC primarily via the parvocellular stream, was unimpaired. Second, we have previously demonstrated intact N1 modulation in schizophrenia by illusory contour stimuli ¹⁹. As opposed to perceptual closure, which depends upon recursive processing between brain regions and thus occurs significantly after the visual N1, illusory contour-related activity overlaps the N1, suggesting that it can be processed during the feedforward sweep ²². Behavioral modulations of perceptual closure, such as repetition and word priming are also relatively intact ⁸, suggesting intact intrinsic function of LOC and modulation by top-down influences. Third, in our path analysis, illness-related effects were observed only at the level of dorsal stream visual cortex, and, to a lesser extent, PFC, suggesting that impaired activation of LOC in this task was driven primarily by impaired input from preceding stages of cortical analysis. Other studies have also demonstrated intact ventral stream function to stimuli such as faces using behavioral ⁵⁵ and fMRI ⁵⁶ measures.

Finally, this study assesses closure-related activity relative to traditional neuropsychological measures. Performance IQ in the WAIS is made up of two indices: POI and PSI. Reduced Performance IQ is a hallmark of schizophrenia. In the present sample, significant reductions in both POI and PSI were observed relative to normative values, although reductions in PSI were significantly more robust (p=.003) consistent with prior literature ⁵⁷. Nevertheless, here, impaired activation of LOC correlated specifically with overall POI impairment, as well as reduced PA scores, but not with impairments in PSI, WMI or BVMT (Table 2). Thus as expected, LOC impairment was most related to cognitive difficulties in perceptual organization.

A significant correlation was also observed between hippocampal activation and PA scores, and a nearly significant correlation with POI was also observed (Table 2), suggesting that the interaction between hippocampus and LOC may also be critical for successful completion of these tasks. In terms of subtests, a somewhat different pattern of correlations was observed between regions, with LOC activation correlating only with MR, and hippocampal activation correlating with both PC and MR. The differential pattern of correlation may reflect the different mnemonic requirements of the PC vs. MR subtests, as hippocampal activation correlated also with performance on the BVMT-R, whereas LOC did not. As with LOC, hippocampal activity in this task did not correlate significantly with performance on either the PSI or WMI, suggesting relative independence of perceptual closure mechanisms from other aspects of cognitive dysfunction.

In summary, the present study represents the first multimodal imaging study of impaired perceptual closure ability in schizophrenia, and highlights the importance of distributed network dysfunction in the pathophysiology of cognitive dysfunction. In the case of perceptual closure the critical network of regions begins in dorsal stream visual cortex, and encompasses PFC, hippocampus and ventral stream visual regions as well. Deficits in early stages of processing contribute significantly to impairments in subsequent frontal, hippocampal and ventral stream processing. Dysfunction within this network contributes as well to overall performance IQ impairments in schizophrenia, with particular relevance to functions relating to processing of complex visual scenes. Overall, sensory input dysfunction must be considered a strong contributor to neurocognitive dysfunction in schizophrenia.