NC and SZ subjects had similar median reaction time (RT) and error rate patterns (). In general, subjects made more errors (F(1,22) = 4.99; p < 0.05) and had shorter RTs (F(1,22) = 9.92; p < 0.01) on Square than No-Square trials. This shorter RT for Squares also was present within the SZ group (F(1,11) = 7.61; p < 0.05) but not within the NC group (F(1,11) = 2.32; NS). Whereas the SZ patients’ low overall error rate (3.9%) indicated that they did not have any particular difficulties with the task, they did make more errors (F(1,22) = 5.97; p < 0.05) and had longer RTs (F(1,22) = 9.06; p < 0.01) than the NC group.
The most notable group difference in the VEPs () was that the P1 component was larger in amplitude for NC than SZ subjects across stimulus conditions (F(1,22) = 6.22; p < 0.05). There was a trend for N1 component amplitude to be larger in the NC than the SZ group (F(1,22) = 3.11; p = 0.09), and across all subjects, the N1 was significantly larger for Square than No-Square stimuli (F(1,22) = 8.66; p < 0.01).
For both the NC and SZ groups, visual stimuli evoked an early phase-locking response in the 24–48 Hz range that began as early as 80 msec and did not overlap with the low-frequency VEP at frontal, central, and occipital sites (). Previous studies have differentiated this early evoked gamma band response into anterior and posterior components (Tallon-Baudry et al., 1997
), and in this study, different effects on the response were indeed found at occipital and fronto-central sites.
Figure 2 Group average time-frequency maps of phase locking. A, Midline frontal (Fz), central (Cz), and occipital (Oz) sites. The early evoked gamma band response (circled in yellow) can be distinguished from the low-frequency VEP. B, Square minus No-Square difference (more ...)
The most striking difference between the SZ and NC groups was in the occipital component of the response (group X stimulus; F(1,22) = 7.29; p < 0.05). In the NC group data, Square stimuli evoked this response maximally at the left occipital site, but there was no response to the No-Square stimuli (F(1,11) = 10.2; p < 0.01). For the SZ subjects, however, neither stimulus evoked the response (F(1,11) = 0.515; NS). These group differences are highlighted in . To try to determine the functional significance of the occipital phase-locking response, we computed nonparametric Spearman’s ρ correlation coefficients between the Square minus No-Square differences in phase synchrony and RT across subjects in the NC group. This correlation was negative and statistically significant (ρ = −0.677; p < 0.05; two-tailed), indicating that the larger the response, the larger the RT advantage for Square compared with No-Square stimuli (). No such correlation was found for the SZ group. These data suggest that the occipital phase-locking response may reflect a feature-binding mechanism in visual cortex that underlies more efficient task performance for healthy individuals, but not persons with schizophrenia.
The fronto-central component of the early evoked gamma band response was abnormal in topography, latency, and frequency for the SZ group. Whereas for both groups this response was larger for No-Square than Square stimuli (NC: F(1,11) = 5.70; p < 0.05; SZ: F(1,11) = 5.19; p < 0.05), it was larger at central than frontal electrode sites for SZ (group X site; F(1,22) = 10.0; p < 0.01; SZ: F(1,11) = 24.5; p < 0.001) but not NC subjects (F(1,11) = 0.00695; NS). The latency of the fronto-central response to Square stimuli was longer for SZ than NC subjects (116.9 vs 99.2 msec; group X stimulus; F(1,22) = 5.66; p < 0.05; group: F(1,22) = 24.1; p < 0.0001), but there was no difference for No-Square stimuli (117.3 vs 119.1 msec; F(1,22) = 0.046; NS). Last, the frequency of the response to No-Square stimuli was lower for SZ than NC subjects at central sites (35.3 vs 42.4 Hz; group X stimulus X site; F(1,22) = 5.08; p < 0.05; group: F(1,22) = 4.45; p < 0.05), but there was no difference for Square stimuli (44.4 vs 41.5 Hz; F(1,22) = 1.04; NS).
In summary, the early evoked gamma band response differed between the SZ and NC groups in several ways, being absent for the occipital component and abnormal in topography, latency, and frequency for the fronto-central component. These results support the hypothesis that gamma band synchronization is abnormal in schizophrenia. The different patterns observed for the VEPs and phase-locking responses indicate that the latter are not artifacts of the low-frequency VEPs.
To test for potential confounds with medication, we computed Spearman’s ρ
correlations between antipsychotic medication dosage in chlorpromazine equivalents and the four effects listed above. None of the correlations approached significance. Next, we analyzed whether any of the above effects were related to error rate, age, illness duration, or SZ symptoms as measured by the Positive and Negative Symptom Scale (PANSS) (Kay et al., 1987
). Whereas there were no correlations for errors, age, or illness duration, the frequency of the No-Square response at central sites correlated negatively with major categories of the PANSS: total symptoms (−0.72; p
< 0.01), positive symptoms (−0.58; p
< 0.05), negative symptoms (−0.62; p
< 0.05), and general symptoms (−0.68; p
< 0.05). Particularly notable in view of our hypothesis about neural synchrony are the correlations with specific items, such as delusions (−0.68; p
< 0.05), conceptual disorganization (−0.66; p
< 0.05), and poor attention (−0.83; p
< 0.001). Thus, the lower the frequency of the early evoked gamma band response, the greater the degree of SZ symptoms.
Phase coherence data
Next, we examined phase coherence, which is indicative of coupling between distant brain regions (Varela et al., 2001
). Periods of coherence change relative to the baseline (coherence increases and decreases) were computed between each pair of electrodes for each group using a statistical parametric mapping method (see Materials and Methods). The number of coherence changes for each group and condition are presented in . It can be seen that the NC group had far more coherence increases than the SZ group, particularly in the 20–26 Hz bin. Inspection of the coherence time-frequency maps revealed that this effect was because of the presence of the VEP in this frequency range, so subsequent analyses focused on the higher frequency bins.
Three main differences between the groups were observed (). First, the NC group had more coherence increases than the SZ group overall (249 vs 91) and in both the Square (χ2(1) = 89.5; p < 0.0001) and the No-Square (χ2(1) = 8.99; p < 0.01) conditions. Second, the NC and SZ groups had different frequency band distributions of coherence decreases in the Square condition (χ2(2) = 113; p < 0.0001); the SZ distribution peaked at 37–44 Hz, and the NC distribution peaked at 48–57 Hz. Third, in the Square minus No-Square comparison, coherence changes were mainly decreases in coherence for the SZ group. The frequency band distribution of coherence decreases differed between groups (χ2(2) = 115; p < 0.0001): SZ coherence decreases showed a peak at 37–44 Hz (χ2(2) = 17.0, p < 0.001), where there were significantly more coherence decreases than the NC group (χ2(1) = 30.9; p < 0.0001), especially in the Square condition (). In summary, the phase coherence data point to abnormalities of coherence in long-distance neural synchronization in schizophrenia.
The spatio-temporal distributions of coherence changes in the 37–44 Hz bin () revealed several patterns. For the NC group, coherence changes began in the 0–75 msec interval and were concentrated in the 75–150 msec interval, coinciding with the early evoked gamma band response, and showed a roughly similar pattern of anterior–posterior coherence increases in the Square and No-Square conditions. Consequently, there were few differences in coherence in the Square minus No-Square comparison. For the SZ group, the onset of coherence changes was delayed, not beginning until after 75 msec. In the Square condition, the number of coherence changes (mainly decreases) was largest in the 225–300 msec interval and involved mostly inter-hemispheric interactions between posterior sites. In contrast, in the No-Square condition, the coherence changes were predominantly anterior–posterior increases in the 75–150 and 150–225 msec intervals, more in agreement in latency and directionality with those in the NC data. In the Square minus No-Square comparison, the SZ subjects displayed a consistent pattern of interhemispheric coherence decreases between posterior sites across the 0–300 msec period, in contrast to the NC group. The electrodes most often involved in these coherence decreases were P7 and P8, which lie over posterior temporal visual areas. Thus, phase coherence changes generally had a delayed onset for SZ compared with NC subjects and differed maximally between the groups in the Square condition, in which the SZ pattern consisted of coherence decreases between posterior sites, contrasting with the NC pattern of coherence increases along the anterior–posterior axis.
These results demonstrate that long-range neural synchronization in the upper beta and gamma bands of the EEG is impaired in schizophrenia, particularly in the 40 Hz range. As in the phase-locking data, the illusory square stimuli were especially associated with a lack of coherence. It should be noted that while possible effects of volume conduction cannot be ruled out from these data, the calculation of phase coherence relative to the prestimulus baseline reduces such a confound by eliminating any overall level of coherence. Also, volume conduction would be expected to bias the spatial distribution of coherence patterns to adjacent electrodes, but this pattern was not observed (cf. Rodriguez et al., 1999