Evidence is accumulating that patients with schizophrenia exhibit relatively severe deficits in early visual sensory processing within the dorsal stream, while processing within the ventral stream appears to be relatively more intact. Here, illusory contour (IC) processing was investigated in a cohort of schizophrenia patients and age-matched healthy controls using high-density visual evoked potentials (VEPs), spatiotemporal topographic analyses and the Local Auto-Regressive Average distributed linear inverse source estimation. IC processing was assessed because it is now known to be an excellent metric of early processing within regions of the ventral visual stream. Results in the present study show that IC processing (106–194 ms) is spared in patients with schizophrenia, providing strong evidence that early ventral stream processing is essentially normal. This is so despite equally strong evidence that early dorsal stream processing is severely impaired in this population, as indexed by a robust decrement in amplitude of the P1 component in patients and a large topographic difference between groups for this component (54–104 ms). Source analysis confirmed that the flow of activity into the dorsal stream was substantially decreased in patients. As such, these results suggest that some aspects of early ventral processing are not entirely reliant on intact inputs from the dorsal stream. Lastly, we show that later phases of visual processing (240–400 ms) also rely on the activity of different brain networks in controls and patients, with the latter recruiting strong frontal activity perhaps as compensation for impaired ventral stream processing during this period. We interpret the present findings in the context of a two-stage processing model. Under this model, it is suggested that the second stage of ventral stream processing is dependent on the fidelity of inputs from the dorsal visual stream and that impairment of this critical modulatory input may underlie the failure of ‘higher-level’ ventral stream processes in this population.
Perhaps the most obvious outward manifestations of schizophrenia are the severe anomalies in cognitive functioning that are typical of this devastating disorder. It is not surprising then that the vast majority of research studies in schizophrenia have concentrated on the so-called ‘higher-order’ cognitive functions — functions such as working memory, executive control, attention and abstract reasoning to name a few (e.g. Goldman-Rakic, 1994; Goldberg and Gold, 1995; Weinberger and Gallhofer, 1997; Green, 1998). For a long time, it was not fully considered that there might be equally severe deficits in early sensory processing. Recent studies, however, have begun to show that some of the most elementary sensory-perceptual processes may also be seriously impacted in schizophrenia. For example, patients exhibit severe deficits in such simple tasks as tone-matching, phoneme recognition, weight discrimination and smooth pursuit eye-movements (e.g. Braff et al., 1992; Waldo et al., 1994; Clementz, 1996; Shelley et al., 1999; Javitt et al., 1999; Cienfuegos et al., 1999; Leitman et al., 2005). Since our higher-order cognitive processes are largely dependent on the fidelity of information input from early sensory-perceptual stages of processing, it will be vital to determine the extent and nature of these sensory-level impairments in schizophrenia. It is entirely plausible that some apparent deficits in cognitive function will ultimately trace their origins to relatively basic sensory-perceptual dysfunction.
Early sensory processing deficits in the auditory system of schizophrenia patients are now well documented and have been the subject of extensive physiological investigation (e.g. Javitt et al., 1993, 1995, 1998, 2000; Alain et al., 2002; Michie et al., 2002; Salisbury, 2002). On the other hand, while behavioral studies have outlined apparent dysfunction in visual sensory-perceptual tasks in patients (e.g. Green et al., 1994; O'Donnell et al., 1996; Schwartz et al., 2001; Keri et al., 2002; Tek et al., 2002), a relatively modest number of studies have investigated the underlying neurophysiology (e.g. Roemer et al., 1978; Connolly et al., 1983; Jibiki et al., 1991; Bruder et al., 1998; Butler et al., 2001; Foxe et al., 2001; Doniger et al., 2002). Nonetheless, an intriguing pattern of deficits has been uncovered that suggests that visual sensory dysfunction in schizophrenia may have a relatively specific etiology. This pattern, which we will briefly describe here, is fascinating both for its clinical implications in patients with schizophrenia and for the potentially revealing information it provides for our understanding of basic visual functioning.
Using steady-state visual evoked potentials (ssVEPs) as the dependent measure, Butler et al. (2001) showed that patients exhibited a robust deficit in their responses to magnocellularly biased stimuli, whereas responses were relatively intact for parvocellularly biased stimuli. The magnocellular system is known to project predominantly to the dorsal visual stream (see Merigan and Maunsell, 1993), the so-called ‘where’ pathway, which courses dorsally from the primary visual area (V1) into the parietal and superior temporal cortices. Magnocellular impairment suggested that patients might be specifically impaired in visuospatial tasks, which are usually associated with processing in this dorsal stream, a notion borne out by other recent studies (e.g. Cadenhead et al., 1998; Schwartz et al., 1999). On the other hand, Butler's data suggested that parvocellular function might be relatively more intact. The parvocellular system projects mainly (although by no means exclusively) to the ventral visual stream, the so-called ‘what’ pathway, which courses from V1 into the inferior occipital and lateral occipito-temporal cortices (e.g. Ungerleider and Mishkin, 1982). This stream is primarily associated with object-recognition processes — the finer grained analysis of object information.
In a subsequent study using high-density event-related potential (ERP) recordings, Foxe et al. (2001) found that generation of one of the most robust components of the VEP, the so-called ‘P1’ component, was significantly impaired in patients. This VEP component is considered to be an index of early sensory processing in extrastriate visual areas, and is seen to occur with typical peak latency in the range of ∼70–110 ms (see Foxe and Simpson, 2002). The P1 was found to be significantly more impaired over the dorsal occipito-parietal scalp than it was over the ventral occipito-temporal scalp. As the P1 represents the activity of generators in both the dorsal and ventral visual streams (e.g. Simpson et al., 1995; Woldorff et al., 1997; Murray et al., 2001), this result suggested that the dorsal stream contributions to P1 were relatively more impaired than the ventral, in apparent support of the findings by Butler et al. (2001). In contrast to this P1 reduction, however, and of critical importance to the present study, Foxe et al. (2001) found that the N1 component of the VEP appeared to be of entirely normal amplitude and latency in patients, despite the fact that the preceding P1 appeared so severely impaired. The N1 component, typically observed in the latency range from ∼140 to 200 ms, is believed to be primarily generated in structures of the ventral visual stream, as evidenced by intracranial grid-electrode recordings (e.g. Allison et al., 1999) and scalp topographic studies (e.g. Doniger et al., 2000, 2001; Bentin et al., 1999). Thus, the presence of an essentially normal N1 component structure in patients also appeared to suggest that processing in the ventral visual stream was relatively more preserved in schizophrenia. From a basic research perspective, this was a fascinating finding, essentially suggesting that generation of the P1 and N1 components of the VEP can occur with a fair degree of independence from each other, rather than representing successive and causally related stages of processing within the visual hierarchy. We will come to this observation in turn.
To this point, our studies largely suggested that the ventral visual stream was comparatively well preserved in patients with schizophrenia. Therefore, in a third study, we investigated one of the critical functional processes of the ventral visual stream — so-called ‘perceptual closure’ (Doniger et al., 2002). Very often, in natural viewing settings, the image of an object that actually impinges upon the receptors of the retina can be significantly degraded, due to everyday environmental factors such as occlusion, shadowing and novel orientations. Nonetheless, humans are highly adept at recognizing objects even under these fairly adverse viewing circumstances. The term ‘perceptual closure’ has been used to refer to this fundamental ability to interpolate or ‘fill in’ the missing information in our retinal input (e.g. Bartlett, 1916; Snodgrass and Feenan, 1990; Foley et al., 1997). In a series of studies in healthy control subjects, an ERP component (Ncl — negativity for closure) was defined that was evoked when subjects were ‘closing’ degraded (fragmented) line-drawings of common objects (Doniger et al., 2000, 2001). This Ncl component occurred with peak latency in the time-range from ∼ 270 to 320 ms, and with a characteristic scalp topography over lateral occipital cortices. It was interpreted as representing neural processes underlying the effortful extraction of object identity within the lateral occipital complex (LOC) of the ventral visual stream. The findings led us to propose a model of object processing, whereby the initial N1 component represented a relatively automatic ‘perceptual’ level of object processing in the LOC and the subsequent Ncl represented an effortful ‘conceptual’ level of processing within the same structure. This second level of processing was invoked when the initial automatic processing represented during the N1 time-frame failed to fully reveal a given object's identity (Doniger et al., 2001; for a similar model, see Tulving and Schacter, 1990). When we conducted a similar ‘perceptual closure’ experiment in patients, we found that they were profoundly impaired in their ability to generate the Ncl and that this failure was correlated with their behavioral performance. That is, we found that patients were largely unable to ‘close’ fragmented pictures. Indeed, in order for most patients to reliably identify a given object, practically the entire image needed to be presented, and patients were highly susceptible to even the mildest levels of fragmentation employed.
This latter study at first appears to be at odds with our initial findings (Butler et al., 2001; Foxe et al., 2001). That is, whereas our initial studies suggested relative sparing of the ventral stream as represented by essentially normal N1 generation, the results of Doniger et al. (2002) appeared to suggest rather profound impairment in a fundamental process of the ventral visual stream. A closer look at the results of Doniger is warranted. While the Ncl modulation was almost entirely absent in the patients in this study, the initial N1 component was found to be, once more, of normal amplitude and latency. Again, this finding would appear to suggest that the first round of processing in the ventral LOC was not particularly impaired in patients. Further, Doniger et al. (2002), like Foxe et al. (2001), showed a severe deficit in generation of the earlier P1 component, suggesting early processing deficits in the dorsal visual stream.
This pattern of results left open the possibility that early processing in the ventral visual stream, processing at the relatively automatic level represented by the N1, was unimpaired in patients, whereas later and more effortful ‘conceptual’ level processing in this stream was impaired. However, as we believe that the N1 and Ncl are largely generated within the same complex of ventral stream visual areas, this pattern of effects also suggested that whatever the nature of the impairment in ventral processing that we had uncovered, it was unlikely to be due to fundamentally impaired neuronal mechanisms within the LOC itself or to a decrement in initial afferent flow into these regions. The fact that we have repeatedly found the N1 component to be of normal amplitude and latency in patients, however, does not of itself provide proof that this automatic object processing stage is unimpaired. Critically, stimulus parameters that explicitly modulate this component have not yet been tested in this population and it remains possible that, despite the apparent ‘normality’ of the patient N1, early processing in the LOC may also be disordered.
Therefore, the present study was designed to explicitly assess this early ventral visual processing component. A recent study in healthy controls from our laboratory (Murray et al., 2002) showed that a major portion of the neural processes responsible for the construction of basic illusory contours (ICs) are carried out within the LOC during this early processing phase. This study showed a robust bilateral modulation of the N1 component for IC figures. Scalp-topographic mapping, inverse source analysis and combined functional magnetic resonance imaging of this effect all pointed to major generators within the LOC. The early timing of the IC effect, which began at just 90 ms over the lateral occipital scalp, led us to surmise that this basic object processing function was happening within the ventral visual stream during, or shortly following, the initial input stages of processing. As such, this effect is an excellent dependent measure for assessing early ventral stream function during the putatively automatic stage of processing.
Here, we investigate IC processing in a cohort of schizophrenia patients and age-matched healthy control subjects. We use the methods of high-density electrical mapping and inverse source analysis to assess these processes, which are an index of early and putatively automatic processing within the LOC. Our data show that IC processing is spared in patients with schizophrenia and provide strong evidence that early ventral stream processing in the LOC is essentially normal. This is so despite equally strong evidence that early dorsal stream processing may be severely impaired in this population. As such, these results suggest that some aspects of early ventral processing are not entirely reliant on inputs from the dorsal stream. In light of previous results, where we have shown that later phases of processing within the ventral stream are severely impaired in patients with schizophrenia (Doniger et al., 2002), we interpret the present findings in the context of a two-stage processing model. Under this model, it is suggested that the second stage of processing in the LOC is dependent on the fidelity of inputs from the dorsal visual stream and that impairment of this critical modulatory input may underlie the failure of ‘higher-level’ perceptual closure processes in this population.
Materials and Methods
Sixteen chronic, medicated patients (two female), aged 23–50 years (mean = 40.9 ± 6.2 years) and meeting DSM-IV criteria for schizophrenia participated. Demographic and clinical information for the patients is given in Table 1. That all patients were receiving medication at the time of testing could be a limiting factor in this study (see e.g. Spohn et al., 1977). However, visual processing deficits in schizophrenia have been shown irrespective of whether patients were medicated or not (Butler et al., 1996), and there was no significant correlation between neuroleptic dose and behavioral or ERP measures. Also, previous findings of a P1 deficit in the absence of any N1 reduction argue against a generalized medication effect (Foxe et al., 2001; Doniger et al., 2002), and recent studies of early auditory processing have shown no effects of haloperidol on ERP amplitudes (Pekkonen et al., 2002a). Nonetheless, a medication effect as a factor in the present findings cannot be completely ruled out.
Control subjects comprised 17 (six female) paid volunteers, aged 23–57 years (mean = 40.6 ± 10.9 years). The mean age of patients and control subjects did not significantly differ [t(31) = 0.09; P = 0.93]. Thirteen of the 16 patients and 15 of the 17 control subjects were right-handed as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971). All subjects had normal or corrected-to-normal vision. The Institutional Review Board of the Nathan Kline Institute for Psychiatric Research approved all experimental procedures, and all patients were recruited from facilities associated with the Institute. Written, informed consent was provided by all subjects once the procedures of the experiment were fully explained.
A single rater performed Positive and Negative Symptoms Scale (PANSS) ratings. Factors were defined according to White et al. (1997). All patients were receiving antipsychotic medication at the time of testing. However, no significant correlations were observed between antipsychotic dose and any of the experimental variables evaluated in this study. Controls were free of psychiatric illness or symptoms by self-report using criteria from the SCID-NP (see Spitzer et al., 1992), and all reported no history of alcohol or substance abuse.
Stimuli and Task
Subjects were instructed to centrally fixate an array of Kanizsa-type (Kanizsa, 1976) ‘pacmen’ inducers that were oriented in one of two manners to either form or not form an illusory shape (‘IC present’ and ‘IC absent’ conditions; see Fig. 1a). Five shapes were used: square, circle, triangle, pentagon and star. Inducers (circular, subtending 3° of visual angle) appeared black on a gray background and were presented on a computer monitor (Iiyama Vision Master Pro 502, model no. A102GT) located 114 cm from the subject. In order to produce illusory shapes of the same maximal width and height (6° in either plane), the eccentricity of inducers varied slightly across shapes (for further details, see Murray et al., 2002).
The timing of presentations was such that each stimulus appeared for 500 ms, followed by a blank gray screen for 1000 ms. Then a ‘Y|N’ response prompt appeared and remained on the screen until a response was made, allowing subjects to control stimulus delivery. Another blank screen (1000 ms duration) followed responses. Subjects were instructed to press one button for a ‘No’ response, indicating that they did not perceive a shape ‘pop-out’ from the background, or a second button for a ‘Yes’ response, indicating that they perceived a shape. IC present and IC absent inducer configurations were randomly presented and were equally probable. Subjects were encouraged to take breaks between blocks to maintain high concentration and prevent fatigue. Use of the response prompt was to diminish the impact of motor responses on the sensory VEP.
Subjects' behavioral responses were recorded through right-hand button presses. Continuous EEG was acquired through Neuroscan Synamps (Neurosoft Inc., Sterling, VA) from 64 scalp electrodes (impedances <5 kΩ, nose reference, 0.05–100 Hz band-pass filter, 500 Hz digitization). Trials (−100 to +500 ms peri-stimulus epochs) with blinks and eye movements were rejected on the basis of horizontal and vertical DC electro-oculogram. An artifact rejection criterion of ±60 μV was used at all scalp sites to reject trials with excessive EMG or other noise transients. Only trials with correct responses and meeting these criteria were included in analyses. Within the patient group, the average (±SD) number of accepted trials for each stimulus condition was 376 ± 258, and for the control subjects it was 519 ± 220. Epochs of continuous EEG (100 ms pre- to 500 ms post-stimulus onset) were averaged from each subject for both the IC present and IC absent stimulus conditions separately to compute the VEP. Baseline was defined as the −100 to +20 ms epoch. Data from artifact electrodes from each subject and condition were interpolated (Perrin et al., 1989). Following this procedure and prior to group-averaging, each subject's data were 40 Hz low-pass filtered, down-sampled to a common 61-channel montage (see Fig. 1) and recalculated against the average reference. In addition, data were normalized to their mean global field power (GFP; Lehmann and Skrandies, 1980) over the −100 to +500 ms period.
As detailed above, we had specific hypotheses regarding both the early sensory processing period (the time-frame of the P1 component) and the early phase of object processing in the LOC (the time-frame of the N1 component). Two varieties of analyses were conducted: the first tested for modulations in VEP waveforms and the second for changes in scalp topography, and by extension brain generators. The analysis of VEP waveforms was conducted for electrodes positioned at the center of the topographic distributions of the P1 and N1 components, which can be seen in Figure 1 for each population and experimental condition. Five symmetrical pairs of scalp-sites over parieto-occipital scalp were selected (O1/O2, PO3/PO4, PO5/PO6, P1, P2, and P3/P4 — see the bottom leftmost map in Fig. 1). Area measures (versus the 0 μV baseline) were taken during time windows identified from the topographic pattern analysis described below, ±10 ms at each end of this period to allow a buffer to minimize the possibility that different topographies from any individual's data were averaged across time. These area measures were then submitted to a repeated measures multivariate analysis of variance (MANOVA). The between-subjects factor was group (patients, controls) and the within-subjects factors were contour (IC present, IC absent), hemiscalp (left, right) and electrode (five pairs of scalp sites as detailed above). All significance levels were two-tailed, with a preset α- level for significance at P < 0.05.
The second family of analysis entailed submitting the 500 ms post-stimulus period of the VEPs from each condition and population to a topographic pattern (i.e. map) analysis. Maps were compared over time within and between conditions and populations, since topographic changes indicate differences in the brain's active generators. The methods applied here have been described in detail elsewhere (e.g. Murray et al., 2004a,b; see also Lehmann and Skrandies, 1980; Lehmann, 1987). Briefly, this method is independent of the reference electrode and is insensitive to pure amplitude modulations across conditions (topographies of normalized maps are compared). A modified cross-validation criterion determined the number of maps that explained the whole group-averaged data set (Pascual-Marqui et al., 1995; the formula is listed in the Appendix). In conceptual terms, this criterion provides a metric of the trade-off between increasing the number of maps to increase the amount of variance in the collective data set one explains and the consequent loss in degrees of freedom. The pattern of maps observed in the group-averaged data from both populations and experimental conditions was statistically tested by comparing each of these maps with the moment-by-moment scalp topography of individual subjects' VEPs from each condition. Each time point was labeled according to the map with which it best correlated (i.e. that which yielded a larger spatial correlation value; see the Appendix for the formula; details are given in Brandeis et al., 1995) — a procedure hereafter referred to as ‘fitting’. It is important to note that this is a labeling procedure. It is therefore not forcibly the case that the spatial correlation value for a given subject and condition is statistically significant versus that of other maps fitted over the same period. Rather, this procedure provides an index of whether different patterns of maps observed in the group-averaged data are statistically reliable across subjects and/or conditions. That is, this fitting yields a measure of the frequency (in milliseconds) of a given map's observation in the data from each subject of each population and experimental condition (as well as the timing of such observations, though we note that absent values are a possibility when a given map is not observed in a individual subject's data), which are then submitted to a MANOVA with the between-subjects factor of group (patients, controls) and the within-subjects factors of contour (IC present, IC absent) and map. This revealed if a given experimental condition and/or population's responses are more often described by one map versus another, and therefore if different generator configurations better accounted for particular experimental conditions/populations.
Lastly, we registered the timing of effects with the onset of the VEP from each condition and population, which was determined in the following manner. We first calculated the 95% confidence interval of the mean GFP for each condition and population during the pre-stimulus period. We used this value as a measure of ‘baseline’ activity. The time-point when the mean GFP minus its 95% confidence interval exceeded this baseline value for at least the next 20 ms (e.g. Guthrie and Buchwald, 1991) was labeled as VEP onset.
Estimates of the intracranial sources of the effects in the data were derived through the use of a distributed linear inverse solution, based on a Local Auto-Regressive Average (LAURA) model of the unknown current density in the brain (Grave de Peralta et al., 2001). LAURA uses a realistic head model with a solution space of 4024 lead field nodes, selected from a 6 × 6 × 6 mm grid equally distributed within the gray matter of the Montreal Neurological Institute's (MNI's) average brain. Like other inverse solutions of this family, LAURA is capable of dealing with multiple simultaneously active sources of a priori unknown location and makes no assumptions regarding the number or location of active sources (for a recent comparison of inverse solution methods, see Michel et al., 2004). This linear distributed inverse solution selects the source configuration that better mimics the biophysical behavior of electric vector fields and produces a unique estimator of the current source density vector inside the brain. That is, the estimated activity at one point depends on the activity at neighboring points as described by electromagnetic laws (for details, see Grave de Peralta and Gonzalez Andino, 2002). For the analysis presented here, we used the cubic power of the inverse distance. The LAURA method increases up to 32% (versus the 20% reported for other inverse solution approaches) the number of sources with zero localization error and decreases the maximum error whilst keeping the average error lower than 1 for almost any eccentricity within the solution space (for details, see Grave de Peralta and Gonzalez Andino, 2002). It is important to note that these estimations provide visualization of the likely underlying sources and do not represent a statistical analysis.
Both subject groups correctly indicated IC presence versus absence with a high degree of accuracy (patients = 89.8 ± 12.9%; controls = 93.9 ± 10.2%) and performance was statistically indistinguishable between subject groups [t(31) = 1.02; P = 0.3]. Since a response prompt was used in this paradigm to diminish the influence of motor responses on the sensory VEP, there were no reaction times to be analyzed.
Figure 1 depicts isocontour voltage maps of the group-averaged VEPs over the 50–390 ms post-stimulus epoch for both healthy control and schizophrenia patient populations, as well as both IC present and IC absent experimental conditions. In all cases, clearly identifiable P1 and N1 components were exhibited. A distinct negative-going C1 component was not observed with the present stimulus configuration. This is as expected, as the stimuli are presented centrally, spanning all four visual quadrants. In order to evoke a strong C1 component, retinotopically specific single-quadrant stimulation is typically necessary (e.g. Clark et al., 1995; Foxe and Simpson, 2002). The cruciform configuration of the four banks of calcarine cortex (which tend to form a macroscopic closed field) generally cause a significant reduction of the C1 response.
For healthy controls, VEP onset (see Materials and Methods for details) was at 46 and 40 ms for the IC present and IC absent conditions, respectively. For schizophrenia patients, these values were 54 and 40 ms. Therefore, no robust difference was observable in the timing of the onset of the VEP across subject groups. In contrast, visual inspection of these maps was suggestive of both between-group differences during the P1 component as well as within-group differences during the N1 component as a function of stimulus condition, consistent with our previous studies (Foxe et al., 2001; Murray et al., 2002, 2004a; Doniger et al., 2002). In addition, the topographic pattern analysis revealed that nine different scalp topographies accounted for the collective 500 ms post-stimulus periods across populations and experimental conditions (Fig. 2). The global explained variance accounted for by these nine scalp topographies was 97.33%, and is calculated as the sum of the explained variances at each time point across populations and conditions weighted by each time point's global field power (see Appendix for full formula). Periods of stable scalp topography extended from ∼0–52, 54–104, 106–194, 196–238, 240–400 and 400–500 ms, and constitute a more observer-independent manner of defining VEP components (for a discussion, see e.g. Michel et al., 2004). That is, scalp topography did not vary randomly across time, but rather remained in a particular configuration for a period of time. Others have referred to this as functional microstates of the brain (e.g. Lehmann and Skrandies, 1980), and a similar phenomenon is observable at the level of single unit recordings (Rolls and Tovee, 1994). We present our electrophysiological results as a function of particular VEP components.
Early Sensory Processing (the P1 Component)
To test for VEP waveform modulations during the P1 component (64–94 ms), a four-way MANOVA (with factors of group, contour, hemiscalp and scalp-site) was conducted as described above. As expected, there was a main effect of Group [F(1,31) = 5.258; P = 0.029], such that P1 amplitude was significantly diminished in patients relative to controls. This effect is illustrated in the upper bar graph of Figure 1, where the area measures from all 10 electrodes have been averaged together for illustrative purposes. VEP waveforms at sites PO3 and PO4, where this effect was largest are shown in Figure 1. Neither effect of contour [F(1,31) = 0.66; P = 0.42] nor hemiscalp (P = 0.95) was significant. Only the interaction between contour and electrode reached the 0.05 significance criterion [F(1.93,59.7) = 6.34; P = 0.003].
To statistically test for topographic (i.e. generator) modulations during the P1 component, we conducted the fitting procedure (see Materials and Methods for details) over the 54–104 ms period with the corresponding scalp topographies identified from the group-averaged data. We found that one scalp topography better accounted for the VEPs from healthy controls, whereas a different scalp topography better accounted for the VEPs from schizophrenia patients (gray and black boxes, respectively, in Fig. 2), independently of experimental condition. This was confirmed statistically with a MANOVA using the frequency of each map observed in the data from each individual as our repeated measure. There was a significant interaction between group and map [F(1,31) = 19.89; P < 0.001; see left bar graph in Fig. 2], indicating that different configurations of active brain areas generated the P1 in controls and patients. No other effects or interactions reached the 0.05 significance criterion (all Ps > 0.25).
Figure 3 (top panel) shows the results of the LAURA source analysis during the P1 period. Given the results of the above topographic analysis, data from each subject and experimental condition were averaged across time (54–104 ms) to obtain a single data point that was then applied to the LAURA source estimation. These estimations were then group-averaged for each group and experimental condition, separately. We would iterate that these source estimations are a visualization only, and do not represent a statistical analysis, though we would refer the interested reader to a recent review and comparison of different source estimation approaches (Michel et al., 2004). Source estimations from healthy controls for both the IC present and IC absent conditions show bilateral lateral-occipital activity that extends superiorly and medially, as well as posterior occipital activity. By contrast, source estimations from schizophrenia patients for both experimental conditions show comparatively weaker activity in these areas that was particularly evident for the left lateral occipital and posterior occipital sources.
Illusory Contour Processing (the N1 Component)
For the N1 component, areas (versus the 0 μV baseline) were measured over the 116–184 ms epoch and subjected to the above MANOVA. In contrast to the observations with the P1 component, the main effect of group did not approach significance [F(1,31) = 1.52; P = 0.23], indicating that N1 amplitude from patients did not significantly differ from that of healthy controls. However, there was a highly robust main effect of contour [F(1,31) = 68.79; P < 0.00001], with greater magnitude responses to IC than to NC stimuli. This confirms that IC processing occurs over this epoch in both populations. These effects are illustrated in the lower bar graph of Figure 1, where the area measures from all 10 scalp sites have been averaged together for illustration purposes. Further, the lack of a significant interaction between the factors of group and contour [F(1,31) = 0.03; P = 0.86] provides no evidence that IC stimuli were initially processed differentially by patients and control subjects over parieto-occipital scalp sites. The interaction between the effects of contour and hemiscalp approached our significance criterion [F(1,31) = 2.90; P = 0.10], suggesting that IC processing was more pronounced over right hemiscalp sites for both experimental groups. In contrast, the main effect of hemiscalp did not reach our significance criterion [F(1,31) = 0.04; P = 0.85], showing that the N1 component is highly bilateral in both populations. In addition, the results of the topographic pattern analysis, identified the same scalp topography for both populations and experimental conditions over the 106–194 ms period (see Fig. 2), suggesting that IC sensitivity follows from a modulation in response strength rather than a change in the underlying generator configuration (see also Murray et al., 2002, 2004a; Pegna et al., 2002).
The middle panel of Figure 3 shows the results of the LAURA source analysis during the N1 period. Data from each subject and experimental condition were averaged across time (106–194 ms) to obtain a single data point that was then applied to the LAURA source estimation. These estimations were then group-averaged for healthy controls and schizophrenia patients as well as each experimental condition, separately. In all cases, bilateral lateral-occipital and posterior occipital sources were observed.
The Ncl Component
Lastly, for the Ncl component, areas (versus the 0 μV baseline) were measured over the 250–390 ms epoch and subjected to the above MANOVA. No main effect or interaction reached our significance criterion (all Ps > 0.15). However, the results of the topographic pattern analysis identified distinct scalp topographies for control and patient populations over the 240–400 ms period that were independent of the experimental condition (white and black stippled boxes in Fig. 3, respectively), repeating the pattern observed during the P1 component. This was confirmed statistically with a MANOVA using the frequency of each map's observed in the data from each individual as our repeated measure. There was a significant interaction between group and map [F(1,31) = 4.17; P < 0.05; see right bar graph in Fig. 3], indicating that different configurations of active brain areas generated the Ncl in controls and patients. No other effects or interactions reached the 0.05 significance criterion (all P-values >0.25).
The bottom panel of Figure 3 shows the results of the LAURA source analysis during the Ncl period. Data from each subject and experimental condition were averaged across time (240–400 ms) to obtain a single data point that was then applied to the LAURA source estimation. These estimations were then group-averaged for healthy controls and schizophrenia patients as well as each experimental condition, separately. Responses from healthy controls included bilateral activity in regions of the lateral occipital, superior parietal and posterior occipital cortices. In contrast, responses from schizophrenia patients did not include robust sources within these regions of the right hemisphere. In addition, responses from patients yielded sources within right inferior frontal cortex that were not evident in the corresponding responses from healthy controls.
The present results demonstrate that neural processing of basic IC stimuli is substantially intact in patients with schizophrenia. That is, we found robust bilateral enhancement of activity during the N1 component of the visual evoked potential when subjects' viewed IC figures relative to non-contour control stimuli — an enhancement that was just as strong in patients as it was in control subjects. Further, behavioral results showed that patients were no different to controls in their ability to recognize the simple presence or absence of ICs. Our previous studies in healthy controls (Murray et al., 2002, 2004a) showed that the neural processes responsible for the construction of basic ICs were primarily carried out within structures of the ventral visual stream — the so-called lateral occipital complex (LOC; see also Mendola et al., 1999; Halgren et al., 2003). The very early timing of these processes (which began at ∼90 ms) suggested that IC stimuli were likely being processed during the initial phases of activity within these ventral stream regions, at a relatively automatic stage of processing. As such, the present findings, showing an entirely similar pattern of effects in patients, suggest that initial processing within LOC remains relatively unimpaired in schizophrenia.
The successful processing of ICs appears to be intact despite the presence of clear and considerable deficits for patients during an earlier phase of visual processing; that is, during the P1 component (54–104 ms). This latter finding replicates previous research from our laboratory where similar P1 decrements were first observed in schizophrenia patients (Foxe et al., 2001; Doniger et al., 2002), and with a more recent replication of this finding by Spencer et al. (2003). It also extends these findings by assessing the relative onset timing of this deficit, which is shown to lag the absolute onset of the VEP response in visual cortices by just ∼10–15 ms. There was no robust difference in the actual instant of VEP response onset between the two groups, suggesting that the P1 amplitude reduction is not due to a general delay in visual processing in schizophrenia. By extension, the implication is that the initial afferent volley through the visual system is likely intact. The present study further extends previous results by providing statistical evidence of a topographic modulation between VEP responses from healthy controls and schizophrenia patients (see also Foxe et al., 2001). A topographic modulation such as this is indicative of a change in the configuration of active brain regions (see e.g. Fender, 1987). The early timing of this deficit, its general scalp topography over dorsal occipital scalp and inverse source-estimation all support the possibility of an early visual processing deficit that is most pronounced in regions of the dorsal visual stream, as has been suggested by previous psychophysical (Butler et al., 1996; Chen et al., 1999a,b; Green et al., 1999; Schwartz et al., 1999; Butler et al., 2001) and neurophysiological research (Foxe et al., 2001; Doniger et al., 2002).
One intriguing implication of this pattern of results is that dysfunction of early signal transmission through dorsal stream areas has little or no impact upon subsequent IC processing within ventral stream areas (LOC: e.g. Mendola et al., 1999; Murray et al., 2002, 2004a). This is a somewhat surprising result, as it is now well-known that the dorsal and ventral visual systems have extensive cross-connectivity and that the ventral ‘object-recognition’ stream receives strong modulatory inputs from the faster magnocellularly biased dorsal stream (e.g. Vidyasagar, 1999; Schroeder et al., 1998, 2001). The current dataset suggests, however, that such modulatory inputs are not critical to at least some portion of the early object-processing functions, accomplished during the initial phase of ventral stream activity. However, the notion of a failure of cross-stream modulation is consonant with our previous findings regarding the later phases of processing involved in perceptual closure processes (the Ncl component; see e.g. Doniger et al., 2002). We would posit that the fact that IC processing within the LOC is intact, and that we have repeatedly found the N1 component to be of normal latency and amplitude in schizophrenia, makes it unlikely that these regions have significant intrinsic structural dysfunction. Why, then, do these regions produce normal early activity (N1) but subsequently fail to produce a normal Ncl component? One plausible explanation is that processing within these structures beyond the initial automatic phase is indeed dependent upon the fidelity of modulatory inputs from the dorsal visual stream. Recall that our initial two-stage model of processing posited that N1 represented early automatic ventral processing whereas Ncl represented a subsequent phase of more effortful ‘conceptual’ processing (see Doniger et al., 2001; Murray et al., 2002, 2004a). It is of note that we also find substantial differences between patients and controls during the Ncl time-frame in the present study, with patients showing statistically significant differences in topography (and by inference in underlying generator configuration) during the late processing phase within LOC. As such, the present results extend our model, suggesting that dorsal stream inputs may not be critical for the automatic phase but are important for the later phase of processing.
A measure of support for this two-stage model is to be found in data from neurological lesion patients suffering from so-called hemispatial neglect syndrome. This is a condition that predominantly results from vascular lesions to dorsal stream structures in the right inferior parietal or temporo-parietal cortices (e.g. Vallar and Perani, 1986; Mesulam, 2000), and it is prototypically marked by an impaired ability to perceive or orient to stimuli presented in contralesional spatial locations (e.g. Mesalum, 1981). It has been shown that performance on the line-bisection task, which is often used as a sensitive measure of the extent of attentional dysfunction in this population, was substantially improved in these patients by the presence of a surrounding IC shape, even when the IC inducers were presented within the neglected hemifield (Mattingly et al., 1995, 1997; Vuilleumier and Landis, 1998; Olk et al., 2001; Vuilleumier et al., 2001). This benefit, however, was not seen for patients in whom the lesion included the lateral occipital cortex (Vuilleumier et al., 2001). One clear implication of this facilitation in performance is that IC processing occurs pre-attentively and/or automatically (i.e. occurs without explicit awareness). A second implication is that substantial dorsal stream lesions do not disrupt successful early-stage ventral processing for ICs. Thus, certain types or modes of object recognition, in both schizophrenia patients as well as neurologically normal subjects, may not critically rely on intact inputs from the dorsal visual processing stream.
Further support for interpreting IC processes and perceptual closure processes as relying on different modes of object recognition is found in VEP studies of selective attention to an object's hierarchical features (Han et al., 1999, 2000a,b). More specifically, P1 amplitude is enhanced for the processing of local versus global structures (e.g. a large ‘A’ composed from smaller ‘X’s). In conjunction with psychophysical evidence from patients with inferior parietal lesions (Robertson, 1991), it was concluded that this P1 modulation reflects activity within dorsal stream structures. Interpreting the present P1 decrement in this manner would thus suggest that IC and perceptual closure processes may be distinguishable in terms of their reliance on global vs. local grouping mechanisms, respectively. In support, patients' performance is selectively impaired on tasks requiring analysis of visual stimuli at a local level (Carter et al., 1996; Granholm et al., 1999). In contrast, psychophysical evidence from normal subjects would indicate that global, Gestalt features supercede local stimulus properties in the detection of IC shapes (e.g. Navon, 1977; Parks, 1980a,b; Ware, 1981; Sekuler, 1994). That is, the intact IC processes that are unaffected by the P1 decrement may rely on more global functions, whereas perceptual closure processes that are impaired in parallel with the P1 component may rely on more local processes.
Compensatory Involvement of Frontal Cortex?
One unanticipated result of our source analysis was particularly intriguing. During the later phase of processing, in the time-frame of the Ncl (240–400 ms), patients showed statistically greater activation of frontal cortex, particularly in right inferior frontal regions. As detailed above, this ‘extra’ activity occurred during a time-frame when patients exhibited dysfunctional processing within posterior visual sensory regions. As this was not expressly hypothesized here, these results must be interpreted with a fair degree of caution and they await replication in future studies. Nonetheless, one possibility is that in order to compensate for deficiencies in object processing abilities, frontal regions are recruited in patients. A similar model has emerged in the aging literature where it has been shown that older adults who are more resistant to decline in memory and executive functioning show increased recruitment of frontal brain regions (e.g. Buckner, 2004; see also Cabeza et al., 2000).
Hemispheric Laterality Effects in Schizophrenia
Results of studies aimed at assessing potential abnormalities in laterality in schizophrenia have provided an often-confusing picture (for an excellent review, see Green et al., 2003). Nonetheless, the bulk of studies have tended to support greater left hemisphere dysfunction than right (e.g. Gur, 1978; Wexler et al., 1991; Carter et al., 1996), although there is certainly no shortage of studies that have also implicated the right hemisphere (e.g. Kucharska-Pietura et al., 2002). Electrophysiological results have tended to be more uniform in implicating the left hemisphere. For example, using ERPs, a number of groups have assessed left and right hemisphere contributions to the relatively late occurring auditory P3 component, and the weight of evidence seems to indicate a greater decrement in left hemisphere contributions to this component (e.g. Morstyn et al., 1983; Faux et al., 1988; Salisbury et al., 1998, 1999). This finding also appears to correlate well with structural decrements in the left hemisphere (see McCarley et al., 2002; see also Kasai et al., 2003a,b). Bruder et al. (1999) found similar left hemisphere dysfunction in the generation of the earlier auditory N2 component evoked by syllabic stimuli. Studies of the laterality of the auditory change-detection potential (the mismatch negativity) have also suggested that impairment is greater in left hemisphere generators (e.g. Hirayasu et al., 1998; Pekkonen et al., 2002b; Youn et al., 2003), showing that potential left hemisphere deficits are not limited to higher-order cognitive processes, but may also occur at relatively early sensory processing stages (see also Roemer et al., 1979). In the visual sensory modality, there has been substantially less study of laterality effects. But, here too, there is some evidence for greater left hemisphere dysfunction in schizophrenia. For example, Roemer et al. (1978), recording simple VEPs to flashed checkerboard stimuli, found that their schizophrenic patients showed significantly reduced response stability during the relatively early sensory processing period (50–150 ms) and that this reduction in stability was significantly greater over the left occipital scalp. Of course, this is the time-frame of the visual P1 and the early portion of the visual N1 component. The present results accord well with this notion of an especially large deficit in early left hemispheric visual processing. Source analysis of the earliest phase of the P1 component, in the latency range of just 54–104 ms, demonstrated severe diminution of activity within the left hemisphere whereas right hemispheric activity was relatively more intact. However, it should be pointed out that right hemispheric activity, while appearing to be more preserved, is not of normal amplitude, so it would not be correct to consider this a purely left hemisphere deficit. Further, source analysis also highlights an additional distinction in this early activation pattern, that of dorsal versus ventral stream contributions to the early VEP. It is clear from our results (see also Foxe et al., 2001) that dorsal stream generators are substantially more affected than are ventral stream structures and that this is the case for dorsal generators in both of the hemispheres. Thus, while we also find that the left hemisphere is particularly impaired in patients with schizophrenia, our data show that this impairment is not exclusive to this hemisphere, and that a potentially more useful distinction is that between the dorsal and ventral visual streams.
Gamma-band Responses to IC Stimuli and Feature Binding
The majority of previous electrophysiological studies of IC processing have concentrated on activity in the ‘induced’ oscillatory (frequency) domain, and in particular upon gamma-band activity (∼40 Hz). These studies have targeted the putative role of synchronized gamma-band activity as a neuronal mechanism for feature binding (see e.g. Tallon-Baudry and Bertrand, 1999; Engel and Singer, 2001). An increase in induced gamma band (30–50Hz) activity is typically observed when comparing IC presence versus absence (e.g. Tallon-Baudry et al., 1996, 1997; Herrmann et al., 1999; Csibra et al., 2000; Herrmann and Mecklinger, 2000), and this finding is most commonly interpreted as evidence for bottom-up binding of coherent visual features (e.g. Tallon-Baudry et al., 1996, 1997; Tallon-Baudry and Bertrand, 1999). However, others have suggested that such oscillations index higher-order processes such as attention (e.g. Pulvermüller et al., 1997; Csibra et al., 2000) or target selection (Hermann et al., 1999). Likewise, there is little consensus on either the timing or source(s) of these oscillations. Some have reported only late (∼280 ms) gamma effects (Tallon-Baudry et al., 1996, 1997; Tallon-Baudry and Bertrand, 1999; Csibra et al., 2000), whereas others have also observed somewhat earlier (∼150 ms) phase-locked effects (Hermann et al., 1999). Moreover, in the studies of Tallon-Baudry and colleagues (Tallon-Baudry et al., 1996, 1997; Tallon-Baudry and Bertrand, 1999), effects were focused over central–posterior scalp sites, whereas Hermann et al. (1999) and Csibra et al. (2000) observed their effects frontally. We have argued previously (Murray et al., 2002) that the timing of IC effects in the broadband ERP (which begin as early as 90 ms) challenge claims that gamma oscillations represent bottom-up feature binding (Tallon-Baudry et al., 1996, 1997) since the effects we find in the broadband ERP substantially precede the induced gamma-band effects described in all previous studies, in some cases by as much as 200 ms (Tallon-Baudry et al., 1996, 1997). Further, the interpretation of such gamma effects as ‘bottom-up’ does not easily fit with what is known about the temporal trajectory of activation across the visual hierarchy, which is considerably faster than can be reconciled with these results (see Schroeder et al., 1998, 2001; Foxe and Simpson, 2002).
Although many reports have shown late gamma effects in the absence of gamma modulation during the early sensory processing phase, a recent study by Spencer et al. (2003) is an exception in this regard. In this study, which investigated the gamma-band response to IC stimuli in both healthy controls and schizophrenia patients, early gamma-modulations were seen in response to the IC stimuli over visual cortices in control subjects (∼80 ms). Compellingly, their schizophrenia patients did not show this differential gamma response. In keeping with the above models, the authors interpreted this as evidence for a less efficient feature-binding mechanism in schizophrenia. First, it is unclear why this study differed from all the previous studies in terms of the early timing and scalp topography of the gamma response, as the stimuli and task parameters used were substantially similar in all respects. Second, it should also be pointed out that if their schizophrenia patients did indeed have a deficit in feature-binding mechanisms, it had no effect on their abilities to recognize the presence of ICs. Just as in the present study, Spencer et al. (2003) found no differences in behavioral accuracy for their patients relative to controls, with patients achieving an accuracy rate of 96.1% when identifying contours. They go on to argue that this lack of behavioral decrement may be due to patients' reliance on some ‘later stage of analysis’. The present data, however, show that schizophrenia patients produce similarly early processing effects to controls when the traditional ERP frequency band is considered and we would note that this ERP effect was also present in the Spencer study. As such, given the lack of any behavioral accuracy differences between groups and the presence of entirely normal IC modulation of the ERP, we feel that it is unlikely that early stage processing for such stimuli is disordered in this population. The functional role of early gamma-band (∼80 ms) activity over visual cortices will clearly need further exploration and the relative timing of induced gamma activity and the effects seen in the broad band ERP will need to be studied together.
Appendix: Formulae Referred to in this Manuscript
1. Global explained variance (GEV)
2. Cross-validation criterion
3. Spatial correlation
We express our sincere appreciation to Beth Higgins for her ever-excellent technical expertise and for editing portions of this manuscript, to Denis Brunet for development of Cartool analysis software, and to Rolando Grave de Peralta Menendez and Sara Gonzalez Andino for the development of the LAURA inverse solution. Prof. Christoph Michel, Dr Sophie Molholm and Dr Pamela Butler provided valuable comments on an earlier version of the manuscript for which we are indebted. This work was supported in part by grants from the National Institute of Mental Health (MH63434 and MH65350 to J.J.F.; MH49334 to D.C.J.).
1Program in Cognitive Neuroscience, Department of Psychology, The City College of the City University of New York, North Academic Complex (NAC) 138th St. & Convent Avenue, New York, NY 10031, USA, 2The Cognitive Neurophysiology Laboratory, Nathan S. Kline Institute for Psychiatric Research, Program in Cognitive Neuroscience and Schizophrenia, 140 Old Orangeburg Road, Orangeburg, NY 10962, USA, 3The Functional Electrical Neuroimaging Laboratory, Division Autonome de Neuropsychologie and Service de Radiodiagnostic et Radiologie Interventionnelle, Centre Hospitalier Universitaire Vaudois, Hôpital Nestlé, 5 Av. Pierre-Decker, 1011 Lausanne, Switzerland and 4Department of Psychiatry, New York University School of Medicine, 550 1st Avenue, New York, NY 10016, USA