Auditory change detection in schizophrenia: sources of activity, related neuropsychological function and symptoms in patients with a first episode in adolescence, and patients 14 years after an adolescent illness-onset

The study protocol was reviewed and approved by both the board of the University of Essen Psychiatry Clinics and the Ethics Committee of the Faculty of Medicine according to the criteria of the Declaration of Helsinki (00-2-1357-Y). After the procedures had been described all subjects and care-givers gave written informed consent. A first-episode of DSM-IV schizophrenia was diagnosed in 28 adolescent inpatients (early-onset: EOS) on the basis of a clinical interview and hospital records first on the ward and then by research group personnel. This was confirmed 6 months later to exclude affective, schizoaffective and schizophreniform psychoses. Eighteen outpatients who had their first break as adolescents in our clinic on average 14.4y before (SD 3.5, range 8.1–19.7: S-14Y) were also recruited: 11 showed a partial remission (CGI 3–5), and 7 a chronic course or no remission (CGI 6–8). They had a mean of 4.6 hospitalizations (range 1–12). Of 34 patients contacted, 9 declined and 7 were too ill: participants did not differ on mean age, gender, illness-severity or social function [45]. Patients were excluded for other major psychiatric or somatic illness, alcohol abuse in the last 5 years and current substance abuse other than nicotine. Two age-matched comparison groups were recruited by advertisement and paid for participation (C-EOS, C-14Y). None had used drugs affecting the central nervous system or had a history of neurological or psychiatric illness. Gender balance was not considered significant as gender does not influence MMN [1, 46‐48]. All had normal or corrected to normal vision, and normal hearing on audiometric testing. In the S-14Y group 15 were receiving stable doses of antipsychotic medication. Of the EOS patients 15 had received the same treatment for >3 days, and 13 were examined without medication (Table 1). Assessment of the positive and negative symptoms in patients [49], SCID-II interviews and testing took place in the same week. Neuropsychological testing included a short IQ (4 WAIS sub-tests [50]), trail-making, digit-span forwards and backwards, logical memories, and visual reproduction.

An auditory oddball sequence (3 sinusoidal tones, 76 dBspl) was presented over 1600 trials. There was a pseudorandom sequence of standards (800 Hz, 80 ms, 10 ms rise/fall, p = 80%), frequency deviants (600 Hz, p = 10%) and duration deviants (40 ms, p = 10%) with each deviant preceded by at least one standard (stimulus onset asynchrony 850–1050 ms, mean 950 ms). During 4 blocks of 200 trials (passive auditory condition) subjects performed a simple visual red/green circle discrimination on a PC (50:50; subtending 3.8° at 1.5 m, changing at random every 1100 ms). Responses to the green target alternated between hands between blocks. Tone detection is not suppressed during various concurrent visual processes [51], nonetheless the stimulus onsets were controlled so as not to coincide. Four further audio-visual trial-blocks were presented with response to the frequency deviant (active auditory condition). This permitted an analysis of duration-deviant MMN in a state of attention to the auditory modality.

An EEG was recorded (Neuroscan, El Paso) from 31 tin electrodes (Fpz, Fz, Cz, Pz, Oz, FCz, CPz, F3, F7, F4, F8, FT7, FT8, C3, C4, P3, P4, CP3, CP4, T3, T4, T5, T6, TP7, TP8, M1, M2) in an electrode cap (Electro-Cap International: modified 10–20 system) with impedance <5 kΩ. A vertical EOG was recorded from the supra-orbital ridge of the right eye and a horizontal EOG from the outer canthus of the right and left eye to monitor blink-and eye-movements for rejection of artefacts (>50 μV) in EOG leads. Electrodes were referenced to linked-earlobes and the average-reference recomputed offline. A band-pass filter was set at 0.1–100 Hz. Data were digitised with 16-bit resolution, sampled at 500 Hz and stored on a hard disk. Records were epoched separately for each tone type with 100 ms prestimulus baseline and linear detrended. A 30 Hz low-pass filter (24 dB/octave) was used offline.

The MMN waveform was derived by subtracting the ERP to standard tones from those elicited by the duration-deviant. Peaks were sought automatically from 90 to 225 ms after stimulus-onset. The MMN was averaged across blocks for each group and the mean amplitude was computed in successive 30 ms windows from average-refenced data (105–135–165–195–225 ms) for the active and passive condition, respectively. An initial repeated-measures analysis of variance (ANOVA) was carried out for 4 subject groups (between subject factor group) using data from 29 electrodes. The 135–165 and 165–195 ms windows covering the MMN peak (factor window), where inspection showed there to be potential group differences (Figure 1), were analysed by an ANOVA on the data from an array of 12 electrodes. The electrodes were arranged in two within-factors of 3 saggital chains (factor side: i.e. left; F3, FC3, C3, CP3: midline Fz, FCz, Cz, CPz: right; F4, FC4, C4, CP4)) and 4 coronal rows (factor row:frontal; fronto-central; central; centro-parietal). This reduced the degrees of freedom that were also adjusted with the Greenhouse-Geisser epsilon. The MMN in the active condition as well as the latency were analysed by ANOVAs using the same factors (for the latter excluding the factor window). Where significant main effects or interactions involved more than two factor levels, additional ANOVAs were carried out to test simple effects. The influence of antipsychotic medication was examined with ANOVA and Spearman correlations in the EOS group. Exploratory correlations were sought for 3 sets of MMN measures (frontal, mastoid and dipole-moments) with the CGI sum scores, the main clusters of SANS/SAPS symptoms, and neuropsychological measures of processes putatively related to MMN measures (digit-span, trail-making)[52, 53]. Trends are described at P < .05 and type-1 corrected correlations at P < .002.

Brain electrical source analysis (BESA) using a four-shell head-model [54] was used to compute dipoles based on the average-referenced ERP from 20 ms before to 40 ms after the MMN peak. Modelling requires iterative fitting of the dipole location and orientation in a spherical head-model until the difference between the recorded and the calculated surface data is minimised (least square fit [55]). Efficacy was enhanced by seeding with previously published solutions [3, 42, 56]. The goodness of fit is expressed by the residual variance (RV), not explained by the model. Final solutions showed an RV of <2% (Figure 2).

To calculate statistical differences between groups, the individual solutions were re-calculated for each subject. Waveforms were high- (1.5 Hz, 6 db/octave) and low-pass filtered (15 Hz, 12 db/octave). Data were discarded if a criterion of RV<4% was not reached. This reflects the decreased signal-to-noise ratio of an individual's waveform. Two C-EOS, 6 EOS, 3 C-14Y, and 5 S-14Y subjects were removed (Ns = 20, 23, 15, 13 respectively). Dipole-locations were compared between groups with a MANOVA including the 4-level factor dipole as well as the between-subject factor group. The x, y, and z-coordinates of each dipole [57] and the phi and theta orientation angles were compared with post-hoc F-Tests. Group differences of peak dipole-strength and latency were determined by univariate ANOVAs using a 2-level between-subject factor for the younger and older groups and a 4-level within-subject factor dipole (left/right, frontal/temporal lobe loci).

Initial analyses with 29 electrodes showed group differences for MMN amplitude (F_84,2296 = 3.5, P < .001, ε =.16) in the 135–165 ms window (Figure 1). A trend difference continued in the 165–195 ms window (p = 0.1). EOS patients had a smaller MMN than the young controls (F_28,1344 = 6.6, P < .001, ε = .16). The reduction in the older patients vs. their controls attained trend significance (F_28,952 = 1.9, P = .1, ε = .15), but it should be noted that the older controls had a smaller MMN than the younger controls (F_28,1064 = 2.8, P < .02, ε = .18). The MMN amplitudes did not differ between patient groups, and thus provide no evidence for a deterioration nor an improvement between onset (EOS) and later stages of the illness (S-14Y).

The ANOVA using mean amplitudes in two consecutive time windows confirmed the fronto-centrally pronounced MMN topography (factor row: F_3,246 = 56.8, P < .001). MMN amplitudes in the frontal and fronto-central rows did not differ but were larger than in the other rows (Fs_1,82 > 4.8, Ps < .03). Further MMN amplitudes were largest at the midline electrodes (factor side: F_2,164 = 3.3, P = .043). However, the topography did vary with the time window (window × side: F_2,164 = 7.8, P = .001). The MMN was broadly distributed in the 135–165 ms window (no significant differences between the midline, left and right side): but in the 165–195 ms window the midline MMN amplitude was larger than on the left or right side (Fs_1,82>9.7, Ps < .003).

Most importantly, the 4 subject groups differed significantly from each other (F_3,82 = 5.1, P = .003). Simple effects confirmed differences between the EOS group and their adolescent controls (F_1,48 = 15.2, P < .001) as well as between the adolescent and the adult control groups (F_1,38 = 6.0, P = .019). Effects were specific to time window and row (window × row × group: F_9,246 = 2.6, P = .041). The group effect was pronounced in the 135–165 ms interval (group: F_3,82 = 7.2, P < .001) and reduced to a trend in the 165–195 ms window (group: F_3,82 = 2.4, P = .072). At 135–165 ms tests of simple effects revealed differences between EOS patients and their controls at all but the centro-parietal row (Fs_1,48>13.5, Ps < .002). Differences between the control groups were most pronounced at the frontal and fronto-central row (Fs_1,38>7.5, Ps < .011). When compared to their adult controls the S-14Y group showed a smaller MMN amplitude only at the frontal row (F_1,34 = 4.8, P = .036).

ANOVA revealed significant differences in MMN peak latency between groups (F_3,82 = 7.6, P < .001). Especially the S-14Y group showed an increased MMN latency, both, when compared to their adult controls (F_1,34 = 18.7, P < .001) and to the younger patient group (F_1,44 = 12.1, P = .001: Table 3).

An ANOVA comparing MMN amplitudes in 15 EOS patients with and 13 without medication was not significant (F_8,192 = 1.3, P = .27, ε = .23). MMN amplitude tended to decrease more in the second window (165–195 ms) with dose of antipsychotic medication (CPZ equivalents: Fz, r = +.58, P = .03, FCz, r = +.52, P = .06). In other words there was a modest tendency for medication to sharpen the shape of the MMN waveform around its peak latency.

The 60 ms of activity around the MMN peak was well explained by the 4-dipole model [3]. With minor adjustments of orientation and location, the group-solution model showed a RV<1.8% (best fit 0.9%) for each group (Table 4). The left temporal source was close to the border of the superior temporal gyrus with the insula and parietal cortex (BA 22, 29), and the right temporal source lay slightly more rostro-ventrally on the border of the medial and superior temporal gyri (BA 21, 22, 42). The left frontal source lay posteriorly in the anterior cingulate cortex (BA 24). The right frontal source was on the border of the inferior and mid-frontal gyri (BA 10, 44, 47: Figure 2).

Source locations calculated from the individual solutions (similar to the group solutions) differed significantly between the groups (Wilks Λ = .16, F_12,22 = 9.9, P < .001). Potentially reflecting post-adolescent maturation the left temporal source was more lateral (10 mm) and ventral (12 mm) in the older vs. the younger controls (Fs_1,33>19.8, Ps < .001), while on the right it was more medial (9 mm) and ventral (14 mm: Fs_1,33>13.3, Ps < .001). The frontal source in the older controls was more posterior in the left cingulate (12 mm), and on the right it was more anterior (10 mm) and medial (13 mm) vs. the younger controls (Fs_1,33>26.2, Ps < .001).

Compared to the young controls each dipole source in the EOS group was shifted slightly (Λ = .14, F_12,30 = 15.4, P < .001). The left temporal source was more lateral (8 mm), and that on the right was 15 mm more ventral (Fs_1,41 = 14.9, 48.0, P < .001). This is reminiscent of the adult control loci, but 11 mm more rostral. In the EOS patients the left cingulate source lay 15 mm more posterior, and the right frontal source 18 mm more medial than in the C-EOS group (F_1,41 = 50.6, 73.5, P < .001: Figure 3). These values exceeded those in the same direction for healthy adults, and were less lateral (left cingular) and less rostral (right frontal).

Differences in the dipole locations in the S-14Y vs. C-14Y subjects were, like the MMN waveform, less marked than those in the younger groups (Λ = .25, F_12,15 = 3.9, P = .008: Figure 4). The right temporal source was 10 mm more anterolateral and the left cingular source 7 mm more medial than in the older controls (F_1,26>15.6, P < .001). These changes lay on different dimensions from those described between young patients and controls. Differences between older and younger patients that could reflect maturation or illness progression were evident (Λ = .14, F_12,23 = 11.8, P < .001). The left temporal source was markedly more ventral (18 mm) and the right temporal source slightly more lateral (7 mm) in the older patients (F_1,34 = 70.6, 15.1, P < .001). In contrast to the temporal source, the left frontal locus was 13 mm more dorsal and the right frontal locus 10 mm more rostral in the older than the younger patients (F_1,34 = 26.6, 14.5, P < .001: Figure 3).

Dipole moments appeared weaker in patients (F_3,201 = 4.4, P = .008). But ANOVA showed there were interactions with location rather than group. The right frontal source was always the weakest (vs. left cingulum; t₇₀ = 3.1, P = .003). Likewise the slightly later dipole-latencies in patients were not significant (F_9,201 = 1.2, P = 0.34). This is attributable to the variation recorded (SD 19–31 ms). However across groups mean latencies were longer for the right frontal source (173 vs. left frontal 161, left temporal 161, right temporal 163 ms: t₇₀>2.8, P < .05) illustrating a bottom up sequence for activation. Finally although the dipole-orientation for the whole group solutions clearly differed between groups (see frontal and cingulate sources in Figure 2) comparisons of the means of the individual solutions proved variable, and ANOVA showed no significant group differences (see group data in Table 4).

Group differences similar to those in the passive condition, were found with attention to the auditory modality (F_3,77 = 3.9, P = .012). An interaction of group × window indicated that a significant reduction of MMN amplitude in the EOS vs. C-EOS group was delayed to the 165–195 ms window (F_1,43 = 5.2, P = .027). Differences for EOS vs. S-14Y (F_1,39 = 7.1, P = .011) and C-EOS vs. C-15Y (F_1,38 = 9.6, P = .004) were retained for the 135–165 ms window. But comparisons between the older subjects were non-significant.

Fitting the dipole-model from the passive to the active condition, resulted in a good fit for the young subjects (C-EOS, RV 2.1%, best fit 1.4%: EOS, RV 3.5%, best fit 2.1%). Peak-dipole activity in the older subjects also fitted well the 60 ms envelope around the peak (RV 5.1%, best fits 1.2–4.0%).

Correlations were sought between MMN amplitude at Fz and the mastoids with the CGI and the 9 clusters of SANS/SAPS symptom ratings. EOS but not S-14Y patients showed decreasing amplitude with increasing CGI severity at frontal (r = +0.58, P = .002) and mastoid sites (r = -0.04, P = .035). In both patient groups associations of MMN with symptoms were restricted to 3 clusters. In EOS patients reduced MMN at the mastoids related to anergia (r = -0.42 to -0.7, P = .003–.0001) and flat affect (r = -0.43, P = .03). Quite separately in S-14Y patients there were indications that hallucinations related to decreased MMN at frontal sites (r = -0.54, P = .02), while, unexpectedly, increased amplitudes at the mastoids related to anergia ratings (r = +0.5 to +0.6, P = .03-.008).

Correlations for MMN amplitude with z-scores for trail-making measures of set-switching and digit-span measures of working memory were explored. Trails B-A scores were unrelated to MMN amplitude in healthy adolescents, but were negatively related to MMN amplitude at both mastoid sites in both time windows in EOS patients (r = -0.36 to -0.53, p = .07-.005). No correlations were evident in the older groups. There were no relationships to dipole-moments in the young subjects. But an association with the left cingular dipole in healthy adults (r = +0.67, P = .006) is of interest in view of trends in both older groups for an association of trails performance with temporal lobe dipole-moments in the left hemisphere (r = +0.5, P = .06–.08).

Surprisingly, in the C-EOS group digit-span scores correlated negatively with left and right mastoid MMN amplitudes in both time windows (r = -0.50 to-0.61, P = .01–.04). In contrast in EOS patients the correlations tended in the opposite direction at mastoid sites (r = +0.43, P = .03) and FCz (r = -0.4, P = .05). These measures were not correlated in the older subjects. There was no evidence of an association of MMN characteristics with other indicators of short-term memory (logical memories or verbal reproduction). Isolated relationships with dipole-moments could not be distinguished from chance.

In young first-episode patients sources in both the left and right auditory cortices appeared more to the left, and in the right hemisphere more ventral than in healthy teenagers. These are the same directions of change observed for the more caudally located sources in healthy 30- vs. healthy 17-year-olds. In these controls this likely reflects developmental white-matter expansion and grey-matter density increases [60, 61] that can continue in the post-adolescent period [62]. The superficially similar result in EOS patients was unexpected. But this could reflect asymmetric ventricle expansion [63, 64] that can be marked in early-onset patients [65].

In the older patients the left temporal source was more ventral than in controls or either group of young subjects. Indeed, their right temporal source changed in association with illness progression, being more anterolateral than in either the EOS or the C-14Y groups.

The reductions of MMN amplitude (significant), smaller dipole moments in the temporal lobe and slightly longer latencies (both non-significant) in adolescent patients are very similar to alterations reported for first-episode and adult patients [1, 66]. There are reports of a loss of right MMN lateralization and reduced dipole-moments associated with location changes on the left [9, 35, 36]. Such changes are consistent with grey-matter loss in the left superior temporal gyrus, and specific to patients with a first episode of schizophrenia rather than other psychoses [67]. The location changes on the right, reported here, might be associated with the adolescent onset of illness. Grey matter reductions in the right superior temporal gyrus have indeed been reported for early-onset patients [68]. This would be consistent with earlier right hemisphere neurodevelopment [69], with the anomalies remaining 14 years after the adolescent onset.

Frontal MMN sources varied more between groups on the rostrocaudal axis rather than on the lateral axis, as seen for sources in the temporal lobe. In the young patients the right frontal source was more caudo-medial than in their controls, but with similar lateral coordinates to the older controls. The dipoles were more rostrally located in younger and older controls. This more caudal locus in the patients is consistent with data for the frequency-deviant MMN [42] and was tentatively interpreted to reflect delayed development in that part of the forebrain which matures latest [60]. Indeed, prefrontal grey-matter is decreased specifically in adult first-episode patients with schizophrenia, rather than affective psychoses [70]. This could account for sources remaining in a caudal location. The right frontal source of the older patients was located similarly to that in the controls. Yet as this source lay more rostral to that in the EOS patients, it probably reflects normal developmental expansion in the third decade of life.

Possibly also reflecting the illness and its duration the left cingular source in the S-14Y group was more medial than the controls, and more dorsal than in the young patients. But like the younger patients the cingular source was more caudally located than in the controls. It is intriguing that the left cingular source in young patients was more medio-caudally located than in the young healthy controls, but in a similar locus to the older controls. This result has some similarity with the results for the other dipoles (above). But it differs from changes we reported for the detection of frequency-deviants [42], where the patients' dipoles were located rostral to the controls. Thus we re-examined the data with low resolution electromagnetic tomography (LORETA [71]). This confirmed a rather rostrally located bilateral activation of the medial frontal gyri, albeit at the expense of right frontal activity, precisely in the two analyses showing the most rostral BESA solutions (namely the frequency-deviant source in the older patients and the duration-deviant source in the controls). While the degree of blending of activated regions in the LORETA solution may be accounted for by the smoothing effect of the underlying algorhythm, it remains difficult to account for the different directions of source changes seen in patients with the two types of deviant stimulus.

Studies of normal subjects demonstrate that temporal lobe contributions to MMN represent the short-term memory templates for standard tones against which deviants can be contrasted. Frontal sources, active after deviant detection, initiate a switch to controlled processing to allow the significance for response to be assessed [6, 7]. The cingulate source, active simultaneously with those in the auditory cortices, may be involved in monitoring the ongoing situation prior to the later engagement of the inferior frontal regions when discord is detected. This is similar to the inferences from studies of the processing of incongruent stimuli [72]. Generically, the function is similar to the mismatch of expectations registered by the error-related negativity that also has BESA-calculated dipole-sources in the cingulate cortex [73].

Trail-making abilities to switch between sets are associated with prefrontal function [74]. Exploratory analyses indicated that in the more severely ill young patients a reduced MMN from the mastoids was associated with poor trail-making scores. This is taken as broadly supporting the attribution of the ability to switch between channels of information processing to one of the functions represented in MMN. (We note there is considerable reason to doubt whether MMN recorded at the mastoids uniquely reflects the output of supratemporal generators, discussion in [75].)

Imaging studies show that digit-span measures of working memory rely on recruiting activity in a network that includes the right dorsolateral prefrontal and anterior cingulate cortices [76]. Despite one negative report for healthy subjects [77] correlations with MMN were anticipated (see introduction). Superficially the correlations for improved digit-span measures with decreased mastoid measures of MMN in healthy subjects and vice versa in patients are the opposite of what would be expected if one predicts an association with MMN's mnemonic role. However, if the main function of deviant detection by MMN (at the mastoid) is the facility for switching (previous paragraph), then in the context of remembering a series of digits, longer spans with fewer errors would be more likely in those showing a less marked MMN. The association of longer spans with larger MMN in patients could reflect either the impaired switching facility or a relatively improved mnemonic role for MMN, perhaps reflecting the function of the supratemporal generator. This interpretation accords with the negative association reported for verbal memory performance with MR-activity in the right prefrontal region just posterior to the dipole source discussed here [78].

It is remarkable that the associations for SANS/SAPS symptom-clusters with frontal and mastoid MMN were largely restricted to negative symptoms of anergia and positive hallucinations. The former was associated with reduced mastoid MMN in the EOS group, as reported elsewhere for duration-deviant MMN in the right hemisphere [1]. In the S-14Y patients hallucination ratings were related negatively to measures at frontal electrodes, also reported previously for left hemisphere MMN recordings [36]. This is consistent with the attribution of these symptoms to impaired functional connectivity between the temporal and frontal lobes [79, 80]. The prominent role for negative symptoms in the EOS group may reflect their prevalence and significance in the development of psychosis in adolescence, when the earlier onset has a poorer prognosis [81] and patients frequently show reduced frontal volumes [82]. However, we cannot exclude that this role would apply to chronic patients with prominent negative symptoms considering a report that MMN deficits in patients ill for 18 years accounted for 42% of the variance in general assessments of function [83]. Thus, here a positive association with MMN for the negative symptoms of anergia or apathy in the S-14Y group was unexpected. Speculatively this could be a sign of the sort of strategy (withdrawal) acquired by patients who have had to cope (relatively succesfully) with this illness for many years. This requires further investigation.

Sources of MMN activity in the frontal, cingulate and temporal cortices have been proposed without detailed localization [1, 2, 84]. Details were restricted to MEG studies of the auditory cortices [33] until the report from Jemel et al. [3]. The reliability of such models depends on the signal-to-noise ratio. A reasonable effect size requires data for 6 electrodes per dipole from a thousand trials [85, 86]. This was achieved here for the group solution (1174–1792 trials), where the stability (RV<2%) was assisted by being hypothesis led [3, 42]. For the individual solutions, that provided a measure of variance between subjects, the data from 80% of the subjects still explained 96% of the variance in this model. Accuracy was facilitated by the digitization of recording sites with respect to individual skull markers. Further, we checked the stability of the group solution over successive millimetres away from the point solution [87] and found they were usually stable over 2–6 mm along each dimension. This order of magnitude for activated tissue is physiologically more plausible than the calculated point solution. Another estimate of the volume activated, or of the error in calculating its centroid is provided by the variability between subjects. Despite the poor signal-to-noise ratio this was usually well below a centimetre (Table 3). This upper limit for the model's accuracy matches that obtained from known modelled sources [88, 89]. But, in addition to the problems of inter-individual stability of the model, error is introduced in the placing of sources on the brain atlas. Discrepancies between MRI anatomical representations of the locus of an individual's source and its localization on an averaged template can be marked. One study of this issue reported differences in nearly two thirds of the cases examined [90]. We have partly corrected for this by representing sources on the anatomy of one of our subjects (figure 4). But without constructing an average brain image from the participants, we have not accounted for the anatomical variability between our subjects.