We begin with a description of the general ERP patterns observed in the data of our 15 control subjects. shows data from group-averaged waveforms taken from an anterior and a posterior electrode. The timing and scalp topographies of the components identified here are highly consistent with those found in the literature.19
We were able to reliably identify components as early as the Po
, which peaked at approximately 15 ms poststimulus over the frontal, frontocentral and central scalp sites. The next component we saw clearly was the P20
, which peaked posteriorly at about 21 ms. The Pa
was present at about 38 ms over frontal and frontocentral sites. The P1 peaked at about 46 ms at frontocentral and central sites. The N1 peaked at about 95 ms over the central scalp.
Fig. 2: Early components of the auditory evoked potential in a group of healthy control subjects (n = 15). Waveforms are derived from group averages taken from a representative anterior site (grey waveform) and a posterior electrode (black waveform). (more ...)
We compared the electrophysiological responses of healthy control subjects to those of our patient group (). As in the control subjects, the classical AEP componentry was evident and the earliest peaks (P0, P20 and Pa) all had identical latencies to those of the control subjects. Very slight delays in peak latency were noted for the P1, which peaked at approximately 50 ms frontocentrally and centrally. The N1 peaked over central scalp sites at about 95 ms. By fully describing and characterizing middle latency components of the AEP over the entire scalp surface of control subjects and schizophrenia patients, we found evidence for the earliest point of auditory sensory dysfunction in schizophrenia from time points as early as the first identifiable MLR component (P0).
Fig. 3: Auditory evoked potentials in control subjects (n = 15) are blue and in patients (n = 21) are red (interstimulus interval = 500 ms). Data from 10 electrodes spanning anterior to posterior scalp sites are presented.
As above, the P0 had a peak latency of approximately 15 ms in both groups and was evident at both the frontal and parietal scalp sites. The ANOVA revealed a significant main effect of group (F1,34 = 35.65, p = 0.031), reflecting the fact that the amplitude of P0 was significantly attenuated for patients. No other main effect or interaction was significant (for a summary of ANOVA results, see ). Topographic mapping showed a similar parietal distribution for the P0 in both groups (, row A). A subtraction map (far right panel) shows the map of the difference between control subjects and patients, indicating a largely similar distribution to that of the base P0 map. The exploratory statistical cluster plots () indicated that this difference began very early, and assessment of the running t tests showed an onset at just 8 ms. We conducted a posthoc ANOVA over the 8–12 ms window to better characterize this effect. A main effect of group was significant (F1,34 = 8.99, p = 0.005).
Fig. 4: Voltage scalp topographies from the control and patient group-averaged waveforms, for the Po, P20, Pa, and P1, and the corresponding difference between the control subject and the patient.
Fig. 5: The Np65: A pronounced negative-going deflection over the posterior scalp, peaking at 65 ms. This is shown for patients (red) and control subjects (blue), from right and left occipital sites where the response was maximal.
In our data, the P20 was represented by a positivity that peaked at about 21 ms posteriorly and centrally in both groups. The ANOVA revealed a significant main effect of group (F1,34 = 7.47, p = 0.01). It also revealed a significant effect of hemisphere (F1,34 = 6.43, p = 0.016), attributable to greater amplitudes at this latency over the left hemisphere for both control subjects and patients. As with the P0, the topographic maps (, row B) show a strong parietal positivity, with a concomitant frontopolar negativity for control subjects. The P20 distribution for patients was very similar, with a parietal positivity. The frontopolar negativity was substantially weaker. This was particularly evident in the subtraction map, where the largest difference between groups was concentrated over the frontopolar and inferior-frontal scalp. The subtraction map also suggested that a lateral inferotemporal source might be implicated during this timeframe.
The Pa had a peak latency of approximately 38 ms over anterior scalp sites for both patients and control subjects. An ANOVA for this component revealed a significant main effect of group (F1,34 = 12.16, p = 0.001), again reflecting substantial attenuation of Pa amplitude in patients. Topographic mapping showed a frontocentral positivity for both groups, broadly distributed across both hemispheres, with a concomitant occipital negativity (, row C). The subtraction map had a clearly dissociable distribution from that of the base Pa maps. It revealed a pair of bilateral positive foci that were more lateralized and concentrated more posteriorly over the central scalp. The fact that this difference map is dissociable from the base Pa map indicates the multiple generator configurations that must underlie the activity pattern during this processing period. The maps suggest that the Pa reflects both lateral activity as well as contributions from more fronto–central generators. It also suggests that the fronto–central aspect of the response is more preserved in patients, whereas the bulk of the deficit for this group is in the more lateral responses.
As with the Pa, the P1 component likely reflects the activity of multiple simultaneously active generators. Over frontocentral regions, the P1 had a peak latency of about 46 ms for control subjects and patients. A robust main effect of group was found (F1,33 = 6.68, p = 0.014). The topographies of the P1 for control subjects showed a central positivity with a concomitant occipital negativity, whereas the patients' positivity was distinctly more frontal and had a similar occipital negativity (, row D). Here, the subtraction map was distinctly different from either base P1 map, again suggesting contributions from more posterior generators.
In our investigation of the earliest components that reveal differences between patients and control subjects, we used the data to make estimates of power (). We hypothesized that there would be significant attenuation of AEPs in patients, compared with control subjects, from time points as early as the first recognizable component (the P0). However, we anticipated the effect size for this difference to be quite small. Indeed the power analysis revealed the lowest level of power for the P0; nevertheless, this low was a healthy 75%. While this is slightly below 80%, we are confident in this medium-sized effect, with highly significant differences in the amplitude of the P0 of patients and control subjects.
To better detect effects between patients and control subjects in this large data set, we computed a statistical cluster plot (see Methods). This served as a hypothesis-generating tool; it provided us with a snapshot of where and when significant differences were occurring between groups. Thus, we observed the clear and distinct clusters corresponding in latency to known auditory components (P0, P20, Pa, P1, N1, N2) (). This reveals differences at the earliest discernible component, which appear to propagate up the auditory processing stream. By and large, the statistical cluster plots correspond to the effects uncovered by our ANOVAs, for which we had a priori hypotheses. In addition, the plot revealed an unpredicted difference at about 65 ms, which we have termed the Np65; this component is discussed below. Substantial amplitude differences were observed at the latency of the N2, as we would expect, given that patient deficits at later latencies follow the earlier deficits, which are the focus of our study.
Fig. 6: Statistical cluster plot of the results of the point-wise running t tests comparing the amplitudes of the patient and control subject auditory evoked potentials. Time, with respect to stimulus onset, is presented on the x axis and topographic (more ...)
A previously uncharacterized negative-going component was discovered in the course of this investigation. Appearing posteriorly at about 65 ms, it was very pronounced in our control subjects. Here, we refer to this component as the Np65. This component was highly attenuated in patients. The amplitude of the Np65 was measured within a 6-ms window centred on its peak latency of 62 ms at 3 electrode sites on the left and 3 on the right. An ANOVA revealed a robust effect of group (F1, 34 = 11.25, p = 0.002) (). The component is clearly dissociable from the known AEPs.
Results of dipole source analysis I: subcortical–cortical distinction
We used the BESA genetic algorithm dipole analysis module to determine the location of the generators of the middle latency components. A symmetrically constrained pair of dipoles was allowed to freely fit for a 6-ms time window around the first peak, the P0 (12–18 ms). The resulting model, explaining 95.8% of the variance, placed sources bilaterally in the thalamus (Talairach coordinates x = 13, y = –7, z = 7). Given previous findings of major contributions from the primary auditory cortex to the middle latency potentials (see Discussion), we constrained a pair of dipoles to the primary auditory cortices bilaterally (46, –24, 12) to test whether such a model could also satisfactorily explain P0 activity. This model explained only 70.3% of the variance. Further, when the dipoles were allowed to freely fit from this starting point in the primary auditory cortex, they returned to the thalamus (, panel A).
Fig. 7: Dipole source analysis I. Panel A: source analysis of the P0 in control subjects and patients reliably models the generators bilaterally in the thalamus; this 2-dipole model explains 95.8% of the variance for control subjects and 87.6% for patients (more ...)
In the next step, we fixed the first dipole pair in their original thalamic locations and then a symmetrically constrained second dipole pair was fit to a 6-ms time window around the next discernible peak, the P20 (18–24 ms). Bilateral sources were modelled in the insula (45, 10, 15), with 98.2% of the variance explained. These sources fell approximately 10 mm anterior to the primary auditory cortex. As a test, dipoles were fixed in the core of primary auditory cortices bilaterally (46, –24, 12); this model explained 96.8% of the variance.
A third symmetrically constrained dipole pair was allowed to fit to a 12-ms time window around the next discernible peak, the Pa
(32–44 ms). This model placed bilateral sources in the region of the inferior parietal lobule (IPL; 43, –28, 25) and explained 99.2% of the variance across the 32–44 ms epoch. Given that it was only explaining 1.0% more of the variance than the 4-dipole model, these sources likely reflect subtle contributions from posterior regions. Previous electrophysiological investigations49,50
have shown Heschl's gyrus to be the generator of the Pa
. Once more, we fixed this dipole pair in the primary auditory cortex (46, –24, 12) and found that it still explained 94.3% of the variance for this epoch.
When the epoch around the P1 was selected (50–72 ms), the 6-dipole solution already explained 98.7% of the variance, rendering further dipoles uninformative. For the entire epoch of interest, encompassing the P0, P20, Pa and P1 (0–72 ms), the model explained 98.6% of the variance. For the epoch 0–50 ms, which excludes the P1 contribution, 98.3% of the variance continued to be explained.
Next, we applied this model to our patient data (, panel B). For the 0–72 ms epoch, the model explained 97.8% of the variance. Excluding the P1, the model explained 98% of the variance for the epoch 0–50 ms.
Results of dipole source analysis II: exploration of dorsal-ventral dissociation
To explore where the differences between patients and control subjects are localized, we employed the following strategy: first, we separately source localized the P1 for control subjects and patients to provide us with a metric of 1 well-known auditory component, the major generators of which are known to be in and around the core region of the auditory cortices10,51–54
(, panel A). We placed a pair of symmetric dipoles in the primary auditory cortex (42, –28, 12); the model explained 92.2% of the variance. For the patients, this model explained 94.0% of the variance.
Fig. 8: Source analysis II. Panel A: source analysis of the P1 for both patients and control subjects reveals remarkably similar localizations of the generators for both groups. Panel B: the green and purple dipoles demarcate the modelled source of the (more ...)
Next, we sought to localize the generator of one early significant difference between the AEP of patients and control subjects — the Pa (, panel B). Using a difference wave (control subjects minus patients) for this model, we freely fit a pair of symmetric dipoles to the 25–45 ms time frame to determine where the generators for the early difference were localized, relative to those of the P1. By comparing this with the P1 source analysis, we hypothesized that we would have an indicator as to whether the differences between patients and control subjects were more apparent in areas falling in the dorsal or the ventral auditory pathway. With 88.6% of the variance explained, we found bilateral stable generators that were distinctly superior and posterior to those of the P1 for control subjects and patients (47, –18, 24), a trajectory that we would expect to see for components in the dorsal stream. This suggests that these earliest differences can be explained by a trend that seems to favour the dorsal stream. Our first dipole source model (Dipole Source Analysis I) corroborates this finding, in that model bilateral sources for the Pa were localized exactly 13 mm posterior and superior to the primary auditory cortex in the IPL.
When these source localization results for differences in patients and control subjects were contrasted with the same for the ERPs, it was clear that, in both, a dorsal–ventral distinction could be observed. In the average waveforms, there were clearly observable differences between anterior and posterior (corresponding to ventral and dorsal auditory pathways, respectively) AEPs generated in patients versus control subjects. For example, the difference between patients' and subjects' P1 at an anterior site (F2) failed to reach statistical significance (p = 0.270): here, patients' P1 (0.753, SD 0.705) was 25% smaller than that of control subjects (1.008, SD 0.649). However, the difference in P1 amplitude at a posterior site (P3) was significant (p = 0.000): the P1 of patients (–0.003, SD 0.367) was 87% smaller than that of control subjects (0.346, SD 0.452) ().