Looking Under the Hood: Using Time-Frequency Analysis to Bridge the Gap Between Event-Related Potential Biomarkers of Psychosis and Underlying Neural Circuitry

Reduction in mismatch negativity (MMN) amplitude is among the more consistently observed biomarkers of schizophrenia. The MMN is an event-related potential (ERP) that is elicited when a sequence of identical auditory stimuli is infrequently interrupted by a stimulus that deviates from the standard along one or more dimensions (e.g., pitch, duration, intensity). This automatic change detection response, which is calculated as the difference between the auditory ERPs evoked by standard and deviant stimuli (see Figure 1A, B), is often conceptualized as a prediction error signal reflecting the integrity of sensory predictive coding processes (1). Robust reductions in MMN amplitude are reliably observed in people with schizophrenia; moreover, MMN reduction prior to the onset of psychosis among clinical high risk (CHR) individuals appears to distinguish those who are at imminent risk for experiencing above-threshold psychotic symptoms (i.e., hallucinations and delusions) from CHR individuals who do not convert to schizophrenia [for a review and meta-analysis, see (2)]. Altogether, such findings are consistent with the predictive coding theory of psychosis framework intended to account for the experience of hallucinations and delusions that are characteristic of the disorder (3).

Figure 1.

Schematic drawing of the event-related potential (ERP) components that can be elicited during a mismatch negativity (MMN) paradigm, as well as time-frequency metrics that can be extracted. (A) ERP in response to standard (gray) and deviant (purple) tones. (B) The MMN, which is the (deviant − standard) difference wave. (C) Example of a theta oscillation with low (0.5 dB) amplitude (gray) and a theta oscillation with a large (1.0 dB) amplitude (purple) with intertrial coherence (ITC) = 1.0. (D) Example of 2 trials with identical amplitude and high ITC (ITC = 0.9). (E) Example of 2 trials with identical amplitude and low ITC (ITC = 0.1).

Given that the predictive validity of the MMN as a marker of psychosis risk is reasonably strong compared with many other candidate biomarkers, a sensible next question concerns the underlying oscillatory and non-oscillatory neural dynamics that give rise to this signal. On this issue, the MMN itself can reveal relatively little; this is because the temporal averaging process required to isolate the MMN from the background electroencephalogram obscures neural activity that is not consistently time locked to the onset of the auditory stimuli. In contrast, time-frequency analytical approaches such as wavelet-based analysis permit examination of the underlying neural oscillations by calculating the amplitude and phase of individual frequencies on a trial-by-trial basis before temporal averaging occurs. The resulting signal is now no longer a 1-dimensional array of microvolt values as a function of time (as in the case of the MMN) but rather a 2-dimensional array in which the average magnitude (or alternately, phase coherence across trials) is estimated for each frequency and at each time point. In doing so, the amplitude (Figure 1C) and coherence (Figure 1D, E) of canonical frequency bands such as theta (4–7 Hz), alpha (8–13 Hz), beta (14–30 Hz), and gamma (31–50 Hz) in response to auditory prediction violations can be examined to understand how neural dynamics support this predictive coding phenomenon in health and in serious mental illness.

In a study recently published in Biological Psychiatry: Global Open Science, Hua et al. (4) examined the characteristics of theta (amplitude and phase coherence) in response to standard versus deviant stimuli in a large sample of CHR individuals from NAPLS-2 (North American Prodrome Longitudinal Study 2). This study included 77 CHR individuals who later converted to psychosis (CHR-C), 238 CHR individuals who did not convert to psychosis within a 24-month follow-up period (CHR-NC), and 241 healthy comparison (HC) individuals. Previously, these authors published a report that MMN amplitude at baseline significantly predicted the emergence of a psychotic disorder in this sample of CHR individuals (5), replicating and extending similar findings from other groups [see (2) for a review]. In the current study, the authors extend this work to examine the role of theta oscillations evoked by standard and deviant stimuli: If theta disruption can account for MMN impairments in schizophrenia and individuals at elevated risk for the disorder as others have suggested [e.g., (6)], the neural processes that give rise to this biomarker of psychosis risk can be more readily elucidated. However, it must first be established that 1) like MMN, reductions in theta amplitude and coherence can be observed in CHR samples, and 2) there is significant shared variance between theta metrics and MMN amplitude with respect to its ability to predict conversion to psychosis. Such findings would be foundational for future work attempting to understand how the neural circuitry that produces MMN impairment also confers a greater risk for conversion to psychosis among individuals at elevated clinical risk.

The authors made several important observations. First, they found that theta amplitude (or power) and intertrial coherence (ITC) were greater for deviant compared with standard stimuli, replicating previous work showing that these signals exhibit a mismatch response pattern similar to the MMN. Second, ITC was reduced among CHR-C individuals compared with HC and CHR-NC individuals, whereas power was only reduced in CHR-C individuals compared with HC individuals across both stimulus types. Importantly, there was no group × stimulus type interaction observed for either measure, indicating that the magnitude of the theta mismatch response was just as large among CHR-C individuals as it was among HC and CHR-NC individuals, despite having lower power and ITC overall. This represents a noteworthy departure from the pattern observed in MMN; whereas MMN may be thought of as a prediction error signal that can be used to index altered predictive coding processes that are hypothesized to give rise to hallucinations and delusions, the theta abnormalities observed in CHR-C individuals appear to be somewhat independent of this deviance detection process. Consistent with this interpretation, the authors found that MMN amplitude was more strongly correlated with the average power and ITC across stimulus types rather than with the deviant − standard difference score. This is further underscored by the fact that group differences in MMN magnitude became nonsignificant after accounting for average power and ITC but not the difference in power and ITC. When the authors examined the association between theta metrics and time to conversion among CHR-C individuals, they found that smaller average ITC (but not power) across stimulus types significantly predicted a shorter time to conversion, even after controlling for the degree of unusual thought content and paranoia.

The current work yields critical insights regarding alterations in the neural circuitry that increase risk for psychosis among CHR individuals. If a meaningful portion of the variance in the MMN biomarker and its capacity to predict conversion can be explained by deficient theta responses to any auditory input among CHR-C individuals, it follows that the local circuit interactions responsible for theta modulation constitute a critical biological pathway by which the disorder emerges. On this question, a growing body of work implicates the somatostatin-expressing class of GABAergic (gamma-aminobutyric acidergic) inhibitory neurons in the production and modulation of theta oscillations [see (7) for a review]. Unlike the parvalbumin-containing class of GABAergic inhibitory neurons, which are implicated in higher-frequency (e.g., 40 Hz) cortical oscillations, the physiology of somatostatin-expressing interneurons is such that they exert influence over longer windows of integration and are accordingly associated with lower-frequency oscillations including theta (7). If expression of this class of interneurons is reduced in schizophrenia as has been reported in postmortem studies (8), this may explain why stimulus-evoked theta modulation is disrupted in schizophrenia and individuals at elevated clinical risk. In a similar vein, somatostatin-expressing interneuron disruption may also account for the consistently observed finding that ketamine—a non-competitive NMDA receptor antagonist that blocks glutamatergic action on interneurons—reduces MMN amplitude in psychiatrically healthy individuals (9). Altogether, this work lends converging support to the hypothesis that disruption in this cortical circuitry both produces deficient theta modulation in response to sensory input and confers risk for developing psychosis.

One important caveat to this framework is that the results from the study by Hua et al. (4) do not suggest that individual differences in theta modulation can fully, or even mostly, account for individual differences in MMN amplitude; average ITC and power only accounted for 28% and 14% of the variance in MMN, respectively. These observations are consistent with intracranial studies indicating that other frequency bands, including beta and gamma, are implicated in the auditory novelty detection response (10) [see (6) for a review]. Thus, the suggestion is not that theta oscillations can fully account for MMN amplitude but rather that disrupted theta oscillations in CHR individuals can account for the relevant group differences in MMN magnitude that directly relate to risk for psychosis. Future work may expand upon these observations to directly test the predictive value of oscillations evoked by nonspecific auditory stimulation compared with oscillations elicited as part of the deviance detection process itself.

The study by Hua et al. (4) highlights the utility of adding time-frequency decomposition to supplement traditional ERP analyses. In applying this approach to a mismatch negativity paradigm, the authors revealed distinct effects of theta amplitude and intertrial coherence that establish a mechanistic link between neural circuit abnormalities and a key biomarker of psychosis risk. Such work is not only essential for advancing our capacity to predict who will go on to experience psychosis but also points to neurophysiological targets that may be engaged in treatment and interventions aimed at illness prevention.