Aberrant hierarchical prediction errors are associated with transition to psychosis: A computational single-trial analysis of the mismatch negativity

Daniel J. Hauke, Colleen E. Charlton, André Schmidt, John Griffiths, Scott W. Woods, Judith M. Ford, Vinod H. Srihari, Volker Roth, Andreea O. Diaconescu, Daniel H. Mathalon
doi: https://doi.org/10.1101/2022.12.20.22283712

Abstract

Background Mismatch negativity (MMN) reductions are among the most reliable biomarkers for schizophrenia and have been associated with increased risk for conversion to psychosis in individuals at clinical high risk for psychosis (CHR-P). Here, we adopt a computational approach to develop a mechanistic model of MMN reductions in CHR-P individuals and patients early in the course of schizophrenia (ESZ).

Methods Electroencephalography (EEG) was recorded in 38 CHR-P individuals (15 converters), 19 ESZ patients (≤5 years), and 44 healthy controls (HC) during three different auditory oddball MMN paradigms including 10% duration-, frequency-, or double-deviants, respectively. We modelled sensory learning with the hierarchical Gaussian filter and extracted precision-weighted prediction error trajectories from the model to assess how the expression of hierarchical prediction errors modulated EEG amplitudes over sensor space and time.

Results Both low-level sensory and high-level volatility precision-weighted prediction errors were altered in CHR-P and ESZ groups compared to HC. Furthermore, low-level precision-weighted prediction errors were significantly different in CHR-P that later converted to psychosis compared to non-converters.

Conclusions Our results implicate altered processing of hierarchical prediction errors as a computational mechanism in early psychosis consistent with predictive coding accounts of psychosis. This computational model appears to capture pathophysiological mechanisms relevant to early psychosis and the risk for future psychosis in CHR-P individuals, and may serve as a predictive biomarker and mechanistic target for novel treatment development.

Introduction

Often without our awareness, our brain continuously learns about the environment that surrounds us. The mismatch negativity (MMN) is a neurophysiological index of such implicit learning commonly measured with electroencephalography (EEG). It refers to a brain response that is elicited automatically when an auditory stimulus violates a statistical regularity in the recent auditory environment,1 for example when a series of low tones is unexpectedly interrupted by a high tone (Figure 1A). Formally, the MMN is a transient negative wave deflection in the event-related potential (ERP) elicited by infrequent auditory deviant stimuli randomly interspersed among frequent standard stimuli that is most easily identified between 100-250ms following stimulus onset in the deviant-standard ERP difference wave (Figure 1B). 2,3 Importantly, this neurophysiological assessment is very feasible in a wide range of clinical settings.

Figure 1.
Figure 1. Mismatch negativity.

A Example stimulus sequence to elicit auditory mismatch negativity. Violation of statistical regularity (established through repetition of low tones) elicits mismatch negativity. B Mismatch negativity is typically identified in the ERP difference wave obtained by subtracting response to the frequent and therefore more predictable tone (standard) from the response to the infrequent unpredictable tone (deviant).

MMN amplitude reductions have been replicated in numerous studies of patients with schizophrenia.4 A number of pharmacological challenge studies investigating the neuroreceptor basis of the MMN showed N-methyl-D-aspartate receptor (NMDAR) antagonists (e.g., phencyclidine or ketamine) to reduce MMN amplitude in rodents,5 monkeys,6 and humans7–9 (see 10 for an overview). Together, these results implicate NMDAR dysfunction as a pathophysiological mechanism underlying MMN amplitude reduction and a wide range of clinical symptoms and cognitive impairments in schizophrenia.11–17

Interest in MMN amplitude reductions has increased in recent years as a possible early warning sign for impending psychosis. MMN reductions are already present in individuals at clinical high-risk for psychosis (CHR-P), likely reflecting vulnerability for a progression to full psychosis as MMN amplitude reductions were found to be more pronounced in CHR-P individuals who later converted to a psychotic disorder.18–22 Despite its clinical potential, the mechanisms that account for these MMN alterations in the CHR-P population remain poorly understood.

One of the biggest challenges in early detection and intervention research lies in development of novel medications to delay or even prevent the transition to psychosis in CHR-P individuals.23 This challenge has been attributed to a lack of mechanistic models of pathophysiological processes, especially in the CHR-P population.23 In a previously published study, we found MMN amplitudes in ESZ patients and CHR-P individuals to be reduced,21 and further showed MMN deficits to be greater in those CHR-P individuals who subsequently converted to psychosis, relative to non-converters followed-up for at least 12 months. Here, we apply a computational approach24 to the EEG data from this prior study to develop a mechanistic model of altered information processing as a basis for MMN amplitude reductions in CHR-P individuals and ESZ patients.

Methods and Materials

Participants

Participants included 19 early-illness schizophrenia patients (ESZ; ≤5 years since initial hospitalization or initiation of antipsychotic medication), 38 CHR-P, and 44 healthy controls (HC).21 Clinical outcomes for the 38 CHR-P participants were tracked over 24 months, resulting in 15 CHR-P who converted to full psychosis and 16 CHR-P who did not convert but were followed for at least 12 months. Seven CHR-P dropped out before the 12-month follow-up and were, therefore, excluded from analyses comparing CHR-P converters and non-converters, since their clinical outcomes were less certain. ESZ participants were referred from the Yale Specialized Treatment Early in Psychosis (STEP) Clinic or from community clinicians. CHR-P were recruited from the Yale Psychosis Prodrome Research (PRIME) Clinic and HC through advertisements and word-of-mouth. The study was approved by the Institutional Review Board of Yale University and all adult participants provided written informed consent. For minors, parents provided written informed consent and minors provided written assent.

Inclusion and exclusion criteria

ESZ patients met DSM-IV criteria for schizophrenia based on a Structured Clinical Interview for DSM-IV25 administered by a trained research assistant. CHR-P participants met the Criteria of Psychosis-Risk Syndromes based on the Structured Interview for Psychosis-Risk Syndromes (SIPS).26,27 Potential participants for all groups were excluded from the study if they fulfilled the following criteria: substance dependence or abuse within the past year, a history of significant medical or neurological illness or a head injury resulting in loss of consciousness, and abnormal audiometric testing. Additionally, HC who met criteria for any past or current DSM-IV Axis I disorder or had a first-degree relative with a psychotic disorder were excluded.

Task

Participants performed an unrelated primary task (silently reading a book) while presented with three different auditory oddball paradigms presented in fixed order. Each paradigm comprised two runs of 875 tones each (1750 tones in total), including 90% standard tones (50ms, 633 Hz) and either 10% duration (100ms), 10% frequency (1000 Hz), or 10% duration + frequency double deviants (100ms and 1000 Hz). All tones were presented at 78 dB in fixed pseudorandomised order with 5ms rise/fall times and 510ms stimulus onset asynchrony through Etymotic ER3-A insert earphones (Etymotic Research, Inc., Elk Grove Village, Illinois).

EEG data processing

EEG was recorded using a 20-channel electrode cap with a standard 10-20 montage (Physiometrix, Inc., North Billerica, Massachusetts) and additional mastoid and nose electrodes with linked-ear reference and an FPz ground. Signals were digitised at 1000 Hz with a Neuroscan Synamps amplifier (Neuroscan, Herndon, Virginia). Electro-oculograms were recorded from electrodes located above and below the left eye and at the outer canthi of both eyes.

EEG preprocessing consisted of high-pass filtering with a Butterworth filter (0.5 Hz), downsampling (256 Hz), low-pass filtering using a Butterworth filter (30 Hz), followed by epoching into 500ms segments around tone onsets (−100 to 400ms), baseline correction (−100 to 0ms), and eyeblink correction using principal component analysis with one component. Eyeblink components of all participants were manually inspected and eyeblink detection thresholds adjusted, if necessary, followed by rejection of remaining artefactual trials (using a ±100 μV amplitude threshold). Preprocessing and statistical analyses were implemented in MATLAB (version: 2020b; https://mathworks.com) using the SPM12 toolbox (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/).

Computational Modelling

We modelled implicit sensory learning about the tone sequences using a 3-level binary Hierarchical Gaussian Filter (HGF).28,29 This model assumes that participants make inferences about a number of hidden environmental states (Figure 2, left). In the context of the oddball paradigm, based on each trial input (standard or deviant tones), a participant needs to infer three hidden states, structured as follows: The lowest level state corresponds to the tone probability. On each trial k a tone can either be deviant Embedded Image or a standard tone Embedded Image. This state can be described by a Bernoulli distribution that is linked to the state at the second level x2(k) through the unit sigmoid transformation:

Figure 2.
Figure 2. Computational analysis pipeline.

Trial-by trial trajectories of low- and high-level precision-weighted prediction errors were computed using the Hierarchical Gaussian Filter28,29 (left). In a first level analysis, precision-weighted prediction errors were used as parametric regressors to explain EEG amplitude variations at each point in sensor space and peristimulus time (PST) following the tone presentation across trials within each participant (middle). First level statistics were carried to the second level to obtain statistical parametric maps over 2D sensor space and peristimulus time (right). EEG: Electroencephalography.

Embedded Image with Embedded Image

Embedded Image represents the unbounded tendency towards standard or deviant tones (−∞, +∞) or the tone tendency and is specified by a normal distribution: Embedded Image

The state at the third level Embedded Image expresses the (log) volatility of the environment over time and is also modelled using a normal distribution: Embedded Image

Participants’ beliefs about these hidden states at level i of the hierarchy and on trial k are denoted with Embedded Image and updated after each new tone according to the following update equation: Embedded Image where Embedded Image is the expectation or belief at trial k and level i of the hierarchy, Embedded Image is the precision (inverse of the variance) from the level below (the hat symbol denotes that this precision has not been updated yet and is associated with the prediction before hearing a new tone), Embedded Image is the updated precision at the current level, and Embedded Image is a prediction error (PE) expressing the discrepancy between the expected and the experienced outcome.

In line with a previous study examining the effects of ketamine on sensory learning in a roving paradigm,24 we focused our analysis on low-level precision-weighted PEs about the tone tendency (ϵ2) and high-level precision-weighted PEs about the volatility of the environment (ϵ3), where the precision-weighted PE Embedded Image on each trial k and at level i of the hierarchy is defined as (cf. Eq. 5): Embedded Image

We implemented this model using the ‘tapas_ehgf_binary’ function from the HGF toolbox (version 6.0), which is made available as open-source code as part of the TAPAS30 software collection (version: 5.1.0; https://github.com/translationalneuromodeling/tapas/releases/tag/v5.1.0) in MATLAB (version: 2020b; https://mathworks.com). We used this recently developed enhanced version of the HGF to improve sensitivity to learning about environmental volatility. The main distinction with respect to earlier versions of the HGF is that the posterior means Embedded Image are updated before the precisions Embedded Image at level i of the hierarchy. For more details on the update equations, see 28,29 for the original HGF and the ‘tapas_ehgf_binary’ function for the eHGF.

We optimised the parameters of the perceptual model assuming that each participant acted as an ideal Bayesian observer minimising the surprise for a given input sequence using the ‘tapas_bayes_optimal_binary’ function. Ideally model parameters should be estimated based on both sensory input and participants’ behavioural responses to estimate how participants’ learning deviates from an ideal Bayesian observer. However, this was not possible, because the MMN paradigm is a passive task that does not require participants to make responses.

The prior settings (mean, variance) for this optimisation were (−3, 4) for the evolution rate ω2 and (2, 4) for meta-volatility ω3. The coupling strength κ2 was fixed to log(1). Posterior parameter estimates are summarised in Table 1.

View this table:
Table 1. Summary of posterior parameter estimates.

Statistical analyses

Demographic and clinical variables were analysed in R (version: 4.04; https://www.r-project.org/) using R-Studio (version: 1.4.1106; https://www.rstudio.com/). We report uncorrected p-values for either ANOVAs or χ2-tests where appropriate. Post hoc tests were Bonferroni-corrected (α = 0.05).

First level analysis

We extracted the trajectories of low-level precision-weighted PEs about the tone tendency ϵ2 and high-level precision-weighted PEs about the volatility of the environment ϵ3. Trial-by trial magnitude estimates of the absolute value of low-level precision-weighted PEs |ϵ2| or high-level precision-weighted PEs ϵ3 were included as parametric regressors to explain trial-by-trial variation in EEG amplitude (Figure 2) as done previously.24 The absolute value of ϵ2 was chosen because it expresses Bayesian surprise independent of the physical characteristics of a tone such as a specific frequency. The general linear model at the first level consisted of an intercept term and either (z-standardized) low or high-level precision-weighted PE trajectories as predictors and EEG amplitude across sensors and peri-stimulus time as the response variable. For each precision-weighted PE, we tested the null hypothesis that the parameter estimate was zero at each sensor and time point using an F-test. Statistical analyses were restricted to 100 to 400ms post-stimulus time.

Second level analysis

First-level statistics were converted into images and smoothed using a Gaussian kernel (FWHM: 16 mm x 16 mm) to ensure that the assumptions of Gaussian random field theory were met.31,32 Smoothed images were carried to the second level to compare groups using different factorial designs for each precision-weighted PE to obtain statistical parametric maps over 2D sensor space and peri-stimulus time (Figure 2). Each factorial design included group as between- and MMN oddball paradigm as within-subject factor, as well as age as a covariate. To ensure that the equal slope assumption for age was met, we masked out voxels that showed a significant group-by-age interaction. Multiple testing correction was implemented using Gaussian random field theory31,32 and we report p-values corrected for peak-(ppFWE) or cluster-level (pcFWE) family-wise error rates using a cluster defining threshold of p<0.00133 unless stated otherwise.

Results

Demographic and clinical characteristics

Demographic and clinical characteristics are displayed in Table 2 (see also 21 for more information).

View this table:
Table 2. Demographic and clinical characteristics

Group differences in low-level precision-weighted prediction errors

We observed a significant group effect on the expression of low-level precision-weighted PEs about the tone tendency ϵ2 peaking at 105ms over left, central channels (F=20.795, pcFWE<0.001) and at 109ms over frontal channels (F=15.656, ppFWE<0.001). Closer inspection of the first effect revealed that the difference between small and large low-level precision-weighted PEs was reduced in central channels in ESZ compared to CHR-P (peak: 152ms, t=4.923, pcFWE =0.001; Figure 3) and ESZ vs HC (peak: 105ms, t=6.427, pcFWE<0.001; Figure 3). The second effect again suggested a reduced difference between small and large low-level precision-weighted PEs. However, this effect was expressed over frontal channels in ESZ vs HC (peak: 109ms, t=5.594, ppFWE<0.001; Figure 3). The timing of these effects coincided with the timing of the MMN21 suggesting that MMN reductions may reflect disturbances in precision-weighted PE updating processes as hypothesised previously.34,35

Figure 3.
Figure 3. Low-level precision-weighted prediction errors ε2.

A-D: Displayed are maximum intensity projections highlighting significant voxels of t-contrasts testing for pairwise group differences in the expression of low-level precision-weighted prediction errors ε2 about the tone tendency. Times displayed on y-axis indicate earliest and latest significant voxel. p-values were corrected for peak-(ppFWE; black dashed-line) or cluster-level (pcFWE) family-wise error rates (FWE) using a cluster defining threshold of p<0.001 (highlighted by coloured area). Note, that p-values in D are small-volume corrected for the group effect on ε2 between HC and ESZ (i.e., significant voxels in plots A OR B; black colour illustrates masked out voxels in D). For illustration, difference waveforms (10% highest −10% lowest ε2 trials) are shown across groups for a channel close to the peak effect. HC: Healthy controls. CHR-P: Individuals at clinical high risk for psychosis. CHR-C: Converters. CHR-NC: Non-converters. ESZ: Early-illness schizophrenia patients (≤5 years since initial hospitalization or initiation of antipsychotic medication). PST: Peristimulus time following tone presentation. Note that the statistical analysis window was restricted to 100-400ms following each tone (standard and deviants).

Group differences in high-level precision-weighted prediction errors

The expression of high-level precision-weighted PEs about the volatility of the environment ϵ3 also showed a significant effect of group peaking at 125ms over right, central channels (F=17.277, pcFWE=0.005). Pairwise comparisons revealed stronger correlations of high-level precision-weighted PEs with EEG amplitudes in HC compared to ESZ over frontal channels (peak: 125ms, t=3.931, ppFWE=0.027) and during a later time window over posterior central channels (peak: 344ms, t=3.821, pcFWE=0.018; Figure 4), which was also significant when comparing CHR-P to ESZ (peak: 340ms, t=3.621, pcFWE=0.046; Figure 4). Furthermore, we found that the difference between small and large precision-weighted PEs was reduced during an early time window in ESZ vs CHR-P (peak: 129ms, t=5.014, pcFWE=0.008; Figure 4) and in ESZ vs HC (peak: 125ms, t=5.728, pcFWE=0.002; Figure 4). While the early cluster again coincided with the time window of the MMN21; the latter cluster rather fell into the P3a time window, raising the question whether the P3a may also reflect PE-related processing (cf. Figure 1).

Figure 4.
Figure 4. High-level precision-weighted prediction errors.

A-D: Displayed are maximum intensity projections highlighting significant voxels of t-contrasts testing for pairwise group differences in the expression of high-level prediction errors ϵ3 about environmental volatility. Times displayed on y-axis indicate earliest and latest significant voxel. p-values were corrected for peak-(ppFWE; black dashed-line) or cluster-level (pcFWE) family-wise error rates (FWE) using a cluster defining threshold of p<0.001 (highlighted by coloured area). For illustration, difference waveforms (10% highest −10% lowest ε3 trials) are shown across groups for a channel close to the peak effect. HC: Healthy controls. CHR-P: Individuals at clinical high risk for psychosis. ESZ: Early-illness schizophrenia patients (≤5 years since initial hospitalization or initiation of antipsychotic medication). PST: Peristimulus time following tone presentation. Note that the statistical analysis window was restricted to 100-400ms following each tone (standard and deviants).

Group differences between converters and non-converters

Lastly, when comparing CHR-P converters to non-converters, we found a significant group effect on the expression of low-level precision-weighted PEs ϵ2 peaking at 137ms over left, central channels (F=12.722, pcFWE=0.040; small-volume corrected for the group effect on ϵ2 between HC and ESZ). In CHR-P individuals who later transitioned to psychosis, the difference between small and large low-level precision-weighted PEs was reduced (peak: 137ms, t=3.567, pcFWE=0.022; small-volume corrected for the group effect on ϵ2 between HC and ESZ; Figure 3D).

Discussion

The objective of this study was to develop and test a mechanistic model of altered information processing as a basis for MMN amplitude reductions in CHR-P individuals and ESZ patients. We obtained three major findings: First, we observed altered expression of low-level precision-weighted PEs about the tone tendency between HC and ESZ and in CHR-P compared to ESZ. Second, we also identified changes in the expression of high-level precision-weighted PEs about the volatility of the environment in ESZ compared to both HC and CHR-P during an early time window (at about 100-175ms peristimulus time), as well as during a later time window (at about 320-380ms). Third, the expression of low-level precision-weighted PEs was significantly altered in those CHR-P that later converted to a psychotic disorder compared to non-converters suggesting that this computational model appears to capture relevant pathophysiological mechanisms and may constitute a useful tool to predict transition to psychosis in CHR-P individuals.

Theoretical implications for the predictive coding account of psychosis

Our results are in line with the predictive coding account of psychosis that postulates that disturbances in hierarchical PE processing may contribute to psychotic symptoms.14,36 Our finding of alterations in the expression of precision-weighted PEs in central channels in patients with schizophrenia may suggest that patients experience aberrantly salient PEs37 in response to familiar stimuli (standard tones). Furthermore, changed expression of hierarchical PEs in frontal channels could signal a decrease in the precision of priors in frontal regions, as proposed in the predictive coding account of psychosis.14

Possible cortical generators of aberrant prediction errors

The network of cortical regions thought to be involved in generating the MMN response includes bilateral primary auditory cortices (A1), superior temporal gyri (STG) and inferior frontal gyri (IFG).34,38–40 It is possible that the different spatiotemporal clusters that were identified in our study may be caused by different cortical generators, for example the correlation between precision-weighted PEs and EEG amplitudes in ESZ expressed in central channels may originate in A1 or STG, while the second cluster, which was identified over frontal regions, possibly suggests involvement of IFG. However, due to volume-conduction effects, we will have to formally test this hypothesis using source modelling in the future. Adams and colleagues41 recently investigated the neural mechanisms of schizophrenia using dynamic causal modelling. They found remarkably consistent findings across a wide range of paradigms implicating reductions in a model parameter that serves to amplify the magnitude of activity or synaptic gain of pyramidal cells. Notably, this study also included an MMN paradigm, in which the authors identified reduction of pyramidal gain in IFG specifically. This finding suggests that the expression of hierarchical PEs over frontal channels may be altered due to a reduction of the precision-weight rather than changes in the PE component of the precision-weighted PEs.

Are aberrant precision-weighted prediction errors related to alterations in neurotransmission?

The dysconnectivity hypothesis13,16,42–44 postulates that NMDAR-mediated modulation of synaptic gain is altered in schizophrenia. In line with this account, Weber et al.24 found that ketamine administration led to a reduced expression of high-level precision-weighted Pes about the volatility of the environment in central channels, similar to our results. However, their results suggest that low-level precision-weighted PEs are unaffected by ketamine.

Several neurotransmitters interact with NMDAR to dynamically control synaptic gain and neuroplasticity. Altered expression of precision-weighted PEs in ESZ, as identified in our study over early auditory regions, could reflect changes in cholinergic neurotransmission. Two recent studies implicate acetylcholine in regulating synaptic gain or — according to the predictive coding account and the dysconnectivity hypothesis — regulating sensory precision in early auditory regions.35,45 The first study employed a Kalman filter (i.e., a 2-level HGF) to model changes in participants that were administered galantamine, which enhances cholinergic neurotransmission.35 The authors argued that galantamine may increase the precision of sensory PEs. The second study by Schöbi and colleagues45 modelled changes in between- and within-region connectivity including synaptic gain during a pharmacological manipulation using muscarinic receptor antagonist scopolamine or muscarinic receptor agonist pilocarpine in rats.45 The authors found dose-dependent changes in synaptic gain, but also changes in inter-regional connectivity between A1 and secondary auditory cortex. Moreover, changes in muscarinic receptor density among schizophrenia patients have been frequently reported 46–49 and Scarr et al.48 proposed that there may be a subgroup of schizophrenia patients specifically characterized by decreased cortical muscarine receptor expression. These results support a potential role of cholinergic neurotransmission in precision-weighted PE signalling. Beyond cholinergic processes, glutamatergic neurotransmission at AMPA receptors may be involved, but its precise role still needs to be clarified.

Clinical Implications

Interestingly, our results suggest that the expression of low-level precision-weighted PEs is blunted in CHR-P that later converted to a psychotic disorder compared to CHR-P that did not convert. This finding highlights potential applications of this computational approach to prediction of psychosis in CHR-P individuals. Furthermore, if the neurotransmitter systems that are involved in computing precision-weighted PEs during the MMN paradigm can be identified, this approach may be useful for identifying critical time windows for preventative interventions or for predicting treatment response to pharmacological interventions that target either glutamatergic neurotransmission like d-serine, which has shown promising results in a recent clinical trial,50 or cholinergic neurotransmission, for example involving muscarinic (M1, M4) agonist xanomeline.51

Limitations

A few limitations of this study merit attention. First, the oddball paradigms in this study were not well-suited to separate low- and high-level PEs because environmental volatility was not manipulated explicitly throughout the task. Future studies should include explicit manipulations of volatility (i.e., changes in deviant probabilities over the course of the experiment, rather than a 90% stable probability) to better distinguish between different levels of hierarchical inference. Secondly, we assumed that participants acted as ideal Bayesian observers without taking subject-specific deviation from an ideal Bayesian observer into account by estimating subject-specific parameters based on both sensory input and behavioural responses. This is an inherent limitation of the passive MMN oddball paradigm. In light of this limitation, our results are more challenging to interpret. Group differences could arise because different groups are better explained by different models or model parameter values. While MMN amplitude reductions have been frequently replicated in schizophrenia,4 future studies should also investigate the representation of precision-weighted PEs using active oddball paradigms that require participants to detect and respond to infrequent target stimuli, such as the paradigm used in a recent study,52 which found that target P3b amplitudes were predictive of conversion to psychosis.

Future directions

Future studies should determine the cortical generators of changes in the expression of hierarchical precision-weighted PEs in the clinical high risk for psychosis state and early schizophrenia. Moreover, the biological implementation of these computations needs to be clarified further, for example through the use of models that include greater physiological detail to bridge the algorithmic description that our modelling approach offers and its physiological implementation in the brain.53 Dynamic causal models for electrophysiological data have been highlighted as computational assays that may allow to infer receptor densities of neuronal populations.13,54 These models have been validated in studies investigating NMDAR antibody encephalitis,55 dopaminergic action on NMDARs,56 and manipulations of cholinergic neurotransmission,45 and thus, constitute a promising way forward. Additionally, there is a need for more pharmacological studies in both animals and humans to map the relationship between hierarchical precision-weighted PEs and different neurotransmitter systems that are targeted by antipsychotic medication.

Conclusions

In this study, we examined the computational mechanisms underlying pre-attentive auditory deviance processing in the clinical high risk for psychosis state and early schizophrenia and found evidence for aberrant expression of precision-weighted PEs at different levels of hierarchical inference. Our results suggest that the expression of low-level precision-weighted PEs is significantly altered in individuals at clinical high risk for psychosis that will later transition to psychosis, highlighting that this computational modelling approach captures relevant pathophysiological mechanisms and may prove useful for predicting transition to psychosis in CHR-P individuals.

Data Availability

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Disclosure

Dr. Mathalon is a consultant for Gilgamesh Pharmaceuticals, Neurocrine Biosciences, and Recognify Life Sciences. All other authors reported no biomedical financial interests or potential conflicts of interest.

Acknowledgements

This work was supported by the National Institutes of Health (R01 MH076989 to DHM), the Swiss National Science Foundation (Doc.Mobility, 200054 to DJH; Ambizione, PZ00P3_167952 to AOD), the Brain and Behavior Research Foundation (to DHM), the Krembil Foundation (to AOD), and the U.S. Department of Veterans Affairs (Veterans Affairs Senior Research Career Award, 1IK6CX002519 to JMF).

A previous version of this article was published as part of DJH’s PhD thesis57 and made available as a preprint on https://www.medrxiv.org/.

References

  1. 1.↵
    Näätänen R, Paavilainen P, Rinne T, Alho K. The mismatch negativity (MMN) in basic research of central auditory processing: A review. Clin Neurophysiol. 2007;118(12):2544–2590. doi:10.1016/J.CLINPH.2007.04.026
    OpenUrlCrossRefPubMedWeb of Science
  2. 2.↵
    Näätänen R, Gaillard AWK, Mäntysalo S. Early selective-attention effect on evoked potential reinterpreted. Acta Psychol (Amst). 1978;42(4):313–329.
    OpenUrlCrossRefPubMedWeb of Science
  3. 3.↵
    Fitzgerald K, Todd J. Making sense of mismatch negativity. Front Psychiatry. 2020;11:468.
    OpenUrlCrossRefPubMed
  4. 4.↵
    Erickson MA, Ruffle A, Gold JM. A meta-analysis of mismatch negativity in schizophrenia: from clinical risk to disease specificity and progression. Biol Psychiatry. 2016;79(12):980–987. doi:10.1016/j.biopsych.2015.08.025.A
    OpenUrlCrossRef
  5. 5.↵
    Ehrlichman RS, Maxwell CR, Majumdar S, Siegel SJ. Deviance-elicited changes in event-related potentials are attenuated by ketamine in mice. J Cogn Neurosci. 2008;20(8):1403–1414. doi:10.1162/jocn.2008.20097
    OpenUrlCrossRefPubMed
  6. 6.↵
    Javitt DC, Steinschneidertt M, Schroedertt CE, Arezzott JC. Role of cortical N-methyl-D-aspartate receptors in auditory sensory memory and mismatch negativity generation: Implications for schizophrenia. Proc Natl Acad Sci U S A. 1996;93:11962–11967.
    OpenUrlAbstract/FREE Full Text
  7. 7.↵
    Schmidt A, Bachmann R, Kometer M, et al. Mismatch Negativity Encoding of Prediction Errors Predicts S-ketamine-Induced Cognitive Impairments. Neuropsychopharmacology. 2012;37(4):865–875. doi:10.1038/npp.2011.261
    OpenUrlCrossRefPubMedWeb of Science
  8. 8.
    Umbricht D, Schmid L, Koller R, Vollenweider FX, Hell D, Javitt DC. Ketamine-Induced Deficits in Auditory and Visual Context-Dependent Processing in Healthy Volunteers. Arch Gen Psychiatry. 2000;57:1139–1147.
    OpenUrlCrossRefPubMedWeb of Science
  9. 9.↵
    Heekeren K, Daumann J, Neukirch A. Mismatch negativity generation in the human 5HT 2A agonist and NMDA antagonist model of psychosis. Psychopharmacology (Berl). 2008;300:77–88. doi:10.1007/s00213-008-1129-4
    OpenUrlCrossRef
  10. 10.↵
    Todd J, Harms L, Schall U, Michie PT. Mismatch negativity: Translating the potential. Front Psychiatry. 2013;4:1–22. doi:10.3389/fpsyt.2013.00171
    OpenUrlCrossRefPubMed
  11. 11.↵
    Corlett PR, Honey GD, Krystal JH, Fletcher PC. Glutamatergic model psychoses: Prediction error, learning, and inference. Neuropsychopharmacology. 2011;36:294–315.
    OpenUrlCrossRefPubMedWeb of Science
  12. 12.
    Corlett PR, Honey GD, Fletcher PC. Prediction error, ketamine and psychosis: An updated model. J Psychopharmacol. 2016;30(11):1145–1155. doi:10.1177/0269881116650087
    OpenUrlCrossRefPubMed
  13. 13.↵
    Stephan KE, Baldeweg T, Friston KJ. Synaptic plasticity and dysconnection in schizophrenia. Biol Psychiatry. 2006;59:929–939. doi:10.1016/j.biopsych.2005.10.005
    OpenUrlCrossRefPubMedWeb of Science
  14. 14.↵
    Sterzer P, Adams RA, Fletcher P, et al. The predictive coding account of psychosis. Biol Psychiatry. Published online 2018. doi:10.1016/j.biopsych.2018.05.015
    OpenUrlCrossRefPubMed
  15. 15.
    Moghaddam B, Javitt D. From Revolution to Evolution: The Glutamate Hypothesis of Schizophrenia and its Implication for Treatment. Neuropsychopharmacology. Published online 2012:4–15. doi:10.1038/npp.2011.181
    OpenUrlCrossRefPubMedWeb of Science
  16. 16.↵
    Friston KJ, Brown HR, Siemerkus J, Stephan KE. The dysconnection hypothesis (2016). Schizophr Res. 2016;176(2-3):83–94. doi:10.1016/j.schres.2016.07.014
    OpenUrlCrossRefPubMed
  17. 17.↵
    Luby ED, Gottlieb JS, Cohen BD, Rosenbaum G, Domino EF. Model psychoses and schizophrenia. Am J Psychiatry. 1962;119(1):61–67.
    OpenUrlCrossRefPubMedWeb of Science
  18. 18.↵
    Bodatsch M, Ruhrmann S, Wagner M, et al. Prediction of psychosis by mismatch negativity. Biol Psychiatry. 2011;69:959–966. doi:10.1016/j.biopsych.2010.09.057
    OpenUrlCrossRefPubMedWeb of Science
  19. 19.
    Bodatsch M, Brockhaus-Dumke A, Klosterkötter J, Ruhrmann S. Forecasting Psychosis by Event-Related Potentials—Systematic Review and Specific Meta-Analysis. Biol Psychiatry. 2015;77(11):951–958. doi:https://doi.org/10.1016/j.biopsych.2014.09.025
    OpenUrlCrossRefPubMed
  20. 20.
    Hamilton HK, Boos AK, Mathalon DH. Electroencephalography and Event-Related Potential Biomarkers in Individuals at Clinical High Risk for Psychosis. Biol Psychiatry. 2020;88(4):294–303. doi:https://doi.org/10.1016/j.biopsych.2020.04.002
    OpenUrlPubMed
  21. 21.↵
    Perez VB, Woods SW, Roach BJ, et al. Automatic Auditory Processing Deficits in Schizophrenia and Clinical High-Risk Patients: Forecasting Psychosis Risk with Mismatch Negativity. Biol Psychiatry. 2014;75(6):459–469. doi:10.1016/j.biopsych.2013.07.038
    OpenUrlCrossRefPubMed
  22. 22.↵
    Haigh SM, Coffman BA, Salisbury DF. Mismatch Negativity in First-Episode Schizophrenia: A Meta-Analysis. Clin EEG Neurosci. 2016;48(1):3–10. doi:10.1177/1550059416645980
    OpenUrlCrossRefPubMed
  23. 23.↵
    Davies C, Cipriani A, Ioannidis JPA, et al. Lack of evidence to favor specific preventive interventions in psychosis: a network meta-analysis. World Psychiatry. 2018;17(2):196–209. doi:10.1002/wps.20526
    OpenUrlCrossRef
  24. 24.↵
    Weber LA, Diaconescu AO, Mathys C, et al. Ketamine Affects Prediction Errors about Statistical Regularities: A Computational Single-Trial Analysis of the Mismatch Negativity. J Neurosci. 2020;(June). doi:10.1523/JNEUROSCI.3069-19.2020
    OpenUrlAbstract/FREE Full Text
  25. 25.↵
    First M, Spitzer R, Gibbon M, Williams J. Structured Clinical Interview for DSM-IV Axis I Disorders (SCID), Research Version, Patient Edition with Psychotic Screen. Biometrics Research, New York State Psychiatric Institute; 1997.
  26. 26.↵
    Miller TJ, Ph D, Mcglashan TH, et al. Prospective Diagnosis of the Initial Prodrome for Schizophrenia Based on the Structured Interview for Prodromal Syndromes : Preliminary Evidence of Interrater Reliability and Predictive Validity. Am J Psychiatry. 2002;19:863–865.
    OpenUrl
  27. 27.↵
    Miller TJ, McGlashan TH, Rosen JL, et al. Prodromal assessment with the structures interview for prodromal syndromes and the scale of prodromal symptoms: Predictive validity, interrater reliability, and training to reliability. Schizophr Bull. 2003;29(4):703–715.
    OpenUrlCrossRefPubMedWeb of Science
  28. 28.↵
    Mathys CD, Daunizeau J, Friston KJ, Stephan KE. A Bayesian foundation for individual learning under uncertainty. Front Hum Neurosci. 2011;5(May):1–20. doi:10.3389/fnhum.2011.00039
    OpenUrlCrossRefPubMed
  29. 29.↵
    Mathys CD, Lomakina EI, Daunizeau J, et al. Uncertainty in perception and the Hierarchical Gaussian filter. Front Hum Neurosci. 2014;8(November):1–24. doi:10.3389/fnhum.2014.00825
    OpenUrlCrossRefPubMed
  30. 30.↵
    Frässle S, Aponte EA, Bollmann S, et al. TAPAS: An open-source software package for Translational Neuromodeling and Computational Psychiatry. Front Psychiatry. 2021;12:680811.
    OpenUrlCrossRef
  31. 31.↵
    Kiebel SJ, Friston KJ. Statistical parametric mapping for event-related potentials: I. Generic considerations. Neuroimage. 2004;22(2):492–502.
    OpenUrlCrossRefPubMedWeb of Science
  32. 32.↵
    Worsley KJ, Marrett S, Neelin P, Vandal AC, Friston KJ, Evans AC. A unified statistical approach for determining significant signals in images of cerebral activation. Hum Brain Mapp. 1996;4(1):58–73.
    OpenUrlCrossRefPubMedWeb of Science
  33. 33.↵
    Flandin G, Friston KJ. Analysis of family-wise error rates in statistical parametric mapping using random field theory. Hum Brain Mapp. 2019;40(7):2052–2054. doi:10.1073/pnas.1602413113
    OpenUrlCrossRef
  34. 34.↵
    Garrido MI, Kilner JM, Stephan KE, Friston KJ. The mismatch negativity: A review of underlying mechanisms. Clin Neurophysiol. 2009;120(3):453–463. doi:10.1016/j.clinph.2008.11.029
    OpenUrlCrossRefPubMedWeb of Science
  35. 35.↵
    Moran RJ, Campo P, Symmonds M, Stephan KE, Dolan RJ, Friston KJ. Free energy, precision and learning: The role of cholinergic neuromodulation. J Neurosci. 2013;33(19):8227–8236. doi:10.1523/JNEUROSCI.4255-12.2013
    OpenUrlAbstract/FREE Full Text
  36. 36.↵
    Fletcher PC, Frith CD. Perceiving is believing: A Bayesian approach to explaining the positive symptoms of schizophrenia. Nat Rev Neurosci. 2009;10(1):48–58. doi:10.1038/nrn2536
    OpenUrlCrossRefPubMedWeb of Science
  37. 37.↵
    Kapur S. Psychosis as a state of aberrant salience: a framework linking biology, phenomenology, and pharmacology in schizophrenia. Am J Psychiatry. 2003;160:13–23. doi:Research Support, Non-U.S. Gov’t Review
    OpenUrlCrossRefPubMedWeb of Science
  38. 38.↵
    Doeller CF, Opitz B, Mecklinger A, Krick C, Reith W, Schröger E. Prefrontal cortex involvement in preattentive auditory deviance detection:: neuroimaging and electrophysiological evidence. Neuroimage. 2003;20(2):1270–1282.
    OpenUrlCrossRefPubMedWeb of Science
  39. 39.
    Marco-Pallarés J, Grau C, Ruffini G. Combined ICA-LORETA analysis of mismatch negativity. Neuroimage. 2005;25(2):471–477.
    OpenUrlCrossRefPubMedWeb of Science
  40. 40.↵
    Opitz B, Rinne T, Mecklinger A, Von Cramon DY, Schröger E. Differential contribution of frontal and temporal cortices to auditory change detection: fMRI and ERP results. Neuroimage. 2002;15(1):167–174.
    OpenUrlCrossRefPubMedWeb of Science
  41. 41.↵
    Adams RA, Pinotsis D, Tsirlis K, et al. Computational Modeling of Electroencephalography and Functional Magnetic Resonance Imaging Paradigms Indicates a Consistent Loss of Pyramidal Cell Synaptic Gain in Schizophrenia. Biol Psychiatry. 2022;91(2):202–215. doi:10.1016/j.biopsych.2021.07.024
    OpenUrlCrossRef
  42. 42.↵
    Friston KJ. The disconnection hypothesis. Schizophr Res. 1998;30:115–125.
    OpenUrlCrossRefPubMedWeb of Science
  43. 43.
    Friston KJ, Frith CD. Schizophrenia: a disconnection syndrome. Clin Neurosci. 1995;3(2):89–97.
    OpenUrlCrossRefPubMedWeb of Science
  44. 44.↵
    Stephan KE, Friston KJ, Frith CD. Dysconnection in schizophrenia: From abnormal synaptic plasticity to failures of self-monitoring. Schizophr Bull. 2009;35(3):509–527. doi:10.1093/schbul/sbn176
    OpenUrlCrossRefPubMedWeb of Science
  45. 45.↵
    Schöbi D, Homberg F, Frässle S, et al. Model-based prediction of muscarinic receptor function from auditory mismatch negativity responses. Neuroimage. 2021;237(January). doi:10.1016/j.neuroimage.2021.118096
    OpenUrlCrossRef
  46. 46.↵
    Scarr E, Hopper S, Vos V, et al. Low levels of muscarinic M1 receptor–positive neurons in cortical layers III and V in Brodmann areas 9 and 17 from individuals with schizophrenia. J Psychiatry Neurosci. 2018;43(5):338–346.
    OpenUrl
  47. 47.
    Scarr E, Craig JM, Cairns MJ, et al. Decreased cortical muscarinic M1 receptors in schizophrenia are associated with changes in gene promoter methylation, mRNA and gene targeting microRNA. Transl Psychiatry. 2013;3(2):e230–e230.
    OpenUrl
  48. 48.↵
    Scarr E, Cowie TF, Kanellakis S, Sundram S, Pantelis C, Dean B. Decreased cortical muscarinic receptors define a subgroup of subjects with schizophrenia. Mol Psychiatry. 2009;14(11):1017–1023.
    OpenUrlCrossRefPubMedWeb of Science
  49. 49.↵
    Dean K, Murray RM. Environmental risk factors for psychosis. Dialogues Clin Neurosci. 2005;7(1):69–80.
    OpenUrlPubMed
  50. 50.↵
    Kantrowitz JT, Woods SW, Petkova E, et al. D-serine for the treatment of negative symptoms in individuals at clinical high risk of schizophrenia: a pilot, double-blind, placebo-controlled, randomised parallel group mechanistic proof-of-concept trial. The Lancet Psychiatry. 2015;2(5):403–412. doi:10.1016/S2215-0366(15)00098-X
    OpenUrlCrossRef
  51. 51.↵
    Paul SM, Yohn SE, Popiolek M, Miller AC, Felder CC. Muscarinic acetylcholine receptor agonists as novel treatments for schizophrenia. Am J Psychiatry. 2022;179(9):611–627.
    OpenUrl
  52. 52.↵
    Hamilton HK, Roach BJ, Bachman PM, et al. Association between P300 responses to auditory oddball stimuli and clinical outcomes in the psychosis risk syndrome. JAMA Psychiatry. 2019;76(11):1187–1197. doi:10.1001/jamapsychiatry.2019.2135
    OpenUrlCrossRef
  53. 53.↵
    Marr D. Vision. Freeman; 1982.
  54. 54.↵
    Stephan KE, Mathys C. Computational approaches to psychiatry. Curr Opin Neurobiol. 2014;25. doi:10.1016/j.conb.2013.12.007
    OpenUrlCrossRefPubMed
  55. 55.↵
    Symmonds M, Moran CH, Leite MI, et al. Ion channels in EEG: isolating channel dysfunction in NMDA receptor antibody encephalitis. Brain. Published online 2018:1691–1702. doi:10.1093/brain/awy107
    OpenUrlCrossRefPubMed
  56. 56.↵
    Moran RJ, Symmonds M, Stephan KE, Friston KJ, Dolan RJ. An in vivo assay of synaptic function mediating human cognition. Curr Biol. 2011;21(15):1320–1325. doi:10.1016/j.cub.2011.06.053
    OpenUrlCrossRefPubMed
  57. 57.↵
    Hauke DJ. Hierarchical Bayesian Inference in Psychosis. Published online 2022. https://edoc.unibas.ch/89780/
  58. 58.
    Kay SR, Fiszbein A A OL. The positive and negative syndrome scale (PANSS) for schizophrenia. Schizophr Bull. 1987;13(2):261–276.
    OpenUrlCrossRefPubMedWeb of Science
  59. 59.
    Crovitz HF, Zener K. A group-test for assessing hand-and eye-dominance. Am J Psychol. 1962;75(2):271–276.
    OpenUrlCrossRefPubMedWeb of Science
Back to top
Posted December 20, 2022.
Download PDF
Data/Code
Email
Aberrant hierarchical prediction errors are associated with transition to psychosis: A computational single-trial analysis of the mismatch negativity
Daniel J. Hauke, Colleen E. Charlton, André Schmidt, John Griffiths, Scott W. Woods, Judith M. Ford, Vinod H. Srihari, Volker Roth, Andreea O. Diaconescu, Daniel H. Mathalon
medRxiv 2022.12.20.22283712; doi: https://doi.org/10.1101/2022.12.20.22283712
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area