40-Hz Auditory Steady-State Responses Characterize Circuit Dysfunctions and Predict Clinical Outcomes in Clinical High-Risk for Psychosis Participants: A Magnetoencephalography Study

,

positive (PV1) GABA (gamma-aminobutyric acid) interneurons and NMDA receptors (NMDARs) are impaired in ScZ (13), highlighting the potential of gamma-band oscillations to provide insights into circuit abnormalities.
Currently, there is robust evidence that both amplitude and phase of 40-Hz ASSRs are impaired in ScZ (14,15) which have been linked to deficits in GABAergic neurotransmission (16).There is mixed evidence, however, for the presence of 40-Hz ASSR deficits in both individuals at a clinical high-risk for psychosis (CHR-P) (17)(18)(19)(20) and patients with first-episode psychosis (FEP) (19,21).In addition, it is unclear whether 40-Hz ASSRs could constitute a possible biomarker for the early detection and diagnosis of emerging psychosis.This is important because CHR-P participants, who are identified on the basis of attenuated psychotic symptoms (APSs), brief limited intermittent psychotic symptoms, or functional decline with genetic risk (22) as well as self-experienced perceptual and cognitive disturbances known as basic symptoms (23), are characterized by cognitive deficits (24) and impairments in social and role functioning (25) that predict clinical outcomes (26,27).In addition, only a minority of CHR-P participants (~20% over a 3-year period) will develop a psychotic episode (28).
To address the potential of 40-Hz ASSRs as a biomarker for the early detection and diagnosis of emerging psychosis, we applied a state-of-the-art MEG approach to examine 40-Hz ASSRs in CHR-P participants as well as in patients with FEP.MEG is characterized by an improved signal-to-noise ratio for measurements of high-frequency oscillations compared with electroencephalography (EEG) (29) and is ideally suited for source reconstruction.Source-level estimation of 40-Hz ASSR has been shown to be higher at source level than at sensor level (30,31).Crucially, source estimation allows the unmixing of signal contributions from different source generators that are obscured at sensor level by individual differences in source orientation and field spread (32).
In this study, we investigated MEG-recorded virtual channel 40-Hz ASSRs in a group of 116 CHR-P participants and 33 patients with FEP as well as in 38 participants with substance abuse and affective disorders (clinical high-risk for psychosisnegative group [CHR-N]) and compared spectral power and phase-coherence measures with 49 healthy control (HC) subjects.We predicted that FEP and CHR-P groups would be characterized by a circumscribed dysfunction of 40-Hz ASSRs in the auditory cortex and subcortical brain regions, which would be closely linked to clinical outcomes.Specifically, we hypothesized that impaired 40-Hz ASSRs would predict persistence of APSs as well as transition to psychosis in CHR-P participants.
CHR-P status at baseline was established by ultra-high risk criteria according to the Comprehensive Assessment of At-Risk Mental States (CAARMS) Interview (22) and the Cognitive Disturbances and Cognitive-Perceptive Basic Symptoms criteria according to the Schizophrenia Proneness Instrument, Adult version (SPI-A) (23).Patients with FEP were assessed with the Structured Clinical Interview for DSM-5 (34) and with the Positive and Negative Syndrome Scale (35).For all groups except patients with FEP, cognition was assessed with the Brief Assessment of Cognition in Schizophrenia (BACS) (36).See Tables 1-3 for demographic and clinical data.
The study was approved by the ethical committees of University of Glasgow and the National Health Service Research Ethics Committee Greater Glasgow & Clyde.All participants provided written informed consent.

Clinical Follow-up
Participants meeting CHR-P criteria were reassessed at intervals of 3, 6, 9, 12, 18, 24, 30, and 36 months to examine persistence of ultra-high risk criteria and transition to psychosis.Persistence of ultra-high risk criteria was operationalized by the continued presence of APSs up to 12 months in this study (see Supplement and Table S1).

Stimuli and Task
Auditory stimuli were 1000-Hz carrier tones (duration: 2 s), amplitude modulated (AM) at 40 Hz (31), presented binaurally through inner ear tubes, with an interstimulus interval of on average 2 seconds (jittered between 1.5 and 2.5 s, equal distribution).Participants were instructed to fixate a translucent screen (viewing distance: 75 cm).Participants received one block of 100 of 40-Hz AM tones (ripple tones) (Figure 1A).To control for potential attentional differences, 10 additional identical sounds with equal intensity levels over time (flat tones) were interspersed, serving as targets to respond to by button press.These target trials were not included in the analyses of the MEG data.

Task Performance
Analysis of task data included percentage of correctly detected flat-tone targets, mean reaction times of correct responses, and false alarms.

40-Hz ASSRs During Emerging Psychosis
Biological Psychiatry inspection and applying independent component analysisbased removal of eye blink, eye movement, and electrocardiographic artifacts.
The main analyses, however, focused on data transformed into source space because regional specificity at each sensor is compromised by field spread through inputs from multiple sources and interindividual differences in temporal cortex folding (32).The basic source-level analyses steps included the following: 1) determination of regions showing stimulusentrained 40-Hz ASSR signal; 2) determination of a subset of regions showing main group effects (HC, CHR-P, CHR-N, FEP); 3) identification of group differences (post hoc pairwise group comparisons); and 4) contributions of ASSR impairments toward predicting transition to psychosis and/or 1-year persistence of APSs.
First, MEG data were coregistered with the T1 MRI scans, using anatomical landmarks (nasion, bilateral preauricular points) and head-shape data collected using a Polhemus 3D Fasttrack digitization system (Polhemus, Colchester, VT).A 3D grid of 5-mm resolution was then created and linearly warped into a single-shell volume conductor model.

Statistical Analyses
Main Effects of Stimulation and Group.Main effects of stimulation were determined from dynamic imaging of coherent sources whole-brain activity, using Monte Carlo permutationbased dependent-sample t test (a = 0.05, one-sided, false discovery rate [FDR]-corrected) across all 236 participants, with baseline and stimulus-induced 40-Hz ASSR power as paired samples.Main effects of group included linear regression analysis (backward-selection method, a = 0.05, two-sided, 1000 sample bootstrapped) on 40-Hz amplitude and 40-Hz ITPC virtual channel data (250-2000 ms averaged relch data) from all regions significantly entrained to the 40-Hz stimulation.
Group Differences.Group differences in MEG-included trial numbers, behavioral task performance, and demographic and clinical as well as cognition data were assessed with nonparametric Kruskal-Wallis tests, a level 0.05, two-sided, with post   For sensor-level ASSR data (power and ITPC relch data), averaged across 39-41 Hz and 250-2000 ms, were submitted to nonparametric cluster-based permutation (n = 1000), independent-sample F tests in FieldTrip to find clusters of sensors with main group differences.In addition, a more restricted analysis, using Kruskal-Wallis testing, was conducted on averaged data from four right frontal-temporal sensors covering the maximum ASSR responses in control subjects (Figure S1).
Virtual channel 40-Hz ASSR data from entrained brain regions across participants were subjected to a nonparametric permutation (n = 5000) t test, tested against the null hypothesis of no difference from the HC group, using estimation statistics (41).Confidence intervals (CIs) were obtained from the central 95% of the resampling distribution.To account for skewed data distribution, data were bias corrected and accelerated bootstrap corrected.Additional contrasts were conducted for CHR-P participants with persistent APSs (APS-P) versus nonpersistent symptoms (APS-NP) within 1-year follow-up as well as between CHR-P participants who transitioned to psychosis (CHR-P-T) versus those who did not (CHR-P-NT).For all contrasts, effect sizes were computed using Cohen's d, and FDR correction was applied to control subjects for multiple comparisons.

Demographic/Clinical Data and Task Performance
The CHR-P group consisted of 30 participants who met SPI-A criteria, 31 participants with CAARMS criteria, and 55 participants who had a combination of SPI-A and CAARMS criteria (Table 3).The FEP group included significantly more male participants than the HC (p = .004),CHR-N (p = .002),and CHR-P groups (p , .001) (Table 1).Patients with FEP were significantly older than CHR-P participants (p = .015).CHR-P participants had significantly fewer years of education (p = .032)and significantly lower BACS composite (p = .014),token motor (p , .001), and symbol coding (p = .002)scores than HC subjects.The FEP, CHR-P, and CHR-N groups had significantly lower GAF scores than HC participants and each other (all p values , .001).In addition, the CHR-P group had lower scores than the HC group in global role functioning (p , .001).Both CHR-P and CHR-N groups were also characterized by lower social functioning (CHR-N, p = .003;CHR-P, p , .001).Patients with FEP produced more behavioral false alarms than HC subjects on rare targets during the ASSR task (p = .044).

Follow-up Outcomes
Follow-up data were available for 110 of the 116 CHR-P individuals (see Supplement).Thirty-four CHR-P participants continued to meet APS criteria at 12 months (APS-P), whereas 39 CHR-P participants did not (APS-NP).Compared with APS-NPs, the APS-P group scored significantly higher on CAARMS severity (p = .001)at baseline (p = .028)(Table S2).A total of 13 participants (11.2%) made a transition to psychosis after a mean follow-up period of 18 months (Table S3).Transitioned CHR-P-T participants had significantly lower GAF (p = .034)and GF social scores (p = .023)at baseline than the CHR-P-NT group.

Main Effect of 40-Hz ASSR Stimulation and Main Group Effects
Across all groups, 40-Hz AM tones induced increased sustained (250-2000 ms) 40-Hz power (Figure 1B, lower panel) in 10 brain regions (F values = 2.5-4.6,p = .002,95% CI = 20.042 to 0.005, FDR corrected), including RHES, RSTG, RMTG, right Rolandic operculum, right supramarginal gyrus, RHIP, right parahippocampal gyrus, RTHA, left Rolandic operculum, and left supramarginal gyrus.For these ROIs, virtual channel time-series data were computed using linearly constrained minimum variance beamformers (42), with normalized lead fields and a regularization parameter of 20% to attenuate leakage from nearby sources.Time series were computed separately for each voxel within each ROI and then combined into one time series per ROI using the first singular value decomposition component across the single-voxel data.Data were then submitted to time-frequency Fourier analysis (frequency resolution 0.25 Hz, sliding window of 500 ms, step size 25 ms, Hanning tapered), and ITPC was computed.In addition, 40-Hz ASSR-amplitude responses were computed by bandpass filtering single-trial virtual channel data with a sharp Butterworth filter (range 39.6-40.4Hz, two-pass, rolloff = 4) (see Figure 1A).The filtered data were enveloped and averaged across trials.All virtual channel data were expressed as relch from baseline activity (2500 to 0 ms).
To reduce data dimensionality, linear regression analyses (backward selection) on 40-Hz amplitude and ITPC virtual channel data were used to select a limited set of ROIs from the 10 stimulus-entrained ROIs for post hoc analyses.For the 40-Hz ITPC data, a significant model was found including RHES activity only (F 1,234 = 5.15, p = .024,adjusted R 2 = 0.017).For 40-Hz amplitude data, however, a combination of all ROIs except the LROL contributed significant variance to explaining main group differences (F 8,235 = 2.17, p = .031,adjusted R 2 = 0.045).An additional Bonferroni correction was applied to these data (a level adjusted to 0.0055 [0.05/9 ROIs]), which left a model for main group differences containing only RHES, RSTG, RTHA, and RHIP (F 4,231 = 3.95, p = .004,adjusted R 2 = 0.048).

Post Hoc Group Differences at Sensor Level
Sensor-level data showed stimulus-induced 40-Hz power and ITPC increases over right and left frontal-temporal sensors (Figure S1).Compared with the HC group, both power and ITPC in the 40-Hz range were reduced over right temporal sensors (Figure S1) in CHR-P and FEP but not in the CHR-N group.However, these differences were not significant.

Classification of APS
In summary, while persistence of APS is predicted by 40-Hz ASSR impairments in RHIP, RSTG, and RMTG, transitioning to FEP is predicted by RTHA impairments.

MEG Activity in Subcortical Regions
Estimation of subcortical activity potentially suffers from source leakage from cortical sources and/or effects of depthbias corrections.To validate the RTHA and RHIP results, we therefore examined group differences also in left thalamus and left hippocampus.There were no main group differences in these ROIs (left thalamus: CHR-N, p = .31,CHR-P, p = .41,FEP, p = .08;left hippocampus: CHR-N, p = .06,CHR-P, p = .73,FEP, p = .45).

DISCUSSION
Robust evidence exists for impairments in 40-Hz ASSRs in patients with ScZ (14), which is consistent with evidence for alterations in neural circuits that are involved in the generation of gamma-band rhythms, such as PV1 interneurons (13,16) and NMDAR-mediated neurotransmission (43).However, it is currently unclear whether 40-Hz ASSRs are impaired during emerging psychosis and can predict clinical outcomes.To address these fundamental issues, we implemented a state-ofthe-art MEG approach to examine which brain regions contribute toward 40-Hz ASSR deficits in a large CHR-P sample and to determine whether such impairments predict persistence of APS-status and transition to psychosis.Consistent with our hypothesis, we found overlapping reductions in 40-Hz ASSRs in FEP and CHR-P participants in RHES, RTHA, and RHIP, consistent with the prominent contribution of 40-Hz ASSR generators in the right hemisphere (44).Moreover, we found evidence for circumscribed correlations between 40-Hz ASSR amplitudes in the RSTG and severity of APS as well as impaired functioning in the CHR-P group.Deficits in RHES were in line with previous observations in ScZ that localized 40-Hz ASSR impairments to the primary auditory cortex and superior temporal cortex (45,46), although deficits have also been observed in the left hemisphere (14).These results contrast, however, with recent EEG studies that found intact 40-Hz ASSRs in both CHR-P (19,20) and FEP (19,47) groups.
One reason for our different findings may constitute the larger sample size in this study.We also implemented a novel analysis approach that involved narrow bandpass filtering of single-trial MEG activity, which, compared with more typically used time-frequency analyses, is less affected by trade-offs in time/frequency resolution.Importantly, the use of MEG allowed us to compute virtual channel source-level ASSRs from cortical and subcortical regions entrained to the stimulation frequency, thus increasing signal-to-noise ratio of ASSR estimates (31).
Interestingly, our whole-brain source-level approach revealed impairments beyond cortical auditory processing areas in CHR-P and FEP groups in RTHA and RHIP.The significant thalamic contribution toward 40-Hz ASSR deficits is consistent with several lines of evidence.Thalamic 40-Hz ASSRs have been observed with positron emission tomography, functional magnetic resonance imaging, and EEG/MEG (8,9,48).Furthermore, experimentally induced thalamic evoked gamma oscillations have been shown to have a direct effect on the auditory cortex ASSR responses to click trains (49).Similarly, the hippocampus exhibits an intrinsic 40-Hz rhythm that can be entrained by both internal network (50) and external signals such as ASSR stimulation (7,51).
Importantly, there is consistent evidence for circuit changes implicating deficits in GABAergic neurotransmission in ScZ in auditory as well as thalamic and hippocampal regions.PV1 interneurons are reduced in the hippocampus (52) and thalamic reticular nucleus (53) in ScZ, while in the auditory cortex, levels of the GABA-synthesizing enzyme GAD65 are decreased (54).Altered synaptic proteins implicating AMPA receptor subunits but not NMDARs have been shown to be altered in the auditory cortex in ScZ (55).Currently, the precise contributions of glutamatergic neurotransmission and GABAergic interneurons toward circuit deficits and aberrant gamma-band oscillations remain unclear, however.One possibility is that circuit deficits are due to a primary dysfunction in inhibitory interneurons in ScZ (56).In addition, evidence exists that impaired inhibition could be the result of NMDAR hypofunctioning on PV1 interneurons (57) or reduced NMDAR drive on pyramidal cells (58).
Crucially, our finding is that 40-Hz ASSR impairments in CHR-P participants represent a potential biomarker for predicting clinical outcomes.Specifically, we show that reductions in 40-Hz ASSRs predicted transition to psychosis in CHR-P participants with good accuracy similar to data from event-related potentials (59) or functional magnetic resonance imaging (60).In addition, MEG-recorded 40-Hz ASSRs distinguished CHR-P participants with persistence of APSs at 12months from those who remitted with excellent accuracy.Predicting persistence of APSs is relevant because there is evidence to suggest that CHR-P participants who continue to experience APSs have poorer outcomes than CHR-P participants with transient APSs (61).
The finding that 40-Hz ASSR in RTHA predicts persistence of APSs is important, because only a minority of CHR-P participants will develop psychosis (62), and a large number of individuals will remit from CHR-P status (63).Moreover, this finding is consistent with previous findings that thalamic abnormalities are a potential biomarker for early detection and diagnosis (60).Finally, we did not find 40-Hz ASSR impairments in CHR-N individuals, suggesting that impaired gammaband oscillations are specifically associated with the CHR-P and FEP phenotypes.

Strengths and Limitations
This study has several limitations.First, we localized MEG activity to the thalamus and hippocampus.While localization of subcortical generators remains challenging, we would like to note that reconstruction of MEG time courses in the thalamus and hippocampus have been demonstrated before by our group (64,65) and others (66,67).Moreover, our analyses revealed that there were no differences in similar depth regions in the opposite hemisphere, suggesting that signal leakage did not contribute to our observations.Second, the sample size of transitioned cases was relatively small (n = 13).Accordingly, further studies are required that replicate our findings in independent samples using nested cross-validation (68).
In conclusion, this study provides novel evidence on deficits in 40-Hz ASSRs in CHR-P participants and patients with FEP in auditory as well as thalamic and hippocampal regions.Crucially, the current findings highlight that 40-Hz ASSRs predict clinical outcomes in CHR-P participants, including transition to psychosis as well as persistence of APSs.Together, these findings highlight the potential of MEG as a novel approach to identify circuit dysfunctions and biomarker for clinical outcomes in psychosis.

ACKNOWLEDGMENTS AND DISCLOSURES
The study was supported by the Medical Research Council (Grant No. MR/ L011689/1) and the ERA-NET project 01EW2007A.RK was supported by the Neurosciences Foundation.

a
Except for sex statistical testing, which are based on c 2 tests, all other tests are based on nonparametric Kruskal-Wallis H-tests: a = 0.05, twosided, adjusted for ties, post hoc Bonferroni-corrected for multiple comparisons.b BACS scores for clinical groups were standardized to control group data, controlled for sex.c Multiple categories possible.

40 -
Hz ASSRs During Emerging Psychosis hoc pairwise comparisons Bonferroni-corrected for type I errors.BACS data were first z-normalized to the HC data.
Analyses of CHR-P Individuals.Correlations between 40-Hz ASSR data and clinical/cognitive baseline assessment data (CAARMS severity, CAARMS perceptual abnormality, SPI-A severity, SPI-A acoustic perceptual disturbances, Global Assessment of Functioning [GAF], global functioning [GF] role and social, BACS token motor, symbol coding, and composite scores) in the CHR-P group were investigated using linear regression.In addition, linear discriminant analyses (LDAs) were conducted (SPSS version 26; IBM Corp., Armonk, NY) to assess classification accuracy of 40-Hz ASSR data for APS persistence and transition to psychosis in CHR-P participants.The analyses included the following contrasts: 1) APS-P (n = 34) versus APS-NP (n = 39) and 2) CHR-P-T (n = 13) versus CHR-P-NT (n = 97), using ITPC and amplitude measures from all ten 40-Hz entrained ROIs as dependent variables.The 40-Hz ASSR data were z-normalized to HC subjects.LDA results were cross-validated with a leave-one-out method.For both models, Wilks' lambda/c 2 tests, standardized canonical discriminant function coefficients, and a receiver operating characteristic curve were evaluated.

Figure 1 .
Figure 1.Task paradigm and brain regions showing stimulus-entrained 40-Hz responses.(A) Task paradigm and an example of 40-Hz narrow bandpass-filtered signals from the four main groups in right Heschl's gyrus.Data show clear entrainment to stimulation frequency in all main groups.(B) Whole-brain significant increases of 40-Hz induced power (250-2000 ms vs. 500 ms baseline) across all participants.Orange color represents F values uncorrected for multiple comparisons, whereas the red colors indicate false discovery rate-corrected brain activity.CHR-N, clinical-high-risk-negative clinical controls; CHR-P, CHR positive; FEP, patients with first-episode psychosis; HC, healthy control; L, left; R, right.

Figure 2 .
Figure 2. Main group differences in the 40-Hz auditory steady-state response (ASSR) signals.(A) Top row are group averaged 40-Hz amplitude data (relative change from baseline) from right hemisphere Heschl's gyrus (RHES), right thalamus (RTHA), and right hippocampus (RHIP), with error bars representing SEM.Stimulus onset at time 0 and offset at 2000 ms.Below the respective Cumming estimation plots with data distribution swarm plots and group difference data distributions (compared with healthy control [HC] subjects).*False discovery rate-corrected significant contrasts.(B) As panel (A) but for intertrial phase coherence (ITPC) data in RHES.In addition, time-frequency plots of ITPC data are shown for RHES for HC and the contrasts of HC with all three main clinical groups.Data represent relative change (relch) from baseline, similar to the amplitude data.(C) The main four regions of interest for which virtual channel data were computed and statistically examined for group differences.CHR-N, clinical-high-risk negative; CHR-P, CHR positive; FEP, first-episode psychosis; RSTG, right superior temporal gyrus.

Table 2 .
FEP-Specific Clinical Data