Mismatch negativity and polygenic risk scores for schizophrenia and bipolar disorder

Objective: Auditory mismatch negativity (MMN) impairment is a candidate endophenotype in psychotic disorders, yet the genetic underpinnings remain to be clarified. Here, we examined the relationships between auditory MMN and polygenic risk scores (PRS) for individuals with psychotic disorders, including schizophrenia spectrum disorders (SSD) and bipolar disorder (BD) and in healthy controls (HC). Methods: Genotyped and clinically well-characterized individuals with psychotic disorders (n = 102), including SSD (n = 43) and BD (n = 59), and HC (n = 397) underwent a roving MMN paradigm. In addition MMN, we measured the memory traces of the repetition positivity (RP) and the deviant negativity (DN), which is believed to reflect prediction encoding and prediction error signals, respectively. SCZ and BD PRS were computed using summary statistics from the latest genome-wide association studies. The relationships between the MMN, RP, and DN and the PRSs were assessed with linear regressions. Results: We found no significant association between the SCZ or BD PRS and grand average MMN in the psychotic disorders group or in the HCs group (all p > 0.05). SCZ PRS and BD PRS were negatively associated with RP in the psychotic disorders group ( β = (cid:0) 0.46, t = (cid:0) 2.86, p = 0.005 and β = (cid:0) 0.29, t = (cid:0) 0.21, p = 0.034, respectively). No significant associations were found between DN and PRS. Conclusion: These findings suggest that genetic variants associated with SCZ and BD may be associated with MMN subcomponents linked to predictive coding among patients with psychotic disorders. Larger studies are needed to confirm these findings and further elucidate the genetic underpinnings of MMN impairment in psychotic disorders.


Introduction
Schizophrenia (SCZ) and bipolar disorder (BD) are severe mental illnesses with a combined prevalence of ~3 % (McGrath et al., 2008).The disorders are highly heritable (Gordovez and McMahon, 2020;Trifu et al., 2020), have onset in late adolescence or early adulthood (Solmi et al., 2022), are associated with markedly reduced life expectancy (Hjorthoj et al., 2017;Chan et al., 2022), and large societal costs (Collaborators, 2022).Despite decades of advanced research on SCZ and BD, the precise neural underpinnings of these disorders remain unknown and established biomarkers for routine clinical practice are missing (Friston et al., 2016;Stahl, 2018;Sigitova et al., 2017).To facilitate the study of biological foundations of complex psychiatric disorders, the endophenotype concept was introduced by Gottesman and Shields in the 1970's.Endophenotypes are intermediate neurobehavioral phenotypes with a more proximal relation to genes than the corresponding clinical constructs (Gottesman and Gould, 2003;Gottesman and Shields, 1973).
One candidate neurophysiological endophenotype for psychotic disorders is the auditory mismatch negativity (MMN) (Light et al., 2015;Thaker, 2008).MMN is an event-related potential (ERP) is a negative wave in the summated electroencephalogram with a maximal amplitude over frontocentral scalp regions peaking 100-250 ms following an unexpected sensory stimulus within a stream of familiar events.MMN is typically induced in auditory stimulus paradigms by a rare (deviant) sound when preceded by two or more identical standard sounds (Naatanen et al., 2007;Garrido et al., 2009).
MMN is the difference between an increasingly positive wave elicited by recurring standard stimuli, i.e., the repetition positivity (RP) (Haenschel et al., 2005;Baldeweg et al., 2004;McCleery et al., 2019;Cooper et al., 2013), and a negative wave elicited by a deviant stimulus, i.e., the deviant negativity (DN) (McCleery et al., 2019).The amplitude of both the RP and the DN increase with a higher number of standard stimuli prior to the deviant, and the slopes of these increments are termed the memory trace effect (Baldeweg et al., 2004;Haenschel et al., 2005;Javitt et al., 1998;McCleery et al., 2019).The memory trace of RP is believed to reflect the strength of encoding of a standard stimulus and indexes the degree of expectation of its recurrence (McCleery et al., 2019;Garrido et al., 2009;Haenschel et al., 2005;Baldeweg et al., 2004), whereas the DN is presumed to reflect the magnitude of detected violations of sensory regularities (Heilbron and Chait, 2018;Garrido et al., 2009;McCleery et al., 2019).The use of a roving MMN paradigm (Cowan et al., 1993;Winkler et al., 1996) allows for the analyses of the RP and DN (McCleery et al., 2019;Baldeweg, 2006;Baldeweg et al., 2004;Cowan et al., 1993).In roving paradigms, a standard stimulus is presented repeatedly until a new and physically different stimulus is presented, which elicits an MMN.The new stimulus is, however, also repeatedly presented and eventually encoded as a new standard.Thus, every change in stimulus elicits a new MMN and the memory trace (MMN-MT) can henceforth be computed by estimating the degree of change of MMN amplitude according to the number of prior standards before a new stimulus (Cowan et al., 1993;Baldeweg and Hirsch, 2015;McCleery et al., 2019;Fryer et al., 2020).Previous studies of the memory trace effects for MMN, RP, and DN in SCZ have reported inconsistent results (Baldeweg and Hirsch, 2015;Baldeweg et al., 2004;Javitt et al., 1998;Baldeweg et al., 2006;Coffman et al., 2017;McCleery et al., 2019).
It has been increasingly acknowledged that the genetic underpinnings of electrophysiological endophenotypes mirrors the complexity of their related behavioral traits (Iacono et al., 2017).Consistently, genome-wide associations studies (GWAS) of electrophysiological endophenotypes have revealed a large number of single nucleotide polymorphisms (SNP) with low effect sizes (Iacono et al., 2017;Flint and Munafo, 2007;Iacono et al., 2014).Likewise, recent GWAS underscore the polygenic nature of SCZ (Ripke et al., 2014;Trubetskoy et al., 2022) and BD (Mullins et al., 2020;Stahl et al., 2019) and a high degree of genetic overlap between the two disorders (Brainstorm et al., 2018;Cross-Disorder Group of the Psychiatric Genomics et al., 2013).The large number of common variants associated with these disorders has enabled the computation of disease-related polygenic risk scores (PRS), which represent aggregates of phenotypeassociated SNPs weighted by their effect sizes at different significance thresholds (Crouch and Bodmer, 2020).PRS thus gives an estimate of individual polygenic load in relation to the phenotype in question (Collister et al., 2022;Konuma and Okada, 2021).
Despite the well-documented impairment of MMN in psychotic disorders and their substantial heritability, no previous study has examined the relationship between MMN and SCZ or BD PRS.In the present study, we examined the relationships between PRS and MMN, RP, and DN amplitudes in individuals with SCZ or BD and HC.Based on the hypothesis that MMN impairment may in part reflect a genetically determined vulnerability to develop psychotic disorders (Earls et al., 2016;Jessen et al., 2001;Ranlund et al., 2016;Michie et al., 2002), but see (Erickson et al., 2016), we hypothesized that higher SCZ and BD PRS would be associated with impairments the MMN, RP, and DN in both patients and controls.

Participants
The current work was based on the Thematically Organized Psychosis (TOP) research study.We recruited 43 participants with schizophrenia spectrum disorders (SSD) which included the diagnoses of schizophrenia: n = 20; psychosis not otherwise specified (NOS): n = 16; schizoaffective disorder: n = 3; schizophreniform disorder: n = 4.We also included 59 participants with BD (BD type 1: n = 27; BD type 2: n = 31; BD NOS: n = 1).The patients were recruited from psychiatric hospitals and outpatient clinics in the greater Oslo area.All participants were between 18 and 65 years and met DSM IV criteria for a psychotic or bipolar disorder.There was a prerequisite that the patient had capacity for informed consent.Exclusion criteria for the patients were IQ < 70, neurological illness, brain injury or history of severe head trauma with loss of consciousness or other significant medical illness affecting brain function, as was mental disorders attributable to the effects of a substance or another medical condition.HCs (n = 397) were recruited through national records, social media, and advertisements in a regional newspaper in the same catchment area.The exclusion criteria for HCs were the same as for the patients and also included a history of drug or alcohol abuse or dependency, psychosis, BD, or major depressive disorder, or having a first degree relative diagnosed with a psychotic or bipolar disorder.

Clinical assessments
All patients were examined by a trained clinician with clinical interviews for DSM-IV.
(SCID-I) (Association, 2013), Positive And Negative Syndrome Scale (PANSS) (Kay et al., 1987) and Inventory of Depressive Symptoms -Clinician version (IDS-C) (Rush et al., 1996).Mood was further assessed on the day of the EEG recording using Montgomery-Aasberg Depression Rating Scale (MADRS) (Montgomery and Asberg, 1979) and Young Mania Rating Scale (YMRS) (Young et al., 1978).Age of onset, lifetime number of psychotic and affective symptoms and level of functioning, as assessed by the Global Assessment of Functioning (GAF-F/S, split version) (Pedersen et al., 2007) were obtained.

EEG acquisition and preprocessing
Scalp EEG data was amplified (− 3 dB at 417 Hz low-pass, DC- Pentz et al. coupled) and digitized (2048 Hz) from 72 Ag/AgCl active electrodes, including 64 active channels positioned according to the international 10-5 system, using a Biosemi Active-Two amplifier (BioSemi, Amsterdam, The Netherlands).All electrode offsets were kept below ±30 μV.
The offline analyses were conducted in Matlab 2017a (Mathworks Inc., Natick, MA) using EEGLAB version 14.1.2b(Delorme and Makeig, 2004) and in-house scripts.The data was first low-pass filtered at 40 Hz, then downsampled to 512 Hz, subjected to a high-pass filter at 0.5 Hz, and rereferenced to the average of left and right mastoids.Line noise and bad channels were removed using the PREP pipeline (Bigdely-Shamlo et al., 2015) and affected channels were interpolated using a robust average reference.The mean number of interpolated channels was 7.0 (standard deviation (SD) 5.1, range 1-29).Epochs were then extracted from -100 ms to 400 ms relative to stimulus onset.Independent component analysis (ICA) was conducted using binica and independent components representing artifacts were identified using ICLabel (Pion-Tonachini et al., 2019).Components with label probabilities <30 % brain and >50 % non-brain were removed, thus resulting in the removal of 22.3 components on average (SD 6.2, range 6-59).Epochs were then baselinecorrected by subtracting the mean amplitude from -50 ms to 0 ms relative to stimulus onset and epochs with amplitudes exceeding ±100 μV were rejected.

Mismatch negativity paradigm
The participants were seated in a comfortable chair during the EEG data collection.Hearing threshold was tested on both sides using a pure sinusoidal 1000 Hz tone and participants with a threshold of >40 dB were excluded.MMN was obtained by a roving paradigm, scripted and presented using PsychToolbox in Matlab (Brainard, 1997) while the participants were reading a magazine to divert attention from the presented stimuli.Pure sinusoidal tones were presented binaurally (80 dB, 5 ms rise and fall times) by earphones (ER-2, Etymotic Research, Inc., Elk Grove Village, IL, USA) with a fixed stimulus onset asynchrony of 400 ms.In the roving paradigm, trains of identical standard auditory stimuli with regards to pitch and duration were presented in a pseudorandom fashion with 2, 6, or 18 repetitions alternating with a new train of stimuli with different physical properties.This makes the first tone of a new train acting as a deviant stimulus relative to last tone of the previous stimulus train, as described by Baldeweg and colleagues (Baldeweg et al., 2004).The pitch of the employed 24 tones ranged from 700 to 1250 Hz, and the duration of the tones was either 50 or 100 ms.For each new tone sequence, both the frequency and the duration of the tones changed, thus making the first tone of a new sequence a "double deviant", accounting for 11.5 % of the total number of stimuli.
Average ERP waves from the FCz electrode were computed both for the deviant and the immediately preceding standard stimulus, as well as the difference wave by subtracting the former from the latter.MMN was defined as the mean amplitude of the difference wave between 100 and 200 ms post-stimulus.The roving paradigm allows for the extraction of separate mismatch responses to the deviants presented after 2, 6, or 18 standard repetitions, hereafter referred to as 2-MMN, 6-MMN, and 18-MMN, respectively, and the MMN memory trace (MMN-MT), here defined as the 18-MMN minus the 2-MMN.The mean numbers of trials for 2-MMN, 6-MMN, and 18-MMN were 84.7 (SD 5.3, range 24-89), 81.4 (SD 5.2, range 20-88), and 83.6 (SD 5.2, range 24-90), respectively.A grand average MMN was calculated by averaging the 2-, 6-, and 18-MMNs.The mean amplitudes between 100 and 200 ms post-stimulus for the 2nd and 18th standard repetition waveforms were extracted and RP was calculated by subtracting the former from the latter.Similarly the mean amplitudes of the ERPs of the deviant following the 2nd and 18th standard was extracted and the DN was calculated by subtracting the former from the latter.
During data collection, we discovered an error in the Matlab-based script, which resulted in the presentation of a 1050 Hz 50 ms tone instead of a 100 ms tone.This affected 6 of the 91 2-repetitions trials, 10 of the 88 6-repetitions trials, and 10 of the 90 18-repetitions trials, in 245 of the participants.These trials were excluded from the data of these individuals.

Polygenic risk scores
DNA was obtained by blood samples from all participants.DNA analysis was performed separately for each participant at deCODE Genetics, Reykjarvik, Iceland using either Human OmniExpress-12, Human OmniExpress-24 or Global Screening Array kits (Illumna Inc., San Diego, CA, USA).Intensity files were then converted in GenomeStudio and imported to PLINK (Purcell et al., 2007) for quality control.Individuals were removed if they had discordant sex information, coverage <80 % and heterozygosity rates >5 standard deviation.Variants were only retained if genotyping rates were >5 %, Hardy-Weinberg equilibrium p > 1 × 10 − 5 , and non-significant batch effects (FDR > 0,5).Quality controlled genotypes were then phased using Eagle (Loh et al., 2016), and missing variant were imputed with Minimac3 (Das et al., 2016) using data from Haplotype Reference Consortium (HRC) (Loh et al., 2016), version 1.1 as reference.All variants with information score lower than 0.8 and minor allele frequency lower than 1 % were removed.Individual genotypes with <75 % confidence were set to missing.Variants in linkage equilibrium (R 2 < 0.1) were then selected to impute each individual sex and compute genetic principal components for each participant.
For computation of PRS for the SCZ and BD phenotype, we used PRS-CS, that utilizes a Bayesian framework and places continuous shrinkage priors on the effect sizes of the SNPs in the discovery GWAS summary statistics (Ge et al., 2019), and Plink 2.0 (Purcell et al., 2007).PRS-CS has been shown to outperform existing methods across a wide range of genetic architectures (Ge et al., 2019).PRS were computed for all participants regardless of ethnicity using summary statistics from two recent GWAS of SCZ (Trubetskoy et al., 2022) (n = 304,404; 76,785 cases, 243,649 controls) and BD (Mullins et al., 2021) (n = 413,466; 41,917 cases, 371,549 controls), from which local samples had been excluded.The European samples from the 1000 Genomes Phase III were used as the LD reference panel (Genomes Project et al., 2015).

Statistical analyses
All statistical analyses were run in R version 4.2.2 ('R Core Team (2020).R: A language and environment for statistical computing.R Foundation for Statistical Computing, Vienna, Austria.URL https:// www.R-project.org/.')https://www.R-project.org/('R Core Team (2020).R: A language and environment for statistical computing.R Foundation for Statistical Computing, Vienna, Austria.URL https:// www.R-project.org/.').First, to compare clinical and demographic variables between groups we ran analysis of variance (ANOVA) and t-tests for continuous variables and chi-square test of independence for categorical variables.Where homogeneity of variance or normality was violated, Welch ANOVA was used followed by Tukey's HSD post-hoc tests.We then ran linear regressions with MMN as dependent variable against PRS for SCZ and BD in the total sample.Age, sex, group (patients vs. controls), genotyping batch and the first 10 genetic principal components were used as covariates.The two-tailed significance level corrected for eight analyses (SCZ and BD PRS and grand average MMN, MMN-MT, RP, and DN) was set to α = 0.05/8 = 0.00625.Any significant group x PRS interactions were followed by post-hoc analyses run separately for HCs, for the combined patient group (BD + SCZ), BD and SSD.

Demographics and clinical variables
HC had significantly higher mean age compared to BD and SSD and A.B. Pentz et al. there was a predominance of female participants in HC and BD, but not in SSD.As expected, the SSD group had significantly lower IQ and had higher scores on the scales of psychotic symptom burden and lower on global functioning compared to the BD group (Table 1).The participants of the present work are also part of a larger study of MMN in SSD and BD where grand average MMN was significantly reduced in SCZ when compared to HCs and BD (Pentz et al., 2023).Further, we found a significant effect of diagnosis on grand average MMN amplitude (F(2,494) = 3.264, p = 0.039) and reduced MMN in SSD vs. BD (p = 0.043), but no significant difference in SSD vs. HCs (p = 0.48), nor between BD and HCs (p = 0.08).There was no significant effect of diagnosis on MMN-MT, RP or DN (all p's > 0.1), (Supplementary Figs.1-3).

Discussion
In the present study we investigated the relationship between psychotic disorder PRS and MMN, DN and RP, with three main findings.First, within the combined patient group we found negative associations between SCZ PRS and BD PRS and RP, indicating that higher PRS may be associated with RP impairment.Second, also within the combined patient group, we found a negative association between BD PRS and the memory trace of the MMN (MMN-MT), suggesting that higher PRS are Fig. 3. Scatterplots visualizing the associations between mismatch negativity(MMN) and schizophrenia-PRS in (A) the combined patient group and (B) within BD, HC and SSD (C) associations between bipolar disorder-PRS and MMN in the combined patient group and (D) within BD, HC, SSD.PRS = polygenic risk score, BD = bipolar disorder, HC = healthy controls, SSD = schizophrenia spectrum disorder.linked to reduced MMN enhancement with increasing number of standards prior to deviant presentation.Finally, we did not find significant associations between grand average MMN and SCZ or BD PRS.

The RP, DN, and MMN memory traces
The MMN memory trace effect in psychotic samples has only been investigated in a few studies up to now, some of which have shown reduced MMN-MT in SCZ (Javitt et al., 1998;Baldeweg and Hirsch, 2015;Baldeweg et al., 2004), and this reduction may be evident only in patients with poorer cognitive functioning (Baldeweg et al., 2004;Baldeweg and Hirsch, 2015).
The MMN memory trace has been suggested to reflect a short-term plasticity process (Baldeweg and Hirsch, 2015), that is impaired by ketamine and therefore assumed to be NMDAR-dependent (Schmidt et al., 2012).A recent study did, however, not replicate reduced MMN-MT in SCZ (McCleery et al., 2019).In our study, we found a trend level association with BD PRS and reduced MMN-MT as well as a negative association between SCZ PRS and BD PRS with RP in the combined patient group.This may suggest that common variants associated with psychotic disorders are also associated with prediction error signaling as well as with the generation of sensory predictions itself.
The same direction of effect were found for both SCZ and BD PRS.Whereas this is consistent with the relatively high correlation between the two polygenic risk scores (r = 0.56) found in our sample, it was somewhat surprising that the association between the memory trace indices was stronger for BD PRS than SCZ PRS.While we are not aware of any previous studies investigating MMN-MT or RP in BD, MMN is clearly more impaired in SCZ than in BD (Umbricht and Krljes, 2005;Erickson et al., 2016;Hermens et al., 2018).We would therefore expect any associations to be stronger for SCZ PRS than for BD PRS.However, the relatively small sample in our study and the extensive genetic overlap between SCZ and BD (Cross-Disorder Group of the Psychiatric Genomics et al., 2013;Grotzinger et al., 2022) makes it plausible that minor differences in the study population may yield significant results in one PRS and not the other.The results therefore suggest that MMN-MT and RP may be genetically influenced indices of predictive coding in psychotic disorders, transcending diagnostic boundaries.

MMN and polygenic risk scores
We did not find a correlation between PRS scores and grand average MMN in the total sample, nor did we find evidence suggesting that polygenic risk scores influence MMN amplitude differently in patients versus controls.The genetic underpinning of MMN in psychotic disorders has until now been evaluated by comparing MMN amplitudes in first degree relatives of probands with SCZ.The results have been mixed (Jessen et al., 2001;Michie et al., 2002;Ranlund et al., 2016;Bramon et al., 2004;Donaldson et al., 2021;Hong et al., 2012;Magno et al., 2008;Nishimura et al., 2016;Erickson et al., 2016;Earls et al., 2016).Two meta-analyses have been conducted (Erickson et al., 2016;Earls et al., 2016).Erickson et al. found a modest non-significant reduction (d = 0.26, p = 0.053) of MMN in first relatives of SCZ compared to HC, and not statistically different from patients with first episode psychosis.The authors concluded that whereas MMN impairment may possibly be a weak marker of genetic risk of schizophrenia, it mainly reflects transition to or progression of the disease (Erickson et al., 2016).However, Earls et al. (Earls et al., 2016) did indeed find reduced MMN in relatives of SCZ, but this was only significant at the frontal midline (FCz) electrode.FCz is the location where the MMN amplitude tends to be largest (Naatanen and Kahkonen, 2009;Kujala et al., 2007).This is also where the MMN deficit is most pronounced in SCZ (Koshiyama et al., 2021a;Baldeweg et al., 2002), hence the location where group differences is most likely to be detected (Javitt et al., 1998).Together, these findings may indicate clustering of MMN impairment within SCZ families, but the effect size appears low (Erickson et al., 2016;Earls et al., 2016).The question then arises whether we have a large enough sample size to detect such an effect.Polygenic risk scores explain low percentages of the phenotypic variance for severe mental disorders, 7.7 % for SCZ (Legge et al., 2021) and 4.75 % for BD (Mullins et al., 2021).Thus, studies with larger sample sizes and improved GWAS may be needed to clarify the relationship between MMN and polygenic risk scores.

MMN: a candidate endophenotype?
An endophenotype is associated with the illness in the population, is heritable, mainly state-independent, and co-segregate with the illness within families (Gottesman and Gould, 2003;Juli et al., 2021;Iacono, 2018).Despite the somewhat inconsistent results in first degree relatives of SCZ reviewed above, twin studies appear to show that MMN is highly heritable (Hall et al., 2006a;Hall et al., 2006b) in non-psychiatric samples.Furthermore genome-level studies in healthy participants has suggested that variants related to glutamatergic neurotransmission may influence MMN amplitude (Kawakubo et al., 2011;Bhat et al., 2021;Lin et al., 2014).Our findings showing associations between psychotic disorders PRS and RP suggest that the addition of MMN subcomponents and their memory trace indices may improve the endophenotypic valence of MMN.Previous research has shown that MMN-MT was associated with reduced working and episodic memory and reduced only in SCZ, but not in Alzheimer's disease, at conventional interstimulus intervals compared to HC, despite considerably worse cognitive abilities in the latter group, indicating that MMN-MT may tap into disorder-specific synaptic pathology (Baldeweg and Hirsch, 2015).While apparently also dependent upon NMDAR function (Rosch et al., 2019), the memory trace of RP, but not that of DN, has been found to be augmented by nicotinergic stimulation (Baldeweg et al., 2006), making it plausible that the encoding of prediction and prediction error signaling are modulated by different transmitter systems (Baldeweg et al., 2006), and at different levels in the auditory processing chain (Garrido et al., 2008;Koshiyama et al., 2021b).

Strengths and limitations
The current findings come with several strengths and limitations.Concerning the strengths, the present study included a well characterized cohort of SCZ and BD patients.Another strength is the use of a roving paradigm that is effective in controlling for physical standarddeviant differences, which is not controlled for in conventional oddball paradigms.Regarding the limitations, the patient sample of the present study is of moderate size and with limited statistical power, thus increasing the probability of both type 1 and type 2 errors.Furthermore, our work included a relatively high-functioning patient sample with moderate impairment which may have lower polygenic risk scores compared to more severely affected individuals (Lin et al., 2023).This may have further reduced our ability to detect associations between MMN and PRS.Another limitation is that we only investigated people of European ancestry in the PRS analyses.Thus, we cannot generalize the results to non-Europeans.

Conclusion
In the current study we did not find significant associations between PRS of SCZ and BD and MMN amplitude.There was however a significant association between BD PRS and reduced memory trace effect of MMN in the combined patient groups.Furthermore, SCZ PRS was associated with reduced memory trace of the repetition positivity (RP) in the combined patient group, with a trend in the same direction for BD PRS.The results suggests that the addition of the memory trace indices may add endophenotypic valence to MMN in severe mental illnesses.

Declaration of competing interest
Torbjørn Elvsåshagen.is a consultant to BrainWaveBank and Synovion and received speaker's honoraria from Lundbeck and Janssen Cilag.
Ole Andreassen. is a consultant to HealthLytix and received speaker's honoraria from Lundbeck.The other authors declare no competing interests.

Fig. 4 .
Fig. 4. Scatterplots visualizing associations between the memory trace of mismatch negativity(MMN-MT) and schizophrenia-PRS in (A) the combined patient group and (B) within BD, HC and SSD (C) associations between bipolar disorder-PRS and MMN in the combined patient group and (D) within BD, HC, SSD.PRS = polygenic risk score, BD = bipolar disorder, HC = healthy controls, SSD = schizophrenia spectrum disorder.

Fig. 5 .
Fig. 5. Scatterplots visualizing the associations between the memory trace of repetition positivity(RP) and schizophrenia-PRS in (A) the combined patient group and (B) within BD, HC and SSD (C) associations between bipolar disorder-PRS and RP in the combined patient group and (D) within BD, HC, SSD.PRS polygenic risk score, BD = bipolar disorder, HC = healthy controls, SSD = schizophrenia spectrum disorder.

Table 1
Clinical and demographic characteristics.