O1.3. A COMPUTATIONAL TRIAL-BY-TRIAL EEG ANALYSIS OF HIERARCHICAL PRECISION-WEIGHTED PREDICTION ERRORS

Abstract Background Action optimisation relies on learning about past decisions and on accumulated knowledge about the stability of the environment. In Bayesian models of learning, belief updating is informed by multiple, hierarchically related, precision-weighted prediction errors (pwPEs). Recent work suggests that hierarchically different pwPEs may be encoded by specific neurotransmitters such as dopamine (DA) and acetylcholine (ACh). Abnormal dopaminergic and cholinergic modulation of N-methyl-D-aspartate (NMDA) receptors plays a central role in the dysconnection hypothesis, which considers impaired synaptic plasticity a central mechanisms in the pathophysiology of schizophrenia. Methods To probe the dichotomy between DA and ACh and to investigate timing parameters of pwPEs, we tested 74 healthy male volunteers performing a probabilistic reward associative learning task in which the contingency between cues and rewards changed over 160 trials between 0.8 and 0.2. Furthermore, the current study employed pharmacological interventions (amisulpride / biperiden / placebo) and genetic analyses (COMT and ChAT) to probe DA and ACh modulation of these computational quantities. The study was double-blind and between-subject. We inferred, from subject-specific behavioural data, a low-level choice PE about the reward outcome, a high-level PE about the probability of the outcome as well as the respective precision-weights (uncertainties) and used them, in a trial-by-trial analysis, to explain electroencephalogram (EEG) signals (64 channels). Behavioural data was modelled implementing three versions of the Hierarchical Gaussian Filter (HGF), a Rescorla-Wagner model, and a Sutton model with a dynamic learning rate. The computational trajectories of the winning model were used as regressors in single-subject trial-by-trial GLM analyses at the sensor level. The resulting parameter estimates were entered into 2nd-level ANOVAs. The reported results were family-wise error corrected at the peak-level (p<0.05) across the whole brain and time window (outcome phase: 0 - 500ms). Results A three-level HGF best explained the data and was used to compute the computational regressors for EEG analyses. We found a significant interaction between pharmacology and COMT for the high-level precision-weight (uncertainty). Specifically: - At 276 ms after outcome presentation the difference between Met/Met and Val/Met was more positive for amisulpride than for biperiden over occipital electrodes. - At 274ms and 278 ms after outcome presentation the difference between Met/Met and Val/Met was more negative over fronto-temporal electrodes for amisulpride than for placebo, and for amisulpride than for biperiden, respectively. No significant results were detected for the other computational quantities or for the ChAT gene. Discussion The differential effects of pharmacology on the processing of high-level precision-weight (uncertainty) were modulated by the DA-related gene COMT. Previous results linked high-level PEs to the cholinergic basal forebrain. One possible explanation for the current results is that high-level computational quantities are represented in cholinergic regions, which in turn are influenced by dopaminergic projections. In order to disentangle dopaminergic and cholinergic effects on synaptic plasticity further analyses will concentrate on biophysical models (e.g. DCM). This may prove useful in detecting pathophysiological subgroups and might therefore be of high relevance in a clinical setting.

as potentially identifying candidate biomarkers to guide diagnosis, treatment and prognosis. Whilst specialized inflammatory marker assays have been found to be associated with outcome and treatment, these tests are not typically available in clinical practice. We sought to establish whether routine inflammatory markers are associated with clinical characteristics and outcomes in patients with schizophrenia and related disorders. Methods: A multi-site cohort study of patients admitted to an acute psychiatric ward between January 2013 and December 2016 within a large Mental Health Trust was undertaken. Cases were identified from an electronic database containing full clinical records. Inclusion criteria were patients aged 18 and 65 years with a discharge diagnosis of schizophrenia, or related disorder and a routine blood test within 3 days of admission. Exclusion criteria were diagnoses of drug-induced psychosis, organic brain disorder, or admission during the perinatal period. Pro-inflammatory (white blood cell total and differential count, C-reactive protein) and anti-inflammatory markers (albumin) recorded during the admission were extracted. Clinical characteristics were based upon the Health of the Nation Outcome Scale, a 12-item clinician rated tool contemporaneously rated at admission and discharge. Results: A total of 968 patients met the inclusion criteria. 309 patients were female and mean age was 38 years. The most frequent ethnicities were White, Black African, Black Other and Black Caribbean and the commonest diagnoses were schizophrenia, unspecified non-organic psychosis and schizoaffective disorder. Mean interval from admission to admission blood test was 0.8 days. Patients with affective psychosis had a significantly higher white cell count, monocyte count and lymphocyte count than patients with non-affective psychosis on admission. Furthermore, among patients with affective psychosis, a partial correlation controlling for age, body mass index, blood pressure, physical health and smoking status found a significant association between symptom severity and monocyte count. There was a highly significant association between both neutrophil count and white cell count with hallucinatory symptoms. There was also a highly significant positive association between C-reactive protein and self-injurious behaviour, replicating recently published findings in smaller samples. There was a significant reduction in overall psychiatric symptoms over the course of admission, which was significantly associated with admission monocyte count. A partial correlation found white cell count and neutrophil count at admission were associated with reductions in hallucinatory symptoms. Eosinophil count was significantly associated with admission length. Discussion: In a large cohort of patients admitted due to psychotic disorder, pro-inflammatory markers were associated with affective psychosis and overall symptom severity, and predicted admission length and reduction in symptom severity. The study supports an association between immune dysregulation and psychosis. Furthermore, the study highlights the role of routinely and inexpensively measured peripheral inflammatory markers as potential diagnostic and prognostic biomarkers in psychosis.

ETH Zurich; 2 University of Zurich & ETH Zurich
Background: Action optimisation relies on learning about past decisions and on accumulated knowledge about the stability of the environment. In Bayesian models of learning, belief updating is informed by multiple, hierarchically related, precision-weighted prediction errors (pwPEs). Recent work suggests that hierarchically different pwPEs may be encoded by specific neurotransmitters such as dopamine (DA) and acetylcholine (ACh). Abnormal dopaminergic and cholinergic modulation of N-methyl-D-aspartate (NMDA) receptors plays a central role in the dysconnection hypothesis, which considers impaired synaptic plasticity a central mechanisms in the pathophysiology of schizophrenia. Methods: To probe the dichotomy between DA and ACh and to investigate timing parameters of pwPEs, we tested 74 healthy male volunteers performing a probabilistic reward associative learning task in which the contingency between cues and rewards changed over 160 trials between 0.8 and 0.2. Furthermore, the current study employed pharmacological interventions (amisulpride / biperiden / placebo) and genetic analyses (COMT and ChAT) to probe DA and ACh modulation of these computational quantities. The study was double-blind and between-subject. We inferred, from subject-specific behavioural data, a low-level choice PE about the reward outcome, a high-level PE about the probability of the outcome as well as the respective precision-weights (uncertainties) and used them, in a trial-by-trial analysis, to explain electroencephalogram (EEG) signals (64 channels). Behavioural data was modelled implementing three versions of the Hierarchical Gaussian Filter (HGF), a Rescorla-Wagner model, and a Sutton model with a dynamic learning rate. The computational trajectories of the winning model were used as regressors in single-subject trial-by-trial GLM analyses at the sensor level. The resulting parameter estimates were entered into 2nd-level ANOVAs. The reported results were family-wise error corrected at the peak-level (p<0.05) across the whole brain and time window (outcome phase: 0 -500ms). Results: A three-level HGF best explained the data and was used to compute the computational regressors for EEG analyses. We found a significant interaction between pharmacology and COMT for the high-level precisionweight (uncertainty). Specifically: -At 276 ms after outcome presentation the difference between Met/Met and Val/Met was more positive for amisulpride than for biperiden over occipital electrodes. -At 274ms and 278 ms after outcome presentation the difference between Met/Met and Val/Met was more negative over fronto-temporal electrodes for amisulpride than for placebo, and for amisulpride than for biperiden, respectively.
No significant results were detected for the other computational quantities or for the ChAT gene. Discussion: The differential effects of pharmacology on the processing of high-level precision-weight (uncertainty) were modulated by the DA-related gene COMT. Previous results linked high-level PEs to the cholinergic basal forebrain. One possible explanation for the current results is that high-level computational quantities are represented in cholinergic regions, which in turn are influenced by dopaminergic projections. In order to disentangle dopaminergic and cholinergic effects on synaptic plasticity further analyses will concentrate on biophysical models (e.g. DCM). This may prove useful in detecting pathophysiological subgroups and might therefore be of high relevance in a clinical setting.

O1.4. CEREBROSPINAL FLUID FINDINGS IN TWINS WITH PSYCHOTIC SYMPTOMS -NOVEL FINDINGS AND FUTURE PROSPECTS
Viktoria Johansson* ,1 1 Karolinska Institutet Background: Schizophrenia and bipolar disorder are severe mental disorders with unknown etiology. Our research group has studied biomarkers in the cerebrospinal fluid (CSF) of twins with schizophrenia and bipolar disorder to be able to determine the genetic and environmental influences. In brain disorders, CSF is the most appropriate substrate to study as it may reflect the brain biochemistry better than blood. In this presentation I aim to give an overview of our findings and their relation to psychotic disorders. I intend to present our most recent preliminary finding and to discuss future prospects. Methods: We studied CSF-markers from a cohort of 50 monozygotic (MZ) and dizygotic (DZ) twins with schizophrenia or bipolar disorder. The twins have gone through diagnostic assessments and have been extensively phenotyped with questionnaires, symptom scales for psychiatric symptoms as well as neuropsychological testing. We have analyzed monoamines, microglia-, neurodegenerative-, kynurenine-, and inflammatory markers using immunoassays and high-performance liquid chromatography techniques. We have also studied microscopic structures with scanning electron microscopy. Results: One of our main findings was that soluble cluster of differentiation 14 protein (sCD14) was higher in twins with schizophrenia or bipolar disorder compared to their not affected co-twins. A later analysis showed that the difference within the discordant twin-pairs was higher in the DZ twin pairs (β=28697.1, t=3.20, p=0.024) compared with the MZ twin pairs (β=5577.5, t=2.10, p=0.081) suggesting that genetic components along with unique environmental effects have an influence on the higher sCD14 levels in patients with schizophrenia and bipolar disorder. We also found that sCD14 was higher in those patients with more psychotic symptoms. In our study on microscopic structures in CSF we found that the structures were prevalent not only in the patients with schizophrenia and bipolar disorder but also in their not affected co-twins. The finding suggests that genetic factors may be partly involved in the formation of the structures. Discussion: We have analyzed inflammatory and neurodegerative markers in the CSF of twins with psychotic disorders to be able to study genetic and environmental influences. Our results indicate that sCD14 may have an influence on microglia activation in psychosis. We have continued with analyses on the correlations between all the markers, the monoamine metabolites and associations with symptoms and cognitive ability and the preliminary results from these analyses will be presented.
To conclude CSF analyses for biomarkers in twins may result in extended knowledge regarding the genetic and environmental relationships. Our unique twin data gives us the possibility to study CSF-markers in relation to psychiatric symptoms and cognitive measures. For future studies it would be of interest to assemble twin-samples from several research groups to be able to study research questions regarding gene and environment interactions. Background: Schizophrenia is a disabling and often unremitting mental illness with an unknown cause that is characterized by heterogeneity in psychotic symptom presentation, cognitive deficits and treatment response. There is accumulating evidence for the role of inflammation in the etiology of schizophrenia. Inflammatory markers have been identified in the brains and peripheral blood of chronically ill patients with schizophrenia and in first episode patients and these markers have been associated with structural and functional brain abnormalities and cognitive deficits. Intercellular adhesion molecule 1 (ICAM-1) is a transmembrane protein expressed on endothelial cells which binds to leukocyte receptors that promotes transmigration of white blood cells into tissue. While peripheral inflammatory markers are altered in people with schizophrenia relative to controls, the extent to which ICAM-1 is elevated in the brains of people with schizophrenia and peripheral levels of soluble ICAM-1 (sICAM-1) is increased in relation to cognitive impairment in schizophrenia is unknown. Methods: In a post-mortem cohort, 8 mRNAs relating to BBB function and 3 immune cell markers were measured by qPCR in the prefrontal cortex of