Hostname: page-component-cd9895bd7-jn8rn Total loading time: 0 Render date: 2024-12-13T16:49:48.692Z Has data issue: false hasContentIssue false

Personality changes during adolescence predict young adult psychosis proneness and mediate gene–environment interplays of schizophrenia risk

Published online by Cambridge University Press:  28 October 2024

Linda A. Antonucci
Affiliation:
Department of Translational Biomedicine and Neuroscience – University of Bari Aldo Moro, Bari, Italy
Alessandra Raio
Affiliation:
Department of Translational Biomedicine and Neuroscience – University of Bari Aldo Moro, Bari, Italy
Gianluca Christos Kikidis
Affiliation:
Department of Translational Biomedicine and Neuroscience – University of Bari Aldo Moro, Bari, Italy
Alessandro Bertolino
Affiliation:
Department of Translational Biomedicine and Neuroscience – University of Bari Aldo Moro, Bari, Italy Psychiatry Unit – Policlinico di Bari, Bari, Italy
Antonio Rampino
Affiliation:
Department of Translational Biomedicine and Neuroscience – University of Bari Aldo Moro, Bari, Italy Psychiatry Unit – Policlinico di Bari, Bari, Italy
Tobias Banaschewski
Affiliation:
Department of Child and Adolescent Psychiatry and Psychotherapy, Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
Arun L.W. Bokde
Affiliation:
Discipline of Psychiatry, School of Medicine and Trinity College Institute of Neuroscience, Trinity College Duin, Dublin, Ireland
Sylvane Desrivières
Affiliation:
Centre for Population Neuroscience and Precision Medicine (PONS), Institute of Psychiatry, Psychology & Neuroscience, SGDP Centre, King's College London, UK
Herta Flor
Affiliation:
Institute of Cognitive and Clinical Neuroscience, Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany Department of Psychology, School of Social Sciences, University of Mannheim, Mannheim, Germany
Antoine Grigis
Affiliation:
NeuroSpin, CEA, Université Paris-Saclay, Gif-sur-Yvette, France
Hugh Garavan
Affiliation:
Departments of Psychiatry and Psychology, University of Vermont, Burlington, Vermont, USA
Andreas Heinz
Affiliation:
Department of Psychiatry and Psychotherapy CCM, Charité – Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany
Jean-Luc Martinot
Affiliation:
Institut National de la Santé et de la Recherche Médicale, INSERM U 1299 “Trajectoires développementales & psychiatrie”, University Paris-Saclay, CNRS, France Ecole Normale Supérieure Paris-Saclay, Centre Borelli; Gif-sur-Yvette, France
Marie-Laure Paillère Martinot
Affiliation:
Institut National de la Santé et de la Recherche Médicale, INSERM U 1299 “Trajectoires développementales & psychiatrie”, University Paris-Saclay, CNRS, France Ecole Normale Supérieure Paris-Saclay, Centre Borelli; Gif-sur-Yvette, France Sorbonne University, Department of Child and Adolescent Psychiatry, Pitié-Salpêtrière Hospital, Paris; France
Eric Artiges
Affiliation:
Institut National de la Santé et de la Recherche Médicale, INSERM U 1299 “Trajectoires développementales & psychiatrie”, University Paris-Saclay, CNRS, France Ecole Normale Supérieure Paris-Saclay, Centre Borelli; Gif-sur-Yvette, France Psychiatry Department, EPS Barthélémy Durand, Etampes; France
Frauke Nees
Affiliation:
Department of Child and Adolescent Psychiatry and Psychotherapy, Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany Institute of Cognitive and Clinical Neuroscience, Central Institute of Mental Health, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany Institute of Medical Psychology and Medical Sociology, University Medical Center Schleswig Holstein, Kiel University, Kiel, Germany
Dimitri Papadopoulos Orfanos
Affiliation:
NeuroSpin, CEA, Université Paris-Saclay, Gif-sur-Yvette, France
Luise Poustka
Affiliation:
Department of Child and Adolescent Psychiatry and Psychotherapy, University Medical Centre Göttingen, Göttingen, Germany
Sarah Hohmann
Affiliation:
Department of Child and Adolescent Psychiatry, Psychotherapy and Psychosomatics, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
Juliane H. Fröhner
Affiliation:
Department of Psychiatry and Neuroimaging Center, Technische Universität Dresden, Dresden, Germany
Michael N. Smolka
Affiliation:
Department of Psychiatry and Neuroimaging Center, Technische Universität Dresden, Dresden, Germany
Nilakshi Vaidya
Affiliation:
Centre for Population Neuroscience and Stratified Medicine (PONS), Department of Psychiatry and Neuroscience, Charité Universitätsmedizin Berlin, Germany
Henrik Walter
Affiliation:
Department of Psychiatry and Psychotherapy CCM, Charité – Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany
Robert Whelan
Affiliation:
School of Psychology and Global Brain Health Institute, Trinity College Dublin, Ireland
Gunter Schumann
Affiliation:
Centre for Population Neuroscience and Stratified Medicine (PONS), Department of Psychiatry and Neuroscience, Charité Universitätsmedizin Berlin, Germany Centre for Population Neuroscience and Precision Medicine (PONS), Institute for Science and Technology of Brain-inspired Intelligence (ISTBI), Fudan University, Shanghai, China
Catharina A. Hartman
Affiliation:
Interdisciplinary Center Psychopathology and Emotion regulation, Department of Psychiatry, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
Giulio Pergola*
Affiliation:
Department of Translational Biomedicine and Neuroscience – University of Bari Aldo Moro, Bari, Italy Lieber Institute for Brain Development, Johns Hopkins Medical Campus, Baltimore, MD, USA Department of Psychiatry and Behavioral Science – John Hopkins University, Baltimore, MD, USA
*
Corresponding author: Giulio Pergola; Email: giulio.pergola@uniba.it
Rights & Permissions [Opens in a new window]

Abstract

Background

Psychotic symptoms in adolescence are associated with social adversity and genetic risk for schizophrenia. This gene–environment interplay may be mediated by personality, which also develops during adolescence. We hypothesized that (i) personality development predicts later Psychosis Proneness Signs (PPS), and (ii) personality traits mediate the association between genetic risk for schizophrenia, social adversities, and psychosis.

Methods

A total of 784 individuals were selected within the IMAGEN cohort (Discovery Sample-DS: 526; Validation Sample-VS: 258); personality was assessed at baseline (13–15 years), follow-up-1 (FU1, 16–17 years), and FU2 (18–20 years). Latent growth curve models served to compute coefficients of individual change across 14 personality variables. A support vector machine algorithm employed these coefficients to predict PPS at FU3 (21–24 years). We computed mediation analyses, including personality-based predictions and self-reported bullying victimization as serial mediators along the pathway between polygenic risk score (PRS) for schizophrenia and FU3 PPS. We replicated the main findings also on 1132 adolescents recruited within the TRAILS cohort.

Results

Growth scores in neuroticism and openness predicted PPS with 65.6% balanced accuracy in the DS, and 69.5% in the VS Mediations revealed a significant positive direct effect of PRS on PPS (confidence interval [CI] 0.01–0.15), and an indirect effect, serially mediated by personality-based predictions and victimization (CI 0.006–0.01), replicated in the TRAILS cohort (CI 0.0004–0.004).

Conclusions

Adolescent personality changes may predate future experiences associated with psychosis susceptibility. PPS personality-based predictions mediate the relationship between PRS and victimization toward adult PPS, suggesting that gene–environment correlations proposed for psychosis are partly mediated by personality.

Type
Original Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
Copyright © The Author(s), 2024. Published by Cambridge University Press

Introduction

Current research refers to psychosis proneness signs (PPS) as unusual psychotic-like experiences (PLEs), like perceptual abnormalities or persecutory ideation (Bourgin et al., Reference Bourgin, Tebeka, Mallet, Mazer, Dubertret and Le Strat2020). PPS occur in both at-psychosis-risk (Bonnett, Varese, Smith, Flores, & Yung, Reference Bonnett, Varese, Smith, Flores and Yung2019) and non-clinical populations (Loch et al., Reference Loch, Wang, Rossler, Tofoli, Silveira and Andrade2011): up to 26.7% of the general population reports at least one PPS (Bourgin et al., Reference Bourgin, Tebeka, Mallet, Mazer, Dubertret and Le Strat2020), without seeking psychiatric help (Yung et al., Reference Yung, Nelson, Baker, Buckby, Baksheev and Cosgrave2009). Interestingly, PPS occur more frequently among young individuals (McGrath et al., Reference McGrath, Saha, Al-Hamzawi, Alonso, Bromet, Bruffaerts and Kessler2015) until young adulthood. This age stage also represents the typical onset period for psychosis-spectrum disorders, including schizophrenia (Solmi et al., Reference Solmi, Radua, Olivola, Croce, Soardo, Salazar de Pablo and Fusar-Poli2022). Although PPS do not meet the threshold for full-blown psychosis (Mark & Toulopoulou, Reference Mark and Toulopoulou2016), their occurrence during adolescence may precede later psychotic onset (Nuevo, Van Os, Arango, Chatterji, & Ayuso-Mateos, Reference Nuevo, Van Os, Arango, Chatterji and Ayuso-Mateos2013). When PPS persist into young adulthood, the individual risk for full-blown psychosis increases (McGrath et al., Reference McGrath, Saha, Al-Hamzawi, Alonso, Andrade, Borges and Kessler2016). Thus, PPS represent an early target to investigate how psychosis susceptibility develops over time (Howes & Murray, Reference Howes and Murray2014). Investigating PPS may refine early identification and prediction of individual risk.

In the context of the high schizophrenia heritability, estimated up to 80% (Lichtenstein et al., Reference Lichtenstein, Yip, Bjork, Pawitan, Cannon, Sullivan and Hultman2009), exposure to adverse environmental factors contributes to the risk of developing schizophrenia and other psychoses (van Os, Kenis, & Rutten, Reference van Os, Kenis and Rutten2010). For schizophrenia, the gap between the genetic risk explainable in terms of genetic variants and the heritability estimates in twins is remarkable. Thus, disentangling different types of gene-environment interplay (Zwicker, Denovan-Wright, & Uher, Reference Zwicker, Denovan-Wright and Uher2018) is important for early identification and primary prevention (Pergola, Penzel, Sportelli, & Bertolino, Reference Pergola, Penzel, Sportelli and Bertolino2023; Plomin, DeFries, & Loehlin, Reference Plomin, DeFries and Loehlin1977; Uher & Zwicker, Reference Uher and Zwicker2017): indeed, gene–environment correlations require the environment to enact genetic risk; the environment may, in turn, be modulated by early intervention programs.

Notably, schizophrenia genetic risk loci have previously been associated with ‘neuroticism’ and ‘openness’ personality traits (Smeland et al., Reference Smeland, Wang, Lo, Li, Frei, Witoelar and Andreassen2017). Personality traits are heritable (Vukasović & Bratko, Reference Vukasović and Bratko2015), and they develop during adolescence (Roberts, Caspi, & Moffitt, Reference Roberts, Caspi and Moffitt2001). Late adolescence is also the period in which PPS typically emerge. Adolescent PPS have been associated with personality characteristics, especially with schizotypal traits (Barrantes-Vidal, Racioppi, & Kwapil, Reference Barrantes-Vidal, Racioppi, Kwapil, Thompson and Broome2020; Fonseca-Pedrero, Ortuno-Sierra, Inchausti, Rodriguez-Testal, & Debbane, Reference Fonseca-Pedrero, Ortuno-Sierra, Inchausti, Rodriguez-Testal and Debbane2019). However, prior research did not examine the identified relationships in the frame of gene–environment interplay models to explain psychopathology development. We reasoned that static (e.g. genetics) and dynamic (e.g. personality) intrinsic factors might be modulated by extrinsic factors (e.g. social adversities) when contributing to PPS. Pergola et al. (Reference Pergola, Papalino, Gelao, Sportelli, Vollerbergh, Grattagliano and Bertolino2019) already reported evidence of an evocative gene-environment correlation in schizophrenia. In that study, the polygenic risk score (PRS) for schizophrenia (PGC2 wave), cumulating the effects of many risk variants, was associated with the subsequent frequency of sub-clinical psychotic symptoms. This effect was mediated by peer and teacher reports of an established schizophrenia environmental factor, i.e. bullying victimization (Armitage et al., Reference Armitage, Morneau-Vaillancourt, Pingault, Andlauer, Paquin, Langevin and Boivin2022; Woolway et al., Reference Woolway, Smart, Lynham, Lloyd, Owen, Jones and Legge2022). However, it is still unexplained how genetic factors could act on the social environment; one hypothesis considers evocative gene-environment correlations (Lella, Antonucci,, & Pergola, Reference Lella, Antonucci and Pergola2023). Peer victimization is not necessary or sufficient per se for a diagnosis of schizophrenia; it may represent an environmental risk factor potentially correlated with genetic risk for schizophrenia (Woolway et al., Reference Woolway, Smart, Lynham, Lloyd, Owen, Jones and Legge2022). We hypothesized that adolescent personality evolution may interact with such exposure in predicting psychosis vulnerability, and potentially be associated with genetic susceptibility, as well (Millan et al., Reference Millan, Andrieux, Bartzokis, Cadenhead, Dazzan, Fusar-Poli and Weinberger2016). Importantly, genetics, social challenges, and personality evolution are intertwined, and their relationship may be non-linear, e.g. detectable only in the presence of high genetic risk (Pergola et al., Reference Pergola, Papalino, Gelao, Sportelli, Vollerbergh, Grattagliano and Bertolino2019). Non-linear interactions between multiple variables are best captured by machine learning approaches when aiming to achieve individual-level predictions (Dwyer, Falkai, & Koutsouleris, Reference Dwyer, Falkai and Koutsouleris2018). Therefore, we aimed to deliver longitudinal calculators of psychosis risk that quantify the predictive potential of individual personality trajectories by combining multivariate and change-tracking techniques (i.e. Latent Growth Curve Models-LGCM).

Here, we aimed to investigate whether personality development during adolescence plays a role in gene–environment interplays predicting subsequent PPS. We analyzed a naturalistic population recruited within the IMAGEN consortium, generating a multivariate risk calculator based on adolescent patterns of personality change aimed at predicting PPS severity in adulthood. Next, we associated our algorithm predictions with the risk of developing emotional and behavioral disorders to explore the wider clinical relevance of our personality-based risk calculator. Finally, we investigated the association of polygenic risk for schizophrenia with subsequent PPS severity. We hypothesized that this relationship was mediated by personality-based multivariate predictions and bullying. To test our findings' generalizability, we replicated the main results on another independent cohort of young individuals (TRAILS). The study is outlined in Fig. 1.

Figure 1. The outline of the study. (1) Individual trajectories of change during adolescence based on personality traits were estimated via latent growth curve models. The two derived trait-related coefficients of change per participant fed a machine learning algorithm as longitudinal predictors of psychosis proneness signs (PPS) in young adulthood. Longitudinal decision scores extracted from the generated models were used: (2) to predict clinical outcomes other than PPS in adulthood; (3) as a longitudinal interface between individual polygenic risk for schizophrenia and bullying victimization across the pathway toward final PPS.

(Figure representing machine learning analyses adapted from Dwyer et al., Reference Dwyer, Falkai and Koutsouleris2018).

PPS, Psychosis Proneness Signs.

Methods

Sample and assessment

We selected 784 individuals recruited within the IMAGEN study (Schumann et al., Reference Schumann, Loth, Banaschewski, Barbot, Barker and Buchel2010), from a wider naturalistic cohort of 13-years-old adolescents undergoing four assessment waves, at Baseline (BL) and at two, four, and six years after BL completion (i.e. follow up 1, 2, 3 – FU1, FU2, FU3) (see online Supplementary Information–SI, Section 1). Parents and adolescents gave written consent and verbal assent, respectively. We selected participants based on the full availability of the individual total score for the Community Assessment of Psychic Experiences – 42 (CAPE-42) (Stefanis et al., Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis, Stefanis and Van Os2002), used to assess PPS at FU3 as the last available time point (see online SI, Section 2). The IMAGEN cohort was randomly split (2:1 ratio) into a discovery sample (N = 526; Table 1A) and a validation sample (N = 258; Table 1B). To test our findings generalizability, we replicated IMAGEN models on an external cohort of 1546 adolescents (Table 1C), selected from the Tracking Adolescents' Individual Lives Survey (TRAILS) study (see online SI, Section 12.1), based on CAPE-42 full data availability at the last accessible time point (wave 3-w3). Details about subsequent steps of participant retention in both cohorts are depicted in a Consort Chart (online Supplementary Figure 1).

Table 1. Demographic characteristics of: (A) IMAGEN Discovery Sample; (B) IMAGEN Validation Sample; (C) TRAILS Replication cohort

BL, Baseline; FU1/FU2/FU3, Follow-Up 1/2/3; H-PPS, Higher-Psychosis Proneness Signs; L-PPS, Lower-Psychosis Proneness Signs; M/F, Male/Female; n.a., not assessed; SD, Standard Deviation. (*) marks p < 0.05, (**) marks p < 0.01, (***) marks p < 0.001.

In both cohorts, each individual was assigned to a higher- or lower-PPS group, based on the median of the FU3 CAPE-42 total score distribution, which for the IMAGEN cohort was calculated in the Discovery sample and used as a cut-off also in the Validation sample (so that data remained independent of cut-off determination). Two-sample t tests and χ2 tests assessed differences between Higher-PPS and Lower-PPS (IMAGEN Discovery/Validation samples: Table 1A/1B; TRAILS replication cohort: Table 1C; α = 0.05).

For each IMAGEN participant, a total of 14 personality scores (listed in online Supplementary Table 1) were computed at BL, FU1, and FU2 based on the items from the following self-report questionnaires (see online SI, Section 3 for details):

  • the NEO Five-Factory Inventory (Costa & McCrae, Reference Costa and McCrae1992), which assesses personality on five main dimensions (Neuroticism, Extraversion, Openness, Agreeableness, and Conscientiousness), i.e. the so-called Big Five traits (Raad & Perugini, Reference Raad and Perugini2002);

  • the Temperament and Character Inventory–Revised (Farmer & Goldberg, Reference Farmer and Goldberg2008), based on the Cloninger comprehensive model of temperament and character (Cloninger, Svrakic, & Przybeck, Reference Cloninger, Svrakic and Przybeck1993). The IMAGEN version provided measures for Novelty Seeking and its four temperamental subcomponents (exploratory excitability, impulsiveness, extravagance, and disorderliness);

  • the substance use risk profile scale (Woicik, Stewart, Pihl, & Conrod, Reference Woicik, Stewart, Pihl and Conrod2009), investigating the role of our main personality traits (Hopelessness, Anxiety Sensitivity, Impulsivity, Sensation Seeking) as potential risk factors for addictive behaviors and co-morbid psychopathology development.

Descriptive statistics and between-group comparisons, performed via two-sample t tests (α = 0.05), are reported for each selected score at each time point in online Supplementary Table 2.

The three IMAGEN waves of personality data collection were used as input variables to estimate individual trajectories of change in both the discovery and validation samples (see Methods, Section ‘Latent Growth Curve Models pipeline’).

In the TRAILS cohort, a total of 12 personality variables (listed in online Supplementary Table 13), including both w3 cross-sectional and w3-w1 longitudinal scores (see online SI, Section 12.2), were analyzed through the very same machine learning pipeline implemented on IMAGEN data (see Methods, Section ‘Machine learning pipeline’).

Generation of the longitudinal personality-based risk calculator

Latent growth curve models pipeline

To compute personality-based trajectories of individual change over time, we implemented LGCMs (Burant, Reference Burant2016) via the R statistics (https://cran.r-project.org/) ‘lavaan’ package. A total of 14 LGCMs were computed separately in both Discovery and Validation samples, one for each personality variable assessed at BL, FU1, and FU2 (listed in online Supplementary Table 1). Incomplete personality information at FU1 (missing data proportion: 7.03–9.33%) was handled via full information maximum likelihood (Allison, Reference Allison1987; Felt, Depaoli, & Tiemensma, Reference Felt, Depaoli and Tiemensma2017). FU3 personality information was excluded from LGCMs computation so that PPS outcomes were solely predicted based on prior time points.

LGCMs computation details are reported in online SI, Section 4.1. Details about goodness-of-fit indices and metrics are reported in online SI, Section 4.2.

From each of the 14 general LGCMs implemented in both the discovery and validation samples, two individual latent coefficients of change (i.e. an intercept factor and a slope factor for each participant) were derived. The 28 personality-based coefficients of change estimated per participant fed a supervised machine learning algorithm built to predict PPS severity levels at FU3 (see Methods, 2.2.2).

Machine learning pipeline

We carried out our machine learning analytic strategy via NeuroMiner, version 1.05 (www.pronia.eu/neurominer/) (Antonucci et al., Reference Antonucci, Pergola, Pigoni, Dwyer, Kambeitz-Ilankovic, Penzel and Bertolino2020; Koutsouleris et al., Reference Koutsouleris, Kahn, Chekroud, Leucht, Falkai, Wobrock and Hasan2016). The 28 individual personality-based coefficients of change computed via LGCMs (see Methods, Section ‘Latent Growth Curve Models pipeline’) fed a Support Vector Machine algorithm aimed at predicting Higher- v. Lower-PPS at FU3 in our Discovery sample. We adopted a repeated nested cross-validation design (Antonucci et al., Reference Antonucci, Raio, Pergola, Gelao, Papalino, Rampino and Bertolino2021; Koutsouleris et al., Reference Koutsouleris, Dwyer, Degenhardt, Maj, Urquijo-Castro and Sanfelici2021) (see online SI, Section 5 for details) to generate a multivariate personality-based longitudinal risk calculator of psychosis proneness. Permutation analyses served to assign statistical significance to both PPS and extra labels prediction performance (Golland & Fischl, Reference Golland and Fischl2003; Koutsouleris et al., Reference Koutsouleris, Kahn, Chekroud, Leucht, Falkai, Wobrock and Hasan2016) (see online SI, Section 6), measured through balanced accuracy (BAC). To assess the generalizability of our personality-based risk calculator, we applied the discovery prediction model to our independent validation sample via out-of-sample cross-validation (OOCV) (Antonucci et al., Reference Antonucci, Raio, Pergola, Gelao, Papalino, Rampino and Bertolino2021; Haas et al., Reference Haas, Antonucci, Wenzel, Ruef, Biagianti, Paolini and Kambeitz-Ilankovic2021).

To verify that generated predictions were not affected by potential confounds, we correlated individual prediction scores extracted from the discovery risk calculator with BL measures for cognitive performance and substance use (see online SI, Section 5.1. for details). All significant p values were <0.05, false discovery rate (FDR)-corrected (Benjamini & Hochberg, Reference Benjamini and Hochberg1995).

Personality-based predictions beyond psychosis proneness symptoms

To investigate whether the predictive performance of our personality-based risk calculator was associated with clinical readouts outside PPS, we used ANOVAs on four different categories of prediction (i.e. two groups of correctly predicted v. two groups of misclassified individuals for higher- or lower-PPS) and the risk to develop emotional, conduct, hyperactivity and any other disorders at FU3. Such risk was estimated for individuals included in both the discovery and validation samples through the Development and Well-being Assessment Algorithm (Goodman, Ford, Richards, Gatward, & Meltzer, Reference Goodman, Ford, Richards, Gatward and Meltzer2000) from the Strength and Difficulties Questionnaire items (Goodman, Reference Goodman1997). For details, see online SI, Sections 7,8.

Personality-based predictions within the schizophrenia risk trajectories

To assess the total effect between predictor and outcome, we preliminarily tested the association between polygenic liability for several psychiatric conditions and psychosis proneness via correlation analyses (see online SI, Section 11); then, we investigated how putative relationships between polygenic risk for schizophrenia, exposure to victimization, and personality-based predictions may influence PPS. To this aim, we carried out a serial mediation analysis and two moderated mediation analyses via R statistics, using PROCESS Macro v. 4 (Hayes, Reference Hayes2017) (see online SI, Section 11 for details). All the models, depicted in online Supplementary Figure 2, were implemented only on 653 individuals with available genetic data out of the 784 included participants (see online Supplementary Table 3 for descriptive statistics). Individual genetic risk for schizophrenia was estimated using genetic data to calculate a PRS for schizophrenia (Purcell et al., Reference International, Purcell, Wray, Stone, Visscher, O'Donovan and Sklar2009; Pergola et al., Reference Pergola, Papalino, Gelao, Sportelli, Vollerbergh, Grattagliano and Bertolino2019) (see online SI, Section 9). To stratify individuals for chronic victimization, we computed the rank product of victimization total scores at BL, FU1, and FU2 (listed in online Supplementary Table 1, descriptive statistics in online Supplementary Table 4), collected through the Bully Questionnaire (details in online SI, Section 10). This score entered in our mediation framework as a second-order mediator or a moderator factor.

To test the modeled pathway generalizability, the very same serial mediation design was implemented on 1132 TRAILS individuals with available genetic data out of the 1546 included participants (online Supplementary Table 15). We generated two different models, respectively including the rank product of victimization scores at w2 and w3 as reported:

Complete details are reported in online SI, Section 12.2 and 12.3.

Results

Demographic differences between samples

Demographic characteristics are separately reported for IMAGEN discovery and validation samples and for TRAILS replication cohort. The discovery sample featured a higher proportion of females in H-PPS than in L-PPS (χ2 = 5.28; p = 0.02) (Table 1A), whereas the TRAILS replication cohort featured a lower proportion (χ2 = 23.17; p < 0.001) (Table 1C). The validation sample included slightly older individuals in H-PPS than in L-PPS at FU2 (t = 2; p = 0.046) and at FU3 (t = 2.45; p = 0.015) (Table 1B).

Machine learning results

In the discovery sample, the risk calculator based on the 28 individual personality coefficients of change predicted higher-PPS v. lower-PPS at FU3 with a cross-validated BAC of 65.6% (p = 0.01) (detailed metrics in Table 2). Only PPS predictions showed a significant permuted p-value (p < 0.05). Features with the highest probability of being selected for prediction included individual intercept coefficients for neuroticism, followed by individual slope and intercept coefficients for openness (online Supplementary Figure 3). Figure 2 shows raw trajectories of change over time for neuroticism and openness for the first 10 individuals from the upper ⩾0.95 and the lower ⩽0.05 limits of the ensemble prediction probability scores distribution: the most prototypical individuals from, respectively, the higher-PPS and the lower-PPS prediction class, showed opposite longitudinal patterns (i.e. increasing v. decreasing) for neuroticism, whereas they showed only mean differences between higher v. lower openness levels over time.

Table 2. Validated performance of the personality-based risk calculator predicting Higher- v. Lower Psychosis Proneness Signs (PPS) in both Discovery and Validation samples

Figure 2. Raw trajectories of change over time for NEO neuroticism scores (left panel) and NEO openness scores (right panel) for the first ten individuals from the upper and the lower limits of the Ensemble Probability Prediction (EPP) scores distribution: notably, as an EPP score from 0.95 above estimated a probability to be assigned to the Higher-Pyschosis Proneness Signs (PPS) class in the 95% of the generated models, we considered such individuals as highly prototypical for such severity class (red lines; the red line in bold depicts the group mean trajectory); as an EPP score from 0.5 below estimated a probability to be assigned to the Higher-PPS class in the 5% of the generated models, we considered such individuals as highly prototypical of the Lower-PPS severity class (green lines; the green line in bold depicts the group mean trajectory) BL, Baseline; FU1, Follow-Up 1; FU2, Follow-Up 2; NEO, NEO Five Factor Inventory.

OOCV results (Methods, Section ‘Machine learning pipeline’) revealed that the discovery risk calculator showed high personality-based prediction performance also when applied to validation unseen individuals (BAC: 69.5%; detailed metrics in Table 2). Neither the discovery nor the validation predictions demonstrated inflation due to information leakage (see online SI, Section 5.2. for details). No significant association between prediction scores from the discovery risk calculator and scores at BL for cognitive performance (online Supplementary Table 8) and substance use (online Supplementary Table 9) emerged, suggesting that predictions were not affected by neuropsychological or substance use-related variables.

Personality-based predictions beyond psychosis proneness symptoms

In both discovery and validation samples, ANOVAs revealed a significant main effect of the prediction rate on scores for emotional and hyperactivity disorders risk at FU3 (all p < 0.001, specific F-statistics reported in online Supplementary Figure 4). Only individuals correctly predicted by the algorithm as Higher-PPS at FU3 also showed significantly higher risk scores at the same time point: specifically, this pattern occurred only for emotional disorders risk in the Discovery sample (online Supplementary Figure 4A) and for both emotional and hyperactivity disorders risk in the Validation sample (online Supplementary Figure 4C and 4D) (all p < 0.001). Only in the Discovery sample, correctly predicted and misclassified Higher-PPS individuals did not differ from each other in terms of estimated risk for hyperactivity disorders, both showing significantly higher scores at FU3 (online Supplementary Figure 4B) (all p < 0.05). No significant effect of the prediction index on scores of estimated risk for conduct (Discovery F = 0.68; p = 0.56; Validation F = 0.21; p = 0.88) or other disorders (Discovery F = 0.84; p = 0.72; Validation F = 0.32; p = 0.61) emerged at FU3.

Personality-based predictions within the schizophrenia risk trajectories

Our preliminary correlation analyses indicated that only the PRS for schizophrenia was significantly associated with PPS (ρ = 0.12; p = 0.008, FDR-corrected). Thus, we included it as the only predictor in our mediation models. As shown in Fig. 3A, for the IMAGEN serial mediation model, the total effect of the schizophrenia PRS on FU3 PPS severity levels was significant (β = 0.14; p < 0.001; bootstrapped 95% CI 0.06–0.22). Moreover, the direct effect of the schizophrenia PRS on FU3 PPS severity levels was significant when both the mediators were taken into account (β = 0.08; p = 0.02; bootstrapped 95% CI 0.01–0.15; 57.1% of the total effect explained (Fairchild, Mackinnon, Taborga, & Taylor, Reference Fairchild, Mackinnon, Taborga and Taylor2009; VanderWeele, Reference VanderWeele2013)). The remaining 42.9% of the total effect of the schizophrenia PRS on PPS was funneled in an indirect effect. As shown in online Supplementary Table 10, a significant indirect effect emerged within the serial mediation pathway including both personality-based predictions and victimization (β = 0.004, bootstrapped 95% CI 0.0006–0.01; 2.8% of effect explained). Also the pathway including only personality-based predictions showed a significant indirect effect, confirming the association discovered via machine learning (β = 0.05, bootstrapped 95% CI 0.02–0.09; 35.7% of effect explained). Instead, no significant indirect effect emerged when victimization was considered as the only mediator across the PRS-PPS pathway (β = −0.002, bootstrapped 95% CI 0.004−0.01), suggesting the relevance of personality scores in the serial mediation. No moderation effects from victimization emerged on the pathways toward final PPS (see online SI, Section 11 for detailed results).

Figure 3. Findings from the serial mediation models, investigating the role of personality-based machine learning predictions and the rank product of Bullying Victimization (BV) within the pathway between polygenic risk for schizophrenia and final Psychosis Proneness Signs. Figure 3A depicted the model generated on IMAGEN data, Fig. 3B and 3C depicted replication models generated on TRAILS data, respectively including children-reported and parents-reported BV information. Direct effects (standardized coefficients) are shown. Red arrows represent relationships returning significant direct effects. The grey arrows represent not significant direct effects. Indirect effects for each model are reported in online Supplementary Table 10 (IMAGEN model), 17, and 18 (TRAILS replication models).

FU3, Follow-Up 3; PPS, Psychosis Proneness Signs; w3, wave 3. *marks p < 0.05; **marks p < 0.01; ***marks p < 0.001.

In the replication on TRAILS data, as shown in online Supplementary Table 17 and 18, we found a significant indirect effect within the serial mediation pathway including both personality-based predictions and victimization, independently of the informants (Fig. 3B model: β = 0.002, bootstrapped 95% CI 0.0004–0.004; 20% of variance explained; Figure 3C model: β = 0.001, bootstrapped 95% CI 0.0002–0.003; 10% of variance explained). Moreover, the pathway including only personality-based predictions showed a significant indirect effect in both serial mediation models (Fig. 3B model: β = 0.022, bootstrapped 95% CI 0.005–0.04; Figure 3C model: β = 0.02, bootstrapped 95% CI 0.006–0.04). On the other hand, no significant indirect effect emerged when victimization was considered as the only mediator across the PRS-PPS pathway (Fig. 3B model: β = −0.003, bootstrapped 95% CI −0.01 to −0.003; Figure 3C model: β = 0.002, bootstrapped 95% CI −0.003 to −0.008).

Discussion

This study aimed to investigate whether patterns of neurodevelopmental changes in dynamic intrinsic factors like personality may predict adult PPS and whether this prediction ability has a role in gene-environment interplays relevant to schizophrenia. After separating our population based on PPS severity levels, significant between-group differences for raw personality features emerged already at the cross-sectional level. However, findings from our longitudinal risk calculator showed that, in the Discovery sample, higher and lower PPS levels in young adulthood can be accurately predicted through personality and temperament trajectories of change across adolescence. Interestingly, these factors, traditionally considered relatively stable (Hampson & Goldberg, Reference Hampson and Goldberg2006), have been recently reconceptualized as sensitive to change and adaptation, especially throughout adolescence (Caspi, Roberts, & Shiner, Reference Caspi, Roberts and Shiner2005; Durbin et al., Reference Durbin, Hicks, Blonigen, Johnson, Iacono and McGue2016; Zohar, Zwir, Wang, Cloninger, & Anokhin, Reference Zohar, Zwir, Wang, Cloninger and Anokhin2019). Our longitudinal fingerprint supports this notion and, providing individual predictions, complements group-level cross-sectional evidence of associations between PPS in healthy adolescents and personality (Wiltink et al., Reference Wiltink, Nelson, Velthorst, Wigman, Lin and Baksheev2015) or temperament (Nitzburg et al., Reference Nitzburg, Gopin, Peters, Karlsgodt, Malhotra and DeRosse2016). The good validation performance of our risk calculator, when applied to unseen individuals, corroborates its generalizability potential. Indeed, the validation BAC of 69.5% is consistent with the standards reported in previous meta-analyses on machine learning prognostic models built on at-risk individuals (Sanfelici, Dwyer, Antonucci, & Koutsouleris, Reference Sanfelici, Dwyer, Antonucci and Koutsouleris2020).

Neuroticism (i.e. the vulnerability to emotional instability and negative emotionality (Begemann, Boyette,, Kwast,, & Sommer, Reference Begemann, Boyette, Kwast and Sommer2020)) and openness (i.e. the availability to new ideas and experiences (Begemann et al., Reference Begemann, Boyette, Kwast and Sommer2020)) emerged as the two personality traits most contributing to longitudinal predictions of PPS levels. Their prominent role in our algorithm is consistent with large-scale studies identifying genetic risk loci shared between schizophrenia and these two personality traits (Lo et al., Reference Lo, Hinds, Tung, Franz, Fan, Wang and Chen2017; Smeland et al., Reference Smeland, Wang, Lo, Li, Frei, Witoelar and Andreassen2017). Accordingly, these traits are associated with psychosis-related manifestations (i.e. symptoms or subclinical PPS severity, and perceived quality of life) in clinical (Franquillo et al., Reference Franquillo, Guccione, Angelini, Carpentieri, Ducci and Caretti2021) and healthy populations (Wiltink et al., Reference Wiltink, Nelson, Velthorst, Wigman, Lin and Baksheev2015). Notably, the evolution of these traits over adolescence contributed to the predictions, suggesting that tracking these traits over time enhances early identification. When we compared the most prototypical Higher-PPS and Lower-PPS individuals, the strict separation of longitudinal neuroticism trajectories between the two groups suggested that individuals with higher starting neuroticism scores, and consistently high scores over time, have higher future risk for psychosis. Higher neuroticism is frequently self-reported by individuals prone to negative emotionality and distress (Boyette et al., Reference Boyette, Korver-Nieberg, Verweij, Meijer, Dingemans, Cahn and de Haan2013). This trait may represent a trans-diagnostic risk factor for future psychopathology. High scores on openness have been associated with positive schizotypy and magical thinking (Wiltink et al., Reference Wiltink, Nelson, Velthorst, Wigman, Lin and Baksheev2015) in non-clinical samples (Larøi, DeFruyt,, van Os,, Aleman,, & Van der Linden, Reference Larøi, DeFruyt, van Os, Aleman and Van der Linden2005; Ross, Lutz, & Bailley, Reference Ross, Lutz and Bailley2002). High openness characterized Higher-PPS individuals in our study but evolved in more heterogeneous within- and between-group trajectories over time. In summary, our findings suggest that the evolution of specific personality traits during adolescence might contribute to future PPS. From a clinical perspective, self-reported neuroticism correlates with maladaptive strategies of emotion regulation (e.g. rumination, self-blaming, suppression (Ludwig, Werner, & Lincoln, Reference Ludwig, Werner and Lincoln2019)). The same phenomena are reported at different stages of psychosis (Chapman et al., Reference Chapman, Visser, Mittal, Gibb, Coles and Strauss2020; Vines et al., Reference Vines, Bridgwater, Bachman, Hayes, Catalano and Jalbrzikowski2022) and associated with delusion severity in psychotic patients (Garety et al., Reference Garety, Freeman, Jolley, Dunn, Bebbington, Fowler and Dudley2005). It follows that psychoeducational skills training, promoting effective management of negative emotions in such individuals, might contribute to delaying or softening the negative consequences of PPS.

Interestingly, although our algorithm was specific for PPS predictions, neuroticism and openness changes showed potential clinical relevance also for other risk conditions potentially associated with PPS. Indeed, individuals correctly classified as Higher-PPS based on personality changes consistently showed a significantly higher estimated risk of developing emotional and hyperactivity disorders in young adulthood, as also reported previously (Christiansen et al., Reference Christiansen, Beck, Bilenberg, Wienecke, Astrup and Lundbye-Jensen2019; Fox, Sheffield, & Woodward, Reference Fox, Sheffield and Woodward2021).

Finally, we integrated our longitudinal personality-based predictions in a serial mediation framework including both genetic and environmental features. We found that PPS predictions based on adolescent personality changes were associated with self-reported victimization, and both prediction and victimization serially mediated the relationship between schizophrenia PRS and adult PPS severity. The schizophrenia PRS was also the only one significantly associated with PPS. We cannot exclude that the large sample size of the latest schizophrenia GWAS and the high heritability of this disorder played a role in such a significant effect. Interestingly, the serial mediation effect also emerged when the same mediation design was replicated on the data collected within the external TRAILS cohort, regardless of the different victimization informants (e.g. children or their parents) (Lella et al., Reference Lella, Antonucci and Pergola2023) and further methodological differences, confirming the association discovered via machine learning. It is noteworthy that Pergola et al. (Reference Pergola, Papalino, Gelao, Sportelli, Vollerbergh, Grattagliano and Bertolino2019) described a similar gene-environment correlation without including personality. That study found a victimization-mediated effect of schizophrenia PRS on the frequency of PPS developed at the last accessible time point, including a correlation between PRS and bullying victimization. The PRS-bullying association in TRAILS was not significant when considering personality as a mediator. This evidence suggests that the mechanism of translating genetic risk into environmental risk involves personality traits and their development during adolescence. The lack of a direct PRS-PPS association in TRAILS, whereas it was significant in IMAGEN, is of interest to future research on the assessment of potential critical periods: indeed, demographic differences between the two cohorts (IMAGEN BL-FU3 age: 13.9–22 y.o.; TRAILS w1-w3 age: 11.1–16.2 y.o.; Table 1) overlap with the mixed results on genetic associations obtained in the ALSPAC cohort when data from 13-year olds were included (Schoeler et al., Reference Schoeler, Choi, Dudbridge, Baldwin, Duncan, Cecil and Pingault2019) or not (Riglin et al., Reference Riglin, Hammerton, Heron, Collishaw, Arseneault, Thapar and Thapar2019). Genetic effects on social patterns characteristic of adolescence may still be too small for detection with the sample sizes employed here in very young individuals. This age specificity would be consistent with other evidence of effects emerging in functional brain imaging only in later adolescence (Passiatore et al., Reference Passiatore, Antonucci, DeRamus, Fazio, Stolfa, Sportelli and Pergola2023; Zalesky et al., Reference Zalesky, Pantelis, Cropley, Fornito, Cocchi, McAdams and Gogtay2015). Nonetheless, the successful replication of the rest of the pathways within the model speaks in favor of its validity and generalizability.

In summary, our results explain the previously reported link (Pergola et al., Reference Pergola, Papalino, Gelao, Sportelli, Vollerbergh, Grattagliano and Bertolino2019) between victimization and genetic risk to develop later PPS as a mechanism based on personality. While the PRS of the risk carriers cannot influence the behavior of other individuals, it can influence personality traits and thus the behavioral manifestation of risk, hence explaining the significant associations with victimization. Personality appears as a privileged interface between the polygenic risk for schizophrenia and social adversities. Genetic risk carriers express personality traits predictive of PPS and also suffer a greater environmental burden. This hypothesis was further corroborated by the peer victimization significance as mediator and not moderator, tested on the main IMAGEN cohort data, at variance with previous evidence (Guloksuz et al., Reference Guloksuz, Pries, Delespaul, Kenis, Luykx, Lin and van Os2019). Thus, the environment in this model amplifies genetic risk as in an evocative gene-environment correlation framework, whereby the reactions of the peers to heritable traits expressed by the carrier enhance risk. The small indirect effect on final PPS captured by the risk pathway, including personality and victimization, suggests that the measures of social environment employed picked up only a very modest portion of the environmental contribution to PPS. Thus, while victimization accounts only for a minor proportion of PPS risk, other gene-environment correlation mechanisms may contribute to reduce the heritability gap, identify pathways of genetic risk to environmental exposure, and devise personalized interventions to foster resilience.

Limitations

To the best of our knowledge, our model of gene–environment correlation informed by personality is the first of its kind. Thus, despite the stringent cross-validation strategy employed and the external replication of our results implemented on TRAILS data, validations in wider naturalistic populations are needed to fulfill generalizability requirements and findings translation into clinical practice (Sanfelici et al., Reference Sanfelici, Dwyer, Antonucci and Koutsouleris2020). As bullying is not the only social stressor potentially contributing to psychosis risk, future replications might model PPS vulnerability pathways including additional core factors of social adversity, such as childhood trauma, discrimination, and exclusion (Varchmin, Montag, Treusch, Kaminski, & Heinz, Reference Varchmin, Montag, Treusch, Kaminski and Heinz2021). Additional replications are also recommended for data collections that provide (i) more than three time points to model nonlinear personality change (Durbin et al., Reference Durbin, Hicks, Blonigen, Johnson, Iacono and McGue2016), and (ii) a more extensive personality assessment, tapping into traits based on different personality theories.

Conclusions

Our findings outline a dynamic model of PPS risk development based on a ‘personality-enriched’ gene-environment correlation pathway: genetic factors may act on PPS through an effect on adolescent personality changes, which in turn could modulate the exposure to negative social interactions. For this reason, our findings highlight the importance of monitoring the evolution of personality traits over time. We found that the increase or stability of high neuroticism and openness across adolescence are relevant to future psychosis proneness. Personality development is strictly entangled with environmental challenges. Thus, we envision that real-world programs aimed at softening the downstream effects of an increased genetic risk for psychosis might be oriented toward (i) early identification and management of maladaptive social environments during adolescence and (ii) promoting adaptation to social adversities in subjects with high neuroticism and openness, particularly when victimized by peers.

Disclosures

Prof Pergola received lecture fees from Lundbeck. Prof. Bertolino received consulting fees from Biogen and lecture fees from Otsuka, Janssen, and Lundbeck. Prof. Rampino received travel fees from Lundbeck. Dr Banaschewski served in an advisory or consultancy role for eye level, Infectopharm, Lundbeck, Medice, Neurim Pharmaceuticals, Oberberg GmbH, Roche, and Takeda. He received conference support or speaker's fee by Janssen, Medice and Takeda. He received royalties from Hogrefe, Kohlhammer, CIP Medien, and Oxford University Press; the present work is unrelated to these relationships. Dr Poustka served in an advisory or consultancy role for Roche and Viforpharm and received speaker's fee by Shire. She received royalties from Hogrefe, Kohlhammer and Schattauer. The present work is unrelated to the above grants and relationships. The other authors report no biomedical financial interests or potential conflicts of interest.

Supplementary material

The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724002198.

Funding statement

This work received support from the funding initiative FAIR-Future Artificial Intelligence Research (PNRR ‘Partenariati Estesi’) for the project H97G22000210007 to GP and LAA, Horizon Europe Seeds 2021 (Next Generation EU - MUR D.M. 737/2021) for the project H99J21017550006 to GP, LAA, AR and AB, EXPRIVIA for the research program: ‘Artificial intelligence, genetics and transcriptomics’ to GP, and the Apulian regional government for the project: ‘Early Identification of Psychosis Risk’ to AB.

This work also received funding from the European Union through the FP6 Integrated Project IMAGEN (Reinforcement-related behaviour in normal brain function and psychopathology) (LSHM-CT- 2007-037286), the Horizon 2020 funded ERC Advanced Grant ‘STRATIFY’ (Brain network based stratification of reinforcement-related disorders) (695313), Human Brain Project (HBP SGA 2, 785907, and HBP SGA 3, 945539), the Medical Research Council Grant ‘c-VEDA’ (Consortium on Vulnerability to Externalizing Disorders and Addictions) (MR/N000390/1), the National Institute of Health (NIH) (R01DA049238, A decentralized macro and micro gene-by-environment interaction analysis of substance use behavior and its brain biomarkers), the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London, the Bundesministeriumfür Bildung und Forschung (BMBF grants 01GS08152; 01EV0711; Forschungsnetz AERIAL 01EE1406A, 01EE1406B; Forschungsnetz IMAC-Mind 01GL1745B), the Deutsche Forschungsgemeinschaft (DFG grants SM 80/7-2, SFB 940, TRR 265, NE 1383/14-1), the Medical Research Foundation and Medical Research Council (grants MR/R00465X/1 and MR/S020306/1), the National Institutes of Health (NIH) funded ENIGMA (grants 5U54EB020403-05 and 1R56AG058854-01), NSFC grant 82150710554 and European Union funded project ‘environMENTAL’, grant no: 101057429. Further support was provided by grants from: - the ANR (ANR-12-SAMA-0004, AAPG2019 - GeBra), the Eranet Neuron (AF12-NEUR0008-01 - WM2NA; and ANR-18-NEUR00002-01 - ADORe), the Fondation de France (00081242), the Fondation pour la Recherche Médicale (DPA20140629802), the Mission Interministérielle de Lutte-contre-les-Drogues-et-les-Conduites-Addictives (MILDECA), the Assistance-Publique-Hôpitaux-de-Paris and INSERM (interface grant), Paris Sud University IDEX 2012, the Fondation de l'Avenir (grant AP-RM-17-013), the Fédération pour la Recherche sur le Cerveau; the National Institutes of Health, Science Foundation Ireland (16/ERCD/3797), U.S.A. (Axon, Testosterone and Mental Health during Adolescence; RO1 MH085772-01A1) and by NIH Consortium grant U54 EB020403, supported by a cross-NIH alliance that funds Big Data to Knowledge Centres of Excellence. ImagenPathways ‘Understanding the Interplay between Cultural, Biological and Subjective Factors in Drug Use Pathways’ is a collaborative project supported by the European Research Area Network on Illicit Drugs (ERANID). This paper is based on independent research commissioned and funded in England by the National Institute for Health Research (NIHR) Policy Research Programme (project ref. PR-ST-0416-10001). The views expressed in this article are those of the authors and not necessarily those of the national funding agencies or ERANID.

This research is also part of the project ‘TRacking Adolescents' Individual Lives Survey (TRAILS)’. Participating centers of TRAILS include various departments of the University Medical Center and University of Groningen, the University of Utrecht, the Radboud University Medical Center, and the Parnassia Psychiatric Institute, all in the Netherlands. TRAILS has been financially supported by various grants from the Netherlands Organization for Scientific Research (NWO), ZonMW, GB-MaGW, the Dutch Ministry of Justice, the European Science Foundation, the European Research Council, BBMRI-NL, and the participating universities. We are grateful to everyone who participated in this research or worked on this project to make it possible.

Ethical standards

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Footnotes

*

Denotes equal contribution of the authors.

References

Achenbach, T. M., & Rescorla, L. A. (2006). The Achenbach System of Empirically Based Assessment. In R. P. Archer (Ed.), *Forensic uses of clinical assessment instruments* (pp. 229262). Mahwah, New Jersey, US: Lawrence Erlbaum Associates Publishers.Google Scholar
Allison, P. D. (1987). Estimation of linear models with incomplete data. Sociological Methodology, 17, 71103.CrossRefGoogle Scholar
Antonucci, L. A., Pergola, G., Pigoni, A., Dwyer, D., Kambeitz-Ilankovic, L., Penzel, N., … Bertolino, A. (2020). A pattern of cognitive deficits stratified for genetic and environmental risk reliably classifies patients with schizophrenia from healthy control subjects. Biological Psychiatry, 87(8), 697707. doi:10.1016/j.biopsych.2019.11.007CrossRefGoogle ScholarPubMed
Antonucci, L. A., Raio, A., Pergola, G., Gelao, B., Papalino, M., Rampino, A., … Bertolino, A. (2021). Machine learning-based ability to classify psychosis and early stages of disease through parenting and attachment-related variables is associated with social cognition. BMC Psychology, 9(1), 47. doi:10.1186/s40359-021-00552-3CrossRefGoogle ScholarPubMed
Armitage, J. M., Morneau-Vaillancourt, G., Pingault, J. B., Andlauer, T. F. M., Paquin, S., Langevin, S., … Boivin, M. (2022). A multi-informant and multi-polygenic approach to understanding predictors of peer victimisation in childhood and adolescence. Journal of Child Psychiatry & Psychology Advances, 2(1), e12063. doi:10.1002/jcv2.12063Google ScholarPubMed
Barrantes-Vidal, N., Racioppi, A., & Kwapil, T. R. (2020). Schizotypy, schizotypal personality, and psychosis risk. In Thompson, A. D. & Broome, M. R. (Eds.), Risk factors for psychosis (pp. 8199). San Diego, CA: Elsevier Academic Press.CrossRefGoogle Scholar
Begemann, M.J.H., Boyette, L.-L., Kwast, A.K., & Sommer, I.E.C. (2020). Personality across the psychosis continuum: A fine-grained perspective. Schizophrenia Bulletin Open, 1(1), sgaa064.CrossRefGoogle Scholar
Benjamini, Y., & Hochberg, Y. (1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B (Methodological), 57(1), 289300.CrossRefGoogle Scholar
Bonnett, L. J., Varese, F., Smith, C. T., Flores, A., & Yung, A. R. (2019). Individualised prediction of psychosis in individuals meeting at-risk mental state (ARMS) criteria: Protocol for a systematic review of clinical prediction models. Diagnostic and Prognostic Research, 3, 21. doi:10.1186/s41512-019-0066-5CrossRefGoogle ScholarPubMed
Bourgin, J., Tebeka, S., Mallet, J., Mazer, N., Dubertret, C., & Le Strat, Y. (2020). Prevalence and correlates of psychotic-like experiences in the general population. Schizophrenia Research, 215, 371377. doi:10.1016/j.schres.2019.08.024CrossRefGoogle ScholarPubMed
Boyette, L.-L., Korver-Nieberg, N., Verweij, K., Meijer, C., Dingemans, P., Cahn, W., & de Haan, L. (2013). Associations between the Five-Factor Model personality traits and psychotic experiences in patients with psychotic disorders, their siblings and controls. Psychiatry Research, 210(2), 491497.CrossRefGoogle ScholarPubMed
Burant, C. J. (2016). Latent growth curve models: Tracking changes over time. International Journal of Aging & Human Development, 82(4), 336350. doi:10.1177/0091415016641692CrossRefGoogle ScholarPubMed
Caspi, A., Roberts, B. W., & Shiner, R. L. (2005). Personality development: Stability and change. Annual Review of Psychology, 56, 453484. doi:10.1146/annurev.psych.55.090902.141913CrossRefGoogle ScholarPubMed
Chapman, H. C., Visser, K. F., Mittal, V. A., Gibb, B. E., Coles, M. E., & Strauss, G. P. (2020). Emotion regulation across the psychosis continuum. Development and Psychopathology, 32(1), 219227. doi:10.1017/S0954579418001682CrossRefGoogle ScholarPubMed
Christiansen, L., Beck, M. M., Bilenberg, N., Wienecke, J., Astrup, A., & Lundbye-Jensen, J. (2019). Effects of exercise on cognitive performance in children and adolescents with ADHD: Potential mechanisms and evidence-based recommendations. Journal of Clinical Medicine, 8(6), 841. doi:10.3390/jcm8060841CrossRefGoogle ScholarPubMed
Cloninger, C. R., Svrakic, D. M., & Przybeck, T. R. (1993). A psychobiological model of temperament and character. Archives of General Psychiatry, 50(12), 975990. doi:10.1001/archpsyc.1993.01820240059008CrossRefGoogle ScholarPubMed
Costa, P.T., & McCrae, R.R. (1992). Normal personality assessment in clinical practice: The NEO Personality Inventory. Psychological Assessment, 4(1), 5.CrossRefGoogle Scholar
Durbin, C. E., Hicks, B. M., Blonigen, D. M., Johnson, W., Iacono, W. G., & McGue, M. (2016). Personality trait change across late childhood to young adulthood: Evidence for nonlinearity and sex differences in change. European Journal of Personality, 30(1), 3144. doi:10.1002/per.2013CrossRefGoogle Scholar
Dwyer, D. B., Falkai, P., & Koutsouleris, N. (2018). Machine learning approaches for clinical psychology and psychiatry. Annual Review of Clinical Psychology, 14, 91118. doi:10.1146/annurev-clinpsy-032816-045037CrossRefGoogle ScholarPubMed
Fairchild, A. J., Mackinnon, D. P., Taborga, M. P., & Taylor, A. B. (2009). R2 effect-size measures for mediation analysis. Behavior Research Methods, 41(2), 486498. doi:10.3758/BRM.41.2.486CrossRefGoogle ScholarPubMed
Farmer, R.F., & Goldberg, L.R. (2008). A psychometric evaluation of the revised Temperament and Character Inventory (TCI-R) and the TCI-140. Psychological Assessment, 20(3), 281.CrossRefGoogle ScholarPubMed
Felt, J. M., Depaoli, S., & Tiemensma, J. (2017). Latent growth curve models for biomarkers of the stress response. Frontiers in Neuroscience, 11, 315. doi:10.3389/fnins.2017.00315CrossRefGoogle ScholarPubMed
Fonseca-Pedrero, E., Ortuno-Sierra, J., Inchausti, F., Rodriguez-Testal, J. F., & Debbane, M. (2019). Beyond clinical high-risk state for psychosis: The network structure of multidimensional psychosis liability in adolescents. Frontiers in Psychiatry, 10, 967. doi:10.3389/fpsyt.2019.00967CrossRefGoogle Scholar
Fox, V., Sheffield, J. M., & Woodward, N. D. (2021). Attention-deficit/hyperactivity disorder in youth with psychosis spectrum symptoms. Schizophrenia Research, 237, 141147. doi:10.1016/j.schres.2021.08.027CrossRefGoogle ScholarPubMed
Franquillo, A. C., Guccione, C., Angelini, G., Carpentieri, R., Ducci, G., & Caretti, V. (2021). The role of personality in schizophrenia and psychosis: A systematic review. Clinical Neuropsychiatry, 18(1), 2840. doi:10.36131/cnfioritieditore20210103Google ScholarPubMed
Garety, P. A., Freeman, D., Jolley, S., Dunn, G., Bebbington, P. E., Fowler, D. G., … Dudley, R. (2005). Reasoning, emotions, and delusional conviction in psychosis. Journal of Abnormal Psychology, 114(3), 373384. doi:10.1037/0021-843X.114.3.373CrossRefGoogle ScholarPubMed
Golland, P., & Fischl, B. (2003). Permutation tests for classification: Towards statistical significance in image-based studies. Information Processing in Medical Imaging, 18, 330341. doi:10.1007/978-3-540-45087-0_28CrossRefGoogle ScholarPubMed
Goodman, R. (1997). The strengths and difficulties questionnaire: A research note. Journal of Child Psychology and Psychiatry, 38(5), 581586. doi:10.1111/j.1469-7610.1997.tb01545.xCrossRefGoogle ScholarPubMed
Goodman, R., Ford, T., Richards, H., Gatward, R., & Meltzer, H. (2000). The Development and Well-Being Assessment: Description and initial validation of an integrated assessment of child and adolescent psychopathology. Journal of Child Psychology and Psychiatry, 41(5), 645655.CrossRefGoogle ScholarPubMed
Guloksuz, S., Pries, L. K., Delespaul, P., Kenis, G., Luykx, J. J., Lin, B. D., … van Os, J. (2019). Examining the independent and joint effects of molecular genetic liability and environmental exposures in schizophrenia: Results from the EUGEI study. World Psychiatry, 18(2), 173182. doi:10.1002/wps.20629CrossRefGoogle ScholarPubMed
Haas, S. S., Antonucci, L. A., Wenzel, J., Ruef, A., Biagianti, B., Paolini, M., … Kambeitz-Ilankovic, L. (2021). A multivariate neuromonitoring approach to neuroplasticity-based computerized cognitive training in recent onset psychosis. Neuropsychopharmacology, 46(4), 828835. doi:10.1038/s41386-020-00877-4CrossRefGoogle ScholarPubMed
Hampson, S. E., & Goldberg, L. R. (2006). A first large cohort study of personality trait stability over the 40 years between elementary school and midlife. Journal of Personality and Social Psychology, 91(4), 763779. doi:10.1037/0022-3514.91.4.763CrossRefGoogle ScholarPubMed
Hayes, A. F. (2017). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York: Guilford publications.Google Scholar
Howes, O. D., & Murray, R. M. (2014). Schizophrenia: An integrated sociodevelopmental-cognitive model. Lancet (London, England), 383(9929), 16771687. doi:10.1016/S0140-6736(13)62036-XCrossRefGoogle ScholarPubMed
International, Schizophrenia Consortium, Purcell, S. M., Wray, N. R., Stone, J. L., Visscher, P. M., O'Donovan, M. C., … Sklar, P. (2009). Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature, 460(7256), 748752. doi:10.1038/nature08185Google Scholar
Koutsouleris, N., Dwyer, D. B., Degenhardt, F., Maj, C., Urquijo-Castro, M. F., Sanfelici, R., … the PRONIA Consortium. (2021). Multimodal machine learning workflows for prediction of psychosis in patients with clinical high-risk syndromes and recent-onset depression. JAMA Psychiatry, 78(2), 195209. doi:10.1001/jamapsychiatry.2020.3604CrossRefGoogle ScholarPubMed
Koutsouleris, N., Kahn, R. S., Chekroud, A. M., Leucht, S., Falkai, P., Wobrock, T., … Hasan, A. (2016). Multisite prediction of 4-week and 52-week treatment outcomes in patients with first-episode psychosis: A machine learning approach. The Lancet. Psychiatry, 3(10), 935946. doi:10.1016/S2215-0366(16)30171-7CrossRefGoogle ScholarPubMed
Larøi, F., DeFruyt, F., van Os, J., Aleman, A., & Van der Linden, M. (2005). Associations between hallucinations and personality structure in a non-clinical sample: Comparison between young and elderly samples. Personality and Individual Differences, 39(1), 189200.CrossRefGoogle Scholar
Lella, A., Antonucci, L.A., Pergola, G. (2023). The interpretation of discrepancies between peer victimization experiences reported by different informants in capturing victimization-related genetic liability. A commentary on Armitage et al. (2022). Journal of Child Psychology and Psychiatry Advances, 3(1), e12137.Google ScholarPubMed
Lichtenstein, P., Yip, B. H., Bjork, C., Pawitan, Y., Cannon, T. D., Sullivan, P. F., & Hultman, C. M. (2009). Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: A population-based study. Lancet (London, England), 373(9659), 234239. doi:10.1016/S0140-6736(09)60072-6CrossRefGoogle ScholarPubMed
Lo, M. T., Hinds, D. A., Tung, J. Y., Franz, C., Fan, C. C., Wang, Y., … Chen, C. H. (2017). Genome-wide analyses for personality traits identify six genomic loci and show correlations with psychiatric disorders. Nature Genetics, 49(1), 152156. doi:10.1038/ng.3736CrossRefGoogle ScholarPubMed
Loch, A. A., Wang, Y. P., Rossler, W., Tofoli, L. F., Silveira, C. M., & Andrade, L. H. (2011). The psychosis continuum in the general population: Findings from the Sao Paulo Epidemiologic Catchment Area Study. European Archives of Psychiatry and Clinical Neuroscience, 261(7), 519527. doi:10.1007/s00406-011-0204-8CrossRefGoogle ScholarPubMed
Ludwig, L., Werner, D., & Lincoln, T. M. (2019). The relevance of cognitive emotion regulation to psychotic symptoms—A systematic review and meta-analysis. Clinical Psychology Review, 72, 101746. doi:10.1016/j.cpr.2019.101746CrossRefGoogle ScholarPubMed
Mark, W., & Toulopoulou, T. (2016). Psychometric properties of “community assessment of psychic experiences”: Review and meta-analyses. Schizophrenia Bulletin, 42(1), 3444. doi:10.1093/schbul/sbv088Google ScholarPubMed
McGrath, J. J., Saha, S., Al-Hamzawi, A., Alonso, J., Bromet, E. J., Bruffaerts, R., … Kessler, R. C. (2015). Psychotic experiences in the general population: A Cross-National analysis based on 31261 respondents from 18 countries. JAMA Psychiatry, 72(7), 697705. doi:10.1001/jamapsychiatry.2015.0575CrossRefGoogle Scholar
McGrath, J. J., Saha, S., Al-Hamzawi, A. O., Alonso, J., Andrade, L., Borges, G., … Kessler, R. C. (2016). Age of onset and lifetime projected risk of psychotic experiences: Cross-national data from the world mental health survey. Schizophrenia Bulletin, 42(4), 933941. doi:10.1093/schbul/sbw011CrossRefGoogle ScholarPubMed
Millan, M. J., Andrieux, A., Bartzokis, G., Cadenhead, K., Dazzan, P., Fusar-Poli, P., … Weinberger, D. (2016). Altering the course of schizophrenia: Progress and perspectives. Nature Reviews. Drug Discovery, 15(7), 485515. doi:10.1038/nrd.2016.28CrossRefGoogle Scholar
Nitzburg, G.C., Gopin, C.B., Peters, B.D., Karlsgodt, K.H., Malhotra, A.K., & DeRosse, P. (2016). The relationship between temperament and character and psychotic-like experiences in healthy children and adolescents. European Psychiatry : the Journal of the Association of European Psychiatrists, 31, 6065.CrossRefGoogle ScholarPubMed
Nuevo, R., Van Os, J., Arango, C., Chatterji, S., & Ayuso-Mateos, J. L. (2013). Evidence for the early clinical relevance of hallucinatory-delusional states in the general population. Acta psychiatrica Scandinavica, 127(6), 482493. doi:10.1111/acps.12010CrossRefGoogle ScholarPubMed
Passiatore, R., Antonucci, L. A., DeRamus, T. P., Fazio, L., Stolfa, G., Sportelli, L., … Pergola, G. (2023). Changes in patterns of age-related network connectivity are associated with risk for schizophrenia. Proceedings of the National Academy of Sciences of the United States of America, 120(32), e2221533120. doi:10.1073/pnas.2221533120CrossRefGoogle ScholarPubMed
Pergola, G., Papalino, M., Gelao, B., Sportelli, L., Vollerbergh, W., Grattagliano, I., & Bertolino, A. (2019). Evocative gene–environment correlation between genetic risk for schizophrenia and bullying victimization. World Psychiatry, 18(3), 366367. doi:10.1002/wps.20685CrossRefGoogle ScholarPubMed
Pergola, G., Penzel, N., Sportelli, L., & Bertolino, A. (2023). Lessons learned from parsing genetic risk for schizophrenia into biological pathways. Biological Psychiatry, 94(2), 121130. doi:10.1016/j.biopsych.2022.10.009CrossRefGoogle ScholarPubMed
Plomin, R., DeFries, J. C., & Loehlin, J. C. (1977). Genotype–environment interaction and correlation in the analysis of human behavior. Psychological Bulletin, 84(2), 309322.CrossRefGoogle ScholarPubMed
Raad, B. d., & Perugini, M. (Eds.). (2002). Big five factor assessment: Introduction. In Big five assessment (pp. 118). Cambridge, MA: Hogrefe & Huber Publishers.Google Scholar
Riglin, L., Hammerton, G., Heron, J., Collishaw, S., Arseneault, L., Thapar, A. K., … Thapar, A. (2019). Developmental contributions of schizophrenia risk alleles and childhood peer victimization to early-onset mental health trajectories. Americal Journal of Psychiatry, 176(1), 3643. doi:10.1176/appi.ajp.2018.18010075CrossRefGoogle ScholarPubMed
Roberts, B. W., Caspi, A., & Moffitt, T. E. (2001). The kids are alright: Growth and stability in personality development from adolescence to adulthood. Journal of Personality and Social Psychology, 81(4), 670683.CrossRefGoogle ScholarPubMed
Ross, S. R., Lutz, C. J., & Bailley, S. E. (2002). Positive and negative symptoms of schizotypy and the Five-factor model: A domain and facet level analysis. Journal of Personality Assessment, 79(1), 5372. doi:10.1207/S15327752JPA7901_04CrossRefGoogle Scholar
Sanfelici, R., Dwyer, D. B., Antonucci, L. A., & Koutsouleris, N. (2020). Individualized diagnostic and prognostic models for patients with psychosis risk syndromes: A meta-analytic view on the state of the art. Biological Psychiatry, 88(4), 349360. doi:10.1016/j.biopsych.2020.02.009CrossRefGoogle ScholarPubMed
Schoeler, T., Choi, S. W., Dudbridge, F., Baldwin, J., Duncan, L., Cecil, C. M., … Pingault, J. B. (2019). Multi-Polygenic score approach to identifying individual vulnerabilities associated with the risk of exposure to bullying. JAMA Psychiatry, 76(7), 730738. doi:10.1001/jamapsychiatry.2019.0310CrossRefGoogle ScholarPubMed
Schumann, G., Loth, E., Banaschewski, T., Barbot, A., Barker, G., Buchel, C., … the IMAGEN consortium. (2010). The IMAGEN study: Reinforcement-related behaviour in normal brain function and psychopathology. Molecular Psychiatry, 15(12), 11281139. doi:10.1038/mp.2010.4CrossRefGoogle ScholarPubMed
Smeland, O. B., Wang, Y., Lo, M. T., Li, W., Frei, O., Witoelar, A., … Andreassen, O. A. (2017). Identification of genetic loci shared between schizophrenia and the Big Five personality traits. Scientific Reports, 7(1), 2222. doi:10.1038/s41598-017-02346-3CrossRefGoogle ScholarPubMed
Solmi, M., Radua, J., Olivola, M., Croce, E., Soardo, L., Salazar de Pablo, G., … Fusar-Poli, P. (2022). Age at onset of mental disorders worldwide: Large-scale meta-analysis of 192 epidemiological studies. Molecular Psychiatry, 27(1), 281295. doi:10.1038/s41380-021-01161-7CrossRefGoogle ScholarPubMed
Stefanis, N. C., Hanssen, M., Smirnis, N. K., Avramopoulos, D. A., Evdokimidis, I. K., Stefanis, C. N., … Van Os, J. (2002). Evidence that three dimensions of psychosis have a distribution in the general population. Psychological Medicine, 32(2), 347358. doi:10.1017/s0033291701005141CrossRefGoogle ScholarPubMed
Uher, R., & Zwicker, A. (2017). Etiology in psychiatry: Embracing the reality of poly-gene–environmental causation of mental illness. World Psychiatry, 16(2), 121129. doi:10.1002/wps.20436CrossRefGoogle ScholarPubMed
VanderWeele, T. J. (2013). Policy-relevant proportions for direct effects. Epidemiology (Cambridge, Mass.), 24(1), 175176. doi:10.1097/EDE.0b013e3182781410CrossRefGoogle ScholarPubMed
van Os, J., Kenis, G., & Rutten, B. P. (2010). The environment and schizophrenia. Nature, 468(7321), 203212. doi:10.1038/nature09563CrossRefGoogle ScholarPubMed
Varchmin, L., Montag, C., Treusch, Y., Kaminski, J., & Heinz, A. (2021). Traumatic events, social adversity and discrimination as risk factors for psychosis - an umbrella review. Frontiers in Psychiatry, 12, 665957. doi:10.3389/fpsyt.2021.665957CrossRefGoogle ScholarPubMed
Vines, L., Bridgwater, M., Bachman, P., Hayes, R., Catalano, S., & Jalbrzikowski, M. (2022). Elevated emotion reactivity and emotion regulation in individuals at clinical high risk for developing psychosis and those diagnosed with a psychotic disorder. Early Intervention in Psychiatry, 16(7), 724735. doi:10.1111/eip.13212CrossRefGoogle ScholarPubMed
Vukasović, T., & Bratko, D. (2015). Heritability of personality: A meta-analysis of behavior genetic studies. Psychological Bulletin, 141(4), 769785. doi:10.1037/bul0000017CrossRefGoogle ScholarPubMed
Wiltink, S., Nelson, B., Velthorst, E., Wigman, J. T. W., Lin, A., Baksheev, G., … Yung, A. R. (2015). The relationship between personality traits and psychotic like experiences in a large non-clinical adolescent sample. Personality and Individual Differences, 73, 9297.CrossRefGoogle Scholar
Woicik, P. A., Stewart, S. H., Pihl, R. O., & Conrod, P. J. (2009). The Substance Use Risk Profile Scale: A scale measuring traits linked to reinforcement-specific substance use profiles. Addictive Behaviors, 34(12), 10421055. doi:10.1016/j.addbeh.2009.07.001CrossRefGoogle Scholar
Woolway, G. E., Smart, S. E., Lynham, A. J., Lloyd, J. L., Owen, M. J., Jones, I. R., … Legge, S. E. (2022). Schizophrenia polygenic risk and experiences of childhood adversity: A systematic review and meta-analysis. Schizophrenia Bulletin, 48(5), 967980. doi:10.1093/schbul/sbac049CrossRefGoogle ScholarPubMed
Yung, A. R., Nelson, B., Baker, K., Buckby, J. A., Baksheev, G., & Cosgrave, E. M. (2009). Psychotic-like experiences in a community sample of adolescents: Implications for the continuum model of psychosis and prediction of schizophrenia. The Australian and New Zealand Journal of Psychiatry, 43(2), 118128. doi:10.1080/00048670802607188CrossRefGoogle Scholar
Zalesky, A., Pantelis, C., Cropley, V., Fornito, A., Cocchi, L., McAdams, H., … Gogtay, N. (2015). Delayed development of brain connectivity in adolescents with schizophrenia and their unaffected siblings. JAMA Psychiatry, 72(9), 900908. doi:10.1001/jamapsychiatry.2015.0226CrossRefGoogle ScholarPubMed
Zohar, A. H., Zwir, I., Wang, J., Cloninger, C. R., & Anokhin, A. P. (2019). The development of temperament and character during adolescence: The processes and phases of change. Development and Psychopathology, 31(2), 601617. doi:10.1017/S0954579418000159CrossRefGoogle ScholarPubMed
Zwicker, A., Denovan-Wright, E. M., & Uher, R. (2018). Gene-environment interplay in the etiology of psychosis. Psychological Medicine, 48(12), 19251936. doi:10.1017/S003329171700383XCrossRefGoogle ScholarPubMed
Figure 0

Figure 1. The outline of the study. (1) Individual trajectories of change during adolescence based on personality traits were estimated via latent growth curve models. The two derived trait-related coefficients of change per participant fed a machine learning algorithm as longitudinal predictors of psychosis proneness signs (PPS) in young adulthood. Longitudinal decision scores extracted from the generated models were used: (2) to predict clinical outcomes other than PPS in adulthood; (3) as a longitudinal interface between individual polygenic risk for schizophrenia and bullying victimization across the pathway toward final PPS.(Figure representing machine learning analyses adapted from Dwyer et al., 2018).PPS, Psychosis Proneness Signs.

Figure 1

Table 1. Demographic characteristics of: (A) IMAGEN Discovery Sample; (B) IMAGEN Validation Sample; (C) TRAILS Replication cohort

Figure 2

Table 2. Validated performance of the personality-based risk calculator predicting Higher- v. Lower Psychosis Proneness Signs (PPS) in both Discovery and Validation samples

Figure 3

Figure 2. Raw trajectories of change over time for NEO neuroticism scores (left panel) and NEO openness scores (right panel) for the first ten individuals from the upper and the lower limits of the Ensemble Probability Prediction (EPP) scores distribution: notably, as an EPP score from 0.95 above estimated a probability to be assigned to the Higher-Pyschosis Proneness Signs (PPS) class in the 95% of the generated models, we considered such individuals as highly prototypical for such severity class (red lines; the red line in bold depicts the group mean trajectory); as an EPP score from 0.5 below estimated a probability to be assigned to the Higher-PPS class in the 5% of the generated models, we considered such individuals as highly prototypical of the Lower-PPS severity class (green lines; the green line in bold depicts the group mean trajectory) BL, Baseline; FU1, Follow-Up 1; FU2, Follow-Up 2; NEO, NEO Five Factor Inventory.

Figure 4

Figure 3. Findings from the serial mediation models, investigating the role of personality-based machine learning predictions and the rank product of Bullying Victimization (BV) within the pathway between polygenic risk for schizophrenia and final Psychosis Proneness Signs. Figure 3A depicted the model generated on IMAGEN data, Fig. 3B and 3C depicted replication models generated on TRAILS data, respectively including children-reported and parents-reported BV information. Direct effects (standardized coefficients) are shown. Red arrows represent relationships returning significant direct effects. The grey arrows represent not significant direct effects. Indirect effects for each model are reported in online Supplementary Table 10 (IMAGEN model), 17, and 18 (TRAILS replication models).FU3, Follow-Up 3; PPS, Psychosis Proneness Signs; w3, wave 3. *marks p < 0.05; **marks p < 0.01; ***marks p < 0.001.

Supplementary material: File

Antonucci et al. supplementary material

Antonucci et al. supplementary material
Download Antonucci et al. supplementary material(File)
File 2.2 MB