Genome-wide Polygenic Scores for Multiple Psychiatric and Common Traits Identify Preadolescent Youth with Risk for Suicide

Suicide is a leading cause of death in youth worldwide, but identifying which youth are at high risk for suicide remains challenging. We constructed genome-wide polygenic scores (GPSs) from 24 psychiatric disorders and common traits from 8,212 US preadolescent children ages 9 to 10 and investigated their associations and predictive utility with suicidality (suicidal ideation and attempt). We identified three GPSs significantly associated with youth suicidality: ADHD (P=2.83x10 -4 ; odds ratio=1.12), general happiness with a belief that life is meaningful (P=1.30x10 -3 ; odds ratio=0.89) and autism spectrum disorder (ASD) (P=1.81x10 3 ; odds ratio=1.14). We also found a significant gene-by-environment interaction such that the GPS of ASD in the context of early life stress substantially increased suicidal ideation (P=1.39x10 -2 , odds ratio=1.11). Machine learning models showed, in predicting suicidal ideation, a receiver operators characteristics-area under the curve (ROC-AUC) of 0.72, and, in suicidal attempts, a ROC-AUC of 0.765. By providing the first quantitative account of the polygenic and environmental factors of suicidality in a large, representative population of preadolescent youth, this study shows the potential utility of the GPSs in investigating youth suicidality for early screening, intervention, and prevention. interaction with ELS. These results corroborate previous findings of preadolescent

Introduction suicidality (suicidal ideation and/or attempt) 25,26 potentially through the epigenetic 23 mechanisms 27,28 . Testing whether and to what extents ELS and genetic factors together act 24 synergistically on youth suicide will offer a much-needed insight into the biological pathway 25 of suicide and an actionable target for intervention. No literature has yet to report this. The 26 predictive utility of genetic factors had never been examined while adjusting for potential environmental confounders, such as early life stress or sociodemographic characteristics. 28 Determining these factors and their interaction may help elucidate early screening of risk for 29 suicide. We test the extent to which GPSs 22,23 for common traits and psychiatric disorders, 30 and their interaction with ELS is linked to the risk for suicide in young children. 32 33 We used data from the Adolescent Brain and Cognitive Development (ABCD) study, 34 a nationwide multisite prospective, longitudinal study. The enrolled sample includes 11,827 35 preadolescent children ages of 9 to 10 years old in the US across 21 sites recruited between 36 2015 and 2019. Among the samples, 1,656 case and 10,171 controls were found for suicidal 37 ideation, and 124 cases and 11,703 controls were found for suicide attempt (Figure 1). 38 Preadolescent children with suicidal ideation or attempts showed similar sociodemographic, 39 behavioral, clinical characteristics to those without suicidal ideation or attempts ( Table 1). 40 For GPS generation, we selected 24 psychiatric and common traits that are known to 41 be related to suicidality including personality, cognitive, psychological traits, and psychiatric 42 disorders that are known to be broadly related to suicidality: general happiness, 29,30 43 insomnia, 31 depression, 32 risk behaviors, 33 risk tolerance, educational attainment, 34,35 44 cognitive performance, 33,34 snoring, 36 worry, 37 IQ, 34 cannabis usage, 38,39 drink per week, 40 45 smoker, 41 Attention deficit hyperactivity disorder (ADHD), 42 Autism spectrum disorder 46 (ASD), 43 major depressive disorder (MDD), 44 schizophrenia, 45 bipolar disorder, 46 posttraumatic stress disorder (PTSD). 47 The GPS for each individual was computed as a sum of 48 their SNPs adjusting for first 10 principal components, with each SNP being weighted by the 49 effect from each GWAS summary statistic. 48 Since we wanted to fully incorporate the effects 50 of genome-wide SNPs, we used no thresholding of p-value significance and included all the 51 available SNPs with GWAS p-value < 1.

52
Out of the 24 GPSs, in suicidal ideation, a greater GPS of ADHD significantly 53 correlated with a greater likelihood (P=2.83x10 -4 , odds ratio (OR)=1.12; FDR significance). 54 Likewise, the ADHD GPS significantly correlated with active (P=9.75x10 -4 , OR=1.14) and 55 passive suicidal ideation (P=6.63x10 -4 , OR=1.13), respectively (Table 2, Figure 2). In active 56 suicidal ideation, a greater GPS of ASD correlated with a greater likelihood of suicidal 57 ideation (P=1.81x10 -3 , OR=1.14). Conversely, in passive suicidal ideation, a less GPS of OR=1.14) remained significant and even became stronger in terms of effect size compared 72 to the original analysis including every controls (Supplementary Table 3(a)). The sex-73 stratified analysis with healthy controls identified a significant association between active 74 suicidal ideation and the ADHD GPS (P=4.49x10 -4 , OR=1.31) in the female population 75 surviving the FDR significance (Supplementary Table 3(b)). 76 We tested whether the multiple GPSs contributed to the improved prediction of youth 77 suicidality using machine learning. The following input features were used: 24 GPSs, socio-78 demographic information (sex, age, marital status, parental education, study site),

95
We tested the utility of the GPS in identifying preadolescent youth at risk for suicide.

96
Using large-scale, representative samples of youth, we found novel associations between 97 youth suicidality and GPS for ADHD, ASD --positive associations --and general happiness -98 -a negative association. We next found significant gene-by-environment interactions 99 between the GPS of ASD and ELS, 25 a known risk factor for youth suicidality, together acting 100 synergistically on youth suicidality. In our data-driven predictive modeling, together with the 101 self-reported questionnaires for psychopathology (e.g., CBCL), intelligence, family 102 environment, and socio-demographic variables, inclusion of the multiple GPSs permitted to 103 classify preadolescent youth with suicidality with moderate accuracy. In sum, this study 104 sheds light on the genetic approach to youth suicidality for better understanding of the 105 etiologic pathway and for prevention and intervention.

106
Using the multi-GPS method, we discovered the association of youth suicidality with 107 the genetic profiles of psychiatric disorders, i.e., ADHD and ASD, particularly relevant to 108 childhood psychopathology, and of a common trait, i.e., general happiness. For ADHD and 109 ASD, a greater polygenic score correlated with a greater likelihood of suicidal ideation, 110 whereas for general happiness, a smaller polygenic score correlates with a greater likelihood 111 of passive suicide ideation. This is in line with the literature, suggesting that ADHD is 112 associated with suicidal attempts in youth. 49 Our GPS results may provide genomic evidence 113 for the link between ADHD and youth suicidality. The explained variance (McFadden's 114 pseudo-R 2 ) 50 of suicidal phenotypes by ADHD GPS was approximately 1.5% in children.

115
This estimation is higher than the results from past PRS studies of suicidality, which showed The GPS of ASD not only correlated with suicidal ideation, but also showed a 119 significant interaction with ELS. These results corroborate previous findings of preadolescent youth with ASD being at an increased risk for suicidality. 54,55 One study shows that youth 121 with ASD are 28 times more likely to endorse suicidal thoughts or behaviors than their 122 unaffected peers. 56 This strong link between risk of suicidality and elevated autistic traits 123 could be explained by specific behavioral attributes of both phenotypes, such as poor 124 socialization and problem-solving skill, or increased levels of anxiety. 55

175
Study design and participants 176 The study sample includes 11,827 preadolescent children ages of 9 to 10 years old in the were found for suicidal ideation, and 124 cases and 11,703 controls were found for suicide 180 attempt (Table 1). Ethical approval for the study was obtained from Seoul National

202
For GPS generation, we selected 24 psychiatric and common traits that are known to be 203 related to suicidality including personality, cognitive, psychological traits, and psychiatric 204 disorders that are known to be broadly related to suicidality: general happiness, 29,30 205 insomnia, 31 depression, 32 risk behaviors, 33 risk tolerance, educational attainment, 34,35 206 cognitive performance, 33,34 snoring, 36 worry, 37

Competing Interests Declaration
None of the authors have significant competing financial, professional, or personal interests that might have influenced the performance or presentation of the work described in the manuscript.

Contents:
Supplementary Table 1 Supplementary Table 2 Supplementary