Characterizing Developmental Trajectories and the Role of Neuropsychiatric Genetic Risk Variants in Early-Onset Depression

Key Points Question Do neuropsychiatric disorder genetic risk variants influence developmental trajectories of depression in youth? Findings In this population-based study including 7543 adolescents, distinct depression trajectory classes were identified. A later-adolescence–onset class (17.3% of the sample) showed a typical depression trajectory and was associated with major depressive disorder risk alleles, and an early-adolescence–onset class (9.0%) showed clinically significant symptoms at age 12 years and was associated with schizophrenia and attention-deficit hyperactivity disorder genetic risk, childhood attention-deficit hyperactivity disorder, and neurodevelopmental traits. Meaning Depression in youth is heterogeneous; findings are consistent with emerging evidence for a neurodevelopmental component to some cases of depression and that this component is more likely when onset is very early.

Psychosis-Like Symptom Interview 5,6 which assesses the presence of the psychotic experiences of hallucinations, delusions and thought interference. A total score was calculated (range 0, 12) including only those experiences rated as "definitely present" by trained interviewers 7 . Evidence to date suggests that psychotic experiences are weak predictors of psychotic disorder and are associated with most psychiatric disorders including depression 8,9 . Mothers and fathers reported on their own and their biological parents' current and past history of severe depression and schizophrenia by questionnaire at 12 weeks gestation. Familial loading for depression and schizophrenia was calculated as the number of family members with a history of depression or schizophrenia weighted by relatedness (first or second-degree relative) 10 . Maternal education (assessed during pregnancy) was used as an indicator of socioeconomic position. This was coded as achieved A levels (which is roughly equivalent to a US high school diploma) or a university degree (1) versus did not achieve this level (0).

Polygenic risk scores: additional details
Best guess genotype data underwent additional marker and individual quality control.
Individuals were excluded based on heterozygosity (greater than 4x standard deviation from sample mean), excessive relatedness (measured by mean kinship against all individuals in the analysis; excessive relatedness was defined as 3x increase standard deviation compared to the sample mean) and genotype missingness (>2%). (These analyses are described in the genotypeqc package at https:github.com/ricanney/stata). Markers were excluded if minor count less than 5, SNP missingness (>2%), Hardy-Weinberg equilibrium (p≤10 -10 ) and deviation from reference MAF (>10%). Ancestry informative principle components were generated from linkage independent ancestry informative markers using the bim2eigenvec package. Training GWAS for PRS were cleaned using the summaryqc package and processed using summaryqc2prePRS. All markers in training GWAS for PRS were mapped to hg19 and nomenclature standardised to the 1000 genomes reference panel based on chromosome location and IUPAC (International Union of Pure and Applied Chemistry e.g http://www.bioinformatics.org/sms2/iupac.html) genotype code.
Calculating bipolar disorder polygenic risk scores Scores were derived from bipolar disorder weights for 29,684 SNPs. Risk alleles were defined as those associated with case status in the most recent Psychiatric Genomics Consortium analysis of bipolar disorder 11 at a threshold of P < .05 as this threshold maximally capture phenotypic variance 11 . The genome-wide association study discovery sample size was: 20,352 cases and 31,358 controls. The PRS was standardized prior to analysis so odds ratios represent a one standard deviation change. Complete measures (6)

eAppendix 1. Determining the Optimal Number of Classes
In line with most recommendations we selected the optimal model on the basis of a number of factors: 1) The meaning, interpretability and face validity of classes based on existing knowledge and theory.
2) Fit indices: the SABIC and the VLMR-LRT have been recommended as the most robust fit indices 12 .
3) The size of the smallest class: a minimum of around 5% of the sample. In the models that allow for some missing observations, the SABIC also starts to level out at three classes. We judged the face validity of the three class model to be superior to that of the four class model which included similar classes to those in the three class model with the addition of a persistent moderately low class. The proportions in the smallest class was similar for the models including individuals with two or more and three or more measures of depressive symptoms (9%) but, as might be expected from selective attrition, was lower when restricting the sample to those with complete data. Entropy, a measure of classification uncertainty, was modest-good for the three class solution including those with 2+ and 3+ measures but increased for the full cases models suggesting this is driven by missing data uncertainty. Footnote to eTable 2b: PRS = polygenic risk score. All PRS are standardized. All correlations are significant at p<.01. Inverse probability weighting

History of Psychiatric Disorder
We used inverse probability weighting (IPW) to investigate the possibility of biased associations due to non-random missingness of genetic data. This approach has been recommended over alternative methods for dealing with missing data (such as multiple imputation) in situations where blocks of data are missing (as is often the case in ALSPAC where missingness of a variable is often due to non-participation in a clinic assessment visit). We also elected to use this approach because it was not valid to impute values for unobserved genetic data. IPW involves weighting complete cases by the inverse probability of their being a complete case and involves specifying a missingness model in order to account for any bias in patterns of association due to missing data 13