Polygenic score prediction of psychopathology dimensions and diagnoses within and between families

Polygenic score prediction of psychopathology dimensions and diagnoses within and between families | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Polygenic score prediction of psychopathology dimensions and diagnoses within and between families Yujing Lin, Francesca Procopio, Engin Keser, Kaito Kawakami, Thalia Eley, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8232693/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract The population prediction of polygenic scores (PGS) was found to capture two distinct processes: within-family prediction (individual-specific genetic differences between family members, such as siblings) and between-family prediction (family-level genetic differences, including assortative mating and ancestry). We quantify between-family prediction as the extent to which population prediction exceeds within-family prediction. While between-family prediction was found to be substantial for cognitive traits, its magnitude for psychopathology remains underexplored. Using 3300 unrelated individuals and 1600 dizygotic twin pairs at age 26 in the UK-based Twins Early Development Study, we examined within-family and population-level PGS prediction for eight psychopathologies, assessed as dimensions and diagnoses. Despite limited statistical power, within-family prediction is broadly comparable to population prediction, accounting for 72.4% of population estimates for dimensions and 78.0% for diagnoses. Only anxiety dimensions showed a significant prediction difference, suggesting some between-family contributions. We conclude that, overall, population PGS for psychopathology primarily reflect within-family genetic effects. Biological sciences/Psychology Biological sciences/Genetics/Behavioural genetics Figures Figure 1 Figure 2 Figure 3 Introduction One promise of the genomic era of psychopathology was that it would enable individual risk prediction from genetic information 1 . Through collaborative efforts over the past two decades, we can now predict a significant portion of variance in psychiatric liability using our DNA 2 . The foundations of this progress come from genome-wide association study (GWAS), which associated thousands of common genetic variants, specifically single nucleotide polymorphisms (SNPs), one at a time with psychiatric disorders 3,4 . These studies reveal that genetic effects collectively (called SNP heritability ) account for a significant proportion of psychiatric liability—for instance, 8.4% of the variance in depression liability 3 . To translate these population-level discoveries into individual risk predictions, polygenic score (PGS) applies GWAS-estimated SNP effect sizes to individuals’ specific genetic variants to estimate genetic liability to psychopathology 5 . For example, the depression PGS can predict 5.8% of its liability variance, approaching SNP heritability while reflecting a more stringent out-of-sample prediction 3 . However, recent research has questioned the interpretation of this predictive power 6,7 . These challenges arise because population-based GWAS and PGS predictions include within-family as well as between-family genetic contributions. Within-family effects reflect the random segregation of alleles during meiosis, which leads to genetic differences among family members and, therefore, trait differences. They could be calculated for example using within-sibling pair differences in PGS to predict trait differences 8 . The within-sibship design effectively excludes shared family-level effects such as passive gene-environment correlation, assortative mating, and population stratification, including socioeconomic status (SES) and ethnicity that may operate through shared familial genetic or environmental pathways 9–12 . In contrast, between-family effects are the shared familial effects included in population-level GWAS and PGS predictions, reflecting broader familial advantages or disadvantages. Here, we operationally define between-family effects as the extent to which population genetic prediction exceeds within-family prediction, making the two types of effects conceptually mutually exclusive 13 . Between-family effects have consistently been reported for cognitive traits, including educational attainment, educational achievement, general cognitive ability (g), and noncognitive abilities 8,13–15 . On average, the within-family prediction is only half the population prediction, highlighting the substantial role of between-family genetic effects in predicting cognitive development. In contrast, between-family effects have so far not been identified for psychopathology (using a sibling comparison design), including attention-deficit hyperactivity disorder (ADHD), schizophrenia, externalizing, and aggression, as well as personality 8,16–18 . A practical challenge in identifying between-family contributions is the limited variance explained by PGS for psychopathology compared to educational and cognitive outcomes 19 , constraining the ability to effectively distinguish within-family from population predictions. More fundamentally, the weak between-family genetic contributions to psychopathology are to be expected because between-family factors are less salient for these phenotypes. For example, assortative mating is much less for psychopathology compared to education and cognition 11,20,21 . As psychopathology PGS move toward individual risk prediction 22–24 , clinicians and researchers may assume these scores reflect individual genetic risk in a narrow sense—the genetic differences that would distinguish one sibling from another (i.e., within-family prediction). While within-family prediction distinguishes family members, both within-family and between-family effects contribute to predicting individual differences in an unrelated population. That is, population PGS prediction also captures between-family effects, which operate through intertwined genetic and environmental pathways and reflect broader familial risk profiles that are already used in mental health care for individual risk prediction 25 . Regardless of the magnitude of between-family contributions in psychopathology, distinguishing the sources of PGS prediction is important for researchers to understand what the prediction captures and for clinicians to use PGS in risk stratification within and across families. Therefore, in the present study, we systematically compare within-family and population-level prediction using polygenic scores for eight psychiatric disorders: autism, bipolar, ADHD, eating disorders, anxiety, alcohol use, depression, and post-traumatic stress disorder (PTSD). These disorders were selected based on the availability of both dimensional and diagnostic measures, adequate sample sizes, and representation of diverse psychiatric phenotypes. We used up to 3313 unrelated individuals (~65% female) and up to 1616 full dizygotic (DZ) twin pairs (~63% female) from the UK Twins Early Development Study (TEDS) 26 . We analyzed all eight psychopathology outcomes both as continuous current or lifetime dimensions and dichotomous diagnostic history (either professional or derived from clinical criteria) at age 26, except for alcohol use, which was only assessed dimensionally. We hypothesized that, unlike educational or cognitive outcomes, within-family PGS prediction for psychopathology outcomes would be similar in magnitude to population predictions across both dimensions and diagnoses. Results We systematically compared population PGS prediction to within-family prediction for eight continuous psychopathology dimensions and seven diagnoses (pre-registered at www.osf.io/pg9w6). Population-level estimates were calculated by regressing participants’ psychopathology outcomes on their respective PGS using unrelated individuals. Within-family estimates were calculated by regressing differences in sibling psychopathology outcomes on differences in their PGS, effectively controlling for between-family variations 27 , using full DZ twin pairs. Between-family genetic effects were operationally defined as and calculated from the extent to which population effects exceed within-family effects. We used linear regressions for dimensions and logistic regressions for diagnoses. All analyses were adjusted for multiple testing using False Discovery Rate (FDR). Age, sex, genotyping chip, and the first ten genetic principal components were included as covariates. PGS were derived from the latest available GWAS summary statistics (Table 1). All psychopathology outcomes were assessed at age 26. Power analyses (Supplementary Note) indicated that for dimensional outcomes, both population-level (assuming N = 3000 unrelated individuals) and within-family analyses (1500 DZ pairs) had adequate power (>80%) to detect effects explaining ~1% variance. For diagnostic outcomes (assuming 5% prevalence), population-level analyses had >80% power to detect log-odds of 0.250. Within-family power was evaluated under the assumption of a roughly balanced proportion of concordant and discordant DZ pairs, which provided >80% power to detect log-odds of 0.150. We also examined power to detect differences between the two estimates: for dimensions, differences ≥0.088 correlation units were detectable with 80% power, whereas for diagnoses, power was more limited (67.2%) to detect log-odds differences>0.200. Table 1. Sample characteristics and study designs Traits N (full DZ twin pairs/unrelated individuals) % diagnosed 1 GWAS 2 Measures (Continuous Outcomes) Autism 1482/3055 1.3% Grove et al., 2019 Ritvo Autism and Asperger Diagnostic Scale (RAADS-6) Bipolar 789/1608 0.8% O’Connell et al., 2025 Mood Disorder Questionnaire (MDQ-6) ADHD 1483/3054 1.4% Demontis et al., 2023 Conners-11 Eating Disorder 449/926 3 10.7% Watson et al., 2019 4 UK Biobank Anxiety 1560/3204 26.3% Friligkou et al., 2024 Generalised Anxiety Disorder – Dimensional (GAD-D) Alcohol Use 1422/2920 5 0% Mallard et al., 2022 Alcohol Use Disorders Identification Test (AUDIT) Depression 1616/3313 36.3% Adams et al., 2025 Moods and Feelings Questionnaire (MFQ-13) PTSD 1503/3096 4.2% Nievergelt et al., 2024 Post-Traumatic Stress Disorder Checklist (PCL-6) Note . Descriptive statistics are available in Table S1. Dimensions reflect participants’ current or lifetime symptoms or traits of psychopathology. Diagnoses refer to whether participants have ever met clinical criteria. 1 All case prevalence rates reflect the TEDS sample. Diagnoses for depression and anxiety were derived from the Composite International Diagnostic Interview–Short Form (CIDI-SF) for Depression (CIDID) and Anxiety (CIDIA), respectively. The eating disorder diagnosis was adapted from the UK Biobank screener developed by the Psychiatric Genomics Consortium (PGC) eating disorder group, based on DSM-5 criteria. The remaining were based on any self-report professional diagnoses. 2 Out-of-sample discovery GWAS for anxiety, alcohol use, and PTSD were based on dimensions, while the remaining traits used case-control GWAS. 3 The relatively high case prevalence for eating disorders reflects both the inclusion of all subtypes and the smaller sample size for this phenotype. 4 The dimensional measure for eating behaviors was an additive score from the adapted UK Biobank screener. 5 Alcohol use was assessed only as a dimension. PGS Prediction of Dimensional Outcomes Across eight continuous psychopathology outcomes at age 26, population PGS predictions were modest (blue bars in Figure 1A). PTSD, depression, and anxiety showed the largest associations (β standardized = 0.159, 0.152, and 0.149, respectively), followed by alcohol use (0.131), eating behaviors (0.092), ADHD (0.067), and autism (0.039). Bipolar symptoms showed no significant associations. Standard errors for beta estimates, variance explained, and FDR-adjusted p-values are available in Table S2. Within-family PGS predictions were broadly comparable to population PGS predictions, accounting for 72.4% of the population predictions (orange bars in Figure 1A; Table S2). PTSD again showed the strongest and the only significant association within families (β standardized = 0.099). We computed ratios only for outcomes with significant population prediction to avoid instability when the denominator approaches zero (purple dots in Figure 2, Table S3). Only the anxiety dimension showed a significant difference (△β standardized = 0.117) between its population and within-family estimates, though with a wide 95% confidence interval of 0.050 to 0.185. This may suggest some contributions of familial influences to individual differences in anxiety dimensional outcomes. To confirm the robustness of these results, we also fit mixed-effects models in which each DZ twin pair’s mean PGS (capturing population variation) and the deviation of each twin from that mean (capturing within-family variation) were entered simultaneously as predictors 8 . Although not identical in specification, the pair-mean and pair-deviation effects conceptually and empirically parallel population and within-family effects, respectively. The mixed-effects model results (Figure 3A; Table S4) were consistent with our mean pair difference regression results, with within-family estimates accounting for a similar proportion (72.4%) of the population estimates. The anxiety dimension also showed the largest difference between the two estimates, although not statistically distinguishable from each other in the mixed-effects framework. PGS Prediction of Diagnostic Outcomes Similarly, we compare the population and within-family PGS predictions for seven psychopathology diagnostic outcomes, excluding alcohol use. At the population level, PTSD again demonstrated the strongest prediction with a log-odds of 0.548, followed by anxiety (0.300), depression (0.267), and eating disorders (0.202), as shown in Figure 1B (blue bars). Autism, ADHD, and bipolar disorder showed nonsignificant predictions. At the within-family level, the average log-odds was smaller at 0.192 (orange bars, Figure 1B). None of the estimates reached statistical significance. Based on our pair difference regression outputs, within-family PGS prediction accounted for half (49.9%) of the population prediction (Figure 2; yellow dots), which ought to be interpreted with caution due to power constraints. Population logistic regression relies on the number of cases to distinguish case–control differences, and within-family regression relies on comparing discordant DZ twin pairs. As shown in Table S1, both case prevalence and the number of discordant pairs were limited, restricting the prediction precision. The mixed-effects model faces the same challenge of low case prevalence when estimating population effects; however, it improves power for within-family effects by jointly modeling the population and within-family estimates rather than relying on discordant twins. For diagnostic outcomes with relatively higher prevalence, such as depression and anxiety, these estimates would be more reliable. Using this framework, within-family prediction accounted for 78.0% of the corresponding population estimates (yellow dots in Figure 3B), suggesting that population and within-family estimates remain broadly comparable for diagnostic outcomes. To compare dimensional and diagnostic outcomes, we note that their outputs are on two different scales (standardized beta vs. log-odds), whereas the within-family-to-population ratio provides a scale-free comparison. While we suggest interpreting the pair difference regression result for diagnostic outcomes with caution due to limited power, overall, within-family PGS prediction accounted for about three-quarters of the population prediction across both dimensional and diagnostic psychopathology outcomes. Sex differences We have additionally tested for any sex differences between males and females for population PGS prediction (Supplementary Figure S1), as well as among same-sex male DZ twins, same-sex female DZ twins, and opposite-sex DZ twins for within-family prediction (Supplementary Figure S2). No significant sex differences were detected. We further explored whether between-family effects differed by sex. Overall, we found no consistent evidence for sex-specific between-family effects across psychopathology outcomes, except for some isolated contrasts that showed nominal differences. For example, pair difference regression showed significant between-family effects when comparing female population estimates to opposite-sex within-family estimates for the anxiety dimension, but not for same-sex comparisons or among males (Table S5). The findings were not replicated in mixed-effects models. For eating disorder diagnoses, mixed-effects models identified significant between-family effects among females but not males, with females and males differing significantly at the within-family level (Table S6). This pattern was not observed in the pair difference regression model. No sex-specific between-family effects were detected for the rest of the psychopathology outcomes across both analytical methods. Discussion We systematically compared within-family and population polygenic score predictions for psychopathology, uniquely examining both dimensional traits and clinical diagnoses within the same cohort in early adulthood at age 26. We found that within-family PGS prediction accounted for approximately three-quarters of population prediction on average across dimensional and diagnostic outcomes. Within the bounds of our statistical power, dimensional anxiety was the only phenotype showing a significant difference between the two predictions, which we refer to as between-family effects and which we operationally define as the extent to which population prediction exceeds within-family prediction. Our findings are consistent with a growing literature on family-based PGS studies, which has reported that PGS prediction at the population level is similar to the within-family level for externalizing behaviors, aggression, schizophrenia, neuroticism, and ADHD as dimensional outcomes 8,17,18 . Our results extend the evidence to clinical diagnoses, providing a more complete picture of what PGS prediction captures across different types of psychopathology measures. However, our absence of evidence should not be taken as evidence of the absence of between-family effects. In our study, population PGS explained on average 1.3% of the variance in dimensions and 1.0% in diagnoses, with relatively strong prediction for depression, PTSD, and anxiety. Although GWAS discovery efforts have significantly improved prediction at the population level 2 , the statistical power to separate within-family prediction from these modest population predictions remains limited. Given this limited power, it is also important to note that empirical evidence for the sources contributing to between-family variations is generally small in psychopathology. One source is assortative mating, referring to non-random selection of partners who are genetically and phenotypically similar in traits 11 . Assortative mating can increase (additive) genetic variation between but not within families. Although its effects tend to be stronger for cognitive ability and educational outcomes than for psychopathology outcomes 11,20,21 , emerging evidence suggests moderate to strong assortative mating for certain disorders, such as substance use disorder, ADHD, and obsessive–compulsive disorder 33 . However, its contributions to between-family effects in PGS prediction remain unclear. Population stratification, another source of between-family effects, occurs when ancestral differences correlate with both genotype and phenotype. A genetic variant may appear associated with a disorder because it is more common in an ancestral group with a higher incidence of that disorder. This pattern produces differences between families but is shared within families among siblings. Population stratification has been found for cognitive, educational, and anthropometric outcomes but under-investigated for psychopathology 34 . A third source is passive genotype-environment correlation. It occurs when biological parents transmit both their genes and a correlated rearing environment to their children. Some evidence of passive genotype-environment correlation exists for psychopathology, although with small effect sizes 35–37 . In our results, only the anxiety dimension showed a detectable difference between population (β standardized = 0.149) and within-family (β standardized = 0.032) predictions in the pair difference regression. This result was not replicated in the mixed-effects model, suggesting that any between-family influences on anxiety are likely modest and should be interpreted with caution pending replication. Family-level influences on anxiety have been reported previously in classic twin studies. Classic twin studies use the ACE model to partition the sources of individual differences into three components: additive genetic effects (A), shared environmental effects (C)—environmental factors that make siblings similar above and beyond shared genetics—and unique environmental effects (E)—factors that make siblings different 38 . These components are inferred by comparing the similarity of MZ twins, who share all their genes, with that of DZ twins, who share on average half. Because twin studies estimate genetic influences through within-family comparisons, between-family effects are absorbed into the shared environmental component (C). The presence of between-family influences on anxiety dimensions aligns with documented C in twin studies 39–41 . Although evidence for C was primarily found during childhood and tends to decrease with age 38 , some evidence for persistence into adulthood remains 41 . Additionally, assortative mating for anxiety outcomes offers a possible mechanistic explanation for the observed between-family contributions 21,33 . Our study also has limitations. First, our psychopathology outcomes are derived from self-report data, which may be subject to informant effects. Future work linking to medical records will provide a valuable complementary perspective. Second, our analyses focused on a relatively homogenous white European-ancestry cohort, limiting generalizability to other populations. Third, our participants were assessed at age 26, representing the latest wave of data collection while restricting inferences to this developmental stage. Fourth, although we explored sex differences, we did not detect consistent evidence across outcome types or modeling approaches. Future research may provide new insights by including the sex chromosomes in PGS construction 42 , which we excluded due to the current methodological constraints of many PGS tools. Fifth, despite adequate power to detect ~1% variance explained at the population level, about half of the population effects fell below this threshold. Low case prevalence further introduced instability, especially to the diagnostic outcomes, producing anomalous results such as depression's within-family estimate accounting for half the population estimate despite non-significant differences, and some within-family estimates exceeding population estimates. We used the ratio of within-family to population estimates as a scale-free, model-free metric to compare results across dimensional and diagnostic outcomes and across modeling approaches. Despite these limitations, our findings indicate that PGS prediction for psychopathology is largely similar at the population and within-family levels, with within-family prediction accounting for about three-quarters of population prediction. Although between-family effects were small in our study, they should not be dismissed as bias within this conceptual framework. We caution against terminology such as ‘gap,’ ‘inflation,’ or ‘attenuation’ in describing the difference between population and within-family estimates, as these terms may imply that population prediction is less valid or meaningful. Likewise, referring to within-family effects as ‘direct’ and between-family effects as ‘indirect’ oversimplifies the intertwined genetic and environmental pathways that contribute to PGS prediction. Both types of effects are informative depending on the research or clinical purpose: within-family effects index genetic differences that distinguish siblings within families, whereas between-family effects index genetic differences structured across families. Together, they comprise the population predictive utility of PGS, which is useful to understanding genetic contributions to psychopathology, informing translational applications in mental health care, and enabling targeted early intervention. Conclusions In this study, we systematically compared within-family and population polygenic score prediction of psychopathology across dimensional and diagnostic outcomes in approximately 3300 unrelated individuals and 1600 dizygotic twin pairs in early adulthood. We found that within-family prediction accounted for roughly three-quarters of population prediction, indicating that individual differences in psychiatric liability are primarily driven by genetic variation that differentiates individuals within families. This contrasts with cognitive and educational outcomes, where between-family influences account for about half of the population predictive power. Although anxiety dimensions showed some evidence of between-family prediction, this result warrants caution pending replication. More broadly, our findings support a conceptualization in which within-family and between-family predictions both provide meaningful sources of prediction depending on the scientific or clinical aim and, together, they define the population utility of PGS for understanding and improving mental health. Methods Participants Participants were drawn from the Twins Early Development Study (TEDS), a longitudinal cohort of twins born in England and Wales between 1994 and 1996 26 . A total of 13,759 families have taken part in the study, with genotype data available for 10,346 individuals. Detailed information on data collection across study waves is provided in the TEDS online data dictionary (https://www.teds.ac.uk/datadictionary/home.htm). The TEDS sample is broadly representative of the UK population for this birth cohort. Ethical approval was granted by the King’s College London Research Ethics Committee (references: PNM/09/10–104 and HR/DP-20/2122060), and informed consent was obtained from participants at each wave of data collection. The present study focused on genotyped participants of European ancestry assessed at approximately age 26 (mean age = 26.416 years, standard deviation, or SD = 0.923). We used two partially overlapping subsamples: population estimates were based on up to 3334 unrelated individuals (one twin per family; ~65.2% female), and within-family estimates were based on up to 1631 dizygotic (DZ) twin pairs (~63.3% female). Phenotypic Measures All mental health conditions were assessed using self-report questionnaires (https://datadictionary.teds.ac.uk/pdfs/26yr/26yr_mhq_coding.pdf). Each condition was analyzed both as a dimensional trait and, where available, as a dichotomous clinical diagnosis. Autism was assessed using the RAADS-6, a 6-item abbreviated version of the Ritvo Autism and Asperger Diagnostic Scale, developed by TEDS researchers from the RAADS-14 43 . The scale includes three items on social communication challenges and three on behavioral challenges, capturing both current and past autism-related characteristics. Higher scores reflect more autistic symptoms. Bipolar disorder was measured using the 6-item Mood Disorder Questionnaire (MDQ), a screening tool for bipolar spectrum disorders commonly used in outpatient settings 44 . Unlike other measures, the MDQ specifically asks whether participants have ever experienced certain symptoms, capturing lifetime experiences. ADHD symptoms were assessed using the Conners 3rd Edition Self-Report Scale 45 . To focus on adult ADHD traits, only the 11 inattention items were included. Eating disorders were measured using a screening tool developed by the Psychiatric Genomics Consortium Eating Disorders Working Group, identical to the measure used in the UK Biobank. Based on DSM-5 criteria, it screens for lifetime anorexia nervosa, bulimia nervosa, and binge eating disorder. A continuous score was derived by averaging available items for the participants who answered at least half of the items. Anxiety symptoms were assessed using the 10-item Generalized Anxiety Disorder-Dimensional (GAD-D) scale, capturing symptom severity over the past seven days 46 . Alcohol use was measured using the 7-item Alcohol Use Disorders Identification Test (AUDIT), which includes items on both lifetime and past-year alcohol consumption 47 . Higher scores indicate more severe patterns of use. Depression was measured using the 13-item Moods and Feelings Questionnaire (MFQ-13), assessing symptoms experienced over the past two weeks 48 . Post-traumatic stress disorder (PTSD) was assessed using the 6-item PTSD Checklist (PCL-6), a shortened version of the full scale that captures symptoms experienced in the past month 49 . Diagnostic history for autism, bipolar disorder, ADHD, and PTSD was collected via a self-report drop-down question asking: “Have you EVER been diagnosed with one or more of the following mental health problems or neurodevelopmental disorders by a professional, even if you don't have it currently?” Diagnostic status for depression and anxiety was derived from the Composite International Diagnostic Interview–Short Form (CIDI-SF), a validated structured interview administered as a questionnaire, using the DSM-5 criteria 50 . For depression, participants had to meet the following criteria: 1) report at least one of the two core symptoms (depressed mood or loss of interest), 2) report that depressive feelings lasted most or all of the day and occurred most or every day, 3) endorse at least five of eight key symptoms (including the core symptoms plus changes in weight/appetite, sleep, energy, concentration, feelings of worthlessness, or thoughts of death), and 4) report that symptoms interfered with everyday life. For anxiety, participants had to meet the following criteria: 1) report at least one of the two core symptoms of either a period of at least one month feeling worried, tense, or anxious most of the time, or a time when they worried much more than most people would in their situation, 2) report that worrying continued for 6 months or more and occurred most days, 3) report either worrying about more than one thing or having many worries on their mind, 4) report difficulty controlling the worry (difficult to stop worrying, often cannot put worries out of mind, or often cannot control worrying), 5) endorse at least three of six associated symptoms (restlessness or feeling on edge, being easily tired, difficulty concentrating, irritability, muscle tension, or sleep problems), and 6) report that symptoms interfered with everyday life. Eating disorder diagnoses were derived from questionnaire responses using the DSM-5 criteria. Due to low sample size, we analyzed three eating disorder diagnoses without distinguishing between subtypes in the main analyses: anorexia nervosa, bulimia nervosa, and binge-eating disorder. Eating disorder subtypes were determined first before we combined them for the present study. Specifically, for anorexia nervosa, participants had to meet three core criteria: 1) a lowest body mass index of 18.55 or below, 2) feeling fat or being afraid of gaining weight during periods of low weight, and 3) self-esteem dependent on body weight or shape. While anorexia nervosa can be further classified into restricting versus binge-eating/purging subtypes based on additional compensatory behaviors, these subtypes were not analyzed separately. For bulimia nervosa, participants had to meet the following criteria: 1) binge eating at least weekly for over 3 months, 2) feeling a loss of control during binge eating, 3) using compensatory behaviors (fasting, vomiting, laxatives/pills, or compulsive exercise) to offset overeating, 4) using these compensatory behaviors independently of binge eating or low weight periods, 5) self-esteem dependent on body weight or shape, and 6) binge eating occurring outside of low-weight episodes. For binge-eating disorder, participants had to meet the following criteria: 1) binge eating at least weekly for over 3 months, 2) feeling a loss of control during binge eating, 3) binge eating associated with at least three characteristics (eating rapidly, eating until uncomfortably full, eating large amounts when not hungry, eating alone due to embarrassment, or feeling disgusted/depressed/guilty afterward), 4) feeling distressed about binge eating, 5) not using compensatory behaviors (fasting, vomiting, compulsive exercise, or pills), and 6) binge eating occurring outside of low-weight episodes. Genotypic Measures Genotyping, Imputation, and Quality Control. DNA samples for TEDS participants were collected over five waves between 1998 and 2015 using cheek swabs and saliva kits. Genotyping was conducted in two phases using the Affymetrix Genome-Wide Human SNP Array 6.0 and, later, the Illumina HumanOmniExpressExome-8v1.2 arrays. Rigorous quality control procedures were applied to remove samples and SNPs with low call rates, inconsistencies in sex, ancestry outliers, and poor-quality markers. Imputation was performed using EAGLE2 and the Sanger Imputation Service with the Haplotype Reference Consortium (release 1.1) panel 51–53 . After harmonizing across platforms and excluding poorly imputed or discordant SNPs, the final dataset included 10,346 individuals and 7.36 million SNPs. Full details are provided in the TEDS data dictionary (https://datadictionary.teds.ac.uk/studies/dna.htm) and previous TEDS publications 8,13 Polygenic Score Calculation. Polygenic scores were constructed using LDpred2-auto, a Bayesian approach that adjusts GWAS summary statistics for linkage disequilibrium using as an external reference panel HapMap3+ 54 . This method estimates both SNP-based heritability and the proportion of SNPs with non-zero effects (i.e., polygenicity) from the summary statistics. We included all quality-controlled common SNPs shared between the TEDS sample and the reference panel (~1.1 million SNPs), ensuring broad genomic coverage and reliable imputation quality. Summary statistics used for score construction are listed in Table 1 3,4,28–32,55 . Statistical Analyses The study was pre-registered at www.osf.io/pg9w6. The current analyses focused on comparing within-family and population-level PGS prediction for psychopathology conditions, as outlined in the pre-registration. A previous study using a similar design has focused on cognitive traits 13 . Additionally, this study extends the preregistration by including comparisons between continuous trait scores and binary diagnostic outcomes. Specifically, we added logistic regression analysis to the originally planned linear regression. The additional analysis was necessary because psychopathology research tends to use a categorical diagnosis based on clinical thresholds 5 . To our knowledge, a combined approach of dimensions and diagnoses has not been applied to within-family PGS studies before. By overcoming methodological challenges to integrate both outcome types, we provide a more complete picture of how well polygenic scores predict psychopathology across the full range from dimensional symptoms to clinical diagnoses, and offer an example for future research seeking to examine genetic prediction for both dimensional and diagnostic measures. Phenotypic scores were derived differently for psychopathology dimensions and diagnoses (https://datadictionary.teds.ac.uk/studies/derived_variables/26yr_derived_variables.htm#zmhcidid). For dimensions, we computed the average score across all available items for each participant. Individuals were retained if they had completed at least half of the items on a given scale. For example, a participant needed to answer at least four out of seven items to be included in analyses. Diagnoses were coded directly from questionnaire responses. For most disorders, including ADHD, autism, bipolar disorder, and PTSD, participants indicated whether they had ever received a professional diagnosis by checking applicable boxes from a list of conditions. Case status was coded as positive if the condition was checked and negative if it was not checked. For depression, anxiety, and eating disorders, diagnoses were derived from questionnaire responses using the DSM-5 criteria, requiring participants to endorse specific combinations of core symptoms, duration requirements, and functional impairment (detailed criteria are provided in the Phenotypic Measures section above). Missing data were handled using a conservative probabilistic approach. Diagnoses were coded as positive only when participants met all required criteria or when they met most criteria and had minimal missing data on items with high endorsement rates among those meeting other criteria. For example, in depression and anxiety, a positive diagnosis was assigned if participants met the screening criteria and endorsed all but one of the remaining criteria, with the missing criterion being one that was frequently endorsed (70 to 90% endorsement rate) by others who screened positive. However, certain essential criteria—such as the 6-month duration requirement for anxiety—were never imputed, and missing data on these items resulted in a missing diagnosis. Negative diagnoses were assigned when participants either failed screening criteria or endorsed too few symptoms to meet diagnostic thresholds, even accounting for missing items. For eating disorders, missing data handling was more conservative, with minimal imputation allowed only for non-essential criteria with very high endorsement rates (>80 to 85%) when all other criteria were clearly met. In all cases, diagnoses remained coded as missing when there was insufficient information to confidently determine case status. Before carrying out the main analysis, we conducted a sensitivity analysis to examine whether covariates, including age, sex, zygosity, and birth order, were associated with our phenotypes. We used independent samples t-tests for continuous psychopathology conditions and chi-squared tests for dichotomous outcomes. If a covariate can significantly predict our psychopathology outcomes, it would be statistically controlled in the subsequent model as a covariate to avoid confounding. Significant age and sex effects were identified for some outcomes, whereas no significant effects were found for zygosity or birth order (Table S7). Therefore, age and sex were included as covariates in all models, along with genotyping platforms and the first ten genetic principal components, to control for potential age, sex, and batch effects and for population stratification. Additionally, all continuous phenotypes and polygenic scores are standardized (mean = 0, SD = 1) before being included in the model to ensure comparability among different measures. All analyses were adjusted for multiple testing using False Discovery Rate (FDR). Population-level genetic effects were estimated by regressing participants’ psychopathology outcomes on their respective PGS using unrelated individuals. Within-family genetic effects were estimated by regressing differences in sibling psychopathology outcomes on differences in their PGS, effectively controlling for between-family effects 27 , using full DZ twin pairs. Between-family genetic effects were operationally defined as the extent to which population effects exceed within-family effects. We used linear regressions for dimensions and logistic regressions for diagnoses. Age, sex, and the first ten genetic principal components were included as covariates. PGS were derived from the latest available GWAS summary statistics (Table 1). All psychopathology outcomes were assessed at age 26. Population-level estimates were calculated using one unrelated individual per twin pair (randomly selected). Coefficients were standardized from linear regressions of dimensions and from logistic analyses reported as log-odds for diagnoses. Within-family estimates were based on differences within DZ twin pairs. For dimensions, pair differences in both phenotype and PGS were calculated using the scaled data, with no additional scaling applied to preserve the original magnitude of the differences for comparison purposes. Re-scaling pair difference composites would distort the comparability with population estimates by inflating the variation of the within-family distribution. Since we used separate models to estimate population and within-family prediction, maintaining the same scale across both models was essential for valid comparisons. Additionally, we retained signed differences rather than absolute values to preserve directionality for regression analyses. Linear regression was also used to estimate within-family effects for dimensional traits. For diagnoses, pair differences are only meaningful in discordant twin pairs; therefore, within-family logistic regression models included only discordant DZ twins. Similar to the population-level analyses, coefficients were presented as standardized betas for dimensional outcomes and log-odds for diagnostic outcomes. To facilitate comparisons between linear and logistic regression estimates, we calculated variance explained for both the full model (including covariates) and a reduced model (with only the predictor). For linear models, variance explained (R²) follows the standard liability approach, whereas for logistic models, a pseudo-R² was used. We adopted McFadden’s R², which is based on maximum-likelihood estimation. While both outcome types yield an R² measure, caution is warranted when directly comparing R² from linear models with pseudo-R² from logistic models. Statistical significance for population and within-family estimates was determined directly from pair difference regression outputs. To estimate standard errors and construct 95% percentile confidence intervals, we used non-parametric bootstrapping with 10,000 iterations. This approach captures the empirical sampling distribution without distributional assumptions. Confidence intervals were constructed using the 2.5 th and 97.5 th percentiles of the bootstrap distribution, which may result in asymmetric intervals when the underlying sampling distribution is skewed. Moreover, for diagnoses, we used stratified bootstrapping to ensure the presence of cases in each subsample, addressing the issue of low case prevalence in our population sample. For between-family prediction, as it represents the difference between population and within-family effects, we used non-parametric bootstrapping (10,000 iterations) to generate 95% confidence intervals for the difference and tested whether these intervals excluded zero. We also computed within-family-to-population ratios from phenotypes with non-zero population prediction to avoid instability due to division by values close to zero. Confidence intervals for the ratios were also obtained from the bootstrapped results. These comparisons allowed us to evaluate both the magnitude and statistical significance of between-family effects. Finally, we conducted sex-stratified analyses to explore potential sex differences in prediction. Population-level models were run separately for males and females. Within-family models were stratified into same-sex DZ female pairs, same-sex DZ male pairs, and opposite-sex DZ pairs. All phenotypes and polygenic scores were also not re-scaled within each sex-stratified subsample for comparison purposes. Given the two subgroups for population estimates and three for within-family estimates, we performed a two-by-three comparison to evaluate differences in between-family contributions. Mixed-Effects Model. To complement our pair-differences regression analyses, we also conducted linear mixed-effects analyses for dimensional outcomes and generalized linear mixed-effects analyses for diagnostic outcomes to estimate within-family and population-level polygenic score effects simultaneously. This approach accounts for the nested structure of the data by incorporating random intercepts for family clustering, allowing use of the full sample of DZ twins. Within-family effects were modeled using each individual’s deviation from their twin pair mean PGS, while population effects were captured by the pair mean PGS. This decomposition isolates within-family variation from population variation and enables direct comparison among the two types of genetic effects. We continue to use our operational definition of between-family effects as the extent to which population estimates exceed within-family estimates, allowing for a comparison similar to our main regression approach. It is important, however, to clarify the interpretation of the population estimate derived from the pair mean PGS predictor. Conceptually, this estimate reflects population genetic effects by treating each twin pair as a single unit—leveraging information from both twins rather than from a single individual, as in analyses of unrelated participants. Statistically, population estimates obtained using pair mean PGS within a mixed-effects framework are typically larger, though not always significantly larger, than those from unrelated individuals. This occurs because averaging PGS across twins reduces random measurement error. Consequently, population-level estimates from mixed-effects models could be interpreted as approximate upper bounds of the true population effects. Additionally, age, sex, genotyping chip, and the first ten genetic principal components were included as covariates in all models. All continuous covariates (age and the first ten principal components), dimensional scores, and polygenic scores were standardized before calculating the pair-level composite. The within-family (pair PGS deviation from the mean) and population (pair mean PGS) effects did not undergo extra standardization. Similar to our pair difference regression approach, this is conducted to preserve the interpretability of the results. By standardizing the individual-level PGS first, the pair means and deviations are already based on standardized inputs, allowing them to be interpreted in terms of standard deviations. Not to standardize the pair composites again ensure these composites directly reflect the change in the outcome for a one standard deviation change in the original, underlying standardized PGS. Finally, the mixed-effects model and the pair difference regression estimate within-family and population effects on related but not directly comparable scales. Although both approaches use standardized PGS inputs, their coefficients differ in interpretation due to different ways of variance partitioning of the models, leading to different units of regression coefficients. Therefore, we compare results across models primarily using the within-family-to-population ratio, which provides a scale-free comparison indicating the contributions of within-family genetic effects in the population PGS prediction. Declarations Acknowledgement We gratefully acknowledge the ongoing contribution of the participants in the Twins Early Development Study (TEDS) and their families. TEDS is supported by the UK Medical Research Council (MR/V012878/1 and previously MR/M021475/1). For the purposes of open access, the author has applied a Creative Commons Attribution (CC BY) license to any Accepted Author Manuscript version arising from this submission. References Plomin, R. & Vassos, E. What clinicians should know about the contribution of modern behavioral genetics to psychiatric problems. Psychol. Med. 55 , e83 (2025). Abdellaoui, A., Yengo, L., Verweij, K. J. H. & Visscher, P. M. 15 years of GWAS discovery: Realizing the promise. Am. J. Hum. Genet. 110 , 179–194 (2023). Adams, M. J. et al. Trans-ancestry genome-wide study of depression identifies 697 associations implicating cell types and pharmacotherapies. Cell 0 , (2025). O’Connell, K. S. et al. Genomics yields biological and phenotypic insights into bipolar disorder. Nature 1–12 (2025) doi:10.1038/s41586-024-08468-9. Lewis, C. M. & Vassos, E. Polygenic Scores in Psychiatry: On the Road From Discovery to Implementation. Am. J. Psychiatry 179 , 800–806 (2022). Howe, L. J. et al. Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects. Nat. Genet. 54 , 581–592 (2022). Veller, C. & Coop, G. M. Interpreting population- and family-based genome-wide association studies in the presence of confounding. PLOS Biol. 22 , e3002511 (2024). Selzam, S. et al. Comparing within- and between-family polygenic score prediction. Am. J. Hum. Genet. 105 , 351–363 (2019). Abdellaoui, A., Dolan, C. V., Verweij, K. J. H. & Nivard, M. G. Gene–environment correlations across geographic regions affect genome-wide association studies. Nat. Genet. 54 , 1345–1354 (2022). Brumpton, B. et al. Avoiding dynastic, assortative mating, and population stratification biases in Mendelian randomization through within-family analyses. Nat. Commun. 11 , 3519 (2020). Horwitz, T. B., Balbona, J. V., Paulich, K. N. & Keller, M. C. Evidence of correlations between human partners based on systematic reviews and meta-analyses of 22 traits and UK Biobank analysis of 133 traits. Nat. Hum. Behav. 7 , 1568–1583 (2023). Zhou, Q. et al. Gene-environment correlation: the role of family environment in academic development. Mol. Psychiatry 1–10 (2024) doi:10.1038/s41380-024-02716-0. Lin, Y. et al. Polygenic score prediction within and between sibling pairs for intelligence, cognitive abilities, and educational traits from childhood to early adulthood. Intell. Cogn. Abil. https://icajournal.scholasticahq.com/article/140654-polygenic-score-prediction-within-and-between-sibling-pairs-for-intelligence-cognitive-abilities-and-educational-traits-from-childhood-to-early-adul (2025). Malanchini, M. et al. Genetic associations between non-cognitive skills and academic achievement over development. Nat. Hum. Behav. 1–13 (2024) doi:10.1038/s41562-024-01967-9. Okbay, A. et al. Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nat. Genet. 54 , 437–449 (2022). Schwaba, T., Nivard, M. G., Tucker-Drob, E. M. & Boomsma, D. I. Personality Genomics. Annu. Rev. Psychol. https://doi.org/10.1146/annurev-psych-013125-042402 (2025) doi:10.1146/annurev-psych-013125-042402. Tanksley, P. T. et al. Do polygenic indices capture “direct” effects on child externalizing behavior problems? Within-family analyses in two longitudinal birth cohorts. Clin. Psychol. Sci. 21677026241260260 (2024) doi:10.1177/21677026241260260. van der Laan, C. M. et al. Direct and Indirect Genetic Effects on Aggression. Biol. Psychiatry Glob. Open Sci. 3 , 958–968 (2023). Plomin, R. Celebrating a century of research in behavioral genetics. Behav. Genet. https://doi.org/10.1007/s10519-023-10132-3 (2023) doi:10.1007/s10519-023-10132-3. Torvik, F. A. et al. Modeling assortative mating and genetic similarities between partners, siblings, and in-laws. Nat. Commun. 13 , 1108 (2022). Torvik, F. A. et al. Non-random mating patterns within and across education and mental and somatic health. Nat. Commun. 15 , 10505 (2024). Agrawal, A. et al. The Psychiatric Genomics Consortium: discoveries and directions. Lancet Psychiatry 0 , (2025). Trinidad, S. B., Fullerton, S. M., Cohn, B., Crosslin, D. R. & Jarvik, G. P. Polygenic risk scores in the clinic: Health-system leaders and primary care providers weigh in. Genet. Med. 27 , 101575 (2025). Visscher, P. M., Yengo, L., Cox, N. J. & Wray, N. R. Discovery and implications of polygenicity of common diseases. Science 373 , 1468–1473 (2021). Pedersen, C. B. et al. Absolute and relative risks of mental disorders in families: a Danish register-based study. Lancet Psychiatry 12 , 590–599 (2025). Lockhart, C. et al. Twins Early Development Study (TEDS): A genetically sensitive investigation of mental health outcomes in the mid‐twenties. JCPP Adv. 3 , e12154 (2023). Lello, L., Raben, T. G. & Hsu, S. D. H. Sibling validation of polygenic risk scores and complex trait prediction. Sci. Rep. 10 , 13190 (2020). Grove, J. et al. Identification of common genetic risk variants for autism spectrum disorder. Nat. Genet. 51 , 431–444 (2019). Demontis, D. et al. Genome-wide analyses of ADHD identify 27 risk loci, refine the genetic architecture and implicate several cognitive domains. Nat. Genet. 55 , 198–208 (2023). Watson, H. J. et al. Genome-wide association study identifies eight risk loci and implicates metabo-psychiatric origins for anorexia nervosa. Nat. Genet. 51 , 1207–1214 (2019). Friligkou, E. et al. Gene discovery and biological insights into anxiety disorders from a large-scale multi-ancestry genome-wide association study. Nat. Genet. 56 , 2036–2045 (2024). Mallard, T. T. et al. Item-Level Genome-Wide Association Study of the Alcohol Use Disorders Identification Test in Three Population-Based Cohorts. Am. J. Psychiatry 179 , 58–70 (2022). Fan, C. C. et al. Spousal correlations for nine psychiatric disorders are consistent across cultures and persistent over generations. Nat. Hum. Behav. 1–9 (2025) doi:10.1038/s41562-025-02298-z. Haworth, S. et al. Apparent latent structure within the UK Biobank sample has implications for epidemiological analysis. Nat. Commun. 10 , 333 (2019). Jaffee, S. R. & Price, T. S. Gene–environment correlations: a review of the evidence and implications for prevention of mental illness. Mol. Psychiatry 12 , 432–442 (2007). Machlitt‐Northen, S. et al. Polygenic scores for schizophrenia and major depression are associated with psychosocial risk factors in children: evidence of gene–environment correlation. J. Child Psychol. Psychiatry 63 , 1140–1152 (2022). Rice, F., Lewis, G., Harold, G. T. & Thapar, A. Examining the role of passive gene–environment correlation in childhood depression using a novel genetically sensitive design. Dev. Psychopathol. 25 , 37–50 (2013). Plomin, R., DeFries, J. C., Knopik, V. S. & Neiderhiser, J. M. Top 10 replicated findings from behavioral genetics. Perspect. Psychol. Sci. 11 , 3–23 (2016). Boomsma, D. I., Van Beijsterveldt, C. E. M. & Hudziak, J. J. Genetic and environmental influences on Anxious/Depression during childhood: a study from the Netherlands Twin Register. Genes Brain Behav. 4 , 466–481 (2005). Ehringer, M. A., Rhee, S. H., Young, S., Corley, R. & Hewitt, J. K. Genetic and Environmental Contributions to Common Psychopathologies of Childhood and Adolescence: A Study of Twins and Their Siblings. J. Abnorm. Child Psychol. 34 , 1–17 (2006). Nes, R. B., Røysamb, E., Reichborn-Kjennerud, T., Harris, J. R. & Tambs, K. Symptoms of Anxiety and Depression in Young Adults: Genetic and Environmental Influences on Stability and Change. Twin Res. Hum. Genet. 10 , 450–461 (2007). Thomas, J. T. et al. Sex-stratified genome-wide association meta-analysis of major depressive disorder. Nat. Commun. 16 , 7960 (2025). Eriksson, J. M., Andersen, L. M. & Bejerot, S. RAADS-14 Screen: validity of a screening tool for autism spectrum disorder in an adult psychiatric population. Mol. Autism 4 , 49 (2013). Hirschfeld, R. M. A. et al. Development and Validation of a Screening Instrument for Bipolar Spectrum Disorder: The Mood Disorder Questionnaire. Am. J. Psychiatry 157 , 1873–1875 (2000). Conners, C. K. Conners 3rd Edition (Conners 3) Manual . (New York: MHS Assessments, 2008). Lebeau, R. T. et al. A dimensional approach to measuring anxiety for DSM‐5. Int. J. Methods Psychiatr. Res. 21 , 258–272 (2012). Reinert, D. F. & Allen, J. P. The Alcohol Use Disorders Identification Test: An Update of Research Findings. Alcohol. Clin. Exp. Res. 31 , 185–199 (2007). Angold, A., Costello, E. J., Messer, S. C. & Pickles, A. Development of a short questionnaire for use in epidemiological studies of depression in children and adolescents. Int. J. Methods Psychiatr. Res. 5 , 237–249 (1995). Weathers, F. W., Litz, B. T., Herman, D. S., Huska, J. A. & Keane, T. M. The PTSD Checklist (PCL): Reliability, Validity, and Diagnostic Utility . (Annual Convention of the International Society for Traumatic Stress Studies, San Antonio., 1993). Kessler, R. C., Andrews, G., Mroczek, D., Ustun, B. & Wittchen, H.-U. The World Health Organization Composite International Diagnostic Interview short-form (CIDI-SF). Int. J. Methods Psychiatr. Res. 7 , 171–185 (1998). Durbin, R. Efficient haplotype matching and storage using the positional Burrows–Wheeler transform (PBWT). Bioinformatics 30 , 1266–1272 (2014). Loh, P.-R. et al. Reference-based phasing using the Haplotype Reference Consortium panel. Nat. Genet. 48 , 1443–1448 (2016). McCarthy, S. A reference panel of 64,976 haplotypes for genotype imputation. Nat. Genet. 48 , 1279–1283 (2016). Privé, F., Arbel, J. & Vilhjálmsson, B. J. LDpred2: better, faster, stronger. Bioinformatics 36 , 5424–5431 (2021). Nievergelt, C. M. et al. Genome-wide association analyses identify 95 risk loci and provide insights into the neurobiology of post-traumatic stress disorder. Nat. Genet. 56 , 792–808 (2024). Additional Declarations There is NO Competing Interest. Supplementary Files WFBFPSSIsubmitted251125.docx Supplementary information: Polygenic score prediction of psychopathology dimensions and diagnoses within and between families Cite Share Download PDF Status: Under Review Version 1 posted Unknown event 14 Jan, 2026 Editorial decision: Reject before peer review 12 Dec, 2025 Editor assigned by journal 10 Dec, 2025 First submitted to journal 28 Nov, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8232693","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":552424201,"identity":"f086a182-03d5-4a49-8fcc-5aba66542be1","order_by":0,"name":"Yujing Lin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2klEQVRIiWNgGAWjYHCChANAQoaNgfkwA0MBWISZKC08bAxsyQwMBsRpAQMeIDImTovBjYSHhwt+MfDwSfd8NvhhwCDP38BjbEBAS8LhmX1Ah8mc3ZzYY8BgOOMAj3ECQS28PUAtErmbDwMdxrgB6MIDRGrJeQzSYk+cFp4fYC3MyUAtiSAteB0meeYB0JYGCaCWNGPDHgOJ5BmH2Yrxep/veE7yZ54/NnLyM5IfS/yosLHtb2/eLIFPi8IBngQGxja4GgnCESnfwH6AgeEPAVWjYBSMglEwsgEAb9pBhSg7XRgAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0007-1688-1948","institution":"King's College London","correspondingAuthor":true,"prefix":"","firstName":"Yujing","middleName":"","lastName":"Lin","suffix":""},{"id":552424202,"identity":"db0e91c7-a3dc-4bea-88f5-c1a7f48e69ab","order_by":1,"name":"Francesca Procopio","email":"","orcid":"","institution":"King's College London","correspondingAuthor":false,"prefix":"","firstName":"Francesca","middleName":"","lastName":"Procopio","suffix":""},{"id":552424203,"identity":"a3364871-210a-46a4-8fd7-7e26d2b26538","order_by":2,"name":"Engin Keser","email":"","orcid":"https://orcid.org/0009-0001-8100-688X","institution":"King's College London","correspondingAuthor":false,"prefix":"","firstName":"Engin","middleName":"","lastName":"Keser","suffix":""},{"id":552424204,"identity":"38e971ca-599f-489a-bf30-154be3212c89","order_by":3,"name":"Kaito Kawakami","email":"","orcid":"","institution":"King's College London","correspondingAuthor":false,"prefix":"","firstName":"Kaito","middleName":"","lastName":"Kawakami","suffix":""},{"id":552424205,"identity":"19713867-560d-4b85-aa80-ceceb944eaf4","order_by":4,"name":"Thalia Eley","email":"","orcid":"https://orcid.org/0000-0001-6458-0700","institution":"King's College London","correspondingAuthor":false,"prefix":"","firstName":"Thalia","middleName":"","lastName":"Eley","suffix":""},{"id":552424206,"identity":"0d60610d-17a1-42dd-ae24-7745c4cea36d","order_by":5,"name":"Kaili Rimfeld","email":"","orcid":"https://orcid.org/0000-0001-5139-065X","institution":"Royal Holloway, University of London","correspondingAuthor":false,"prefix":"","firstName":"Kaili","middleName":"","lastName":"Rimfeld","suffix":""},{"id":552424207,"identity":"095ac450-8216-4d29-94ad-bbb476462a9f","order_by":6,"name":"Margherita Malanchini","email":"","orcid":"https://orcid.org/0000-0002-7257-6119","institution":"Queen Mary University of London","correspondingAuthor":false,"prefix":"","firstName":"Margherita","middleName":"","lastName":"Malanchini","suffix":""},{"id":552424208,"identity":"0096f2fd-d9bc-4b23-a83c-06e6b0528ade","order_by":7,"name":"Robert Plomin","email":"","orcid":"https://orcid.org/0000-0002-0756-3629","institution":"Institute of Psychiatry, Psychology and Neuroscience, King's College London, London","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"","lastName":"Plomin","suffix":""}],"badges":[],"createdAt":"2025-11-28 18:25:44","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8232693/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8232693/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97505246,"identity":"7a211131-9c3c-415d-8fa8-ef4880629d3a","added_by":"auto","created_at":"2025-12-05 08:05:14","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":78719,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison of population (blue) and within-family (orange) PGS prediction for eight psychopathologies\u003c/strong\u003e. Estimates are presented as standardized beta coefficients for dimensions and as log-odds for diagnoses. Error bars represent 95% confidence intervals (CIs) calculated as the 2.5\u003csup\u003eth\u003c/sup\u003e and 97.5\u003csup\u003eth\u003c/sup\u003e percentiles of 10,000 bootstrap iterations; their potential asymmetry reflects the underlying distribution of the bootstrapped samples. The statistical significance for population and within-family estimates was assessed through formal tests, with FDR-adjusted p-values reported in Table S2, which also includes standardized betas, total model R², and incremental R² for the PGS predictor. Between-family effects are defined operationally as the extent to which the population estimate exceeds the within-family estimate. A significant between-family effect occurs when the population estimate is statistically greater than the within-family estimate, indicated by asterisks (**). Note that the asterisks denote significant between-family differences, not the significance of the individual population or within-family estimates themselves. Detailed results for between-family differences, including 95% CIs and FDR-adjusted p-values, are provided in Table S3.\u003c/p\u003e\n\u003cp\u003e** FDR-adjusted \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.01\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8232693/v1/09296198eaee4b9f0cdb49ec.png"},{"id":97505244,"identity":"aa731feb-84e4-46b6-89fb-4c6750323ef8","added_by":"auto","created_at":"2025-12-05 08:05:14","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":60849,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRatio of within-family to population estimates for psychopathology dimensions (purple) and diagnoses (yellow)\u003c/strong\u003e. The point estimates represent the within-family-to-population ratio for each psychopathology outcome. For psychopathology outcomes where the population-level estimate is not significant, the ratio is not calculated because the denominator is not distinguishable from zero. Ratios involving small or non-significant within-family estimates—though included in the analyses—should also be interpreted cautiously, as uncertainty in the numerator can inflate the variability of the ratio. The error bars are the 95% confidence intervals generated from 10,000 bootstrap iterations. A ratio below 1 indicates that the population-level estimate is larger than the within-family estimate. Full estimates are provided in Table S3.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8232693/v1/4e82e17c0fa8886bdfa2b018.png"},{"id":97505245,"identity":"70ab0e24-a059-485d-b608-0f1202dcda8b","added_by":"auto","created_at":"2025-12-05 08:05:14","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":68056,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMixed-effects model comparisons of population (cyan) and within-family (red) PGS predictions for eight psychopathologies\u003c/strong\u003e. Beta coefficients are shown for dimensions and log-odds for diagnoses. Error bars represent 95% confidence intervals. In mixed-effects model, population and within-family effects are estimated jointly using DZ twin pair’s mean PGS score (pair mean) and the deviation of each DZ twin’s score from that mean (pair deviation), respectively. The pair mean captures population variation by treating each family as an individual unit, while the pair deviation captures within-family variation. For visualization clarity, x-axes are clipped at ±2, affecting three psychopathology diagnoses (autism, bipolar disorder, ADHD). None of the estimates from these three diagnoses reached significance after multiple testing correction. No significant differences were detected between pair mean and pair deviation estimates for any measure. Complete estimates are provided in Table S4.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8232693/v1/44817e42b46ad858d81f2081.png"},{"id":98429026,"identity":"4c6c359b-9efe-4c3b-a455-0c95d7235d0d","added_by":"auto","created_at":"2025-12-17 16:42:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1024603,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8232693/v1/ed65cbcd-acdd-4f3e-b8db-28050484a34d.pdf"},{"id":97505247,"identity":"4e6ac5b5-8718-455d-aa22-68395983a2da","added_by":"auto","created_at":"2025-12-05 08:05:15","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":530072,"visible":true,"origin":"","legend":"Supplementary information: Polygenic score prediction of psychopathology dimensions and diagnoses within and between families","description":"","filename":"WFBFPSSIsubmitted251125.docx","url":"https://assets-eu.researchsquare.com/files/rs-8232693/v1/952c491597812644cbe07ed1.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Polygenic score prediction of psychopathology dimensions and diagnoses within and between families","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOne promise of the genomic era of psychopathology was that it would enable individual risk prediction from genetic information\u0026nbsp;\u003csup\u003e1\u003c/sup\u003e. Through collaborative efforts over the past two decades, we can now predict a significant portion of variance in psychiatric liability using our DNA\u0026nbsp;\u003csup\u003e2\u003c/sup\u003e. The foundations of this progress come from genome-wide association study (GWAS), which associated thousands of common genetic variants, specifically single nucleotide polymorphisms (SNPs), one at a time with psychiatric disorders\u0026nbsp;\u003csup\u003e3,4\u003c/sup\u003e. These studies reveal that genetic effects collectively (called \u003cem\u003eSNP heritability\u003c/em\u003e) account for a significant proportion of psychiatric liability—for instance, 8.4% of the variance in depression liability\u0026nbsp;\u003csup\u003e3\u003c/sup\u003e. To translate these population-level discoveries into individual risk predictions, polygenic score (PGS) applies GWAS-estimated SNP effect sizes to individuals’ specific genetic variants to estimate genetic liability to psychopathology\u0026nbsp;\u003csup\u003e5\u003c/sup\u003e. For example, the depression PGS can predict 5.8% of its liability variance, approaching SNP heritability while reflecting a more stringent out-of-sample prediction\u0026nbsp;\u003csup\u003e3\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHowever, recent research has questioned the interpretation of this predictive power\u0026nbsp;\u003csup\u003e6,7\u003c/sup\u003e. These challenges arise because population-based GWAS and PGS predictions include within-family as well as between-family genetic contributions. Within-family effects reflect the random segregation of alleles during meiosis, which leads to genetic differences among family members and, therefore, trait differences. They could be calculated for example using within-sibling pair differences in PGS to predict trait differences\u0026nbsp;\u003csup\u003e8\u003c/sup\u003e. The within-sibship design effectively excludes shared family-level effects such as passive gene-environment correlation, assortative mating, and population stratification, including socioeconomic status (SES) and ethnicity that may operate through shared familial genetic or environmental pathways\u0026nbsp;\u003csup\u003e9–12\u003c/sup\u003e. In contrast, between-family effects are the shared familial effects included in population-level GWAS and PGS predictions, reflecting broader familial advantages or disadvantages. Here, we operationally define between-family effects as the extent to which population genetic prediction exceeds within-family prediction, making the two types of effects conceptually mutually exclusive\u0026nbsp;\u003csup\u003e13\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBetween-family effects have consistently been reported for cognitive traits, including educational attainment, educational achievement, general cognitive ability (g), and noncognitive abilities\u0026nbsp;\u003csup\u003e8,13–15\u003c/sup\u003e. On average, the within-family prediction is only half the population prediction, highlighting the substantial role of between-family genetic effects in predicting cognitive development.\u003c/p\u003e\n\u003cp\u003eIn contrast, between-family effects have so far not been identified for psychopathology (using a sibling comparison design), including attention-deficit hyperactivity disorder (ADHD), schizophrenia, externalizing, and aggression, as well as personality\u0026nbsp;\u003csup\u003e8,16–18\u003c/sup\u003e. A practical challenge in identifying between-family contributions is the limited variance explained by PGS for psychopathology compared to educational and cognitive outcomes\u0026nbsp;\u003csup\u003e19\u003c/sup\u003e, constraining the ability to effectively distinguish within-family from population predictions. More fundamentally, the weak between-family genetic contributions to psychopathology are to be expected because between-family factors are less salient for these phenotypes. For example, assortative mating is much less for psychopathology compared to education and cognition\u0026nbsp;\u003csup\u003e11,20,21\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs psychopathology PGS move toward individual risk prediction\u0026nbsp;\u003csup\u003e22–24\u003c/sup\u003e, clinicians and researchers may assume these scores reflect individual genetic risk in a narrow sense—the genetic differences that would distinguish one sibling from another (i.e., within-family prediction). While within-family prediction distinguishes family members, both within-family and between-family effects contribute to predicting individual differences in an unrelated population. That is, population PGS prediction also captures between-family effects, which operate through intertwined genetic and environmental pathways and reflect broader familial risk profiles that are already used in mental health care for individual risk prediction\u0026nbsp;\u003csup\u003e25\u003c/sup\u003e. Regardless of the magnitude of between-family contributions in psychopathology, distinguishing the sources of PGS prediction is important for researchers to understand what the prediction captures and for clinicians to use PGS in risk stratification within and across families.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTherefore, in the present study, we systematically compare within-family and population-level prediction using polygenic scores for eight psychiatric disorders: autism, bipolar, ADHD, eating disorders, anxiety, alcohol use, depression, and post-traumatic stress disorder (PTSD). These disorders were selected based on the availability of both dimensional and diagnostic measures, adequate sample sizes, and representation of diverse psychiatric phenotypes. We used up to 3313 unrelated individuals (~65% female) and up to 1616 full dizygotic (DZ) twin pairs (~63% female) from the UK Twins Early Development Study (TEDS)\u0026nbsp;\u003csup\u003e26\u003c/sup\u003e. We analyzed all eight psychopathology outcomes both as continuous current or lifetime dimensions and dichotomous diagnostic history (either professional or derived from clinical criteria) at age 26, except for alcohol use, which was only assessed dimensionally. We hypothesized that, unlike educational or cognitive outcomes, within-family PGS prediction for psychopathology outcomes would be similar in magnitude to population predictions across both dimensions and diagnoses.\u0026nbsp;\u003c/p\u003e"},{"header":"Results ","content":"\u003cp\u003eWe systematically compared population PGS prediction to within-family prediction for eight continuous psychopathology dimensions and seven diagnoses (pre-registered at www.osf.io/pg9w6). Population-level estimates were calculated by regressing participants\u0026rsquo; psychopathology outcomes on their respective PGS using unrelated individuals. Within-family estimates were calculated by regressing differences in sibling psychopathology outcomes on differences in their PGS, effectively controlling for between-family variations \u003csup\u003e27\u003c/sup\u003e, using full DZ twin pairs. Between-family genetic effects were operationally defined as and calculated from the extent to which population effects exceed within-family effects. We used linear regressions for dimensions and logistic regressions for diagnoses. All analyses were adjusted for multiple testing using False Discovery Rate (FDR). Age, sex, genotyping chip, and the first ten genetic principal components were included as covariates. PGS were derived from the latest available GWAS summary statistics (Table 1). All psychopathology outcomes were assessed at age 26.\u003c/p\u003e\n\u003cp\u003ePower analyses (Supplementary Note) indicated that for dimensional outcomes, both population-level (assuming N = 3000 unrelated individuals) and within-family analyses (1500 DZ pairs) had adequate power (\u0026gt;80%) to detect effects explaining ~1% variance. For diagnostic outcomes (assuming 5% prevalence), population-level analyses had \u0026gt;80% power to detect log-odds of 0.250. Within-family power was evaluated under the assumption of a roughly balanced proportion of concordant and discordant DZ pairs, which provided \u0026gt;80% power to detect log-odds of 0.150. We also examined power to detect differences between the two estimates: for dimensions, differences \u0026ge;0.088 correlation units were detectable with 80% power, whereas for diagnoses, power was more limited (67.2%) to detect log-odds differences\u0026gt;0.200.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1. Sample characteristics and study designs\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTraits\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eN (full DZ twin pairs/unrelated individuals)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e% diagnosed\u0026nbsp;\u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGWAS\u0026nbsp;\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMeasures (Continuous Outcomes)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAutism\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1482/3055\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGrove et al., 2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRitvo Autism and Asperger Diagnostic Scale (RAADS-6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBipolar\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e789/1608\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eO\u0026rsquo;Connell et al., 2025\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMood Disorder Questionnaire (MDQ-6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eADHD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1483/3054\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eDemontis et al., 2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eConners-11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEating Disorder\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e449/926\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003csup\u003e3\u0026nbsp;\u003c/sup\u003e10.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eWatson et al., 2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003csup\u003e4\u0026nbsp;\u003c/sup\u003eUK Biobank\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAnxiety\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1560/3204\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e26.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eFriligkou et al., 2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGeneralised Anxiety Disorder \u0026ndash; Dimensional (GAD-D)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAlcohol Use\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1422/2920\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003csup\u003e5\u0026nbsp;\u003c/sup\u003e0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMallard et al., 2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAlcohol Use Disorders Identification Test (AUDIT)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eDepression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1616/3313\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e36.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAdams et al., 2025\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMoods and Feelings Questionnaire (MFQ-13)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePTSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1503/3096\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNievergelt et al., 2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePost-Traumatic Stress Disorder Checklist (PCL-6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eNote\u003c/em\u003e. \u0026nbsp;Descriptive statistics are available in Table S1. Dimensions reflect participants\u0026rsquo; current or lifetime symptoms or traits of psychopathology. Diagnoses refer to whether participants have ever met clinical criteria.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e1\u0026nbsp;\u003c/sup\u003eAll case prevalence rates reflect the TEDS sample.\u0026nbsp;Diagnoses for depression and anxiety were derived from the Composite International Diagnostic Interview\u0026ndash;Short Form (CIDI-SF) for Depression (CIDID) and Anxiety (CIDIA), respectively. The eating disorder diagnosis was adapted from the UK Biobank screener developed by the Psychiatric Genomics Consortium (PGC) eating disorder group, based on DSM-5 criteria. The remaining were based on any self-report professional diagnoses.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e2\u003c/sup\u003e Out-of-sample discovery GWAS for anxiety, alcohol use, and PTSD were based on dimensions, while the remaining traits used case-control GWAS.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e3\u003c/sup\u003e The relatively high case prevalence for eating disorders reflects both the inclusion of all subtypes and the smaller sample size for this phenotype.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e4\u003c/sup\u003e The dimensional measure for eating behaviors was an additive score from the adapted UK Biobank screener.\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e5\u0026nbsp;\u003c/sup\u003eAlcohol use was assessed only as a dimension.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePGS Prediction of Dimensional Outcomes\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAcross eight continuous psychopathology outcomes at age 26, population PGS predictions were modest (blue bars in Figure 1A). PTSD, depression, and anxiety showed the largest associations (\u0026beta;\u003csub\u003estandardized\u003c/sub\u003e = 0.159, 0.152, and 0.149, respectively), followed by alcohol use (0.131), eating behaviors (0.092), ADHD (0.067), and autism (0.039). Bipolar symptoms showed no significant associations. Standard errors for beta estimates, variance explained, and FDR-adjusted p-values are available in Table S2.\u003c/p\u003e\n\u003cp\u003eWithin-family PGS predictions were broadly comparable to population PGS predictions, accounting for 72.4% of the population predictions (orange bars in Figure 1A; Table S2). PTSD again showed the strongest and the only significant association within families (\u0026beta;\u003csub\u003estandardized\u003c/sub\u003e = 0.099). We computed ratios only for outcomes with significant population prediction to avoid instability when the denominator approaches zero (purple dots in Figure 2, Table S3). Only the anxiety dimension showed a significant difference (△\u0026beta;\u003csub\u003estandardized\u003c/sub\u003e = 0.117) between its population and within-family estimates, though with a wide 95% confidence interval of 0.050 to 0.185. This may suggest some contributions of familial influences to individual differences in anxiety dimensional outcomes.\u003c/p\u003e\n\u003cp\u003eTo confirm the robustness of these results, we also fit mixed-effects models in which each DZ twin pair\u0026rsquo;s mean PGS (capturing population variation) and the deviation of each twin from that mean (capturing within-family variation) were entered simultaneously as predictors \u003csup\u003e8\u003c/sup\u003e. Although not identical in specification, the pair-mean and pair-deviation effects conceptually and empirically parallel population and within-family effects, respectively.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe mixed-effects model results (Figure 3A; Table S4) were consistent with our mean pair difference regression results, with within-family estimates accounting for a similar proportion (72.4%) of the population estimates. The anxiety dimension also showed the largest difference between the two estimates, although not statistically distinguishable from each other in the mixed-effects framework.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePGS Prediction of Diagnostic Outcomes\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSimilarly, we compare the population and within-family PGS predictions for seven psychopathology diagnostic outcomes, excluding alcohol use. At the population level, PTSD again demonstrated the strongest prediction with a log-odds of 0.548, followed by anxiety (0.300), depression (0.267), and eating disorders (0.202), as shown in Figure 1B (blue bars). Autism, ADHD, and bipolar disorder showed nonsignificant predictions. At the within-family level, the average log-odds was smaller at 0.192 (orange bars, Figure 1B). None of the estimates reached statistical significance. Based on our pair difference regression outputs, within-family PGS prediction accounted for half (49.9%) of the population prediction (Figure 2; yellow dots), which ought to be interpreted with caution due to power constraints.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePopulation logistic regression relies on the number of cases to distinguish case\u0026ndash;control differences, and within-family regression relies on comparing discordant DZ twin pairs. As shown in Table S1, both case prevalence and the number of discordant pairs were limited, restricting the prediction precision. The mixed-effects model faces the same challenge of low case prevalence when estimating population effects; however, it improves power for within-family effects by jointly modeling the population and within-family estimates rather than relying on discordant twins. For diagnostic outcomes with relatively higher prevalence, such as depression and anxiety, these estimates would be more reliable. Using this framework, within-family prediction accounted for 78.0% of the corresponding population estimates (yellow dots in Figure 3B), suggesting that population and within-family estimates remain broadly comparable for diagnostic outcomes.\u003c/p\u003e\n\u003cp\u003eTo compare dimensional and diagnostic outcomes, we note that their outputs are on two different scales (standardized beta vs. log-odds), whereas the within-family-to-population ratio provides a scale-free comparison. While we suggest interpreting the pair difference regression result for diagnostic outcomes with caution due to limited power, overall, within-family PGS prediction accounted for about three-quarters of the population prediction across both dimensional and diagnostic psychopathology outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSex differences\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe have additionally tested for any sex differences between males and females for population PGS prediction (Supplementary Figure S1), as well as among same-sex male DZ twins, same-sex female DZ twins, and opposite-sex DZ twins for within-family prediction (Supplementary Figure S2). No significant sex differences were detected.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe further explored whether between-family effects differed by sex. Overall, we found no consistent evidence for sex-specific between-family effects across psychopathology outcomes, except for some isolated contrasts that showed nominal differences. For example, pair difference regression showed significant between-family effects when comparing female population estimates to opposite-sex within-family estimates for the anxiety dimension, but not for same-sex comparisons or among males (Table S5). The findings were not replicated in mixed-effects models. For eating disorder diagnoses, mixed-effects models identified significant between-family effects among females but not males, with females and males differing significantly at the within-family level (Table S6). This pattern was not observed in the pair difference regression model. No sex-specific between-family effects were detected for the rest of the psychopathology outcomes across both analytical methods.\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe systematically compared within-family and population polygenic score predictions for psychopathology, uniquely examining both dimensional traits and clinical diagnoses within the same cohort in early adulthood at age 26. We found that within-family PGS prediction accounted for approximately three-quarters of population prediction on average across dimensional and diagnostic outcomes. Within the bounds of our statistical power, dimensional anxiety was the only phenotype showing a significant difference between the two predictions, which we refer to as between-family effects and which we operationally define as the extent to which population prediction exceeds within-family prediction.\u003c/p\u003e\n\u003cp\u003eOur findings are consistent with a growing literature on family-based PGS studies, which has reported that PGS prediction at the population level is similar to the within-family level for externalizing behaviors, aggression, schizophrenia, neuroticism, and ADHD as dimensional outcomes\u0026nbsp;\u003csup\u003e8,17,18\u003c/sup\u003e. Our results extend the evidence to clinical diagnoses, providing a more complete picture of what PGS prediction captures across different types of psychopathology measures.\u003c/p\u003e\n\u003cp\u003eHowever, our absence of evidence should not be taken as evidence of the absence of between-family effects. In our study, population PGS explained on average 1.3% of the variance in dimensions and 1.0% in diagnoses, with relatively strong prediction for depression, PTSD, and anxiety. Although GWAS discovery efforts have significantly improved prediction at the population level\u0026nbsp;\u003csup\u003e2\u003c/sup\u003e, the statistical power to separate within-family prediction from these modest population predictions remains limited.\u003c/p\u003e\n\u003cp\u003eGiven this limited power, it is also important to note that empirical evidence for the sources contributing to between-family variations is generally small in psychopathology. One source is assortative mating, referring to non-random selection of partners who are genetically and phenotypically similar in traits\u0026nbsp;\u003csup\u003e11\u003c/sup\u003e. Assortative mating can increase (additive) genetic variation between but not within families. Although its effects tend to be stronger for cognitive ability and educational outcomes than for psychopathology outcomes\u0026nbsp;\u003csup\u003e11,20,21\u003c/sup\u003e, emerging evidence suggests moderate to strong assortative mating for certain disorders, such as substance use disorder, ADHD, and obsessive–compulsive disorder\u0026nbsp;\u003csup\u003e33\u003c/sup\u003e. However, its contributions to between-family effects in PGS prediction remain unclear. Population stratification, another source of between-family effects, occurs when ancestral differences correlate with both genotype and phenotype. A genetic variant may appear associated with a disorder because it is more common in an ancestral group with a higher incidence of that disorder. This pattern produces differences between families but is shared within families among siblings. Population stratification has been found for cognitive, educational, and anthropometric outcomes but under-investigated for psychopathology\u0026nbsp;\u003csup\u003e34\u003c/sup\u003e. A third source is passive genotype-environment correlation. It occurs when biological parents transmit both their genes and a correlated rearing environment to their children. Some evidence of passive genotype-environment correlation exists for psychopathology, although with small effect sizes\u0026nbsp;\u003csup\u003e35–37\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn our results, only the anxiety dimension showed a detectable difference between population (β\u003csub\u003estandardized\u003c/sub\u003e = 0.149) and within-family (β\u003csub\u003estandardized\u003c/sub\u003e = 0.032) predictions in the pair difference regression. This result was not replicated in the mixed-effects model, suggesting that any between-family influences on anxiety are likely modest and should be interpreted with caution pending replication. Family-level influences on anxiety have been reported previously in classic twin studies. Classic twin studies use the ACE model to partition the sources of individual differences into three components: additive genetic effects (A), shared environmental effects (C)—environmental factors that make siblings similar above and beyond shared genetics—and unique environmental effects (E)—factors that make siblings different\u0026nbsp;\u003csup\u003e38\u003c/sup\u003e. These components are inferred by comparing the similarity of MZ twins, who share all their genes, with that of DZ twins, who share on average half. Because twin studies estimate genetic influences through within-family comparisons, between-family effects are absorbed into the shared environmental component (C). The presence of between-family influences on anxiety dimensions aligns with documented C in twin studies\u0026nbsp;\u003csup\u003e39–41\u003c/sup\u003e. Although evidence for C was primarily found during childhood and tends to decrease with age\u0026nbsp;\u003csup\u003e38\u003c/sup\u003e, some evidence for persistence into adulthood remains\u0026nbsp;\u003csup\u003e41\u003c/sup\u003e. Additionally, assortative mating for anxiety outcomes offers a possible mechanistic explanation for the observed between-family contributions\u0026nbsp;\u003csup\u003e21,33\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur study also has limitations. First, our psychopathology outcomes are derived from self-report data, which may be subject to informant effects. Future work linking to medical records will provide a valuable complementary perspective. Second, our analyses focused on a relatively homogenous white European-ancestry cohort, limiting generalizability to other populations. Third, our participants were assessed at age 26, representing the latest wave of data collection while restricting inferences to this developmental stage. Fourth, although we explored sex differences, we did not detect consistent evidence across outcome types or modeling approaches. Future research may provide new insights by including the sex chromosomes in PGS construction\u0026nbsp;\u003csup\u003e42\u003c/sup\u003e, which we excluded due to the current methodological constraints of many PGS tools. Fifth, despite adequate power to detect ~1% variance explained at the population level, about half of the population effects fell below this threshold. Low case prevalence further introduced instability, especially to the diagnostic outcomes, producing anomalous results such as depression's within-family estimate accounting for half the population estimate despite non-significant differences, and some within-family estimates exceeding population estimates. We used the ratio of within-family to population estimates as a scale-free, model-free metric to compare results across dimensional and diagnostic outcomes and across modeling approaches.\u003c/p\u003e\n\u003cp\u003eDespite these limitations, our findings indicate that PGS prediction for psychopathology is largely similar at the population and within-family levels, with within-family prediction accounting for about three-quarters of population prediction. Although between-family effects were small in our study, they should not be dismissed as bias within this conceptual framework. We caution against terminology such as ‘gap,’ ‘inflation,’ or ‘attenuation’ in describing the difference between population and within-family estimates, as these terms may imply that population prediction is less valid or meaningful. Likewise, referring to within-family effects as ‘direct’ and between-family effects as ‘indirect’ oversimplifies the intertwined genetic and environmental pathways that contribute to PGS prediction. Both types of effects are informative depending on the research or clinical purpose: within-family effects index genetic differences that distinguish siblings within families, whereas between-family effects index genetic differences structured across families. Together, they comprise the population predictive utility of PGS, which is useful to understanding genetic contributions to psychopathology, informing translational applications in mental health care, and enabling targeted early intervention.\u003c/p\u003e"},{"header":"Conclusions ","content":"\u003cp\u003eIn this study, we systematically compared within-family and population polygenic score prediction of psychopathology across dimensional and diagnostic outcomes in approximately 3300 unrelated individuals and 1600 dizygotic twin pairs in early adulthood. We found that within-family prediction accounted for roughly three-quarters of population prediction, indicating\u0026nbsp;that individual differences in psychiatric liability are primarily driven by genetic variation that differentiates individuals within families. This contrasts with cognitive and educational outcomes, where between-family influences account for about half of the population predictive power. Although anxiety dimensions showed some evidence of between-family prediction, this result warrants caution pending replication. More broadly, our findings support a conceptualization in which within-family and between-family predictions both provide meaningful sources of prediction depending on the scientific or clinical aim and, together, they define the population utility of PGS for understanding and improving mental health.\u003c/p\u003e"},{"header":"Methods ","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eParticipants \u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants were drawn from the Twins Early Development Study (TEDS), a longitudinal cohort of twins born in England and Wales between 1994 and 1996 \u003csup\u003e26\u003c/sup\u003e. A total of 13,759 families have taken part in the study, with genotype data available for 10,346 individuals. Detailed information on data collection across study waves is provided in the TEDS online data dictionary (https://www.teds.ac.uk/datadictionary/home.htm). The TEDS sample is broadly representative of the UK population for this birth cohort. Ethical approval was granted by the King’s College London Research Ethics Committee (references: PNM/09/10–104 and HR/DP-20/2122060), and informed consent was obtained from participants at each wave of data collection.\u003c/p\u003e\n\u003cp\u003eThe present study focused on genotyped participants of European ancestry assessed at approximately age 26 (mean age = 26.416 years, standard deviation, or SD = 0.923). We used two partially overlapping subsamples: population estimates were based on up to 3334 unrelated individuals (one twin per family; ~65.2% female), and within-family estimates were based on up to 1631 dizygotic (DZ) twin pairs (~63.3% female). \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePhenotypic Measures \u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll mental health conditions were assessed using self-report questionnaires (https://datadictionary.teds.ac.uk/pdfs/26yr/26yr_mhq_coding.pdf). Each condition was analyzed both as a dimensional trait and, where available, as a dichotomous clinical diagnosis.\u003c/p\u003e\n\u003cp\u003eAutism was assessed using the RAADS-6, a 6-item abbreviated version of the Ritvo Autism and Asperger Diagnostic Scale, developed by TEDS researchers from the RAADS-14 \u003csup\u003e43\u003c/sup\u003e. The scale includes three items on social communication challenges and three on behavioral challenges, capturing both current and past autism-related characteristics. Higher scores reflect more autistic symptoms.\u003c/p\u003e\n\u003cp\u003eBipolar disorder was measured using the 6-item Mood Disorder Questionnaire (MDQ), a screening tool for bipolar spectrum disorders commonly used in outpatient settings \u003csup\u003e44\u003c/sup\u003e. Unlike other measures, the MDQ specifically asks whether participants have ever experienced certain symptoms, capturing lifetime experiences.\u003c/p\u003e\n\u003cp\u003eADHD symptoms were assessed using the Conners 3rd Edition Self-Report Scale \u003csup\u003e45\u003c/sup\u003e. To focus on adult ADHD traits, only the 11 inattention items were included.\u003c/p\u003e\n\u003cp\u003eEating disorders were measured using a screening tool developed by the Psychiatric Genomics Consortium Eating Disorders Working Group, identical to the measure used in the UK Biobank. Based on DSM-5 criteria, it screens for lifetime anorexia nervosa, bulimia nervosa, and binge eating disorder. A continuous score was derived by averaging available items for the participants who answered at least half of the items. \u003c/p\u003e\n\u003cp\u003eAnxiety symptoms were assessed using the 10-item Generalized Anxiety Disorder-Dimensional (GAD-D) scale, capturing symptom severity over the past seven days \u003csup\u003e46\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eAlcohol use was measured using the 7-item Alcohol Use Disorders Identification Test (AUDIT), which includes items on both lifetime and past-year alcohol consumption \u003csup\u003e47\u003c/sup\u003e. Higher scores indicate more severe patterns of use.\u003c/p\u003e\n\u003cp\u003eDepression was measured using the 13-item Moods and Feelings Questionnaire (MFQ-13), assessing symptoms experienced over the past two weeks \u003csup\u003e48\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003ePost-traumatic stress disorder (PTSD) was assessed using the 6-item PTSD Checklist (PCL-6), a shortened version of the full scale that captures symptoms experienced in the past month \u003csup\u003e49\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eDiagnostic history for autism, bipolar disorder, ADHD, and PTSD was collected via a self-report drop-down question asking: “Have you EVER been diagnosed with one or more of the following mental health problems or neurodevelopmental disorders by a professional, even if you don't have it currently?”\u003c/p\u003e\n\u003cp\u003eDiagnostic status for depression and anxiety was derived from the Composite International Diagnostic Interview–Short Form (CIDI-SF), a validated structured interview administered as a questionnaire, using the DSM-5 criteria \u003csup\u003e50\u003c/sup\u003e. For depression, participants had to meet the following criteria: 1) report at least one of the two core symptoms (depressed mood or loss of interest), 2) report that depressive feelings lasted most or all of the day and occurred most or every day, 3) endorse at least five of eight key symptoms (including the core symptoms plus changes in weight/appetite, sleep, energy, concentration, feelings of worthlessness, or thoughts of death), and 4) report that symptoms interfered with everyday life. \u003c/p\u003e\n\u003cp\u003eFor anxiety, participants had to meet the following criteria: 1) report at least one of the two core symptoms of either a period of at least one month feeling worried, tense, or anxious most of the time, or a time when they worried much more than most people would in their situation, 2) report that worrying continued for 6 months or more and occurred most days, 3) report either worrying about more than one thing or having many worries on their mind, 4) report difficulty controlling the worry (difficult to stop worrying, often cannot put worries out of mind, or often cannot control worrying), 5) endorse at least three of six associated symptoms (restlessness or feeling on edge, being easily tired, difficulty concentrating, irritability, muscle tension, or sleep problems), and 6) report that symptoms interfered with everyday life. \u003c/p\u003e\n\u003cp\u003eEating disorder diagnoses were derived from questionnaire responses using the DSM-5 criteria. Due to low sample size, we analyzed three eating disorder diagnoses without distinguishing between subtypes in the main analyses: anorexia nervosa, bulimia nervosa, and binge-eating disorder. Eating disorder subtypes were determined first before we combined them for the present study. Specifically, for anorexia nervosa, participants had to meet three core criteria: 1) a lowest body mass index of 18.55 or below, 2) feeling fat or being afraid of gaining weight during periods of low weight, and 3) self-esteem dependent on body weight or shape. While anorexia nervosa can be further classified into restricting versus binge-eating/purging subtypes based on additional compensatory behaviors, these subtypes were not analyzed separately. For bulimia nervosa, participants had to meet the following criteria: 1) binge eating at least weekly for over 3 months, 2) feeling a loss of control during binge eating, 3) using compensatory behaviors (fasting, vomiting, laxatives/pills, or compulsive exercise) to offset overeating, 4) using these compensatory behaviors independently of binge eating or low weight periods, 5) self-esteem dependent on body weight or shape, and 6) binge eating occurring outside of low-weight episodes. For binge-eating disorder, participants had to meet the following criteria: 1) binge eating at least weekly for over 3 months, 2) feeling a loss of control during binge eating, 3) binge eating associated with at least three characteristics (eating rapidly, eating until uncomfortably full, eating large amounts when not hungry, eating alone due to embarrassment, or feeling disgusted/depressed/guilty afterward), 4) feeling distressed about binge eating, 5) not using compensatory behaviors (fasting, vomiting, compulsive exercise, or pills), and 6) binge eating occurring outside of low-weight episodes. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGenotypic Measures \u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGenotyping, Imputation, and Quality Control.\u003c/strong\u003e DNA samples for TEDS participants were collected over five waves between 1998 and 2015 using cheek swabs and saliva kits. Genotyping was conducted in two phases using the Affymetrix Genome-Wide Human SNP Array 6.0 and, later, the Illumina HumanOmniExpressExome-8v1.2 arrays. Rigorous quality control procedures were applied to remove samples and SNPs with low call rates, inconsistencies in sex, ancestry outliers, and poor-quality markers. Imputation was performed using EAGLE2 and the Sanger Imputation Service with the Haplotype Reference Consortium (release 1.1) panel \u003csup\u003e51–53\u003c/sup\u003e. After harmonizing across platforms and excluding poorly imputed or discordant SNPs, the final dataset included 10,346 individuals and 7.36 million SNPs. Full details are provided in the TEDS data dictionary (https://datadictionary.teds.ac.uk/studies/dna.htm) and previous TEDS publications \u003csup\u003e8,13\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePolygenic Score Calculation.\u003c/strong\u003e Polygenic scores were constructed using LDpred2-auto, a Bayesian approach that adjusts GWAS summary statistics for linkage disequilibrium using as an external reference panel HapMap3+ \u003csup\u003e54\u003c/sup\u003e. This method estimates both SNP-based heritability and the proportion of SNPs with non-zero effects (i.e., polygenicity) from the summary statistics. We included all quality-controlled common SNPs shared between the TEDS sample and the reference panel (~1.1 million SNPs), ensuring broad genomic coverage and reliable imputation quality. Summary statistics used for score construction are listed in Table 1 \u003csup\u003e3,4,28–32,55\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eStatistical Analyses \u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was pre-registered at www.osf.io/pg9w6. The current analyses focused on comparing within-family and population-level PGS prediction for psychopathology conditions, as outlined in the pre-registration. A previous study using a similar design has focused on cognitive traits \u003csup\u003e13\u003c/sup\u003e. Additionally, this study extends the preregistration by including comparisons between continuous trait scores and binary diagnostic outcomes. Specifically, we added logistic regression analysis to the originally planned linear regression. The additional analysis was necessary because psychopathology research tends to use a categorical diagnosis based on clinical thresholds \u003csup\u003e5\u003c/sup\u003e. To our knowledge, a combined approach of dimensions and diagnoses has not been applied to within-family PGS studies before. By overcoming methodological challenges to integrate both outcome types, we provide a more complete picture of how well polygenic scores predict psychopathology across the full range from dimensional symptoms to clinical diagnoses, and offer an example for future research seeking to examine genetic prediction for both dimensional and diagnostic measures.\u003c/p\u003e\n\u003cp\u003ePhenotypic scores were derived differently for psychopathology dimensions and diagnoses (https://datadictionary.teds.ac.uk/studies/derived_variables/26yr_derived_variables.htm#zmhcidid). For dimensions, we computed the average score across all available items for each participant. Individuals were retained if they had completed at least half of the items on a given scale. For example, a participant needed to answer at least four out of seven items to be included in analyses. \u003c/p\u003e\n\u003cp\u003eDiagnoses were coded directly from questionnaire responses. For most disorders, including ADHD, autism, bipolar disorder, and PTSD, participants indicated whether they had ever received a professional diagnosis by checking applicable boxes from a list of conditions. Case status was coded as positive if the condition was checked and negative if it was not checked. For depression, anxiety, and eating disorders, diagnoses were derived from questionnaire responses using the DSM-5 criteria, requiring participants to endorse specific combinations of core symptoms, duration requirements, and functional impairment (detailed criteria are provided in the Phenotypic Measures section above). Missing data were handled using a conservative probabilistic approach. Diagnoses were coded as positive only when participants met all required criteria or when they met most criteria and had minimal missing data on items with high endorsement rates among those meeting other criteria. For example, in depression and anxiety, a positive diagnosis was assigned if participants met the screening criteria and endorsed all but one of the remaining criteria, with the missing criterion being one that was frequently endorsed (70 to 90% endorsement rate) by others who screened positive. However, certain essential criteria—such as the 6-month duration requirement for anxiety—were never imputed, and missing data on these items resulted in a missing diagnosis. Negative diagnoses were assigned when participants either failed screening criteria or endorsed too few symptoms to meet diagnostic thresholds, even accounting for missing items. For eating disorders, missing data handling was more conservative, with minimal imputation allowed only for non-essential criteria with very high endorsement rates (\u0026gt;80 to 85%) when all other criteria were clearly met. In all cases, diagnoses remained coded as missing when there was insufficient information to confidently determine case status.\u003c/p\u003e\n\u003cp\u003eBefore carrying out the main analysis, we conducted a sensitivity analysis to examine whether covariates, including age, sex, zygosity, and birth order, were associated with our phenotypes. We used independent samples t-tests for continuous psychopathology conditions and chi-squared tests for dichotomous outcomes. If a covariate can significantly predict our psychopathology outcomes, it would be statistically controlled in the subsequent model as a covariate to avoid confounding. Significant age and sex effects were identified for some outcomes, whereas no significant effects were found for zygosity or birth order (Table S7). Therefore, age and sex were included as covariates in all models, along with genotyping platforms and the first ten genetic principal components, to control for potential age, sex, and batch effects and for population stratification. Additionally, all continuous phenotypes and polygenic scores are standardized (mean = 0, SD = 1) before being included in the model to ensure comparability among different measures. All analyses were adjusted for multiple testing using False Discovery Rate (FDR).\u003c/p\u003e\n\u003cp\u003ePopulation-level genetic effects were estimated by regressing participants’ psychopathology outcomes on their respective PGS using unrelated individuals. Within-family genetic effects were estimated by regressing differences in sibling psychopathology outcomes on differences in their PGS, effectively controlling for between-family effects \u003csup\u003e27\u003c/sup\u003e, using full DZ twin pairs. Between-family genetic effects were operationally defined as the extent to which population effects exceed within-family effects. We used linear regressions for dimensions and logistic regressions for diagnoses. Age, sex, and the first ten genetic principal components were included as covariates. PGS were derived from the latest available GWAS summary statistics (Table 1). All psychopathology outcomes were assessed at age 26.\u003c/p\u003e\n\u003cp\u003ePopulation-level estimates were calculated using one unrelated individual per twin pair (randomly selected). Coefficients were standardized from linear regressions of dimensions and from logistic analyses reported as log-odds for diagnoses. Within-family estimates were based on differences within DZ twin pairs. \u003c/p\u003e\n\u003cp\u003eFor dimensions, pair differences in both phenotype and PGS were calculated using the scaled data, with no additional scaling applied to preserve the original magnitude of the differences for comparison purposes. Re-scaling pair difference composites would distort the comparability with population estimates by inflating the variation of the within-family distribution. Since we used separate models to estimate population and within-family prediction, maintaining the same scale across both models was essential for valid comparisons. Additionally, we retained signed differences rather than absolute values to preserve directionality for regression analyses. Linear regression was also used to estimate within-family effects for dimensional traits. For diagnoses, pair differences are only meaningful in discordant twin pairs; therefore, within-family logistic regression models included only discordant DZ twins. Similar to the population-level analyses, coefficients were presented as standardized betas for dimensional outcomes and log-odds for diagnostic outcomes. \u003c/p\u003e\n\u003cp\u003eTo facilitate comparisons between linear and logistic regression estimates, we calculated variance explained for both the full model (including covariates) and a reduced model (with only the predictor). For linear models, variance explained (R²) follows the standard liability approach, whereas for logistic models, a pseudo-R² was used. We adopted McFadden’s R², which is based on maximum-likelihood estimation. While both outcome types yield an R² measure, caution is warranted when directly comparing R² from linear models with pseudo-R² from logistic models.\u003c/p\u003e\n\u003cp\u003eStatistical significance for population and within-family estimates was determined directly from pair difference regression outputs. To estimate standard errors and construct 95% percentile confidence intervals, we used non-parametric bootstrapping with 10,000 iterations. This approach captures the empirical sampling distribution without distributional assumptions. Confidence intervals were constructed using the 2.5\u003csup\u003eth\u003c/sup\u003e and 97.5\u003csup\u003eth\u003c/sup\u003e percentiles of the bootstrap distribution, which may result in asymmetric intervals when the underlying sampling distribution is skewed. Moreover, for diagnoses, we used stratified bootstrapping to ensure the presence of cases in each subsample, addressing the issue of low case prevalence in our population sample.\u003c/p\u003e\n\u003cp\u003eFor between-family prediction, as it represents the difference between population and within-family effects, we used non-parametric bootstrapping (10,000 iterations) to generate 95% confidence intervals for the difference and tested whether these intervals excluded zero. We also computed within-family-to-population ratios from phenotypes with non-zero population prediction to avoid instability due to division by values close to zero. Confidence intervals for the ratios were also obtained from the bootstrapped results. These comparisons allowed us to evaluate both the magnitude and statistical significance of between-family effects. \u003c/p\u003e\n\u003cp\u003eFinally, we conducted sex-stratified analyses to explore potential sex differences in prediction. Population-level models were run separately for males and females. Within-family models were stratified into same-sex DZ female pairs, same-sex DZ male pairs, and opposite-sex DZ pairs. All phenotypes and polygenic scores were also not re-scaled within each sex-stratified subsample for comparison purposes. Given the two subgroups for population estimates and three for within-family estimates, we performed a two-by-three comparison to evaluate differences in between-family contributions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMixed-Effects Model.\u003c/strong\u003e To complement our pair-differences regression analyses, we also conducted linear mixed-effects analyses for dimensional outcomes and generalized linear mixed-effects analyses for diagnostic outcomes to estimate within-family and population-level polygenic score effects simultaneously. This approach accounts for the nested structure of the data by incorporating random intercepts for family clustering, allowing use of the full sample of DZ twins.\u003c/p\u003e\n\u003cp\u003eWithin-family effects were modeled using each individual’s deviation from their twin pair mean PGS, while population effects were captured by the pair mean PGS. This decomposition isolates within-family variation from population variation and enables direct comparison among the two types of genetic effects. We continue to use our operational definition of between-family effects as the extent to which population estimates exceed within-family estimates, allowing for a comparison similar to our main regression approach. It is important, however, to clarify the interpretation of the population estimate derived from the pair mean PGS predictor. Conceptually, this estimate reflects population genetic effects by treating each twin pair as a single unit—leveraging information from both twins rather than from a single individual, as in analyses of unrelated participants. Statistically, population estimates obtained using pair mean PGS within a mixed-effects framework are typically larger, though not always significantly larger, than those from unrelated individuals. This occurs because averaging PGS across twins reduces random measurement error. Consequently, population-level estimates from mixed-effects models could be interpreted as approximate upper bounds of the true population effects.\u003c/p\u003e\n\u003cp\u003eAdditionally, age, sex, genotyping chip, and the first ten genetic principal components were included as covariates in all models. All continuous covariates (age and the first ten principal components), dimensional scores, and polygenic scores were standardized before calculating the pair-level composite. The within-family (pair PGS deviation from the mean) and population (pair mean PGS) effects did not undergo extra standardization. Similar to our pair difference regression approach, this is conducted to preserve the interpretability of the results. By standardizing the individual-level PGS first, the pair means and deviations are already based on standardized inputs, allowing them to be interpreted in terms of standard deviations. Not to standardize the pair composites again ensure these composites directly reflect the change in the outcome for a one standard deviation change in the original, underlying standardized PGS. \u003c/p\u003e\n\u003cp\u003eFinally, the mixed-effects model and the pair difference regression estimate within-family and population effects on related but not directly comparable scales. Although both approaches use standardized PGS inputs, their coefficients differ in interpretation due to different ways of variance partitioning of the models, leading to different units of regression coefficients. Therefore, we compare results across models primarily using the within-family-to-population ratio, which provides a scale-free comparison indicating the contributions of within-family genetic effects in the population PGS prediction.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe gratefully acknowledge the ongoing contribution of the participants in the Twins Early Development Study (TEDS) and their families. TEDS is supported by the UK Medical Research Council (MR/V012878/1 and previously MR/M021475/1).\u003c/p\u003e\n\u003cp\u003eFor the purposes of open access, the author has applied a Creative Commons Attribution (CC BY) license to any Accepted Author Manuscript version arising from this submission.\u003c/p\u003e"},{"header":"References ","content":"\u003col\u003e\n \u003cli\u003ePlomin, R. \u0026amp; Vassos, E. What clinicians should know about the contribution of modern behavioral genetics to psychiatric problems. \u003cem\u003ePsychol. Med.\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, e83 (2025).\u003c/li\u003e\n \u003cli\u003eAbdellaoui, A., Yengo, L., Verweij, K. J. H. \u0026amp; Visscher, P. M. 15 years of GWAS discovery: Realizing the promise. \u003cem\u003eAm. J. Hum. Genet.\u003c/em\u003e \u003cstrong\u003e110\u003c/strong\u003e, 179\u0026ndash;194 (2023).\u003c/li\u003e\n \u003cli\u003eAdams, M. J. \u003cem\u003eet al.\u003c/em\u003e Trans-ancestry genome-wide study of depression identifies 697 associations implicating cell types and pharmacotherapies. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e0\u003c/strong\u003e, (2025).\u003c/li\u003e\n \u003cli\u003eO\u0026rsquo;Connell, K. S. \u003cem\u003eet al.\u003c/em\u003e Genomics yields biological and phenotypic insights into bipolar disorder. \u003cem\u003eNature\u003c/em\u003e 1\u0026ndash;12 (2025) doi:10.1038/s41586-024-08468-9.\u003c/li\u003e\n \u003cli\u003eLewis, C. M. \u0026amp; Vassos, E. Polygenic Scores in Psychiatry: On the Road From Discovery to Implementation. \u003cem\u003eAm. J. Psychiatry\u003c/em\u003e \u003cstrong\u003e179\u003c/strong\u003e, 800\u0026ndash;806 (2022).\u003c/li\u003e\n \u003cli\u003eHowe, L. J. \u003cem\u003eet al.\u003c/em\u003e Within-sibship genome-wide association analyses decrease bias in estimates of direct genetic effects. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e54\u003c/strong\u003e, 581\u0026ndash;592 (2022).\u003c/li\u003e\n \u003cli\u003eVeller, C. \u0026amp; Coop, G. M. Interpreting population- and family-based genome-wide association studies in the presence of confounding. \u003cem\u003ePLOS Biol.\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, e3002511 (2024).\u003c/li\u003e\n \u003cli\u003eSelzam, S. \u003cem\u003eet al.\u003c/em\u003e Comparing within- and between-family polygenic score prediction. \u003cem\u003eAm. J. Hum. Genet.\u003c/em\u003e \u003cstrong\u003e105\u003c/strong\u003e, 351\u0026ndash;363 (2019).\u003c/li\u003e\n \u003cli\u003eAbdellaoui, A., Dolan, C. V., Verweij, K. J. H. \u0026amp; Nivard, M. G. Gene\u0026ndash;environment correlations across geographic regions affect genome-wide association studies. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e54\u003c/strong\u003e, 1345\u0026ndash;1354 (2022).\u003c/li\u003e\n \u003cli\u003eBrumpton, B. \u003cem\u003eet al.\u003c/em\u003e Avoiding dynastic, assortative mating, and population stratification biases in Mendelian randomization through within-family analyses. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 3519 (2020).\u003c/li\u003e\n \u003cli\u003eHorwitz, T. B., Balbona, J. V., Paulich, K. N. \u0026amp; Keller, M. C. Evidence of correlations between human partners based on systematic reviews and meta-analyses of 22 traits and UK Biobank analysis of 133 traits. \u003cem\u003eNat. Hum. Behav.\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 1568\u0026ndash;1583 (2023).\u003c/li\u003e\n \u003cli\u003eZhou, Q. \u003cem\u003eet al.\u003c/em\u003e Gene-environment correlation: the role of family environment in academic development. \u003cem\u003eMol. Psychiatry\u003c/em\u003e 1\u0026ndash;10 (2024) doi:10.1038/s41380-024-02716-0.\u003c/li\u003e\n \u003cli\u003eLin, Y. \u003cem\u003eet al.\u003c/em\u003e Polygenic score prediction within and between sibling pairs for intelligence, cognitive abilities, and educational traits from childhood to early adulthood. \u003cem\u003eIntell. Cogn. Abil.\u003c/em\u003e https://icajournal.scholasticahq.com/article/140654-polygenic-score-prediction-within-and-between-sibling-pairs-for-intelligence-cognitive-abilities-and-educational-traits-from-childhood-to-early-adul (2025).\u003c/li\u003e\n \u003cli\u003eMalanchini, M. \u003cem\u003eet al.\u003c/em\u003e Genetic associations between non-cognitive skills and academic achievement over development. \u003cem\u003eNat. Hum. Behav.\u003c/em\u003e 1\u0026ndash;13 (2024) doi:10.1038/s41562-024-01967-9.\u003c/li\u003e\n \u003cli\u003eOkbay, A. \u003cem\u003eet al.\u003c/em\u003e Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e54\u003c/strong\u003e, 437\u0026ndash;449 (2022).\u003c/li\u003e\n \u003cli\u003eSchwaba, T., Nivard, M. G., Tucker-Drob, E. M. \u0026amp; Boomsma, D. I. Personality Genomics. \u003cem\u003eAnnu. Rev. Psychol.\u003c/em\u003e https://doi.org/10.1146/annurev-psych-013125-042402 (2025) doi:10.1146/annurev-psych-013125-042402.\u003c/li\u003e\n \u003cli\u003eTanksley, P. T. \u003cem\u003eet al.\u003c/em\u003e Do polygenic indices capture \u0026ldquo;direct\u0026rdquo; effects on child externalizing behavior problems? Within-family analyses in two longitudinal birth cohorts. \u003cem\u003eClin. Psychol. Sci.\u003c/em\u003e 21677026241260260 (2024) doi:10.1177/21677026241260260.\u003c/li\u003e\n \u003cli\u003evan der Laan, C. M. \u003cem\u003eet al.\u003c/em\u003e Direct and Indirect Genetic Effects on Aggression. \u003cem\u003eBiol. Psychiatry Glob. Open Sci.\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 958\u0026ndash;968 (2023).\u003c/li\u003e\n \u003cli\u003ePlomin, R. Celebrating a century of research in behavioral genetics. \u003cem\u003eBehav. Genet.\u003c/em\u003e https://doi.org/10.1007/s10519-023-10132-3 (2023) doi:10.1007/s10519-023-10132-3.\u003c/li\u003e\n \u003cli\u003eTorvik, F. A. \u003cem\u003eet al.\u003c/em\u003e Modeling assortative mating and genetic similarities between partners, siblings, and in-laws. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 1108 (2022).\u003c/li\u003e\n \u003cli\u003eTorvik, F. A. \u003cem\u003eet al.\u003c/em\u003e Non-random mating patterns within and across education and mental and somatic health. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 10505 (2024).\u003c/li\u003e\n \u003cli\u003eAgrawal, A. \u003cem\u003eet al.\u003c/em\u003e The Psychiatric Genomics Consortium: discoveries and directions. \u003cem\u003eLancet Psychiatry\u003c/em\u003e \u003cstrong\u003e0\u003c/strong\u003e, (2025).\u003c/li\u003e\n \u003cli\u003eTrinidad, S. B., Fullerton, S. M., Cohn, B., Crosslin, D. R. \u0026amp; Jarvik, G. P. Polygenic risk scores in the clinic: Health-system leaders and primary care providers weigh in. \u003cem\u003eGenet. Med.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 101575 (2025).\u003c/li\u003e\n \u003cli\u003eVisscher, P. M., Yengo, L., Cox, N. J. \u0026amp; Wray, N. R. Discovery and implications of polygenicity of common diseases. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e373\u003c/strong\u003e, 1468\u0026ndash;1473 (2021).\u003c/li\u003e\n \u003cli\u003ePedersen, C. B. \u003cem\u003eet al.\u003c/em\u003e Absolute and relative risks of mental disorders in families: a Danish register-based study. \u003cem\u003eLancet Psychiatry\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 590\u0026ndash;599 (2025).\u003c/li\u003e\n \u003cli\u003eLockhart, C. \u003cem\u003eet al.\u003c/em\u003e Twins Early Development Study (TEDS): A genetically sensitive investigation of mental health outcomes in the mid‐twenties. \u003cem\u003eJCPP Adv.\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, e12154 (2023).\u003c/li\u003e\n \u003cli\u003eLello, L., Raben, T. G. \u0026amp; Hsu, S. D. H. Sibling validation of polygenic risk scores and complex trait prediction. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 13190 (2020).\u003c/li\u003e\n \u003cli\u003eGrove, J. \u003cem\u003eet al.\u003c/em\u003e Identification of common genetic risk variants for autism spectrum disorder. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 431\u0026ndash;444 (2019).\u003c/li\u003e\n \u003cli\u003eDemontis, D. \u003cem\u003eet al.\u003c/em\u003e Genome-wide analyses of ADHD identify 27 risk loci, refine the genetic architecture and implicate several cognitive domains. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 198\u0026ndash;208 (2023).\u003c/li\u003e\n \u003cli\u003eWatson, H. J. \u003cem\u003eet al.\u003c/em\u003e Genome-wide association study identifies eight risk loci and implicates metabo-psychiatric origins for anorexia nervosa. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 1207\u0026ndash;1214 (2019).\u003c/li\u003e\n \u003cli\u003eFriligkou, E. \u003cem\u003eet al.\u003c/em\u003e Gene discovery and biological insights into anxiety disorders from a large-scale multi-ancestry genome-wide association study. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, 2036\u0026ndash;2045 (2024).\u003c/li\u003e\n \u003cli\u003eMallard, T. T. \u003cem\u003eet al.\u003c/em\u003e Item-Level Genome-Wide Association Study of the Alcohol Use Disorders Identification Test in Three Population-Based Cohorts. \u003cem\u003eAm. J. Psychiatry\u003c/em\u003e \u003cstrong\u003e179\u003c/strong\u003e, 58\u0026ndash;70 (2022).\u003c/li\u003e\n \u003cli\u003eFan, C. C. \u003cem\u003eet al.\u003c/em\u003e Spousal correlations for nine psychiatric disorders are consistent across cultures and persistent over generations. \u003cem\u003eNat. Hum. Behav.\u003c/em\u003e 1\u0026ndash;9 (2025) doi:10.1038/s41562-025-02298-z.\u003c/li\u003e\n \u003cli\u003eHaworth, S. \u003cem\u003eet al.\u003c/em\u003e Apparent latent structure within the UK Biobank sample has implications for epidemiological analysis. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 333 (2019).\u003c/li\u003e\n \u003cli\u003eJaffee, S. R. \u0026amp; Price, T. S. Gene\u0026ndash;environment correlations: a review of the evidence and implications for prevention of mental illness. \u003cem\u003eMol. Psychiatry\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 432\u0026ndash;442 (2007).\u003c/li\u003e\n \u003cli\u003eMachlitt‐Northen, S. \u003cem\u003eet al.\u003c/em\u003e Polygenic scores for schizophrenia and major depression are associated with psychosocial risk factors in children: evidence of gene\u0026ndash;environment correlation. \u003cem\u003eJ. Child Psychol. Psychiatry\u003c/em\u003e \u003cstrong\u003e63\u003c/strong\u003e, 1140\u0026ndash;1152 (2022).\u003c/li\u003e\n \u003cli\u003eRice, F., Lewis, G., Harold, G. T. \u0026amp; Thapar, A. Examining the role of passive gene\u0026ndash;environment correlation in childhood depression using a novel genetically sensitive design. \u003cem\u003eDev. Psychopathol.\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 37\u0026ndash;50 (2013).\u003c/li\u003e\n \u003cli\u003ePlomin, R., DeFries, J. C., Knopik, V. S. \u0026amp; Neiderhiser, J. M. Top 10 replicated findings from behavioral genetics. \u003cem\u003ePerspect. Psychol. Sci.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 3\u0026ndash;23 (2016).\u003c/li\u003e\n \u003cli\u003eBoomsma, D. I., Van Beijsterveldt, C. E. M. \u0026amp; Hudziak, J. J. Genetic and environmental influences on Anxious/Depression during childhood: a study from the Netherlands Twin Register. \u003cem\u003eGenes Brain Behav.\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 466\u0026ndash;481 (2005).\u003c/li\u003e\n \u003cli\u003eEhringer, M. A., Rhee, S. H., Young, S., Corley, R. \u0026amp; Hewitt, J. K. Genetic and Environmental Contributions to Common Psychopathologies of Childhood and Adolescence: A Study of Twins and Their Siblings. \u003cem\u003eJ. Abnorm. Child Psychol.\u003c/em\u003e \u003cstrong\u003e34\u003c/strong\u003e, 1\u0026ndash;17 (2006).\u003c/li\u003e\n \u003cli\u003eNes, R. B., R\u0026oslash;ysamb, E., Reichborn-Kjennerud, T., Harris, J. R. \u0026amp; Tambs, K. Symptoms of Anxiety and Depression in Young Adults: Genetic and Environmental Influences on Stability and Change. \u003cem\u003eTwin Res. Hum. Genet.\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 450\u0026ndash;461 (2007).\u003c/li\u003e\n \u003cli\u003eThomas, J. T. \u003cem\u003eet al.\u003c/em\u003e Sex-stratified genome-wide association meta-analysis of major depressive disorder. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 7960 (2025).\u003c/li\u003e\n \u003cli\u003eEriksson, J. M., Andersen, L. M. \u0026amp; Bejerot, S. RAADS-14 Screen: validity of a screening tool for autism spectrum disorder in an adult psychiatric population. \u003cem\u003eMol. Autism\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 49 (2013).\u003c/li\u003e\n \u003cli\u003eHirschfeld, R. M. A. \u003cem\u003eet al.\u003c/em\u003e Development and Validation of a Screening Instrument for Bipolar Spectrum Disorder: The Mood Disorder Questionnaire. \u003cem\u003eAm. J. Psychiatry\u003c/em\u003e \u003cstrong\u003e157\u003c/strong\u003e, 1873\u0026ndash;1875 (2000).\u003c/li\u003e\n \u003cli\u003eConners, C. K. \u003cem\u003eConners 3rd Edition (Conners 3) Manual\u003c/em\u003e. (New York: MHS Assessments, 2008).\u003c/li\u003e\n \u003cli\u003eLebeau, R. T. \u003cem\u003eet al.\u003c/em\u003e A dimensional approach to measuring anxiety for DSM‐5. \u003cem\u003eInt. J. Methods Psychiatr. Res.\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 258\u0026ndash;272 (2012).\u003c/li\u003e\n \u003cli\u003eReinert, D. F. \u0026amp; Allen, J. P. The Alcohol Use Disorders Identification Test: An Update of Research Findings. \u003cem\u003eAlcohol. Clin. Exp. Res.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 185\u0026ndash;199 (2007).\u003c/li\u003e\n \u003cli\u003eAngold, A., Costello, E. J., Messer, S. C. \u0026amp; Pickles, A. Development of a short questionnaire for use in epidemiological studies of depression in children and adolescents. \u003cem\u003eInt. J. Methods Psychiatr. Res.\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 237\u0026ndash;249 (1995).\u003c/li\u003e\n \u003cli\u003eWeathers, F. W., Litz, B. T., Herman, D. S., Huska, J. A. \u0026amp; Keane, T. M. \u003cem\u003eThe PTSD Checklist (PCL): Reliability, Validity, and Diagnostic Utility\u003c/em\u003e. (Annual Convention of the International Society for Traumatic Stress Studies, San Antonio., 1993).\u003c/li\u003e\n \u003cli\u003eKessler, R. C., Andrews, G., Mroczek, D., Ustun, B. \u0026amp; Wittchen, H.-U. The World Health Organization Composite International Diagnostic Interview short-form (CIDI-SF). \u003cem\u003eInt. J. Methods Psychiatr. Res.\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 171\u0026ndash;185 (1998).\u003c/li\u003e\n \u003cli\u003eDurbin, R. Efficient haplotype matching and storage using the positional Burrows\u0026ndash;Wheeler transform (PBWT). \u003cem\u003eBioinformatics\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, 1266\u0026ndash;1272 (2014).\u003c/li\u003e\n \u003cli\u003eLoh, P.-R. \u003cem\u003eet al.\u003c/em\u003e Reference-based phasing using the Haplotype Reference Consortium panel. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e48\u003c/strong\u003e, 1443\u0026ndash;1448 (2016).\u003c/li\u003e\n \u003cli\u003eMcCarthy, S. A reference panel of 64,976 haplotypes for genotype imputation. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e48\u003c/strong\u003e, 1279\u0026ndash;1283 (2016).\u003c/li\u003e\n \u003cli\u003ePriv\u0026eacute;, F., Arbel, J. \u0026amp; Vilhj\u0026aacute;lmsson, B. J. LDpred2: better, faster, stronger. \u003cem\u003eBioinformatics\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 5424\u0026ndash;5431 (2021).\u003c/li\u003e\n \u003cli\u003eNievergelt, C. M. \u003cem\u003eet al.\u003c/em\u003e Genome-wide association analyses identify 95 risk loci and provide insights into the neurobiology of post-traumatic stress disorder. \u003cem\u003eNat. Genet.\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, 792\u0026ndash;808 (2024).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"molecular-psychiatry","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"mp","sideBox":"Learn more about [Molecular Psychiatry](http://www.nature.com/mp/)","snPcode":"41380","submissionUrl":"https://mts-mp.nature.com/cgi-bin/main.plex","title":"Molecular Psychiatry","twitterHandle":"@molpsychiatry","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8232693/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8232693/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe population prediction of polygenic scores (PGS) was found to capture two distinct processes: within-family prediction (individual-specific genetic differences between family members, such as siblings) and between-family prediction (family-level genetic differences, including assortative mating and ancestry). We quantify between-family prediction as the extent to which population prediction exceeds within-family prediction. While between-family prediction was found to be substantial for cognitive traits, its magnitude for psychopathology remains underexplored.\u003c/p\u003e\n\u003cp\u003eUsing 3300 unrelated individuals and 1600 dizygotic twin pairs at age 26 in the UK-based Twins Early Development Study, we examined within-family and population-level PGS prediction for eight psychopathologies, assessed as dimensions and diagnoses.\u003c/p\u003e\n\u003cp\u003eDespite limited statistical power, within-family prediction is broadly comparable to population prediction, accounting for 72.4% of population estimates for dimensions and 78.0% for diagnoses. Only anxiety dimensions showed a significant prediction difference, suggesting some between-family contributions. We conclude that, overall, population PGS for psychopathology primarily reflect within-family genetic effects.\u003c/p\u003e","manuscriptTitle":"Polygenic score prediction of psychopathology dimensions and diagnoses within and between families","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-05 08:05:05","doi":"10.21203/rs.3.rs-8232693/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"transferred","content":"Molecular Psychiatry","date":"2026-01-14T15:55:27+00:00","index":"","fulltext":""},{"type":"decision","content":"Reject before peer review","date":"2025-12-12T10:48:52+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-10T23:55:30+00:00","index":"","fulltext":""},{"type":"submitted","content":"Nature Mental Health","date":"2025-11-28T18:13:09+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"molecular-psychiatry","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"mp","sideBox":"Learn more about [Molecular Psychiatry](http://www.nature.com/mp/)","snPcode":"41380","submissionUrl":"https://mts-mp.nature.com/cgi-bin/main.plex","title":"Molecular Psychiatry","twitterHandle":"@molpsychiatry","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"8cc30d14-eace-457c-806e-ba005638c71d","owner":[],"postedDate":"December 5th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":58791006,"name":"Biological sciences/Psychology"},{"id":58791007,"name":"Biological sciences/Genetics/Behavioural genetics"}],"tags":[],"updatedAt":"2026-03-21T21:15:16+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-05 08:05:05","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8232693","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8232693","identity":"rs-8232693","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}