Can Psychiatric Genetics Advance Without Incorporating a Life Course Perspective?

Abstract

Psychiatric disorders unfold over the life course; however, genomic studies of these conditions overwhelmingly rely on phenotypes collected at a single time point, often in adulthood. Therefore, genome-wide association studies (GWASs) of psychiatric conditions may miss genetic variants with time-varying relevance to etiology, prevention, and treatment, such as those that influence trajectories of symptoms and behaviors, age at onset, course of treatment response, and the co-evolution of comorbidities. With recent advances in longitudinal biobanks and analytic tools, we posit that incorporating a life course perspective in psychiatric genetics will enable critically relevant insights into each of these areas of investigation. We propose that the current inconsistent portability of polygenic scores across age groups can be reconciled through the design of carefully considered longitudinal GWASs in age-diverse samples. Pioneering longitudinal GWASs in psychiatry have revealed novel genomic signals associated with time-dependent phenotypes that are distinct from those influencing lifetime diagnosis, suggesting that the study of longitudinal phenotypes will complement cross-sectional approaches and empower biological and therapeutic discoveries. Advances in post-GWAS functional annotation resources and analytic approaches now enable us to contextualize the genetic contributions to psychiatric disorders as dynamic age- and exposure-dependent processes. Although longitudinal GWASs pose unique challenges with regard to data availability, selection bias, and missing data, integrating temporality into psychiatric genetics at scale is now attainable and promises to reveal novel biology and therapeutic opportunities for psychiatric conditions.

Psychiatric disorders are among the most complex health conditions to diagnose and treat, in part because we lack clinically actionable biomarkers and individuals vary widely in age at onset, symptom trajectories across the lifespan, and treatment response (1,2). For decades, studies have attempted to understand how genetic variation contributes to the risk, development, and course of psychiatric disorders, with the ultimate goals of elucidating biological mechanisms, improving risk prediction, and informing prevention and intervention strategies across the life course. However, much of this work has relied on stable features of illness or assumed some correspondence between adult and youth presentations. Inevitably, this focus, necessitated by sample availability, may have obscured development-dependent genetic factors.

Timing is a critical, yet underutilized, axis in all areas of psychiatric genetics. Current genome-wide association studies (GWASs) of psychiatric conditions are largely based on static phenotypes, conducted by associating germline genetic variation with case/control status or quantitative phenotypes collected at a single point in time, often in middle-aged adults (Figure 1). Our reliance on fixed-in-time designs is understandable. GWASs require large sample sizes, often at the cost of phenotyping depth, breadth, and study duration. Adults are easier to recruit and consent and exhibit less frequent attrition than other age groups (3). Given that three-quarters of psychiatric conditions first manifest by early adulthood (4), GWASs targeting data collected after the period of greatest onset risk represent a more reliable return on investment. However, these GWASs may miss genetic factors that influence how symptoms evolve and conditions co-evolve over time, when conditions first onset, and how symptoms vary in response to treatments or developmentally delimited gene-environment interactions (Figure 2).

Figure 1.

Age-related heterogeneity across study samples used in psychiatric condition genome-wide association studies (GWASs) that reported age information. (A) Average ages reported for study samples used in the Nievergelt et al. (143) GWAS of posttraumatic stress disorder (PTSD). (B) Age ranges reported for study samples used in the Walters et al. (144) GWAS of alcohol dependence. When a maximum age was not reported (e.g., ages 18 years or older), it was set to 99 years. When available, the average participant age for each study sample is indicated by a circle, with the circle size representing study sample size. (C) Range and average age of onset for study samples was included in the Kalman et al. (62) GWAS of bipolar disorder age of onset. Circles indicate the average age of participants, and the black line indicates the reported average age of onset. These examples show opportunities to re-analyze existing data to gain new insights into variance around age at onset. See Table S1 for expansion of study abbreviations.

Figure 2.

Psychiatric conditions and their underlying symptoms are dynamic across the life course and between individuals. Example of symptoms of a psychiatric condition in 4 people across the life course. Individuals differ in their age at symptom emergence, response to nongenetic triggers (e.g., trauma), symptom trajectories over time, age at diagnosis, and response to treatment. Genome-wide association study (GWAS) case-control group assignment depends on when in life individuals are ascertained for GWAS relative to their particular symptom trajectories and may therefore miss genetic associations with these development-dependent phenotypes.

The genetic architecture of the same apparent phenotype may diverge over the life course (5,6), and twin studies have shown that the heritability of some (but not all) psychiatric symptoms or conditions increases throughout childhood and adolescence and decreases through adulthood (7–9). Changes in heritability may be partially attributable to temporal activation or inactivation of genetic factors over time, as has been reported for anxiety (10,11), depressive symptoms (10), and certain substance use phenotypes (12). There is also support for a suite of genetic influences that stably impact how related disorders coalesce across development (13). There is a push to expand the developmental perspective in psychiatric genetics (14–16), but prior constraints on data availability and computational methods have slowed progress. With recent increases in cross-sectional and longitudinal data available across age groups (Box 1) and advances in scalable longitudinal GWAS methods (see the Supplement), efforts to use temporally dynamic data to accelerate our understanding of psychiatric conditions are worth considering.

Box 1. The Growing Landscape of Longitudinal Biobank Data.

The past decade has seen an explosion of biobank data, and the longitudinal data in these resources are growing. Large-scale, population-based prospective studies such as the UK Biobank, All of Us Research Program, the ALSPAC, the ABCD study, and the MoBa have recruited participants at different life stages, and these participants are being followed prospectively as they age. In addition, administrative health records linked to biosamples, under the banners of PsycheMERGE, FinnGen, and iPSYCH, provide additional data sources to characterize the longitudinal view of psychiatric genetics. Summarizing all pertinent datasets is beyond the scope of this review. Instead, we highlight searchable catalogs of longitudinal datasets of relevance to psychiatric genetics.

Landscaping International Longitudinal Datasets is a Wellcome Trust–funded initiative that catalogs studies of participants who have been followed over time, which could be used to improve our understanding of depression, anxiety, and/or psychosis, including their trajectories and opportunities for early intervention. The catalog includes more than 3000 datasets spanning 146 different countries, 33% of which have more than 8000 participants at ascertainment, and 67 datasets have biological and genetic data (138).

Given the plethora of datasets, harmonization efforts are essential to improving comparability of measures across studies so that they can be combined in a larger analysis. Maelstrom Research (https://www.maelstrom-research.org/) (139,140) hosts a catalog of more than 441 individual studies organized by consortia and research networks, many of which focus on longitudinal data collection and have harmonization protocols in place. Study metadata, including availability of biosamples and psychiatric outcome data, are easily searchable on the platform. LongITools (https://longitools.org/) (141) is a European-based project that has harmonized 24 different longitudinal datasets on cardiometabolic conditions from across the life course. The project is part of the European Human Exposome Network and includes multiple datasets and analysis tools for longitudinal data (142).

ABCD, Adolescent Brain Cognitive Development; ALSPAC, Avon Longitudinal Study of Parents and Children; MoBa, Norwegian Mother, Father and Child Cohort Study.

In this review, we examine how longitudinal approaches can complement cross-sectional GWASs to advance our understanding of the genetic architecture of psychiatric disorders. First, we present polygenic analyses and show how they do (and do not) capture age-dependent condition liability and how they can be improved for age-diverse investigations. Second, we highlight longitudinal GWAS efforts that are in early stages. Third, we describe new resources and approaches to incorporate life course information into post-GWAS analyses through developmental annotation and time-varying exposures. Fourth, we discuss practical considerations for longitudinal analyses. Fifth, we map the path forward to more comprehensive integration of developmental variability into gene discovery. Our aim is to show that overcoming the challenges inherent in incorporating a life course perspective is feasible and will pay dividends by improving inferences about when and how genetic influences operate across development and disease course.

PREDICTIVE PERFORMANCE OF POLYGENIC SCORES ACROSS AGE GROUPS

A major advantage of GWASs is the aspirational application of polygenic scores (PGSs) to predict an individual’s susceptibility (compared with a reference population) to a psychiatric condition. PGSs alone are not sufficiently predictive for individual-level clinical decisions. Instead, they may support 1) risk stratification for public health and prevention efforts and 2) integration into prediction models that combine genetic, clinical, environmental, and developmental data for shared decision-making. For example, a model that included PGSs identified adolescents who may benefit most from alcohol prevention programs (17). However, the across-development portability of PGSs is predicated on the notion of stable genetic influence. Fortunately, many examples exist in which PGSs derived from GWASs of psychiatric disorder susceptibility conducted with middle-aged adults are predictive of the same condition, its symptoms, or behavioral correlates at earlier ages (18–27). This portability across age groups suggests that PGSs derived from cross-sectional GWASs in middle-aged adults primarily capture genetic effects on temporally stable manifestations of psychiatric conditions.

There is value in demonstrating how genetic risk for susceptibility to a psychiatric condition associates with the same condition and its symptoms across the life course, but such investigations may miss transient biological signals important to psychiatric conditions at certain developmental time points. For example, PGSs for schizophrenia derived from GWASs of adults explain nearly 19% of the variance in childhood-onset schizophrenia, a rare and severe form of the disorder (28), and 7% in other adult cohorts (29) but <0.5% psychosis symptom variance in older children and adolescents, in whom such symptoms are often transient and only sometimes precede schizophrenia (30). These findings raise etiologically relevant questions; for example, is novel genetic variation responsible for liability to the same disorder at different developmental stages or, despite some prima facie similarity, are certain disorder features distinct at different developmental stages? PGSs derived from lifetime disorders (e.g., diagnosed depression in adults) measured in adult cohorts also show poor predictive performance of developmentally delimited symptoms thought to precede those disorders (e.g., internalizing behavior in children) (31). In this sense, available PGSs derived from adult GWASs may assist in diagnosis and risk prediction for stable and more severe conditions but fall short for individuals with transient or evolving symptoms.

Future well-powered longitudinal GWASs and GWASs of age-specific symptomatology may partially address this gap. Analyses using existing PGSs can help optimize GWAS design by identifying which symptoms and behaviors do or do not correlate with genetic risk for psychiatric condition susceptibility at different life stages. For example, PGSs for attention-deficit/hyperactivity disorder (ADHD) associate with earlier developmental milestone achievement (e.g., age at first independent walking) (32), and PGSs for bipolar disorder associate with conduct and oppositional defiant difficulties in childhood (33).

Phenome-wide association studies (PheWASs) of PGSs highlight changes in pleiotropy throughout development (34). For example, age-stratified PheWASs of a PGS for ADHD using electronic health record (EHR) data have shown associations with numerous health conditions throughout the lifespan, including ADHD diagnosis in childhood and extending to substance use disorders in adulthood, even in the absence of an ADHD diagnosis (35). Likewise, a PGS indexing liability toward externalizing symptoms is associated with relevant psychiatric-related behaviors from toddlerhood through early adulthood (e.g., childhood conduct problems, early adolescent, and young adult substance use) (36). Psychopathology in late childhood and adolescence is better predicted by PGSs indexing broad-spectrum psychopathology as opposed to PGSs for specific psychiatric conditions (37,38). While the field generally hedges their bet on stable genetic attributes, innovation may come from picking apart the transient genetic variability because it may index a cohort that, if identified and treated early, could maximally benefit.

GWASs OF TIME-DEPENDENT PHENOTYPES

Current GWASs focus on identifying genetic variants associated with susceptibility to psychiatric disorders. Less is known about the genetic architecture of the time-dependent manifestations of these conditions, such as subclinical symptom trajectories, age of onset, treatment response, and the co-evolution of comorbidities. Increasing evidence outside of psychiatry shows that the genetics of disease susceptibility is often distinct from that of disease progression (39–44). Although longitudinal GWASs in psychiatry have so far yielded modest heritability and few genome-wide significant loci (45,46), largely reflecting the smaller sample sizes and analytic complexity inherent to repeated-measures designs, these limitations do not argue against their value but instead underscore how cross-sectional and longitudinal GWASs effectively complement each other. Below, we summarize findings from the limited number of studies that have investigated the genetics of time-dependent phenotypes in psychiatry and give directions on how such phenotypes could be leveraged for genomic discovery in psychiatry in the future.

Symptom Evolution Across the Life Course.

Recent calls for dimensional and transdiagnostic approaches in psychiatric genetics (47,48) recognize the developmental evolution of psychiatric symptoms and behaviors. One approach to identifying age-specific genetic effects is to conduct serial GWASs within the same individuals, using age-specific measures as the outcome. With sufficient sample size (e.g., in studies of body mass index [BMI] totaling >28,000 individuals at each time point across early childhood), this approach has identified novel genetic markers associated with BMI at specific ages that were distinct from those in adulthood (42,49). Smaller sample sizes currently preclude similar discoveries in psychiatry (50). However, extracting temporally stable components of psychiatric phenotypes, which often show higher heritability and genetic signal (50,51), is a pragmatic first step. A second strategy is to “stitch together” values of the trait across the life course by applying meta-regression analyses (see the Supplement) to data measured in different individuals at different life stages. Using this approach with neuroimaging data from 15,640 participants ages 4 to 99 years (in which each participant had repeated measures spanning at least 6 months) from 40 cohorts, Brouwer et al. (52) showed that some loci have age-dependent associations with changes in brain structure.

Beyond age-specific effects, how a symptom changes over time has important implications for disorder etiology and outcomes (53). For example, the rate of escalation in recreational substance use at specific life stages predicts the likelihood of transitioning to a substance use disorder (54,55), and rapid versus slower age-related cognitive decline associates with different biomarkers, lifestyle factors, and health features (56). Such trajectory phenotypes may be constructed by categorizing individuals into subgroups using longitudinal data (e.g., early-onset, high-frequency use vs. late-onset, low-frequency use of substances) or by deriving continuous values of rate of change over time (i.e., the trajectory’s slope between measurements) (see the Supplement). While we know that psychiatric phenotype trajectories are heritable (13,57–59), these have been investigated less frequently in GWASs because of the high computational burden inherent to longitudinal data analysis (see the Supplement). Existing examples remain sparse and underpowered, including studies of substance use (60), impulsivity (50), cognition in older adults (61), and brain structure (52). However, these small-scale longitudinal GWASs already reveal that a trait and its change over time are genetically overlapping but separable characteristics that yield distinct biological associations.

Age of Onset

Age of onset associates with prognosis for some psychiatric conditions, with early disorder onset often correlating with worsening long-term outcomes (62–64). Case-only GWASs have shown that age of onset is a heritable trait (62,65–67) and that genetic modifiers of age of onset for some conditions are separate from those conferring susceptibility (23,65,68). The genetic correlation between early- versus late-onset depression in adults from the UK Biobank is 0.76 (63); GWASs have identified unique genetic variants associated with childhood versus late-diagnosed ADHD (69); and multivariate analyses have shown that childhood- and late-diagnosed ADHD have different genetic correlations with behavioral, psychiatric, cognitive, and health outcomes (70). Therefore, reporting the mean/median age at diagnosis or age of onset of participants included in GWASs is a valuable but not yet common practice (Figure 1).

Instead of stratifying groups based on age of onset, time-to-event survival models for GWASs have been fundamental to accounting for disorder variability by age-of-onset effects (see the Supplement), although they have predominantly been applied to somatic disorders in older cohorts [e.g., (71–74)]. For psychiatric conditions, Pedersen et al. (75) demonstrated that integrating age of onset into a GWAS of 143,000 iPSYCH participants using the age-dependent liability threshold model increased power for some psychiatric disorders, yielding loci detection that either exceeded (i.e., for ADHD and depression) or was on par with (i.e., for schizophrenia and autism spectrum disorder [ASD]) standard case/control GWASs. The extent to which these age-of-onset GWASs will benefit genetic discovery for conditions with younger age of onset (i.e., most psychiatric disorders) needs further assessment [e.g., trade-offs between power and computational burden (76)]. Nevertheless, even among adult cohorts, understanding genetic influences on the pace of disorder onset could provide valuable insights for early prevention.

Treatment Response

Treatment response is an inherently time-dependent phenotype. Although GWASs of medication response in psychiatry are emerging, inconsistent results across studies of depression (77–79) and schizophrenia (80,81) exemplify the challenges in characterizing this phenotype. Nonetheless, a common finding is that even though disorder GWASs are enriched for medication targets (82–84), the genetic architecture of medication response is largely distinct from that of susceptibility to the index condition, even when considering disorder severity (85). Research leveraging longitudinal primary care records to characterize medication-resistant depression (77,78) has identified unique genetic contributions beyond those captured by diagnosis alone. A GWAS of schizophrenia treatment response (N = ~20,000) versus nonresponse (N = ~10,000) replicated findings from GWASs of schizophrenia susceptibility (80); however, modeling medication response using repeated pharmacokinetic measures of clozapine (i.e., drug treatment) metabolism and conducting GWASs with repeated-measures methods revealed 8 distinct loci with direct relevance to clinical treatment (81). Because GWASs of treatment response lead to pharmacogenetic testing in the clinic (86,87), results may not generalize across age groups. Effective pharmacological treatments for depression, for example, vary as individuals age (16), and genetic analyses may provide insights into why this is the case.

Comorbidity Modeling

Psychiatric conditions rarely manifest in isolation. They often co-evolve with other conditions to form complex heterogeneous profiles (88). These overlapping and evolving patterns, including diagnostic switching over time, are critical to understanding disease mechanisms and tailoring therapeutic approaches. Advances in computing power and the growth of longitudinal administrative health datasets comprising multiple data types (e.g., diagnoses, prescriptions, laboratory values) are fueling new studies of longitudinal comorbidity modeling in somatic diseases (see the Supplement) (89). The most notable example in psychiatry is age-dependent topic modeling in the UK Biobank and All of Us Research Program to uncover latent structures defined by co-evolving patterns of comorbidity with depression over time (90). Comorbidity-defined disease subtypes were genetically distinct, as determined by associations with PGSs and specific loci, suggesting that longitudinal comorbidity trajectories capture etiologically meaningful heterogeneity. Caution is needed when applying these methods because this type of modeling largely relies on unsupervised machine learning. Reproducibility and interpretability will be key metrics for model evaluation. Other, more straightforward methods, such as genomic structural equation modeling, offer opportunities to identify genetic sources of across-time differences in the factorial structure of and measurement variance of disorder co-occurrence. For example, a recent cross-disorder GWAS of 14 psychiatric conditions revealed 5 genomic factors, one of which exclusively captured the coheritability between the childhood neurodevelopmental conditions of ASD and ADHD (91). This neurodevelopmental factor correlated weakly (genetic correlation [r_g] < 0.30) with the other factors, except for the internalizing disorders factor (r_g = 0.55). While this study hints at developmental distinctions, whether these same 5 factors representing genetic comorbidity would be evident at other ages remains an interesting and unanswered question.

DEVELOPMENTALLY INFORMED POST-GWAS APPROACHES

Annotation of GWAS Loci to Genes, Cells, and Tissues Across the Life Course

Functional annotation is a cornerstone of post-GWAS analysis in which GWAS loci are computationally mapped to gene-regulatory features in different cell types and tissues. Similar to GWASs, however, most postmortem brain functional annotation resources [i.e., the GTEx (Genotype-Tissue Expression) Project (92), PsychENCODE Consortium (93), and CommonMind Consortium (94)] are dominated by samples from middle-aged and older adults. For example, the median age of donor participants in the GTEx is 55 years (95). However, genetic effects on gene expression can be highly time specific (95–98). This suggests that developmental stage–specific genetic effects on gene regulation are missed in existing resources, which could explain why many GWAS-identified variants remain unannotated (99). Large-scale efforts to quantify these genetic effects on genomic features across the life course will likely yield important insights, as shown by a recent cross-ancestry meta-analysis of gene, isoform, and splicing regulation in 672 fetal brain samples spanning 4 to 39 postconception weeks (96). Gene regulation exhibited specificity across trimesters, with the heritability of gene expression and splicing decreasing across gestation. More than 2400 genes were unique to the fetal stage, and fetal brain molecular quantitative trait loci were twice as likely to colocalize with psychiatric trait GWAS loci as corresponding functional genomic annotations derived from adult tissues. Age-diverse samples are on the horizon, owing to efforts such as the Developmental GTEx Project, which aims to catalog 74 tissues from 120 human donors spanning birth to adulthood (100), although deep phenotyping of psychiatric history is likely to be limited in these collections.

Integrating Data on Changing Exposures Across the Life Course

The relationship between certain exposures and psychiatric conditions depends on when the exposure occurred (101,102). Evidence for gene-environment interplay is widespread (103,104), and some effects appear to be developmentally time limited. For example, genetic variants in genes regulating critical periods in development were associated with increased depression risk, and genetic risk was greater in the presence of childhood maltreatment (105). Many genotyped cohorts collect longitudinal data on environmental factors alongside repeated assessments of symptoms and diagnoses (Box 1). These longitudinal exposure data can be incorporated as time-varying covariates in models of time-varying phenotypes (see the Supplement). However, conditioning on time-varying exposures introduces challenges, including collider or overadjustment bias if those exposures lie on the causal pathway from genetic liability to outcome or are themselves affected by prior outcomes or treatment (106). Careful use of causal diagrams (i.e., directed acyclic graphs) and explicit assumptions about temporal ordering are essential when introducing time-varying covariates.

Mendelian randomization (MR) can help prioritize the most relevant time-varying environmental exposures. However, an inherent assumption of conventional MR is the modeling of lifetime exposure and risk. Multivariable MR is an MR extension that directly compares the causal effects of exposures at different life stages (107,108), including in the context of mediation analysis (109). Multivariate MR must be used with caution, however, because model assumptions are unlikely to hold in realistic settings, and the resulting estimates may not reliably reflect causal effects at a particular time point (110). Nonetheless, multivariable MR has valuable applications in many areas of psychiatry. For example, given the associations between adolescent cannabis use and increased incidence and earlier onset of schizophrenia, multivariable MR could be used to identify how the timing of cannabis exposure causally affects psychosis risk.

MR can also be used to test the causal effects of prenatal exposures during critical periods on offspring outcomes (111), which are often confounded in observational studies by unknown or unmeasured factors (e.g., genotype) that jointly affect maternal exposures and offspring outcomes (112–115). In approximately 15,000 mothers in the MoBa (Norwegian Mother, Father and Child Cohort) study, PGSs for ADHD, autism, and schizophrenia were associated with pregnancy-related predisposing factors thought to affect offspring neurodevelopment, such as depression and anxiety symptoms, coffee consumption, and smoking (112). These findings replicated earlier analyses in >7900 mothers in ALSPAC (Avon Longitudinal Study of Parents and Children) (116). Therefore, MR studies have an important role to play in inferring the causality of prenatal exposures and in focusing prevention efforts on risk factors with strong causal evidence.

ANALYTICAL CHALLENGES TO OVERCOME

Repeated-measures analysis across the life course in large-scale, population-based samples will require technical innovations. Recent GWAS methods explicitly model longitudinal phenotypes to enable the study of symptom evolution over time. The TrajGWAS method (117) handles repeated-measures data from hundreds of thousands of individuals to identify genetic variants affecting the trait mean, analogous to a traditional GWAS, as well as variants influencing within-subject variability over time (i.e., within-subject fluctuations around an individual’s mean value, which may inform disease course and progression). In proof-of-principle analyses of cardiometabolic biomarkers in the UK Biobank participants, approximately 8% of single nucleotide polymorphisms identified by TrajGWAS associated with only within-subject variability and would not have been detected by standard GWAS approaches. FEMA-Long is another GWAS framework for repeated measures (118). Key innovations include the use of unstructured covariance matrices, which allows heritability and genetic correlations to vary across time, support for nonlinear phenotype modeling, and interaction terms to identify time-dependent genetic effects. When applied to infant length, weight, and BMI during the first year of life, FEMA-Long was thousands of times faster than the standard approach of multilevel modeling and revealed time-dependent genetic effects. However, both TrajGWAS and FEMA-Long rely on individual-level data to estimate within-subject covariance structures. While this design maximizes statistical efficiency, it restricts scalability for large-scale meta-analyses.

Misclassification in GWASs reduces statistical power for gene discovery and dilutes heritability estimates and genetic correlations (119,120). Several groups have shown that using longitudinal data from EHRs can yield more accurate phenotypes and reproducible genetic associations than a single survey questionnaire measure (121–124). Therefore, using reports across multiple time points to verify diagnoses and improve the power of psychiatric disorder GWASs represents a straightforward opportunity. However, longitudinal data can be sparse and influenced by selection bias. Genetic associations with study participation and attrition can distort inference if unmodeled (125), including when diagnosis is a condition for inclusion in a longitudinal study, resulting in collider (index event) bias (106). Remedies to reduce bias from nonrandom follow-up include the use of inverse probability weighting and linkage to administrative records, as well as multiple imputation to account for uncertainty in missing phenotypes. Even a limited number of imputations (e.g., 5–10) can improve parameter estimation in large-scale analyses (126,127). Generative models of disease progression may also boost genomic discovery by extrapolating long-term disease trajectories from short-term symptom or biomarker data (128). Although computationally demanding, these approaches will be essential to prevent biases arising from nonrandom missing data in longitudinal studies.

Accounting for relatedness in longitudinal GWASs poses unique challenges because both familial and within-individual correlations must be modeled. New methods such as SPA_GRM (129) use saddlepoint approximation (SPA) within a genetic relationship matrix (GRM) framework to control for these dependencies while modeling time-structured random effects. Because longitudinal biobank data increasingly include multiple family members sampled over time, such methods will be essential to ensure unbiased estimation of genetic effects.

A PATH FORWARD

The gold-standard design to capture the life course is a prospective longitudinal study. De novo phenotype collection at a longitudinal scale is a costly and resource-intensive undertaking, and funding for these initiatives may not be immediately available. Extending genomic data collection to cohorts that have already been longitudinally phenotyped could prove to be a worthwhile investment. Creative solutions for longitudinal phenotype acquisition in biobanked samples should also be supported. For example, well-curated passive data sources such as EHRs can provide researchers not only with medical status but also with longitudinal information on biomarkers and other nonpathogenic traits. Phenotype collection could also be expedited using digital strategies such as wearable technologies and remote assessments (i.e., mobile apps, telehealth) (130,131). The extensive scale of longitudinal EHR data in biobanks must be balanced against the challenges of working with these data. They are well suited for wide rather than deep phenotyping, there are concerns surrounding data accuracy and known biases that disproportionately affect marginalized groups (132), and thoughtful statistical and quality control methods are required to address inevitable missing data, especially for youth and elderly cohorts, which can reduce power and introduce selection bias (133). Finally, many data repositories have access restrictions and lengthy wait times that bottleneck research studies, although efforts to expedite data sharing without compromising participants’ privacy are continually evolving (134).

Collaborative research is essential to advancing the longitudinal research agenda described in this review. At a minimum, researchers should engage with experts in pediatric and geriatric research who can advise on age-appropriate phenotype matching and data collection practices suited to each age group. Furthermore, casting a broad net across psychopathology domains is important given the phenotypic covariance of symptoms across diagnostic boundaries (135), as well as the possibility of heterotypic continuity, in which an underlying liability may manifest differently across developmental periods [e.g., early anxiety disorders as risk factors for later depression (136)]. We anticipate that initiatives such as the Psychiatric Genomics Consortium, PsycheMERGE, and the Lifecourse GWAS Consortium (137), which are composed of more than a dozen working groups focused on different disorders and research areas, will continue to play a central role in fostering researcher cross talk, project collaborations, and data sharing and harmonization.

CONCLUSIONS

We have discussed the advantages of and considerations for incorporating a life course perspective into genetic analyses of psychiatric conditions and the need to continue longitudinal data collection across diverse age groups. Without incorporating a life course perspective into psychiatric genetics, efforts to pinpoint the optimal timing and strategies for personalized prevention, diagnostics, and intervention will remain imprecise, if not unattainable.

Supplementary Material

Supplementary material cited in this article is available online at https://doi.org/10.1016/j.biopsych.2026.04.004.