Genetic Influences on the Developing Young Brain and Risk for Neuropsychiatric Disorders

Imaging genetics provides an opportunity to discern associations between genetic variants and brain imaging phenotypes. Historically, the field has focused on adults and adolescents; very few imaging genetics studies have focused on brain development in infancy and early childhood (from birth to age 6 years). This is an important knowledge gap because developmental changes in the brain during the prenatal and early postnatal period are regulated by dynamic gene expression patterns that likely play an important role in establishing an individual’s risk for later psychiatric illness and neurodevelopmental disabilities. In this review, we summarize findings from imaging genetics studies spanning from early infancy to early childhood, with a focus on studies examining genetic risk for neuropsychiatric disorders. We also introduce the Organization for Imaging Genomics in Infancy (ORIGINs), a working group of the ENIGMA (Enhancing NeuroImaging Genetics through Meta-Analysis) consortium, which was established to facilitate large-scale imaging genetics studies in infancy and early childhood.

Imaging genetics reveals considerable information about genetic influences on structural and functional imaging phenotypes (1-3), but until recently focused largely on the adolescent or adult human brain (4,5). This is an important limitation because the most dynamic phase of human brain development is from embryonic life through early childhood (6) (Figure 1). Disrupted gene expression in this period can produce lifelong changes in brain morphology and function. Even common genetic variations may affect early neurodevelopmental processes, thereby increasing risk for psychiatric conditions later in life (7). These effects may be detectable in early life via neuroimaging, thereby providing opportunities for identifying at-risk populations in infancy for primary prevention and the development of interventions to adjust adverse trajectories earlier in the clinical sequence. In this paper, we review empirical evidence underlying this hypothesis focusing on magnetic resonance imaging (MRI). Studies using ultrasound (8,9) and studies integrating imaging and epigenetic data (10-13) also provide insights into how genes influence the developing young brain but are beyond the scope of this review. First, we describe the heritability of brain imaging phenotypes in early life. Then, we discuss candidate gene studies of brain structure, function, and connectivity. Next, we review studies characterizing associations between psychiatric risk genes and brain phenotypes in early life using polygenic scores. Then, we discuss genome-wide studies on brain imaging phenotypes in early childhood. Finally, we introduce ORIGINs (the Organization for Imaging Genomics in Infancy), a working group of the ENIGMA (Enhancing NeuroImaging Genetics through Meta-Analysis) consortium, which was established to facilitate large-scale imaging genetics studies in infancy and early childhood.

HERITABILITY
Twin studies have revealed that many brain phenotypes are heritable in early infancy. Genetic effects explain around 85% of the variance in global white matter volume (WMV) and 56% of the variance in global gray matter volume (GMV), at around 1 month of age (14), while heritability of head size is negligible (15). This contrasts with studies of older children and adults where heritabilities greater than 80% are reported for all 3 phenotypes (global WMV, GMV, and head size) (16)(17)(18). Heritability estimates for global cortical surface area (SA) are high in early infancy (78%), while estimates for global cortical thickness (CT) are lower (29%), with significant genetic overlap between the two (19). This differs from adults, among whom SA and CT are both highly heritable (89% and 81%, respectively), with distinct genetic factors contributing to each measure (20)(21)(22). White matter microstructure is moderately heritable in early life, with 30% to 60% of the variability in mean fractional anisotropy (FA) linked to genetic variation and similar estimates for other diffusivity indices (23,24). In adults, estimates for FA range from 72% to 88% (25). Despite large variations in heritability estimates across individual tracts, a single latent measure of white matter microstructure accounts for a great deal of heritable variation in neonates (50%) (26). Similarly, individual differences in regional CT and SA appear to be driven by a common set of genetic factors influencing cortical structure at the global level (19). In both cases, the pattern of results mirrors temporal changes in gene expression which show strong spatial differences in fetal but not postnatal development (27,28). Finally, genetic effects on resting-state functional MRI phenotypes have been observed during the first 2 years of life. Gao et al. (29) reported modest genetic effects on within-network connectivity in neonates, with 3 visual networks and the right frontoparietal network demonstrating above-average effects. At age 1 year, the most heritable networks were the bilateral frontoparietal networks, the salience network, and 2 visual networks. At age 2 years, genetic effects were the strongest for the auditory network. However, genetic effects were not as strong as those reported in adolescents and adults (30)(31)(32). Genetic effects on betweennetwork connectivity are also minimal in neonates (33). Intergenerational transmission of imaging phenotypes has been reported and likely reflects a combination of genetic, epigenetic, and environmental effects (34)(35)(36). One such study examined the intergenerational transmission from mothers to their 5-year-old children and reported significant effects on sulcal phenotypes in the right frontal and parietal cortices (35).
A recurring theme across studies is that heritability is higher in adulthood than in infancy. This might appear paradoxical because interindividual variation in environmental exposures increases with age, but similar patterns are observed for IQ, where increasing heritability during development is called the Wilson effect (37). The Wilson effect is thought to arise from gene-environment correlations that increase with age. In other words, babies and young children have environments thrust upon them, but as they age, they select, modify, and create environments that are correlated with their genetic predispositions (38). Alternatively, higher heritability observed at later ages could reflect stronger heritability of postnatal processes such as myelination and shifts in proportions of white versus gray matter across development. Determining whether a genetic amplification model applies to neuroimaging phenotypes will require large-scale longitudinal studies that address gene-environment interplay across the life span. The ENIGMA plasticity working group has begun tackling this question. Using 5 longitudinal twin cohorts, they demonstrated that rates of brain change are heritable, and heritability estimates of change rates were higher in adults than in children (39). They subsequently identified variants involved in structural brain changes via a genome-wide association study (GWAS) (40). However, their studies did not include infants or toddlers.

Candidate Gene Approaches
Traditional candidate gene studies test for associations between phenotypic outcomes and variation within specific genes selected for suspected roles in organ development or physiology. In imaging genetics, selection is often based on hypothesized involvement in psychiatric disease. The first candidate gene study of brain imaging phenotypes in neonates focused on global and local brain tissue volumes and several genes with known roles in brain development and putative links to psychiatric disease including DISC1, COMT, NRG1, ESR1, and BDNF (41). Many reported effects mirrored findings in adults; others were unique to infancy. For the BDNF Val/Met polymorphism, Met1 neonates had decreased volumes in regions of the right occipital cortex, left hippocampus, parahippocampus, fusiform gyrus, and inferior temporal gyrus and increased volumes in the motor and somatosensory cortex (41). We highlight this result because a recent study partially replicated the original findings. Specifically, Kawasaki et al. (42) reported that Met1 neonates had significantly smaller relative hippocampal volumes.
The largest traditional candidate gene study to include children under 6 years of age focused on Klotho, a gene linked to age-related decline. A significant interaction between Klotho allele status (rs9536314) and age was observed for total brain volume and total GMV, with KLS-VS heterozygotes having larger volumes in early childhood but not in later childhood/ adolescence. Among girls, KL-VS heterozygotes had less WMV than noncarriers, whereas among boys, heterozygotes had greater WMV than noncarriers. No effects were significant in a replication cohort that did not include children younger  Figure 1. Early neurodevelopment is a sensitive period for accumulating transdiagnostic risk for psychiatric disorders. Neurodevelopment in the human brain begins from approximately 2 weeks after conception. Neurogenesis and neural migration are primarily prenatal processes, and neurite outgrowth is minimal after 4 years of age. Myelination and synaptic pruning continue beyond 6 years of age (indicated by the small arrows). Genetic influences on these various processes contribute to psychiatric risk, which accumulates across development and may not manifest until later in life (large orange arrow). than 6 years of age (43), supporting the importance of conducting imaging genetics studies in early life to unveil effects that have been found to be absent in older cohorts.
In addition to age, genetic effects on neurodevelopment may vary based on factors such as prematurity and family history. For example, Krishnan et al. (44) hypothesized that polymorphisms in DLG4 would moderate responses to perinatal inflammation and their impact on white matter microstructure based on gene network analysis of the microglial transcriptomic response to injury in mouse models and complementary, data-driven analysis of protein-protein interactions, transcription factors, and human brain gene expression. The team discovered a specific variant in DLG4 (rs17203281) associated with FA in preterm individuals in 2 independent cohorts. Van Steenwinckel et al. (45) adopted a similar approach, identifying key genes and gene networks in animal models of neuroinflammation-induced hypomyelination and then testing for associations in preterm infants. The researchers revealed that Wnt pathway genes were collectively associated with cerebral structural connectivity. In addition, a study of 13 candidate genes revealed that ARVCF, previously linked to schizophrenia, and FADS2, previously linked to intelligence, were associated with white matter FA in preterm infants (46). These studies highlight the importance of considering potential interactions between genetic variation and early-life environmental exposures given that neither DLG4 nor the Wnt pathway genes would be expected to impact diffusion tensor imaging phenotypes in the absence of perinatal inflammation. With regard to family history, Douet et al. (47) reported that the effects of variants in ERBB4 differed in children with and without a family history of schizophrenia and/ or bipolar disorder. The TT variant for rs7598440 had more pronounced effects on age-related changes (3-20 years) in CT and SA in children with a family history; these children showed steeper increases in frontal and temporal SA in both early and late childhood.
Several studies explicitly tested for gene-environment interactions using the candidate gene approach. COMT single nucleotide polymorphisms (SNPs) moderated the association between antenatal maternal anxiety and prefrontal and parietal CT in neonates (48). The BDNF genotype (Val66Met) moderates associations between methylation patterns and neonatal hippocampal and amygdala volumes (49). FKBP5, which regulates the hypothalamic-pituitary-adrenal axis, moderates the association between antenatal maternal depressive symptoms and neonatal right hippocampal volume (50). For oxytocin receptor (OXTR) gene variant rs53576, a sex-specific main effect was seen for neonatal hippocampal volume. Left hippocampal volumes were larger in GG-homozygotes than A-allele carriers in boys only. Prenatal maternal anxiety interacted with genotype in both sexes: higher maternal anxiety was associated with larger hippocampal volumes in A-allele carriers (51). Additional details on these studies are found in Table 1. A graphical representation (PhenoGram) (52) of genes and associated phenotypes is given in Figure 2.
Interestingly, imaging genetics studies of infants and young children began at a time when microarray genotyping began making large-scale genotyping practical. The GWAS era quickly highlighted weaknesses in the traditional candidate gene approach. Well-powered GWASs failed to support involvement of many traditional candidate genes in psychiatric disorders. This may partly reflect addressable methodological weaknesses including failure to control for population stratification and thereby increasing the risk of false-positive associations due to differences in ancestry. However, the key disadvantage of the candidate gene approach is likely poor candidate selection, given the inadequacy of current knowledge about underlying biological processes. Subsequent meta-analyses of candidate gene studies relevant to psychiatry, including imaging genetics studies conducted in older populations, revealed poor replicability, false-positive associations, overestimation of effect sizes, and publication bias (53). While the imaging genetics literature for infants and young children is not extensive enough to allow meta-analyses, existing studies likely have similar weaknesses.
One promising approach to addressing these problems is to focus future studies on variants that are robustly associated with mental and neurological disorders or adult brain imaging phenotypes in large-scale GWASs. The e4 allele of the apolipoprotein E (APOE) gene meets this criterion. Not only is e4 the strongest known risk variant for Alzheimer's disease, it also has well-documented effects on brain structure and cognition in healthy individuals (54,55). In neonates, APOE e3e4 heterozygotes have significantly lower volumes in temporal regions, compared with e3 homozygotes, and lower volume in the frontal and parietal lobes. e3e4 heterozygotes have significantly greater volumes in specific parietal, frontal, and occipital areas (41). Infant ε4 carriers have lower white matter myelin water fraction (MWF) and GMV measurements in the precuneus, posterior/middle cingulate, lateral temporal, and medial occipitotemporal regions, areas which are preferentially affected by Alzheimer's disease, and greater MWF and GMV measurements in frontal regions (56). Decreased myelin in ε4 carriers in the corticospinal tract, the splenium of the corpus callosum, and frontal white matter, observed in the previous study, was also reported in a longitudinal analysis from the same group in which children were followed from birth to age 5.5 years (57). Regions in which ε4 carriers had greater MWF early on had a decreased rate of MWF development until age 5.5 years, allowing noncarriers to catch up and surpass ε4 carriers at around 3 years of age. In another study, age-related changes in brain structures and cognition were observed to vary depending on genotype, with the smallest hippocampi in ε2ε4 children, the lowest hippocampal FA in younger ε4ε4 children, the largest medial orbitofrontal cortical areas in ε3ε4 children, and age-dependent thinning of entorhinal cortex in ε4ε4 children (58). All these studies suggest that Alzheimer's disease is a neurodevelopmental disorder as well as a neurodegenerative one.
Six SNPs robustly associated with subcortical volumes in adult GWASs have recently been tested for effects in neonates. An association between rs945270 (an intergenic locus downstream of the kinectin 1 [KTN1] gene) and putamen volume was reported, suggesting that at least some variants have detectable effects across the life span (59).
A significant challenge when performing a candidate gene study informed by existing GWASs is how to prioritize genes for follow-up. Fortunately, an increasing array of in silico tools for searching GWAS literature and performing functional characterization of variants can assist with this task (60-62). The ε4ε4 and ε2ε4 genotypes may negatively influence brain development and brain aging at the extremes of age.

Genetic Influences on Young Brain
Another challenge is that candidate gene studies informed by GWASs still focus on only a few selected genes/polymorphisms which account for only a fraction of variants involved in psychiatric risk. One approach to overcoming this challenge is to use polygenic risk scores.

Polygenic Risk Score Approaches
Polygenic risk scores (PRSs) estimate an individual's susceptibility to a complex trait based on prior GWAS summary statistics (63). Efforts such as the Psychiatric Genomics Consortium have produced many well-powered GWASs of psychiatric and neurodevelopmental conditions including schizophrenia, major depressive disorder (MDD), bipolar disorder, attention-deficit/hyperactivity disorder, and autism spectrum disorder (ASD) (64). Summary statistics are often freely available, and researchers can obtain individual data from controlled-access repositories. By examining associations between PRSs and neuroimaging phenotypes measured in infancy and early childhood, researchers are clarifying how genetic risk for these conditions manifests in early life, thereby providing new insights into the etiology of psychiatric and neurodevelopmental disorders. This is a key step in identifying individuals who may benefit from early intervention.
One of the first studies to use this approach in early life was conducted by Xia et al. (65), who found that PRSs for schizophrenia and ASD were not associated with neonatal total brain volume. Like Xia et al. (65), Cullen et al. (59) did not observe associations between neonatal volumes and PRSs for schizophrenia. However, PRSs for schizophrenia were negatively associated with regional GMV and WMV and total WMV in neonates in a different study (66). PRSs for ASD were associated with greater CT and reduced white matter connectivity in children (3 to w14 years) (67). In preterm infants, a PRS for 5 conditions (ASD, attention-deficit/hyperactivity disorder, bipolar disorder, MDD, and schizophrenia) predicted reduced volume of the lentiform nucleus, which plays a key role in motor control, cognition, and emotion (68). The authors hypothesized that genetic risk for psychiatric disorders increased vulnerability to abnormal lentiform development in the context of perinatal stress associated with preterm birth but did not include term infants for comparison. Other studies have used PRSs to probe relationships between early-life adversity, genetic risk, and neurodevelopment. Ursini et al. (69) used transcriptomic data to create placental genomic risk scores (PlacGRSs) for schizophrenia. PlacGRSs were calculated like traditional PRSs, but only used markers in genes highly expressed in placenta and differentially expressed in placentae from complicated, compared with normal, pregnancies. PlacGRS was negatively associated with neonatal brain volume in children with perinatal complications, especially in boys. No significant associations were observed for PlacGRSs and nonplacental GRSs for other disorders and traits associated with early-life complications, suggesting that the link between placental biology, genetic risk, perinatal environmental risk, and early brain development outcomes is relatively unique to schizophrenia.
Another important perinatal stressor is maternal depression. Many studies report associations of maternal depressive symptoms with neuroimaging outcomes in early life, but it is  (71). However, the interaction became nonsignificant after correction for multiple comparisons. The Finnish team also investigated associations of an MDD PRS with infant striatal volumes and found sex-specific effects: the MDD PRS was positively associated with caudate volumes in boys but negatively associated with caudate volumes in girls (72). They did not observe significant interaction effects of the PRS with prenatal maternal depressive symptoms for any dorsal striatal volumes. PRSs can also be used to understand how molecular pathways shape individual differences in neurodevelopment. For example, a PRS for serum testosterone was recently found to be positively associated with total SA development in female infants (73). Researchers interested in this application of PRSs may implement expression-based PRSs (ePRSs) rather than traditional PRSs. ePRSs integrate genotype data with transcriptomic data to predict expression levels of a particular gene or gene network. Morgunova et al. (74) used this approach to investigate relationships between a coexpression network of the DCC gene, which is robustly associated with multiple psychiatric conditions, and total brain volume in both neonates and older children. Higher ePRSs for the DCC coexpression network were associated with larger brain volumes. A study from GUSTO investigated how genes involved in inflammation interact with maternal depression to shape neonatal brain morphology. They created separate ePRSs for 22 cytokine and chemokine genes expressed in fetal brain and found that ePRSs for TNFRSF19, IL17RB, BMPR1B, IL1RAP, and CXCR4 moderated the impact of maternal depression on specific subcortical volumes and regional CT (75). Using longitudinal data from the same cohort, investigators revealed an age-dependent involvement for transmembrane receptor (TGF-b) variants in moderating effects of prenatal maternal depressive symptoms on amygdala volume (76).
Finally, PRSs have been used to investigate how genetic variants linked to adult and adolescent brain morphology influence early brain development. Xia et al. (65) found that PRSs for WMV and GMV in adolescence showed positive associations with neonatal WMV and GMV, respectively, although the overall proportion of variation explained was low. Morgunova et al. (74) calculated PRSs for brain volume using data from UK Biobank, ENIGMA, CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology), and the Early Growth Genetics Consortium. These PRSs did not predict brain volume in their neonate and school-age community cohorts. The first large-scale GWAS of adult intracranial volume tested whether a polygenic score generated from 7 genome-wide significant loci predicted head growth in children of European ancestry who were followed prenatally until 6 years of age (77). The investigators found an age-dependent effect in which the PRS became more predictive in older children, suggesting that adult brain volume is strongly shaped by genetic influences operating in early childhood. Cullen et al. (59) found robust Figure 2. PhenoGram of genes associated with brain imaging phenotypes in infants and young children from candidate gene studies and genomewide association studies. CT, cortical thickness; FA, fractional anisotropy; GMV, gray matter volume; ICV, intracranial volume; MWF, myelin water fraction; PFC, prefrontal cortex; SA, surface area; TBV, total brain volume; WM, white matter; WMV, white matter volume. associations between PRSs for adult brainstem, hippocampus, putamen, and thalamus volumes and neonatal volumes, suggesting some stability across the life course.
Reviewed studies highlight the potential of PRS-based investigations of neuroimaging phenotypes in infancy and early childhood ( Table 2). They also reveal the importance of considering effects of sex and ancestry. A major limitation of this approach is the lack of sufficiently powered GWASs conducted in non-European populations. European ancestry GWASs do not transfer well to other ancestries and can lead to unpredictable biases (78). Another limitation of PRS studies is that they are based on existing GWASs and hence limited by the power of current datasets. In other words, PRS-based studies, like candidate gene studies, are constrained by current biological knowledge. To fully understand how DNA variants influence brain development, well-powered GWASs of infants and young children encompassing multiple ancestries are needed.

Genome-wide Association Studies
Hypothesis-free GWASs can identify new associations and overturn prior assumptions. However, GWASs of neuroimaging outcomes in infants and young children are very limited. The first GWAS of healthy infants identified several common variants associated with neonatal brain structure (65). An intronic SNP in IGFBP7 was significantly associated with GMV. An intronic SNP in WWOX fell just short of genome-wide significance for WMV. Many top associations tagged transcriptional regulators expressed during brain development (KLF13, LMCD1, TOX3, and TBX4). The investigators also compared their results to large-scale neuroimaging GWASs in adolescents and adults and concluded that genetic determinants of global brain volumes are highly distinct at different ages. In a subsequent GWAS of white matter microstructure in neonates, an intronic SNP in PSMF1 was significantly associated with a tractography-based factor capturing shared variation in FA across 44 white matter bundles (26). Additional loci nearing genome-wide significance were in or near genes with roles in axon growth and guidance, fasciculation, and myelination including B3GAT1, TENM2, NFATC1, and MAP3K13. The above studies are the first of their kind, and replication is crucial. Furthermore, these studies were not large enough to generate stable SNP-wise heritability estimates or evaluate genomic correlations between infant neuroimaging phenotypes and psychiatric disorders.
A smaller study in preterm individuals used genome-wide data and pathway-based and network-based approaches (79). The peroxisome proliferator-activated receptor signaling pathway was found to have a role in white matter development, with 5 genes implicated (AQP7, ME1, PLIN1, SLC27A1, and ACAA1). This inspired the team to examine the peroxisome proliferator-activated receptor pathway in a larger cohort of preterm children. Using machine learning analysis, they uncovered 3 genes associated with cerebral connectivity (PPARG, ITGA6, and FXR1) (80). GWASs are summarized in Table 3 and included in the Figure 2 PhenoGram (52). Gene functions and associated neurological phenotypes/conditions for candidate genes and genes identified via GWASs are provided in Table 4.
General strengths and limitations of GWASs have been reviewed in detail elsewhere (81). In terms of GWASs in infants and young children, the primary limitations of existing studies include their being 1) underpowered due to small samples, 2) mostly cross-sectional rather than longitudinal, 3) only focused on neonates and preterm infants, and 4) the fact that individuals of non-European ancestry constitute such a small proportion of the total samples.

Rigor and Reproducibility
When we consider the rigor and reproducibility of published imaging genetics studies in infants and young children, insufficient power and sample size are significant concerns. Sample sizes for early childhood imaging genetics studies are low, with a mean of 365 for candidate gene studies (median = 216), a mean of 225 for PRS-based studies (median = 168), and a mean of 344 for GWASs (median = 371.5). Consequently, existing studies are powered to detect variants with large effect sizes. It is likely that most variants impacting infant brain imaging phenotypes will explain between 0.1% and 1% of the variance, like other complex traits (82). Furthermore, small samples can produce unstable results, and homogenous sampling can generate statistical inferences that do not represent the overall population. Independent replication is essential to validate results and improve estimation of effects but is currently rare due to difficulties recruiting and scanning large groups of infants.
To improve rigor and reproducibility and fully understand how DNA variants influence brain development in infancy and early childhood across diverse populations and the implications for future research and clinical care, large longitudinal studies are needed. ORIGINs was founded to facilitate such work.

ORGANIZATION FOR IMAGING GENOMICS IN INFANCY
ORIGINs includes investigators from different centers around the world (16 sites, 19 cohorts, 5 countries) who are engaged in neuroimaging research in infancy and early childhood. Our goal is to determine how genetic and environmental factors influence development of brain morphometry, anatomical and functional connectivity, and cognitive and emotional function from birth to age 6 years. In 2020, we received National Institutes of Health funding to create the largest-ever imaging genomics dataset focused on infancy and early childhood. In subsequent sections, we briefly describe who will be included in this dataset, what is being measured, and our data analysis plans.  Neonates with a genetic risk score less than or equal to the median showed a positive association between antenatal maternal depressive symptoms and right hippocampal volume. Neonates with a genetic risk score greater than the median showed a negative association between antenatal maternal depressive symptoms and right hippocampal volume.  In neonates with a high GES of TGFBR1, higher levels of prenatal maternal depressive symptoms were associated with a smaller right amygdala volume. In children with a low GES of TGFBR1, greater prenatal maternal depressive symptoms predicted greater left and right amygdala volumes at 6 years of age.

Genetic Influences on Young Brain
Biological Psychiatry May 15, 2023; 93:905-920 www.sobp.org/journal University of Washington, Washington University in St. Louis, King's College London, and National University of Singapore. Eight cohorts have completed initial data collection and 11 are actively scanning. We estimate that 51% of the sample will be female, 49% will be male, 0.05% will be American Indian, Alaska Natives, Native Hawaiians, or Pacific Islanders, 11% will be Asian, 20% will be Black, 6% will be more than one ancestry, and 62% will be White. Individuals identifying as Hispanic/Latinx are expected to make up 15% of the cohort.

Data Measurements
Demographic and Medical History. Health history and demographic information of participants were provided by parents or guardians and/or extracted from medical records. The information includes birth outcomes (gestational age, birth weight), sex, socioeconomic factors (maternal education, total family income), and family history of medical and neuropsychiatric disorders.
Genomic Data. Most participating sites have used/are using saliva samples for DNA extraction. Two cohorts (Drakenstein Child Health Study and GUSTO) used umbilical cord and venous blood specimens. To harmonize data across genotyping platforms, we will impute genomes to a common set of SNPs using the Michigan Imputation Server (83). See the Supplement for details on harmonization and genotyping platforms used by each cohort.
Behavioral Assessments. In a subset of participants (w3800), we will examine 3 behavioral traits-impulsivity/ distractibility, anxiety, and aggressive behavior-that can be reliably measured in very young children and are relevant to multiple psychiatric disorders. Behavioral traits will be measured using age-appropriate versions of the Child Behavior Checklist (84) and the Behavior Assessment System for Children, Second Edition (85,86).
Image Acquisition and Quality Control. 3T Siemens scanners (Allegra, Tim Trio, Verio, Skyra, and Prisma) and comparable sequences (Tables S3-S6) were used with all cohorts except for the dHCP (developing Human Connectome Project) and GUSTO. dHCP data were acquired on a Philips Achieva, and newborn T2 structural MRI acquisition for GUSTO was on a 1.5T General Electric scanner. To ensure consistent processing across datasets with the same tools and appropriately standardized parameter settings, all structural, diffusion, and functional connectivity data will be processed at a central site. T1 and T2 structural MRI, diffusion tensor imaging, and resting-state functional MRI acquisition parameters, platform, and harmonization pipeline for each site are detailed in the Supplement.
Data Analysis Plan. Extracted neuroimaging measures will be analyzed using nonlinear growth models. Growth model parameters, which we refer to as "developmental imaging phenotypes," will be used to test effects of genetic variants on structural brain development and connectivity using a multivariate GWAS approach. Canonical correlation analysis will be used to identify association patterns between genetically influenced neurodevelopmental traits and clinically salient behaviors. The data analysis and data sharing plan is detailed in the Supplement. A schematic for data analysis is provided as Figure 3.

CONCLUSIONS
Imaging genetics studies of infants and young children have provided evidence that variants associated with psychiatric disorders influence early neurodevelopment both independently and through interactions with environmental factors. In addition, GWASs of neonates and preterm infants have revealed new genes, variants, and molecular pathways implicated in brain development. However, most findings have not been independently replicated. Existing studies, regardless of design, are relatively small and do not encompass diverse ancestries. The ORIGINs initiative is addressing these limitations by creating and harmonizing the Genetic Influences on Young Brain   (114), memory (115) ADHD, attention-deficit/hyperactivity disorder; GABA, gamma-aminobutyric acid; HPA, hypothalamic-pituitary-adrenal; IGF, insulin-like growth factor; PTSD, posttraumatic stress disorder; T2DM, type 2 diabetes mellitus.

Overall Sample Size 6809
Aim 1a: Harmonized Genomes 6400 largest and most diverse imaging genomics dataset focused on infancy and early childhood to date. This dataset will help reveal how genetic risk for psychiatric disease manifests across infancy and early childhood, in terms of brain structure and function, and assist in early identification of at-risk individuals. Ultimately, identifying genes and molecular pathways associated with early neuroimaging phenotypes could lead to the development of novel prophylactics against complex psychiatric illness.