Heritability of Psychological Traits and Developmental Milestones in Infancy

This systematic review and meta-analysis investigates the correlation of genetic and shared and nonshared environment factors with infant developmental milestones.


Introduction
[4] Investment in early childhood is argued to be one of the most effective economic strategies through promoting long-term socioeconomic and health outcomes. 5Investment before age 2 years, in particular, appears to be associated with the greatest rate of return for investment. 5This is reflected in an increasing policy focus globally on the first thousand and one days from conception to age 2 years. 6idence suggests that complex traits are substantially but not entirely, heritable. 7nsequently, to gain understanding of the development of traits in infancy, it is important to draw on literature examining genetic and environmental factors associated with infant trait variation.The quantitative genetic method most widely and comprehensively performed in infancy is the classical twin design, which has been used for more than a century to partition phenotypic variance into additive genetic variance (heritability) and variance in the shared and nonshared environment.
Family studies comparing biologically related siblings or parent-offspring pairs are typically unable to separate genetics from shared environment.In contrast, the classical twin design can provide separate estimates of heritability (the proportion of trait variation explained by genetic differences) and shared and nonshared environment.Twin studies are more feasible than adoption studies (which compare degree of resemblance between adoptees and their birth parents with resemblance between adoptees and their adoptive parents) to conduct at scale during infancy because adoption often occurs later in childhood.This has resulted in a far smaller and less comprehensive body of evidence in infancy from adoption studies than twin studies.The molecular genetic literature on infant traits is also small; the first genome-wide association study of infant traits was only recently conducted, 8 and most molecular genetic studies in infancy have used candidate gene association methods, which in general have failed to yield replicable findings. 9landmark meta-analysis, 7 synthesizing virtually all twin studies of complex traits (predominantly psychiatric, metabolic, and cognitive traits) found a heritability of 49% across the lifespan when all traits and age groups were combined.The analysis combined data from infants and older children, calculating pooled estimates for children aged 0 to 11 years.Infancy is a rapid and sensitive period of development that deserves special focus.To address this, we conducted the first, to our knowledge, meta-analysis of twin studies of psychological and developmental functioning, disability, and health in infancy (birth to age 2 years), calculating pooled estimates of heritability and shared and nonshared environment.

Methods
This study protocol was registered with PROSPERO (record number, CRD42019151532), and the systematic review and meta-analysis were performed in line with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline 2020 statement and Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guideline proposal for reporting.Given that the review involved the synthesis of anonymized information available in the public domain, it was exempt, according to University College London Research Ethics Committee (UCL REC) regulations, from requirements for ethics review by the UCL REC and the need for informed consent.

Search Strategy
PubMed and PsycINFO databases were searched on November 30, 2018; February 5, 2020; and February 11, 2021, for twin studies (a genetically informed design described in the eMethods in the Supplement) of psychological traits and developmental milestones in infancy, using the search terms in eTable 1 in the Supplement.Search results were imported into EndNote software version 9 (Clarivate).C.A. reviewed duplicates identified by EndNote, deleting true duplicates, and screened titles and abstracts of identified records against inclusion and exclusion criteria (eTable 2 in the Supplement).Full texts were retrieved for nonexcluded records, and these, along with references of

Statistical Analysis
In the R package metafor, 12 we conducted two 3-level multilevel random-effects models (incorporating sampling variance, within-cohort variance in outcome measurements, and betweencohort variance) on twin correlations weighted by sample size from 10 categories of the ICF-CY containing data from 5 or more twin cohorts (eMethods in the Supplement).Zygosity was included as a moderator, with the dizygotic (DZ) group coded as the reference category in the first model to obtain a pooled monozygotic (MZ) twin correlation (MZ r) and standard error.The second model was identical but reparameterized with the MZ group as the reference category, producing a pooled DZ twin correlation (DZ r) and standard error.To allow for differences in variability in MZ and DZ subsets, models had a random error structure creating separate study-level and outcome error terms for MZ and DZ twins.As detailed in the eMethods in the Supplement, using pooled correlations and variances from the multilevel model, we calculated estimates for how much of the variation in each ICF-CY category was explained by additive genetic factors (A), the shared environment (C), and the nonshared environment (E, known collectively as ACE estimates) from standard univariate twin models estimated in the meta-analytic context using the R package metaSEM. 13We used 95% CIs around the pooled estimates in the twin study meta-analysis.Forest plots for analyses were produced using the R package metafor version 2.4-0 for R statistical software version 4.0.2(R Project for Statistical Computing). 12We calculated I 2 for each of 3 levels in multilevel models.According to Cochrane guidelines, I 2 Յ 40% suggests low heterogeneity, while I 2 = 30%-60% suggests moderate heterogeneity and I 2 Ն 50% indicates substantial or considerable heterogeneity. 14To reduce heterogeneity, analysis steps were repeated in 10 ICF-CY subcategories (with data from Ն5 samples) and 3 ICF-CY categories (with separate data from parents and observers from Ն5 samples) by parent and observer subgroup (for 6 meta-analyses in total) given that differences in rater have been found to be associated with differences in heritability estimates. 15,16 ran Egger tests of publication bias using the standard error as the estimator, and created funnel plots, plotting effect sizes against standard errors. 17In line with Cochrane recommendations, publication bias tests were run only on estimates in trait categories containing at least 10 estimates. 14ger tests of publication bias were 2-sided and were considered significant at P < .05.Data analysis was conducted March through September 2021.

Results
We identified 5047 publications (4675 publications in databases and 372 publications in references).

Analysis of Phenotypes by Category
Among 10 categories of infant psychological and developmental functioning, disability and health displayed in Table 1 and defined in the ICF-CY, 11 there were enough data from independent samples for meta-analysis (Ն5 samples).Results are reported in Remaining categories had lower estimates with CIs above 0 (pooled e 2 range, 0.18-0.33)(Table 2).

Analysis of Phenotypes by Subcategory and Rater
To reduce heterogeneity, we analyzed 10 subcategories of the ICF-CY (with data from Ն5 samples) and 3 phenotypic categories (with separate parent and observer data from Ն5 samples) by rater (for 6 subgroups: 3 with parent report and 3 with observer report).Full findings are reported in eResults and eTables 8 and 9 in the Supplement.Parent-rated phenotypes in the 3 examined categories (psychomotor and emotional functions and basic interpersonal interactions) had higher heritability and lower nonshared estimates than observer ratings and comparable shared environment estimates.

Publication Bias
Possible publication bias was detected in the unexpected direction across all categories.Findings are in eResults, eTables 10 to 11, and eFigures 14 to 18 in the Supplement.

Discussion
Drawing on a systematically-retrieved pooled sample of 79 044 twins, this systematic review and meta-analysis found evidence that most domains of functioning, disability, and health in psychological and developmental milestones were heritable in infancy and had moderate to high nonshared estimates.Contrary to evidence in older ages, 7 shared environment estimates were high across several important domains of infant development.

Heritability
Consistent with evidence in older samples, 7 all meta-analyzed categories had heritability estimates with 95% CIs above 0 in infancy, apart from sleep and language functions.Estimates were high (Ն40%) for important areas of development (psychomotor, attention, and emotional functions; family relationships [attachment and dependency]; and complex interpersonal interactions [behavioral problems]), suggesting that these categories may be particularly suitable candidates for gene mapping.
High heritability in infancy of attention functions was consistent with the high heritability of attention-deficit/hyperactivity disorder (ADHD) and ADHD traits in older samples. 157In accordance with the very high heritability of autism, 158 a neurodevelopmental condition involving differences in social interaction, social cues in relationships, and regulating behaviors within interactions were among the most heritable subcategories.Absence of evidence that infant language was heritable was consistent with evidence that the heritability of cognition, including language, is low in early development, increasing with age. 159,160e higher heritability of parent-rated than observer-rated phenotypes may be driven by contrast bias in parental reports of phenotypes among children who were DZ twins, exaggerating DZ differences, or by assimilation bias in parental reports of MZ twins, exaggerating MZ similarity. 161rrelated rater bias that inflated MZ and DZ twin similarity equally would lead to inflated shared environment estimates.Without raw data from individual studies, it was not possible to test this by examining variance-covariance structures, which can uncover evidence of contrast and assimilation bias.Overall, our results suggest that individual differences in growth, motor, cognitive, and emotional development may be associated with genetic factors as early as the first 2 years of life.

Shared Environment
Contrary to evidence in older age groups, 7 shared environment estimates had CIs above 0 in several domains and were high for language, sleep, growth maintenance, and basic cognitive functions, reflecting a broader trend noted in the literature that shared environment estimates for language and cognition are higher in early development. 159,160This may have important implications for obesity prevention and efforts to promote intellectual outcomes, which are among the most robust estimators of health and longevity. 162Shared environment estimates had CIs overlapping with 0 for psychomotor, attention, and emotional functions; basic interpersonal interactions; and family relationships.This was consistent with pooled findings in older age groups 7 and evidence that shared environments do not contribute as much to similarity between siblings as genetics and do not contribute as much to differences between siblings as nonshared environments. 163,164

Nonshared Environment
Nonshared environment estimates had 95% CIs above 0 for all phenotypic categories and were high for emotional and attention functions, family relationships, and basic interpersonal interactions.
Higher nonshared estimates for observer ratings than parent ratings were consistent with wider research 165 and may reflect the importance of each twin's unique experience in the expression of phenotypes specifically when rated by observers.Alternatively, given that nonshared estimates also include measurement error, higher observer-rated estimates may reflect increased error in observational measurement.

Limitations
This study has several limitations.Given that research designs all have limitations and biases, establishing robust evidence ideally involves triangulation of methods.However, the classical twin design is currently the only quantitative genetic method that has produced data from enough independent samples to conduct adequately powered meta-analyses across a comprehensive range of infant traits.The generalizability of twin findings may be limited because some infant phenotypes (eg, language and birth weight) develop differently in twins compared with individuals from singleton births. 166,167However, given that our aim was to examine individual differences rather than investigate how and why groups differed, mean differences between twins and singletons may not indicate issues with generalizability.
Although the twin method can be used to examine genotype-environment correlation or interaction, we did not synthesize findings on these outcomes.In twin modeling, ignored interaction between genotype and shared environment is estimated as heritability and ignored interaction between genotype and nonshared environment will be estimated as nonshared environment, potentially contributing to biased estimates.

Figure .
Figure.Ternary Plot of Pooled Heritability and Shared and Nonshared Environment Estimates by Phenotypic Category

Downloaded From: https://jamanetwork.com/ on 09/17/2023 included
Heritability of Psychological Traits and Developmental Milestones in Infancy publications, were screened by C.A. and M.M. Uncertainty about whether publications met inclusion criteria was resolved with senior researchers P.F. and A.R.

Quality Assessment and Data Extraction
containing data on 79 044 twin pairs (31 053 MZ and 47 991 DZ twins), 52 twin cohorts, 21 countries, and 6 continents between 1972 and 2020.The sample included 66 407 twin pairs from Western, educated, industrialized, rich, and democratic countries (in Europe, North America, and Oceania; 84.01%) and

Table 1 .
Examples of Phenotypes by Category Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); ICF-CY, International Classification of Functioning, Disability and Health for Children and Youth.

Table 2
and the Figure.Forest plots for these meta-analyses are reported in eFigures 4 through 13 in the Supplement.More twin samples used in these meta-analyses contained parent-reported data (cohort k = 22) than observer-rated data (cohort k = 12) (eResults and eTable 7 in the Supplement).P = .001).Remaining categories had lower estimates with 95% CIs above 0 (pooled h 2 range, 0.24-0.38),apartfrommental functions of language and sleep functions, which had CIs overlapping 0 (pooled h 2 , 0.24 and 0.35, respectively) (Table2).

Table 2 .
Multilevel Random Effects Models of Phenotypic Categories Abbreviations: DZ, dizygotic; ICF-CY, International Classification of Functioning, Disability and Health, Children and Youth Version; MZ, monozygotic.a Heterogeneity.b Definitions for categories and subcategories can be found in cited ICF-CY manual. 11c Number of independent twin cohorts.d Number of estimates (twin correlations).e MZ twin correlation.f DZ twin correlation.g Heritability.h Shared environment.i Nonshared environment.j Sampling variance.k Within-cohort variance in outcome measurement.l Between-cohort variance. 168 Publications Identified in Systematic Literature Search, Presented Alphabetically by First Author eTable 5. Twin Studies Identified in Systematic Literature Search, Presented Alphabetically by Study Name eTable 6. Phenotypes Identified in Systematic Literature Search Coded Using Classification System eTable 7. Estimates and Cohorts in Phenotypic Categories by Rater eTable 8. Multilevel Random Effects Models of Phenotypic Subcategories eTable 9. Multilevel Random Effects Models of Parent and Observer Ratings of 3 Phenotypic Categories eTable 10.Tests for Publication Bias on Twin Correlations by Phenotype Category eTable 11.Tests for Publication Bias on Phenotype Categories by Variance Component in Phenotype Categories With Ն10 Estimates eFigure 1. Prisma Flow Diagram eFigure 2. Bar Chart of Number of Twin Pairs by Country eFigure 3. Bar Chart of Number of Twin Pairs by Continent eFigure 4. Sleep Functions Forest Plot eFigure 5. Attention Functions Forest Plot eFigure 6. Psychomotor Functions Forest Plot eFigure 7. Emotional Functions Forest Plot eFigure 8. Basic Cognitive Functions Forest Plot eFigure 9. Mental Functions of Language Forest Plot eFigure 10.Growth Maintenance Functions Forest Plot eFigure 11.Basic Interpersonal Interactions Forest Plot eFigure 12. Complex Interpersonal Interactions Forest Plot eFigure 13.Family Relationships Forest Plot eFigure 14.Funnel Plots of Association Between Monozygotic Twin Correlation and Standard Error in Phenotype Categories With Ն10 Estimates eFigure 15.Funnel Plots of Association Between Dizygotic Twin Correlation and Standard Error in Phenotype Categories With Ն10 Estimates eFigure 16.Funnel Plots of Association Between Heritability and Standard Error in Phenotype Categories With Ն10 Estimates eFigure 17.Funnel Plots of Association Between Shared Environment and Standard Error in Phenotype Categories With Ն10 Estimates eFigure 18. Funnel Plots of Association Between Nonshared Environment and Standard Error in Phenotype Categories With Ն10 Estimates eFigure 19.Bar Plot of Quality Assessment Scores eFigure 20.Sustaining Attention Forest Plot eFigure 21.Psychomotor Control Forest Plot eFigure 22. Organization of Psychomotor Functions Forest Plot eFigure 23.Regulation of Emotion Forest Plot eFigure 24.Range of Emotion Forest Plot eFigure 25.Expression of Language Forest Plot eFigure 26.Respect and Warmth in Relationships Forest Plot eFigure 27.Social Cues in Relationships Forest Plot eFigure 28.Regulating Behaviors Within Interactions Forest Plot eFigure 29.Interacting According to Social Rules Forest Plot eFigure 30.Psychomotor Functions (Observer Report) Forest Plot eFigure 31.Psychomotor Functions (Parent Report) Forest Plot eFigure 32.Emotional Functions (Observer Report) Forest Plot eFigure 33.Emotional Functions (Parent Report) Forest Plot eFigure 34.Basic Interpersonal Interactions (Observer Report) Forest Plot eFigure 35.Basic Interpersonal Interactions (Parent Report) Forest Plot eReferences.