Charting the cognitive development of children using adult 'polygenic g scores'

This study demonstrates that adult-derived polygenic g scores effectively chart the cognitive development of children, showing minimal predictive power in toddlerhood that increases substantially to explain 12% of variance in early adulthood while also correlating with educational achievement and faster cognitive growth.

The Gist

Imagine you have a seed. You know that this seed comes from a giant oak tree, but right now, it's just a tiny sprout in the dirt. You want to know: How big will this tree get? How fast will it grow? Will it be the tallest in the forest?

For a long time, scientists could only guess by looking at the soil (the environment) or by waiting decades to see the tree grow. But this new study is like having a magic blueprint hidden inside the seed itself.

Here is the story of that study, told simply:

The Magic Blueprint (The Polygenic Score)

Scientists have discovered that our DNA contains thousands of tiny instructions that influence how smart we are. Individually, each instruction is like a single grain of sand—too small to notice. But when you pile them all together, they form a massive mountain of information.

The researchers combined two huge piles of these "smartness instructions":

1. The "Brain" Pile: Based on tests of adult intelligence.
2. The "School" Pile: Based on how many years of school adults finished.

They mixed these two piles together to create a single, super-powerful "Polygenic g Score." Think of this score as a genetic weather forecast. It doesn't tell you exactly what the weather will be on a specific Tuesday, but it tells you if the season is generally going to be sunny or stormy.

The Experiment: Watching the Sprout Grow

The researchers didn't just look at the blueprint; they watched the actual trees grow. They followed 10,000 British children from the time they were toddlers (age 2) all the way to young adulthood (age 26).

They checked the children's brains, school grades, and behavior at every stage:

• Toddlerhood (Age 2-4): The blueprint was a bit fuzzy here. The children were just starting to talk and play. The genetic forecast wasn't very accurate yet because the "soil" (environment) and the child's own development were still figuring things out.
• Childhood & Adolescence (Age 7-16): As the children grew, the blueprint became clearer. The genetic forecast started to match the reality of their school grades and test scores much better.
• Young Adulthood (Age 25-26): By the time they were adults, the blueprint was incredibly accurate. The genetic score could predict about 12% to 15% of the differences in how smart these adults were. That's a huge amount for something as complex as human intelligence!

The "Snowball" Effect

One of the most fascinating discoveries was how the children grew.

Imagine two children starting a race. One has a "high score" blueprint, and the other has a "low score" blueprint.

• The Start: At age 2, they are running almost side-by-side. The difference is tiny.
• The Race: As they get older, the child with the "high score" blueprint doesn't just stay ahead; they actually start running faster relative to their peers.
• The Result: By adulthood, the gap has widened significantly.

This happens because of a concept called Gene-Environment Correlation. Think of it like this: A child with a genetic knack for learning is more likely to choose to read books, ask their parents questions, and get excited about puzzles. These choices create an environment that makes them even smarter, which makes them choose even more learning activities. It's a snowball rolling down a hill, getting bigger and faster as it goes. The genetic score predicts not just where you start, but how fast you pick up speed.

What About Behavior and School?

The researchers also checked if this "smartness blueprint" predicted other things:

• School Grades: It was a very strong predictor. The better the genetic score, the better the grades, especially during high school exams.
• Behavior Problems: It was a weak predictor. Kids with high scores tended to have fewer behavior problems early on, but the difference wasn't huge later in life.
• Height and Weight: The blueprint had almost nothing to do with how tall or heavy they were.

The Big Takeaway

This study is like finally finding the GPS for human development.

Previously, we thought we had to wait until a child was an adult to know their genetic potential. This study shows that the GPS signal is there from the very beginning (even at age 2), but the signal gets stronger and clearer as the child grows.

In short:

• DNA is fixed: You are born with this blueprint, and it never changes.
• Development is a journey: As you grow, your genes interact with the world, making the blueprint's predictions more accurate over time.
• It's a tool, not a destiny: This score is a powerful tool for scientists to understand how children develop, but it doesn't mean a child's future is written in stone. It just shows the tendency of the path they are likely to walk.

The researchers concluded that by combining these genetic clues, we can now "chart the course" of a child's mind from their first steps all the way to their first job, giving us a clearer picture of human potential than ever before.

Technical Summary

1. Problem Statement

While polygenic scores (PGS) derived from Genome-Wide Association Studies (GWAS) have become powerful tools for predicting behavioral traits, their application in developmental psychology faces a specific challenge: temporal mismatch.

• The Gap: Most GWAS for cognitive traits (intelligence, $g$) and educational attainment (EA) are conducted on adults due to the difficulty of assessing large cohorts of children.
• The Question: Can an adult-derived polygenic score effectively predict cognitive development starting from toddlerhood?
• The Hypothesis: Due to the "Wilson effect" (increasing heritability of IQ with age) and substantial genetic continuity across the lifespan, adult-derived PGS should predict early cognitive traits, with prediction strength increasing as the child's phenotype stabilizes and aligns with the adult GWAS architecture.

2. Methodology

Study Design and Sample

• Cohort: The study utilized data from the Twins Early Development Study (TEDS), a longitudinal cohort of 13,759 families with twins born in England and Wales (1994–1996).
• Participants: The analysis focused on 10,346 participants with genotypic data. For cross-sectional analyses, unrelated individuals (one twin per pair) were selected; for longitudinal and structural equation modeling (SEM), the full twin sample was used to maximize power while accounting for family clustering.
• Age Range: Assessments spanned from age 2 (toddlerhood) to age 26 (early adulthood).

Genetic Measures

• Polygenic Score Construction: The authors created a composite "polygenic g score" by combining two existing scores:
  1. g-PGS: Derived from an adult intelligence GWAS ($N \approx 266,000$).
  2. EA-PGS: Derived from an educational attainment GWAS ($N \approx 765,000$).
• Integration Method: They used SMTPred, a Bayesian method that calculates optimal multi-trait weights based on SNP heritability, genetic correlation, and sample size. The resulting composite was weighted 0.91 for EA-PGS and 0.31 for g-PGS.
• Genotyping: DNA was genotyped on Affymetrix and Illumina platforms, imputed to the Haplotype Reference Consortium, and filtered to ~7.3 million SNPs.

Phenotypic Measures

• Cognitive Abilities: Measured via verbal and nonverbal tasks adapted for age (e.g., parent reports at ages 2–4, WISC-III at age 7, Raven's Matrices at age 14, Pathfinder gamified tests at age 25).
• Educational Achievement: Teacher ratings, GCSE/A-level grades, and university degrees.
• Behavior Problems: Strengths and Difficulties Questionnaire (SDQ) across multiple informants (parents, teachers, self).
• Covariates: Age, sex, genotyping chip, and the first 10 genomic principal components (PCs) were included as covariates.

Statistical Analyses

1. Cross-Sectional Regression: Predicting outcomes at specific ages using the polygenic g score.
2. Confirmatory Factor Analysis (CFA): Extracting cross-time latent factors to isolate stable cognitive traits across development, accounting for measurement method variance (e.g., parent vs. self-report).
3. Latent Growth Curve Modeling (LGCM): Modeling intercepts (baseline levels) and slopes (rates of change) to determine if the PGS predicts initial status and developmental trajectories.
4. Distributional Analysis: Testing for linearity (quadratic terms) and comparing extreme groups (top/bottom deciles and >3 SD).

3. Key Contributions

• Creation of a Composite "Polygenic g Score": By empirically combining EA-PGS and g-PGS, the authors achieved a predictor that outperformed either score alone, explaining up to 12% of the variance in adult general cognitive ability ($g$).
• Developmental Trajectory Mapping: The study provides the most comprehensive charting of PGS prediction from toddlerhood to adulthood, demonstrating that genetic prediction is not static but evolves with the phenotype.
• Theoretical Upper Bound Estimation: By integrating the PGS directly into Structural Equation Models (SEM) at the latent level (removing measurement error), the authors estimated the theoretical maximum predictive power of genetics on $g$, reaching 15%.
• Validation of the Wilson Effect: The study empirically validates the Wilson effect using molecular genetics, showing that prediction strength increases as children age, aligning with the increasing heritability of intelligence.

4. Key Results

Prediction Strength Across Development

• Toddlerhood (Ages 2–4): Prediction was minimal to negligible ($\beta \approx 0.03–0.05$).
• Childhood to Adolescence: Prediction increased steadily.
• Early Adulthood (Age 25): Prediction peaked with a standardized beta of 0.36 for general cognitive ability ($g$) and verbal ability, and 0.28 for nonverbal ability.
• Educational Achievement: Prediction peaked at Age 16 (GCSE) with $\beta = 0.42$ (explaining ~18% of variance), then declined slightly for university grades and years of schooling.

Latent Growth and Trajectories

• Baseline: Higher polygenic g scores predicted higher initial cognitive and educational performance relative to peers.
• Growth: Higher polygenic g scores were associated with steeper relative growth rates (positive slope), suggesting that genetic propensity accelerates cognitive gains relative to the cohort over time.
• Behavior: Higher PGS predicted fewer behavior problems at baseline, but the relative gap narrowed slightly over time (slope effects were small).

Distribution and Linearity

• Linearity: The relationship between PGS and outcomes was linear across the distribution. There were no significant quadratic effects, indicating that extreme scores differ quantitatively, not qualitatively, from the mean.
• Extremes: Individuals in the top 3 SD of the PGS distribution showed cognitive advantages of ~1–2 SD in adulthood compared to the bottom 3 SD.

Subgroup Consistency

• Sex and Zygosity: No significant differences were found in prediction strength between males/females or monozygotic/dizygotic twins.

5. Significance and Implications

• Reverse Engineering Development: The study demonstrates that adult molecular genetic findings can be used "in reverse" to chart the emergence of cognitive traits from birth, bridging the gap between adult GWAS and developmental psychology.
• Measurement vs. Biology: The increasing prediction strength with age is attributed to both the stabilization of phenotypic measures (less noise in adult testing) and the alignment of genetic architecture with adult GWAS cohorts, rather than the emergence of new age-specific genes.
• Predictive Utility: The polygenic g score is a robust tool for identifying children at risk for lower cognitive or educational outcomes as early as age 2, though prediction power is significantly stronger in later childhood and adolescence.
• Limitations: The authors note that the composite score has a verbal bias (due to the heavy weighting of EA-PGS), which may explain why verbal ability prediction slightly outperforms nonverbal prediction. Additionally, the findings are specific to populations of European ancestry.

Conclusion
This paper establishes that adult-derived polygenic scores are powerful, stable predictors of cognitive development from toddlerhood to adulthood. The predictive power is not static; it grows as the child's cognitive phenotype matures, supporting the view that genetic influences on intelligence are continuous and increasingly expressed as individuals actively select environments that align with their genetic propensities.