Genetic data yields millions of SNPs that must be analyzed, and the team used several techniques to narrow down the pool to the SNPs most related to the growth curves and measures of obesity.
“We first evaluated the impact of each SNP individually on obesity-related measures, as a way to remove those that were clearly not related,” said Ana Kenney, graduate student in statistics at Penn State at the time of the research and now a postdoctoral researcher at the University of California, Berkeley. “Some studies choose to stop at this step, however we narrowed down the pool even more by looking at all the remaining SNPs simultaneously and eliminating those that did not appear to have an impact when considered along with others.”
This process yielded 24 SNPs that the researchers incorporated into a polygenic risk score. The scores, built based on growth curves, turned out also to be related to other, more commonly used measures; they were higher in children with higher conditional weight gain—the change in weight gain over the first 6 months — and with rapid infant weight gain — a predictor of obesity later in life.
The research team further narrowed the pool to five of the most “stable” SNPs — the SNPs that had the most impact even when they perturbed the data. From these five SNPS, they produced a second score that could be used as a simpler alternative.
“Although the score with 24 SNPs is more powerful than the score with 5 SNPs, we verified that both are useful measures of obesity risk, and we believe either could be used in a clinical setting,” said Matthew Reimherr, associate professor of statistics at Penn State. “A score that requires fewer SNPs to be typed should make it easier to produce in clinics.”
Notably, the scores produced in this study also predicted obesity in older children and in adults, which the research team verified using publicly available datasets. However, scores produced from other studies that were based on obesity information in adults did not translate to the young children in this study.
“This suggests that the genetic signals related to obesity that we see in early childhood are critical across the lifecourse,” said Ian Paul, professor of pediatrics and public health sciences at Penn State College of Medicine. “However, as people age, they start manifesting other parts of their genetic composition. Scores based on early signals seem to be more robust throughout a person’s lifetime. This highlights the need for more studies that focus on identifying risk and preventing obesity in young children, particularly in the ‘first 1000 days’ spanning pregnancy and the first two years after birth.”
The study also demonstrates that smaller studies that deeply characterize individuals and take advantage of functional data analysis techniques can be a powerful alternative to typical large-scale genetic studies.
“These techniques can open doors to smaller labs with fewer resources,” said Craig. “By working carefully and rigorously to collect longitudinal data from more targeted cohorts, and by using powerful statistical techniques, you can still manage to find useful information with a study that is orders-of-magnitude smaller than typical GWAS studies.”
In addition to Craig, Makova, Chiaromonte, Kenney, Reimherr and Paul, the research team includes Junli Lin, a research associate at Penn State at the time of the research; Leann Birch, late professor of foods and nutrition at the University of Georgia who helped lead INSIGHT; Jennifer Savage, director for the Center of Childhood Obesity Research and associate professor of nutritional sciences at Penn State; and Michele Marini, research technologist and statistician for the Center of Childhood Obesity Research at Penn State.
This work is supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK); the Penn State Eberly College of Science; the Penn State Institute for Computational and Data Sciences; the Penn State Huck Institutes of the Life Sciences; and the Pennsylvania Department of Health using Tobacco CURE funds. Additional support was provided by the National Science Foundation.