First Genome Wide Association Study (GWAS) for Maternally Acting Gene Alleles Identifies New Candidate Genes in Autism

Background: Although several environmental factors have been suggested for autism, specific environmental risk factors have been hard to identify. Gene alleles that act in mothers can alter the fetal environment and contribute to the autism phenotype in their offspring. From the perspective of the fetus, maternal environmental factors and maternal genetic factors acting prenatally are both environmental factors. We and others have reported evidence of maternally acting gene alleles (MAGAs) in autism; however, there has been no systematic study of a maternal prenatal contribution to autism. There are at least 45 reports of MAGAs across all disorders, comprising more then 20 individual genes. This number has nearly tripled since the topic was reviewed in 2003 and most contribute to neurodevelopmental disorders. Three have been reported in autism using a candidate gene approach.

Objectives: To use genome-wide array data to identify MAGAs that contribute to autism.

Methods: Genotype and phenotype data were made available by The Autism Genetic Resource Exchange (AGRE). Genotyping was done by Dr. Hakan Hakonarson at the Children’s Hospital of Philadelphia on the Illumina Hap550 GWAS platform for full or partial trios (affected individuals and parents). Families were restricted to those with at least one child with a diagnosis of "Autism without possible non-idiopathic autism". Pedigrees were trimmed to one nuclear family limited to both parents (if available) and all affected offspring, leaving 825 families with a mean of 3.6 genotyped individuals per family. We used the Weinberg log linear method extended to families with more than one affected child by averaging within families. The log linear method examines symmetric mother-father-child types (e.g., 2-0-1 and 0-2-1, where the numbers are the counts of a specified allele in the mother, father, and child, respectively) to test for asymmetric counts in the parental types (e.g., whether 2-0-1 is much more common than 0-2-1). A Poisson regression model is fit to the counts under the null and alternative hypotheses and the likelihood ratio test applied. The EM algorithm was used to incorporate families with a missing parental genotype.

Results: After QC, there were 825 full and partial trios (3,082 individuals) and 468,568 SNPs. One SNP, rs12487874 (intronic in gene RTFN1), showed genome-wide significance (p = 8.6E-11, genome-wide significance threshold p< 5.0E-08). In addition there was no evidence of a child effect for this SNP. We found that families where the mother has two copies of the allele dramatically outnumbered families where the father has two copies. There was no apparent effect for one copy versus none. Other SNPs of interest were in or near GLI2 (8.28E−08), HHLA2 (1.86E−05), PHF21B (1.44E−05) and TLE1 (1.35E−05). The loci near RTFN1 and GLI2 seemed particularly notable for the lack of SNPs in linkage disequilibrium (LD) with our index SNP. SNP rs7120625 on Chr #11 has a suggestive p-value, but the lack of good p-values for SNPs that are in partial LD makes it less interesting.

Conclusions: This was the first GWAS for MAGAs. Several promising loci were found. The results will need to be replicated.