Deciphering the Associations between Autism and Schizophrenia: Evidence from a Two-Sample Multivariable Mendelian Randomization Study.

Background:

Although autism and schizophrenia are considered distinct diagnostic entities, the conditions seem to be intertwined at a genetic and phenotypic level. Family history of schizophrenia has been associated to autism risk in the offspring, while individuals with autism are at increased risk of psychosis spectrum disorders. Little is known on the possible cofounding role of IQ in the autism-schizophrenia associations, with observational studies providing inconclusive evidence and whole genome approaches suggesting that the two conditions present divergent associations with IQ. There is a need to decipher the role of IQ in the autism-schizophrenia associations, using approaches robust to the limitations of observational and whole genome approaches. Mendelian randomization (MR) is an instrumental variable approach, utilizing genetic variants as proxies for environmental exposures. The method provides robust control over residual or unmeasured confounding and allows the detection of pleiotropy. This is the first study in the field of autism research to utilise MR in order to:

Objectives:

• Assess the bi-directional associations between genetic liabilities for autism and schizophrenia using two-sample MR.
• Examine whether IQ confounds these associations, by applying an extension of MR, multivariable MR.

Methods:

Instruments for genetic liability to autism were based on the latest GWAS meta-analysis of autism. Due to the small number of significant hits identified (5 SNPs, P-value<= 5e-08), a relaxed P-value threshold for SNP inclusion was utilised to increase statistical power and precision of the estimates (P-value<=5e-07, 10 instruments). Instruments for genetic liability to schizophrenia were 128 independent (r²=0.01) genome-wide significant SNPs identified in the latest GWAS of schizophrenia. In the multivariable MR analyses, IQ was entered as a covariate in the models. Instruments for IQ were 246 independent (r²=0.01) SNPs identified in the largest GWAS on intelligence. Estimates were generated using an inverse-variance weighted model. Sensitivity analyses to identify and exclude outliers and test for violations in the MR assumptions, were conducted.

Results:

There was limited evidence to suggest an effect of genetic liability to autism on schizophrenia (crOR: 0.93; 95%C.I.: 0.82- 1.06; p=0.27). After adjusting for IQ, there was evidence consistent with an approximately 24% increase in the odds of schizophrenia for each unit increase in log-odds liability to autism (adjOR: 1.24; 95%C.I.: 1.08- 1.43; p= 0.004). For the reverse direction, the effect of genetic liability to schizophrenia on autism, the was evidence suggesting that one unit increase in log odds liability to schizophrenia was associated with an approximately 19% increase in the odds of autism (crOR: 1.19; 95%C.I.: 1.14- 1.24; p= 1e-14). The effect estimate did not change markedly upon adjusting for IQ in the model (adjOR: 1.26; 95%C.I.: 1.18- 1.35; p= 1e-10).

Conclusions:

In a two-sample MR framework, we found evidence suggesting bi-directional associations between genetic liabilities to autism and schizophrenia, after adjusting for IQ. The meaning of the findings and the possible contribution of IQ in the autism-schizophrenia associations as well as directions for future research will be discussed.