Comment on ‘Large-Scale Cognitive GWAS Meta-Analysis Reveals Tissue-Specific Neural Expression and Potential Nootropic Drug Targets’ by Lam et al.

Abstract

Intelligence and educational attainment are strongly genetically correlated. This relationship can be exploited by Multi-Trait Analysis of GWAS (MTAG) to add power to Genome-wide Association Studies (GWAS) of intelligence. MTAG allows the user to meta-analyze GWASs of different phenotypes, based on their genetic correlations, to identify association's specific to the trait of choice. An MTAG analysis using GWAS data sets on intelligence and education was conducted by Lam et al. (2017). Lam et al. (2017) reported 70 loci that they described as 'trait specific' to intelligence. This article examines whether the analysis conducted by Lam et al. (2017) has resulted in genetic information about a phenotype that is more similar to education than intelligence.

Genetic analysis identifies molecular systems and biological pathways associated with household income

“…Of note are the variables of autism and schizophrenia. As found in previous studies 8,49,51,65,66 schizophrenia showed a small positive genetic correlation with EA (rg=0.06, SE=0.02, P=1.15 × 10 −3 ) whereas, in the present study, income showed a negative genetic correlation with schizophrenia (rg=−0.14, SE=0.02, P=6.49×10 −9 ); the difference between these two genetic correlations was significant (P=6.57×10 −11 ). Autism also showed a positive genetic correlation with EA (rg=0.27, SE = 0.03, P=1.10×10 −15 ) as found previously, 8,51,67 whereas income showed no detectable genetic correlation with autism (rg=0.04, SE=0.05, P=0.37), and this difference was again significant (P=1.17×10 −11 ).…”

A Further Comment on ‘Large-Scale Cognitive GWAS Meta-Analysis Reveals Tissue-Specific Neural Expression and Potential Nootropic Drug Targets’ by Lam et al.

“…Lam et al (2018) also state ‘our leave-one-out analyses (Figure 3 in Lam et al ((2017)) demonstrate that prediction of held-out samples, phenotyped for cognitive ability, are better for MTAG than for either GWAS COG or GWAS EDU alone. This finding supports our interpretation that MTAG is boosting polygenic signal for cognition, and does not support the conclusion of Hill (2018) that the MTAG polygenic signal is ‘indistinguishable from that of education’’. The phrase ‘indistinguishable from that of education’ is attributed to Hill (2018) but it is not found in the manuscript of Hill (2018).…”

“…The third piece of evidence Lam et al (2018) use to show that their MTAG analysis is more similar to cognitive ability than to education, is to claim the magnitude of the genetic correlations with education were reported inaccurately by Hill (2018). In their rebuttal, Lam et al (2018) write that ‘Hill (2018) elides the fact that the calculation method employed by LD score regression is known to sometimes produce values for r g >1, if the variables are so highly similar as to be self-same (Walters, 2016)’. However, Hill (2018) states in his Table 1, Figure 1, and in the publically available scripts used by Hill (2018) to perform his analyses that genetic correlations of greater than 1 are being treated as 1.…”

“…Lam et al (2018) also state that the ‘overall pattern of genetic correlations’ between the three cognitive phenotypes used in Lam et al (2017) — education, the MTAG-intelligence phenotype, and a GWAS composed solely of tests of cognitive ability — is ‘highly similar’. However, attention was called to the results of bipolar disorder and of schizophrenia by Hill (2018) precisely because genetic correlations between education with bipolar disorder and schizophrenia are positive (Bulik-Sullivan et al, 2015; Hill et al, 2015; Okbay et al, 2016), whereas for cognitive ability they are negative for schizophrenia and near zero for bipolar disorder (Hagenaars et al, 2016; Hill et al, 2015). This separation provides the ability to examine whether the associations produced by Lam et al (2017) are indeed trait-specific to cognitive ability as claimed, or are in fact closer to education as shown by Hill (2018).…”

“…In their rebuttal, Lam et al (2018) write that ‘Hill (2018) elides the fact that the calculation method employed by LD score regression is known to sometimes produce values for r g >1, if the variables are so highly similar as to be self-same (Walters, 2016)’. However, Hill (2018) states in his Table 1, Figure 1, and in the publically available scripts used by Hill (2018) to perform his analyses that genetic correlations of greater than 1 are being treated as 1. In addition, the same reference (Walters, 2016) is used in the rebuttal by Lam et al (2018) to state this is a known issue of LDSC regression, as is used by Hill (2018) to justify the decision to report genetic correlations of greater than 1 as 1.…”