GxEsum: a novel approach to estimate the phenotypic variance explained by genome-wide GxE interaction based on GWAS summary statistics for biobank-scale data

Table 3 Estimates obtained from GxEsum analysis using real data

Main trait	Environmental variable	Main additive genetic variance (\( {\sigma}_{g_0}^2 \))	GxE interaction variance (\( {\sigma}_{g_1}^2 \))	p-value for GxE
BMI	Age	0.216 (0.007)	0.004 (0.002)	1.86E−02
	NEU	0.216 (0.007)	0.007 (0.002)	1.61E−05
	PA	0.218 (0.007)	0.003 (0.001)	2.57E−02
	ALC	0.216 (0.007)	0.003 (0.002)	5.98E−02
Hypertension	BMI	0.152 (0.008)	0.006 (0.002)	2.09E−03
	WHR	0.154 (0.008)	0.005 (0.002)	3.21E−02
	BFP	0.151 (0.008)	0.008 (0.003)	2.66E−02
Type 2 diabetes	BMI	0.141 (0.014)	0.085 (0.022)	1.58E−04
	Diastolic BP	0.198 (0.014)	− 0.004 (0.006)	5.38E−01
	Systolic BP	0.204 (0.014)	− 0.006 (0.006)	3.17E−01

We used a quantitative trait (BMI) and binary disease traits (hypertension and type 2 diabetes) because BMI is known to be modulated by age/lifestyle such as NEU, ALC, and PA [8, 22, 23], and hypertension and type 2 diabetes are known to be associated with obese traits such as BMI, WHR, and BP [24, 25]. The p-value is from a Wald test for the estimated GxE variance not being different from zero. The estimates on the observed scale for the binary traits were transformed to those on the liability scale using Robertson transformation [17, 26]. All estimates were from the GxEsum model
NEU neuroticism score, PA physical activity, ALC alcohol intake frequency, WHR waist-hip ratio, BFP body fat percentage, BP blood pressure

ISSN: 1474-760X