TABLE 2.

Genetic variance, interaction variance and their covariance component estimates for six phenotypes across POP2 + POP3 with the covariates PC1 and PC2.

UKBB data field	Phenotype	Covariate	var(g0) (SE)	var(g1) (SE)	cov(g0,g1) (SE)	var(e0) (SE)	P-value by LRT comparing with baseline model (DF = 2)
48	Waist circumference	PC1	0.1802(0.0243)	0.0222(0.0069)	−0.0395(0.0079)	0.7990(0.0256)	2.92E−06
		PC2	0.1789(0.0243)	0.0076(0.0037)	0.0300(0.0078)	0.8147(0.0252)	0.0004
21002	Weight	PC1	0.2537(0.0252)	0.0209(0.0069)	−0.0328(0.0081)	0.7270(0.0257)	0.0002
		PC2	0.2529(0.0252)	0.0077(0.0040)	0.0219(0.0080)	0.7408(0.0252)	0.0252
2443	Diabetes diagnosed by	PC1	0.1688(0.0203)	0.0259(0.0070)	−0.0015(0.0077)	0.7901(0.0218)	6.65E−11
	doctor	PC2	0.1734(0.0204)	0.0162(0.0051)	−0.0005(0.0076)	0.7966(0.0219)	3.73E−08
2139	Age first had sexual	PC1	0.0936(0.0258)	0.0267(0.0086)	−0.0072(0.0087)	0.8795(0.0283)	7.86E−05
	intercourse	PC2	0.0933(0.0258)	0.0153(0.0056)	0.0112(0.0086)	0.8918(0.0278)	0.0071
6138	Qualifications	PC1	0.0937(0.0264)	0.0324(0.0094)	0.0159(0.0091)	0.8715(0.0287)	1.06E−15
		PC2	0.1139(0.0267)	0.0150(0.0057)	0.0137(0.0086)	0.8713(0.0285)	0.0162

The phenotypes were adjusted by basic plus additional confounders of fixed effects and transformed by rank-based INT. The results for behavioral phenotypes age first had sexual intercourse and qualifications have demonstrated strong genetic heterogeneity signals, which are consistent with the findings for POP1 + POP2. The significant results for anthropometric traits (waist circumference and weight) and diabetes diagnosed by doctor were also detected. However, these phenotypes were not discovered across POP1 + POP2 with significant G × P interaction signals. SE denotes standard error. DF denotes degree of freedom.