Based on the genotypic associations, we analyzed the SNP × SNP interactions to explore the potential joint effects of the six SNPs. The results showed that all SNPs were independent from each other (
Supplementary Table S4). We then performed joint-SNP analysis for all possible combinations of these six SNPs to investigate the effects of paired SNPs on Aniso-SE and -AL development (
Supplementary Tables S6–
S20). For joint analysis between
ZFHX1B rs13382811 and
PAX6 rs644242, the wild-type genotypes of
PAX6 rs644242 and
ZFHX1B rs13382811, that is, CC-CC, were used as a reference. Considering the small cell counts of the genotypes of
ZFHX1B rs13382811 TT and
PAX6 rs644242 AA, that is, TT-AA (
n = 3), the Firth regression was applied when it was compared with CC-CC. The TT-AA genotype conferred a 12.43-fold risk toward Aniso-AL development (
P = 0.01) (
Supplementary Table S5), and a mild but insignificant risk toward Aniso-SE development (OR, 1.63; 95% CI, –1.51 to 4.77;
P = 0.76) (
Supplementary Table S5). Because
PAX6 rs644242 showed the strongest association with both Aniso-SE and Aniso-AL in the dominant model, the AA genotype was pooled with the AC genotype for further analysis. The joint AC/AA-TT genotype of
PAX6 rs644242 and
ZFHX1B rs13382811 was associated nominally with Aniso-SE (OR, 4.33; 95% CI, 1.39–13.48;
P = 0.01) (
Fig. A and
Supplementary Table S6) and Aniso-AL (OR, 6.90; 95% CI, 2.23–21.36;
P = 0.001) (
Fig. B and
Supplementary Table S6). The joint AC/AA-CT genotype was also nominally associated with Aniso-AL development (OR, 2.39;
P = 0.001) (
Fig.B and
Supplementary Table S6). In addition, nominal associations were found for the joint genotype of
ZFHX1B rs13382811-
KCNQ5 rs7744813 and Aniso-SE (TT-AA vs. CC-AA, OR, 3.29;
P = 0.007) (
Supplementary Table S8), and that of
ZFHX1B rs13382811-
SNTB1 rs7839488 and Aniso-SE (TT-GG vs. CC-GG, OR, 3.01;
P = 0.02) (
Supplementary Table S9) and Aniso-AL (TT-AG vs. CC-GG, OR, 6.54;
P = 0.002) (
Supplementary Table S9). Nominal and significant associations were also found for the joint genotype of
ZC3H11B rs4373767-
PAX6 rs644242 (TT-AC/AA vs. TT-CC, OR, 2.57;
P = 0.0001; CT-AC/AA vs. TT-CC, OR, 2.05;
P = 0.007; CC-AC/AA vs. TT-CC, OR, 2.64) (
P = 0.03;
Supplementary Table S11),
KCNQ5 rs7744813-
PAX6 rs644242 (AA-AC/AA vs. AA-CC, OR, 1.93;
P = 0.001; CC-AC/AA vs. AA-CC, OR, 3.99;
P = 0.008) (
Supplementary Table S12),
SNTB1 rs7839488-
PAX6 rs644242 (AG-AC/AA vs. GG-CC, OR, 2.47;
P = 0.0001; AA-AC/AA vs. GG-CC, OR, 2.58;
P = 0.03) (
Supplementary Table S13), and
GJD2 rs524952-
PAX6 rs644242 (AA-AC/AA vs. AA-CC, OR, 2.65; P =0 .002; TA-AC/AA vs. AA-CC, OR, 2.42;
P = 0.002; TT-AC/AA vs. AA-CC, OR,2 .22;
P = 0.03) (
Supplementary Table S14) and Aniso-AL.