6. Applications to Lung Cancer Data from the ILCCO/TRICL Consortium
6.7. Results of Joint Tests for Genetic Main and G×E Interaction Effects
We performed four joint tests, namely EHB-GENNJ, as well as CCJ, COJ, MUK-EBJ, on our GWAS data, to search for the association signals possibly missed by classic main effect or pure interaction genome-wide association tests. For all the studies and both exposure models, NE and MH, we only address SNPs with p-values of interaction effect ≤ 10-5 located within ±500kb of coding regions.
In the CE-IARC Study with NE coding, two SNPs on chromosome 15q24-25 rs1051730 and rs8034191 mapped to the nicotine acetylcholine acceptor subunit CHRNA3 and AGPHD1 genes had genome-wide significant p-values applying 𝑇𝐸𝐻𝐵−𝐺𝐸
𝑁𝑁
𝐽 (p = 6.0×10-10, p = 2.4×10
-9). These two markers were previously reported as being in association with lung cancer, identified by the classic main genetic effect test (Amos, Wu et al. 2008, Hung, McKay et al.
2008, Amos, Gorlov et al. 2010, Fehringer, Liu et al. 2012, Timofeeva, Hung et al. 2012) and discussed in Section 6.4. For the same GWAS with MH coding, the same two SNPs rs1051730, and rs8034191 had greater p-values; however, they remained significant on the genome-wide level (p = 5.8×10-9, p = 6.7×10-8). Another signal of two SNPs, rs13106574 (p = 8.5×10-6), rs13149938 (p = 1.8×10-6) on chromosome 4 that both belong to the gene SLC10A6 was discovered in this analysis. The SLC10A6 locus is an important human sodium-dependent organic anion transporter gene, member 6 of the solute carrier family 10 (sodium/bile acid cotransporter family).
Joint 𝑇𝐶𝐶𝐽 identified a SNP on chromosome 5p15 that mapped to the TERT gene: rs2736100 with corresponding p-value of 8.5×10-6. The telomerase reverse transcriptase (TERT) gene is a candidate lung cancer biomarker. Recently, a number of studies reported TERT variant rs2736100 in association with lung cancer impacting differently on lung cancer histology in European populations (Landi, Chatterjee et al. 2009, Truong, Hung et al. 2010, Brenner,
129 Boffetta et al. 2012, Timofeeva, Hung et al. 2012). SNPs of the TERT gene including rs2736100 were also found to be associated with the risk of lung cancer in the Chinese population (Li, Yin et al. 2013, Myneni, Chang et al. 2013). This variant is discussed in Section 6.5. However, it is important to mention that it was previously only identified in a large scale meta-analysis study and that no single study prior to this has reported this signal. We also identified signals on chromosomes 16 and 14 with 𝑇𝐸𝐻𝐵−𝐺𝐸
𝑁𝑁
𝐽 analysis (rs1014744, rs10492572, rs10492573, rs10507886, and rs7982922), see Table 6.4. The ENOX1 protein is the constitutive ENOX family protein with an essential role in the enlargement phase of cell growth (Jiang, Gorenstein et al. 2008). It belongs to the same protein family and is very similar to the ENOX2 gene that expresses on the cell surface of malignancies and is detectable in the serum of patients with cancer (Cho, Chueh et al. 2002, Hostetler, Weston et al. 2009). Three regions on chromosome 13 including 13q14, where ENOX1 is mapped, were reported to influence non-small-cell lung cancer (NSCLC) development (Tamura, Zhang et al. 1997). Another signal in the same analysis (GLC, NE) comprises four SNPs on chromosome 7 rs13244987 (p = 5.1×10-8), rs13438768 (p = 4.2×10
-8), rs847916 (P = 7.9×10-6), rs847918 (P = 6.3×10-8). With the MH model, we found the signals on chromosome 9 as described in Table 6.5 and some additional individual association signals spread along the genome.
In SLRI NE, we identified the following SNPs with 𝑇𝐸𝐻𝐵−𝐺𝐸
𝑁𝑁
𝐽 . Two SNPs rs10517026 (p = 1.98×10-6) and rs10517026 (p = 1.61×10-6) on chromosome 4 mapped to the protein coding region. One marker, namely rs12956176 located in the KLHL14 gene on chromosome 18, had
130 a p-value ≤ 10-6 with 𝑇𝐸𝐻𝐵−𝐺𝐸
𝑁𝑁
𝐽 and ≤ 10-7 with 𝑇𝐶𝐶𝐽 . For the SLRI Study and the MH model, the 𝑇𝐸𝐻𝐵−𝐺𝐸
𝑁𝑁
𝐽 test did not identify any SNPs with p-values ≤ 10-5. The 𝑇𝐶𝐶𝐽 test revealed five SNPs with p-values ≤ 10-5 that belong to genes. Data are in Table 6.5.
In MDACC, MH analysis, SNPs in the SLC24A3 gene (rs1555852, rs2876537, rs4239730) on chromosome 20 form possible association signal 𝑇𝐶𝐶𝐽 (p = 4.6×10-6, p = 3.3×10-6, p = 2.9×10
-6). The SLC24A3 product is known as prostate cancer-associated protein 6.
The most prominent findings of simultaneous testing with each coding (NE, MH), each joint test statistic (EHB-GENNJ, CCJ, COJ, MUK-EBJ) and for each GWAS (CE-IARC, GLC, SLRI, MDACC) are summarized in Table 6.5. In Table 6.5, only SNPs with corresponding p-values
≤ 10-5 for at least one of the joint tests and those located within the known genes or maximum
±500Kb away from the gene were included. Manhattan plots in Figure 6.5 visualize the results for each GWAS. Generally EHB-GENNJ has similar power and as the consequence similar p-values as COJ and MUK-EBJ and smaller p-values compared to CCJ. However when SNP has protective effect against the outcome (negative estimated coefficient of the association with the trait) then CCJ has greater power and as consequence lower p-values for such signals. An example of the later statement are SNPs rs2736100 (TERT, CE-IARC GWAS, MH) and rs9347645 (PARK2, SLRI GWAS, MH) in Table 6.5. Even though the simulation study in Chapter 5 reflected slight power loss of the EHB-GENN compared to both COJ and MUK-EBJ in realdata we observed only minor increase in p-values for the important signals. For example, SNP rs1051730 (CHRNA3, CE-IARC GWAS, NE) has p-value 5.97×10-10 testing with EHB-GENNJ and 5.93×10-10 and 6.42×10-10 for COJ, MUK-EBJ respectively.
131 Table 6.5 Markers indicated by joint tests in ILCCO/TRICL data with p-values ≤ 10-5 for at least one of the joint tests
GWAS E SNP CHR MA Gene p TJEHB-GE(NN) p TJcc p TJco p TJMUK-EB
rs9302935 17 G LOC400618 1,27×10-03 1,93×10-06 1,29×10-03 3,09×10-05
rs1006957 17 T UBB 8,25×10-06 1,35×10-05 8,27×10-06 8,21×10-06
rs4324798 6 A LOC401242 9,35×10-06 3,00×10-05 9,51×10-06 1,30×10-05
rs1076204 11 C ABCC8 3,30×10-05 9,33×10-06 3,48×10-05 1,46×10-05
rs13244987** 7 A LOC645249 5,09×10-08 2,37×10-05 5,04×10-08 5,09×10-08
rs13438768** 7 C ±LOC645249 4,19×10-08 1,32×10-04 4,23×10-08 4,19×10-08
rs847916* 7 G ±SCIN 7,91×10-06 1,05×10-05 7,90×10-06 9,87×10-06
rs847918 7 T ±SCIN 6,34×10-06 1,93×10-05 6,33×10-06 6,89×10-06
rs10849065 12 T B4GALNT3 6,78×10-06 7,79×10-04 6,79×10-06 3,42×10-04
rs7982922** 13 A ENOX1 3,01×10-07 1,20×10-06 3,00×10-07 5,46×10-07
rs10492572** 13 T ENOX1 1,85×10-06 1,15×10-05 1,84×10-06 1,86×10-06
132
NE rs10517026* 4 G LOC100507930 1,98×10-06 4,23×10-07 1,98×10-06 2,40×10-06
rs10517031* 4 G LOC100507930 1,61×10-06 3,39×10-07 1,61×10-06 1,86×10-06
rs12956176** 18 A KLHL14 5,40×10-05 2,25×10-06 5,43×10-05 4,43×10-05
MH rs10517026 4 G LOC100507930 7,01×10-05 6,66×10-06 7,01×10-05 7,13×10-05
rs10517031* 4 G LOC100507930 3,48×10-05 3,28×10-06 3,48×10-05 3,65×10-05
rs9347645 6 C PARK2 2,81×10-03 3,81×10-06 2,82×10-03 1,47×10-03
Abbreviations: GWAS, genome-wide association study; SNP, single nucleotide polymorphism; CHR, chromosome number; MA, minor allele; EHB-GENN, a parametric empirica l hierarchical Bayes approach for G×E interaction; ILCCO, International Lung and Cancer Consortium/ TRICL, Transdisciplinary Research in Cancer of the Lung; ±, denote that SNP locates ±500Kb of the gene; CC, classical case-control interaction estimator; CO, case-only interaction estimator; MUK-EB, empirical Bayes shrinkage estimator; TJ, joint test statistics;
p, p-value associated with the joint test
E = environmental coding (NE = never vs. ever, MH = moderate vs. heavy; Gene = SNP to gene or nearest gene annotation; * denotes SNPs with p-value ≤ 10-5 testing for classica l genetic main effect; ** denotes SNPs with p-value ≤ 10-5 testing for G×E interaction only by EHB-GENN in bold are markers with p-value≤ 10-7;
133 Figure 6.5 Manhattan plots of p-values for SNPs joint effect based on the EHB-GENN test for G×E interaction component
Red line specifies 10-8 level of significance; Green line specifies 10-5 level of significance;
134