PMC:4572492 / 17400-26812 JSONTXT

{"project":"2_test","denotations":[{"id":"26140448-18300295-2053711","span":{"begin":217,"end":219},"obj":"18300295"},{"id":"26140448-19875614-2053712","span":{"begin":599,"end":601},"obj":"19875614"},{"id":"26140448-24057835-2053713","span":{"begin":3137,"end":3139},"obj":"24057835"},{"id":"26140448-20967843-2053713","span":{"begin":3137,"end":3139},"obj":"20967843"},{"id":"26140448-17615390-2053714","span":{"begin":4487,"end":4489},"obj":"17615390"},{"id":"26140448-17615390-2053715","span":{"begin":5017,"end":5019},"obj":"17615390"},{"id":"26140448-24482837-2053716","span":{"begin":8157,"end":8159},"obj":"24482837"},{"id":"26140448-19875614-2053717","span":{"begin":8305,"end":8307},"obj":"19875614"},{"id":"26140448-22466613-2053718","span":{"begin":9033,"end":9034},"obj":"22466613"}],"text":"Results\n\nType 1 Diabetes Complications\nTraditional analysis using the individual location-only test did not yield a genome-wide significant association of p \u003c 5 × 10−8 (as previously proposed by Dudbridge and Gusnanto28) between rs1358030 and HbA1c (p = 2.3 × 10−7), whereas the JLS-Fisher method did (p = 4.9 × 10−8). Genome-wide significance was also not achieved by the scale-only test (p = 0.01), the distribution test (p = 1.7 × 10−7, estimated from 108 permutation replicates), the LRT (p = 2.0 × 10−7), or the JLS-minP test (p = 4.6 × 10−7).\nKnowing the treatment information, Paterson et al.18 was able to explicitly model this interactor and identified rs1358030 as genome-wide significantly associated with HbA1c levels and interacting with treatment type (p = 3.8 × 10−10 and p = 0.013 for the main and interaction effect, respectively).\n\nCystic Fibrosis Lung Disease: Single-Variant Association Analysis\nWe first noted in Table 1 that all methods (the joint and individual tests using the full sample or various sub-samples) consistently highlighted that variants in both SLC9A3 and SLC6A14 are associated with the lung phenotype, but not those in SLC26A9. For the individual location-only test, although the age cut-off of \u003c18 years appears to be ideal for rs17563161 in SLC9A3 resulting in the smallest p value (p = 3.3 × 10−5) even with the smaller sample size (n = 753), a different cut-off (\u003c20 years, n = 830) yields the most significant location-only test result for rs3788766 in SLC6A14 (p = 0.0002). This illustrates the challenge of specifying and modeling interacting exposures in the context of multiple SNPs and genes of interest.\nWith the entire sample of CF-affected individuals (n = 1,409) encompassing a wider age range and thus greater variability of environmental exposures, we indeed observed evidence for scale differences in the lung phenotype (Levene’s test p = 0.01–0.08) for variants in SLC9A3 and SLC6A14, revealing the possibility of GxG or GxE interaction. (Table S3 provides evidence for SNP-by-age interaction effects from regression models directly incorporating age.) Compared with the distribution and LRT joint tests, the JLS-Fisher test consistently provides the smallest p values for variants in SLC9A3 and SLC6A14.\n\nCystic Fibrosis Lung Disease: Gene-Set Association Analysis\nWe observed evidence of an association between the apical gene set and SakNorm (JLS-Fisher permutation p = 0.0099; Figures 1A and 1B and Table 2). Comparison with the individual location-only and scale-only tests showed that this association does not reach statistical significance using the conventional gene-set location test (regression permutation p = 0.0876; Figures 1C and 1D) whereas the variance component contributed a significant result (Levene’s permutation p = 0.0222; Figures 1E and 1F).\nExamination of the SNP- and gene-specific JLS-Fisher p values and rankings showed SLC9A3 to be the top ranked (Table S4), with three of the top four ranked SNPs being annotated to SLC9A3. This provided consistent support for the relationship between SLC9A3 and lung function as previously reported (Table 1).19,29 In the independent French sample (n = 1,232), we observed replication of the apical hypothesis via the JLS-Fisher test (permutation p = 0.0180; Figures 2A and 2B and Table 2). Again, the standard location testing approach, by itself, was not powerful enough to detect the association (regression permutation p = 0.2058; Figures 2C and 2D), and the added contribution from the scale-testing component was beneficial (Levene’s permutation p = 0.0077; Figures 2E and 2F). After excluding all ten genotyped SNPs annotated to SLC9A3, the apical gene-set test in both the CGS and French samples remained significant (Table 2), suggesting that the JLS-Fisher method identified multiple additional associations within the gene set, beyond the known SLC9A3 contribution.\nThree of the five top-ranked genes in the apical gene set were SLC9A3, SLC9A3R2 (MIM: 606553), and EZR (MIM: 123900), with ten genes in total displaying JLS-Fisher p \u003c 0.05 (Table S4). Of these top five, protein product interactions are known between (1) SLC9A3 (also known as NHE3) and SLC9A3R2 (“SLC9A3 regulator 2,” also known as E3KARP or NHERF2), (2) SLC9A3 and EZR, and (3) SLC9A3R2 and EZR. SLC9A3R1, or “SLC9A3 regulator 1” (also known as EBP50 or NHERF1), is recognized as a paralog of SLC9A3R2 with comparable binding sites to both SLC9A3 and EZR (reviewed in Donowitz and Li30); SLC9A3R1 (MIM: 604990) was the 21st ranked. Based on interaction investigations typically involving intestinal and kidney tissues, a current paradigm is that EZR provides anchorages for the SLC9A3 regulators to the actin cytoskeleton, and probably also facilitates early trafficking of SLC9A3 from the Golgi to the cell periphery, with eventual “hand-off” to the SLC9A3 regulators. The SLC9A3 regulators help to maintain SLC9A3 at the apical membrane and facilitate its dynamic shuffling with endosomes and internal vesicles.30\nNotably, variants within the SLC9A3 complex component genes are significantly associated as a set in the CGS sample using the JLS-Fisher or regression tests (p \u003c 0.0001 and p \u003c 0.0001, respectively; Table 2). The French sample provided replicated support for the SLC9A3 complex using the JLS-Fisher test (p = 0.0415), although in this latter sample, SLC9A3R1 is more highly ranked than SLC9A3R2 and EZR. Removal of the four-gene set, SLC9A3, SLC9A3R1/2, and EZR, and the re-testing of the apical gene set with the remaining 151 genes suggests that association(s) beyond the SLC9A3 regulator complexes also exist (JLS-Fisher p = 0.0329 and 0.0201 in the CGS and French samples, respectively).\nSLC9A3 contains multi-membrane spanning motifs in its amino portion to facilitate transport function, with a large cytosolic carboxyl portion that affords regulation with binding sites for multiple interactors. Although there is significant LD that extends throughout the gene (Figure 3), there was evidence suggesting greater association for variants as a group from the region corresponding to the regulatory portion, compared to the transporting portion (regression, Levene’s, and JLS-Fisher permutation p = 0.0391, 0.312, and 0.1254, respectively). Amino acids 586–660 of SLC9A3 bind the SLC9A3 regulators where an exonic nucleotide variant (rs2230437) is associated with lung function (JLS-Fisher p = 7.6 × 10−6). This synonymous change is in high LD (r2 \u003e 0.8) with four other variants, all with similar association evidence (rs11743825, rs41282625, 5:475625:GC_G, and rs11745923 with JLS-Fisher p = 1.4 × 10−6, 2.2 × 10−6, 2.3 × 10−6, and 2.7 × 10−6, respectively), and all in noncoding positions. There are also no associated coding variants in the major EZR binding site from amino acids 519–595. Similarly, there are no coding variants in the respective PDZ domains of the SLC9A3 regulators or in the FERM domain of EZR that bind SLC9A3. The imputation analysis would have captured the major variation in this gene. Collectively, because the constituents or their direct physical interactions would not appear to be affected, disturbed expression with altered stoichiometry of components or dynamic positioning of the SLC9A3 complex might be contributing to lung disease severity.\n\nSimulation\nResults showed that, under the normal model and in most scenarios considered, the LRT and JLS-Fisher methods had similar power and were more powerful than the JLS-minP and distribution tests (Figures 4 and S3) and the individual location-only and scale-only tests (Figures S4 and S5).\nInvestigation of type 1 error (100,000 replicates) under the normal model showed that all individual and joint tests considered here maintained the nominal error rates at the 0.05, 0.005, and 0.0005 levels when MAF was at least 0.1 and there were ≥ 20 individuals within each genotype group (Tables S1 and S2). However, as MAF or the smallest genotype group size decreased, the LRT method demonstrated inflated type 1 error (Tables S1 and S2). Departure from normality also resulted in LRT having inflated type 1 error, as previously discussed in Cao et al.17\nTo study the type 1 error at the genome-wide level, we conducted association analysis of all 866,995 GWAS SNPs available in the T1D HbA1c example18 with permuted phenotype. Results showed that all joint testing methods provide correct type 1 error control in a finite sample (Figures S1A, S1D, and S1G, respectively, for the JLS-Fisher, JLS-minP, and LRT methods; the distribution test by design has the correct type 1 error control). However, when stratified by minor allele frequency (MAF), we observed that, whereas the JLS-Fisher and JLS-minP tests are slightly conservative for SNPs with MAF \u003c 0.1 (Figures S1C and S1F), the LRT method has increased type 1 error rate (Figure S1I) and would probably not be appropriate when the size of the genotype group is limited (n \u003c 20). Permutation analysis of all of the 565,884 GWAS SNPs available in the CF application example5 provided type 1 error results similar to those of the T1D HbA1c example (Figure S2).\nFor the sensitivity analysis, we observed that the individual location and scale tests and the JLS-Fisher and JLS-minP tests were not affected by poorly captured genotypes; they maintained correct type 1 error level irrespective of the proportion of incorrectly assigned genotypes (Table S5)."}