DNA from 871 non-diabetic ESRD cases and 948 unrelated non-renal disease controls were genotyped for 15 MYH9
gene single-nucleotide polymorphisms (SNPs) previously shown to be significantly associated with FSGS and HIVAN.11
On the basis of medical record review, the ESRD case group was classified as containing 696 individuals clinically diagnosed with H-ESRD, 80 with presumed chronic glomerular disease, 15 renal biopsy-proven FSGS, 20 HIVAN, 25 lupus nephritis, and 35 other non-diabetic etiologies of kidney disease. Of the 696 H-ESRD cases, DNA from 262 cases was previously sent to Dr Jeffrey Kopp (NIDDK), and 241 of these samples were included in a prior publication.11
These 262 individuals were retained in our overall analysis to enhance power to detect new polymorphisms independent of the previously reported risk haplotype, and the 434 ‘new’ H-ESRD cases (not previously genotyped) were evaluated separately for replication. contains demographic characteristics in study participants. Cases initiated dialysis therapy at nearly 49 years of age, typical of the African-American non-diabetic population in the United States.1
Among the 948 non-renal disease controls, 558 (58.9%) were asked whether they had high blood pressure, and 34% of them (192/558) reported having hypertension, with mean age at onset 44.0±13.0 years (median 45 years). Frequentist estimation of individual ancestry proportion (FRAPPE) was used to calculate the proportion of African ancestry in cases and controls.14
The 70 ancestry informative markers revealed mean African ancestry proportions of 0.79 (s.d. = 0.10) in controls and 0.81 (s.d. = 0.10) in non-diabetic ESRD cases.
Demographic characteristics of African American non-diabetic ESRD cases and controls
The 15 SNPs genotyped spanned 49.3 kb of MYH9, encompassing the majority of the coding region (). Genotyping success rates for the MYH9 SNPs were 96.2–98.74% in cases and controls. A concordance rate of 99.8% was observed in 465 duplicate samples that were genotyped for quality control purposes. Ancestry informative marker SNPs genotyped at a rate of 93.1–98.5%. summarizes allele frequencies and results of the single SNP genotypic association analysis for the non-diabetic ESRD cases and controls. contains single SNP genotypic association results for the 696 cases with H-ESRD (excluding the 175 cases with chronic glomerular disease). SNP rs2187776 failed to meet Hardy–Weinberg equilibrium in the control population (P=0.0005) and was removed from further analysis. The initial age, gender, and admixture-adjusted analysis revealed single SNP associations with non-diabetic ESRD, most significantly in the recessive model, with 13 of the 14 SNPs analyzed. These associations were comparably significant without adjustment for admixture ( and ).
Gene structure and linkage disequilibrium plot of 49 kb of the MYH9 gene
MYH9 single SNP associations in 871 African Americans with non-diabetic ESRDa
MYH9 single SNP associations in 696 African Americans with H-ESRDa
and contain results of a priori MYH9
E1 haplotype associations in all 871 non-diabetic ESRD cases, as well as limited to the 696 cases with H-ESRD, respectively. This analysis was performed to allow for comparison with results previously reported by Kopp et al.11
in FSGS and HIVAN and to allow for the identification of novel risk loci after adjusting for the published risk haplotype. The age, gender, and ancestry adjusted P
-value for the E1 risk haplotype determined using a weighted logistic regression analysis in all non-diabetic cases of ESRD vs controls was 1.22 × 10−15
, with 71.8% of cases and 57.3% of controls homozygous for the risk haplotype (OR 2.38). Limited to the 696 H-ESRD cases, the ancestry-adjusted E1 haplotype P
-value was 4.52 × 10−12
(OR 2.23); a result that survives stringent Bonferroni correction. A novel association in a second haplotype block L1, (1324) consisting of rs7078, rs12107, rs735853, and 5756129, is also displayed.
Logistic regression results—MYH9 haplotypes (all 871 non-diabetic ESRD cases)a
Logistic regression results—MYH9 haplotypes (696 hypertension-associated ESRD cases)a
and contain the results of single SNP and haplotype association analyses limited to the 434 newly genotyped cases with H-ESRD not analyzed in the earlier report.11
Despite reduced power in the smaller but independent sample, single SNP associations remained highly significant (P
-values ranged 0.05–9.1 × 10−11
recessive; admixture adjusted OR 1.35–3.23), and the 3224 and 1324 haplotypes were also strongly associated (P
= 2.6 × 10−6
and 3.1 × 10−7
, respectively). This analysis replicated the association between MYH9
and H-ESRD in a new sample.
Table 4a MYH9 SNP associations in 434 new H-ESRD cases (excluding subjects evaluated in the study by Kopp et al.11)
Table 4b MYH9 haplotype associations in 434 new H-ESRD cases (excluding subjects evaluated in the study by Kopp et al.11)
and contains SNP association analyses, adjusted for the presence of the powerful E1 risk haplotype (—E1 additive model, 5b—E1 recessive model). After adjustment for the E1 haplotype, rs5756152, rs12107, rs1005570, rs16996674, and rs16996677 remained independently associated with non-diabetic forms of ESRD (P = 0.047–7.1 × 10−6; OR 1.20–1.50, additive). This suggests that although E1 by itself explained a significant portion of the variation, additional SNPs in the region independently contribute to the risk of non-diabetic ESRD.
Logistic regression results for individual SNPs, adjusting for 3224 haplotype (additive model) in all 871 casesa
Logistic regression results for individual SNPs, adjusting for 3224 haplotype (recessive model) in all 871 casesa
contains the results of the multilocus SNP analysis using logistic regression. After adjustment for age, gender, and admixture, three SNPs exhibited independent evidence for association, rs4821480, rs5756152, and rs12107. The estimated OR for an individual SNP in this analysis adjusts for all effects in the model, including all other loci. In addition, only rs4821480 was part of the E1 risk haplotype, associated at a P-value of 7.0 × 10−9. This model demonstrates that after controlling for age, gender, and admixture three SNPs provide discriminating power to distinguish between non-diabetic ESRD cases and non-nephropathy controls.
Results of stepwise logistic regression analysis performed on all SNPs
We examined the contribution of the covariates and SNPs to the C-statistic (area under the receiver-operator characteristic curve).15
The C-statistic is a nonparametric measure of the predictive ability of a model and a function of sensitivity and specificity. It is the probability that for a randomly selected pair of subjects, one with non-diabetic ESRD and one without, the individual with ESRD has a higher predicted probability of disease than the individual without ESRD. A C
= 0.50 is completely random. Using the same sample for all three estimates, the C-statistic was C
= 0.63 for the model with age and gender only, C
= 0.64 for age, gender and admixture estimate, and 0.70 for age, gender, admixture, and the three SNPs. Thus, the three SNPs markedly increase the predictive ability of the model in these data, as well as are individually statistically significant.
The individual SNP association analysis and stepwise association analysis (, and ) identified multiple SNPs outside of the original E1 haplotype that predicted risk for non-diabetic ESRD. In particular, the first four SNPs in , rs7078, rs12107, rs735853, and rs5756129, potentially explain additional risk and comprise a second haplotype block. The 1324 L1 haplotype in this block was present in 72.2% of ESRD cases and 57.5% of controls. A logistic regression model that adjusted for the original E1 (3224) haplotype under a recessive model, as well as for age, gender, and admixture, provides evidence that the second haplotype (1324) explained independent risk for non-diabetic ESRD (P = 0.00009, additive) (). Collectively, it is clear from the individual SNP analysis, the multiple locus analysis, and the multiple haplotype analysis that there were several loci with independent evidence for association to non-diabetic ESRD. The relatively high linkage disequilibrium (LD) within the region complicates the elucidation of these effects. However, these multilocus analyses test for associations conditional on the effects at the genotyped loci that are in LD. Finally, rs5756152 remained an independent predictor of risk, even after adjusting for the 1324 and 3224 risk haplotypes (data not shown).
Results of logistic regression analysis using both associated haplotypes