Association Analysis of Dyslipidemia-Related Genes in Diabetic Nephropathy

Type 1 diabetes (T1D) increases risk of the development of microvascular complications and cardiovascular disease (CVD). Dyslipidemia is a common risk factor in the pathogenesis of both CVD and diabetic nephropathy (DN), with CVD identified as the primary cause of death in patients with DN. In light of this commonality, we assessed single nucleotide polymorphisms (SNPs) in thirty-seven key genetic loci previously associated with dyslipidemia in a T1D cohort using a case-control design. SNPs (n = 53) were genotyped using Sequenom in 1467 individuals with T1D (718 cases with proteinuric nephropathy and 749 controls without nephropathy i.e. normal albumin excretion). Cases and controls were white and recruited from the UK and Ireland. Association analyses were performed using PLINK to compare allele frequencies in cases and controls. In a sensitivity analysis, samples from control individuals with reduced renal function (estimated glomerular filtration rate<60 ml/min/1.73 m2) were excluded. Correction for multiple testing was performed by permutation testing. A total of 1394 samples passed quality control filters. Following regression analysis adjusted by collection center, gender, duration of diabetes, and average HbA1c, two SNPs were significantly associated with DN. rs4420638 in the APOC1 region (odds ratio [OR]  = 1.51; confidence intervals [CI]: 1.19–1.91; P = 0.001) and rs1532624 in CETP (OR = 0.82; CI: 0.69–0.99; P = 0.034); rs4420638 was also significantly associated in a sensitivity analysis (P = 0.016) together with rs7679 (P = 0.027). However, no association was significant following correction for multiple testing. Subgroup analysis of end-stage renal disease status failed to reveal any association. Our results suggest common variants associated with dyslipidemia are not strongly associated with DN in T1D among white individuals. Our findings, cannot entirely exclude these key genes which are central to the process of dyslipidemia, from involvement in DN pathogenesis as our study had limited power to detect variants of small effect size. Analysis in larger independent cohorts is required.


Introduction
Type 1 diabetes mellitus (T1D) has been previously reported to increase the risk of microvascular complications and cardiovascular disease (CVD) [1][2][3]. In contrast to the reduction in cardiovascular mortality within the general US population, the declining trend is less evident in individuals with diabetes [4]. Despite improved disease management strategies, CVD remains the primary cause of death in patients with T1D [5] and a ten-fold increase in risk is reported in those with diabetic nephropathy (DN) relative to those without it [6]. DN is a complex, multifactorial disease and identifying robust genetic risk factors has proved challenging. Several risk factors are common to both CVD and DN, including hypertension, male gender, smoking and modifiable dyslipidemia [5][6][7][8][9][10][11].
Dyslipidemia results from abnormal lipid metabolism with departure from optimum vascular cholesterol and triglyceride levels leading to atherosclerosis, a process of fatty acid plaque deposition in arterial blood vessels. Previous studies reported normal low-density lipoprotein (LDL) and high-density lipoprotein (HDL) cholesterol levels in individuals with T1D, with elevated triglyceride levels more commonly associated with poor glycemic control [12]. This abnormal lipid profile can result from insulin deficiency with subsequent reduction in lipoprotein lipase activity and diminished ability for chylomicron and very low-density lipoprotein (VLDL) clearance [13]. This contrasts with individuals with type 2 diabetes (T2D) who often exhibit reduced HDL levels with a shift in LDL to the more atherogenic dense VLDL particles as a consequence of increased hepatic production. This process is increased by insulin resistance resulting in reduced clearance of VLDL and chylomicrons [14].
Observational studies have identified multiple lipid abnormalities in both incipient and overt DN [15][16][17], although this has not been consistently reported [18]. While the exact mechanism of effect is not fully understood, dyslipidemia has been associated with DN progression as well as increasing cardiovascular risk [19][20]. Supporting evidence implicates insulin resistance as pivotal in the development and/or progression of this condition [21][22][23][24][25]. Potential mechanisms contributing to renal injury in DN have included stimulation of pro-inflammatory and pro-fibrotic cytokine production, cell apoptosis, vasoconstriction and modulation of mesangial cell proliferation [26][27][28]. As such, parallels between mechanisms that underpin atherosclerosis and glomerulosclerosis provide support for investigation of the parameters that contribute to both conditions [29].
While previous evidence demonstrates modulation of lipid profiles through lifestyle changes such as smoking, diet and physical activity, recent studies have also identified common genetic variation as regulators of lipid levels and subsequent dyslipidemia [30][31][32][33][34][35][36][37]. To date, almost 100 genetic loci have been reported in association with serum cholesterol and triglyceride levels [38]. Aulchenko and colleagues highlighted that many of the loci influencing lipid levels and CVD risk had previously been identified in association studies enriched by participants with diabetes [34]. The management of diabetic dyslipidemia, a wellrecognized and modifiable risk factor, is a key element in the multifactorial approach to prevent CVD in individuals with diabetes [23]. In light of the evidence supporting association of these variants with dyslipidemia in individuals with diabetes, we sought to assess the allelic frequency of 53 common single nucleotide polymorphisms (SNPs) in 37 key loci in individuals with DN using a case-control design involving 1467 individuals with T1D. These loci and SNPs were selected on the basis of their functional significance and previous reported association with dyslipidemia [30][31][32][33][34][35][36][37].

Participants
Research ethics approval was obtained from the South and West Multicentre Research Ethics Committee (MREC/98/6/71) and Queens University Belfast Research Ethics Committee. Written informed consent was obtained prior to participation. All recruited individuals were white, had T1D diagnosed before 32 years of age and were born in the UK or Ireland. Patients (n = 718) and controls (n = 749) originated from the Warren 3/UK Genetics of Kidneys in Diabetes (GoKinD) and all-Ireland collections [39]. The definition of DN in cases was based on development of persistent proteinuria (.0.5 g protein/24 h) at least 10 years after diagnosis of T1D, hypertension (blood pressure .135/85 mmHg or treatment with antihypertensive agents) and associated diabetic retinopathy. Controls were individuals with T1D for at least 15 years with normal urinary albumin excretion rates and no evidence of microalbuminuria on repeated testing. In addition, control subjects had not been prescribed antihypertensive drug treatment avoiding possible misclassification of diabetic individuals as 'control phenotypes' when the use of antihypertensive treatment may have reduced urinary albumin excretion into the normal range. Individuals with microalbuminuria were excluded from both case and control groups since it was not possible to be confident in assigning case/control status for such individuals whose urinary albumin excretion might either regress or progress over time [40].

SNP selection and genotyping
SNPs (n = 53) were selected on the basis of previously reported association with dyslipidemia [31][32][33][34] and of minor allele frequency (MAF) exceeding 0.1 in populations of European descent. Genotyping was performed by MassARRAY iPLEX (Sequenom, San Diego, CA, USA) assays according to the manufacturer's instructions. Quality filters for exclusion of SNPs included call rates below 95% and deviation from HWE (P,0.001). DNA samples were excluded if missing genotypes exceeded 10%. Other quality control measures included parent/offspring trio samples, duplicates on plates, random sample allocation to plates, independent scoring of problematic genotypes by two individuals and re-sequencing of selected DNAs to validate genotypes.

Statistical analysis
Clinical characteristics of cases and controls were compared using the z-test for large independent samples and the x 2 test. Association analyses were performed using PLINK (version 1.07; http://pngu. mgh.harvard.edu/,purcell/plink/). Initially a x 2 test for trend (1 df) was used with stratification by collection center. Logistic regression analysis was performed on each SNP with terms for potential confounders (collection center, gender, duration of T1D and HbA1c) included in the model. A sensitivity analysis to minimize potential misclassification of case/control status was performed by excluding samples from those control individuals with an estimated glomerular filtration (eGFR) ,60 ml/min/1.73 m 2 . The level of statistical significance was set at 5% and adjustment for multiple testing performed by permutation test (n = 100,000). Potential genegene interactions between SNPs were assessed using likelihood ratio x 2 tests in the logistic regression with terms for potential confounders (collection center, gender, duration of T1D and HbA1c) included in the model.

Results
The clinical characteristics of the DN cases (n = 718) and diabetic controls (n = 749) genotyped in this study are listed in Table 1. There were more males, higher mean HbA1c and blood pressure values (despite the use of antihypertensive treatment) in the case group compared with the control group. All comparisons were significant at P,0.001 with the exception of age at diagnosis, LDL cholesterol and body mass index which did not differ significantly between groups. Approximately one quarter of cases (26.9%) had end-stage renal disease (ESRD).
A total of 53 SNPs were genotyped using MassARRAY iPLEX technology in 718 cases and 749 controls ( Table 2). We excluded 73 samples (34 cases and 39 controls) from the analysis with $10% missing genotypes. The average call rate for all SNPs analysed was 99.3%. The genotype distribution for each SNP did not deviate significantly from HWE in either cases or controls. No duplicate or Mendelian inconsistencies were observed.
Single marker testing stratified by collection center identified two non-coding SNPs (rs1532624 in Cholesteryl ester transfer protein (CETP) and rs4420638 in Apolipoprotein C-I APOC1) to be significantly associated with DN ( Table 2). In logistic regression analysis with adjustment by collection center, gender, duration of T1D, and average HbA1c as covariates, the significance of both SNPs was maintained (rs1532624: odds ratio [OR] = 0.82; confidence intervals [CI]: 0.69-0.99; P = 0.034; rs4420638: OR = 1.51; CI: 1.19-1.91; P = 0.001). The sensitivity analysis (that includes samples only from those controls with eGFR .60 ml/min/1.73 m 2 ) identified two SNPs significantly associated with DN in the fully adjusted model (rs4420638; P = 0.016 and rs7679; P = 0.027). However, no associations were maintained following correction for multiple testing. Subgroup analyses showed no association of any SNP with ESRD status.
With no prior hypotheses or supporting evidence of potential gene-gene interaction, we assumed a more stringent level of significance (P,0.01). Interactions were assessed using likelihood ratio x 2 tests in the logistic regression with terms for potential confounders (collection center, gender, duration of T1D and HbA1c) included in the model. Seven interaction terms exceeded the minimum threshold set but following correction for multiple testing and examination of the resultant Q-Q plot, none were identified as being worthy of further investigation (Table 3).

Discussion
Dyslipidemia can result through dietary and lifestyle influences or alternatively as a consequence of variation in genes pivotal to lipoprotein metabolism. In persons with diabetes, prolonged elevation of insulin levels often leads to dyslipidemia, a process central to the pathogenesis of atherosclerosis and increasing CVD risk. As previous studies have reported multiple lipid abnormalities in patients with T1D [15][16][17][18][19][20], we evaluated common polymorphic variation previously associated with dyslipidemia, in persons with T1D, both with and without DN. Univariate analysis identified two SNPs associated with DN (rs1532624 in CETP and rs4420638 in APOC1) both of which remained significant following adjustment for collection center, gender, duration of T1D, and average HbA1c. Interestingly, rs4420638 was also significantly associated with DN in the sensitivity analysis using only those samples from diabetic controls with eGFR .60 ml/min/1.73 m 2 . However, following correction for multiple testing, these associations were no longer significant. Although, published data were available from the US GoKinD genome-wide association study, limited coverage on the Affymetrix 500 K genotyping platform across the genomic locations of both CETP and APOC1, prevented in silico indepen-  dent replication or meta-analysis of our top SNPs or any potential proxies (r 2 .0.8) [41].
In previously published studies the definition of the DN phenotype has proved challenging. We do not think it is surprising that cases in our study had persistent proteinuria (macroalbuminuria) despite the use of antihypertensive medication. The use of angiotensin converting enzyme inhibitors (ACEi) or angiotensin receptor blockers (ARBs), typically reduces but does not abolish protein excretion in persons with overt diabetic nephropathy [42][43][44] suggesting that persistent proteinuria is unlikely to be a consequence of suboptimal blood pressure control. The differences in mean blood pressures observed between case and control groups were consistent with findings in clinical practice.
In addition, it has been suggested that some individuals with a very prolonged duration of type 1 diabetes may develop chronic kidney disease (CKD) without albuminuria. Molitch and colleagues [45] identified 89 of the 1,439 individuals recruited to the DCCT/EDIC study that had developed CKD (defined by estimated GFR ,60 ml/min/1.73 m 2 ) after almost 20 years of follow up. Of the 89 individuals with CKD, 21 were classified as normoalbuminuric (albumin excretion rate [AER] ,30); 14 as microalbuminuric (AER: 30-300); and 54 as macroalbuminuric (AER .300). Of note 43% of the normoalbuminuric individuals with CKD were taking ACEi during the study and 14% were taking ARBs at year 13/14 of the EDIC study [45]. The antihypertensive drugs, ACEi and ARBs, can both lower AER and reduce eGFR which may partly explain why the authors found a small number of individuals with normoalbuminuria and reduced eGFR. The normoalbuminuric patients with reduced eGFR were also 4 years older at time of recruitment than the macroalbuminuric patients (30+/27 yr vs. 26+/2 7 yr [45]). Nevertheless, we did attempt to address this issue of diabetic patients having CKD without albuminuria. In a sensitivity analysis, we excluded all those diabetic patients we had originally recruited as normalbuminuric controls in whom the eGFR was ,60 ml/min/1.73 m 2 . We excluded these controls with reduced renal function from our analysis to limit any risk of misclassification of nephropathy status but found this made little difference to the main analysis (Table 2).
CETP is a protein central to the process of dyslipidemia. It acts as a mediator for the transfer of cholesteryl esters from HDL to VLDL or LDL in exchange for triglycerides, reducing serum HDL concentrations [46]. Variation in CETP levels have been correlated with lipid metabolism and insulin resistance in Type 2 diabetes [47], and also in association with the development of obesity [48] and susceptibility to atherosclerosis and other CVD [49]. Recently, Igl and colleagues demonstrated that the genetic influence mediated by rs1532624 could be attenuated by lifestyle factors such as diet or physical activity, highlighting the potential for interaction at this locus [50]. Our study was unable to examine lifestyle influences, as dietary and physical activity measurements were not collected during recruitment. Nonetheless we sought to evaluate the potential for pair-wise gene-gene interaction between the candidate SNPs examined. Several pair-wise interactions which included the CETP and APOB loci were identified but did not remain significant following correction for multiple testing. As no association survived correction for multiple testing, it is unlikely these gene variants play a specific role in the etiology of DN.
Apo C-I is a protein constituent of chylomicrons, VLDL and HDL and while its precise physiological role is not well established, evidence has demonstrated support for its involvement in HDL metabolism through activation and inhibition of other proteins central to lipid metabolism, including CETP [51]. Association of rs4420638 with DN in T1D in this cohort has been previously reported [52].
Improved therapeutic regimens to lower LDL levels using statins have proved beneficial for patients both with and without diabetes with respect to CVD risk. In addition, increasing evidence suggests statins provide therapeutic benefit independent of cholesterol modulation, by improving endothelial and vascular function and reducing inflammation [53].
Common genetic loci explain only a proportion of the variation observed in lipid levels within the general population. Evidence in support of rare variants with potentially large individual effect size continues to grow, and is likely to make a significant impact on the genetic heritability of this condition [36]. Since our study focused only on common variants, untyped, highly penetrant rare variants in these genes could also contribute to DN. This study has insufficient power to detect rare variants particularly if the effect sizes are small in magnitude, such as the odds ratios of 1.2/1.3 which are more commonly found in common complex diseases (Table 4). Future amalgamation of independent cohorts with similar DN phenotypes will enable a more robust evaluation of such loci. In addition, other factors such as copy number variation or indeed epigenetic mechanisms (e.g. DNA methylation, histone modification and microRNAs) may also attenuate gene function affecting these pathways which modulate disease risk.
Although the SNPs assessed in this study were chosen on the basis of previous associations with dyslipdemia there are a number of inherent limitations associated with using 53 SNPs across 37 genes [54]: (1) identification of association does not necessarily equate to functional significance given the concept of linkage disequilibrium (LD). (2) assessing one or two SNPs per gene may provide inadequate representation of the genetic architecture at that locus. (3) patterns of LD can vary significantly within and between different populations and therefore a significant association in one population may not necessarily translate across all populations.
In conclusion, we found no strong association between common variants in genes involved in dyslipidemia and DN. Further work to investigate lifestyle factors which influence genes may identify potential risk factors for susceptibility to DN.