Comprehensive Whole-Genome and Candidate Gene Analysis for Response to Statin Therapy in the Treating to New Targets (TNT) CohortCLINICAL PERSPECTIVE
Background— Statins are effective at lowering low-density lipoprotein cholesterol and reducing risk of cardiovascular disease, but variability in response is not well understood. To address this, 5745 individuals from the Treating to New Targets (TNT) trial were genotyped in a combination of a whole-genome and candidate gene approach to identify associations with response to atorvastatin treatment.
Methods and Results— A total of 291 988 single-nucleotide polymorphisms (SNPs) from 1984 individuals were analyzed for association with statin response, followed by genotyping top hits in 3761 additional individuals. None was significant at the whole-genome level in either the initial or follow-up test sets for association with low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, or triglyceride response. In addition to the whole-genome platform, 23 candidate genes previously associated with statin response were analyzed in these 5745 individuals. Three SNPs in apoE were most highly associated with low-density lipoprotein cholesterol response, followed by 1 in PCSK9 with a similar effect size. At the candidate gene level, SNPs in HMGCR were also significant though the effect was less than with those in apoE and PCSK9. rs7412/apoE had the most significant association (P=6×10−30), and its high significance in the whole-genome study (P=4×10−9) confirmed the suitability of this population for detecting effects. Age and gender were found to influence low-density lipoprotein cholesterol response to a similar extent as the most pronounced genetic effects.
Conclusions— Among SNPs tested with an allele frequency of at least 5%, only SNPs in apoE are found to influence statin response significantly. Less frequent variants in PCSK9 and smaller effect sizes in SNPs in HMGCR were also revealed.
- hydroxymethylglutaryl coenzyme A reductase inhibitors
- myocardial infarction
Received September 9, 2008; accepted January 26, 2009.
The tremendous popularity of statins for lowering low-density lipoprotein cholesterol (LDL-C) and reducing the risk of cardiovascular disease has led to efforts to determine whether the observed variability in lipid lowering is due to genetic effects or other factors such as environment or compliance. On the basis of mechanism of action and pharmacokinetics, a plethora of candidate genes and dozens of single-nucleotide polymorphisms (SNPs) have been reported to be associated with differing aspects of statin response.1–4 Replication of such results has been sporadic except for SNPs in apoE.5 This lack of replication may be due to many confounding factors, but is likely related to the small population sizes frequently used.6 Compliance is also an issue and poor response may be simply due to nonadherence7 rather than by a biological process. In fact, if the number of noncompliers is larger than the number of true nonresponders, the signal would be difficult to detect at all.
Clinical Perspective see p 173
In our previous examination of atorvastatin response in the Atorvastatin Comparative Cholesterol Efficacy and Safety Study (ACCESS),8 SNPs in only 2 genes, apoE and ABCB1, showed a significant association with response. To extend these data and possibly reveal novel loci, we assessed a much larger cohort from the Treating to New Targets (TNT) trial in which DNA samples were available from 5745 individuals of European ancestry.9 After a 10-mg atorvastatin run-in, individuals were randomized to stay at that dose or titrated to 80 mg and response was assessed in both the run-in at 10 mg and during the titration to 80 mg.
To determine whether additional genes might be involved in statin response, we analyzed 1984 individuals who had been genotyped for ≈300 000 SNPs and looked for genetic associations with lipid changes. SNPs appearing most significant were then replicated in the remainder of the cohort, 3761 individuals. Independently, all individuals were genotyped for 111 SNPs in 23 candidate genes chosen based on earlier literature reports.
The TNT trial, of which details were published previously,9 included 10 001 individuals aged 35 to 75 with clinically evident coronary heart disease. Coronary heart disease was defined as having previous myocardial infarction, previous or current angina with objective evidence of atherosclerotic coronary heart disease, or a history of coronary revascularization. At screening, LDL-C, high-density lipoprotein cholesterol (HDL-C), triglycerides (TG), and total cholesterol were measured in all subjects in a fasting state. In addition, blood pressure and body mass index as well as other standard blood chemistries were measured. All laboratory tests were performed at a central laboratory (Medical Research Laboratories, Highland Heights, Ky) certified by the National Heart, Lung, and Blood Institute/Centers for Disease Control Part III Program. No subfraction analysis was done. These were repeated 4 weeks later, at randomization, 3 and 6 months postrandomization, and annually thereafter. Cardiovascular and health history were taken including smoking, diabetes status, age, and self-declared race. Electrocardiograms were also administered. After approval by the institutional review committee, informed consent for genetic analysis was sought on entry into the trial and 5966 DNA samples were obtained from consenting individuals. A subset was chosen for whole-genome analysis based on the cardiovascular events during the course of the trial and those individuals were matched 3:1 with controls based on age, gender, treatment arm, smoking, diabetes, hypertension, baseline lipid values, baseline glucose levels, and screening LDL-C. Primary events were defined as death from coronary heart disease (n=228), nonfatal myocardial infarction (n=551), resuscitation after cardiac arrest (n=51), and fatal or nonfatal stroke (n=272). Primary analysis was restricted to individuals of European ancestry due to lack of power in other groups. Genotyping for the whole-genome scan was carried out on a set of chips with 322 185 SNPs as described previously.10 The quality control criteria consisted of eliminating SNPs with call rate <80%, and SNPs with either all or no heterozygote calls. We also removed samples with call rates <80%. Although these quality criteria are relatively permissive, we did not see evidence of inflated test statistics for SNPs that would fail more strict quality criteria. For the SNPs that were analyzed, the average call rate was 97.9%, the median was 99.4%, 87.1% had call rates ≥95%, and 94.2% had Hardy-Weinberg P>0.001. Principal components analysis was used to confirm that the sample set did not have substantial population structure. Consistent with the reported ancestries of the sample set, the first principal component seemed to identify European substructure and had strongest loadings on SNPs near the lactase locus. We tested the top 20 components for association with case-control status and with LDL response. None of these tests were significant after correction for the number of comparisons. Candidate gene SNPs were genotyped via SNPlex as described by the manufacturer (Applied Biosystems, Foster City, Calif).
To find genetic associations, we fit linear regression models to the quantitative lipid phenotypes. Genotypes were coded as allele counts in regression models. This corresponds to fitting an additive model where each allele makes the same incremental contribution to phenotype. Models included covariates for environmental factors. For LDL-C response, covariates included age, gender, and screening LDL-C levels. No other covariates were found to be significant. The same covariates were used for the HDL-C and TG response except that the screening levels for those lipids instead of LDL-C. We selected transformations for quantitative variables to remove skew and provide more normally distributed residuals in regressions. The primary test for association consisted of an analysis of variance to assess the significance of the amount of variance explained by adding the genotype term to the model. More detailed descriptions of statistical methods and quality control criteria are described elsewhere.10
The TNT study was a trial in which the effect of low versus high dose statin was examined for its effect on cardiovascular outcomes.9 Individuals were selected for inclusion based on lipid cutoffs with 130 to 250 mg/dL LDL-C permissible at the screening visit (in the absence of statin treatment). After 8 weeks of treatment with 10 mg atorvastatin daily, the study design required that patients must have reached 130 mg/dL LDL-C or less (Figure), and DNA was collected from consenting individuals at that point. This ensured that, at least for the initial 8 weeks, patients were able to comply with therapy. Noncompliance with the therapeutic regimen is a critical issue in such studies, and noncompliers can represent a significant fraction of those who apparently do not respond to treatment. Although this design eliminated a small number of extreme poor responders, it had no effect on the normal or high responders. A wide range of responses existed at any given level of LDL-C (Figure). Of the 15 464 patients who attempted entry into the trial, only 648 patients were eliminated due to LDL-C >130 mg/dL at randomization, an additional 70 patients due to known noncompliance, and 35 patients due to myalgia, summing to fewer than 5% of the total screened for the trial. Not all the 648 patients who were eliminated for LDL-C >130 mg/dL were poor responders, because many had a normal response but started with LDL-C that was too high to be brought into range with low-dose statin. Informed consent was not sought until after individuals were accepted to the trial, so that excluded individuals could not be studied.
Two thousand ninety-two patients were selected for genotyping on the whole-genome platform. Individuals of European ancestry who had a primary event during the trial were matched 1:3 with patients without an event. Matching was based on age, gender, treatment arm, smoking, diabetes, hypertension, baseline glucose level, and screening LDL-C to maximally power studies of cardiovascular events rather than lipid response. Because of the relation of LDL-C with primary events, the case-control selection resulted in fewer high responders picked for the whole-genome genotyping than was present in the total population. The variability/dynamic range for statin response is higher (by Levene test, P<0.0001 on both raw values and covariate-adjusted residuals) among the set of TNT individuals not chosen for whole-genome genotyping, caused by a greater tail of high responders in that group. The lower quartiles of response are nearly identical for the groups (improvements of 52.5 and 52.0 mg/dL, respectively), whereas the upper quartiles of response differed to a larger degree (79.0 and 75.5 mg/dL, respectively). The mean difference is <1% of the overall response but statistically significant (P=0.0028).
A summary of demographic and metabolic phenotypes of all patients is shown in Table 1. Cases (defined as having suffered a primary event) and controls (no events) were genotyped in the whole-genome analysis and the replication cohort (comprising a cohort with no events) who were genotyped for candidate genes and for SNPs identified via the whole-genome scan. For patients remaining at 10 mg after randomization, there was no significant additional change in LDL-C. Individuals switching to 80 mg atorvastatin at 8 weeks lowered their LDL-C by a total of 55.7% and 54.8% at 20 and 60 weeks, respectively. Because only half the individuals were titrated to 80 mg, all such studies have less power than with the full cohort who could be studied at 10 mg.
A total of 322 185 SNPs were attempted on 2092 individuals, but after filtering data using quality criteria described elsewhere,10 data for 1984 individuals and 291 998 SNPs remained. Although this analysis focuses on statin-induced LDL-C change, the other primary analysis examined primary events (no significant, replicated associations found, data not shown). In addition, changes in HDL-C and TG were examined though these secondary analyses were not expected to yield positive results due to the minimal response of HDL-C and the high variability of TG levels. For all models, age and gender were included as covariates and quantile-quantile plots for the standard normal distribution were calculated for a variety of mathematical manipulations for initial levels and changes in lipids. For LDL-C change, the best-behaved linear model was achieved with no transformation (Shapiro-Wilks test of normality P=0.008 compared with P<1.0E-14 using log LDL-C change). For screening LDL-C and for both change and screening levels of HDL-C and TG, a log transformation of the data worked best. The 25 SNPs most highly associated with change in LDL-C are listed in Table 2. Two SNPs are found with P<1E-5, about the number expected based on random variation, and an additional 39 SNPs with P<1E-4, somewhat higher than expected. However, 6 SNPs are closely linked in the region adjacent to GRIK4 on chromosome 11 and another pair is closely linked on chromosome 2. Also, 7 SNPs among the 41 have call rates <95%, an indicator of potential artifacts. A summary of SNPs meeting probability value cutoffs as well as those not meeting call rate or Hardy-Weinberg equilibrium criteria is shown in Supplementary Table I. The top 1000 SNPs for the primary trait under study, LDL-C response, is provided in Supplementary Table II. The secondary traits, HDL-C and TG responses, are provided in Supplementary Tables III and IV and were not expected to yield significant results due to the lack of effect (HDL-C) or high variance (TG).
From the whole-genome scan, the only SNPs with P<1×10−6 for any phenotype were single SNPs associated with screening values of HDL-C and TG. These SNPs, in CETP for HDL-C and GCKR for TG, are in genes previously found via whole-genome scans,10,11 demonstrating the relevance of the TNT population for studying lipid phenotypes. No similarly strong associations were found when examining lipid response values at either 8 or 20 weeks. Thus, among those SNPs we examined, none of the lipid response SNPs is significant after correction for multiple testing at a whole-genome level within this cohort. There can always be some real associations embedded within a large number of strong but potentially false positives, regardless of whether results are whole-genome significant, so the top 50 SNPs for each category of lipid response were tested in 3761 additional individuals to see whether trends evident in the whole-genome scan population were confirmed. These individuals were TNT trial participants who had not been genotyped in the whole-genome scan and, as shown in Table 1, are similar to the whole-genome-genotyped cohort but tended to smoke less, have less diabetes and hypertension, and slightly higher HDL-C and lower TG levels.
For LDL-C response SNPs, 1 assay failed SNPlex design, and 2 others were omitted due to near perfect linkage disequilibrium (LD) with another SNP. No SNP identified in the initial scan replicates to the same degree in the second set. Notably, based on the effect sizes and allele frequencies, each of the top 25 SNPs had at least a 99% power to achieve a P<0.05 level of significance in the complementary TNT sample set if the initial observations were true positives with the same effect size. Only 3 SNPs achieved P<0.05, 2 of which have false discovery rates (FDRs) in excess of 25%. These SNPs fare even worse if Bonferroni corrected. The only SNP that seemed to even marginally retain significance was rs6790122, which achieves a marginal FDR of 0.057 among 22 tests. A further replication was attempted in an independently ascertained set of 1248 individuals from the ACCESS trial.8 The selection criteria for this trial were somewhat different than for TNT, but the LDL-C lowering achieved in the 2 populations was similar and the same population was used in another statin response analysis8 and yielded a significant association with LDL-C response. The power to replicate these findings in ACCESS at P<0.05 ranged from 85% to 95%, yet not a single SNP achieved this level of significance (Table 2). The only marginally promising SNP from TNT, rs6790122, had 90% power to replicate in ACCESS, but failed to do so. Its genotypic effect estimate, while in the same direction, was <30% of that observed in TNT, suggesting a real but weak effect.
Previous studies have identified numerous candidate genes associated with statin response1–4 but very few have replicated. However, some SNPs within apoE have replicated consistently and were expected to be detected. In our whole-genome scan, no SNPs in or near the apoE region were significant even at the P<0.05 level. The most strongly associated SNP in our previous study, rs7412, was not on this platform. To determine whether the absence of an association with SNPs in apoE was due to a lack of coverage in the region or to something unique to this population, 29 SNPs within 50 kb of apoE were genotyped in the cohort. In addition, we had previously tested 43 SNPs in 16 genes8 that had been reported in the literature as associated with lipid response so those SNPs with minor allele frequency >5% were retested in TNT. Earlier testing identified SNPs in additional genes (ADAMTS1, FCAR, NPC1L1, PCSK9, PON1, PPARG, and SCAP1)12–18 as potentially associated with statin response and these were also included as candidates.
The candidate SNP analysis was split into 2 stages. In the first stage, only data from the individuals who were genotyped in the whole-genome scan were analyzed so the findings would be directly comparable with the genome-wide analysis. In the second stage, both the individuals examined in the genome-wide analysis as well as all others in the TNT cohort were also analyzed. Because most candidate SNPs were not associated with statin response in the ACCESS study performed earlier, they were not expected to be associated in TNT either. SNPs that achieved an association with LDL-C response with a significance level of P<0.05, uncorrected for multiple testing, in the complete population are included in Table 3. No candidate SNPs were found to be significantly associated with HDL-C or TG response. All 111 candidate SNPs that were genotyped in the whole cohort in Hardy-Weinberg equilibrium are listed in Supplementary Table V. Thirty-eight of the 111 SNPs were also present on the whole-genome scan platform.
In the analysis of the whole-genome subset of individuals, only 4 SNPs met significance criteria of P<0.05 including 3 in apoE. If all these SNPs had been on the whole-genome platform, only rs7412 would have been chosen for follow-up. Indeed, rs7412 would have been the most strongly associated SNP and even passed a Bonferroni multiple testing correction. For rs7412, the nonwhole-genome individuals independently yield a strong association and, when combined for the complete cohort, rs7412 was strongly associated with LDL-C response (P=6E-30). Thus, the top candidate gene SNP identified in TNT is also significant in the whole-genome scan subset, demonstrating the power of that subset to detect signals of that magnitude.
To determine whether SNPs in the apoE region could be genotyped as surrogates for rs7412, more SNPs were genotyped across the apoE region in addition to those examined in the whole-genome scan (Table 4). rs7412 was found at a frequency of only 5.6% and was not in LD with any of the whole-genome SNPs. Indeed, the only SNP tested with r2>0.2 with rs7412 was SNP17 (no dbSNP identifier, also known as ENSSNP5761165),19 with r2=0.5 (Supplementary Figure I). The entire region is noteworthy for its very low level of LD and short LD blocks. Thus, this particular region requires extensive genotyping to capture all potential functional variation. The Perlegen 322K used in this study, the Affymetrix 500K, and the Illumina 317K/550K platforms all omit rs7412 and would not detect a weak phenotypic effect. More robust phenotypes could potentially be detected on all platforms via other SNPs in weak LD with rs7412.
All 17 candidate SNPs with P<0.05 (uncorrected) in the complete cohort are shown in Table 3. Sixteen of the 17 SNPs had the same direction of effect in both subsets. Eleven of those 16 had >30% reduction in effect size for the whole-genome scan subset compared with the remainder of the cohort, consistent with the expectations of reduced effect size in that group. Other than the apoE SNPs, none would have passed whole-genome criteria, but some are significant if examined in the less stringent criteria of candidate genes. The most significant non-apoE SNP, rs11591147, is found in the proprotein convertase subtilisin/kexin 9 gene (PCSK9). Its genotype effect is nearly identical to the best apoE SNP, but its association is less significant due to its low frequency. Several other genes yield 1 or more SNPs with FDRs of <0.1 (HMGCR, ACE, ABCG8, ABCB1, LDLR, and LIPC), but the magnitude of effect is much lower than with apoE or PCSK9 and none would have been selected for replication from a whole-genome scan, regardless of which individuals were used.
The most highly associated candidate SNPs other than those within apoE and PCSK9 include 3 within the HMGCR gene. The first 2, rs10474433 and rs17671591, are in high LD with each other (r2=0.99) and the third, rs6453131, is also linked (r2=0.68). In addition to the 3 HMGCR SNPs listed in Table 3 and 3 HMGCR SNPs in Supplementary Table V that were tested in the full set of individuals, we genotyped 12 more HMGCR SNPs in the nongenome-wide analysis subset that included black individuals. rs17238540 (also reported as SNP29) has previously been found associated with statin response in mixed cohorts or blacks.19,20 Except for the SNPs listed in Table 3, we found no association with any other HMGCR SNP in either individuals of European or black ancestry though the population size of the latter group (n=154) was not sufficient for good power. rs3846662 has also been reported to be associated with alternative splicing of HMGCR,21 but we found no association with statin response.
Despite our extensive efforts to identify novel genetic markers associated with statin response, none was found. The fact that rs7412 in apoE was highly significant (P=6×10−9) in just the whole-genome subset shows that effects of that size would be easily detectable if associated with a common variant in this population. The lack of a stronger effect size among either the whole-genome or candidate gene SNPs indicates that no other common variants should be expected to play a major role in variation in response to atorvastatin. This population (n=1984) is sufficiently large to identify lipid effects such as baseline HDL-C with CETP (rs247616, P=2×10−11, FDR=5×10−6) and baseline TG with GCKR (rs780094, P=1×10−7, FDR=0.039) but not with more complex phenotypes such as cardiovascular events. Even though this whole-genome study was designed to be maximally powered for cardiovascular events, no SNP met whole-genome significance levels (all FDRs >0.9) with that phenotype. The CETP SNP found to be associated with HDL-C in this study, rs247616, is in the same LD bin as rs183180, previously shown to be most highly associated SNP within the CETP gene.22 The GCKR SNP found to be associated with TG, rs780094, is exactly the same SNP found to be most highly associated with TG in another whole-genome scan.11
The strongest genetic effects were observed with apoE and PCSK9, genes previously known to affect statin response. Indeed, in previous studies, only apoE has been identified as associated with LDL-C response across multiple studies in a reproducible manner.5,8 Some studies, generally with very small populations, do not show an association with apoE. The more recently identified PCSK9 is also accumulating reproducible support for an impact on statin response.15,23–25 apoE is found on chylomicrons and VLDL lipoprotein particles in the circulation and can affect their uptake in the liver. Thus, variants of apoE can affect the relative proportion of cholesterol taken up via the diet versus synthesized in the liver. Individuals synthesizing a higher proportion of cholesterol are more susceptible to its inhibition. Similarly, PCSK9 has been found to bind LDLR and affect its degradation, thus impacting cholesterol uptake in the liver and hence its synthesis.
SNPs within HMGCR, the target protein of statins, have also been sometimes found associated with LDL-C response, and we found 3 linked SNPs that achieve significance if examined as candidate SNPs. Two tightly linked SNPs, rs10474433 and rs17671591, are located ≈30 kb upstream of HMGCR and the third, rs6453131, is located in intron 6. None has been previously reported to be associated with statin response. The effect size of these SNPs is much smaller than that found with apoE and PCSK9 SNPs and, if the association is real, the functional SNP is not known. SNP12 (rs17244841) in intron 5 and SNP29 (rs17238540) in intron 18 are 2 highly linked SNPs that have been reported to be associated with LDL-C response in a mixed cohort with pravastatin19 and a black cohort with simvastatin20 but not associated in a European cohort with atorvastatin.8 The lack of association with atorvastatin was replicated in the present cohort though an effect specific to other statins or with specific cohorts cannot be ruled out. The combination of low frequency of SNPs and small effects makes the study of epistatic interactions unworkable. It would not be unreasonable to expect complex interactions among genes but it would likely require more extremes of response.
Although statins have variable effects among individuals (Figure), most respond if compliant with therapy. Of the >15 000 individuals who completed the 8-week run-in period in this TNT trial, a total of <5% were eliminated due to insufficient LDL-C lowering, noncompliance, or myalgia. Similarly, in the ACCESS trial in which exclusion criteria were not based on response to treatment,26 only 65 of 1427 individuals on 10 mg atorvastatin had <20% LDL-C lowering at 6 weeks. With the exception of the <1% of the population homozygous for the rare apoE or PCSK9 SNPs, a combination of age and gender is more effective at predicting statin response than any of the individual genetic factors (Table 5). These effects could be related to an underlying mechanism or simply due to variable compliance among groups. Similar results were found in the ACCESS statin trial, but other biological and genetic factors might still cause some variation.
Because few genes have been reliably implicated in statin response, we chose a whole-genome approach so that the entire breadth of potential mechanisms could be assessed. Although inclusive for most genes, uncommon variants or genes with no functional variation are undetectable in such a study. Thus, the low minor allele frequency of rs7412 in apoE, coupled with its low LD with genotyped SNPs, caused it to be missed in our whole-genome scan. Similarly, none of the PCSK9 SNPs were detected because of their low frequency. There may be other genes with rare variants or weak effects that remain undetected, but there are unlikely to be any genes with a large effect on a significant fraction of the population. Resequencing extreme responders may provide insight into rare, functional variants.
The lack of strong genetic effects across a large population are likely a reflection of the complexity of lipid homeostasis and, except for individuals with rare, high-effect polymorphisms, variability in response is due to a wide range of small effects superimposed on compliance with recommended dosing. It remains to be determined whether other statins will behave similarly because sufficiently large populations have not been reported yet. With respect to predicting efficacy, genetic markers are not presently useful in setting atorvastatin dose and the driving force should continue to be treatment with an appropriate dose titration to attain the guideline-driven LDL-C goal. Moreover, these data emphasize that even comprehensive genetic analyses might miss important associations when minor allele frequencies are low and phenotypic consequences are modest and further illustrate the limits of the predictive power of genetic analyses for similar questions, especially when restricted to common variants.
Sources of Funding
Pfizer provided funding for the trial and subsequent analyses.
Drs Thompson, Hyde, Wood, and Cox and S.A. Paciga are present or former employees of Pfizer. Drs Hinds and Cox are present or former employees of Perlegen. Dr Kastelein has received grant support, lecture fees, and consulting fees from Pfizer. Dr Hovingh reports no disclosures.
Shitara Y, Sugiyama Y. Pharmacokinetic and pharmacodynamic alterations of 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors: drug–drug interactions and interindividual differences in transporter and metabolic enzyme functions. Pharmacol Ther. 2006; 112: 71–105.
LaRosa JC, Grundy SM, Waters DD, Shear C, Barter P, Fruchart JC, Gotto AM, Greten H, Kastelein JJ, Shepherd J, Wenger NK; Treating to New Targets (TNT) Investigators. Intensive lipid lowering with atorvastatin in patients with stable coronary disease. N Engl J Med. 2005; 352: 1425–1435.
Diabetes Genetics Initiative of Broad Institute of Harvard and MIT, Lund University, and Novartis Institutes of BioMedical Research, Saxena R, Voight BF, Lyssenko V, Burtt NP, de Bakker PI, Chen H, Roix JJ, Kathiresan S, Hirschhorn JN, Daly MJ, Hughes TE, Groop L, Altshuler D, Almgren P, Florez JC, Meyer J, Ardlie K, Bengtsson Boström K, Isomaa B, Lettre G, Lindblad U, Lyon HN, Melander O, Newton-Cheh C, Nilsson P, Orho-Melander M, Råstam L, Speliotes EK, Taskinen MR, Tuomi T, Guiducci C, Berglund A, Carlson J, Gianniny L, Hackett R, Hall L, Holmkvist J, Laurila E, Sjögren M, Sterner M, Surti A, Svensson M, Svensson M, Tewhey R, Blumenstiel B, Parkin M, Defelice M, Barry R, Brodeur W, Camarata J, Chia N, Fava M, Gibbons J, Handsaker B, Healy C, Nguyen K, Gates C, Sougnez C, Gage D, Nizzari M, Gabriel SB, Chirn GW, Ma Q, Parikh H, Richardson D, Ricke D, Purcell S. Genome-wide association analysis identifies loci for type 2 diabetes and triglyceride levels. Science. 2007; 316: 1331–1336.
Sabatine MS, Ploughman L, Simonsen KL, Iakoubova OA, Kirchgessner TG, Ranade K, Tsuchihashi Z, Zerba KE, Long DU, Tong CH, Packard CJ, Pfeffer MA, Devlin JJ, Shepherd J, Campos H, Sacks FM, Braunwald E. Association between ADAMTS1 matrix metalloproteinase gene variation, coronary heart disease, and benefit of statin therapy. Arterioscler Thromb Vasc Biol. 2008; 28: 562–567.
Iakoubova OA, Tong CH, Chokkalingam AP, Rowland CM, Kirchgessner TG, Louie JZ, Ploughman LM, Sabatine MS, Campos H, Catanese JJ, Leong DU, Young BA, Lew D, Tsuchihashi Z, Luke MM, Packard CJ, Zerba KE, Shaw PM, Shepherd J, Devlin JJ, Sacks FM. Asp92Asn polymorphism in the myeloid IgA Fc receptor is associated with myocardial infarction in two disparate populations: CARE and WOSCOPS. Arterioscler Thromb Vasc Biol. 2006; 26: 2763–2768.
Naoumova RP, Tosi I, Patel D, Neuwirth C, Horswell SD, Marais AD, van Heyningen C, Soutar AK. Severe hypercholesterolemia in four British families with the D374Y mutation in the PCSK9 gene: long-term follow-up and treatment Response. Arterioscler Thromb Vasc Biol. 2005; 25: 2654–2660.
Chen S, Tsybouleva N, Ballantyne CM, Gotto AM Jr, Marian AJ. Effects of PPARalpha, gamma and delta haplotypes on plasma levels of lipids, severity and progression of coronary atherosclerosis and response to statin therapy in the lipoprotein coronary atherosclerosis study. Pharmacogenetics. 2004; 14: 61–71.
Krauss RM, Mangravite LM, Smith JD, Medina MW, Wang D, Guo X, Rieder MJ, Simon JA, Hulley SB, Waters D, Saad M, Williams PT, Taylor KD, Yang H, Nickerson DA, Rotter JI. Variation in the 3-hydroxyl-3-methylglutaryl coenzyme a reductase gene is associated with racial differences in low-density lipoprotein cholesterol response to simvastatin treatment. Circulation. 2008; 117: 1537–1544.
Medina MW, Gao F, Ruan W, Rotter JI, Krauss RM. Alternative splicing of 3-hydroxy-3-methylglutaryl coenzyme A reductase is associated with plasma low-density lipoprotein cholesterol response to simvastatin. Circulation. 2008; 118: 355–362.
Thompson JF, Wood LS, Pickering EH, Dechairo B, Hyde CL. High-density genotyping and functional SNP localization in the CETP gene. J Lipid Res. 2007; 48: 434–443.
Dubuc G, Chamberland A, Wassef H, Davignon J, Seidah NG, Bernier L, Prat A. Statins upregulate PCSK9, the gene encoding the proprotein convertase neural apoptosis-regulated convertase-1 implicated in familial Hypercholesterolemia. Arterioscler Thromb Vasc Biol. 2004; 24: 1454–1459.
Rashid S, Curtis DE, Garuti R, Anderson NN, Bashmakov Y, Ho YK, Hammer RE, Moon YA, Horton JD. Decreased plasma cholesterol and hypersensitivity to statins in mice lacking Pcsk9. Proc Natl Acad Sci USA. 2005; 102: 5374–5379.
Berge KE, Ose L, Leren TP. Missense mutations in the PCSK9 gene are associated with hypocholesterolemia and possibly increased response to statin therapy. Arteriol Thromb Vasc Biol. 2006; 26: 1094–1100.
Ballantyne CM, Andrews TC, Hsia JA, Kramer JH, Shear C. Correlation of non-high-density lipoprotein cholesterol with apolipoprotein B: effect of 5 hydroxymethylglutaryl coenzyme A reductase inhibitors on non-high-density lipoprotein cholesterol levels. Am J Cardiol. 2001; 88: 265–269.
Statins are effective at lowering low-density lipoprotein cholesterol and reducing risk of cardiovascular disease, but variability in individual response to statins is not well understood. To address this, 5745 individuals from the TNT trial were genotyped in a combination of a whole-genome and candidate gene approach to identify associations with response to atorvastatin treatment. Only 1 SNP, rs7412, in the apoE gene was significantly associated with response to atorvastatin after correction for multiple testing. Age and gender were found to influence low-density lipoprotein cholesterol response to a similar extent as the most pronounced genetic effects. The lack of strong genetic effects across a large sample are likely a reflection of the complexity of lipid homeostasis and, except for individuals with rare, high-effect polymorphisms, variability in response to statins is due to a wide range of small effects superimposed on compliance with recommended dosing. It remains to be determined whether these findings apply to other statins, because studies of sufficiently large populations using other agents have not been reported yet. With respect to predicting efficacy of statins, genetic markers are not presently useful in setting atorvastatin dose, and the driving force should continue to be treatment with an appropriate dose titration to attain the guideline-driven low-density lipoprotein cholesterol goal. Moreover, these data emphasize that even comprehensive genetic analyses might miss important associations when minor allele frequencies are low and phenotypic consequences are modest and further illustrate the limits of the predictive power of genetic analyses for similar questions, especially when restricted to common variants.
The online-only Data Supplement is available at http://circgenetics.ahajournals.org/cgi/content/full/CIRCGENETICS.108.818062/DC1.