Final Report Summary - PREDICTION-ADR (Personalisation of tREatment In Cardiovascular disease through next generation sequencing in Adverse Drug Reactions)
Executive Summary:
The primary aim of PREDICTION-ADR was to develop genetic risk assessment and diagnostic tools for the prediction of adverse drug reactions in two commonly prescribed drug classes that are used to manage cardiovascular risk, namely statins and Angiotensin Converting Enzyme inhibitors (ACEI)/Angiotensin Receptor Blockers (ARB). Drug intolerance is a major factor in lack of adherence and efficacy in both classes of drugs. The ultimate goal is to optimise the management of cardiovascular disease (CVD) with currently used medications. Particularly, we aimed to develop tools that will identify (i) patients with likelihood of experiencing severe adverse reactions to drugs, and (ii) patients who may discontinue drugs due to milder intolerance. The application of such drug response prediction tools is expected to directly improve patient care through individualized management of both cardiovascular risk and disease in addition to reducing health care cost. This work utilised the integration of expertise from leading European groups in the fields of cardiovascular pharmacology, pharmacoepidemiology and molecular genetics, with clinical medicine.
The development of genetic drug response prediction tools involved a clearly defined process of (a) discovery of novel genetic biomarkers by next generation sequencing, (b) using sophisticated statistical analysis to define massively multi-allelic genetic interaction models to predict adverse drug reactions with a high specificity and sensitivity, (c) validation of these markers in population-based data samples and clinical trials, (d) development of a novel diagnostic incorporating the validated drug response prediction tools, and finally, (e) dissemination through engagement with stakeholders, patient groups and exploitation through commercialization by ASPER-Biotech a leading EU genetic diagnostic provider.
Exome sequencing was successfully performed in a large case control study for both ACEI angioedema and statin induced myopathy and the data was analysed by TAXONOMY 3 interaction software and produced an AUC of 85% and 84% for myopathy and angioedema respectively. Rare and common variants were found for both pharmacokinetic and immune genetic variants including a novel common protective variant in SLC01B1 and implicated a novel immune-modulatory gene family, the Leukocyte Immunoglobulin Like Receptors in statin induced and statin independent muscle pain. Results for angioedema highlighted a role for potassium channels in side effects to ACE inhibitors that also include the more common cough reaction. Effects in the HLA-DRB1 gene were seen, in agreement with previous studies on aspirin induced angioedema. A cost effectiveness analysis was performed for genetic testing in new ACEI starts. This came to the conclusion that point of care testing for single drug starts was likely to be not cost effective with tests costing more than 2 EURO even if the test provided an AUC of over 90%. This however does not consider the cost of more common intolerance, and it is possible that pre-emptive whole genome testing at first ever drug start would provide a life-time long term cost effectiveness. The PREDICTION-ADR study will provide genetic algorithms to prevent harm from statins and ACE inhibitors in future implementation of pre-emptive genotyping programmes such as those being piloted by Ubiquitous Pharmacogenetics (http://upgx.eu).
Project Context and Objectives:
The primary aim of PREDICTION-ADR was to develop genetic risk assessment and diagnostic tools for the prediction of adverse drug reactions in two commonly prescribed drug classes that are used to manage cardiovascular risk, namely statins and Angiotensin Converting Enzyme inhibitors (ACEI)/Angiotensin Receptor Blockers (ARB). Drug intolerance is a major factor in lack of adherence and efficacy in both classes of drugs. The ultimate goal is to optimise the management of cardiovascular disease (CVD) with currently used medications. Particularly, we aimed to develop tools that will identify (i) patients with likelihood of experiencing severe adverse reactions to drugs, and (ii) patients who may discontinue drugs due to milder intolerance. The application of such drug response prediction tools is expected to directly improve patient care through individualized management of both cardiovascular risk and disease in addition to reducing health care cost. This work utilised the integration of expertise from leading European groups in the fields of cardiovascular pharmacology, pharmacoepidemiology and molecular genetics, with clinical medicine.
The development of genetic drug response prediction tools involved a clearly defined process of (a) discovery of novel genetic biomarkers by next generation sequencing, (b) using sophisticated statistical analysis to define massively multi-allelic genetic interaction models to predict adverse drug reactions with a high specificity and sensitivity, (c) validation of these markers in population-based data samples and clinical trials, (d) development of a novel diagnostic incorporating the validated drug response prediction tools, and finally, (e) dissemination through engagement with stakeholders, patient groups and exploitation through commercialization by ASPER-Biotech a leading EU genetic diagnostic provider.
The principle basis for the concept of PREDICTION-ADR is clinical need. With the ageing population in the developed and developing world, drug therapy to manage CVD is steadily increasing. Although the incidence of the most severe adverse drug reactions (ADRs) is rare, given the number of prescriptions of CVD drugs globally, the absolute number of ADRs is substantial. Studies have shown that ADRs are one of the most common reasons for hospitalisation in the adult population. It has been proposed that ADRs are the fourth to sixth leading cause of death in hospitalised patients. ADRs also commonly lead to poor compliance and discontinuation of vital therapies and therefore negatively affect the burden of CVD in health care systems worldwide. The current knowledge about possible genetic causes of ADRs is limited. Studying the genetic basis of susceptibility to ADRs may provide possibilities for identification of susceptible individuals through pharmacogenetic testing.
Understanding the molecular basis of ADRs may also make it possible to design safer drugs. Statins primarily reduce the risk of coronary artery disease (CAD) by lowering blood cholesterol through inhibition of the HMG-CoA reductase enzyme. Large clinical trials show a 27% average relative risk reduction of major coronary events. However there is large variability in benefits from statin therapy. This variability can be partially explained by factors such as adherence, gender, age, diet, concomitant drug use and environmental factors, with drug intolerance being a major factor in adherence and concomitant drug use also being a major factor in determining intolerance. Therapy with statins has become relatively cheap because most statins are out of patent which means that even more patients will be treated with them. Next to that, the dosages used in the treatment of patients are increasing, because it has been shown that higher doses are more effective. With more patients being treated with increasing doses the burden of adverse events becomes ever greater.
Project Results:
The PREDICTON-ADR consortium brought together a wide range of investigators from Sweden, Holland, Estonia and the UK to investigate the genetic basis of statin induced muscle damage and angioedema resulting from exposure to ACE-inhibitor drugs. Previous studies had been hampered by lack of consistent phenotype definition, poor sample sizes, and the investigation of limited candidate genes.
The first task of the consortium was to provide a robust analysis of the appropriate phenotypic descriptions for both statin myopathy and ACEI induced angioedema. Workshops held in Liverpool in December 2013 resulted in the publication of authoritative manuscripts in the Journal Clinical Pharmacology and Therapeutics, the leading journal for pharmacogenetics. This informed the study design for our exome sequencing project and cases were ascertained from Sweden, UK and Holland. All cases and controls were subjected to centralised adjudication by the responsible partner (Uppsala for ACEI angioedema and Liverpool for statin induced myopathy). Exome sequencing was performed at three centres using a harmonised protocol and identical capture reagent (Agilent human exome V4).
The comparability and quality of the sequencing at all three sites was confirmed by the sequencing of 8 common samples obtained from Dundee. The sequencing of the cases and controls was completed and gene-based test analysis was performed. This used both the absolute number of variants observed in each gene in each case and control (T1 Burden) and the dispersion of the observed effect sizes across each gene (SKAT-O). For statin myopathy this revealed interesting associations with appropriate loci including the statin transporter SLCO1B1 and the HLA region. In particular a strong association was observed with a novel gain of function variant. Exome sequencing on a discovery cohort of 235 statin cases, 251 ACE cases and 502 controls from 3 different countries (UK, Sweden and Netherland) was performed and analysed through 2 different pipelines: -BWA-MEM/GATK haplotype-caller (Pipeline 1) -BWA-Stampy/Platypus (Pipeleine2).
Quality control excluded variants with poor genotyping as evidenced by genotype call rate < 95%, deviation from the Hardy-Weinberg Equilibrium with p<1x10-6 or failure of other criteria resulting in 550,057 and 558,909 variants from (2) passing all quality criteria, respectively for Statin and ACEI, 527,795 and 509,173 variants from (1) respectively for Statin and ACEI.
Analyses were performed as simple case controls Burden test and SKAT taking into account the sequencing centre as a covariable using RareMetal (Feng et al.). Analyses were restricted to the 38,642 genes fulfilling all quality control filters and to variants or resulting in a stop missense/splicing/gain/loss. Variants were also only considered if they exhibited minor allele frequencies less than 0.01%.
QQ plots of both ACEI and Statin data showed an early separation of the observed from the expected with a depletion, which means that p values were less significant than expected.
Top hits from both analyses (Statin and ACEI) were different between the two pipelines. However the C9 gene encoding Complement 9, ranked in the top ten hits for the 2 pipelines for statin myopathy with 12 common variants between the 2 pipelines, with Burden test p values of 3.17x10-4 (Pipeline 1) and 5.21x10-4 (Pipeline 2), SKAT p values of 3.96x10-3 (Pipeline 1) and 3.79x10-4 (Pipeline 2) respectively.
Besides C9, our strategy revealed other candidate genes for Statin like FAM110A involved in immunology responses that showed SKAT p values of 4.16x10-3 for (2) and 6.68x10-3 for (1) with 7 common variants between the 2 pipelines (2 singulars in (1)), or SLC27A6 and HLA-A (Table 4).
Another important finding was the presence of a common SLCO1B1 “gain of function” variant present at a MAF of 10%. This variant, Pro155Thr was strongly protective against statin myopathy and has been previously demonstrated to protect against Methotrexate toxicity. In addition the Burden test using rare variants in SLC01B1 showed nominal significance (p=0.0224) and indicated a potential action on statin myopathy of at least 5 rare variants (Ala13Glu,Ser51Phe,Ile106Thr,Val235Met,Il499Val) all present at frequencies below 0.005.
A strong determinant of ACEI induced angioedema was seen in the HLA-DRB1 gene with both a common variant Val73Met (MAF 10%) providing exome-wide significance (6X10-7) and with suggestive significance in the rare variant burden test (Gln260Pro and His141Tyr, both MAF). This is consistent with previous reports of an association of HLA-DRB1 with aspirin induced angioedema. Candidate genes for ACEI induced angioedema were identified like SLC22A6 a member of a family of organic anion transporters that mediates the secretion of endogenous and exogenous metabolites including drugs, PPP1R12B a protein phosphatase or CLCN2 involved in chlorine ion transport.
A strong signal in an intron of the “Maxi” potassium channel KCNMA1 was obtained from genome-wide association study (GWAS) analysis of the Uppsala angioedema cohort. This was then replicated in the Danish/Swedish angioedema replication cohort, the Marshfield angioedema GWAS data, the GoDARTS/Rotterdam ACEi discontinuation cohort and two cohorts of ACEI cough. None of the cohorts replicated the strong effect size observed in the discovery GWAS (OR= 2.4 p=4.2X10-8). However, the signal is directionally consistent in all studies with a combined effect estimate of OR=1.44 p=2X10-6). The test for heterogeneity suggested that the initial finding had a strong overestimate, and there was a hint that the signal was weakest for the cough phenotype.
The significant finding in the intolerance phenotype would suggest that other forms of intolerance including angioedema may be influenced by this variant. Exome sequencing revealed that rare variants in KCNMA1 and its heterodimeric partner KCNMB1 are also associated with angioedema. This is relevant to the finding that the associated regulatory protein KCNIP4 was our main finding from the GWAS of ACEI associated cough, underscoring the role of these channels in side effects to ACEI. This work is now being followed up by the Dundee group as part of the IMI project BEAT-DKD, where we are investigating genes and potassium levels in the use of ACEI drugs to protect from chronic kidney disease. This will incorporate drug intolerance features including intolerance to determine individuals that will benefit the most, and suffer the least harm from ACEI drugs.
In WP4 we examined the potential role of gene-gene interactions using the Taxonomy 3 software of the SME partner Adorial. This software had been developed for GWAS data and this was the first use with exome sequence data. Significant effort was assigned to ensure that the Taxonomy3 software would be able to accept exome sequence data with no opportunity for data corruption. This entailed frequent discussions with members of WP3. It is also important that clinical variables are included in the Taxonomy3 analysis, and much interaction with members of WP1 and WP2 was necessary to understand the nature of the variables and develop a suitable software environment to receive these data.
Sequence data generated in WP3 was assessed using our standard two-stage approach. Univariate QC was used to exclude subjects and markers with poor quality data. This posed a significant challenge. Our method is very powerful, but has stringent requirements for the data that can be analysed. Our usual data quality cut-off is 2% of missing data for both subjects and markers, to avoid potential biases in the data due to, for example, poor quality DNA data. Applying this criterion only excluded a few of the subjects but the majority of markers failed this criterion. Rather than risk introducing biases into the data analysis, we retained the QC criterion, and proceeded with analysis of the reduced marker set.
The second stage of our data QC involved co-analysis of the subjects’ data with HapMap data form subjects of known ethnicity to identify an ethnically homogeneous group of cases and controls, minimising patient stratification bias. Although only a small number of markers were common to the dataset from WP3 and the HapMap data, a surprisingly good plot allowed homogeneous case/control groups to be determined.
Taxonomy3 Analysis:
Two types of analysis were carried out on both datasets:
1. ‘Main effect’ analysis, looking at the contribution of markers to case/control separation
2. Interaction analysis, where the contribution of interacting markers is assessed
The analyses were carried out using in-house proprietary software utilising cloud computing capacity where necessary to speed the analysis times.
Predictive Power:
The markers identified in the main effect and interaction analysis were used to determine predictive power using the Leave-One-Our-Crossover Validation method (LOOCV). In this method, each subject in turn is omitted from the analysis, and the analysis parameters used to predict whether the subject is a case or a control.
The results from both adverse event studies were very similar. The output from the Taxonomy3 main effect showed only limited predictive power (AUC=0.59 for myopathy and 0.6 for angioedema), but the predictive power was substantially increased by including the SNP-SNP interactions in the analysis (AUC=0.85 for myopathy and 0.84 for angioedema).
Conclusions:
The fact that most variables did not meet our missing data criterion had a significant impact on our ability to generate genetic associations. However despite this limitation, the predictive power of the identified markers was surprisingly good, although not sufficient to provide useful information for clinical providers. The output of WP6 suggested that higher sensitivities and specificities would be required to have a cost-effective screen for these adverse events.
The predictive power was significantly improved when SNP-SNP interactions are included in the calculation. In our experience, this interaction analysis – a unique feature of Taxonomy3 – typically generates more novel genetic insights, providing greater pathology understanding. This will translate into greater predictive power for adverse events such as myopathy and angioedema. Future work should investigate the use of whole genome sequence or imputed microarray data to improve the marker coverage across the genome.
Work package 5 for population validation looked at common variation in large cohorts and examined the role of genetic variation in determining more common forms of intolerance as marked by features such as mild muscle pain or cough. GWAS data was available from the GoDARTS cohort as well as the Rotterdam study where statin intolerance was defined by discontinuation and drug switching algorithms as well as by mild alteration in CK results. Collaborations were initiated with US EMERGE groups with GWAS data based on EMR data in a similar fashion to the GoDARTS and Rotterdam. In particular the Vanderbilt group defined a GWAS for ACEI cough and the Marshfield group, where we worked together to establish variants in KCNIP as determinants of ACE-I intolerance including cough, but not angioedema.
In addition the JUPITER trial of atorvastatin contributed GWAS data with a phenotype of myalgia. During the early part of the PREDICTION-ADR study a GWAS by Dubé et al. 11 reported a missense variant, Asp247Gly, in the leukocyte immunoglobulin-like receptor subfamily B member 5 gene LILRB5 on chromosome 19 that was associated with circulating serum CK levels. The mean CK levels of Asp247 homozygotes (T/T) were significantly higher. This association was found to be independent of statin use, however there is no known biological mechanism for the variant in determining CK levels. A GWAS by Kristjansson et al. of over 60,000 Icelanders replicated the association of the variant and CK levels 12. The same study also reported the association of the variant with serum LDH levels in a population of over 90,000 Icelanders 12. The LILRB5 variant showed the same direction of effect, i.e. Asp247 homozygotes had higher LDH and CK levels. LDH is often used in conjunction with CK as a marker of tissue damage. The findings suggested the variant might impart a statin independent susceptibility to muscle-based events. This makes LILRB5 a potential marker for susceptibility to the commonly noted muscle-based symptoms attributed to statin intolerance.
These discoveries warranted an investigation into the role of the LILRB5 variant in statin intolerance. Population-based studies use surrogate markers of intolerance, such as elevations in CK, trends in statin treatment, dose reductions, switching or the discontinuation of therapy. Therefore, a priori, we considered two definitions of statin intolerance, one dependent on and one independent of elevated CK levels. We hypothesize that carriers of variants associated with higher muscle enzyme levels (CK and LDH) will also be predisposed to forms of statin intolerance independent of CK levels.
The principal cohort used was the Genetics of Diabetes Audit and Research, Tayside Scotland (GoDARTS). GoDARTS has been previously used to establish pharmacogenetics (PGx) associations of genes such as the hepatic influx transporter SLCO1B1 and statin intolerance 13. GoDARTS contained 11,912 statin users and provides approximately 98,000 person-years of statin exposure, providing an ideal cohort to examine the association of this genetic variant with statin intolerance. Replication was examined in the Clinical Practice Research Datalink (CPRD) STAGE study 14 and clinically adjudicated cases of statin-induced myopathy (SIM) in the European PREDICTION-ADR consortium study. The interaction of this effect with statin use was then studied among participants who developed myalgia in the JUPITER (Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin) randomized clinical trial (RCT) where individuals were allocated rosuvastatin or placebo to assess the relative reduction in vascular events.
Since the LILRB5 Asp247Gly variant was known to be associated with CK levels, a phenotype independent of CK elevations was created in order to determine if the association of the variant with statin intolerance was confounded by the variant’s association with CK levels. This definition was derived from the GAUSS-2 trial, the consensus definition based on recommendations by Banach et al. and recommendations of the National Lipid Association (NLA) in 2014.
These phenotypic definitions of statin intolerance were validated against known SLCO1B1 genotype risk score 13 and the outcome of major adverse cardiovascular event (MACE). Replication was sought from the CPRD-STAGE study, the PREDICTION-ADR myopathy sequencing cohort as well as the JUPITER clinical trial of atorvastatin. Individuals homozygous for Asp247 had 1.96 times the odds of having general statin intolerance compared to carriers of the 247Gly variant. Individuals homozygous for LILRB5 Asp247 had 1.43 times the odds of being intolerant to the lowest dose of a statin (LDI) compared to carriers of the 247Gly variant. In the PREDICTION-ADR cohort, individuals homozygous for LILRB5 Asp247 had higher odds of developing SIM (OR 1.48 P-value =0.025) compared to those carrying the 247Gly variant. In the JUPITER trial individuals homozygous for Asp247 had 1.35 times the odds of developing myalgia (P-value = 0.01) compared to 247Gly variant carriers. Interestingly, interaction between genotype and statin use was significant (P-value = 0.04). An analysis of effect stratified by genotype, showed that a statin-specific myalgia effect is only seen among 247Gly variant carriers. Therefore, while Asp247 homozygotes have an overall higher risk of myalgia, statin-induced myalgia is only observed in 247Gly variant carriers.
A meta-analysis of the association between LILRB5 Asp247Gly and outcomes examined across studies showed that Asp247 homozygotes have 1.34 times the odds of having outcomes associated with statin intolerance compared to carriers of the 247Gly variant (95% CI: 1.16 1.54) P-value 7x10-5.
Therefore we conclude that the LILRB5 Asp247 homozygous genotype is robustly associated with increased risk of outcomes associated with statin intolerance across observational, clinically adjudicated and clinical trial datasets.
A possible mechanism for the role of LILRB5 in muscle pathology is suggested by a recent study by Kuswanto et al. who highlight a role for the immune system in the repair and regeneration of skeletal muscles. They report that the presence and rapid accumulation of T regulatory (Treg) cells is crucial in the repair of damaged skeletal muscles 24. It is also reported that statins increase both the number and suppressive function of CD4+ Foxp3+ Treg cells; Foxp3 is a transcription factor that is the master regulator of Treg immune-suppressive activity. The same study shows sustained increased expression of Foxp3 with statin use. Therefore, mechanisms that induce Foxp3 expression and sustain Treg function are of great interest in understanding muscle homeostasis. We examined mRNA expression data from the GTEx portal and found that the LILRB5 Asp247 variant is associated with FOXP3 mRNA expression in the spleen. FOXP3 lies on the X chromosome, indicating a trans-eQTL effect, with the LILRB5 variant having an indirect immunomodulatory effect on FOXP3 expression. These findings underscore a role for immunogenetics in understanding muscle damage and repair, with LILRB5 Asp247Gly being the first candidate to be found for common statin intolerance.
While the current variant results in an amino acid substitution, it is not clear that the phenotype results from this protein change. This is especially true as this variant is also associated with LILRB5 expression in certain tissues in the GTEx portal database 27,28. We used GTEx to identify the strongest cis¬-eQTLs for LILRB5 expression, and found that SNPs rs1408812 and rs3852892 showed much stronger effects in skeletal muscles and whole blood respectively. However, these polymorphisms have no known associations with CK, LDH or statin intolerance and were not in linkage with the Asp247Gly variant. This demonstrates that the observed phenotypes are not associated with genetically driven variation in LILRB5 expression and therefore the association with the Asp247Gly variant would appear to be mediated by a functional change in the protein due to the amino acid substitution. However, this has to be confirmed by direct experimentation.
Previous studies have shown that CK levels were associated with LILRB5 variants and with all definitions of intolerance, including diagnoses of myalgia, making it a potential confounder. We have ruled out any artefactual associations by including CK measures as a covariate in analyses, where appropriate. Since the general statin intolerance phenotype in GoDARTS and diagnoses of myopathy in CPRD-STAGE and PREDICTION-ADR were entirely dependent on elevated CK, this adjustment was not possible. Additionally, the findings of this study could be impacted by confounders that were unmeasured in the cohorts used. Indeed, the non-significant finding in CPRD-STAGE might be due to the use of population controls with no available clinical or demographic information, limiting inclusion of important covariates in the analysis.
Another source of bias in non-randomized studies could be drug interactions that are known to increase risk of SIM. Inhibition of cytochrome P450 (CYP) 3A4 can increase exposure to simvastatin and atorvastatin several fold. On the other hand, inhibition of hepatic influx transporter OATP1B1 by drugs such as gemfibrozil can increase plasma concentrations of all statins. Due to the increased risk of muscle toxicity with combination therapy, fibrates are not generally recommended for concurrent use with statins for primary lipid control, but are co-prescribed to patients with Type 2 Diabetes (T2D) to control triglyceride levels. Since GoDARTS is primarily a population of T2D patients, fibrates are widely prescribed in the study population. We observe comedications, especially fibrates consistently increased the risk of statin intolerance. However, in adjusted models, the association of the LILRB5 Asp247Gly variant was independent of the effect of these co-medications. Additionally, the effect is observed in the RCT setting.
The JUPITER trial allowed us to examine the association of the LILRB5 Asp247Gly variant in the presence and absence of statins, revealing that homozygosity of Asp247 is associated with increased odds of myalgia regardless of statin allocation. This supports the concept that LILRB5 Asp247 homozygous genotype modulates CK and LDH levels through statin-independent muscle damage. The association of 247Gly carriers with the development of myalgia in JUPITER is suggestive of a more complex biological gene-drug interaction. The observation that 247Gly carriers showed statin-specific myalgia, suggests a subpopulation of individuals who are inherently protected from myalgia, are susceptible to true “statin-induced” myalgia. In observational data such as GoDARTS it is impossible to determine if intolerance to statins is occurring due to statin-specific or non-specific side effects. However, occurrence of intolerance is associated with increased risk of adverse CV outcomes. We believe a recruit-by-genotype trial would be the ideal platform to examine the statin-dependency by LILRB5 Asp247Gly genotypes, and to further explore the underlying immune mechanisms.
Clinical trials have consistently found no evidence of statin-specific myalgia, just as there is no difference in incidence of myalgia between placebo and rosuvastatin arms of the JUPITER trial. The lack of association of statins with muscle pain in RCTs has led to a debate regarding the existence of statin-related muscle symptoms. Indeed the difficulty of ascribing causality in statin related muscle symptoms is highlighted by the Goal Achievement after Utilizing an anti-PCSK9 Antibody in Statin-Intolerant Subjects -3 (GAUSS-3) trial. Data from this trial was the first systematic evaluation of statin-specific myalgia with rechallenge and provided an estimate of 43% of individuals having statin-specific myalgia, and also demonstrated that 37% of intolerant individuals have statin-independent or non-specific myalgia. The LILRB5 Asp247Gly genotype presents a unique opportunity to probe this phenomena of muscle pain specific to statins compared to “constitutive’ muscle pain that appears in LILRB5 Asp247 homozygotes.
Therefore we initiated a recruit by genotype study called Immunostat which is a randomised, placebo-controlled, cross-over study. We will assess the immunological response and tolerability to atorvastatin in relation to LILRB5 rs12975366 single nucleotide polymorphism. We aim to recruit a total of sixty statin-naïve, healthy individuals: thirty with the LILRB5 rs12975366 T/T genotype and thirty with the C/C genotype. All participants will be identified from the GoSHARE cohort. We intend to carry out interim analysis after 15 vs 15 participants to assess for irrefutable results.
Atorvastatin (80mg daily) will be prescribed to all participants. Atorvastatin has been widely used in clinical trials in individuals without hypercholesterolaemia. Therefore we know that it is safe for use in healthy volunteers.
After recruitment, all participants will be commenced on atorvastatin or placebo during treatment period one. Before and at the end of the treatment period, blood samples will be taken and a muscle symptoms questionnaire will be completed to assess the tolerability and immune response to the study drug exposure. After four weeks, the study drug is stopped for a washout period of three weeks before cross-over commences. Thereafter, during treatment period two, the alternate study drug will be started, and tolerability will be assessed similar to that in period one. The study is powered to detect previously published effects of atorvastatin on serum CK levels, and will provide detailed information on the gene drug interaction effect on FOXP3 expressing Treg cells that are known to modulate muscle repair and homeostasis.
Potential Impact:
We evaluated the specifications of a pharmacogenomic test for preventing ACEI induced angioedema in terms of the required specificity, sensitivity and price for achieving cost effectiveness. Our findings indicate that testing all ACEI starters is unlikely to be cost effective as >90% specificity, >93% sensitivity and a low (<€2.00) price would be required.
Our results highlight that limiting testing to high risk populations can be a fruitful endeavour for increasing cost effectiveness. This statement is further supported by a Deterministic sensitivity analysis (DSA) demonstrating a major influence of angioedema incidence on the incremental cost effectiveness ratio. Miller et al. reported a relative risk of 3.88 and 1.45 for people of African-American ethnicity and for women, respectively. In our model this had a profound positive impact on parameter requirements. Further clarification of risk factors, for example women of African-American ethnicity, could prove to lower diagnostic accuracy and test price to more favourable ranges that could warrant actual development of a PGx test for this specific indication.
Nevertheless, individual tests for rare ADRs may not be very efficient. Plumpton and colleagues have shown that single testing is not always cost effective, even when a proper biomarker or SNP is present. Their results indicate that mainly Human Leukocyte Antigen (HLA) polymorphisms are cost-effective single targets. These HLA polymorphisms predispose for hypersensitivity reactions, sometimes leading to very severe ADRs like Stevens-Johnson Syndrome (SJS) and Toxic Epidermal Necrolysis (TEN), induced by carbamazepine, abacavir and allopurinol. Not only are these ADRs more severe with mortality ranging from 10% to 40% for TEN, incidence rates of up to 5% are much higher than incidence rates of ACEI induced angioedema.
There could be a solution to biomarkers that do have value but are too costly to implement separately: Combine many of these tests into a single package or perform them together with a test that will be performed in routine daily practice. In this way, the fixed costs of sampling, transport to a lab and reporting the results are spread and incremental costs per test could decrease dramatically. We can extend the idea of combining tests to whole genome microarray genotyping, exome or whole genome sequencing. Currently, these sequencing techniques considered to be too costly for implementation in routine practice but prices have been falling dramatically. When routine sequencing or genotyping becomes part of daily clinical practice, all future genomic markers will deliver additional benefit to patients, regardless of the rarity of the predictor. Sadly, the full potential value that innovations may deliver in the future cannot be captured in traditional cost effectiveness analysis.
The two most important limitations of our study need to be addressed. Firstly, the DSA indicates a strong influence of the additional cost of antihypertensive treatment. This is the cost of a false positive case. In Dutch practice, switching to another antihypertensive is more expensive than ACEI treatment. This price difference is likely to be country specific. In other jurisdictions where ACEI treatment is more expensive than other antihypertensive treatment, the genotyping strategy would result in drug-cost savings in the event of a (false) positive diagnosis.
Secondly, model parameters were based on multiple studies with different study designs possibly leading to biased estimates. Especially our assessment of mortality risk was based on suboptimal evidence that required some assumptions. However, the DSA indicates a relatively low influence of mortality risk on model outcomes. Utility scores were assessed by estimating the answers to the EQ5D health questionnaire which is clearly sub-optimal. The DSA indicates that these parameters have a negligible effect on the results.
Conclusion.
Our study indicates that testing all patients starting an ACEI to predict angioedema is unlikely to be cost effective as the test should have a high diagnostic accuracy combined with a sub €2.00 cost. Selectively testing only populations that have an increased risk of developing ACEI induced angioedema improves test characteristics needed and price for an incremental cost-effectiveness ratio (ICER) between €20,000 and €80,000. While separate testing for this variation for all ACEI starters or subgroups is not cost-effective, implementing high density microarray genotyping, whole exome or genome sequencing in routine clinical practice will result in economically attractive benefits of finding genetic variations like the one discussed here.
While this formal analysis has indicated the lack of cost effectiveness of point of care testing based on official methods of calculating the cost of the burden of the ADR, it does not take into account the personal suffering of more common intolerances to the drugs. Both statins and ACE inhibitors both have diverse common side effects that are not deemed clinically significant, but do lead to drug discontinuation and poor adherence. It is clear that taking this into account would give the patient or the patient’s physician a different perspective on whether or not a test would be required. This is clearly reflected by the work by the Estonian SME partner, ASPER Biotech who developed and marketed a Statin myopathy test as part of Work Package 7. This was a single marker test based on a very robust pharmacokinetic marker in SLC01B1, and reflects a major genetic defect in a molecular statin pump present in the liver. This is known to prevent the statin going into the liver to be metabolised and then cleared from the body. Accordingly more drug is available to promote muscle toxicity and other dose related toxicities. Asper Biotech reported a healthy uptake of this test from physicians around Europe, and an enthusiasm for the development of more sophisticated tests as they become available.
It is therefore clear that while single testing at an individual drug start has many hurdles to face in terms of cost effectiveness and logistics, it is clear that pre-emptive whole genome testing (sequence or SNP chip) would be a major step to overcome this. GWAS SNP chips are now being advertised for under €30, For the 150 PGx tests currently recommended by the FDA this would represent 20 cents per test. This clearly is way beyond the tipping point of cost effectiveness. Therefore the challenge now lies in determining when pre-emptive genotyping should take place, who should initiate/pay for this. The whole genome assay cost is now trivial, being less than the cost of an evening meal at a city centre restaurant, so it is clear that this could be a personal cost, or driven by the health care provider. Another challenge is to build the decision support mechanisms in place so that the genomic information is taken into account at every relevant clinical event. This would require the genomic data to be available to all clinical managements systems. To this end the PREDICTION-ADR partner, University of Liverpool, is now involved in the EU project entitled “Ubiquitous pharmacogenomics” (http://upgx.eu). This project has now initiated the PREPARE (Pre-emptive Pharmacogenomic Testing for Preventing Adverse Drug Reactions) study to implement and evaluate the impact of pharmacogenomic testing on therapy outcomes in seven European clinical centres.
The goal of PREPARE is to show that pre-emptively testing patients for an entire panel of clinically relevant PGx markers will result in an overall reduction in the number of clinically relevant drug-genotype associated adverse drug reactions. Pre-emptive testing means that the testing is performed before a certain drug is prescribed. This means that the results can be used by your physician or pharmacist to select the correct drug or dose for you. Furthermore, the cost-effectiveness of testing patients for an entire panel of relevant markers at once will be evaluated. Within 3 years, starting from January 1, 2017, 8,100 patients will be pre-emptively tested for more than 40 clinically relevant PGx markers across 13 important pharmacogenes. For 4,050 patients assigned to the study group, their test results will be used by their healthcare providers to guide the dose and drug selection for over 40 commonly prescribed drugs. The other half of patients assigned to the control group will receive standard of care during the study period but will be provided with their test results after the study ends. Data on therapy outcome and other parameters collected during the study period will be analyzed in 2020.
Unfortunately markers from PREDICTION-ADR were not suitably validated for inclusion in PREPARE, however it is anticipated that markers from the PREDICTION-ADR study will be incorporated into second generation large scale implementation schemes that will follow from PREPARE. For example, we will be working to ensure that the Scottish Health Research Register- SHARE (www.registerforshare.org) which has obtained pre-consent for the pre-emptive genotyping of spare blood and use in informing care in over 160000 individuals will introduce the PREPARE plus PREDICTION-ADR markers in their panel. This will transform the landscape of efficacy and safety in prescribing and provide a step change in the implementation of personalised medicine.
List of Websites:
http://www.prediction-adr.eu/
Professor Colin N.A. Palmer
Pat Macpherson Centre for Pharmacogenetics and Pharmacogenomics
Division of Molecular & Clinical Medicine
Level 5, Mailbox 12
Ninewells Hospital and Medical School
Dundee
DD1 9SY
Phone Number: +(44) 01382 383155
Email Address: c.n.a.palmer@dundee.ac.uk
http://medicine.dundee.ac.uk/staff-member/professor-colin-palmer
The primary aim of PREDICTION-ADR was to develop genetic risk assessment and diagnostic tools for the prediction of adverse drug reactions in two commonly prescribed drug classes that are used to manage cardiovascular risk, namely statins and Angiotensin Converting Enzyme inhibitors (ACEI)/Angiotensin Receptor Blockers (ARB). Drug intolerance is a major factor in lack of adherence and efficacy in both classes of drugs. The ultimate goal is to optimise the management of cardiovascular disease (CVD) with currently used medications. Particularly, we aimed to develop tools that will identify (i) patients with likelihood of experiencing severe adverse reactions to drugs, and (ii) patients who may discontinue drugs due to milder intolerance. The application of such drug response prediction tools is expected to directly improve patient care through individualized management of both cardiovascular risk and disease in addition to reducing health care cost. This work utilised the integration of expertise from leading European groups in the fields of cardiovascular pharmacology, pharmacoepidemiology and molecular genetics, with clinical medicine.
The development of genetic drug response prediction tools involved a clearly defined process of (a) discovery of novel genetic biomarkers by next generation sequencing, (b) using sophisticated statistical analysis to define massively multi-allelic genetic interaction models to predict adverse drug reactions with a high specificity and sensitivity, (c) validation of these markers in population-based data samples and clinical trials, (d) development of a novel diagnostic incorporating the validated drug response prediction tools, and finally, (e) dissemination through engagement with stakeholders, patient groups and exploitation through commercialization by ASPER-Biotech a leading EU genetic diagnostic provider.
Exome sequencing was successfully performed in a large case control study for both ACEI angioedema and statin induced myopathy and the data was analysed by TAXONOMY 3 interaction software and produced an AUC of 85% and 84% for myopathy and angioedema respectively. Rare and common variants were found for both pharmacokinetic and immune genetic variants including a novel common protective variant in SLC01B1 and implicated a novel immune-modulatory gene family, the Leukocyte Immunoglobulin Like Receptors in statin induced and statin independent muscle pain. Results for angioedema highlighted a role for potassium channels in side effects to ACE inhibitors that also include the more common cough reaction. Effects in the HLA-DRB1 gene were seen, in agreement with previous studies on aspirin induced angioedema. A cost effectiveness analysis was performed for genetic testing in new ACEI starts. This came to the conclusion that point of care testing for single drug starts was likely to be not cost effective with tests costing more than 2 EURO even if the test provided an AUC of over 90%. This however does not consider the cost of more common intolerance, and it is possible that pre-emptive whole genome testing at first ever drug start would provide a life-time long term cost effectiveness. The PREDICTION-ADR study will provide genetic algorithms to prevent harm from statins and ACE inhibitors in future implementation of pre-emptive genotyping programmes such as those being piloted by Ubiquitous Pharmacogenetics (http://upgx.eu).
Project Context and Objectives:
The primary aim of PREDICTION-ADR was to develop genetic risk assessment and diagnostic tools for the prediction of adverse drug reactions in two commonly prescribed drug classes that are used to manage cardiovascular risk, namely statins and Angiotensin Converting Enzyme inhibitors (ACEI)/Angiotensin Receptor Blockers (ARB). Drug intolerance is a major factor in lack of adherence and efficacy in both classes of drugs. The ultimate goal is to optimise the management of cardiovascular disease (CVD) with currently used medications. Particularly, we aimed to develop tools that will identify (i) patients with likelihood of experiencing severe adverse reactions to drugs, and (ii) patients who may discontinue drugs due to milder intolerance. The application of such drug response prediction tools is expected to directly improve patient care through individualized management of both cardiovascular risk and disease in addition to reducing health care cost. This work utilised the integration of expertise from leading European groups in the fields of cardiovascular pharmacology, pharmacoepidemiology and molecular genetics, with clinical medicine.
The development of genetic drug response prediction tools involved a clearly defined process of (a) discovery of novel genetic biomarkers by next generation sequencing, (b) using sophisticated statistical analysis to define massively multi-allelic genetic interaction models to predict adverse drug reactions with a high specificity and sensitivity, (c) validation of these markers in population-based data samples and clinical trials, (d) development of a novel diagnostic incorporating the validated drug response prediction tools, and finally, (e) dissemination through engagement with stakeholders, patient groups and exploitation through commercialization by ASPER-Biotech a leading EU genetic diagnostic provider.
The principle basis for the concept of PREDICTION-ADR is clinical need. With the ageing population in the developed and developing world, drug therapy to manage CVD is steadily increasing. Although the incidence of the most severe adverse drug reactions (ADRs) is rare, given the number of prescriptions of CVD drugs globally, the absolute number of ADRs is substantial. Studies have shown that ADRs are one of the most common reasons for hospitalisation in the adult population. It has been proposed that ADRs are the fourth to sixth leading cause of death in hospitalised patients. ADRs also commonly lead to poor compliance and discontinuation of vital therapies and therefore negatively affect the burden of CVD in health care systems worldwide. The current knowledge about possible genetic causes of ADRs is limited. Studying the genetic basis of susceptibility to ADRs may provide possibilities for identification of susceptible individuals through pharmacogenetic testing.
Understanding the molecular basis of ADRs may also make it possible to design safer drugs. Statins primarily reduce the risk of coronary artery disease (CAD) by lowering blood cholesterol through inhibition of the HMG-CoA reductase enzyme. Large clinical trials show a 27% average relative risk reduction of major coronary events. However there is large variability in benefits from statin therapy. This variability can be partially explained by factors such as adherence, gender, age, diet, concomitant drug use and environmental factors, with drug intolerance being a major factor in adherence and concomitant drug use also being a major factor in determining intolerance. Therapy with statins has become relatively cheap because most statins are out of patent which means that even more patients will be treated with them. Next to that, the dosages used in the treatment of patients are increasing, because it has been shown that higher doses are more effective. With more patients being treated with increasing doses the burden of adverse events becomes ever greater.
Project Results:
The PREDICTON-ADR consortium brought together a wide range of investigators from Sweden, Holland, Estonia and the UK to investigate the genetic basis of statin induced muscle damage and angioedema resulting from exposure to ACE-inhibitor drugs. Previous studies had been hampered by lack of consistent phenotype definition, poor sample sizes, and the investigation of limited candidate genes.
The first task of the consortium was to provide a robust analysis of the appropriate phenotypic descriptions for both statin myopathy and ACEI induced angioedema. Workshops held in Liverpool in December 2013 resulted in the publication of authoritative manuscripts in the Journal Clinical Pharmacology and Therapeutics, the leading journal for pharmacogenetics. This informed the study design for our exome sequencing project and cases were ascertained from Sweden, UK and Holland. All cases and controls were subjected to centralised adjudication by the responsible partner (Uppsala for ACEI angioedema and Liverpool for statin induced myopathy). Exome sequencing was performed at three centres using a harmonised protocol and identical capture reagent (Agilent human exome V4).
The comparability and quality of the sequencing at all three sites was confirmed by the sequencing of 8 common samples obtained from Dundee. The sequencing of the cases and controls was completed and gene-based test analysis was performed. This used both the absolute number of variants observed in each gene in each case and control (T1 Burden) and the dispersion of the observed effect sizes across each gene (SKAT-O). For statin myopathy this revealed interesting associations with appropriate loci including the statin transporter SLCO1B1 and the HLA region. In particular a strong association was observed with a novel gain of function variant. Exome sequencing on a discovery cohort of 235 statin cases, 251 ACE cases and 502 controls from 3 different countries (UK, Sweden and Netherland) was performed and analysed through 2 different pipelines: -BWA-MEM/GATK haplotype-caller (Pipeline 1) -BWA-Stampy/Platypus (Pipeleine2).
Quality control excluded variants with poor genotyping as evidenced by genotype call rate < 95%, deviation from the Hardy-Weinberg Equilibrium with p<1x10-6 or failure of other criteria resulting in 550,057 and 558,909 variants from (2) passing all quality criteria, respectively for Statin and ACEI, 527,795 and 509,173 variants from (1) respectively for Statin and ACEI.
Analyses were performed as simple case controls Burden test and SKAT taking into account the sequencing centre as a covariable using RareMetal (Feng et al.). Analyses were restricted to the 38,642 genes fulfilling all quality control filters and to variants or resulting in a stop missense/splicing/gain/loss. Variants were also only considered if they exhibited minor allele frequencies less than 0.01%.
QQ plots of both ACEI and Statin data showed an early separation of the observed from the expected with a depletion, which means that p values were less significant than expected.
Top hits from both analyses (Statin and ACEI) were different between the two pipelines. However the C9 gene encoding Complement 9, ranked in the top ten hits for the 2 pipelines for statin myopathy with 12 common variants between the 2 pipelines, with Burden test p values of 3.17x10-4 (Pipeline 1) and 5.21x10-4 (Pipeline 2), SKAT p values of 3.96x10-3 (Pipeline 1) and 3.79x10-4 (Pipeline 2) respectively.
Besides C9, our strategy revealed other candidate genes for Statin like FAM110A involved in immunology responses that showed SKAT p values of 4.16x10-3 for (2) and 6.68x10-3 for (1) with 7 common variants between the 2 pipelines (2 singulars in (1)), or SLC27A6 and HLA-A (Table 4).
Another important finding was the presence of a common SLCO1B1 “gain of function” variant present at a MAF of 10%. This variant, Pro155Thr was strongly protective against statin myopathy and has been previously demonstrated to protect against Methotrexate toxicity. In addition the Burden test using rare variants in SLC01B1 showed nominal significance (p=0.0224) and indicated a potential action on statin myopathy of at least 5 rare variants (Ala13Glu,Ser51Phe,Ile106Thr,Val235Met,Il499Val) all present at frequencies below 0.005.
A strong determinant of ACEI induced angioedema was seen in the HLA-DRB1 gene with both a common variant Val73Met (MAF 10%) providing exome-wide significance (6X10-7) and with suggestive significance in the rare variant burden test (Gln260Pro and His141Tyr, both MAF). This is consistent with previous reports of an association of HLA-DRB1 with aspirin induced angioedema. Candidate genes for ACEI induced angioedema were identified like SLC22A6 a member of a family of organic anion transporters that mediates the secretion of endogenous and exogenous metabolites including drugs, PPP1R12B a protein phosphatase or CLCN2 involved in chlorine ion transport.
A strong signal in an intron of the “Maxi” potassium channel KCNMA1 was obtained from genome-wide association study (GWAS) analysis of the Uppsala angioedema cohort. This was then replicated in the Danish/Swedish angioedema replication cohort, the Marshfield angioedema GWAS data, the GoDARTS/Rotterdam ACEi discontinuation cohort and two cohorts of ACEI cough. None of the cohorts replicated the strong effect size observed in the discovery GWAS (OR= 2.4 p=4.2X10-8). However, the signal is directionally consistent in all studies with a combined effect estimate of OR=1.44 p=2X10-6). The test for heterogeneity suggested that the initial finding had a strong overestimate, and there was a hint that the signal was weakest for the cough phenotype.
The significant finding in the intolerance phenotype would suggest that other forms of intolerance including angioedema may be influenced by this variant. Exome sequencing revealed that rare variants in KCNMA1 and its heterodimeric partner KCNMB1 are also associated with angioedema. This is relevant to the finding that the associated regulatory protein KCNIP4 was our main finding from the GWAS of ACEI associated cough, underscoring the role of these channels in side effects to ACEI. This work is now being followed up by the Dundee group as part of the IMI project BEAT-DKD, where we are investigating genes and potassium levels in the use of ACEI drugs to protect from chronic kidney disease. This will incorporate drug intolerance features including intolerance to determine individuals that will benefit the most, and suffer the least harm from ACEI drugs.
In WP4 we examined the potential role of gene-gene interactions using the Taxonomy 3 software of the SME partner Adorial. This software had been developed for GWAS data and this was the first use with exome sequence data. Significant effort was assigned to ensure that the Taxonomy3 software would be able to accept exome sequence data with no opportunity for data corruption. This entailed frequent discussions with members of WP3. It is also important that clinical variables are included in the Taxonomy3 analysis, and much interaction with members of WP1 and WP2 was necessary to understand the nature of the variables and develop a suitable software environment to receive these data.
Sequence data generated in WP3 was assessed using our standard two-stage approach. Univariate QC was used to exclude subjects and markers with poor quality data. This posed a significant challenge. Our method is very powerful, but has stringent requirements for the data that can be analysed. Our usual data quality cut-off is 2% of missing data for both subjects and markers, to avoid potential biases in the data due to, for example, poor quality DNA data. Applying this criterion only excluded a few of the subjects but the majority of markers failed this criterion. Rather than risk introducing biases into the data analysis, we retained the QC criterion, and proceeded with analysis of the reduced marker set.
The second stage of our data QC involved co-analysis of the subjects’ data with HapMap data form subjects of known ethnicity to identify an ethnically homogeneous group of cases and controls, minimising patient stratification bias. Although only a small number of markers were common to the dataset from WP3 and the HapMap data, a surprisingly good plot allowed homogeneous case/control groups to be determined.
Taxonomy3 Analysis:
Two types of analysis were carried out on both datasets:
1. ‘Main effect’ analysis, looking at the contribution of markers to case/control separation
2. Interaction analysis, where the contribution of interacting markers is assessed
The analyses were carried out using in-house proprietary software utilising cloud computing capacity where necessary to speed the analysis times.
Predictive Power:
The markers identified in the main effect and interaction analysis were used to determine predictive power using the Leave-One-Our-Crossover Validation method (LOOCV). In this method, each subject in turn is omitted from the analysis, and the analysis parameters used to predict whether the subject is a case or a control.
The results from both adverse event studies were very similar. The output from the Taxonomy3 main effect showed only limited predictive power (AUC=0.59 for myopathy and 0.6 for angioedema), but the predictive power was substantially increased by including the SNP-SNP interactions in the analysis (AUC=0.85 for myopathy and 0.84 for angioedema).
Conclusions:
The fact that most variables did not meet our missing data criterion had a significant impact on our ability to generate genetic associations. However despite this limitation, the predictive power of the identified markers was surprisingly good, although not sufficient to provide useful information for clinical providers. The output of WP6 suggested that higher sensitivities and specificities would be required to have a cost-effective screen for these adverse events.
The predictive power was significantly improved when SNP-SNP interactions are included in the calculation. In our experience, this interaction analysis – a unique feature of Taxonomy3 – typically generates more novel genetic insights, providing greater pathology understanding. This will translate into greater predictive power for adverse events such as myopathy and angioedema. Future work should investigate the use of whole genome sequence or imputed microarray data to improve the marker coverage across the genome.
Work package 5 for population validation looked at common variation in large cohorts and examined the role of genetic variation in determining more common forms of intolerance as marked by features such as mild muscle pain or cough. GWAS data was available from the GoDARTS cohort as well as the Rotterdam study where statin intolerance was defined by discontinuation and drug switching algorithms as well as by mild alteration in CK results. Collaborations were initiated with US EMERGE groups with GWAS data based on EMR data in a similar fashion to the GoDARTS and Rotterdam. In particular the Vanderbilt group defined a GWAS for ACEI cough and the Marshfield group, where we worked together to establish variants in KCNIP as determinants of ACE-I intolerance including cough, but not angioedema.
In addition the JUPITER trial of atorvastatin contributed GWAS data with a phenotype of myalgia. During the early part of the PREDICTION-ADR study a GWAS by Dubé et al. 11 reported a missense variant, Asp247Gly, in the leukocyte immunoglobulin-like receptor subfamily B member 5 gene LILRB5 on chromosome 19 that was associated with circulating serum CK levels. The mean CK levels of Asp247 homozygotes (T/T) were significantly higher. This association was found to be independent of statin use, however there is no known biological mechanism for the variant in determining CK levels. A GWAS by Kristjansson et al. of over 60,000 Icelanders replicated the association of the variant and CK levels 12. The same study also reported the association of the variant with serum LDH levels in a population of over 90,000 Icelanders 12. The LILRB5 variant showed the same direction of effect, i.e. Asp247 homozygotes had higher LDH and CK levels. LDH is often used in conjunction with CK as a marker of tissue damage. The findings suggested the variant might impart a statin independent susceptibility to muscle-based events. This makes LILRB5 a potential marker for susceptibility to the commonly noted muscle-based symptoms attributed to statin intolerance.
These discoveries warranted an investigation into the role of the LILRB5 variant in statin intolerance. Population-based studies use surrogate markers of intolerance, such as elevations in CK, trends in statin treatment, dose reductions, switching or the discontinuation of therapy. Therefore, a priori, we considered two definitions of statin intolerance, one dependent on and one independent of elevated CK levels. We hypothesize that carriers of variants associated with higher muscle enzyme levels (CK and LDH) will also be predisposed to forms of statin intolerance independent of CK levels.
The principal cohort used was the Genetics of Diabetes Audit and Research, Tayside Scotland (GoDARTS). GoDARTS has been previously used to establish pharmacogenetics (PGx) associations of genes such as the hepatic influx transporter SLCO1B1 and statin intolerance 13. GoDARTS contained 11,912 statin users and provides approximately 98,000 person-years of statin exposure, providing an ideal cohort to examine the association of this genetic variant with statin intolerance. Replication was examined in the Clinical Practice Research Datalink (CPRD) STAGE study 14 and clinically adjudicated cases of statin-induced myopathy (SIM) in the European PREDICTION-ADR consortium study. The interaction of this effect with statin use was then studied among participants who developed myalgia in the JUPITER (Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin) randomized clinical trial (RCT) where individuals were allocated rosuvastatin or placebo to assess the relative reduction in vascular events.
Since the LILRB5 Asp247Gly variant was known to be associated with CK levels, a phenotype independent of CK elevations was created in order to determine if the association of the variant with statin intolerance was confounded by the variant’s association with CK levels. This definition was derived from the GAUSS-2 trial, the consensus definition based on recommendations by Banach et al. and recommendations of the National Lipid Association (NLA) in 2014.
These phenotypic definitions of statin intolerance were validated against known SLCO1B1 genotype risk score 13 and the outcome of major adverse cardiovascular event (MACE). Replication was sought from the CPRD-STAGE study, the PREDICTION-ADR myopathy sequencing cohort as well as the JUPITER clinical trial of atorvastatin. Individuals homozygous for Asp247 had 1.96 times the odds of having general statin intolerance compared to carriers of the 247Gly variant. Individuals homozygous for LILRB5 Asp247 had 1.43 times the odds of being intolerant to the lowest dose of a statin (LDI) compared to carriers of the 247Gly variant. In the PREDICTION-ADR cohort, individuals homozygous for LILRB5 Asp247 had higher odds of developing SIM (OR 1.48 P-value =0.025) compared to those carrying the 247Gly variant. In the JUPITER trial individuals homozygous for Asp247 had 1.35 times the odds of developing myalgia (P-value = 0.01) compared to 247Gly variant carriers. Interestingly, interaction between genotype and statin use was significant (P-value = 0.04). An analysis of effect stratified by genotype, showed that a statin-specific myalgia effect is only seen among 247Gly variant carriers. Therefore, while Asp247 homozygotes have an overall higher risk of myalgia, statin-induced myalgia is only observed in 247Gly variant carriers.
A meta-analysis of the association between LILRB5 Asp247Gly and outcomes examined across studies showed that Asp247 homozygotes have 1.34 times the odds of having outcomes associated with statin intolerance compared to carriers of the 247Gly variant (95% CI: 1.16 1.54) P-value 7x10-5.
Therefore we conclude that the LILRB5 Asp247 homozygous genotype is robustly associated with increased risk of outcomes associated with statin intolerance across observational, clinically adjudicated and clinical trial datasets.
A possible mechanism for the role of LILRB5 in muscle pathology is suggested by a recent study by Kuswanto et al. who highlight a role for the immune system in the repair and regeneration of skeletal muscles. They report that the presence and rapid accumulation of T regulatory (Treg) cells is crucial in the repair of damaged skeletal muscles 24. It is also reported that statins increase both the number and suppressive function of CD4+ Foxp3+ Treg cells; Foxp3 is a transcription factor that is the master regulator of Treg immune-suppressive activity. The same study shows sustained increased expression of Foxp3 with statin use. Therefore, mechanisms that induce Foxp3 expression and sustain Treg function are of great interest in understanding muscle homeostasis. We examined mRNA expression data from the GTEx portal and found that the LILRB5 Asp247 variant is associated with FOXP3 mRNA expression in the spleen. FOXP3 lies on the X chromosome, indicating a trans-eQTL effect, with the LILRB5 variant having an indirect immunomodulatory effect on FOXP3 expression. These findings underscore a role for immunogenetics in understanding muscle damage and repair, with LILRB5 Asp247Gly being the first candidate to be found for common statin intolerance.
While the current variant results in an amino acid substitution, it is not clear that the phenotype results from this protein change. This is especially true as this variant is also associated with LILRB5 expression in certain tissues in the GTEx portal database 27,28. We used GTEx to identify the strongest cis¬-eQTLs for LILRB5 expression, and found that SNPs rs1408812 and rs3852892 showed much stronger effects in skeletal muscles and whole blood respectively. However, these polymorphisms have no known associations with CK, LDH or statin intolerance and were not in linkage with the Asp247Gly variant. This demonstrates that the observed phenotypes are not associated with genetically driven variation in LILRB5 expression and therefore the association with the Asp247Gly variant would appear to be mediated by a functional change in the protein due to the amino acid substitution. However, this has to be confirmed by direct experimentation.
Previous studies have shown that CK levels were associated with LILRB5 variants and with all definitions of intolerance, including diagnoses of myalgia, making it a potential confounder. We have ruled out any artefactual associations by including CK measures as a covariate in analyses, where appropriate. Since the general statin intolerance phenotype in GoDARTS and diagnoses of myopathy in CPRD-STAGE and PREDICTION-ADR were entirely dependent on elevated CK, this adjustment was not possible. Additionally, the findings of this study could be impacted by confounders that were unmeasured in the cohorts used. Indeed, the non-significant finding in CPRD-STAGE might be due to the use of population controls with no available clinical or demographic information, limiting inclusion of important covariates in the analysis.
Another source of bias in non-randomized studies could be drug interactions that are known to increase risk of SIM. Inhibition of cytochrome P450 (CYP) 3A4 can increase exposure to simvastatin and atorvastatin several fold. On the other hand, inhibition of hepatic influx transporter OATP1B1 by drugs such as gemfibrozil can increase plasma concentrations of all statins. Due to the increased risk of muscle toxicity with combination therapy, fibrates are not generally recommended for concurrent use with statins for primary lipid control, but are co-prescribed to patients with Type 2 Diabetes (T2D) to control triglyceride levels. Since GoDARTS is primarily a population of T2D patients, fibrates are widely prescribed in the study population. We observe comedications, especially fibrates consistently increased the risk of statin intolerance. However, in adjusted models, the association of the LILRB5 Asp247Gly variant was independent of the effect of these co-medications. Additionally, the effect is observed in the RCT setting.
The JUPITER trial allowed us to examine the association of the LILRB5 Asp247Gly variant in the presence and absence of statins, revealing that homozygosity of Asp247 is associated with increased odds of myalgia regardless of statin allocation. This supports the concept that LILRB5 Asp247 homozygous genotype modulates CK and LDH levels through statin-independent muscle damage. The association of 247Gly carriers with the development of myalgia in JUPITER is suggestive of a more complex biological gene-drug interaction. The observation that 247Gly carriers showed statin-specific myalgia, suggests a subpopulation of individuals who are inherently protected from myalgia, are susceptible to true “statin-induced” myalgia. In observational data such as GoDARTS it is impossible to determine if intolerance to statins is occurring due to statin-specific or non-specific side effects. However, occurrence of intolerance is associated with increased risk of adverse CV outcomes. We believe a recruit-by-genotype trial would be the ideal platform to examine the statin-dependency by LILRB5 Asp247Gly genotypes, and to further explore the underlying immune mechanisms.
Clinical trials have consistently found no evidence of statin-specific myalgia, just as there is no difference in incidence of myalgia between placebo and rosuvastatin arms of the JUPITER trial. The lack of association of statins with muscle pain in RCTs has led to a debate regarding the existence of statin-related muscle symptoms. Indeed the difficulty of ascribing causality in statin related muscle symptoms is highlighted by the Goal Achievement after Utilizing an anti-PCSK9 Antibody in Statin-Intolerant Subjects -3 (GAUSS-3) trial. Data from this trial was the first systematic evaluation of statin-specific myalgia with rechallenge and provided an estimate of 43% of individuals having statin-specific myalgia, and also demonstrated that 37% of intolerant individuals have statin-independent or non-specific myalgia. The LILRB5 Asp247Gly genotype presents a unique opportunity to probe this phenomena of muscle pain specific to statins compared to “constitutive’ muscle pain that appears in LILRB5 Asp247 homozygotes.
Therefore we initiated a recruit by genotype study called Immunostat which is a randomised, placebo-controlled, cross-over study. We will assess the immunological response and tolerability to atorvastatin in relation to LILRB5 rs12975366 single nucleotide polymorphism. We aim to recruit a total of sixty statin-naïve, healthy individuals: thirty with the LILRB5 rs12975366 T/T genotype and thirty with the C/C genotype. All participants will be identified from the GoSHARE cohort. We intend to carry out interim analysis after 15 vs 15 participants to assess for irrefutable results.
Atorvastatin (80mg daily) will be prescribed to all participants. Atorvastatin has been widely used in clinical trials in individuals without hypercholesterolaemia. Therefore we know that it is safe for use in healthy volunteers.
After recruitment, all participants will be commenced on atorvastatin or placebo during treatment period one. Before and at the end of the treatment period, blood samples will be taken and a muscle symptoms questionnaire will be completed to assess the tolerability and immune response to the study drug exposure. After four weeks, the study drug is stopped for a washout period of three weeks before cross-over commences. Thereafter, during treatment period two, the alternate study drug will be started, and tolerability will be assessed similar to that in period one. The study is powered to detect previously published effects of atorvastatin on serum CK levels, and will provide detailed information on the gene drug interaction effect on FOXP3 expressing Treg cells that are known to modulate muscle repair and homeostasis.
Potential Impact:
We evaluated the specifications of a pharmacogenomic test for preventing ACEI induced angioedema in terms of the required specificity, sensitivity and price for achieving cost effectiveness. Our findings indicate that testing all ACEI starters is unlikely to be cost effective as >90% specificity, >93% sensitivity and a low (<€2.00) price would be required.
Our results highlight that limiting testing to high risk populations can be a fruitful endeavour for increasing cost effectiveness. This statement is further supported by a Deterministic sensitivity analysis (DSA) demonstrating a major influence of angioedema incidence on the incremental cost effectiveness ratio. Miller et al. reported a relative risk of 3.88 and 1.45 for people of African-American ethnicity and for women, respectively. In our model this had a profound positive impact on parameter requirements. Further clarification of risk factors, for example women of African-American ethnicity, could prove to lower diagnostic accuracy and test price to more favourable ranges that could warrant actual development of a PGx test for this specific indication.
Nevertheless, individual tests for rare ADRs may not be very efficient. Plumpton and colleagues have shown that single testing is not always cost effective, even when a proper biomarker or SNP is present. Their results indicate that mainly Human Leukocyte Antigen (HLA) polymorphisms are cost-effective single targets. These HLA polymorphisms predispose for hypersensitivity reactions, sometimes leading to very severe ADRs like Stevens-Johnson Syndrome (SJS) and Toxic Epidermal Necrolysis (TEN), induced by carbamazepine, abacavir and allopurinol. Not only are these ADRs more severe with mortality ranging from 10% to 40% for TEN, incidence rates of up to 5% are much higher than incidence rates of ACEI induced angioedema.
There could be a solution to biomarkers that do have value but are too costly to implement separately: Combine many of these tests into a single package or perform them together with a test that will be performed in routine daily practice. In this way, the fixed costs of sampling, transport to a lab and reporting the results are spread and incremental costs per test could decrease dramatically. We can extend the idea of combining tests to whole genome microarray genotyping, exome or whole genome sequencing. Currently, these sequencing techniques considered to be too costly for implementation in routine practice but prices have been falling dramatically. When routine sequencing or genotyping becomes part of daily clinical practice, all future genomic markers will deliver additional benefit to patients, regardless of the rarity of the predictor. Sadly, the full potential value that innovations may deliver in the future cannot be captured in traditional cost effectiveness analysis.
The two most important limitations of our study need to be addressed. Firstly, the DSA indicates a strong influence of the additional cost of antihypertensive treatment. This is the cost of a false positive case. In Dutch practice, switching to another antihypertensive is more expensive than ACEI treatment. This price difference is likely to be country specific. In other jurisdictions where ACEI treatment is more expensive than other antihypertensive treatment, the genotyping strategy would result in drug-cost savings in the event of a (false) positive diagnosis.
Secondly, model parameters were based on multiple studies with different study designs possibly leading to biased estimates. Especially our assessment of mortality risk was based on suboptimal evidence that required some assumptions. However, the DSA indicates a relatively low influence of mortality risk on model outcomes. Utility scores were assessed by estimating the answers to the EQ5D health questionnaire which is clearly sub-optimal. The DSA indicates that these parameters have a negligible effect on the results.
Conclusion.
Our study indicates that testing all patients starting an ACEI to predict angioedema is unlikely to be cost effective as the test should have a high diagnostic accuracy combined with a sub €2.00 cost. Selectively testing only populations that have an increased risk of developing ACEI induced angioedema improves test characteristics needed and price for an incremental cost-effectiveness ratio (ICER) between €20,000 and €80,000. While separate testing for this variation for all ACEI starters or subgroups is not cost-effective, implementing high density microarray genotyping, whole exome or genome sequencing in routine clinical practice will result in economically attractive benefits of finding genetic variations like the one discussed here.
While this formal analysis has indicated the lack of cost effectiveness of point of care testing based on official methods of calculating the cost of the burden of the ADR, it does not take into account the personal suffering of more common intolerances to the drugs. Both statins and ACE inhibitors both have diverse common side effects that are not deemed clinically significant, but do lead to drug discontinuation and poor adherence. It is clear that taking this into account would give the patient or the patient’s physician a different perspective on whether or not a test would be required. This is clearly reflected by the work by the Estonian SME partner, ASPER Biotech who developed and marketed a Statin myopathy test as part of Work Package 7. This was a single marker test based on a very robust pharmacokinetic marker in SLC01B1, and reflects a major genetic defect in a molecular statin pump present in the liver. This is known to prevent the statin going into the liver to be metabolised and then cleared from the body. Accordingly more drug is available to promote muscle toxicity and other dose related toxicities. Asper Biotech reported a healthy uptake of this test from physicians around Europe, and an enthusiasm for the development of more sophisticated tests as they become available.
It is therefore clear that while single testing at an individual drug start has many hurdles to face in terms of cost effectiveness and logistics, it is clear that pre-emptive whole genome testing (sequence or SNP chip) would be a major step to overcome this. GWAS SNP chips are now being advertised for under €30, For the 150 PGx tests currently recommended by the FDA this would represent 20 cents per test. This clearly is way beyond the tipping point of cost effectiveness. Therefore the challenge now lies in determining when pre-emptive genotyping should take place, who should initiate/pay for this. The whole genome assay cost is now trivial, being less than the cost of an evening meal at a city centre restaurant, so it is clear that this could be a personal cost, or driven by the health care provider. Another challenge is to build the decision support mechanisms in place so that the genomic information is taken into account at every relevant clinical event. This would require the genomic data to be available to all clinical managements systems. To this end the PREDICTION-ADR partner, University of Liverpool, is now involved in the EU project entitled “Ubiquitous pharmacogenomics” (http://upgx.eu). This project has now initiated the PREPARE (Pre-emptive Pharmacogenomic Testing for Preventing Adverse Drug Reactions) study to implement and evaluate the impact of pharmacogenomic testing on therapy outcomes in seven European clinical centres.
The goal of PREPARE is to show that pre-emptively testing patients for an entire panel of clinically relevant PGx markers will result in an overall reduction in the number of clinically relevant drug-genotype associated adverse drug reactions. Pre-emptive testing means that the testing is performed before a certain drug is prescribed. This means that the results can be used by your physician or pharmacist to select the correct drug or dose for you. Furthermore, the cost-effectiveness of testing patients for an entire panel of relevant markers at once will be evaluated. Within 3 years, starting from January 1, 2017, 8,100 patients will be pre-emptively tested for more than 40 clinically relevant PGx markers across 13 important pharmacogenes. For 4,050 patients assigned to the study group, their test results will be used by their healthcare providers to guide the dose and drug selection for over 40 commonly prescribed drugs. The other half of patients assigned to the control group will receive standard of care during the study period but will be provided with their test results after the study ends. Data on therapy outcome and other parameters collected during the study period will be analyzed in 2020.
Unfortunately markers from PREDICTION-ADR were not suitably validated for inclusion in PREPARE, however it is anticipated that markers from the PREDICTION-ADR study will be incorporated into second generation large scale implementation schemes that will follow from PREPARE. For example, we will be working to ensure that the Scottish Health Research Register- SHARE (www.registerforshare.org) which has obtained pre-consent for the pre-emptive genotyping of spare blood and use in informing care in over 160000 individuals will introduce the PREPARE plus PREDICTION-ADR markers in their panel. This will transform the landscape of efficacy and safety in prescribing and provide a step change in the implementation of personalised medicine.
List of Websites:
http://www.prediction-adr.eu/
Professor Colin N.A. Palmer
Pat Macpherson Centre for Pharmacogenetics and Pharmacogenomics
Division of Molecular & Clinical Medicine
Level 5, Mailbox 12
Ninewells Hospital and Medical School
Dundee
DD1 9SY
Phone Number: +(44) 01382 383155
Email Address: c.n.a.palmer@dundee.ac.uk
http://medicine.dundee.ac.uk/staff-member/professor-colin-palmer