NEW YORK (GenomeWeb) – Transcriptomic data may help predict how patients respond to treatment with statins, according to a new study from Children's Hospital Oakland Research Institute researchers appearing in Genome Biology.
By drawing on expression data from more than 370 participants taking part in the Cholesterol and Pharmacogenomics (CAP) trial, the team led by CHORI's Ronald Krauss found some 100 genes whose expression levels differ between people who are high and low responders to statins. These differences, the team noted, could explain some 12 percent of the variation in response.
Folding in genomic data such as SNPs or eQTLs, Krauss and his colleagues added, could enable them to explain a slightly greater percentage of the variation in statin response.
"We propose novel integrated prediction models to investigate inter-individual variation in statin efficacy using a comprehensive method that combines transcriptomic and genetic information," Krauss and his colleagues wrote in their paper. "With this approach, we explain a substantial percentage of variation in LDL-[cholesterol] response to simvastatin treatment."
Variation in response to statins, which are commonly prescribed to lower patients' cholesterol levels, has been credited to factors such as age and smoking status as well as to genetic factors. Because of statins' potential side effects like muscle pain and liver damage, knowing how people will respond could guide how clinicians prescribe the drugs.
Krauss and his colleagues drew on transcriptomic data gathered from immortalized lymphoblastoid cells that were generated for 372 Caucasian, non-smoking participants of the CAP simvastatin trial whose response to drug treatment had been recorded. The researchers measured gene expression levels in the cells using the Illumina Human-Ref8v3 bead array.
Using non-negative matrix factorization, the researchers split the participants into high- or low-responder groups and designed an algorithm to uncover genes that were differentially expressed between the high and low responders. Of the top 100 genes that differed between the groups, two-thirds were more highly expressed in the high responders while a third were more highly expressed in the low responders.
Two of these genes, Krauss and his colleagues said, are known to be involved in cholesterol metabolism — CYP51A1 encodes a protein that acts as a catalyst during cholesterol biosynthesis, and NFYC encodes a subunit of the NFY complex that acts as a transcription factor that interacts with separate transcription factors involved in cholesterol synthesis and uptake.
A further 48 of the genes were enriched among the genes correlated with the HMGCR gene, which encodes the rate-limiting step of the cholesterol biosynthesis pathway.
Krauss and his colleagues also developed a series of radial-basis support vector machine prediction models to gauge how well this set of expressed genes could predict response to statins. Through this, they found that the 100-gene signature could predict extreme responders with greater power than it could predict response across the full study population. They further calculated that this model could predict 12.3 percent of the variation in statin response.
They also tested whether the addition of eQTLs or SNPs into the model could improve its predicting prowess.
Through drawing on five publicly available datasets, the researchers identified 36 eQTLs to add to the 100-gene signature model. By doing that, the performance of the model improved slightly, the researchers noted, as it could now explain 13.5 percent of the variance in change of LDLC.
Similarly, they examined how the addition of seven SNPs previously linked through GWAS studies to LDLC reduction would affect the 100-gene signature model, finding it could explain 13.8 percent of the variance observed.
Taken together, the gene signature, eQTLs, and SNPs combined into one model could explain 15 percent of the variance in LDLC response, the researchers said.
By replicating their results in a subset of the CAP population, the researchers found that the model had better predictive power for the very high and low responders,
"As genotypic, eQTLs, and phenotypic datasets grow, our approach can provide a promising framework for identifying novel genes, SNPs, and pathways involved in drug response," Krauss and his colleagues added.