Personalized Medicine in Rheumatoid Arthritis: Miles to Go Before We Sleep

“The woods are lovely, dark and deep,

But I have promises to keep,

And miles to go before I sleep,

And miles to go before I sleep.”

Pulitzer Prize winning poet Robert Frost penned these words in 1922 as part of his classic poem “Stopping by Woods on a Snowy Evening”. Scholars have posited that “the woods” represent temptation and sin, while the “miles to go before I sleep” suggest one’s obligations to a higher being, self, loved ones, or society before the end of one’s life. As with many worthy goals, identification of markers predictive of subsequent treatment responses is a challenging and difficult task, and researchers have many hurdles to overcome in this quest. In this issue, Plant et al. publish results of a genome-wide association study (GWAS) of predictors of anti-TNF treatment in rheumatoid arthritis (RA). Inadequate response occurs in 30–40% of RA patients treated with anti-TNF agents. These drugs are expensive, costing as much as $15,000 to $30,000 US dollars per year. Given that the survival of RA patients has improved over the years, and that 0.5 to 1.0% of the US population is affected by RA, the financial cost to the health care system is substantial.

There are reports of genome-wide association studies (GWAS) for a variety of biological traits and human illnesses (1), including responses to medications for various disease states (reviewed in (2)). Other approaches to identify markers predictive of treatment response in human disease include gene expression in affected tissue, measurement of serum protein expression, metabolites of drugs, and others. The study by Plant et al. addresses an important question: Are there genetic predictors of response to anti-TNF therapy among Caucasian RA patients?

Although relatively small compared to many other published GWAS, this is the largest genome-wide investigation of genetic predictors of anti-TNF response in RA to date, including 1,285 participants with RA. The initial analysis focused on 566 anti-TNF-treated RA patients from the Wellcome Trust Case-Control Consortium. Genetic markers associated with change in 28 joint count disease activity score (DAS28) after 6 months of treatment were genotyped in two independent replication cohorts (n = 379 and n = 341) and a combined analysis was performed. All participants in the study were from the Biologics in Rheumatoid Arthritis Genetics and Genomics Study Syndicate (BRAGGSS). Patients were not eligible for this study if they had stopped treatment during the first 6 months for reasons other than inefficacy (presumably toxicity, infection, or other adverse events; or logistic reasons such as difficulties with travel, etc.).

A GWAS was performed using an array of ~500,000 SNPs. Because of the comprehensive, unbiased approach afforded by the genome-wide SNP panel, both positive and negative results are informative. Of 171 genotyped markers demonstrating association with treatment response in the initial GWAS, seven were corroborated in the combined subsequent analyses. These seven SNPs, which when added into predictive models, substantially increase the variance in treatment response accounted for by clinical variables alone.

Despite the strengths of this well-done study, there are weaknesses, which point out some of the difficulties in identifying markers of treatment response to anti-TNF drugs in RA. First, no single SNP consistently replicates at a convincing level of genome-wide significance, with no single SNP reaching P<0.01 in either of their two replication samples. As the authors point out, they chose a relatively lenient threshold for statistical significance at the initial GWAS stage in order to avoid false negatives, but such a strategy increases the likelihood of identifying false positive associations. Of 10 SNPs that showed association in the initial analysis, 3 were not able to be genotyped in subsequent analyses, leaving 7 for full analysis. In the meta-analysis of all 3 cohorts, the strength of the association compared to that in the initial cohort increased for only 3 of the 7 markers (rs12081765, rs17301249 and rs7305646). The strength of association diminished for the remaining four SNPs, which is what would be expected if chance alone were responsible for the original GWAS result. In addition, for 3 of the 4 remaining SNPs (rs4694890, rs1350948 and rs7962316), the effect was in the opposite direction in the stage three cohort, meaning that the allele associated with good response in the initial and stage two cohort was associated with poor response in the stage three cohort, suggesting a spurious result. Finally, applying an aggregate risk score for the 7 SNPs in the stage 3 cohort demonstrated no predictive value, which is unexpected if all SNPs truly contribute to anti-TNF responsiveness.

Genetic variants previously reported to be associated with anti-TNF response in RA were not tested in this analysis. SNPs in the target gene, TNF, have been attractive candidates for pharmacogenetics analyses of TNF inhibitors in RA. There have been conflicting reports of the effect of the TNF-308 SNP on anti-TNF response in RA, including several meta-analyses (3;4). The largest of these meta-analyses failed to demonstrate an association between this polymorphism and treatment response to TNF inhibitors in RA (5). Lui et al. recently reported preliminary associations of SNPs in or near MAFB (chromosome 20); the type I interferon gene IFNk (chromosome 9); and PON1 (chromosome 7); as well as a weak association with a SNP in the IL10 promoter (rs1800896) previously reported to be associated with anti-TNF response (6). In a large study of multiple groups of anti-TNF treated RA patients, Cui et al. reported that a SNP in PTPRC was associated with treatment response (7).

The most important conclusion from this study is that there are no common alleles with large effect size that predict change in DAS28 score following treatment with anti-TNF therapy in RA patients. This is in contrast to genetic factors of RA risk, where common alleles within the MHC region contribute substantially to heritability. This is also in contrast to other pharmacogenetic studies, where common alleles exert large effects on response and toxicity phenotypes [e.g., warfarin metabolism (8), lumiracoxib-related liver injury (9), and ribavirin-induced hemolytic anemia in chronic hepatitis C (10)].

What are the possible explanations for a negative study? First, it is possible that the phenotype of change is DAS28 is not heritable. No studies have been conducted to show that this trait is under genetic control. It is possible that components of the DAS28 (e.g., inflammatory markers) or more extreme phenotypes (e.g., EULAR responder vs non-responder categories) have less heterogeneity and are more heritable. Similarly, it is possible that genetic control of treatment response is different for individual drugs within the class of anti-TNF agents, and that combining the drugs into a single class (as was done here) introduces noise. Second, the phenotype might be heritable, but there are a large number of common alleles with small to moderate effects that influence treatment response. Such a “polygenic” architecture has been observed for quantitative traits such as height (11) and categorical traits such as schizophrenia (12). Larger sample sizes and new statistical methodologies are required to test this hypothesis. One potentially powerful approach is the integration of pathways and gene expression to identify bridges between GWAS data and candidate genes or pathways (2) (13). Third, the phenotype might be heritable, but under the control of rare variants that are not adequately captured on GWAS arrays. With emerging next-generation sequencing technology, it is now feasible to search for and test rare variants in genome-wide studies.

Variability in allele frequencies among different racial/ethnic groups affects power in association studies. The study by Plant et al. was restricted to individuals of European ancestry. While there are currently no data to suggest differential treatment response to anti-TNF agents among different racial/ethnic groups, it is possible that there are common alleles of large effect size that influence response to anti-TNF therapy in other ethnic populations. In patients with advanced non-small-cell lung cancer, for example, genetic variants in the epidermal growth factor receptor (EGFR) influence treatment response to the EGFR tyrosine kinase inhibitor gefitinib. A multicenter, randomized, Phase 3 trial of first-line gefitinib for Japanese patients with advanced non-small-cell lung cancer selected on the basis of EGFR mutations showed improved progression-free survival, with acceptable toxicity, of gefitinib compared to standard chemotherapy (14).

There is a great need for large-scale collaboration to create publically available, large databases of treatment response data and biospecimens to allow discovery, validation, and replication of pharmacogenetic markers of treatment response in RA. Unprecedented collaborative approaches to Alzheimer’s disease have led to great progress in our understanding of biomarkers associated with progression of that disease (15). Similar advances have been made in the understanding of RA, systemic lupus erythematosus, and other autoimmune diseases through large collaborative efforts such as the North American Rheumatoid Arthritis Consortium (NARAC) (16;17); the International Consortium on the Genetics of Systemic Lupus Erythematosus Genetics Consortium (SleGen) (https://slegen.phs.wfubmc.edu/public/index.cfm); the Multiple Autoimmune Disease Genetics Consortium (MADGC) (http://www.madgc.org/madgc/) and the Autoimmune Biomarkers Collaborative Network (ABCoN) (6).

The Treatment Efficacy and Toxicity in Rheumatoid Arthritis Database (TETRAD), a project funded jointly by NIH (5RC2 AR058964) and the Arthritis Foundation, is a collaborative effort of 10 academic medical centers performing a pilot study of prospective data collection and formation of a biorepository of serum/plasma, DNA, and RNA from patients starting biologic agents or methotrexate for RA. For such an effort to be allow investigators performing cutting-edge research to perform well-powered studies of markers of treatment response, many more patients need to be enrolled.

In addition, creative, less expensive ways of identifying patients according to treatment response are needed. Approaches to identify RA patients using natural language processing of electronic medical records data are currently being validated (18). The Electronic Medical Records and Genomics (eMERGE) Network is an NIH-organized and funded consortium of medical research institutions with a stated goal of developing, disseminating, and applying approaches to research that combine DNA biorepositories with electronic medical record (EMR) systems for large-scale, high-throughput genetic research (19). Other NIH initiatives that will either directly or indirectly impact on the quest for markers to allow individualized therapy for RA include the NIH PharmacoGenomics Research Network (PGRN), which will advance the goals of personalized medicine for many diseases (3), and the NIH Patient-Reported Outcomes Measurement Information System (PROMIS) (20), which together may facilitate identification of genetic markers of treatment responses as defined by patients. Such initiatives have been likened to building a highway for personalized medicine (21).

In conclusion, the pursuit of markers of treatment response to biologic agents in RA is a difficult, but worthwhile, task. With persistence, large-scale collaborative efforts, and continued improvements in defining outcomes, genotyping and statistical methods, the goal stratifying RA patients according to likelihood of response to particular medications may be ultimately achieved. Once identified, human genetics should provide an improved understanding of pathogenetic pathways that influence why some patients respond to anti-TNF therapy and others do not. Real progress will become evident when clinically beneficial new therapeutic approaches are incorporated into clinical practice (21). Although we have miles to go before we sleep, progress is being made toward the goal of personalized medicine in RA.