Our results have shown a significant association between CYP2C19 phenotype and cessation of leflunomide due to toxicity in a group of patients with RA, most of whom were concurrently receiving other DMARDs.
The CYP2C19*2 loss-of-function allele (carried by poor, intermediate or unknown metabolizers) has been associated with lower teriflunomide concentration, which is thought to be due to reduced conversion of leflunomide to teriflunomide . If the side effects from leflunomide were solely caused by teriflunomide, we would expect intermediate/poor metabolizers to have a lower incidence of toxicity. This was not the case in this study, and suggests either direct toxicity by leflunomide, or an alternate pathway is competing with the conversion of leflunomide to teriflunomide and results in formation of another metabolite which contributes to side effects. The latter is supported by studies of leflunomide in cultured hepatocytes, which indicated that Cytochrome P450 mediated metabolism is a prerequisite for leflunomide induced liver injury, whereas teriflunomide is directly, albeit less significantly toxic . Metabolites of teriflunomide have not been observed in human plasma, but studies with human liver microsomes have shown that teriflunomide can be converted via hydroxylation to a metabolite (M1), whereas leflunomide can be converted to either teriflunomide or a different metabolite (M2, also by hydroxylation) [13, 14]. If the enzymes responsible for formation of M2 are different from the well characterized conversion pathway of leflunomide to teriflunomide, reduced activity of CYP2C19 is likely to result in increased formation of M2, which may be the toxic metabolite contributing to the side effects associated with leflunomide.
While we provide no direct evidence for either the formation of M1 or M2 in vivo or toxicity associated with M2, we have demonstrated that CYP2C19 phenotypes with reduced activity were associated with higher rates of toxicity. Furthermore, since there has been an association between greater reduction in RA disease activity in patients with higher plasma teriflunomide steady state concentration [3, 15, 16], CYP2C19 intermediate/poor metabolizers may have an especially adverse risk:benefit ratio by virtue of a higher risk of cessation with side effects and a reduced likelihood of therapeutic benefit. In contrast, ultra-rapid metabolizers appear to have a particularly favorable risk:benefit ratio. This may be an important factor for clinicians to consider when they are contemplating which DMARD to prescribe their patients with RA. This finding may also explain why there has been no success in identifying a relationship between adverse drug events and teriflunomide concentration.
A number of factors are considered by both physician and patient when deciding whether to cease any medication. A novel feature of our study protocol is that a DMARD is never stopped or switched for reason of inefficacy, and the only reason to cease a DMARD is for toxicity. However, we must acknowledge that both patient and physician perceptions of side effects may be subjective, and we can not exclude the possibility that inefficacy may have influenced withdrawals. To this end, we did not observe any association between response to leflunomide after three months of treatment and CYP2C19 phenotype (data not shown), suggesting that reduced efficacy is unlikely to be responsible for increased drug cessation in CYP2C19 intermediate and poor metabolizers.
van Roon and colleagues have shown that when leflunomide was given primarily as monotherapy to long term RA patients, 29% withdrew due to an adverse event and 13% ceased due to inefficacy after a median follow-up of 317 days . This compares to a withdrawal rate due to adverse events of 42.3% in our study and addition of another DMARD was required in 5.1% of patients. Since our study had a shorter mean duration from diagnosis to introduction of leflunomide (1.4 vs 9.5 years) and more frequent use of combination DMARDs, the greater efficacy and poorer tolerability observed in our cohort is to be expected and entirely consistent with this previous report .
The results of our study are in contrast to previous studies of leflunomide toxicity that have reported no relationship with CYP2C19 genotype, but did demonstrate an association with SNPs in CYP1A2 and DHODH [6, 7]. There are a number of important differences in both study design and the participants included in our study compared to these previous reports (which were presumably performed on the same patient population) which may account for these discrepancies. First, our study was a retrospective cohort design, which has advantages over retrospective case control studies, which are prone to additional selection/recruitment bias. The previous study reported that only 19% of the group who had not ceased leflunomide were taking another DMARD (most of whom were also taking methotrexate), whereas the number who ceased leflunomide while taking other DMARDs was not reported; such a discrepancy could account for the differing rates of toxicity between genotypes, which is supported by our finding of higher cessation rates in individuals who were continuing triple therapy. Despite not accounting for these important variables, the authors adjusted for factors that did not appear to influence toxicity, such as age, body mass index (BMI) and disease duration. Our approach of adjusting only for variables that were seen to influence cessation due to toxicity is more robust. Finally, by comprehensively assessing information about multiple CYP2C19 genotypes, we were able to consider the entirety of the information regarding CYP2C19 activity concurrently by comparing outcomes related to CYP2C19 phenotype instead of each genotype separately. This is an important point of difference with previous studies, since 20 to 25% of individuals who carry a loss (*2) or gain (*17) of function allele are unknown metabolizers and should, therefore, not be considered equivalent to intermediate or ultra-rapid metabolizers respectively. As such, compared to CYP2C19 phenotypes, the effect size observed with individual genotypes is likely to be lower and significant associations are, therefore, more difficult to demonstrate.
The conversion of leflunomide to teriflunomide is mediated by CYP1A2 (in addition to CYP2C19) , and the immunosuppressive properties are mediated by inhibition of DHODH , but there is currently no biologically plausible explanation for the proposed associations of CYP1A2 and DHODH genotype and leflunomide toxicity [6, 7]. The CYP1A2*1F allele occurs in the promoter region of the CYP1A2 gene (in intron 1), and is thought to be important in induction of enzymatic activity in smokers (smokers who carried one or two C residues had higher enzyme activity compared to those who had the AA genotype), but it does not have an effect on enzymatic activity in non-smokers . If this mechanism is responsible for the higher rate of adverse events, one would expect that this SNP would only have an effect in current smokers. Given that the current smoking rate in our cohort (28%) was higher than that reported by Bohanec Grabar et al. (18%)  a more pronounced effect on toxicity would have been expected in our study, but this was not the case. Smokers who carry an AC or CC genotype are likely to be more efficient at converting lefunomide to teriflunomide, but this does not translate to higher teriflunomide concentrations . Furthermore, given the association between CYP2C19 phenotype and toxicity suggests that intolerance may be due to a toxic metabolite, enhanced CYP1A2 activity (and, therefore, enhanced conversion of leflunomide to teriflunomide) is more likely to be protective of side effects. The SNP in the DHODH gene associated with toxicity by Grabar et al. (rs3413422) is associated with an amino acid substitution in the expressed protein (Lys7Gln) , but the effect of this on the activity of the expressed DHODH protein is not known. While there is currently no established biological plausibility to support the association of DHODH genotype and toxicity, this could change if the functional significance of this polymorphism is determined.
Although it was not statistically significant, reduced activity of the ABCG2 drug efflux pump appeared to be protective against diarrhea. This is despite individuals with reduced ABCG2 activity having higher plasma teriflunomide concentrations, which is likely due to less hepatobiliary recycling and fecal elimination . Reduced secretion of teriflunomide into the gastrointestinal tract may be protective against diarrhea through reduced local effect of teriflunomide on the bowel wall.
Our study did have some weaknesses that must be acknowledged, including the retrospective design, the assessment of patient compliance was limited to direct questioning of study participants and one cannot be certain that all adverse events attributed to leflunomide were both drug related and caused by leflunomide (as opposed to one of the other DMARDs the patient was taking concurrently). Also, where a patient ceased leflunomide due to more than one side effect, no attempt was made to ascertain the relative severity of each side effect (and, therefore, the one primarily responsible for cessation). The small patient numbers also did not allow us to make definitive findings regarding the factors associated with events that occurred at low frequency within the cohort, such as the influence of ABCG2 genotype on cessation due to diarrhea.
We are not aware of any previous observations that have linked side effects to any DMARD (including leflunomide) and anti-CCP antibody and rheumatoid factor positivity. However, it is notable that only a proportion of cases that conform to diagnostic or classification criteria for RA express these antibodies . Accordingly, RA may be regarded as a clinical syndrome, in which a substantial proportion of cases have an autoimmune aetiology, as evidenced by the presence of anti-CCP antibodies (and rheumatoid factor with which it is strongly correlated) in serum and synovial fluid and citrullinated peptides within affected joints . Cases that present with the RA syndrome but lack anti-CCP antibodies and rheumatoid factor presumably have other etiologies that, in contrast to sero-positive RA, may be less dependent upon lymphocyte proliferation to sustain a chronic, pathological antigen-specific immune response. The primary mechanism of action of teriflunomide is via inhibition of DHODH, which prevents activated T-lymphocytes from accumulating sufficient pyrimidines to support DNA synthesis, resulting in decreased cell proliferation . Since methotrexate may also compromise the DHODH synthetic pathway , it is conceivable that sero-positive RA is particularly responsive to leflunomide, especially when given as an adjunct to methotrexate therapy. In addition to the overlapping effects of methotrexate and leflunomide on pyrimidine synthesis, sulfasalazine and leflunomide each have inhibitory effects on nuclear factor-κB [22, 23]. These interactive effects suggest that even lower doses of leflunomide than those used in trials hitherto may yield benefit and be better tolerated than current standard doses, especially in sero-positive RA.
Informed by our observations, future research should investigate whether leflunomide toxicity is mediated by a toxic metabolite, and if so, seeking to identify and quantify the putative toxic metabolite in patient plasma as a basis for defining correlations with toxicity. Also, a prospective evaluation of the factors that influence leflunomide toxicity, which also incorporates measurement of teriflunomide plasma concentrations would seem to be warranted. Such a trial would ideally have higher patient numbers through recruitment from multiple centers that utilize structured DMARD regimens. The value of ABCG2 genotype in predicting diarrhea should also be assessed in a larger patient group.