Present data strongly suggest that the variant ABCG2 c.421C > A (rs2231142) allele increases AUCτ,ss of MPA in stable renal transplant patients (by around 40%, with a high probability that the effect is > 20%) in agreement with proportionally reduced CLT/F,ss. The estimates are consistent based on raw data (patients free of relevant interfering comorbidities and co-medication) and in matched/adjusted analysis, where a number of further potential confounders, “classical” and pharmacogenetic, were controlled for. Considering the latter, we did not account for the SLCO1B3 c.334T > G (rs4149117) and UGT1A9*3 (c.98T > C, rs72551330) SNPs. OATP1B3 mediates MPAG uptake, and variant SLCO1B3 c.334T > G shows around 40% reduced activity in vitro [33]. We identified 4 studies (two in European patients [31, 33], and one each in Chinese [34] and Japanese [35] patients) reporting crude mean±SD dose-adjusted MPA AUCτ,ss in TT/TG vs. GG patients on IR MMF co-treated with CsA (3 cohorts) or macrolactam immunosuppressants (3 cohorts): pooled TT/TG vs. GG differences in the co-treatment subgroups (consistently) and overall suggested a slight tendency of higher exposure (by some 10–15%) in TT/TG subjects (see Figure S2). The most compelling individual study findings were those [31] suggesting by around 24% higher (crude) AUC in 56 TT/TG vs. 111 GG patients co-treated with CsA, and around 18% higher AUC in 54 TT/TG vs. 107 GG patients co-treated with macrolactams. The UGT1A9 c.98T > C SNP results in reduced enzyme activity in vitro [36]. We identified 3 studies (European patients) [32, 33, 37] reporting crude AUCτ,ss in TC vs. TT patients on IR MMF co-treated with CsA (2 cohorts) or with macrolactams (3 cohorts): pooled TC vs. TT differences consistently suggested a mild tendency of higher (by ∼10%) exposure in TC subjects (see Figure S3). The most compelling individual study findings were those [32] reporting around 50% higher AUC (time-averaged estimate of 6 measurements over 1 year) in 5 TC vs. 170 TT patients co-treated with CsA and in 5 TC vs. 158 TT patients co-treated with tacrolimus. The present sensitivity analysis (Fig. 3) demonstrates: even with a marked simultaneous imbalance between ABCG2 c.421C > A variant and wt patients regarding both TT/TG (SLCO1B3) and TC (UGT1A9) genotypes and assuming their maximum reported effects, bias-adjusted estimate of the ABCG2 c.421C > A variant allele effect would still be > 1.25 (i.e., above the conventional upper limit of equivalent exposure). However, it is not very likely that the present estimate was biased by these two SNPs to such an extent: (i) all the reported values were crude, unadjusted values; ii) there is no biologically plausible reason to expect such a huge simultaneous imbalance in prevalence of the two genotypes between ABCG2 variant and wt subjects; (iii) UGT1A9 c.98T > C SNP is rare, and a reasonably expected number of TC subjects in the present sample is 2–3; (iv) population pharmacokinetic models in French [38] and Chinese patients [39] found no association between these two SNPs and MPA clearance. Also, it does not seem likely that other enzyme/transporter SNPs could explain the present observations. Three UGT1A9 promoter SNPs [beyond − 275T > A (rs6714486) and − 2152C > T (rs17868320) that we controlled for] are associated with increased UGT1A9 levels in the liver: − 440C > T (rs2741045), -331T > C (rs2741046) and − 665C > T (rs10176426) [4, 5, 40]. However, studies have failed to provide consistent signals about association of any of these SNPs and exposure to MPA; moreover, rs6714486 and rs17868320 are in complete LD with these SNPs and form two haplotypes (UGT1A9*1l and *1n) [40]. Therefore, by controlling for rs6714486 and rs17868320, one controls also for several SNPs that were not directly genotyped. No consistent signal of association with MPA exposure has been found for several other UGT1A9 SNPs (rs6731242, rs13418420, rs3832043, rs2741049, rs13418420, rs17868323) [4, 5, 39, 41]. Moreover, rs6714486 and rs17868320 are in LD with some of them (haplotypes UGT1A91v and *1w) [40]. Apart from UGT2B7 802C > T (rs7439366), here “represented” by rs7668258 (since in complete LD), studies have consistently failed to yield a clear, reproducible signal of association of any other UGT2B7 SNP and exposure to MPA. The same applies for a number of evaluated UGT1A1, 1A7 and 1A8 SNPs [4, 5, 39, 41]. In the present analysis, we evaluated the effect of one of the ABCG2 polymorphisms (rs2231142). Reduced transporter function has been reported associated with three further SNPs (rs34783571, rs192169062 and rs34264773), for three SNPs no effect on function is reported and for the rest functional consequences are unknown [8]. The estimated global cumulative minor allele prevalence of all “reduced function” SNPs is 0.68%, and for combined “unknown” and “reduced” it is 1.3% [8] – this implies that at most one of the present patients should be reasonably expected to carry any of these SNPs, and it is highly unlikely that this possibility affected the present estimates. Similarly, the three ABCB1 (linked) SNPs controlled for are by far the most prevalent (among Caucasians) coding ABCB1 variants. Cumulative prevalence of other six coding ABCB1 SNPs in Caucasians is around 10% [42], suggesting that at most 6–7 patients in the current sample might have harbored any of those SNPs. In order to be accountable to any relevant part of the present observations, all such (hypothetical) SNPs should have had marked and synergistic effects – not a likely scenario: as recently reviewed [43], most of them have no practical relevance in drug pharmacokinetics. The same is applicable to the ABCC2 SNPs (beyond those controlled for in the present study) and a wide range of investigated ABCC1 and ABCC3 SNPs [43]. Specifically, in respect to MPA, apart from ABCC2 1249G > A (rs2273697) and − 24C > T (rs71762) controlled for in the present analysis, studies have consistently failed to identify a relevant signal of association between MPA exposure and a range of investigated ABCC2 SNPs (rs3740066, rs8187710, rs1885301, rs7910642, rs113646094, rs8187694, rs17222723, rs3740066, rs2804402) and ABCC3 SNPs (rs4793665, rs2277624) [4, 5, 39, 41]. Finally, (apart from the SLCO1B1 c.521T > C, in LD with c.388A > G, controlled in the present study, and already discussed SLCO1B3 c334T > G), no consistent signal of association between a range of SLCO1B1 and 1B3 SNPs and MPA exposure has been detected across numerous individual studies [4, 5, 39, 41]. To attribute the observed effect to these unmeasured but unlikely confounders, one needs to assume their simultaneous synergistic effects. The present sensitivity analysis suggests: even if it existed, and even if really marked (GMR = 1.60), such a (hypothetical) cumulative confounding effect would not completely explain away the observed effect since GMR for the variant ABCG2 c421C > A allele vs. wild type would still be 1.20. Overall, it is justified to state that present data reasonably validly document an effect of the ABCG2 c.421C > A variant allele on steady-state exposure to MPA in renal transplant patients. Discrepancy between the present results and earlier studies not detecting associations between exposure to MPA and ABCG2 c.421C > A SNP might, at least in part, be due to methodological differences. A study in Chinese patients co-treated with CsA reported slightly higher crude dose-adjusted AUCτ,ss in17 variant carriers than in 20 wt controls (30.9±13.0 vs. 27.7±10.7 mg ⋅ h/L) [44]. Our patients were co-treated with CsA or tacrolimus (and matched for CNI and CNI troughs). Neither CsA nor tacrolimus are ABCG2 substrates, but both are ABCG2 inhibitors, and their inhibitory effect might differ, particularly under c.421 SNP (with reduced transporter numbers) [45–47]. Two larger studies (Chinese [48] and Brazilian [49] patients) reporting no association between the c.421 SNP and MPA measured only trough concentrations while present data refer to AUCτ,ss (note: in the present analysis, dose-adjusted MPA troughs tended to be higher in variant carriers, but variability was high), while a Chinese population pharmacokinetic model included only patients co-treated with tacrolimus [39]. Clearly, it is difficult to directly compare results from observational studies differing in methodology and design, sampling populations and sample sizes, outcomes and control of confounding – each should be evaluated on its own merit. We believe that the present analysis reasonably supports a conclusion that the observed difference in AUCτ,ss between the ABCG2 c.421C > A variant and wt subjects is attributable to the fact of variant allele carriage.
Present study is limited by a modestly sized single-center sample, the fact that MPAG was not measured (as not a part of routine TDM), and, relatedly, by no insight into possible mechanisms of the observed effect. A study in Japanese patients [11] reported higher steady-state MPAG concentrations in 44 c.421C > A variant carriers than in 36 wt controls (median 1540 vs. 1195 mg ⋅ h/L; P = 0.029; corresponds to ROM = 1.29), and suggested involvement of ABCG2 in MPAG-MPA recirculation. Current observations (Fig. 1, Table 5) of closely similar Cmax (at around 2 hours post-dose), but clearly larger AUCτ,ss in variant carriers vs. wt controls indirectly support such a possibility: the difference in AUC is primarily due to differences that occurred between 3 and 12 hours post-dose, which is in agreement with hypothetical differences in MPAG recirculation.