Figure S1A . Directed acyclic graph (DAG) representing the setting to estimate the effect of ABCG2 c.421 SNP, i.e., reduced (vs. preserved) ABCG2 function resulting from variant allele carriage (vs. wt homozygosity) on steady-state pharmacokinetics of MPA (MPA PK – outcome, O). The measured “treatment” – ABCG2 c.421C>A genotype - is an instrument (black circle), since ABCG2 activity (actual exposure) is not measured. The causal path (thick black arrow) might be a direct one and/or mediated (dashed black arrow) through an unmeasured (hypothetical) mediator, i.e., MPAG levels. Pale red circles represent confounders (ancestors of both the treatment i.e., actual exposure, and the outcome), blue circles represent ancestor of the outcome and green circles represent ancestors of (actual, but unmeasured) exposure (ABCG2 activity). Gray arrows depict biasing paths. Gray filled/outlined circles represent unmeasured variables - one is a suggested but unmeasured confounder – SLCO1B3 c.334 SNP, and one is unmeasured ancestor of the outcome- UGT1A9 c.98 SNP; the others indicate transporter/enzyme activities (presumably) affected by exposure/outcome ancestors (see text for details).
4. Number of baseline covariates that can interfere with the (tested)ABCG2 c.421 (ABCG2 activity) effect is high (although Figure S1A is somewhat simplified). In a scenario in which the ABCG2 c.421variant would be “treatment”, none of them would meet the “classical” definition of a confounder, since “treatment” is defined at conception, and the current knowledge about possible epigenetic regulation of ABCG2 is virtually non-existing. In such a case, they would qualify as “ancestors of the outcome” (i.e., factors known or suspected to affect MPA PK, thorough different mechanisms [paths]). As illustrated in Figure S1A, when ABCG2 activity is considered as an actual but unobserved “treatment” (but adequately represented by an instrument), then some of these variables should justifiably be considered ancestors of both the “treatment” (may affect ABCG2 activity) and the outcome (may affect MPA PK, by different mechanisms); 5. Variables that may be considered ancestors of both the “treatment” and the “outcome” (depicted in pale red in Figure S1A) include: i) type of CNI (CsA or tacrolimus). They are both (in vitro ) potent ABCG2 inhibitors, but it is possible that in vivo(at therapeutic doses) they differ in their inhibitory effect –in vitro , CsA is particularly (and more) potent when the number of transporter is reduced [12] (as in the case of the ABCG2c.421 SNP). Next, CsA inhibits ABCC2 (ABCC2 activity is another unmeasured variable in this setting) and affects MPAG/MPA recirculation and exposure, while tacrolimus does not [13]. Also, both CsA and tacrolimus may both inhibit and induce ABCB1 activity (a further unmeasured variable) [14], and may differ in this respect, and MPA is a substrate of ABCB1 [15]. Also, CsA, but not tacrolimus, is listed among SLCO inhibitors [15] – thus, it can affect SLCO1B1 and/or 1B3 activity (further unmeasured variables), and MPAG is a substrate of both [1]; ii) ABCB1 2677/345/1236 SNPs (as diplotypes, since in LD) reflect on ABCB1 activity (not measured), hence they affect MPA (outcome), and also the exposure: both CsA and tacrolimus are also ABCB1 substrates [14], hence altered ABCB1 activity may reflect on their trough concentrations, and this may result in a variable effect on ABCG2 activity (exposure); iii) ABCC2 -24 or/and 1249 SNPs may reflect on ABCC2 activity and MPAG is an ABCC2 substrate. Also, although ABCC2 is not considered relevant in CsA and tacrolimus pharmacokinetic pathways [10], ABCC2 (and -24/1249 SNPs) may affect tacrolimus [16] – hence, affect its concentrations which might reflect on its effect on ABCG2 activity; iv) donor’s ABCC2 1249 SNP might reflect on MPA (presumably, by affecting MPAG in the kidney) [17], and, at least theoretically, on tacrolimus [16] (although renal excretion is of minor relevance for tacrolimus [10]), and thus contribute to the variability of tacrolimus effect on ABCG2 activity; v) UGT2B7 -161 SNP is in a complete LD with the UGT1B7 802 SNP [18], hence it “represents” the 802 SNP. By affecting UGT2B7 activity (not measured), it would affect MPA glucuronidation. On the other hand, one study demonstrated direct glucuronidation of CsA and tacrolimus in human gut and liver by UGT2B7 [19] – hence, it might affect CNI concentrations, and, consequently, their effect on ABCG2 activity; vi) SCLO1B1 521 SNP (and linked SNPs) – may affect SLCO1B1 activity (not measured) and MPAG is a substrate to SLCO1B1. However, SLCO1B1 may also transport tacrolimus [20], hence affect its concentrations and the effect on ABCG2; vii) serum albumin levels and diseases that might interfere with pharmacokinetics of MPA and of CNI – Figure S1A is simplified in that these factors were considered jointly (since also possibly inter-related): hypoalbuminemia is known factor affecting exposure to MPA, and various systemic conditions in the early post-transplant stage can be reasonably considered as factors that could affect both exposure to MPA and CNI levels (and, thus, CNI effects on ABCG2 activity); viii) Food (concomitant) may interfere with absorption of both MPA and CNIs (and their concentrations); ix) Renal function and its commonly measurable “proxy” – estimated creatinine clearance (eCrCl) – may reflect on bioavailability of MPA and of CNI (although, this is a minor pathway for CNIs [10]), hence on both “exposure” and the “outcome”; x) age, body mass index (Figure S1B is simplified in that it combines these two demographic factors and omits all possible interconnections between demographics, concomitant morbidity, liver and renal function) – have been found related to bioavailability of all immunosuppressants (at least to some extent; e.g., by reflecting on renal function, liver function, or in any other way), hence they affect both the “exposure” and the “outcome”. 6. Unobserved (known) confounderSLCO1B3 c.334 SNP (and linked SNPs). In complex pharmacogenetic settings, a number of unmeasured/unknown confounders are possible. SLCO1B3 c.334 SNP would qualify as a known (although not unambiguously) possible confounder which however remained unobserved (patients were not genotyped for this SNP). By (presumably) affecting SLCO1B3 activity it would reflect on MPAG (MPAG is a substrate), but it may also reflect on tacrolimus concentrations [20], and thus on its effect on ABCG2 activity. 7. Ancestors of the outcome (depicted in blue in Figure S1A). A number of covariates qualified as ancestors of the outcome (they can be plausibly related to MPA PK, but not to “exposure”, i.e., ABCG2 activity): i) drugs affecting MPA PK (by effects on UGTs, transporters or by any other mechanism) – Figure S1A is simplified in that it considers all such drugs jointly and omits their relationship to “effector molecules”, e.g., UGT enzymes, ABCB1, ABCC2, SLCO1B1/B3 or others; ii) MPA formulation and MPA dose – IR MMF and EC-MPA formulations are not bioequivalent; they deliver different molar doses and MPA concentration-time profiles are not equivalent [21]. Clearly, choice of formulation (specific molar dose and release particulars) affects bioavailability of MPA; iii) UGT1A9 -2152/-275 SNPs (as diplotypes, since in complete LD) reflect on the enzyme activity (unmeasured), and thus on MPA PK; 8. Unobserved (known) ancestor of the outcomeUGT1A9 c.98T>C : variant allele carriage has been suggested (although with high uncertainty) associated with higher exposure to MPA, but its prevalence is very low. 9. Ancestors of exposure (depicted in Figure S1A in green). Variables that affect exposure and have no effect on the outcome (apart that executed through their effect on exposure) are instrumental variables. When actual exposure is quantified, adjustment for instruments worsens or introduces bias, does not remove it [5-7]. In the present study, actual exposure is unobserved, and we use an instrument (ABCG2 c.421C>A genotype). In such a setting, accounting for other instrumental or near-instrumental variables (affect “exposure”, while effect on the outcome is minor) is needed in order for the instrument to adequately represent the (actual) exposure: i) drugs that interfere with ABCG2 activity (beyond CNIs). Figure S1A is simplified in that it considers all such drugs jointly. It also allows for a possibility that these drugs could (by any mechanism) affect CNI concentrations (which could also reflect on ABCG2 activity); ii) CNI concentration (morning trough at the beginning of the 12-hour MPA sampling period) regardless of the CNI type [ln(tacrolimus) troughs rescaled to ln(CsA troughs) scale by linear transformation]. CNI concentration is a descendant of several variables that may affect it. Putting both “CNI type” and “CNI concentration” into the network is reasonable: in some aspects, tacrolimus and CsA differ qualitatively (e.g., CsA inhibits ABCC2 and SLCO1B1, tacrolimus does not), while the effect on “exposure” (ABCG2 activity) might be concentration-dependent (along with a possibility of a qualitative difference between the two); iii) CNI dose directly affects CNI concentrations; iv) CYP3A4/5SNPs (those genotyped in the present sample and other potentially relevant [10]) may affect CNI concentrations; v) drugs affecting pharmacokinetics of CNIs – Figure S1A is simplified in that it considers all such drugs jointly and omits their mechanisms (e.g., effects on CYPs, transporters or any other “effector”).
Figure S1B contains all the same elements as Figure S1A, but depicts the minimal adjustment set required (and sufficient) to block biasing paths, as to (unbiasedly) estimate the causal effect of treatment (ABCG2 c.421C>A variant allele – reduced transporter function) on the outcome (MPA PK): adjustment (by different means) for outcome/exposure ancestors (but not for colliders) [5-7]. The minimal adjustment set includes (variables depicted in Figure S1B as open, black-outlined circles): i) ABCB1 2677/3435/1236 SNPs considered as diplotypes (since in strong LD; 3 levels based on the number of variant alleles) – by matching; ii) ABCC2 -24 and1249 SNPs and also donor’s ABCC2 1249 SNP, dichotomized as variant carriers and wt subjects – by matching (the latter also by statistical adjustment); iii) UGT2B7 -161 SNP (represents the 802 SNP since in a complete LD), dichotomized as variant allele or wt – by matching; iv) UGT1A9 -2152 and -275 SNPs considered as diplotypes (since in complete LD), dichotomized as variant or wt diplotype – by matching; v) SLCO1B1 521 SNP (may include other SNPs in a complete LD), dichotomized as variant or wt – by matching; vi) Food-drug interaction – by clinical procedure: blood sampling after an overnight fast; vii) MPA formulation (IR MMF or EC-MPA) – by matching; viii) MPA dose – all PK parameters calculated using dose-adjusted MPA concentrations; ix) CNI type (CsA or tacrolimus) – by matching; x) CNI concentration (trough) – by matching; xi) Age, BMI – by matching (latter also by statistical adjustment); xii) Renal function – by inclusion criteria (patients had to have by at least 33% improved creatinine vs. post-operative Day 1 with absolute value <300 µmol/L and stable diuresis);