Introduction

Lamotrigine is a commonly used broad-spectrum antiepileptic drug (AED) known for a considerable inter-subject variability in systemic exposure due to variable total body clearance [1,2,3,4,5,6], resulting in a rather wide range of recommended trough concentration in therapeutic drug monitoring (TDM) [3, 5]. It is cleared almost exclusively by hepatic uridine diphosphate glucuronosyltransferases (UGTs), predominantly UGT1A4 with a contribution of UGT2B7 (possible contribution of UGT1A3 and/or UGT1A2 has also been suggested) [2, 3], while ~ 10% is excreted unchanged via kidneys [1, 6]. Consequently, UGT inducers (several antiretrovirals, classical AEDs, and estrogens/gestagens) reduce exposure to lamotrigine up to 40–50%, while valproate (commonly used with lamotrigine) inhibits UGTs [7], reduces clearance by 50–60%, and increases exposure to lamotrigine by approximately twofold [1, 8]. This is reflected in dosing recommendations in co-treated (inducers, valproate) patients [1]. Other “classical” factors also contribute somewhat to variability in lamotrigine clearance [1, 2, 4, 6]: (i) it is reduced in moderate-severe liver failure and moderately decreases with older age and advanced renal failure; (ii) it increases in pregnancy and slightly with increasing body weight; (iii) over the initial 2–3 weeks of treatment, lamotrigine mildly induces its own glucuronidation [1, 6, 7]. Accounting for UGT inducer or valproate use, age and body weight reduce the inter-subject coefficient of variation (%CV) of lamotrigine clearance from 90% to around 45–50%—a still high inter-individual variability [9]. Lamotrigine is a substrate for efflux transporter proteins P-glycoprotein (ABCB1) and breast cancer resistance protein (ABCG2). Limited and equivocal data suggest [2, 10] that single nucleotide polymorphisms (SNPs) ABCB1 1236C > T (rs1128503), 2677G > T/A (rs2032582) and 3435C > T (rs1045642) [in a strong linkage disequilibrium (LD) [10]], and ABCG2 c.421C > T (rs2231142) might affect systemic lamotrigine exposure. However, the main pharmacogenetic “targets” in attempts to understand the variability of lamotrigine clearance are UGT1A4 and UGT2B7 polymorphisms [2]. Both genes are highly polymorphic [11]. The most consistent findings pertain to UGT1A4*3 c.142 T > G (Leu48Val, rs2011425): (i) in vitro, the 48Val variant displays increased glucuronidation (tamoxifen as a probe) [12]; (ii) in vivo, several studies indicated associations between the variant allele/variant homozygosity (GG) and lower exposure and less clinical effect of lamotrigine [2]. Some studies, however, failed to demonstrate such an association (Japanese [13] or Danish patients [14]). Of the UGT2B7 SNPs, most of the (rather limited) in vivo human data pertain to UGT2B7 -161C > T (rs7668258) and UGT2B7 802C > T (rs7439366) [2]. In human liver tissue, UGT2B7 -161C > T is associated with reduced enzyme content and overall reduced glucuronidation capacity [15]. The two SNPs are in a complete LD [16, 17], and a few smaller studies suggested a mildly reduced lamotrigine clearance in heterozygous/variant homozygous subjects [2]. A recent larger study in Danish patients suggested around 9% higher dose-adjusted lamotrigine troughs in the UGT2B7 802C > T variant than in wild-type homozygotes [14], while a study in Mexican patients suggested no relevant association between either of the SNPs and lamotrigine troughs [18]. The apparent inconsistencies could be due to a variety of factors, e.g., ethnic specificities, study designs, sample size, control of confounding, and assessed outcomes. Moreover, considering the large number of SNPs in each of the two genes, attempts to evaluate relevance of any single one of them for bioavailability of lamotrigine might seem meaningless if one does not account (“control”) for all of the others. Obviously, such an effort would require studies including tens of thousands of subjects that are unlikely to ever happen. However, both rs7668258 and rs2011425 are in complete LD with many other SNPs in the respective genes. UGT2B7 -161C > T (rs7668258) is in a complete LD with numerous other UGT2B7 promoter polymorphisms forming two major haplotypes [16] and with a number of other SNPs, and participates in several haplotypes [11]—UGT2B7*1a, *1j, *1 k, *2b, *2c, *2d, *2f. Similarly, UGT1A4*3 c.142 T > G (rs2011425) is in a complete LD with several promoter SNPs, e.g., -219C > T and -163G > A (rs3732219 and rs3732218) to form the UGT1A4*3a haplotype, but also with -419 and -463, and with several other SNPs (form haplotypes *5 and *7a) [11, 12, 19, 20]. Also, at least in Caucasians, rs2011425 is in a complete LD with UGT1A4*2 c.70C > A (rs6755571, Pro24Tre) [21, 22] which in vitro is associated with a reduced enzyme activity [12, 23], but reports about its association with lamotrigine troughs have been ambiguous (e.g., in Scandinavian subjects [14, 24]). Hence, by identification of heterozygous or variant homozygous UGT2B7 -161C > T or UGT1A4*3 c.142 T > G genotype, one identifies subjects with “broader” genetic makeups that differ from that in their respective wild-type (wt) homozygous controls. Elements of these makeups may or may not be related to lamotrigine exposure, and it might not be possible to untangle their individual contributions. Consequently, by contrasting subjects heterozygous or variant homozygous at UGT2B7 -161C > T or UGT1A4*3 c.142 T > G to their wt peers, one may not be able to estimate the effects of these specific polymorphisms, but could still estimate the effects of the respective “broader makeups” represented by these genotypes. In this context, we aimed to estimate the effect of UGT1A4*3 c.142 T > G and of UGT2B7 -161C > T heterozygous/variant homozygous genotypes (i.e., related “broader makeups”) on (dose-adjusted) lamotrigine troughs in adult and adolescent epilepsy patients of Central-Eastern European descent.

Patients and methods

Study outline

Otherwise generally healthy patients on lamotrigine or on combined lamotrigine + valproate therapy undergoing routine TDM after at least 3 weeks of (co-)treatment were genotyped for UGT2B7 -161C > T (rs7668258) and UGT1A4*3 c.142 T > G (rs2011425), and also for two efflux transporter SNPs—ABCG2 c.421C > A (rs2231142) (classified as wt or variant carriers, since only 1.0% of patients were variant homozygous) and ABCB1 1236C > T (rs1128503). Patients were also classified with respect to exposure to valproate as (i) valproate trough = 0 (patients on lamotrigine monotreatment) or below the lower limit of quantification (BLOQ) (20.8 µmol/L); (ii) low valproate, i.e., 0/BLOQ < valproate trough < 364 µmol/L (median of the quantified values, and approximate lower limit of recommended valproate troughs [5]); and (iii) target/high valproate (≥ 364 µmol/L). The study concept was as follows: (i) heterozygous or variant homozygous subjects are considered to differ from the respective wt controls not only regarding the determined genotype, but regarding a “broader makeup” consisting of linked polymorphisms; (ii) these “broader makeups” have no other means of affecting exposure to lamotrigine but by affecting the (respective) UGT enzyme activity; (iii) however, whether or not enzyme activity is affected is of no interest—the outcome of interest are lamotrigine troughs, and “enzyme activity” is considered an unobserved true exposure represented by an instrumental variable, i.e., the UGT2B7 -161C > T or UGT1A4*3 c.142 C > T genotype. To estimate the effects of UGT2B7 -161C > T (i.e., the associated broader makeup), in the entire sample (main effects) we emulated a randomized experiment in which “treated” were heterozygous (CT) and variant homozygous subjects (TT), whereas wt subjects were controls. To estimate the main effects of UGT1A4*3 c.142 T > G (i.e., the associated broader makeup), we emulated a trial in which “treatment” was variant allele carriage (TG or GG; since there were < 1% variant homozygotes) and wt patients were controls. Finally, we emulated two trials to test potential moderation of the polymorphism effects by exposure to valproate, i.e., the genotype*valproate interaction: “treated” were variant carriers (CT/TT in the case of rs7668258, or TG/GG in the case of rs2011425) and controls were their wt peers, and differences were estimated at valproate 0/BLOQ and at valproate > 0/BLOQ. Although cross-sectional, we deemed data as appropriate for the purpose: (i) the presumed cause (genotype/associated broader makeup) preceded the outcome (lamotrigine troughs); (ii) it was plausible to assume no reverse causation, i.e., no effect of the outcome on “treatment”—samples were taken after the initial lamotrigine self-induction had been completed [25]; (iii) it was plausible to assume also no effect of outcome on other possible causes, i.e., confounders/outcome ancestors. This is primarily of interest in the sense of no effect of lamotrigine on valproate levels. Since valproate is partly eliminated by UGTs [26], it has been suggested that valproate-lamotrigine interaction could be bi-directional [27], considering the initial UGT induction by lamotrigine. However, present samples were taken after this process had been completed, and individual and population pharmacokinetic studies have refuted the (hypothetical) effects of lamotrigine on valproate clearance [28, 29]. The same reasoning applies to the lack of effect of the outcome on other UGT enzymes or transporters.

We used inclusion/exclusion criteria and covariate entropy balancing to control for the effects of confounders/outcome ancestors (Table 1) (details in Supplementary Information—Methods to achieve conditional exchangeability [Fig S1, Fig S2]). Since it was reasonable to expect residual confounding (Table 1), the estimated effects were subjected to analysis of sensitivity to unmeasured confounding.

Table 1 Confounders [may affect both the outcome (dose-adjusted lamotrigine troughs) and exposure—activity of UGT enzyme whose polymorphism (UGT2B7 -161 C > T or UGT1A4*3 142 T > G) is used as an instrumental variable] and outcome ancestors (may affect the outcome) considered in an attempt to achieve conditional exchangeability between “treated” and “controls” (see Supplementary Information 1: Methods to achieve conditional exchangeability, with directed acyclic graphs [Fig S1, Fig S2])
Full size table

The study was conducted in line with the Declaration of Helsinki (the 2008 version) and was approved by the Institutional Ethics Committee.

Patients

Consecutive epilepsy patients on lamotrigine (immediate-release tablets) or on combined lamotrigine + valproate (extended-release tablets) regimen with gradual dose titration as per approved labels, scheduled for routine TDM after at least 21 days of (co-)treatment provided blood samples for determination of morning (07:00–09:00 h) lamotrigine/valproate troughs. From initiation of the monotherapy or from initiation of the combined treatment (addition of valproate to lamotrigine, or, less commonly, lamotrigine to pre-existing valproate), patients were seen in 2-week intervals, and at a pre-TDM interview to assess (by self-report) tolerability, treatment compliance and possible violation of the inclusion/exclusion criteria. They were included in the study if: (i) willing to donate blood samples and provided signed informed consent for genotyping of pharmacogenes; (ii) aged ≥ 16 years; (iii) non-smokers or ex-smokers; (iv) not using other AEDs or other drugs known to affect lamotrigine or valproate, and/or activity of UGTs, P-glycoprotein, or ABCG2 within the previous month; (v) had preserved cardiac, renal, and liver function, based on routine assessment. Patients suffering unregulated diabetes mellitus, hypo- or hyperthyroidism, those with a history of or an ongoing malignant disease or any acute illness, pregnant women, and patients with HIV/AIDS were not included.

Bioanalytical methods and genotyping

Plasma lamotrigine was measured using a validated high-performance liquid chromatography with a diode-array detector (Shimadzu, Japan), as described previously [30], while serum valproate was measured by an immunoassay (PETINIA) on a Dimension Expand analyzer (Siemens; calibrator and control samples by Siemens, Germany). Both analytes are included in external quality control schemes (DGKL RfB and UK NEQAS).

Genomic DNA was extracted from 2 mL of whole blood using the FlexiGene DNA Kit (Qiagen, Hilden, Germany), according to the manufacturer’s instructions. Genotyping of MDR1/ABCB1 1236C > T, ABCG2 421C > A, and UGT2B7 -161C > T was performed using TaqMan Drug Metabolism Genotyping assays ID C_7586662_10, ID C_15854163_70, and ID C_27827970_40, respectively, while genotyping of UGT1A4*3 c.142 T > G was performed using Custom TaqMan SNP Genotyping assay (Applied Biosystems, Foster City, CA, USA) by real-time polymerase chain reaction (PCR) genotyping method on the 7500 Real-Time PCR System (Applied Biosystems, Foster City, CA, USA), according to the manufacturer’s instructions. Genotyping of UGT1A4*3 c.142 T > G was confirmed by a PCR–RFLP method on the Gene Amp PCR System 9700 (Applied Biosystems, Foster City, CA, USA) [31].

Weighting and data analysis

To achieve a balance between “treated” and “controls” on measured covariates, we used entropy balancing [32] implemented in package WeightIt [33] in R [34] with average treatment effect (ATE) as the estimand. Entropy balancing is a form of distance matching: the procedure assigns weights under given enforced restrictions on distance between treated and controls (that is, the distance between moments of covariates), taking into account the estimand [35]. To estimate the main effects, balancing was undertaken in the entire sample; to test the genotype*valproate interaction, “treated” and “controls” were balanced separately at each level of exposure to valproate. We used generalized frequentist (robust variance estimator) and Bayesian weighted models to analyze (ln-transformed) dose-adjusted lamotrigine troughs with geometric means ratios (GMRs) as effect measures. In Bayesian analysis, we defined moderate-strength skeptical normal prior for the polymorphism effect [normal (0.0, 0.355)] compatible with the a priori hypothesis of no treatment effect. In models testing the interaction, we additionally defined a moderate-strength normal prior for the effect of valproate [normal (0.693, 0.40)] in line with the expected twice higher, on average, exposure to lamotrigine with valproate co-treatment. We used SAS 9.4 for Windows (SAS Inc., Cary, NC) and R package rstanarm [36]. We used CubeX [37] to evaluate Hardy–Weinberg equilibrium and linkage disequilibrium.

Sensitivity to unmeasured confounding/bias

We considered that bias arising from unmeasured confounders was primarily due to (hypothetical) effects of UGT2B7 and UGT1A4 SNPs that were not accounted for, i.e., the “remaining” genetic makeups besides those consisting of the evaluated SNPs and their linked polymorphisms. We assumed that this hypothetical bias might have “pushed” the observed GMRs to > 1.0 or to < 1.0 with a “moderate” (i.e., 1.25 or 0.80, respectively) or a “strong” effect (i.e., 1.43 or 0.70, respectively): GMRs 1.25/0.80 correspond to standard upper and lower limits of equivalent exposure, while GMRs 1.43/0.70 are their “extended” values applicable to compounds showing high variability, i.e., inter-subject %CV of 50% (corresponds to the “inherent” variability in lamotrigine clearance, after adjusting for age, body weight, and concomitant use of UGT inducers or valproate) [9]. According to the present (incomplete) knowledge, practically all UGT1A4 polymorphisms with a prevalence of around 10–15% (“common”) are in LD with UGT1A4*3 c.142 T > G [11, 19], whereas cumulative prevalence of all other SNPs is around 5–10%. Similarly, the most common (known) UGT2B7 haplotypes/haplotype pairs include UGT2B7 -161C > T [11, 17], while cumulative prevalence of haplotype pairs not including this SNP may be approximated at around 15% [17]. We (conservatively) assumed that the prevalence of these genetic constellations that we did not account for in the present sample was 25% for UGT1A4 and 25% for UGT2B7 (regardless of whether they were considered as “competing instrument” or as “outcome ancestor”). Since their occurrence is independent, the probability of their joint occurrence is 6.25%; hence, we stayed with a more unfavorable scenario with prevalence of 25%. Finally, we assumed that this total prevalence resulted from a marked imbalance between “treated” and “control” subjects of 2:1 and 4:1. Hence, we corrected the observed estimates of the “treatment” effect for unobserved confounding effect [38] of GMR 1.25 and 1.43 (and their reciprocal values), assuming 2:1 and 4:1 imbalance of a biasing set of covariates between “treated” and “controls” assuming its total prevalence of 25% (R package episensr [39]) (see also Supplementary Information – Sensitivity of GMR to unmeasured confounding).

Results

Patients

We included 471 patients, 143 (30.4%) co-treated with valproate and 328 on lamotrigine monotherapy (Table 2). Three co-treated patients had valproate troughs BLOQ; hence, 331 (70.2%) patients had valproate 0/BLOQ, while “low” and “target/high” valproate were seen in 70 patients each (Table 2). Regarding UGT2B7 -161C > T, 50% of the patients were heterozygotes, while wt and variant homozygotes were comparably prevalent (Table 2). Only 4 (0.8%) patients were UGT1A4*3 c.142 T > G variant homozygous, and wt subjects prevailed (77.5%) (Table 2). Variant homozygotes were also sporadic regarding ABCG2 c.421C > A (Table 2). Patient subsets based on UGT2B7 -161C > T and on UGT1A4*3 c.142 T > C genotypes numerically differed with respect to a number of characteristics; however, dose-adjusted lamotrigine troughs apparently only mildly differed across the respective subsets (Table 2).

Table 2 Subject characteristics (raw data) overall and by UGT2B7 -161C > T and UGT1A4*3 c.142 T > G polymorphisms (with standardized differences [d] for balancing variables and the outcome). Data are count (%), median (range), mean ± SD (range), and geometric (geo) mean (%CV) for ln(lamotrigine [LAM]/dose)
Full size table

There were no departures from the Hardy–Weinberg equilibrium for any SNP, and no indication of LD between the ABCG2 and UGT2B7 loci (long arm chromosome 4) (D’ = 0.239, r2 = 0.0068, Chi2 = 3.2).

Balanced/weighted data

In the overall sample, all treated (UGT2B7 -161 CT or TT, or UGT1A4*3 c.142 TG/GG genotype) and respective wt control patients (CC and TT genotypes, respectively) were well balanced (Supplementary Information—Table S1 summarizes information on weights) on all covariates (d = 0.000) and their dose-adjusted lamotrigine troughs were closely similar (Table 3). All comparisons (main effects) yielded GMRs close to 1.0 with CI/CrI within the conventional range of equivalent exposure (Fig. 1A). For both polymorphisms, variant allele carriers (CT/TT or TG/GG) were well balanced on all covariates vs. their respective wt controls at valproate 0/BLOQ and at valproate > 0/BLOQ (Table 4). Dose-adjusted lamotrigine troughs were (expectedly) considerably higher with valproate > 0/BLOQ than with valproate 0/BLOQ (Table 4), and for both polymorphisms, variant carriers and wt controls had closely similar values at both valproate levels (Table 4). All GMRs (variant carriers vs. wt controls) were close to 1.0 (Fig. 2A) while some CIs/CrIs were wide (exceeded the conventional limits of equivalence) (Fig. 2A) due to high inter-subject variability and a limited number of subjects in some of the valproate-by-polymorphism subsets. Overlapping distributions of GMRs (variant carriers vs. wt controls) estimated at the two levels of exposure to valproate (Fig. 2B) illustrate lack of polymorphism*valproate interaction.

Table 3 Subject characteristics by UGT2B7 -161C > T and UGT1A4*3 c.142 T > G genotypes after balancing/weighting. “Treated” are UGT2B7 -161C > T heterozygous or variant homozygous patients and UGT1A4*3 c.142 T > G variant allele carriers (TG/GG) (only 4 patients were variant homozygous), and controls are their respective wild type (wt) subjects. Data are weighted counts (percent), mean ± SD, or geometric mean (%CV) for lamotrigine (LAM) dose-adjusted troughs (on ln-transformed data). Shown are also standardized differences (d) for balancing variables (maximum d for any pairwise comparison) and for the outcome
Full size table
Fig. 1
figure 1

A Differences [as geometric means ratios (GMR) with 95% and 90% confidence/credible intervals] in dose-adjusted lamotrigine troughs between patients heterozygous/variant homozygous at UGT2B7 -161C > T or at UGT1A4*3 c.142 T > G and their respective wild-type (wt) controls: overall (“main effects”) and at different levels of valproate exposure [valproate trough 0 or below the limit of quantification (BLOQ) and valproate trough > 0/BLOQ]. Vertical gray lines indicate GMRs 0.90 and 1.11, a range within which typically GMR point-estimates fall under equivalent exposure; vertical black lines indicate GMRs 0.80 and 1.25, a conventional acceptance range for the 90% CIs around point estimates for a claim of equivalent exposure. B Frequentist sampling distributions (left) and Bayesian posterior distributions (right) (we simulated 40,000 distributions for each) of GMRs for variant allele carriers (i.e., UGT2B7 -161 CT/TT or UGT1A4 c.142 TG/GG) vs. respective wt controls estimated at valproate 0/BLOQ and at valproate > 0/BLOQ. Vertical dashed lines indicate GMR point estimates. The general overlap of estimated effect distributions illustrates their close similarity at both levels of exposure to valproate for both polymorphisms

Full size image
Table 4 Subject characteristics by UGT2B7 -161C > T and UGT1A4*3 c.142 T > G genotypes before and after covariate entropy balancing, separately at different levels of exposure to valproate [valproate 0 or below the limit of quantification (BLOQ); valproate > 0/BLOQ] for evaluation of the effect of variant carriage (CT/TT or TG/GG vs. respective wild type) at different exposure to valproate. Data are (weighted) counts (percent), mean ± SD, or geometric mean (%CV) for lamotrigine (LAM) dose-adjusted troughs (on ln-transformed data) (outcome). Shown are also standardized differences (d) for balancing variables and for the outcome
Full size table
Fig. 2
figure 2

Sensitivity analysis—shown are observed (main) effects (point-estimate geometric means ratios, GMRs) corrected for bias due to unmeasured confounding. We assumed that a set of unmeasured covariates (“biasing set”) had an effect on dose-adjusted lamotrigine troughs and that it could have either increased them or reduced them. We further assumed that the total prevalence of such a set in the current sample was 25%, but with imbalance between “treated” (in the case of UGT2B7 -161C > T polymorphism, treated are either CT or TT subjects; in the case of UGT1A4*3 c.142 T > G, treated are TG/GG subjects) and “control” subjects (CC and TT, respectively) of 2:1 or 4:1 (see Sensitivity to unmeasured confounding for details). A In the case of UGT2B7 polymorphism, previous reports suggested that CT or TT genotypes were associated with higher lamotrigine levels. Hence, it is assumed that the observed GMRs for CT vs. CC subjects (GMR = 1.00) and for TT vs. CC subjects (GMR = 0.97) are due to a biasing effect of unmeasured confounders that “pushed” GMR to < 1.0, and was moderate (GMR = 0.80) or strong (0.70). B In the case of UGT1A4 polymorphism, previous reports suggested that variant allele was associated with lower lamotrigine troughs. Hence, it is assumed that the observed GMR for TG/GG vs. GG subjects (GMR = 0.95) is due to a biasing effect that “pushed” GMR towards 1.0 (i.e., towards > 1.0) and was moderate (GMR = 1.25) or strong (1.43)

Full size image

Sensitivity to unmeasured confounding

Based on previous reports, variant UGT2B7 -161C > T allele should be expected associated with higher exposure to lamotrigine. We hence assumed that the observed GMRs of 1.00 (CT vs. CC) and 0.97 (TT vs. CC) were due to the effect of confounding bias that “pushed” the “true” GMR towards ≤ 1.0 (Fig. 2A): however, even assuming a considerable imbalance in the prevalence of the “biasing” covariates and their moderate (0.80) or strong (0.70) effect, the bias-corrected estimates did not suggest any relevant effect of this polymorphism on dose-adjusted lamotrigine troughs (Fig. 2A). On the other hand, considering previous reports, variant UGT1A4*3 c.142 T > G allele should be expected associated with lower exposure to lamotrigine. We hence assumed that the observed GMR of 0.95 (TG/GG vs. TT) was due to the effect of confounders that “increased” the “true” GMR towards ≥ 1.0 (Fig. 2B): however, even under a huge assumed imbalance in prevalence of the biasing covariates (60% vs. 15%) and with a marked biasing effect (1.43) “corrected” GMR estimate (GMR = 0.804) still did not cross the limit of what is generally considered “a practically relevant difference” (i.e., outside the limits of “equivalent exposure”) (Fig. 2B).

Discussion

Polymorphisms in genes encoding UGT1A4 and UTG2B7—considered the main enzymes in lamotrigine metabolism—have been commonly evaluated in attempts to elucidate sources of inter-individual variability in lamotrigine clearance. The largest body of evidence pertains to UGT2B7 -161C > T (rs7668258) and UGT1A4*3 c.142 T > G (rs2011425), both of which are in vitro associated with altered enzyme activity [12, 15]. In vivo data, however, are equivocal: some studies reported associations between heterozygosity (CT)/variant homozygosity (TT) at UGT2B7 -161C > T with mildly increased lamotrigine levels, and some reported associations between the variant allele at UGT1A4*3 c.142 T > G (TG/GG) and reduced lamotrigine concentrations—but several studies reported no association of either polymorphism with exposure to lamotrigine (reviewed in [2], exemplified in, e.g., [13, 14, 18, 24]). As in any complex setting investigated using observational data, these somewhat inconsistent reports might be due to any one or more of several reasons, e.g., ethnicity-related specifics, sample size, outcome measures, bioanalytical methods, and control of confounding. The present analysis included adult Caucasian epilepsy patients of Central-Eastern European descent and used dose-adjusted lamotrigine troughs obtained through routine TDM as an outcome. We a priori accepted the fact that it was impossible to assess specific relationships between either of the two SNPs and the outcome due to their complete LD with many other polymorphisms within the respective genes, i.e., that genotypes at the two loci were parts of broader “genetic makeups” whose actual “composition” remained unknown (we did not determine genotypes at other respective polymorphisms and, currently, not all linkages among numerous SNPs in UGT2B7 and UGT1A4 genes might be known). Finally, we a priori acknowledged that many polymorphisms were likely not linked to two genotyped polymorphisms and could have been (reasonable) sources of bias. Otherwise, we accounted for a range of classical and (pharmaco)genetic factors known or suspected to affect exposure to lamotrigine by combining inclusion/exclusion criteria and “statistical” adjustment. For the latter, we used a method (covariate entropy balancing) that is model-independent and more appropriate for a given setting than a “standard” regression analysis. For example, in a UGT1A4*3 c.142 T > G TG/GG vs. wt control comparison, considered covariates formed a total of 108 strata (3 × 3 × 3 × 2 × 2), with a further need for adjustment for age and body weight. For a regression model to yield a reasonably accurate “adjusted” estimate of a difference, i.e., one that is not dependent on model extrapolations that might be considerably astray, each stratum would need to contain at least a few “treated” and a few “controls”—which in the present case would not be possible, since there were 106 TG/GG patients—and in each stratum values of age and body weight between “treated” and “controls” would need to at least partly overlap.

Under these circumstances, all observed GMRs (main effects)—for UGT2B7 -161C > T CT or TT vs. wt controls (CC) and for UGT1A4*3 c.142 T > G TG/GG vs. wt controls (TT)—were closely around 1.0 with CIs/CrIs within the classical limits of equivalent exposure. Even GMRs (point-estimates) corrected for a hypothetical considerable biasing effect of unmeasured confounders with (unrealistically) high imbalance between “treated” and “controls” did not signal any practically relevant effect. We assigned this (hypothetical) biasing effect primarily to unmeasured variables pertaining to other potential SNPs in the UGT2B7 and UGT1A4 genes that so far have not been suggested related to exposure to lamotrigine, nor shown linked to the two genotyped SNPs, although it could be viewed as a result of any number of biasing factors. However, based on the current knowledge, those factors that could be identified have likely not contributed to this hypothetical bias. For example, we adjusted for the loss-of-function SNP in the ABCG2 gene (ABCG2 c.421C > A, rs2231142) that apparently moderately affects lamotrigine troughs [40], and for which global minor allele prevalence has been estimated at around 12% [41]. Reduced transporter function has been reported associated with three further ABCG2 SNPs (rs34783571, rs192169062, and rs34264773), for three SNPs no effect on function has been reported, and for the rest, functional consequences are unknown [41]. The cumulative estimated prevalence of combined other (besides rs2231142) “loss-of-function” and “unknown effect on function” SNPs is around 1.0% [41]. This suggests that it would be reasonable to expect at most 5 patients in the current sample bearing any of these “other” SNPs—hence, it is highly unlikely that these (undetermined) SNPs have biased the present results. Similar reasoning is applicable to SNPs in the ABCB1 gene, as well. We adjusted for the ABCB1 1236 T > C (rs1128503) polymorphism which is in a strong LD [10] with two further common coding SNPs -2677 T > G/A (rs2032582) and 3435 T > C (rs104564). In a sample of renal transplant patients from the same general population as in the present study, we recently also observed almost complete LD among these three SNPs [42]. Hence, by controlling for the rs1128503 genotype, one largely controls for the other two SNPs. In Caucasians, these three SNPs are the most prevalent ones and are the most commonly evaluated among numerous ABCB1 SNPs with respect to bioavailability of a range of drugs, but with extremely variable outcomes disabling any consensus [43]. In respect to lamotrigine, several studies tested involvement of individual SNPs or of the haplotype [with T/G/T having higher lamotrigine concentration than C/G(A)/C] in lamotrigine pharmacokinetics [2], but the most recent larger study in Scandinavian patients [44] found no signal that would relate 1236 T > C or 3435 T > C to dose-adjusted lamotrigine troughs. Cumulative prevalence of other six coding ABCB1 SNPs in Caucasians is around 10% [43], suggesting that in the “worst case scenario” at most 50 patients in the current sample might have harbored any of those SNPs. Even if one were to assume that each of them “worked in the same direction” regarding exposure to lamotrigine, and that there was an unrealistically huge imbalance in their simultaneous prevalence between “treated” and “controls,” and their considerable effect, these “other” SNPs could not have relevantly biased the present estimates. Finally, a recent comprehensive systematic review [45] identified a number of studies evaluating SNPs in other ABC transporters in relation to pharmacokinetics and response to a variety of drugs—just to find mostly weak or the none and unreproducible associations, suggesting that the impact of these SNPs on drug pharmacokinetics is generally minor (if any) [45], and this appears applicable to lamotrigine, as well. Based on the current knowledge (reviewed in [2]), it is also reasonable to conclude that polymorphisms in the SCL superfamily transporters are highly unlikely to be relevant for exposure to lamotrigine. Therefore, the hypothetical strong bias used in the present analysis to “correct” the observed estimates might have had different sources, albeit it seems reasonable to assign it to UGT2B7 and/or UGT1A4 SNPs that have not been addressed and are not linked (or are not known to be linked) to the two typed polymorphisms.

In addition to the main effects, the present analysis demonstrates closely similar dose-adjusted troughs between variant carriers (UGT2B -161 CT/TT or UGT1A4*3 TG/GG subjects) and their wt peers at each of the two levels of exposure to valproate, i.e., lack of an interaction between genotype and valproate. In this analysis, genotypes used for adjustment and exposure to valproate were dichotomized since, despite the total number of 471 patients, number of subjects in some of the strata formed by multiple 3-level and multiple 2-level factors was very low. Values in CT/TT or TG/GG patients were equivalent to those in CC or TT patients (respectively) at each of the two levels of exposure to valproate, or point-estimates were within the narrow range between 0.90 and 1.11, with CIs/CrIs slightly exceeding the conventional limits of equivalence. In this respect, it should be noted that even with a GMR of 1.0, with 50% CV (this corresponds to %CV in lamotrigine clearance after adjustment for age, body weight, use of UGT inducers and/or valproate) a sample of 96 vs. 44 or of 30 vs. 110 subjects achieves only around 60% power to “place” the 90%CIs/CrIs within the range 0.80–1.25.

Comparing results across observational studies that differ in sampling populations and methodology is not straightforward—it seems more reasonable to assess each individual study for its own merit. We believe that in the present analysis we generated reasonably unbiased estimates to support a view that heterozygosity or variant homozygosity at UGT2B7 -161C > T (rs7668258) or at UGT1A4*3 c.142 T > G (rs2011425)—each representing a “broader genetic makeup” that differs from that represented by the wt genotype—has no relevant consequences for dose-adjusted lamotrigine troughs in adult epilepsy patients. Present estimates were obtained in Caucasian patients of Central-Eastern European descent (Slavic) and might not hold in other populations, e.g., those in which the typed polymorphisms are potentially linked to different other SNPs, or in which prevalence of functionally relevant non-linked SNPs is considerably different.