ABCC2 rs717620 Emerges as the Only Consistent Genetic Signal in Antiseizure Drug Resistance
A new systematic review and meta-analysis places ABCC2 rs717620 at the center of a long-running question in epilepsy pharmacogenetics: why some patients continue to seize despite appropriate treatment. Across 20 studies involving 2627 people with antiseizure medication resistance and 2639 responsive controls, this promoter-region variant showed a reproducible association with resistance across multiple genetic models, while the other two prespecified polymorphisms, rs2273697 and rs3740066, did not hold up in pooled analyses.
The signal was modest in size but persistent. Under the allelic model, rs717620 carried an odds ratio of 1.25 with a 95% confidence interval of 1.06 to 1.48. The dominant, recessive, and homozygous models all pointed in the same direction, with odds ratios of 1.25, 1.63, and 1.79, respectively. Low-to-moderate heterogeneity and no evidence of small-study bias gave the finding more credibility than isolated single-center reports usually allow.
Why a Transporter Variant Can Matter at the Bedside
ABCC2 encodes a multidrug resistance-associated efflux transporter, part of the blood-brain barrier machinery that can push medications back out of the central nervous system. In epilepsy, that biology matters because a drug that never sustains sufficient brain exposure may fail even when the dose looks adequate on paper. The rs717620 variant sits in the 5′ untranslated region, a location that can plausibly alter gene expression rather than protein structure, making it a credible candidate for changing transporter activity and thereby shaping antiseizure response.
The clinical stakes are not abstract. Drug-resistant epilepsy is defined by failure of two appropriately chosen and tolerated regimens to achieve sustained seizure freedom, and the burden remains substantial despite the arrival of newer agents. A transporter-based mechanism offers one explanation for why some patients cycle through multiple therapies with little durable benefit.
One finding cuts through the noise: rs2273697 and rs3740066 did not show a pooled association with resistance.
Inconsistent Signals, Uneven Study Power
The review incorporated 19 studies for rs2273697 and 20 for rs3740066, yet the broader literature around both variants remained unstable. In one German study, the rs2273697 AA genotype appeared protective, with an odds ratio of 0.19 and a 95% confidence interval of 0.05 to 0.77. Elsewhere, different haplotypes or subgroup-specific findings pointed in the opposite direction. That pattern is common in small genetic association studies: a striking result in one cohort can vanish when tested against a larger pooled dataset or a different ancestry group.
Power calculations explain much of the uncertainty. Most studies were underpowered, and the estimated median sample-size increases needed to reach 80% power were steep: 348% for rs2273697, 322% for rs3740066, and 289% for rs717620. In practical terms, many existing cohorts were simply too small to detect anything but large effects, which makes false negatives likely and unstable positives hard to trust.
Haplotype Findings Point to a More Complex Genetic Architecture
Haplotype analysis added another layer of complexity. Twelve of the 20 studies examined multilocus combinations, but not in a uniform way, and only five reported significant associations. Certain haplotypes, including CAC and the H2/H9 diplotype, were linked to better treatment response, while TGT, AC, and TTGC were associated with resistance in selected cohorts. A GCGC multilocus pattern tracked with better carbamazepine response, whereas TTGC aligned with poor response. These combinations matter because transporter behavior rarely hinges on a single nucleotide alone; nearby variants can jointly influence transcription, RNA stability, or linked regulatory effects.
Still, the haplotype literature remained heterogeneous across populations. Some studies found both genotype-level and haplotype-level effects, some found only one or the other, and several found nothing at all. That unevenness argues against immediate clinical deployment of any one panel without broader validation.
What the Cohorts Actually Looked Like
The included populations were clinically mixed rather than narrowly defined. Mean ages ranged from childhood through mid-adulthood, and most studies included both focal and generalized epilepsy. Carbamazepine and valproic acid were the most frequently studied antiseizure medications, followed by phenobarbital, phenytoin, lamotrigine, and oxcarbazepine. Polytherapy dominated most cohorts, often involving two to four antiseizure drugs per patient, which reflects real-world refractory epilepsy more than a simplified trial population.
Geographically, the evidence base was concentrated in Asia, especially China, which accounted for 9 of the 20 studies. The remainder came from a scattered set of countries, including Germany, Japan, Hong Kong, Austria, Malaysia, Croatia, Mexico, Jordan, South Korea, and Egypt. Several studies used alternative resistance definitions, ranging from failure of two or more regimens to seizure-frequency thresholds or inadequate seizure reduction over at least 12 months. Those differences in case definition can easily reshape genetic associations, especially when the phenotype being studied is already biologically heterogeneous.
What This Means for Genetic Testing in Epilepsy
The pooled evidence supports one cautious conclusion: ABCC2 rs717620 is the most credible of the three tested variants for association with antiseizure medication resistance, but it is not ready to function as a stand-alone clinical biomarker. The effect size is real but small, and the underlying studies are too inconsistent in design, ancestry composition, seizure phenotype, and treatment exposure to justify broad use in routine practice.
The value of this meta-analysis lies partly in what it rules out. rs2273697 and rs3740066 do not appear to offer reliable explanatory power across current datasets, despite isolated signals in individual reports. Future work needs larger consortium-based cohorts, harmonized definitions of resistance, and multilocus haplotype analysis that can capture the transporter biology more faithfully than single-SNP testing alone. Without that scale, the field will keep producing signals that are statistically interesting but clinically thin.