RNA Splicing Errors in Drug Transporter Genes May Predict Cancer Survival Across 33 Tumor Types

Every gene tells a story, but the way that story is edited can determine whether a patient lives or dies. In one of the most sweeping investigations of its kind, researchers in China have traced how alternative splicing—the molecular cutting-and-pasting of RNA transcripts—in a single family of genes shapes cancer outcomes across 33 tumor types. Their findings, published in Clinical Cancer Bulletin, suggest that the splice patterns of ATP-binding cassette (ABC) transporter genes could serve as powerful prognostic markers and, in at least two cases, as causally implicated players in cancer biology.

Alternative splicing is the process by which a single gene produces multiple distinct RNA molecules. After a gene is transcribed into precursor messenger RNA, segments called introns are removed and the remaining exons are stitched together. Depending on which exons are retained, skipped, or joined at alternative boundaries, the same gene can yield proteins with very different functions. Seven major categories of these events are recognized, including exon skipping, intron retention, alternate promoter and terminator usage, and mutually exclusive exon selection. When this editing machinery goes awry in cancer, the consequences can be profound: aberrant splicing has been linked to suppressed apoptosis, enhanced drug resistance, altered DNA damage repair, and uncontrolled cell cycle progression.

The ABC transporter family is among the largest and oldest gene families in the human genome, organized into subfamilies A through G. These proteins act as efflux pumps, using the energy of ATP hydrolysis to move molecules across cell membranes. Their duties range from nutrient uptake and cellular detoxification to lipid balance, signal transduction, antiviral defense, and antigen presentation. In oncology, they are infamous for a darker role: overactive transporters such as ABCB1, ABCC1, and ABCG2 can pump chemotherapy drugs out of tumor cells, producing the multidrug resistance that frustrates treatment. A nonsense mutation in ABCB1 that truncates the P-glycoprotein pump, for example, renders tumors newly sensitive to anticancer drugs, underscoring how central these molecules are to therapeutic response.

To probe how splicing reshapes this family, the team, led by Yidan Zhang and supervised by Xiao Zhu of Guangdong Medical University, first compiled a curated set of 114 ABC transporter genes from the MSigDB, DGIdb, and GeneCards databases. They then mined the TCGA SpliceSeq resource, identifying a staggering 10,617 alternative splicing events across 33 cancer types, drawn from 736 normal control samples and 9,881 patient samples. Using univariate Cox regression, they found that 44 splicing signals from 29 genes were potentially tied to overall survival. LASSO regression, a technique that shrinks weak predictors to zero, narrowed the field, and multivariate Cox analysis ultimately pinned down 14 splicing signals from 11 genes—including PSMA4, PSMD7, UBA52, and PSMF1—that were significantly associated with pan-cancer prognosis.

The prognostic power of these signals was substantial. When patients were stratified by a risk score built from the model, the high-risk group showed dramatically lower overall survival than the low-risk group. The researchers then integrated clinical variables from 1,299 TCGA patients with complete records, confirming that age, tumor grade, and TNM staging were independent prognostic factors alongside the splicing-derived risk score. A nomogram combining these variables predicted one-, three-, five-, and ten-year survival with areas under the curve of 0.695, 0.746, 0.756, and 0.756 respectively—moderate accuracy that the authors themselves caution requires external validation in independent cohorts before clinical deployment.

Perhaps the most striking findings emerged when the splicing signature was overlaid on the tumor immune microenvironment. Using the ESTIMATE algorithm and single-sample gene set enrichment analysis, the team showed that patients in the high-risk group carried heavier immune infiltration, with twelve immune phenotypes positively correlated with risk scores. Checkpoint genes central to modern immunotherapy—CTLA4, CD274 (the gene encoding PD-L1), PDCD1, HAVCR2, and LAG3—were all expressed at higher levels in high-risk patients. Two genes stood out in deeper analysis: PSMA4 and PSMD7. PSMD7, a component of the 19S proteasome, showed that lower expression triggers the activation of potent antitumor immune cells, while higher PSMA4 expression correlated with increased naive CD4 T cells and follicular helper T cells. TIMER database analysis revealed that PSMA4 expression correlated with CD8-positive T cell infiltration in 20 tumor types, and both genes’ expression significantly co-varied with CD274 and CTLA4 across most cancers—hinting that splicing of these transporter-linked genes could help predict which patients will respond to checkpoint inhibitors.

Behind every splicing event stands a splicing factor, the protein that decides where the RNA scissors cut. By mapping the expression of 390 known splicing factors against the prognostic splicing events, the researchers constructed a regulatory network in Cytoscape. It revealed 17 significant splicing events—11 upregulated and 6 downregulated—each controlled by one or more of 50 splicing factors. Notably, some factors exerted dual regulatory effects, with the same factor influencing multiple events and single events being governed by several factors simultaneously. These hub nodes represent potential intervention points: if a splicing factor drives a harmful splice variant, inhibiting that factor could theoretically restore normal transcript production.

Correlation, however, is not causation, and this is where the study makes its boldest methodological move. Mendelian randomization uses genetic variants as natural experiments: because alleles segregate randomly at conception, SNPs associated with a gene’s expression can serve as instruments to test whether that gene genuinely influences disease risk, largely free of confounding. Drawing on eQTL data from the IEU Open GWAS database and outcome data covering 17,254 samples across 33 cancers, the team applied inverse variance weighting, weighted median, and MR-Egger methods, with rigorous sensitivity checks including heterogeneity testing and leave-one-out analysis. The verdict: UBA52 emerged as a protective factor, with its SNPs (rs10414427, rs9908158, rs139767434) associated with reduced risk across all 33 cancers, while ABCB4 acted as a risk factor, its SNPs (rs80351204, rs45493392, rs9275406) linked to increased disease occurrence.

To harden these conclusions, the researchers turned to Bayesian Weighted Mendelian randomization, a technique designed to handle weak polygenic effects and pleiotropy—the situation where a genetic variant influences multiple traits—by downweighting outlier instruments through Bayesian inference. The BWMR results corroborated the standard analysis, with UBA52 (P = 0.038) confirmed as protective and ABCB4 (P = 0.002) confirmed as a risk factor for pan-cancer development. The evidence-of-convergence plots showed rapid stabilization, and posterior weight analysis flagged one observation as potentially confounded, which the Bayesian framework appropriately discounted. Together, the two causal inference approaches elevate UBA52 and ABCB4 from statistical associations to genetically supported candidates for therapeutic targeting.

The authors are candid about limitations. Pan-cancer analyses can mask cancer-type-specific heterogeneity, and the protective or harmful effects of individual splicing events may vary in magnitude or even direction across tumor types. The TCGA and GWAS datasets are dominated by populations not representative of global genetic diversity, and the team calls for multi-ethnic validation in Asian and African cohorts, as well as proteomic confirmation of the functional consequences of the identified splice variants. They also note that PSI values—the quantitative measure of splicing—carry inherent measurement error, and that alternative splicing is shaped by environmental and epigenetic factors no genetic instrument can capture. Still, the convergence of transcriptomic modeling, immune profiling, network biology, and dual causal inference methods marks this study as a template for how splicing biology might finally earn its place in the oncology clinic. If validated, a simple readout of RNA editing patterns in drug transporter genes could one day tell oncologists not only how long a patient is likely to survive, but which immunotherapies their tumor’s molecular editing has primed them to receive.

Subject of Research: Alternative splicing of ABC transporter genes as a pan-cancer prognostic marker and therapeutic target

Article Title: Multi-omics and Mendelian randomization reveal ABC transporter alternative splicing as a pan-cancer prognostic marker and therapeutic target

Article References: Zhang, Y., Wu, J., Lin, Y., Diao, Z., Zhang, X., Yu, L., Cao, Z., & Zhu, X. (2025). Multi-omics and Mendelian randomization reveal ABC transporter alternative splicing as a pan-cancer prognostic marker and therapeutic target. Clinical Cancer Bulletin, 4(1), Article 20. https://doi.org/10.1007/s44272-025-00049-9

Image Credits: AI Generated

DOI: 10.1007/s44272-025-00049-9

Keywords: ABC transporters, alternative splicing, pan-cancer, Mendelian randomization, Bayesian weighted Mendelian randomization, TCGA, tumor microenvironment, immune checkpoints, UBA52, ABCB4, splicing factors, prognostic biomarkers