The therapy works by harvesting a patient’s own T cells and engineering them to express a synthetic receptor that recognizes CD19, a surface protein found on B cells. Nearly all approved CD19 CAR-T products use a binding domain derived from the FMC63 antibody, which recognizes an epitope encoded by exons 3 and 4 of the CD19 gene. When cancer cells lose or alter the CD19 antigen, the engineered T cells have nothing to latch onto, and the malignancy escapes immune destruction. Previous work has shown that a splice variant called CD19Δexon2, in which exon 2 is skipped during RNA processing, is linked to therapy resistance and relapse. What has been missing is a reliable way to identify, before treatment ever begins, which patients carry this risk.

The research team, led by Søren Helweg Dam, Giorgia Moranzoni and Lars Rønn Olsen of the University of Copenhagen, together with colleagues at the Technical University of Denmark, Rigshospitalet and Odense University Hospital, took a computational approach. They mined publicly available RNA-sequencing data from four B-ALL patients treated with CD19 CAR-T therapy—one complete responder, one non-responder, and two who relapsed after an initial response. For each patient, they quantified the percent spliced in (PSI) of CD19 exon 2, a metric that estimates how much of the CD19 messenger RNA retains exon 2 versus skipping it. A lower PSI means a higher abundance of the CD19Δexon2 variant.

The pattern was striking. The patient who relapsed earliest—just 1.5 months after infusion—had the lowest pre-treatment exon 2 PSI, with 21 RNA reads spanning the exon 1–3 junction, a direct signature of exon 2 skipping. The non-responder showed 13 such reads, the slow relapser only one, and the complete responder none at all. After treatment, CD19 read coverage dropped sharply in all patients who initially responded, consistent with effective B-cell depletion, while the non-responder retained substantial CD19 expression and continued exon 1–3 junction usage. Although the cohort is far too small for statistical inference, the qualitative concordance between exon 2 inclusion and clinical outcome suggests that reduced exon 2 retention may be associated with CAR-T therapy failure.

To understand why skipping exon 2 might undermine the therapy, the researchers turned to structural modeling. They retrieved the experimentally resolved structure of canonical CD19 from the AlphaFold Database and predicted the structure of the Δexon2 variant de novo using AlphaFold3. In the canonical protein, the two discontinuous segments of the FMC63 epitope fold into close spatial proximity, forming a contiguous binding surface. Exon 2 skipping disrupts this arrangement, separating the epitope elements and likely impairing antibody accessibility. The finding provides a plausible mechanistic explanation for the experimentally observed reduction in FMC63 binding affinity to the variant protein, positioning exon 2 as a stabilizing element of the functional epitope.

Experimental studies add weight to this model. CD19 isoforms lacking exon 2 exhibit defective folding, abnormal glycosylation, and impaired interaction with the CD81 chaperone complex, causing the misfolded protein to be retained in the endoplasmic reticulum rather than trafficked to the cell surface. Only a small fraction of the variant escapes quality control and reaches the membrane, and that population appears unstable and poorly recognized by CD19-directed antibodies. Reduced CAR recognition of CD19Δexon2 therefore arises from both impaired trafficking and altered surface conformation, leading to markedly diminished antigen density.

Perhaps the most provocative finding is that exon 2 skipping is not exclusively a therapy-induced phenomenon. The team analyzed larger RNA-sequencing datasets from treatment-naïve B-ALL (75 samples from the iCOPE/STAGING study), newly diagnosed DLBCL (553 samples from the GOYA trial), and healthy tissues from the GTEx project (1,048 blood and 255 spleen samples). Exon 2 inclusion was uniformly high in B-ALL and healthy tissues, with median PSI values of 96.3 percent, 96.9 percent and 97.7 percent respectively, and roughly 96 to 97 percent of samples in these groups showed near-complete exon retention. DLBCL, by contrast, displayed marked heterogeneity: the median PSI was 91.6 percent, only 59.5 percent of samples met the 90 percent inclusion threshold, and 86.8 percent showed measurable skipping, including one patient with a PSI as low as 22.8 percent.

These results suggest that CD19Δexon2 is a naturally occurring, low-frequency splice variant present in both malignant and normal B cells, rather than a strictly disease-specific or therapy-induced transcript. Under the selective pressure of CD19-directed CAR-T therapy, malignant cells already expressing the variant may gain a survival advantage and expand, producing the apparent enrichment of the isoform in relapsed disease. This reframes the biology: the variant is not created by the treatment but unmasked by it, which is precisely what makes pre-treatment detection potentially valuable.

The translational implications are considerable. Because bulk RNA sequencing of leukemic cells is increasingly performed as part of the diagnostic workup for B-ALL, exon 2 PSI could be computed as an additional feature at no extra experimental cost. Alternatively, quantitative PCR assays targeting the exon 1–3 junction could provide a rapid and inexpensive test to stratify relapse risk or monitor splicing changes during follow-up. The epitope connection may extend even further: the parental antibody of blinatumomab, a bispecific T-cell engager now moving into front-line B-ALL therapy, competes with FMC63 for CD19 binding, suggesting that structural disruption caused by exon 2 skipping could also compromise blinatumomab-mediated cytotoxicity—though this remains to be formally tested.

The authors are careful to emphasize the limitations. The CAR-T cohort contains only four patients, precluding statistical analysis and any claim of causality or predictive value. Post-treatment samples had sparse read coverage, bulk RNA-seq cannot resolve whether the variant is co-expressed with canonical CD19 in the same cells or confined to distinct subclones, and pre-treatment tumor burden may confound PSI measurements, since normal B cells express predominantly canonical CD19. Protein-level consequences remain to be verified experimentally, and the structural analyses rely on computational modeling. Still, the consistency between structural predictions, patient-level RNA patterns and baseline tissue distributions strengthens the biological plausibility of the findings, and larger prospective studies using targeted long-read sequencing or isoform-specific qPCR could determine whether CD19 exon 2 skipping is merely a correlate of poor response or a functional determinant of resistance. Either outcome would sharpen the understanding of CD19 biology and help clinicians decide which patients might benefit most from multi-target CAR strategies directed against CD19 together with CD20 or CD22.

Subject of Research: CD19 exon 2 alternative splicing as a potential prognostic biomarker for relapse after anti-CD19 CAR-T cell therapy in B-cell malignancies

Article Title: CD19 exon 2 skipping is a potential prognostic correlate of anti-CD19 CAR-T therapy relapse

Article References: Dam, S. H., Moranzoni, G., Høie, M. H., Modvig, S., Wadt, K. A. W., Als-Nielsen, B., Schmiegelow, K., Vitting-Seerup, K., Barnkob, M. B., & Olsen, L. R. (2026). CD19 exon 2 skipping is a potential prognostic correlate of anti-CD19 CAR-T therapy relapse. Frontiers in Molecular Medicine, 6, Article 1763390. https://doi.org/10.3389/fmmed.2026.1763390

Image Credits: AI Generated

DOI: 10.3389/fmmed.2026.1763390

Keywords: CAR-T therapy, CD19, alternative splicing, exon 2 skipping, B-cell acute lymphoblastic leukemia, DLBCL, relapse, biomarker, antigen escape, FMC63 epitope, RNA sequencing, immunotherapy resistance