An Alternatively Spliced Gain-of-Function NT5C2 Isoform Contributes to Chemoresistance in Acute Lymphoblastic Leukemia

Leukemia is the most common cancer in children and adolescents/young adults, accounting for almost one thirds of all pediatric cancers, with the majority of cases classified as B-cell acute lymphoblastic leukemia (B-ALL). Only ∼50% of children with first relapse survive in the long term, and outcomes are appreciably poorer with subsequent relapses. Early bone marrow (BM) relapses occurring within 3 years of diagnosis are particularly difficult to salvage (1). As a result, relapsed or refractory (r/r) high-risk B-ALL cases account for a substantial number of pediatric cancer–related deaths. In addition, although most cases of B-ALL are diagnosed early in life, most deaths from B-ALL (∼80%) occur in adults, largely due to r/r disease. Indeed, adults with B-ALL experience very high relapse rates and long-term event-free survival of less than 50% (2, 3).

What molecular mechanisms are responsible for treatment failures in r/r B-ALL is only partially understood. Previous whole-exome/genome sequencing efforts have identified a dozen of recurrent mutations in relapse-specific targets (4–6). Some of them (e.g., TP53) are known to mediate responses to pan-cancer drugs such as anthracyclines, whereas others are involved in therapeutic responses to specific leukemia therapeutics: glucocorticoids (NR3C1 and WHSC1), antifolates (FPGS), and thiopurines (PRPS1, PRPS2, and NT5C2). The most recent effort to catalog such mutations yielded up to 50 genes significantly enriched at first relapse but often as subclonal or low-recurrence events (7). Even the most recurrently affected drivers are mutated in fewer than 20% of all samples, suggesting a major role for alternative mechanisms of gene dysregulation. One such mechanism is alternative spicing (AS); however, its role in acquired resistance to chemotherapy remains poorly defined.

To bridge this gap, we previously had generated and mined new and existing RNA sequencing (RNA-seq) datasets in search for local splicing variations (LSV) prevalent in B-ALL samples but rare in normal BM cell counterparts. Of note, B-ALL–specific AS was found to affect 15 out of the 20 top leukemia driver genes from the COSMIC database with frequencies far exceeding those of somatic mutations (8). This intersection suggests that AS could be an important mechanism of both leukemogenesis and drug resistance. Here, we provide evidence that a subset of relapses often lacking identifiable acquired mutations has an AS signature simultaneously affecting multiple r/r B-ALL drivers. Among them is a novel NT5C2 exon 6a mRNA variant encoding a gain-of-function nucleotidase isoform with increased enzymatic activity capable of conferring resistance to 6-mercaptopurine (6-MP).

The Modeling Alternative Junction Inclusion Quantification (MAJIQ) algorithm version 4.4 (RRID:SCR_016706; ref. 9) was run, with each sample being quantified individually, on a build comprising 219 TARGET B-ALL samples with de novo junction detection but without intron retention detection, as described previously (10–13). Samples were run against the GENCODE hg38 quantification. For ∆PSI (percent spliced-in) quantification, the MAJIQ PSI and Voila tools were used on 48 diagnostic–relapse paired samples to extract all the LSVs in all the samples with the –show-all flag. Visualization and downstream analyses were conducted in R using the ggplot2 (RRID:SCR_014601), ComplexHeatmap (RRID:SCR_017270), and tidyverse (RRID:SCR_019186) packages. The sashimi plots were generated with the ggsashimi tool. In parallel, raw junction-spanning reads were obtained from the STAR aligner (see Supplementary Materials) and expressed as junction counts per million. Junction counts per millions for select AS events in select samples are included in Supplementary Table S1.

For each target gene, primers corresponding to first and last exons were designed to ensure full coverage of relevant AS events. Typically, 5 ng of cDNA were amplified with LongAmp Taq 2× Master Mix (New England Biolabs) for 25 cycles. The resulting amplicons were subjected to amplicon-seq [#SQK-NBD114.24, Oxford Nanopore Technologies (ONT)] library preparation. Subsequently, each library was loaded into a Spot-ON flow cell version R10 (#FLO-MIN114, ONT) and sequenced in a MinION Mk1C device (ONT) until at least 1,000 reads per sample were obtained. The results were aligned using Minimap2 version 2.24-r1122 (RRID:SCR_018550) and visualized in Integrative Genomics Viewer (RRID:SCR_011793) version 2.12.3.

REH (RRID:CVCL_1650), Nalm-6 (RRID:CVCL_0092), 697 (RRID:CVCL_0079), and MHH-CALL-4 (RRID:CVCL_1410) human B-ALL cell lines originally obtained from commercial repositories were cultured and maintained in conditions recommended by ATCC or Deutsche Sammlung von Mikroorganismen und Zellkulturen (DSMZ). Cell line authentication for REH and MHH-CALL-4 was performed by short tandem repeat profiling. All cell lines were routinely subjected to Sartorius EZ-PCR Mycoplasma Detection Kit testing.

Single-guide RNAs targeting CREBBP exons 25 and 26 and CAS9 protein were obtained from Integrated DNA Technologies (see Supplementary Materials). Cas9 ribonucleoprotein complexes were assembled following manufacturing recommendations. These ribonucleoprotein complexes were transfected into 697 and REH cells via electroporation using the Neon transfection system (Invitrogen Neon Transfection System Model MPK5000) and Neon 10 µL Transfection Kit (#MPK1096, Invitrogen) using the following conditions: 1,700 V, 20 ms, and 2 pulses. Their effects on exon removal were measured by cDNA PCR, with primers amplifying the region between exons 20 and 27 (see Supplementary Materials), and Western blotting.

For FPGS splice switching, a specific Vivo-morpholino (MO) targeting the canonical 3′ splice site of exon 8 was designed (AAA​TTC​CAC​TGG​TCC​GTC​TGA​CCC​C). As a control, the reverse sequence was used in order to maintain sequence composition, as recommended by the manufacturer. Cells were incubated with 2.5 µmol/L MOs for 24 hours before treatment with methotrexate (MTX).

For NT5C2 constructs, the coding sequence of the canonical NT5C2 isoform (NM_001351169.2) was amplified using specific primers (see Supplementary Materials) and cloned into the pLVX backbone (Clontech) modified with the EF1a promoter and the basticidin-resistant gene using NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs). Specific primers and NEBuilder HiFi DNA Assembly Master Mix were used to introduce exon 6a or the R238W mutation into the canonical sequence of NT5C2. For viral particle production, 293T cells (ATCC) grown to 90% confluence in a 10-cm Petri dish were transfected with 10 μg of the NT5C2 constructs, 2.5 μg of the pMD2.G (#12259, Addgene), and 7.5 μg of the psPAX2 (#12260, Addgene) using 60 μL of 25 µmol/L polyethylenimine (PEI) transfection reagent (#24765, Polysciences). Supernatants were collected 48 hours after transfection, passed through 0.45-μm PVDF filters, and incubated for 48 hours with 4 × 10⁶ cells in media containing 10 µg/mL polybrene (#HY-112735, MedChemExpress). After 48 hours, culture media were replaced, and cells were incubated with 10 μg/mL blasticidin for 96 hours.

Recombinant wild-type (WT) and mutant NT5C2 constructs were cloned, expressed, and purified as previously described (14). Briefly, the full-length NT5C2 cDNA constructs with N-terminal hexahistidine (His6) tags were inset into the pET28a-LIC expression vector. Recombinant proteins were expressed in Rosetta 2 (DE3) E. coli cells by induction with 0.5 mmol/L isopropyl-β-D-thiogalactopyranoside overnight at 16°C. Then recombinant proteins were purified using an ÄKTA fast protein liquid chromatography system (GE Healthcare). This was followed by affinity and size-exclusion chromatographies.

5′-Nucleotidase activities of purified WT and mutant NT5C2 proteins were measured in the absence and presence of allosteric activators using a 5′-NT enzymatic test kit (Diazyme) according to the manufacturer’s instructions, as described previously (15). All assays were performed in triplicate and analyzed using a GloMax multidetection system plate reader (Promega).

For treatments with MTX (#S1210, Selleck Chemicals), 6-MP (#852678-5G-A, Sigma Aldrich), mizoribine (#S1384, Selleck Chemicals), doxorubicin (#HY-15142, MedChemExpress), dexamethasone (#D4902, Sigma-Aldrich), and vincristine (#HY-N0488, MedChemExpress), cells were incubated with increasing concentrations of each drug for 72 hours. Vehicle control dilutions corresponding to the highest drug concentration were used as a baseline measure for 100% survival. Cell viability was measured by CellTiter-Glo assays (G7572, Promega) following the manufacturer’s instructions. IC₅₀ values were calculated using Prism software v9.3.1 (GraphPad software) with the log (inhibitor) versus normalized response–variable slope.

Luciferase-transduced REH cells were transduced with NT5C2 WT, E6a, or R238W constructs and selected with blasticidin for 96 hours. Following selection, NOD/SCID/Il2rgtm1wjl/SzJ (NSG) mice were intravenously injected with 1 × 10⁶ parental or modified REH cells and assessed for leukemia engraftment using the Xenogen In Vivo Imaging System. Ten days after B-ALL engraftment was detected, cohorts of mice (n = 5/group) of each REH cell type were randomized for treatment with vehicle control or 6-MP (100 mg/kg/day). 6-MP was administered intraperitoneally daily × 5 days, as described previously (16). Radiance as a surrogate for total leukemia burden was measured by in vivo bioluminescent imaging of mice following i.p. injection of luciferin substrate. Bioluminescent imaging assessment was performed with Living Image software (PerkinElmer) with statistical analysis and data display performed using Prism (GraphPad).

Main RNA-seq datasets analyzed in this study were parts of the TARGet alL phase I and II projects (https://portal.gdc.cancer.gov/projects) and the BLUEPRINT consortium (http://dcc.blueprint-epigenome.eu). TARGET datasets accessed through dbGaP were phs000463.v23.p8 ALL Pilot Phase 1 and phs000464.v23.p8 ALL Expansion Phase 2. The raw long-read sequencing data generated in this study are publicly available in the Sequence Read Archive as Project PRJNA1138448. All other raw data generated in this study are available upon request from the corresponding author.

To gain new insights into molecular mechanisms that render B-ALL chemotherapy ineffective, we analyzed 48 TARGET patients with available paired diagnosis–relapse (D-R) RNA-seq datasets. We called mutations in known r/r driver genes from the RNA-seq data using Genome Analysis Toolkit (see Supplementary Materials). In agreement with prior analyses, we observed that in 12 out of 48 D-R pairs (25%), there were no identifiable relapse-specific mutations in 12 common r/r genes (Fig. 1A). To determine whether differential gene expression could account for the chemoresistant phenotype, we performed principal component analysis on all 96 samples. We observed no separation of diagnostic and relapse samples (Supplementary Fig. S1A, red and cyan dots, respectively), suggesting that differential gene expression as a whole is not a primary driver of B-ALL relapses. However, pervasive AS reported in our earlier B-ALL study (8) could contribute to chemoresistance by altering expression levels of distinct mRNA isoforms. Thus, we asked whether any members of the HNRNP and SRSF/TRA2 superfamilies of splicing factors (SF) strongly implicated in exon inclusion/skipping are affected by mutations or copy number alterations (CNA). Using cBioPortal (17), we identified such CNAs in 14 out of 48 patients (30%). For example, deep deletions of both SRSF4 and SRSF10 genes was found in patient PAPZST and the SRSF3 gene in samples PARAKF and PARJSR (Fig. 1B). Of note, these deletions affected only the latest among longitudinal samples, attesting to their possible role in B-ALL relapses.

To explore the extent and functional consequences of splicing dysregulation, we ran the MAJIQ 2.1 splicing algorithm on paired diagnostic/relapse samples while filtering for LSVs with ≥20% difference in junction inclusion with 95% confidence (Supplementary Fig. S1B). We further filtered for samples without known driver mutations and for LSVs affecting these unmutated driver genes. After performing hierarchical clustering, we discovered a cluster of five relapses with concurrent mis-splicing of multiple chemoresistance genes (Supplementary Fig. S1C, light blue rectangle). To elucidate the underlying molecular mechanisms, we performed gene set enrichment analysis (18) on all transcripts differentially expressed in D-R pairs in that cluster of patients. Top 3 and 7 out of top 12 categories pertained to RNA splicing (Fig. 1C). To determine whether AS is only in active in samples with unmutated drivers of chemoresistance or has a broader impact, we extended MAJIQ splicing analysis to all 48 D-R pairs. Unsupervised clustering rediscovered the same 5 patient samples and assigned additional 12 r/r leukemias to the same cluster for the total of 17 (Fig. 1D, light blue rectangle), potentially implicating AS in ∼1/3 of B-ALL relapses. We obtained further evidence of SF dysregulation at the level of poison exon inclusion known to result in nonsense-mediated decay (19). For example, in all relapse samples, almost 100% of transcripts encoding the aforementioned SRSF10 carried stop codon–containing exon 3a, presumably rendering them nonfunctional (Supplementary Fig. S1D).

We then considered more carefully the nature of AS events contributing to this cluster (Table 1). We noted that many of them mapped to genes conferring resistance to commonly used B-ALL therapeutics: glucocorticoids (NR3C1), thiopurines (PRPS1 and NT5C2), and antifolates (FPGS). In additional, they affected relapse genes associated with more general, not exclusively cancer cell–intrinsic mechanisms. Examples included the master epigenetic regulator CREBBP and its functional interactor SETD2 and the universal pan-cancer suppressor TP53. Overall, the majority of the splicing variants in Table 1 are putative loss-of-function events as they either directly disrupt open reading frames (mimicking nonsense mutations) or render the transcript vulnerable to nonsense-mediated decay. For instance, TP53 exon 9i contains a premature stop codon, and its inclusion results in the previously validated p53β protein isoform with multiple functional defects (20, 21). However, the effects of CREBBP exon 25/26 skipping and the alternative 5′ splice site in FPGS exon 8 (previously reported in the literature; refs. 22, 23) could not be inferred a priori in the context of the loss-of-function model and required experimental investigation.

Although CREBBP mRNA was affected by several AS events, in-frame skipping of exons 25 and 26 was of potential functional significance as it removes 261bp (87aa) from the HAT domain. This AS event was also among the most robust in our analysis of B-ALL relapses: on average, 50% of all CREBBP reads aligned to the ex24-ex27 junction (Fig. 2A). In the representative example in Fig. 2B, fewer than 30% of reads mapped uniquely to the Δex25/26 isoform in the diagnostic sample (32 reads in the blue sashimi plots), whereas in the paired relapse sample, this event predominated (99 reads in the red sashimi plots). To determine the effects of this event on protein function, we designed two sets of single-guide RNAs mapping to the exons in question and used them to remove the exon 25 to 26 segment from 697 B-ALL cells via CRISPR-mediated genome editing. Using semiquantitative RT-PCR, we detected decreased expression of the CREBBP full-length isoform and robust skipping of its exons 25 and/or 26, with combined skipping of both exons being the most readily detectable event (Fig. 2C). This was also apparent when protein levels were analyzed by immunoblotting (Fig. 2D, top). Because the splicing event affects the HAT domain, we chose, as a functional readout, its effects on histone modifications. We observed that CREBBP exon 25/26 skipping resulted in a commensurate reduction in H3K27Ac while preserving total H3 levels (Fig. 2D). These results support the notion that CREBBP exon 25/26 skipping yields a protein isoform deficient in histone acetyl transferase activity.

The AS event affecting FPGS exon 8 was equally pervasive: only in a few pairs of samples was its direction reversed (Fig. 2E). In a representative relapse sample depicted in Fig. 2F, 65% of reads connected to the downstream exon 9 (186 in the red sashimi plot) originated at the alternative 5′ splice site, whereas in the paired diagnostic sample, this ratio was flipped (32 reads in the blue sashimi plot). To model this event in vitro, we designed a MO antisense splice blocker targeting the canonical 5′ splice site, thus forcing the use of the alternative downstream site (see Materials and Methods). In transfected Nalm6 B-ALL cells, this MO almost completely redirected splicing toward the expected downstream site, resulting in partial intron inclusion and rendering exon 8 “poisonous” via inclusion of a stop codon (“PE” in Fig. 2G). Therefore, the use of FPGS MO also resulted in a sharp decrease in FPGS protein levels, as evidenced by immunoblotting (Fig. 2H). As FPGS is a known driver of sensitivity to MTX (24), we tested the effects of MO-induced FPGS AS on cell survival in response to this antifolate drug. In these experiments, redirecting splicing of FPGS exon 8 toward the downstream splice site resulted in essentially complete resistance to MTX and unattainable IC₅₀ values (Fig. 2I). These data solidified our conclusion that AS phenocopies loss-of-function mutations in this gene previously linked to chemoresistance in r/r B-ALL (5).

Across all relapse-associated AS events identified in our analysis (Table 1), the alternative isoform of NT5C2 with the functionally uncharacterized in-frame exon 6a (previously referred to by us as exon 4a; ref. 8) stood out. This microexon is annotated in RefSeq, and its robust inclusion was detected in normal BM B-cell progenitors in our earlier AS study (8) and in some other populations of hematopoietic cells in our recently unveiled MAJIQlopedia, a reference database for splicing variations across human cancers and normal tissues (25). However, it is not included in the Matched Annotation from the NCBI and EMBL-EBI (MANE), which is a broadly used standard for clinical reporting. Of note, increased frequencies of this event were observed in most relapsed B-ALL samples compared with the matched diagnostic specimens (Fig. 3A). In a representative relapse sample depicted in Fig. 3B, >80% of reads originating in the upstream canonical exon 6 (352 in the red sashimi plot) connected to exon 6a, whereas in the paired diagnostic sample, <25% did (80 in the blue sashimi plot). This finding was consistent with our prior data showing lower levels of NT5C2 exon 6a inclusion in de novo leukemias compared with their normal BM progenitors (8).

To determine whether NT5C2 exon 6a can be found in translatable cap-to-poly(A) protein-coding transcripts, we amplified its cDNA and subjected the resultant 1.7-kb fragment (Supplementary Fig. S2A) to targeted resequencing using the ONT approach (see Materials and Methods). Using this assay, we readily detected exon NT5C2 6a inclusion in the majority of reads corresponding to the Philadelphia chromosome–like MHH-CALL4 cell line with the IGH::CRLF2 rearrangement (Supplementary Fig. S2B). Further support for the translatability of this isoform comes from its inclusion in the UniProt database (accession code: A0A6Q8PHP0) and the inclusion of the exon 6a–encoded tryptic peptide SQVAVQKR in PeptideAtlas (accession code: PAp02559851).

Inclusion of exon 6a in the NT5C2 open reading frame results in the addition of eight extra amino acids near the ATP-binding effector site 2. To determine the effect of this alteration on NT5C2 function, we produced the corresponding recombinant protein (Supplementary Fig. S2C and S2D) and tested its enzymatic activity in nucleotidase assays with IMP as a substrate under basal conditions and following allosteric activation with ATP (14). In these cell-free assays, bacterially produced NT5C2ex6a exhibited elevated enzymatic activity compared with canonical NT5C2 protein in response to low concentrations of ATP (Fig. 3C). This biochemical response was consistent with a gain of function mechanism, similar to that observed for most relapsed-associated NT5C2 mutations characterized to date (14).

To rigorously test a potential role of NT5C2 ex6a in 6-MP resistance, we generated REH B-ALL cell models expressing canonical NT5C2, NT5C2 ex6a, and NT5C2 R238W, a validated gain-of-function resistance driver with the most common hotspot mutation (Fig. 3D; ref. 14). Cell viability assays carried out with increasing concentrations of 6-MP demonstrated that expression of the NT5C2 ex6a isoform confers 6-MP resistance comparable to that induced by the NT5C2 R238W mutation and well above that conferred by the canonical isoform of WT NT5C2 (Fig. 3E; Supplementary Fig. S2E), although the effects were somewhat less potent in Nalm-6 cells (Supplementary Fig. S2F). Metabolically, increased enzymatic activity of NT5C2 mutants results in increased clearance of thiopurine mononucleotide monophosphate metabolites but also in depletion of intracellular nucleotide pools, which in turn renders leukemia cells more sensitive to mizoribine, an inosine monophosphate dehydrogenase inhibitor (16). Therefore, we evaluated the relationship between NT5C2 ex6a and sensitivity to this drug. Once again, both NT5C2 R238W-expressing and NT5C2 ex6a-expressing cells showed convergent phenotypes with respect to increased sensitivity to mizoribine treatment vis-a-vis cells expressing the canonical isoform of WT NT5C2 (Fig. 3F; Supplementary Fig. S2G).

We further noted that the NT5C2 exon 6a–encoded peptide SQVAVQKR contains a characteristic SQ motif, which constitutes the recognition site for the DNA-PK/ATR/ATM subfamily of kinases of the PIKK family (26), also reflected in the Kinase Library (27). This observation suggested a potential role for protein phosphorylation at this site in the regulation of NT5C2 activity. Thus, we generated mutant variants of NT5C2 ex6a, in which the ex6a-encoded serine residue was replaced by the common phosphomimetic aspartic acid (S131D) or by alanine (S131A) to serve as a phosphorylation-deficient control. Evaluation of responses to 6-MP treatment of REH cells expressing each of these constructs revealed that whereas the parental NT5C2 E6a isoform and its S131A derivative conferred equivalent levels of 6-MP resistance, the S131D substitution resulted in the markedly increased resistance phenotype, with one log further increase in the IC₅₀ value (Supplementary Fig. S2H). These results suggest that phosphorylation events at the S131 site might augment NT5C2 exon 6a–driven 6-MP resistance.

To directly assess the impact of NT5C2 ex6a inclusion on leukemia cells growth and therapy responses in vivo, we implanted luciferase-expressing REH cells expressing the canonical WT NT5C2 isoform, NT5C2 R238R, and NT5C2 ex6a into NSG mice and imaged animals on days 5, 8, 12, and 14. Previous reports linked hyperactive NT5C2 to a “fitness cost” phenotype and decreased cell proliferation resulting from the depletion of intracellular nucleotide pools (16). Consistent with these observations, expression of NT5C2 R238W and NT5C2 ex6a induced clear growth retardation when compared with expression of the canonical NT5C2 isoform (Supplementary Fig. S3A). Nevertheless, we observed robust leukemia engraftment, allowing us to test the effect of the three different NT5C2 variants on chemotherapy response by treating each experimental group with either vehicle or 6-MP (Fig. 3G). When compared with the initial treatment timepoint (day 10 after engraftment), the WT model shows a negative fold change (FC; plotted on the y-axis in Supplementary Fig. S3B), which was indicative of cell death. In contrast, the E6a model demonstrates slower cell division under treatment, but the FC remains non-negative, signifying that the treatment does not induce cell death. Notably, the E6a model exhibits an identical FC to the R238W mutant model. These findings further support the comparable roles of these two NT5C2 variants in promoting chemoresistance.

RNA-centric regulatory mechanisms, such as formation of fusion transcripts and circular RNAs, RNA editing, and in particular aberrant mRNA splicing, recently have been recognized as key drivers of neoplastic growth (28). Indeed, mutations in genes encoding core SFs (SF3B1, SRSF2, etc.) are common in adult leukemias, raising the possibility that targeting aberrant splicing programs could be a viable therapeutic option. Although such mutations are exceedingly rare in pediatric AML and ALL, a growing body of work by us and other researchers has revealed that inclusion of “poison” and other noncanonical exons in SF-encoding transcripts is among the most consistent splicing aberration in both B-ALL and T-ALL (8, 29), as well as in many solid tumors (19, 30). Additionally, we show that in TARGET r/r B-ALL samples, SRSF- and HNRNP-encoding genes are frequently affected by CNAs such as deep deletions, likely exacerbating AS.

We also report pervasive dysregulation, at the level of mRNA splicing, of key chemosensitivity genes. In most cases (CREBBP, FPGS, and other examples in Table 1) the AS events are of the loss-of-function nature, making it challenging to devise pharmacologic interventions. Still, an RNA-based diagnostic test detecting, for example, the heavy usage of FPGS ex8 alt 5′ site could guide clinical decisions regarding the use of MTX. Thus, we envision that targeted RNA-seq panels, either short- or long-read, will become valuable companion diagnostics, despite known limitations of underlying technologies (31).

The case of NT5C2 is particularly interesting from the therapeutic standpoint since NT5C2 splicing aberrations leading to the inclusion of cryptic exon 6a ultimately create a proteoform with gain-of-function properties. Since gain-of-function NT5C2 variants interfere with cell proliferation (16), it comes as no surprise that the inclusion of NT5C2 exon 6a is actually reduced in de novo leukemias compared with pro-B cells (8) and is regained only upon treatment with thiopurines as a trade-off between survival and proliferation. This complex dynamic was apparent in our in vivo experiments, in which both NT5C2 E6a and R238W leukemias grew more slowly in NSG mice than the controls but effectively resisted treatment with 6-MP. Of interest, at least in one leukemia variant (PARJZZ), NT5C2 exon 6a inclusion coexists with the hotspot mutation R367Q reported earlier (4). This finding is consistent with an evolutionary model wherein nonmutational mechanisms (AS, NT5C2 phosphorylation) confer initial resistance to 6-MP, enable persistence of proliferating cells, and allow eventual emergence of clones with gain-of-function NT5C2 mutations and a constitutive resistance phenotype.

Robust expression of the NT5C2ex6a isoform not only predicts resistance to thiopurines but may suggest additional therapeutic options, such as treatment with the emerging class of direct NT5C2 small-molecule inhibitors such as CRCD2 (32) and also with clinically used inosine monophosphate dehydrogenase inhibitors and immunosuppressive agents mycophenolate mofetil and mizoribine (33). In support of this notion, NT5C2ex6a-expressing REH and NALM6 cells were found to exhibit heightened sensitivity to mizoribine compared with the isogenic cell lines reconstituted with the canonical NT5C2 isoform.

Other ways to therapeutically target NT5C2 dysregulation are likely to exist. Our recently published work elucidated the role of NT5C2 post-transcriptional modification in chemoresistance, with phosphorylation of the Ser-502 residue by a yet-to-be-identified protein kinase increasing its activity against 6-MP (32). Our current data showing that substituting Ser at position 131 with a phosphomimetic Asp augments NT5C2ex6a-induced resistance to 6-MP lend further credence to the model linking NT5C2 phosphorylation and chemoresistance functions. If future research confirms the involvement of ATR/ATM family kinases in phosphorylating the NT5C2 exon 6a-encoded SQ motif, a variety of small-molecule inhibitors of these enzymes (34) will be available for preclinical studies and clinical trials aiming to resensitize r/r B-ALL to chemotherapy and in doing so improve outcomes in this aggressive and often fatal childhood cancer.

M. Torres-Diz reports a patent for PCT/US2023/025031 pending. C. Reglero reports her work was supported by a Leukemia and Lymphoma Society Special Fellow award. M.M. Li reports personal fees from Bayer HealthCare Pharmaceuticals Inc. and Agilent Technologies outside the submitted work. Y. Barash reports grants from the NIH during the conduct of the study and grants from the NIH, CureBRCA, and Basser Institute outside the submitted work. MAJIQ 2.0 used in this study is available for licensing for free for academics and for a fee for commercial usage. Some of the commercial licensing revenue goes to Y. Barash and members of the Barash laboratoy. A. Ferrando reports grants from the NIH, Leukemia and Lymphoma Society, Alex’s Lemonade Stand Foundation, and Irving Institute for Clinical and Translational Research of Columbia University during the conduct of the study and other support from Regeneron Pharmaceuticals outside the submitted work; in addition, A. Ferrando has a patent for NT5C2 inhibitors issued. A. Thomas-Tikhonenko reports grants from the NIH, Pennsylvania Department of Health, St. Baldrick’s Foundation, Stand Up To Cancer (SU2C), The V Foundation for Cancer Research, The Emerson Collective, and Alex’s Lemonade Stand Foundation during the conduct of the study and grants from CureSearch for Childhood Cancer, Pfizer ASPIRE Onc/Hem Program and personal fees from the University of Miami, the Italian Association for Cancer Research, Guidepoint, and Houston Methodist outside the submitted work; in addition, A. Thomas-Tikhonenko has a patent for PCT/US2023/025031 pending. No disclosures were reported by the other authors.

M. Torres-Diz: Conceptualization, data curation, formal analysis, validation, investigation, writing–original draft, writing–review and editing. C. Reglero: Investigation. C.D. Falkenstein: Investigation. A. Castro: Investigation. K.E. Hayer: Data curation, formal analysis. C.M. Radens: Data curation, formal analysis. M. Quesnel-Vallières: Data curation, formal analysis. Z. Ang: Conceptualization, writing–review and editing. P. Sehgal: Resources. M.M. Li: Conceptualization. Y. Barash: Conceptualization, data curation, formal analysis. S.K. Tasian: Supervision. A. Ferrando: Conceptualization, supervision, writing–review and editing. A. Thomas-Tikhonenko: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

The authors are grateful to past and present members of the Barash, Tasian, and Thomas-Tikhonenko laboratories and Penn’s RNA Club for many helpful discussions. This work was supported by grants from the NIH/NCI (U01 CA232563 to A. Thomas-Tikhonenko and Y. Barash, U01 CA232486 and U01 CA243072 to S.K. Tasian, and P30 CA013696, R35 CA210065, and CA216981 to A. Ferrando), Pennsylvania Department of Health SFY22 CURE Non-Formula Collaborative Research on Childhood and Adolescent Blood Cancers (#67-173 to A. Thomas-Tikhonenko), U.S. Department of Defense (CA180683P1 to S.K. Tasian), St. Baldrick’s Foundation [EPICC Team and St. Baldrick’s SU2C Dream Team Translational Cancer Research Grant (SU2C-AACR-DT-27-17) to A. Thomas-Tikhonenko and S.K. Tasian], The V Foundation for Cancer Research (T2018-014 to A. Thomas-Tikhonenko), The Emerson Collective (886246066 to A. Thomas-Tikhonenko), Alex’s Lemonade Stand Foundation (Innovation Awards to ATT and AF), and Leukemia & Lymphoma Society (Translational Research Grant 6455-15 and Screen to Lead Grant 8011-18 to A. Ferrando). The indicated SU2C research grant is administered by the AACR, the scientific partner of SU2C. C. Reglero is a past Special Fellow and S.K. Tasian is a Scholar of the Leukemia & Lymphoma Society. A. Thomas-Tikhonenko and A. Ferrando acknowledge support from the SPROUT Program and Accelerating Cancer Therapeutics Program at CHOP and Columbia University, respectively. M. Torres-Diz and A. Thomas-Tikhonenko further acknowledge support from The Ellen Weisberg Fund: Advancing Breakthroughs in Pediatric Cancer. S.K. Tasian is Joshua Kahan Endowed Chair in Pediatric Leukemia Research, and A. Thomas-Tikhonenko is Mildred L. Roeckle Endowed Chair in Pathology at Children’s Hospital of Philadelphia.